blob: ec5ddccbf82e9dbef3e23b7a6faa72283360eee4 [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>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090016#include <linux/gfp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#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
Mel Gorman33906bc2010-08-09 17:19:16 -070051#define CREATE_TRACE_POINTS
52#include <trace/events/vmscan.h>
53
Linus Torvalds1da177e2005-04-16 15:20:36 -070054struct scan_control {
Linus Torvalds1da177e2005-04-16 15:20:36 -070055 /* Incremented by the number of inactive pages that were scanned */
56 unsigned long nr_scanned;
57
Rik van Riela79311c2009-01-06 14:40:01 -080058 /* Number of pages freed so far during a call to shrink_zones() */
59 unsigned long nr_reclaimed;
60
KOSAKI Motohiro22fba332009-12-14 17:59:10 -080061 /* How many pages shrink_list() should reclaim */
62 unsigned long nr_to_reclaim;
63
KOSAKI Motohiro7b517552009-12-14 17:59:12 -080064 unsigned long hibernation_mode;
65
Linus Torvalds1da177e2005-04-16 15:20:36 -070066 /* This context's GFP mask */
Al Viro6daa0e22005-10-21 03:18:50 -040067 gfp_t gfp_mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -070068
69 int may_writepage;
70
Johannes Weinera6dc60f82009-03-31 15:19:30 -070071 /* Can mapped pages be reclaimed? */
72 int may_unmap;
Christoph Lameterf1fd1062006-01-18 17:42:30 -080073
KOSAKI Motohiro2e2e4252009-04-21 12:24:57 -070074 /* Can pages be swapped as part of reclaim? */
75 int may_swap;
76
Rafael J. Wysockid6277db2006-06-23 02:03:18 -070077 int swappiness;
Nick Piggin408d8542006-09-25 23:31:27 -070078
Andy Whitcroft5ad333e2007-07-17 04:03:16 -070079 int order;
Balbir Singh66e17072008-02-07 00:13:56 -080080
KOSAKI Motohiro5f53e762010-05-24 14:32:37 -070081 /*
82 * Intend to reclaim enough contenious memory rather than to reclaim
83 * enough amount memory. I.e, it's the mode for high order allocation.
84 */
85 bool lumpy_reclaim_mode;
86
Balbir Singh66e17072008-02-07 00:13:56 -080087 /* Which cgroup do we reclaim from */
88 struct mem_cgroup *mem_cgroup;
89
KAMEZAWA Hiroyuki327c0e92009-03-31 15:23:31 -070090 /*
91 * Nodemask of nodes allowed by the caller. If NULL, all nodes
92 * are scanned.
93 */
94 nodemask_t *nodemask;
Linus Torvalds1da177e2005-04-16 15:20:36 -070095};
96
Linus Torvalds1da177e2005-04-16 15:20:36 -070097#define lru_to_page(_head) (list_entry((_head)->prev, struct page, lru))
98
99#ifdef ARCH_HAS_PREFETCH
100#define prefetch_prev_lru_page(_page, _base, _field) \
101 do { \
102 if ((_page)->lru.prev != _base) { \
103 struct page *prev; \
104 \
105 prev = lru_to_page(&(_page->lru)); \
106 prefetch(&prev->_field); \
107 } \
108 } while (0)
109#else
110#define prefetch_prev_lru_page(_page, _base, _field) do { } while (0)
111#endif
112
113#ifdef ARCH_HAS_PREFETCHW
114#define prefetchw_prev_lru_page(_page, _base, _field) \
115 do { \
116 if ((_page)->lru.prev != _base) { \
117 struct page *prev; \
118 \
119 prev = lru_to_page(&(_page->lru)); \
120 prefetchw(&prev->_field); \
121 } \
122 } while (0)
123#else
124#define prefetchw_prev_lru_page(_page, _base, _field) do { } while (0)
125#endif
126
127/*
128 * From 0 .. 100. Higher means more swappy.
129 */
130int vm_swappiness = 60;
Andrew Mortonbd1e22b2006-06-23 02:03:47 -0700131long vm_total_pages; /* The total number of pages which the VM controls */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132
133static LIST_HEAD(shrinker_list);
134static DECLARE_RWSEM(shrinker_rwsem);
135
Balbir Singh00f0b822008-03-04 14:28:39 -0800136#ifdef CONFIG_CGROUP_MEM_RES_CTLR
KAMEZAWA Hiroyukie72e2bd2009-01-07 18:08:23 -0800137#define scanning_global_lru(sc) (!(sc)->mem_cgroup)
KAMEZAWA Hiroyuki91a45472008-02-07 00:14:29 -0800138#else
KAMEZAWA Hiroyukie72e2bd2009-01-07 18:08:23 -0800139#define scanning_global_lru(sc) (1)
KAMEZAWA Hiroyuki91a45472008-02-07 00:14:29 -0800140#endif
141
KOSAKI Motohiro6e901572009-01-07 18:08:15 -0800142static struct zone_reclaim_stat *get_reclaim_stat(struct zone *zone,
143 struct scan_control *sc)
144{
KAMEZAWA Hiroyukie72e2bd2009-01-07 18:08:23 -0800145 if (!scanning_global_lru(sc))
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800146 return mem_cgroup_get_reclaim_stat(sc->mem_cgroup, zone);
147
KOSAKI Motohiro6e901572009-01-07 18:08:15 -0800148 return &zone->reclaim_stat;
149}
150
Vincent Li0b217672009-09-21 17:03:09 -0700151static unsigned long zone_nr_lru_pages(struct zone *zone,
152 struct scan_control *sc, enum lru_list lru)
KOSAKI Motohiroc9f299d2009-01-07 18:08:16 -0800153{
KAMEZAWA Hiroyukie72e2bd2009-01-07 18:08:23 -0800154 if (!scanning_global_lru(sc))
KOSAKI Motohiroa3d8e052009-01-07 18:08:19 -0800155 return mem_cgroup_zone_nr_pages(sc->mem_cgroup, zone, lru);
156
KOSAKI Motohiroc9f299d2009-01-07 18:08:16 -0800157 return zone_page_state(zone, NR_LRU_BASE + lru);
158}
159
160
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161/*
162 * Add a shrinker callback to be called from the vm
163 */
Rusty Russell8e1f9362007-07-17 04:03:17 -0700164void register_shrinker(struct shrinker *shrinker)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165{
Rusty Russell8e1f9362007-07-17 04:03:17 -0700166 shrinker->nr = 0;
167 down_write(&shrinker_rwsem);
168 list_add_tail(&shrinker->list, &shrinker_list);
169 up_write(&shrinker_rwsem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170}
Rusty Russell8e1f9362007-07-17 04:03:17 -0700171EXPORT_SYMBOL(register_shrinker);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172
173/*
174 * Remove one
175 */
Rusty Russell8e1f9362007-07-17 04:03:17 -0700176void unregister_shrinker(struct shrinker *shrinker)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177{
178 down_write(&shrinker_rwsem);
179 list_del(&shrinker->list);
180 up_write(&shrinker_rwsem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181}
Rusty Russell8e1f9362007-07-17 04:03:17 -0700182EXPORT_SYMBOL(unregister_shrinker);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183
184#define SHRINK_BATCH 128
185/*
186 * Call the shrink functions to age shrinkable caches
187 *
188 * Here we assume it costs one seek to replace a lru page and that it also
189 * takes a seek to recreate a cache object. With this in mind we age equal
190 * percentages of the lru and ageable caches. This should balance the seeks
191 * generated by these structures.
192 *
Simon Arlott183ff222007-10-20 01:27:18 +0200193 * If the vm encountered mapped pages on the LRU it increase the pressure on
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194 * slab to avoid swapping.
195 *
196 * We do weird things to avoid (scanned*seeks*entries) overflowing 32 bits.
197 *
198 * `lru_pages' represents the number of on-LRU pages in all the zones which
199 * are eligible for the caller's allocation attempt. It is used for balancing
200 * slab reclaim versus page reclaim.
akpm@osdl.orgb15e0902005-06-21 17:14:35 -0700201 *
202 * Returns the number of slab objects which we shrunk.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203 */
Andrew Morton69e05942006-03-22 00:08:19 -0800204unsigned long shrink_slab(unsigned long scanned, gfp_t gfp_mask,
205 unsigned long lru_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206{
207 struct shrinker *shrinker;
Andrew Morton69e05942006-03-22 00:08:19 -0800208 unsigned long ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209
210 if (scanned == 0)
211 scanned = SWAP_CLUSTER_MAX;
212
213 if (!down_read_trylock(&shrinker_rwsem))
akpm@osdl.orgb15e0902005-06-21 17:14:35 -0700214 return 1; /* Assume we'll be able to shrink next time */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215
216 list_for_each_entry(shrinker, &shrinker_list, list) {
217 unsigned long long delta;
218 unsigned long total_scan;
Dave Chinner7f8275d2010-07-19 14:56:17 +1000219 unsigned long max_pass;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220
Dave Chinner7f8275d2010-07-19 14:56:17 +1000221 max_pass = (*shrinker->shrink)(shrinker, 0, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222 delta = (4 * scanned) / shrinker->seeks;
Andrea Arcangeliea164d72005-11-28 13:44:15 -0800223 delta *= max_pass;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224 do_div(delta, lru_pages + 1);
225 shrinker->nr += delta;
Andrea Arcangeliea164d72005-11-28 13:44:15 -0800226 if (shrinker->nr < 0) {
David Rientjes88c3bd72009-03-31 15:23:29 -0700227 printk(KERN_ERR "shrink_slab: %pF negative objects to "
228 "delete nr=%ld\n",
229 shrinker->shrink, shrinker->nr);
Andrea Arcangeliea164d72005-11-28 13:44:15 -0800230 shrinker->nr = max_pass;
231 }
232
233 /*
234 * Avoid risking looping forever due to too large nr value:
235 * never try to free more than twice the estimate number of
236 * freeable entries.
237 */
238 if (shrinker->nr > max_pass * 2)
239 shrinker->nr = max_pass * 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240
241 total_scan = shrinker->nr;
242 shrinker->nr = 0;
243
244 while (total_scan >= SHRINK_BATCH) {
245 long this_scan = SHRINK_BATCH;
246 int shrink_ret;
akpm@osdl.orgb15e0902005-06-21 17:14:35 -0700247 int nr_before;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248
Dave Chinner7f8275d2010-07-19 14:56:17 +1000249 nr_before = (*shrinker->shrink)(shrinker, 0, gfp_mask);
250 shrink_ret = (*shrinker->shrink)(shrinker, this_scan,
251 gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252 if (shrink_ret == -1)
253 break;
akpm@osdl.orgb15e0902005-06-21 17:14:35 -0700254 if (shrink_ret < nr_before)
255 ret += nr_before - shrink_ret;
Christoph Lameterf8891e52006-06-30 01:55:45 -0700256 count_vm_events(SLABS_SCANNED, this_scan);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257 total_scan -= this_scan;
258
259 cond_resched();
260 }
261
262 shrinker->nr += total_scan;
263 }
264 up_read(&shrinker_rwsem);
akpm@osdl.orgb15e0902005-06-21 17:14:35 -0700265 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266}
267
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268static inline int is_page_cache_freeable(struct page *page)
269{
Johannes Weinerceddc3a2009-09-21 17:03:00 -0700270 /*
271 * A freeable page cache page is referenced only by the caller
272 * that isolated the page, the page cache radix tree and
273 * optional buffer heads at page->private.
274 */
Johannes Weineredcf4742009-09-21 17:02:59 -0700275 return page_count(page) - page_has_private(page) == 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276}
277
278static int may_write_to_queue(struct backing_dev_info *bdi)
279{
Christoph Lameter930d9152006-01-08 01:00:47 -0800280 if (current->flags & PF_SWAPWRITE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 return 1;
282 if (!bdi_write_congested(bdi))
283 return 1;
284 if (bdi == current->backing_dev_info)
285 return 1;
286 return 0;
287}
288
289/*
290 * We detected a synchronous write error writing a page out. Probably
291 * -ENOSPC. We need to propagate that into the address_space for a subsequent
292 * fsync(), msync() or close().
293 *
294 * The tricky part is that after writepage we cannot touch the mapping: nothing
295 * prevents it from being freed up. But we have a ref on the page and once
296 * that page is locked, the mapping is pinned.
297 *
298 * We're allowed to run sleeping lock_page() here because we know the caller has
299 * __GFP_FS.
300 */
301static void handle_write_error(struct address_space *mapping,
302 struct page *page, int error)
303{
Nick Piggina6aa62a2010-07-20 13:24:25 -0700304 lock_page_nosync(page);
Guillaume Chazarain3e9f45b2007-05-08 00:23:25 -0700305 if (page_mapping(page) == mapping)
306 mapping_set_error(mapping, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307 unlock_page(page);
308}
309
Andy Whitcroftc661b072007-08-22 14:01:26 -0700310/* Request for sync pageout. */
311enum pageout_io {
312 PAGEOUT_IO_ASYNC,
313 PAGEOUT_IO_SYNC,
314};
315
Christoph Lameter04e62a22006-06-23 02:03:38 -0700316/* possible outcome of pageout() */
317typedef enum {
318 /* failed to write page out, page is locked */
319 PAGE_KEEP,
320 /* move page to the active list, page is locked */
321 PAGE_ACTIVATE,
322 /* page has been sent to the disk successfully, page is unlocked */
323 PAGE_SUCCESS,
324 /* page is clean and locked */
325 PAGE_CLEAN,
326} pageout_t;
327
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328/*
Andrew Morton1742f192006-03-22 00:08:21 -0800329 * pageout is called by shrink_page_list() for each dirty page.
330 * Calls ->writepage().
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 */
Andy Whitcroftc661b072007-08-22 14:01:26 -0700332static pageout_t pageout(struct page *page, struct address_space *mapping,
333 enum pageout_io sync_writeback)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334{
335 /*
336 * If the page is dirty, only perform writeback if that write
337 * will be non-blocking. To prevent this allocation from being
338 * stalled by pagecache activity. But note that there may be
339 * stalls if we need to run get_block(). We could test
340 * PagePrivate for that.
341 *
Vincent Li6aceb532009-12-14 17:58:49 -0800342 * If this process is currently in __generic_file_aio_write() against
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343 * this page's queue, we can perform writeback even if that
344 * will block.
345 *
346 * If the page is swapcache, write it back even if that would
347 * block, for some throttling. This happens by accident, because
348 * swap_backing_dev_info is bust: it doesn't reflect the
349 * congestion state of the swapdevs. Easy to fix, if needed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350 */
351 if (!is_page_cache_freeable(page))
352 return PAGE_KEEP;
353 if (!mapping) {
354 /*
355 * Some data journaling orphaned pages can have
356 * page->mapping == NULL while being dirty with clean buffers.
357 */
David Howells266cf652009-04-03 16:42:36 +0100358 if (page_has_private(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359 if (try_to_free_buffers(page)) {
360 ClearPageDirty(page);
Harvey Harrisond40cee22008-04-30 00:55:07 -0700361 printk("%s: orphaned page\n", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 return PAGE_CLEAN;
363 }
364 }
365 return PAGE_KEEP;
366 }
367 if (mapping->a_ops->writepage == NULL)
368 return PAGE_ACTIVATE;
369 if (!may_write_to_queue(mapping->backing_dev_info))
370 return PAGE_KEEP;
371
372 if (clear_page_dirty_for_io(page)) {
373 int res;
374 struct writeback_control wbc = {
375 .sync_mode = WB_SYNC_NONE,
376 .nr_to_write = SWAP_CLUSTER_MAX,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700377 .range_start = 0,
378 .range_end = LLONG_MAX,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 .nonblocking = 1,
380 .for_reclaim = 1,
381 };
382
383 SetPageReclaim(page);
384 res = mapping->a_ops->writepage(page, &wbc);
385 if (res < 0)
386 handle_write_error(mapping, page, res);
Zach Brown994fc28c2005-12-15 14:28:17 -0800387 if (res == AOP_WRITEPAGE_ACTIVATE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388 ClearPageReclaim(page);
389 return PAGE_ACTIVATE;
390 }
Andy Whitcroftc661b072007-08-22 14:01:26 -0700391
392 /*
393 * Wait on writeback if requested to. This happens when
394 * direct reclaiming a large contiguous area and the
395 * first attempt to free a range of pages fails.
396 */
397 if (PageWriteback(page) && sync_writeback == PAGEOUT_IO_SYNC)
398 wait_on_page_writeback(page);
399
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400 if (!PageWriteback(page)) {
401 /* synchronous write or broken a_ops? */
402 ClearPageReclaim(page);
403 }
Mel Gorman755f0222010-08-09 17:19:18 -0700404 trace_mm_vmscan_writepage(page,
405 trace_reclaim_flags(page, sync_writeback));
Andrew Mortone129b5c2006-09-27 01:50:00 -0700406 inc_zone_page_state(page, NR_VMSCAN_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407 return PAGE_SUCCESS;
408 }
409
410 return PAGE_CLEAN;
411}
412
Andrew Mortona649fd92006-10-17 00:09:36 -0700413/*
Nick Piggine2867812008-07-25 19:45:30 -0700414 * Same as remove_mapping, but if the page is removed from the mapping, it
415 * gets returned with a refcount of 0.
Andrew Mortona649fd92006-10-17 00:09:36 -0700416 */
Nick Piggine2867812008-07-25 19:45:30 -0700417static int __remove_mapping(struct address_space *mapping, struct page *page)
Christoph Lameter49d2e9c2006-01-08 01:00:48 -0800418{
Nick Piggin28e4d962006-09-25 23:31:23 -0700419 BUG_ON(!PageLocked(page));
420 BUG_ON(mapping != page_mapping(page));
Christoph Lameter49d2e9c2006-01-08 01:00:48 -0800421
Nick Piggin19fd6232008-07-25 19:45:32 -0700422 spin_lock_irq(&mapping->tree_lock);
Christoph Lameter49d2e9c2006-01-08 01:00:48 -0800423 /*
Nick Piggin0fd0e6b2006-09-27 01:50:02 -0700424 * The non racy check for a busy page.
425 *
426 * Must be careful with the order of the tests. When someone has
427 * a ref to the page, it may be possible that they dirty it then
428 * drop the reference. So if PageDirty is tested before page_count
429 * here, then the following race may occur:
430 *
431 * get_user_pages(&page);
432 * [user mapping goes away]
433 * write_to(page);
434 * !PageDirty(page) [good]
435 * SetPageDirty(page);
436 * put_page(page);
437 * !page_count(page) [good, discard it]
438 *
439 * [oops, our write_to data is lost]
440 *
441 * Reversing the order of the tests ensures such a situation cannot
442 * escape unnoticed. The smp_rmb is needed to ensure the page->flags
443 * load is not satisfied before that of page->_count.
444 *
445 * Note that if SetPageDirty is always performed via set_page_dirty,
446 * and thus under tree_lock, then this ordering is not required.
Christoph Lameter49d2e9c2006-01-08 01:00:48 -0800447 */
Nick Piggine2867812008-07-25 19:45:30 -0700448 if (!page_freeze_refs(page, 2))
Christoph Lameter49d2e9c2006-01-08 01:00:48 -0800449 goto cannot_free;
Nick Piggine2867812008-07-25 19:45:30 -0700450 /* note: atomic_cmpxchg in page_freeze_refs provides the smp_rmb */
451 if (unlikely(PageDirty(page))) {
452 page_unfreeze_refs(page, 2);
Christoph Lameter49d2e9c2006-01-08 01:00:48 -0800453 goto cannot_free;
Nick Piggine2867812008-07-25 19:45:30 -0700454 }
Christoph Lameter49d2e9c2006-01-08 01:00:48 -0800455
456 if (PageSwapCache(page)) {
457 swp_entry_t swap = { .val = page_private(page) };
458 __delete_from_swap_cache(page);
Nick Piggin19fd6232008-07-25 19:45:32 -0700459 spin_unlock_irq(&mapping->tree_lock);
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -0700460 swapcache_free(swap, page);
Nick Piggine2867812008-07-25 19:45:30 -0700461 } else {
462 __remove_from_page_cache(page);
Nick Piggin19fd6232008-07-25 19:45:32 -0700463 spin_unlock_irq(&mapping->tree_lock);
Daisuke Nishimurae767e052009-05-28 14:34:28 -0700464 mem_cgroup_uncharge_cache_page(page);
Christoph Lameter49d2e9c2006-01-08 01:00:48 -0800465 }
466
Christoph Lameter49d2e9c2006-01-08 01:00:48 -0800467 return 1;
468
469cannot_free:
Nick Piggin19fd6232008-07-25 19:45:32 -0700470 spin_unlock_irq(&mapping->tree_lock);
Christoph Lameter49d2e9c2006-01-08 01:00:48 -0800471 return 0;
472}
473
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474/*
Nick Piggine2867812008-07-25 19:45:30 -0700475 * Attempt to detach a locked page from its ->mapping. If it is dirty or if
476 * someone else has a ref on the page, abort and return 0. If it was
477 * successfully detached, return 1. Assumes the caller has a single ref on
478 * this page.
479 */
480int remove_mapping(struct address_space *mapping, struct page *page)
481{
482 if (__remove_mapping(mapping, page)) {
483 /*
484 * Unfreezing the refcount with 1 rather than 2 effectively
485 * drops the pagecache ref for us without requiring another
486 * atomic operation.
487 */
488 page_unfreeze_refs(page, 1);
489 return 1;
490 }
491 return 0;
492}
493
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700494/**
495 * putback_lru_page - put previously isolated page onto appropriate LRU list
496 * @page: page to be put back to appropriate lru list
497 *
498 * Add previously isolated @page to appropriate LRU list.
499 * Page may still be unevictable for other reasons.
500 *
501 * lru_lock must not be held, interrupts must be enabled.
502 */
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700503void putback_lru_page(struct page *page)
504{
505 int lru;
506 int active = !!TestClearPageActive(page);
Lee Schermerhornbbfd28e2008-10-18 20:26:40 -0700507 int was_unevictable = PageUnevictable(page);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700508
509 VM_BUG_ON(PageLRU(page));
510
511redo:
512 ClearPageUnevictable(page);
513
514 if (page_evictable(page, NULL)) {
515 /*
516 * For evictable pages, we can use the cache.
517 * In event of a race, worst case is we end up with an
518 * unevictable page on [in]active list.
519 * We know how to handle that.
520 */
Johannes Weiner401a8e12009-09-21 17:02:58 -0700521 lru = active + page_lru_base_type(page);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700522 lru_cache_add_lru(page, lru);
523 } else {
524 /*
525 * Put unevictable pages directly on zone's unevictable
526 * list.
527 */
528 lru = LRU_UNEVICTABLE;
529 add_page_to_unevictable_list(page);
Johannes Weiner6a7b9542009-10-26 16:50:00 -0700530 /*
531 * When racing with an mlock clearing (page is
532 * unlocked), make sure that if the other thread does
533 * not observe our setting of PG_lru and fails
534 * isolation, we see PG_mlocked cleared below and move
535 * the page back to the evictable list.
536 *
537 * The other side is TestClearPageMlocked().
538 */
539 smp_mb();
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700540 }
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700541
542 /*
543 * page's status can change while we move it among lru. If an evictable
544 * page is on unevictable list, it never be freed. To avoid that,
545 * check after we added it to the list, again.
546 */
547 if (lru == LRU_UNEVICTABLE && page_evictable(page, NULL)) {
548 if (!isolate_lru_page(page)) {
549 put_page(page);
550 goto redo;
551 }
552 /* This means someone else dropped this page from LRU
553 * So, it will be freed or putback to LRU again. There is
554 * nothing to do here.
555 */
556 }
557
Lee Schermerhornbbfd28e2008-10-18 20:26:40 -0700558 if (was_unevictable && lru != LRU_UNEVICTABLE)
559 count_vm_event(UNEVICTABLE_PGRESCUED);
560 else if (!was_unevictable && lru == LRU_UNEVICTABLE)
561 count_vm_event(UNEVICTABLE_PGCULLED);
562
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700563 put_page(page); /* drop ref from isolate */
564}
565
Johannes Weinerdfc8d632010-03-05 13:42:19 -0800566enum page_references {
567 PAGEREF_RECLAIM,
568 PAGEREF_RECLAIM_CLEAN,
Johannes Weiner645747462010-03-05 13:42:22 -0800569 PAGEREF_KEEP,
Johannes Weinerdfc8d632010-03-05 13:42:19 -0800570 PAGEREF_ACTIVATE,
571};
572
573static enum page_references page_check_references(struct page *page,
574 struct scan_control *sc)
575{
Johannes Weiner645747462010-03-05 13:42:22 -0800576 int referenced_ptes, referenced_page;
Johannes Weinerdfc8d632010-03-05 13:42:19 -0800577 unsigned long vm_flags;
Johannes Weinerdfc8d632010-03-05 13:42:19 -0800578
Johannes Weiner645747462010-03-05 13:42:22 -0800579 referenced_ptes = page_referenced(page, 1, sc->mem_cgroup, &vm_flags);
580 referenced_page = TestClearPageReferenced(page);
Johannes Weinerdfc8d632010-03-05 13:42:19 -0800581
582 /* Lumpy reclaim - ignore references */
KOSAKI Motohiro5f53e762010-05-24 14:32:37 -0700583 if (sc->lumpy_reclaim_mode)
Johannes Weinerdfc8d632010-03-05 13:42:19 -0800584 return PAGEREF_RECLAIM;
585
586 /*
587 * Mlock lost the isolation race with us. Let try_to_unmap()
588 * move the page to the unevictable list.
589 */
590 if (vm_flags & VM_LOCKED)
591 return PAGEREF_RECLAIM;
592
Johannes Weiner645747462010-03-05 13:42:22 -0800593 if (referenced_ptes) {
594 if (PageAnon(page))
595 return PAGEREF_ACTIVATE;
596 /*
597 * All mapped pages start out with page table
598 * references from the instantiating fault, so we need
599 * to look twice if a mapped file page is used more
600 * than once.
601 *
602 * Mark it and spare it for another trip around the
603 * inactive list. Another page table reference will
604 * lead to its activation.
605 *
606 * Note: the mark is set for activated pages as well
607 * so that recently deactivated but used pages are
608 * quickly recovered.
609 */
610 SetPageReferenced(page);
611
612 if (referenced_page)
613 return PAGEREF_ACTIVATE;
614
615 return PAGEREF_KEEP;
616 }
Johannes Weinerdfc8d632010-03-05 13:42:19 -0800617
618 /* Reclaim if clean, defer dirty pages to writeback */
Johannes Weiner645747462010-03-05 13:42:22 -0800619 if (referenced_page)
620 return PAGEREF_RECLAIM_CLEAN;
621
622 return PAGEREF_RECLAIM;
Johannes Weinerdfc8d632010-03-05 13:42:19 -0800623}
624
Mel Gormanabe4c3b2010-08-09 17:19:31 -0700625static noinline_for_stack void free_page_list(struct list_head *free_pages)
626{
627 struct pagevec freed_pvec;
628 struct page *page, *tmp;
629
630 pagevec_init(&freed_pvec, 1);
631
632 list_for_each_entry_safe(page, tmp, free_pages, lru) {
633 list_del(&page->lru);
634 if (!pagevec_add(&freed_pvec, page)) {
635 __pagevec_free(&freed_pvec);
636 pagevec_reinit(&freed_pvec);
637 }
638 }
639
640 pagevec_free(&freed_pvec);
641}
642
Nick Piggine2867812008-07-25 19:45:30 -0700643/*
Andrew Morton1742f192006-03-22 00:08:21 -0800644 * shrink_page_list() returns the number of reclaimed pages
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 */
Andrew Morton1742f192006-03-22 00:08:21 -0800646static unsigned long shrink_page_list(struct list_head *page_list,
Andy Whitcroftc661b072007-08-22 14:01:26 -0700647 struct scan_control *sc,
648 enum pageout_io sync_writeback)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649{
650 LIST_HEAD(ret_pages);
Mel Gormanabe4c3b2010-08-09 17:19:31 -0700651 LIST_HEAD(free_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 int pgactivate = 0;
Andrew Morton05ff5132006-03-22 00:08:20 -0800653 unsigned long nr_reclaimed = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654
655 cond_resched();
656
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 while (!list_empty(page_list)) {
Johannes Weinerdfc8d632010-03-05 13:42:19 -0800658 enum page_references references;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659 struct address_space *mapping;
660 struct page *page;
661 int may_enter_fs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662
663 cond_resched();
664
665 page = lru_to_page(page_list);
666 list_del(&page->lru);
667
Nick Piggin529ae9a2008-08-02 12:01:03 +0200668 if (!trylock_page(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669 goto keep;
670
Nick Piggin725d7042006-09-25 23:30:55 -0700671 VM_BUG_ON(PageActive(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672
673 sc->nr_scanned++;
Christoph Lameter80e43422006-02-11 17:55:53 -0800674
Nick Pigginb291f002008-10-18 20:26:44 -0700675 if (unlikely(!page_evictable(page, NULL)))
676 goto cull_mlocked;
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700677
Johannes Weinera6dc60f82009-03-31 15:19:30 -0700678 if (!sc->may_unmap && page_mapped(page))
Christoph Lameter80e43422006-02-11 17:55:53 -0800679 goto keep_locked;
680
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 /* Double the slab pressure for mapped and swapcache pages */
682 if (page_mapped(page) || PageSwapCache(page))
683 sc->nr_scanned++;
684
Andy Whitcroftc661b072007-08-22 14:01:26 -0700685 may_enter_fs = (sc->gfp_mask & __GFP_FS) ||
686 (PageSwapCache(page) && (sc->gfp_mask & __GFP_IO));
687
688 if (PageWriteback(page)) {
689 /*
690 * Synchronous reclaim is performed in two passes,
691 * first an asynchronous pass over the list to
692 * start parallel writeback, and a second synchronous
693 * pass to wait for the IO to complete. Wait here
694 * for any page for which writeback has already
695 * started.
696 */
697 if (sync_writeback == PAGEOUT_IO_SYNC && may_enter_fs)
698 wait_on_page_writeback(page);
Andrew Morton4dd4b922008-03-24 12:29:52 -0700699 else
Andy Whitcroftc661b072007-08-22 14:01:26 -0700700 goto keep_locked;
701 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702
Johannes Weinerdfc8d632010-03-05 13:42:19 -0800703 references = page_check_references(page, sc);
704 switch (references) {
705 case PAGEREF_ACTIVATE:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 goto activate_locked;
Johannes Weiner645747462010-03-05 13:42:22 -0800707 case PAGEREF_KEEP:
708 goto keep_locked;
Johannes Weinerdfc8d632010-03-05 13:42:19 -0800709 case PAGEREF_RECLAIM:
710 case PAGEREF_RECLAIM_CLEAN:
711 ; /* try to reclaim the page below */
712 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714 /*
715 * Anonymous process memory has backing store?
716 * Try to allocate it some swap space here.
717 */
Nick Pigginb291f002008-10-18 20:26:44 -0700718 if (PageAnon(page) && !PageSwapCache(page)) {
Hugh Dickins63eb6b92008-11-19 15:36:37 -0800719 if (!(sc->gfp_mask & __GFP_IO))
720 goto keep_locked;
Hugh Dickinsac47b002009-01-06 14:39:39 -0800721 if (!add_to_swap(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722 goto activate_locked;
Hugh Dickins63eb6b92008-11-19 15:36:37 -0800723 may_enter_fs = 1;
Nick Pigginb291f002008-10-18 20:26:44 -0700724 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725
726 mapping = page_mapping(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727
728 /*
729 * The page is mapped into the page tables of one or more
730 * processes. Try to unmap it here.
731 */
732 if (page_mapped(page) && mapping) {
Andi Kleen14fa31b2009-09-16 11:50:10 +0200733 switch (try_to_unmap(page, TTU_UNMAP)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734 case SWAP_FAIL:
735 goto activate_locked;
736 case SWAP_AGAIN:
737 goto keep_locked;
Nick Pigginb291f002008-10-18 20:26:44 -0700738 case SWAP_MLOCK:
739 goto cull_mlocked;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740 case SWAP_SUCCESS:
741 ; /* try to free the page below */
742 }
743 }
744
745 if (PageDirty(page)) {
Johannes Weinerdfc8d632010-03-05 13:42:19 -0800746 if (references == PAGEREF_RECLAIM_CLEAN)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747 goto keep_locked;
Andrew Morton4dd4b922008-03-24 12:29:52 -0700748 if (!may_enter_fs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749 goto keep_locked;
Christoph Lameter52a83632006-02-01 03:05:28 -0800750 if (!sc->may_writepage)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751 goto keep_locked;
752
753 /* Page is dirty, try to write it out here */
Andy Whitcroftc661b072007-08-22 14:01:26 -0700754 switch (pageout(page, mapping, sync_writeback)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 case PAGE_KEEP:
756 goto keep_locked;
757 case PAGE_ACTIVATE:
758 goto activate_locked;
759 case PAGE_SUCCESS:
Andrew Morton4dd4b922008-03-24 12:29:52 -0700760 if (PageWriteback(page) || PageDirty(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761 goto keep;
762 /*
763 * A synchronous write - probably a ramdisk. Go
764 * ahead and try to reclaim the page.
765 */
Nick Piggin529ae9a2008-08-02 12:01:03 +0200766 if (!trylock_page(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 goto keep;
768 if (PageDirty(page) || PageWriteback(page))
769 goto keep_locked;
770 mapping = page_mapping(page);
771 case PAGE_CLEAN:
772 ; /* try to free the page below */
773 }
774 }
775
776 /*
777 * If the page has buffers, try to free the buffer mappings
778 * associated with this page. If we succeed we try to free
779 * the page as well.
780 *
781 * We do this even if the page is PageDirty().
782 * try_to_release_page() does not perform I/O, but it is
783 * possible for a page to have PageDirty set, but it is actually
784 * clean (all its buffers are clean). This happens if the
785 * buffers were written out directly, with submit_bh(). ext3
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700786 * will do this, as well as the blockdev mapping.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 * try_to_release_page() will discover that cleanness and will
788 * drop the buffers and mark the page clean - it can be freed.
789 *
790 * Rarely, pages can have buffers and no ->mapping. These are
791 * the pages which were not successfully invalidated in
792 * truncate_complete_page(). We try to drop those buffers here
793 * and if that worked, and the page is no longer mapped into
794 * process address space (page_count == 1) it can be freed.
795 * Otherwise, leave the page on the LRU so it is swappable.
796 */
David Howells266cf652009-04-03 16:42:36 +0100797 if (page_has_private(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798 if (!try_to_release_page(page, sc->gfp_mask))
799 goto activate_locked;
Nick Piggine2867812008-07-25 19:45:30 -0700800 if (!mapping && page_count(page) == 1) {
801 unlock_page(page);
802 if (put_page_testzero(page))
803 goto free_it;
804 else {
805 /*
806 * rare race with speculative reference.
807 * the speculative reference will free
808 * this page shortly, so we may
809 * increment nr_reclaimed here (and
810 * leave it off the LRU).
811 */
812 nr_reclaimed++;
813 continue;
814 }
815 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816 }
817
Nick Piggine2867812008-07-25 19:45:30 -0700818 if (!mapping || !__remove_mapping(mapping, page))
Christoph Lameter49d2e9c2006-01-08 01:00:48 -0800819 goto keep_locked;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820
Nick Piggina978d6f2008-10-18 20:26:58 -0700821 /*
822 * At this point, we have no other references and there is
823 * no way to pick any more up (removed from LRU, removed
824 * from pagecache). Can use non-atomic bitops now (and
825 * we obviously don't have to worry about waking up a process
826 * waiting on the page lock, because there are no references.
827 */
828 __clear_page_locked(page);
Nick Piggine2867812008-07-25 19:45:30 -0700829free_it:
Andrew Morton05ff5132006-03-22 00:08:20 -0800830 nr_reclaimed++;
Mel Gormanabe4c3b2010-08-09 17:19:31 -0700831
832 /*
833 * Is there need to periodically free_page_list? It would
834 * appear not as the counts should be low
835 */
836 list_add(&page->lru, &free_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 continue;
838
Nick Pigginb291f002008-10-18 20:26:44 -0700839cull_mlocked:
Hugh Dickins63d6c5a2009-01-06 14:39:38 -0800840 if (PageSwapCache(page))
841 try_to_free_swap(page);
Nick Pigginb291f002008-10-18 20:26:44 -0700842 unlock_page(page);
843 putback_lru_page(page);
844 continue;
845
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846activate_locked:
Rik van Riel68a223942008-10-18 20:26:23 -0700847 /* Not a candidate for swapping, so reclaim swap space. */
848 if (PageSwapCache(page) && vm_swap_full())
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800849 try_to_free_swap(page);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700850 VM_BUG_ON(PageActive(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851 SetPageActive(page);
852 pgactivate++;
853keep_locked:
854 unlock_page(page);
855keep:
856 list_add(&page->lru, &ret_pages);
Nick Pigginb291f002008-10-18 20:26:44 -0700857 VM_BUG_ON(PageLRU(page) || PageUnevictable(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 }
Mel Gormanabe4c3b2010-08-09 17:19:31 -0700859
860 free_page_list(&free_pages);
861
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862 list_splice(&ret_pages, page_list);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700863 count_vm_events(PGACTIVATE, pgactivate);
Andrew Morton05ff5132006-03-22 00:08:20 -0800864 return nr_reclaimed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865}
866
Andy Whitcroft5ad333e2007-07-17 04:03:16 -0700867/*
868 * Attempt to remove the specified page from its LRU. Only take this page
869 * if it is of the appropriate PageActive status. Pages which are being
870 * freed elsewhere are also ignored.
871 *
872 * page: page to consider
873 * mode: one of the LRU isolation modes defined above
874 *
875 * returns 0 on success, -ve errno on failure.
876 */
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700877int __isolate_lru_page(struct page *page, int mode, int file)
Andy Whitcroft5ad333e2007-07-17 04:03:16 -0700878{
879 int ret = -EINVAL;
880
881 /* Only take pages on the LRU. */
882 if (!PageLRU(page))
883 return ret;
884
885 /*
886 * When checking the active state, we need to be sure we are
887 * dealing with comparible boolean values. Take the logical not
888 * of each.
889 */
890 if (mode != ISOLATE_BOTH && (!PageActive(page) != !mode))
891 return ret;
892
Johannes Weiner6c0b1352009-09-21 17:02:59 -0700893 if (mode != ISOLATE_BOTH && page_is_file_cache(page) != file)
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700894 return ret;
895
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700896 /*
897 * When this function is being called for lumpy reclaim, we
898 * initially look into all LRU pages, active, inactive and
899 * unevictable; only give shrink_page_list evictable pages.
900 */
901 if (PageUnevictable(page))
902 return ret;
903
Andy Whitcroft5ad333e2007-07-17 04:03:16 -0700904 ret = -EBUSY;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800905
Andy Whitcroft5ad333e2007-07-17 04:03:16 -0700906 if (likely(get_page_unless_zero(page))) {
907 /*
908 * Be careful not to clear PageLRU until after we're
909 * sure the page is not being freed elsewhere -- the
910 * page release code relies on it.
911 */
912 ClearPageLRU(page);
913 ret = 0;
914 }
915
916 return ret;
917}
918
Christoph Lameter49d2e9c2006-01-08 01:00:48 -0800919/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920 * zone->lru_lock is heavily contended. Some of the functions that
921 * shrink the lists perform better by taking out a batch of pages
922 * and working on them outside the LRU lock.
923 *
924 * For pagecache intensive workloads, this function is the hottest
925 * spot in the kernel (apart from copy_*_user functions).
926 *
927 * Appropriate locks must be held before calling this function.
928 *
929 * @nr_to_scan: The number of pages to look through on the list.
930 * @src: The LRU list to pull pages off.
931 * @dst: The temp list to put pages on to.
932 * @scanned: The number of pages that were scanned.
Andy Whitcroft5ad333e2007-07-17 04:03:16 -0700933 * @order: The caller's attempted allocation order
934 * @mode: One of the LRU isolation modes
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700935 * @file: True [1] if isolating file [!anon] pages
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 *
937 * returns how many pages were moved onto *@dst.
938 */
Andrew Morton69e05942006-03-22 00:08:19 -0800939static unsigned long isolate_lru_pages(unsigned long nr_to_scan,
940 struct list_head *src, struct list_head *dst,
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700941 unsigned long *scanned, int order, int mode, int file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942{
Andrew Morton69e05942006-03-22 00:08:19 -0800943 unsigned long nr_taken = 0;
Mel Gormana8a94d12010-08-09 17:19:17 -0700944 unsigned long nr_lumpy_taken = 0;
945 unsigned long nr_lumpy_dirty = 0;
946 unsigned long nr_lumpy_failed = 0;
Wu Fengguangc9b02d92006-03-22 00:08:23 -0800947 unsigned long scan;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948
Wu Fengguangc9b02d92006-03-22 00:08:23 -0800949 for (scan = 0; scan < nr_to_scan && !list_empty(src); scan++) {
Andy Whitcroft5ad333e2007-07-17 04:03:16 -0700950 struct page *page;
951 unsigned long pfn;
952 unsigned long end_pfn;
953 unsigned long page_pfn;
954 int zone_id;
955
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956 page = lru_to_page(src);
957 prefetchw_prev_lru_page(page, src, flags);
958
Nick Piggin725d7042006-09-25 23:30:55 -0700959 VM_BUG_ON(!PageLRU(page));
Nick Piggin8d438f92006-03-22 00:07:59 -0800960
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700961 switch (__isolate_lru_page(page, mode, file)) {
Andy Whitcroft5ad333e2007-07-17 04:03:16 -0700962 case 0:
963 list_move(&page->lru, dst);
KAMEZAWA Hiroyuki2ffebca2009-06-17 16:27:21 -0700964 mem_cgroup_del_lru(page);
Nick Piggin7c8ee9a2006-03-22 00:08:03 -0800965 nr_taken++;
Andy Whitcroft5ad333e2007-07-17 04:03:16 -0700966 break;
Nick Piggin46453a62006-03-22 00:07:58 -0800967
Andy Whitcroft5ad333e2007-07-17 04:03:16 -0700968 case -EBUSY:
969 /* else it is being freed elsewhere */
970 list_move(&page->lru, src);
KAMEZAWA Hiroyuki2ffebca2009-06-17 16:27:21 -0700971 mem_cgroup_rotate_lru_list(page, page_lru(page));
Andy Whitcroft5ad333e2007-07-17 04:03:16 -0700972 continue;
973
974 default:
975 BUG();
976 }
977
978 if (!order)
979 continue;
980
981 /*
982 * Attempt to take all pages in the order aligned region
983 * surrounding the tag page. Only take those pages of
984 * the same active state as that tag page. We may safely
985 * round the target page pfn down to the requested order
986 * as the mem_map is guarenteed valid out to MAX_ORDER,
987 * where that page is in a different zone we will detect
988 * it from its zone id and abort this block scan.
989 */
990 zone_id = page_zone_id(page);
991 page_pfn = page_to_pfn(page);
992 pfn = page_pfn & ~((1 << order) - 1);
993 end_pfn = pfn + (1 << order);
994 for (; pfn < end_pfn; pfn++) {
995 struct page *cursor_page;
996
997 /* The target page is in the block, ignore it. */
998 if (unlikely(pfn == page_pfn))
999 continue;
1000
1001 /* Avoid holes within the zone. */
1002 if (unlikely(!pfn_valid_within(pfn)))
1003 break;
1004
1005 cursor_page = pfn_to_page(pfn);
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001006
Andy Whitcroft5ad333e2007-07-17 04:03:16 -07001007 /* Check that we have not crossed a zone boundary. */
1008 if (unlikely(page_zone_id(cursor_page) != zone_id))
1009 continue;
Minchan Kimde2e7562009-09-21 17:01:43 -07001010
1011 /*
1012 * If we don't have enough swap space, reclaiming of
1013 * anon page which don't already have a swap slot is
1014 * pointless.
1015 */
1016 if (nr_swap_pages <= 0 && PageAnon(cursor_page) &&
1017 !PageSwapCache(cursor_page))
1018 continue;
1019
KAMEZAWA Hiroyukiee993b12009-06-16 15:33:24 -07001020 if (__isolate_lru_page(cursor_page, mode, file) == 0) {
Andy Whitcroft5ad333e2007-07-17 04:03:16 -07001021 list_move(&cursor_page->lru, dst);
KAMEZAWA Hiroyukicb4cbcf2009-06-23 08:57:55 +09001022 mem_cgroup_del_lru(cursor_page);
Andy Whitcroft5ad333e2007-07-17 04:03:16 -07001023 nr_taken++;
Mel Gormana8a94d12010-08-09 17:19:17 -07001024 nr_lumpy_taken++;
1025 if (PageDirty(cursor_page))
1026 nr_lumpy_dirty++;
Andy Whitcroft5ad333e2007-07-17 04:03:16 -07001027 scan++;
Mel Gormana8a94d12010-08-09 17:19:17 -07001028 } else {
1029 if (mode == ISOLATE_BOTH &&
1030 page_count(cursor_page))
1031 nr_lumpy_failed++;
Andy Whitcroft5ad333e2007-07-17 04:03:16 -07001032 }
1033 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034 }
1035
1036 *scanned = scan;
Mel Gormana8a94d12010-08-09 17:19:17 -07001037
1038 trace_mm_vmscan_lru_isolate(order,
1039 nr_to_scan, scan,
1040 nr_taken,
1041 nr_lumpy_taken, nr_lumpy_dirty, nr_lumpy_failed,
1042 mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043 return nr_taken;
1044}
1045
Balbir Singh66e17072008-02-07 00:13:56 -08001046static unsigned long isolate_pages_global(unsigned long nr,
1047 struct list_head *dst,
1048 unsigned long *scanned, int order,
1049 int mode, struct zone *z,
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001050 int active, int file)
Balbir Singh66e17072008-02-07 00:13:56 -08001051{
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001052 int lru = LRU_BASE;
Balbir Singh66e17072008-02-07 00:13:56 -08001053 if (active)
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001054 lru += LRU_ACTIVE;
1055 if (file)
1056 lru += LRU_FILE;
1057 return isolate_lru_pages(nr, &z->lru[lru].list, dst, scanned, order,
Johannes Weinerb7c46d12009-09-21 17:02:56 -07001058 mode, file);
Balbir Singh66e17072008-02-07 00:13:56 -08001059}
1060
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061/*
Andy Whitcroft5ad333e2007-07-17 04:03:16 -07001062 * clear_active_flags() is a helper for shrink_active_list(), clearing
1063 * any active bits from the pages in the list.
1064 */
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001065static unsigned long clear_active_flags(struct list_head *page_list,
1066 unsigned int *count)
Andy Whitcroft5ad333e2007-07-17 04:03:16 -07001067{
1068 int nr_active = 0;
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001069 int lru;
Andy Whitcroft5ad333e2007-07-17 04:03:16 -07001070 struct page *page;
1071
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001072 list_for_each_entry(page, page_list, lru) {
Johannes Weiner401a8e12009-09-21 17:02:58 -07001073 lru = page_lru_base_type(page);
Andy Whitcroft5ad333e2007-07-17 04:03:16 -07001074 if (PageActive(page)) {
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001075 lru += LRU_ACTIVE;
Andy Whitcroft5ad333e2007-07-17 04:03:16 -07001076 ClearPageActive(page);
1077 nr_active++;
1078 }
Mel Gorman1489fa12010-08-09 17:19:33 -07001079 if (count)
1080 count[lru]++;
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001081 }
Andy Whitcroft5ad333e2007-07-17 04:03:16 -07001082
1083 return nr_active;
1084}
1085
Nick Piggin62695a82008-10-18 20:26:09 -07001086/**
1087 * isolate_lru_page - tries to isolate a page from its LRU list
1088 * @page: page to isolate from its LRU list
1089 *
1090 * Isolates a @page from an LRU list, clears PageLRU and adjusts the
1091 * vmstat statistic corresponding to whatever LRU list the page was on.
1092 *
1093 * Returns 0 if the page was removed from an LRU list.
1094 * Returns -EBUSY if the page was not on an LRU list.
1095 *
1096 * The returned page will have PageLRU() cleared. If it was found on
Lee Schermerhorn894bc312008-10-18 20:26:39 -07001097 * the active list, it will have PageActive set. If it was found on
1098 * the unevictable list, it will have the PageUnevictable bit set. That flag
1099 * may need to be cleared by the caller before letting the page go.
Nick Piggin62695a82008-10-18 20:26:09 -07001100 *
1101 * The vmstat statistic corresponding to the list on which the page was
1102 * found will be decremented.
1103 *
1104 * Restrictions:
1105 * (1) Must be called with an elevated refcount on the page. This is a
1106 * fundamentnal difference from isolate_lru_pages (which is called
1107 * without a stable reference).
1108 * (2) the lru_lock must not be held.
1109 * (3) interrupts must be enabled.
1110 */
1111int isolate_lru_page(struct page *page)
1112{
1113 int ret = -EBUSY;
1114
1115 if (PageLRU(page)) {
1116 struct zone *zone = page_zone(page);
1117
1118 spin_lock_irq(&zone->lru_lock);
1119 if (PageLRU(page) && get_page_unless_zero(page)) {
Lee Schermerhorn894bc312008-10-18 20:26:39 -07001120 int lru = page_lru(page);
Nick Piggin62695a82008-10-18 20:26:09 -07001121 ret = 0;
1122 ClearPageLRU(page);
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001123
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001124 del_page_from_lru_list(zone, page, lru);
Nick Piggin62695a82008-10-18 20:26:09 -07001125 }
1126 spin_unlock_irq(&zone->lru_lock);
1127 }
1128 return ret;
1129}
1130
Andy Whitcroft5ad333e2007-07-17 04:03:16 -07001131/*
Rik van Riel35cd7812009-09-21 17:01:38 -07001132 * Are there way too many processes in the direct reclaim path already?
1133 */
1134static int too_many_isolated(struct zone *zone, int file,
1135 struct scan_control *sc)
1136{
1137 unsigned long inactive, isolated;
1138
1139 if (current_is_kswapd())
1140 return 0;
1141
1142 if (!scanning_global_lru(sc))
1143 return 0;
1144
1145 if (file) {
1146 inactive = zone_page_state(zone, NR_INACTIVE_FILE);
1147 isolated = zone_page_state(zone, NR_ISOLATED_FILE);
1148 } else {
1149 inactive = zone_page_state(zone, NR_INACTIVE_ANON);
1150 isolated = zone_page_state(zone, NR_ISOLATED_ANON);
1151 }
1152
1153 return isolated > inactive;
1154}
1155
1156/*
Mel Gorman66635622010-08-09 17:19:30 -07001157 * TODO: Try merging with migrations version of putback_lru_pages
1158 */
1159static noinline_for_stack void
Mel Gorman1489fa12010-08-09 17:19:33 -07001160putback_lru_pages(struct zone *zone, struct scan_control *sc,
Mel Gorman66635622010-08-09 17:19:30 -07001161 unsigned long nr_anon, unsigned long nr_file,
1162 struct list_head *page_list)
1163{
1164 struct page *page;
1165 struct pagevec pvec;
Mel Gorman1489fa12010-08-09 17:19:33 -07001166 struct zone_reclaim_stat *reclaim_stat = get_reclaim_stat(zone, sc);
Mel Gorman66635622010-08-09 17:19:30 -07001167
1168 pagevec_init(&pvec, 1);
1169
1170 /*
1171 * Put back any unfreeable pages.
1172 */
1173 spin_lock(&zone->lru_lock);
1174 while (!list_empty(page_list)) {
1175 int lru;
1176 page = lru_to_page(page_list);
1177 VM_BUG_ON(PageLRU(page));
1178 list_del(&page->lru);
1179 if (unlikely(!page_evictable(page, NULL))) {
1180 spin_unlock_irq(&zone->lru_lock);
1181 putback_lru_page(page);
1182 spin_lock_irq(&zone->lru_lock);
1183 continue;
1184 }
1185 SetPageLRU(page);
1186 lru = page_lru(page);
1187 add_page_to_lru_list(zone, page, lru);
1188 if (is_active_lru(lru)) {
1189 int file = is_file_lru(lru);
1190 reclaim_stat->recent_rotated[file]++;
1191 }
1192 if (!pagevec_add(&pvec, page)) {
1193 spin_unlock_irq(&zone->lru_lock);
1194 __pagevec_release(&pvec);
1195 spin_lock_irq(&zone->lru_lock);
1196 }
1197 }
1198 __mod_zone_page_state(zone, NR_ISOLATED_ANON, -nr_anon);
1199 __mod_zone_page_state(zone, NR_ISOLATED_FILE, -nr_file);
1200
1201 spin_unlock_irq(&zone->lru_lock);
1202 pagevec_release(&pvec);
1203}
1204
Mel Gorman1489fa12010-08-09 17:19:33 -07001205static noinline_for_stack void update_isolated_counts(struct zone *zone,
1206 struct scan_control *sc,
1207 unsigned long *nr_anon,
1208 unsigned long *nr_file,
1209 struct list_head *isolated_list)
1210{
1211 unsigned long nr_active;
1212 unsigned int count[NR_LRU_LISTS] = { 0, };
1213 struct zone_reclaim_stat *reclaim_stat = get_reclaim_stat(zone, sc);
1214
1215 nr_active = clear_active_flags(isolated_list, count);
1216 __count_vm_events(PGDEACTIVATE, nr_active);
1217
1218 __mod_zone_page_state(zone, NR_ACTIVE_FILE,
1219 -count[LRU_ACTIVE_FILE]);
1220 __mod_zone_page_state(zone, NR_INACTIVE_FILE,
1221 -count[LRU_INACTIVE_FILE]);
1222 __mod_zone_page_state(zone, NR_ACTIVE_ANON,
1223 -count[LRU_ACTIVE_ANON]);
1224 __mod_zone_page_state(zone, NR_INACTIVE_ANON,
1225 -count[LRU_INACTIVE_ANON]);
1226
1227 *nr_anon = count[LRU_ACTIVE_ANON] + count[LRU_INACTIVE_ANON];
1228 *nr_file = count[LRU_ACTIVE_FILE] + count[LRU_INACTIVE_FILE];
1229 __mod_zone_page_state(zone, NR_ISOLATED_ANON, *nr_anon);
1230 __mod_zone_page_state(zone, NR_ISOLATED_FILE, *nr_file);
1231
1232 reclaim_stat->recent_scanned[0] += *nr_anon;
1233 reclaim_stat->recent_scanned[1] += *nr_file;
1234}
1235
Mel Gorman66635622010-08-09 17:19:30 -07001236/*
Wu Fengguange31f36982010-08-09 17:20:01 -07001237 * Returns true if the caller should wait to clean dirty/writeback pages.
1238 *
1239 * If we are direct reclaiming for contiguous pages and we do not reclaim
1240 * everything in the list, try again and wait for writeback IO to complete.
1241 * This will stall high-order allocations noticeably. Only do that when really
1242 * need to free the pages under high memory pressure.
1243 */
1244static inline bool should_reclaim_stall(unsigned long nr_taken,
1245 unsigned long nr_freed,
1246 int priority,
1247 struct scan_control *sc)
1248{
1249 int lumpy_stall_priority;
1250
1251 /* kswapd should not stall on sync IO */
1252 if (current_is_kswapd())
1253 return false;
1254
1255 /* Only stall on lumpy reclaim */
1256 if (!sc->lumpy_reclaim_mode)
1257 return false;
1258
1259 /* If we have relaimed everything on the isolated list, no stall */
1260 if (nr_freed == nr_taken)
1261 return false;
1262
1263 /*
1264 * For high-order allocations, there are two stall thresholds.
1265 * High-cost allocations stall immediately where as lower
1266 * order allocations such as stacks require the scanning
1267 * priority to be much higher before stalling.
1268 */
1269 if (sc->order > PAGE_ALLOC_COSTLY_ORDER)
1270 lumpy_stall_priority = DEF_PRIORITY;
1271 else
1272 lumpy_stall_priority = DEF_PRIORITY / 3;
1273
1274 return priority <= lumpy_stall_priority;
1275}
1276
1277/*
Andrew Morton1742f192006-03-22 00:08:21 -08001278 * shrink_inactive_list() is a helper for shrink_zone(). It returns the number
1279 * of reclaimed pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280 */
Mel Gorman66635622010-08-09 17:19:30 -07001281static noinline_for_stack unsigned long
1282shrink_inactive_list(unsigned long nr_to_scan, struct zone *zone,
1283 struct scan_control *sc, int priority, int file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284{
1285 LIST_HEAD(page_list);
KOSAKI Motohiroe247dbc2010-08-09 17:19:28 -07001286 unsigned long nr_scanned;
Andrew Morton05ff5132006-03-22 00:08:20 -08001287 unsigned long nr_reclaimed = 0;
KOSAKI Motohiroe247dbc2010-08-09 17:19:28 -07001288 unsigned long nr_taken;
1289 unsigned long nr_active;
KOSAKI Motohiroe247dbc2010-08-09 17:19:28 -07001290 unsigned long nr_anon;
1291 unsigned long nr_file;
KOSAKI Motohiro78dc5832009-06-16 15:31:40 -07001292
Rik van Riel35cd7812009-09-21 17:01:38 -07001293 while (unlikely(too_many_isolated(zone, file, sc))) {
KOSAKI Motohiro58355c72009-10-26 16:49:35 -07001294 congestion_wait(BLK_RW_ASYNC, HZ/10);
Rik van Riel35cd7812009-09-21 17:01:38 -07001295
1296 /* We are about to die and free our memory. Return now. */
1297 if (fatal_signal_pending(current))
1298 return SWAP_CLUSTER_MAX;
1299 }
1300
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302 lru_add_drain();
1303 spin_lock_irq(&zone->lru_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304
KOSAKI Motohiroe247dbc2010-08-09 17:19:28 -07001305 if (scanning_global_lru(sc)) {
1306 nr_taken = isolate_pages_global(nr_to_scan,
1307 &page_list, &nr_scanned, sc->order,
1308 sc->lumpy_reclaim_mode ?
1309 ISOLATE_BOTH : ISOLATE_INACTIVE,
1310 zone, 0, file);
1311 zone->pages_scanned += nr_scanned;
KOSAKI Motohirob35ea172009-09-21 17:01:36 -07001312 if (current_is_kswapd())
KOSAKI Motohiroe247dbc2010-08-09 17:19:28 -07001313 __count_zone_vm_events(PGSCAN_KSWAPD, zone,
1314 nr_scanned);
1315 else
1316 __count_zone_vm_events(PGSCAN_DIRECT, zone,
1317 nr_scanned);
1318 } else {
1319 nr_taken = mem_cgroup_isolate_pages(nr_to_scan,
1320 &page_list, &nr_scanned, sc->order,
1321 sc->lumpy_reclaim_mode ?
1322 ISOLATE_BOTH : ISOLATE_INACTIVE,
1323 zone, sc->mem_cgroup,
1324 0, file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325 /*
KOSAKI Motohiroe247dbc2010-08-09 17:19:28 -07001326 * mem_cgroup_isolate_pages() keeps track of
1327 * scanned pages on its own.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328 */
KOSAKI Motohiroe247dbc2010-08-09 17:19:28 -07001329 }
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07001330
Mel Gorman66635622010-08-09 17:19:30 -07001331 if (nr_taken == 0) {
1332 spin_unlock_irq(&zone->lru_lock);
1333 return 0;
1334 }
KOSAKI Motohiroe247dbc2010-08-09 17:19:28 -07001335
Mel Gorman1489fa12010-08-09 17:19:33 -07001336 update_isolated_counts(zone, sc, &nr_anon, &nr_file, &page_list);
KOSAKI Motohiroe247dbc2010-08-09 17:19:28 -07001337
1338 spin_unlock_irq(&zone->lru_lock);
1339
1340 nr_reclaimed = shrink_page_list(&page_list, sc, PAGEOUT_IO_ASYNC);
1341
Wu Fengguange31f36982010-08-09 17:20:01 -07001342 /* Check if we should syncronously wait for writeback */
1343 if (should_reclaim_stall(nr_taken, nr_reclaimed, priority, sc)) {
KOSAKI Motohiroe247dbc2010-08-09 17:19:28 -07001344 congestion_wait(BLK_RW_ASYNC, HZ/10);
1345
1346 /*
1347 * The attempt at page out may have made some
1348 * of the pages active, mark them inactive again.
1349 */
Mel Gorman1489fa12010-08-09 17:19:33 -07001350 nr_active = clear_active_flags(&page_list, NULL);
KOSAKI Motohiroe247dbc2010-08-09 17:19:28 -07001351 count_vm_events(PGDEACTIVATE, nr_active);
1352
1353 nr_reclaimed += shrink_page_list(&page_list, sc, PAGEOUT_IO_SYNC);
1354 }
1355
1356 local_irq_disable();
1357 if (current_is_kswapd())
1358 __count_vm_events(KSWAPD_STEAL, nr_reclaimed);
1359 __count_zone_vm_events(PGSTEAL, zone, nr_reclaimed);
1360
Mel Gorman1489fa12010-08-09 17:19:33 -07001361 putback_lru_pages(zone, sc, nr_anon, nr_file, &page_list);
Andrew Morton05ff5132006-03-22 00:08:20 -08001362 return nr_reclaimed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363}
1364
Martin Bligh3bb1a8522006-10-28 10:38:24 -07001365/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366 * This moves pages from the active list to the inactive list.
1367 *
1368 * We move them the other way if the page is referenced by one or more
1369 * processes, from rmap.
1370 *
1371 * If the pages are mostly unmapped, the processing is fast and it is
1372 * appropriate to hold zone->lru_lock across the whole operation. But if
1373 * the pages are mapped, the processing is slow (page_referenced()) so we
1374 * should drop zone->lru_lock around each page. It's impossible to balance
1375 * this, so instead we remove the pages from the LRU while processing them.
1376 * It is safe to rely on PG_active against the non-LRU pages in here because
1377 * nobody will play with that bit on a non-LRU page.
1378 *
1379 * The downside is that we have to touch page->_count against each page.
1380 * But we had to alter page->flags anyway.
1381 */
KAMEZAWA Hiroyuki1cfb4192008-02-07 00:14:37 -08001382
Wu Fengguang3eb41402009-06-16 15:33:13 -07001383static void move_active_pages_to_lru(struct zone *zone,
1384 struct list_head *list,
1385 enum lru_list lru)
1386{
1387 unsigned long pgmoved = 0;
1388 struct pagevec pvec;
1389 struct page *page;
1390
1391 pagevec_init(&pvec, 1);
1392
1393 while (!list_empty(list)) {
1394 page = lru_to_page(list);
Wu Fengguang3eb41402009-06-16 15:33:13 -07001395
1396 VM_BUG_ON(PageLRU(page));
1397 SetPageLRU(page);
1398
Wu Fengguang3eb41402009-06-16 15:33:13 -07001399 list_move(&page->lru, &zone->lru[lru].list);
1400 mem_cgroup_add_lru_list(page, lru);
1401 pgmoved++;
1402
1403 if (!pagevec_add(&pvec, page) || list_empty(list)) {
1404 spin_unlock_irq(&zone->lru_lock);
1405 if (buffer_heads_over_limit)
1406 pagevec_strip(&pvec);
1407 __pagevec_release(&pvec);
1408 spin_lock_irq(&zone->lru_lock);
1409 }
1410 }
1411 __mod_zone_page_state(zone, NR_LRU_BASE + lru, pgmoved);
1412 if (!is_active_lru(lru))
1413 __count_vm_events(PGDEACTIVATE, pgmoved);
1414}
KAMEZAWA Hiroyuki1cfb4192008-02-07 00:14:37 -08001415
Andrew Morton1742f192006-03-22 00:08:21 -08001416static void shrink_active_list(unsigned long nr_pages, struct zone *zone,
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001417 struct scan_control *sc, int priority, int file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418{
KOSAKI Motohiro44c241f2009-09-21 17:01:35 -07001419 unsigned long nr_taken;
Andrew Morton69e05942006-03-22 00:08:19 -08001420 unsigned long pgscanned;
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -07001421 unsigned long vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422 LIST_HEAD(l_hold); /* The pages which were snipped off */
Wu Fengguang8cab4752009-06-16 15:33:12 -07001423 LIST_HEAD(l_active);
Christoph Lameterb69408e2008-10-18 20:26:14 -07001424 LIST_HEAD(l_inactive);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425 struct page *page;
KOSAKI Motohiro6e901572009-01-07 18:08:15 -08001426 struct zone_reclaim_stat *reclaim_stat = get_reclaim_stat(zone, sc);
KOSAKI Motohiro44c241f2009-09-21 17:01:35 -07001427 unsigned long nr_rotated = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428
1429 lru_add_drain();
1430 spin_lock_irq(&zone->lru_lock);
KAMEZAWA Hiroyukie72e2bd2009-01-07 18:08:23 -08001431 if (scanning_global_lru(sc)) {
Johannes Weiner8b25c6d2010-05-24 14:32:40 -07001432 nr_taken = isolate_pages_global(nr_pages, &l_hold,
1433 &pgscanned, sc->order,
1434 ISOLATE_ACTIVE, zone,
1435 1, file);
KAMEZAWA Hiroyuki1cfb4192008-02-07 00:14:37 -08001436 zone->pages_scanned += pgscanned;
Johannes Weiner8b25c6d2010-05-24 14:32:40 -07001437 } else {
1438 nr_taken = mem_cgroup_isolate_pages(nr_pages, &l_hold,
1439 &pgscanned, sc->order,
1440 ISOLATE_ACTIVE, zone,
1441 sc->mem_cgroup, 1, file);
1442 /*
1443 * mem_cgroup_isolate_pages() keeps track of
1444 * scanned pages on its own.
1445 */
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001446 }
Johannes Weiner8b25c6d2010-05-24 14:32:40 -07001447
Johannes Weinerb7c46d12009-09-21 17:02:56 -07001448 reclaim_stat->recent_scanned[file] += nr_taken;
KAMEZAWA Hiroyuki1cfb4192008-02-07 00:14:37 -08001449
Wu Fengguang3eb41402009-06-16 15:33:13 -07001450 __count_zone_vm_events(PGREFILL, zone, pgscanned);
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001451 if (file)
KOSAKI Motohiro44c241f2009-09-21 17:01:35 -07001452 __mod_zone_page_state(zone, NR_ACTIVE_FILE, -nr_taken);
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001453 else
KOSAKI Motohiro44c241f2009-09-21 17:01:35 -07001454 __mod_zone_page_state(zone, NR_ACTIVE_ANON, -nr_taken);
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07001455 __mod_zone_page_state(zone, NR_ISOLATED_ANON + file, nr_taken);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456 spin_unlock_irq(&zone->lru_lock);
1457
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 while (!list_empty(&l_hold)) {
1459 cond_resched();
1460 page = lru_to_page(&l_hold);
1461 list_del(&page->lru);
Rik van Riel7e9cd482008-10-18 20:26:35 -07001462
Lee Schermerhorn894bc312008-10-18 20:26:39 -07001463 if (unlikely(!page_evictable(page, NULL))) {
1464 putback_lru_page(page);
1465 continue;
1466 }
1467
Johannes Weiner645747462010-03-05 13:42:22 -08001468 if (page_referenced(page, 0, sc->mem_cgroup, &vm_flags)) {
KOSAKI Motohiro44c241f2009-09-21 17:01:35 -07001469 nr_rotated++;
Wu Fengguang8cab4752009-06-16 15:33:12 -07001470 /*
1471 * Identify referenced, file-backed active pages and
1472 * give them one more trip around the active list. So
1473 * that executable code get better chances to stay in
1474 * memory under moderate memory pressure. Anon pages
1475 * are not likely to be evicted by use-once streaming
1476 * IO, plus JVM can create lots of anon VM_EXEC pages,
1477 * so we ignore them here.
1478 */
Wu Fengguang41e20982009-10-26 16:49:53 -07001479 if ((vm_flags & VM_EXEC) && page_is_file_cache(page)) {
Wu Fengguang8cab4752009-06-16 15:33:12 -07001480 list_add(&page->lru, &l_active);
1481 continue;
1482 }
1483 }
Rik van Riel7e9cd482008-10-18 20:26:35 -07001484
KOSAKI Motohiro5205e562009-09-21 17:01:44 -07001485 ClearPageActive(page); /* we are de-activating */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 list_add(&page->lru, &l_inactive);
1487 }
1488
Andrew Mortonb5557492009-01-06 14:40:13 -08001489 /*
Wu Fengguang8cab4752009-06-16 15:33:12 -07001490 * Move pages back to the lru list.
Andrew Mortonb5557492009-01-06 14:40:13 -08001491 */
Johannes Weiner2a1dc502008-12-01 03:00:35 +01001492 spin_lock_irq(&zone->lru_lock);
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001493 /*
Wu Fengguang8cab4752009-06-16 15:33:12 -07001494 * Count referenced pages from currently used mappings as rotated,
1495 * even though only some of them are actually re-activated. This
1496 * helps balance scan pressure between file and anonymous pages in
1497 * get_scan_ratio.
Rik van Riel7e9cd482008-10-18 20:26:35 -07001498 */
Johannes Weinerb7c46d12009-09-21 17:02:56 -07001499 reclaim_stat->recent_rotated[file] += nr_rotated;
Rik van Riel556adec2008-10-18 20:26:34 -07001500
Wu Fengguang3eb41402009-06-16 15:33:13 -07001501 move_active_pages_to_lru(zone, &l_active,
1502 LRU_ACTIVE + file * LRU_FILE);
1503 move_active_pages_to_lru(zone, &l_inactive,
1504 LRU_BASE + file * LRU_FILE);
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07001505 __mod_zone_page_state(zone, NR_ISOLATED_ANON + file, -nr_taken);
Christoph Lameterf8891e52006-06-30 01:55:45 -07001506 spin_unlock_irq(&zone->lru_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507}
1508
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001509static int inactive_anon_is_low_global(struct zone *zone)
KOSAKI Motohirof89eb902009-01-07 18:08:14 -08001510{
1511 unsigned long active, inactive;
1512
1513 active = zone_page_state(zone, NR_ACTIVE_ANON);
1514 inactive = zone_page_state(zone, NR_INACTIVE_ANON);
1515
1516 if (inactive * zone->inactive_ratio < active)
1517 return 1;
1518
1519 return 0;
1520}
1521
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001522/**
1523 * inactive_anon_is_low - check if anonymous pages need to be deactivated
1524 * @zone: zone to check
1525 * @sc: scan control of this context
1526 *
1527 * Returns true if the zone does not have enough inactive anon pages,
1528 * meaning some active anon pages need to be deactivated.
1529 */
1530static int inactive_anon_is_low(struct zone *zone, struct scan_control *sc)
1531{
1532 int low;
1533
KAMEZAWA Hiroyukie72e2bd2009-01-07 18:08:23 -08001534 if (scanning_global_lru(sc))
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001535 low = inactive_anon_is_low_global(zone);
1536 else
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001537 low = mem_cgroup_inactive_anon_is_low(sc->mem_cgroup);
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001538 return low;
1539}
1540
Rik van Riel56e49d22009-06-16 15:32:28 -07001541static int inactive_file_is_low_global(struct zone *zone)
1542{
1543 unsigned long active, inactive;
1544
1545 active = zone_page_state(zone, NR_ACTIVE_FILE);
1546 inactive = zone_page_state(zone, NR_INACTIVE_FILE);
1547
1548 return (active > inactive);
1549}
1550
1551/**
1552 * inactive_file_is_low - check if file pages need to be deactivated
1553 * @zone: zone to check
1554 * @sc: scan control of this context
1555 *
1556 * When the system is doing streaming IO, memory pressure here
1557 * ensures that active file pages get deactivated, until more
1558 * than half of the file pages are on the inactive list.
1559 *
1560 * Once we get to that situation, protect the system's working
1561 * set from being evicted by disabling active file page aging.
1562 *
1563 * This uses a different ratio than the anonymous pages, because
1564 * the page cache uses a use-once replacement algorithm.
1565 */
1566static int inactive_file_is_low(struct zone *zone, struct scan_control *sc)
1567{
1568 int low;
1569
1570 if (scanning_global_lru(sc))
1571 low = inactive_file_is_low_global(zone);
1572 else
1573 low = mem_cgroup_inactive_file_is_low(sc->mem_cgroup);
1574 return low;
1575}
1576
Rik van Rielb39415b2009-12-14 17:59:48 -08001577static int inactive_list_is_low(struct zone *zone, struct scan_control *sc,
1578 int file)
1579{
1580 if (file)
1581 return inactive_file_is_low(zone, sc);
1582 else
1583 return inactive_anon_is_low(zone, sc);
1584}
1585
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001586static unsigned long shrink_list(enum lru_list lru, unsigned long nr_to_scan,
Christoph Lameterb69408e2008-10-18 20:26:14 -07001587 struct zone *zone, struct scan_control *sc, int priority)
1588{
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001589 int file = is_file_lru(lru);
1590
Rik van Rielb39415b2009-12-14 17:59:48 -08001591 if (is_active_lru(lru)) {
1592 if (inactive_list_is_low(zone, sc, file))
1593 shrink_active_list(nr_to_scan, zone, sc, priority, file);
Rik van Riel556adec2008-10-18 20:26:34 -07001594 return 0;
1595 }
1596
Rik van Riel33c120e2008-10-18 20:26:36 -07001597 return shrink_inactive_list(nr_to_scan, zone, sc, priority, file);
Christoph Lameterb69408e2008-10-18 20:26:14 -07001598}
1599
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600/*
Shaohua Li76a33fc2010-05-24 14:32:36 -07001601 * Smallish @nr_to_scan's are deposited in @nr_saved_scan,
1602 * until we collected @swap_cluster_max pages to scan.
1603 */
1604static unsigned long nr_scan_try_batch(unsigned long nr_to_scan,
1605 unsigned long *nr_saved_scan)
1606{
1607 unsigned long nr;
1608
1609 *nr_saved_scan += nr_to_scan;
1610 nr = *nr_saved_scan;
1611
1612 if (nr >= SWAP_CLUSTER_MAX)
1613 *nr_saved_scan = 0;
1614 else
1615 nr = 0;
1616
1617 return nr;
1618}
1619
1620/*
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001621 * Determine how aggressively the anon and file LRU lists should be
1622 * scanned. The relative value of each set of LRU lists is determined
1623 * by looking at the fraction of the pages scanned we did rotate back
1624 * onto the active list instead of evict.
1625 *
Shaohua Li76a33fc2010-05-24 14:32:36 -07001626 * nr[0] = anon pages to scan; nr[1] = file pages to scan
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001627 */
Shaohua Li76a33fc2010-05-24 14:32:36 -07001628static void get_scan_count(struct zone *zone, struct scan_control *sc,
1629 unsigned long *nr, int priority)
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001630{
1631 unsigned long anon, file, free;
1632 unsigned long anon_prio, file_prio;
1633 unsigned long ap, fp;
KOSAKI Motohiro6e901572009-01-07 18:08:15 -08001634 struct zone_reclaim_stat *reclaim_stat = get_reclaim_stat(zone, sc);
Shaohua Li76a33fc2010-05-24 14:32:36 -07001635 u64 fraction[2], denominator;
1636 enum lru_list l;
1637 int noswap = 0;
1638
1639 /* If we have no swap space, do not bother scanning anon pages. */
1640 if (!sc->may_swap || (nr_swap_pages <= 0)) {
1641 noswap = 1;
1642 fraction[0] = 0;
1643 fraction[1] = 1;
1644 denominator = 1;
1645 goto out;
1646 }
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001647
Vincent Li0b217672009-09-21 17:03:09 -07001648 anon = zone_nr_lru_pages(zone, sc, LRU_ACTIVE_ANON) +
1649 zone_nr_lru_pages(zone, sc, LRU_INACTIVE_ANON);
1650 file = zone_nr_lru_pages(zone, sc, LRU_ACTIVE_FILE) +
1651 zone_nr_lru_pages(zone, sc, LRU_INACTIVE_FILE);
Hugh Dickinsb9627162009-01-06 14:39:41 -08001652
KAMEZAWA Hiroyukie72e2bd2009-01-07 18:08:23 -08001653 if (scanning_global_lru(sc)) {
KOSAKI Motohiroeeee9a82009-01-07 18:08:17 -08001654 free = zone_page_state(zone, NR_FREE_PAGES);
1655 /* If we have very few page cache pages,
1656 force-scan anon pages. */
Mel Gorman41858962009-06-16 15:32:12 -07001657 if (unlikely(file + free <= high_wmark_pages(zone))) {
Shaohua Li76a33fc2010-05-24 14:32:36 -07001658 fraction[0] = 1;
1659 fraction[1] = 0;
1660 denominator = 1;
1661 goto out;
KOSAKI Motohiroeeee9a82009-01-07 18:08:17 -08001662 }
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001663 }
1664
1665 /*
KOSAKI Motohiro58c37f62010-08-09 17:19:51 -07001666 * With swappiness at 100, anonymous and file have the same priority.
1667 * This scanning priority is essentially the inverse of IO cost.
1668 */
1669 anon_prio = sc->swappiness;
1670 file_prio = 200 - sc->swappiness;
1671
1672 /*
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001673 * OK, so we have swap space and a fair amount of page cache
1674 * pages. We use the recently rotated / recently scanned
1675 * ratios to determine how valuable each cache is.
1676 *
1677 * Because workloads change over time (and to avoid overflow)
1678 * we keep these statistics as a floating average, which ends
1679 * up weighing recent references more than old ones.
1680 *
1681 * anon in [0], file in [1]
1682 */
KOSAKI Motohiro58c37f62010-08-09 17:19:51 -07001683 spin_lock_irq(&zone->lru_lock);
KOSAKI Motohiro6e901572009-01-07 18:08:15 -08001684 if (unlikely(reclaim_stat->recent_scanned[0] > anon / 4)) {
KOSAKI Motohiro6e901572009-01-07 18:08:15 -08001685 reclaim_stat->recent_scanned[0] /= 2;
1686 reclaim_stat->recent_rotated[0] /= 2;
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001687 }
1688
KOSAKI Motohiro6e901572009-01-07 18:08:15 -08001689 if (unlikely(reclaim_stat->recent_scanned[1] > file / 4)) {
KOSAKI Motohiro6e901572009-01-07 18:08:15 -08001690 reclaim_stat->recent_scanned[1] /= 2;
1691 reclaim_stat->recent_rotated[1] /= 2;
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001692 }
1693
1694 /*
Rik van Riel00d80892008-11-19 15:36:44 -08001695 * The amount of pressure on anon vs file pages is inversely
1696 * proportional to the fraction of recently scanned pages on
1697 * each list that were recently referenced and in active use.
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001698 */
KOSAKI Motohiro6e901572009-01-07 18:08:15 -08001699 ap = (anon_prio + 1) * (reclaim_stat->recent_scanned[0] + 1);
1700 ap /= reclaim_stat->recent_rotated[0] + 1;
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001701
KOSAKI Motohiro6e901572009-01-07 18:08:15 -08001702 fp = (file_prio + 1) * (reclaim_stat->recent_scanned[1] + 1);
1703 fp /= reclaim_stat->recent_rotated[1] + 1;
KOSAKI Motohiro58c37f62010-08-09 17:19:51 -07001704 spin_unlock_irq(&zone->lru_lock);
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001705
Shaohua Li76a33fc2010-05-24 14:32:36 -07001706 fraction[0] = ap;
1707 fraction[1] = fp;
1708 denominator = ap + fp + 1;
1709out:
1710 for_each_evictable_lru(l) {
1711 int file = is_file_lru(l);
1712 unsigned long scan;
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001713
Shaohua Li76a33fc2010-05-24 14:32:36 -07001714 scan = zone_nr_lru_pages(zone, sc, l);
1715 if (priority || noswap) {
1716 scan >>= priority;
1717 scan = div64_u64(scan * fraction[file], denominator);
1718 }
1719 nr[l] = nr_scan_try_batch(scan,
1720 &reclaim_stat->nr_saved_scan[l]);
1721 }
Wu Fengguang6e08a362009-06-16 15:32:29 -07001722}
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001723
KOSAKI Motohiro5f53e762010-05-24 14:32:37 -07001724static void set_lumpy_reclaim_mode(int priority, struct scan_control *sc)
1725{
1726 /*
1727 * If we need a large contiguous chunk of memory, or have
1728 * trouble getting a small set of contiguous pages, we
1729 * will reclaim both active and inactive pages.
1730 */
1731 if (sc->order > PAGE_ALLOC_COSTLY_ORDER)
1732 sc->lumpy_reclaim_mode = 1;
1733 else if (sc->order && priority < DEF_PRIORITY - 2)
1734 sc->lumpy_reclaim_mode = 1;
1735 else
1736 sc->lumpy_reclaim_mode = 0;
1737}
1738
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001739/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001740 * This is a basic per-zone page freer. Used by both kswapd and direct reclaim.
1741 */
Rik van Riela79311c2009-01-06 14:40:01 -08001742static void shrink_zone(int priority, struct zone *zone,
Andrew Morton05ff5132006-03-22 00:08:20 -08001743 struct scan_control *sc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744{
Christoph Lameterb69408e2008-10-18 20:26:14 -07001745 unsigned long nr[NR_LRU_LISTS];
Christoph Lameter86959492006-03-22 00:08:18 -08001746 unsigned long nr_to_scan;
Christoph Lameterb69408e2008-10-18 20:26:14 -07001747 enum lru_list l;
KOSAKI Motohiro01dbe5c2009-01-06 14:40:02 -08001748 unsigned long nr_reclaimed = sc->nr_reclaimed;
KOSAKI Motohiro22fba332009-12-14 17:59:10 -08001749 unsigned long nr_to_reclaim = sc->nr_to_reclaim;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750
Shaohua Li76a33fc2010-05-24 14:32:36 -07001751 get_scan_count(zone, sc, nr, priority);
KAMEZAWA Hiroyuki1cfb4192008-02-07 00:14:37 -08001752
KOSAKI Motohiro5f53e762010-05-24 14:32:37 -07001753 set_lumpy_reclaim_mode(priority, sc);
1754
Rik van Riel556adec2008-10-18 20:26:34 -07001755 while (nr[LRU_INACTIVE_ANON] || nr[LRU_ACTIVE_FILE] ||
1756 nr[LRU_INACTIVE_FILE]) {
Lee Schermerhorn894bc312008-10-18 20:26:39 -07001757 for_each_evictable_lru(l) {
Christoph Lameterb69408e2008-10-18 20:26:14 -07001758 if (nr[l]) {
KOSAKI Motohiroece74b22009-12-14 17:59:14 -08001759 nr_to_scan = min_t(unsigned long,
1760 nr[l], SWAP_CLUSTER_MAX);
Christoph Lameterb69408e2008-10-18 20:26:14 -07001761 nr[l] -= nr_to_scan;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762
KOSAKI Motohiro01dbe5c2009-01-06 14:40:02 -08001763 nr_reclaimed += shrink_list(l, nr_to_scan,
1764 zone, sc, priority);
Christoph Lameterb69408e2008-10-18 20:26:14 -07001765 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766 }
Rik van Riela79311c2009-01-06 14:40:01 -08001767 /*
1768 * On large memory systems, scan >> priority can become
1769 * really large. This is fine for the starting priority;
1770 * we want to put equal scanning pressure on each zone.
1771 * However, if the VM has a harder time of freeing pages,
1772 * with multiple processes reclaiming pages, the total
1773 * freeing target can get unreasonably large.
1774 */
KOSAKI Motohiro338fde92009-12-14 17:59:15 -08001775 if (nr_reclaimed >= nr_to_reclaim && priority < DEF_PRIORITY)
Rik van Riela79311c2009-01-06 14:40:01 -08001776 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001777 }
1778
KOSAKI Motohiro01dbe5c2009-01-06 14:40:02 -08001779 sc->nr_reclaimed = nr_reclaimed;
1780
Rik van Riel556adec2008-10-18 20:26:34 -07001781 /*
1782 * Even if we did not try to evict anon pages at all, we want to
1783 * rebalance the anon lru active/inactive ratio.
1784 */
MinChan Kim69c85482009-06-16 15:32:44 -07001785 if (inactive_anon_is_low(zone, sc) && nr_swap_pages > 0)
Rik van Riel556adec2008-10-18 20:26:34 -07001786 shrink_active_list(SWAP_CLUSTER_MAX, zone, sc, priority, 0);
1787
Andrew Morton232ea4d2007-02-28 20:13:21 -08001788 throttle_vm_writeout(sc->gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789}
1790
1791/*
1792 * This is the direct reclaim path, for page-allocating processes. We only
1793 * try to reclaim pages from zones which will satisfy the caller's allocation
1794 * request.
1795 *
Mel Gorman41858962009-06-16 15:32:12 -07001796 * We reclaim from a zone even if that zone is over high_wmark_pages(zone).
1797 * Because:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001798 * a) The caller may be trying to free *extra* pages to satisfy a higher-order
1799 * allocation or
Mel Gorman41858962009-06-16 15:32:12 -07001800 * b) The target zone may be at high_wmark_pages(zone) but the lower zones
1801 * must go *over* high_wmark_pages(zone) to satisfy the `incremental min'
1802 * zone defense algorithm.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804 * If a zone is deemed to be full of pinned pages then just give it a light
1805 * scan then give up on it.
1806 */
KOSAKI Motohirobb21c7c2010-06-04 14:15:05 -07001807static bool shrink_zones(int priority, struct zonelist *zonelist,
Andrew Morton05ff5132006-03-22 00:08:20 -08001808 struct scan_control *sc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001809{
Mel Gormandd1a2392008-04-28 02:12:17 -07001810 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001811 struct zone *zone;
KOSAKI Motohirobb21c7c2010-06-04 14:15:05 -07001812 bool all_unreclaimable = true;
KAMEZAWA Hiroyuki1cfb4192008-02-07 00:14:37 -08001813
Mel Gormand4debc62010-08-09 17:19:29 -07001814 for_each_zone_zonelist_nodemask(zone, z, zonelist,
1815 gfp_zone(sc->gfp_mask), sc->nodemask) {
Con Kolivasf3fe6512006-01-06 00:11:15 -08001816 if (!populated_zone(zone))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001817 continue;
KAMEZAWA Hiroyuki1cfb4192008-02-07 00:14:37 -08001818 /*
1819 * Take care memory controller reclaiming has small influence
1820 * to global LRU.
1821 */
KAMEZAWA Hiroyukie72e2bd2009-01-07 18:08:23 -08001822 if (scanning_global_lru(sc)) {
KAMEZAWA Hiroyuki1cfb4192008-02-07 00:14:37 -08001823 if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL))
1824 continue;
KOSAKI Motohiro93e4a892010-03-05 13:41:55 -08001825 if (zone->all_unreclaimable && priority != DEF_PRIORITY)
KAMEZAWA Hiroyuki1cfb4192008-02-07 00:14:37 -08001826 continue; /* Let kswapd poll it */
KAMEZAWA Hiroyuki1cfb4192008-02-07 00:14:37 -08001827 }
Nick Piggin408d8542006-09-25 23:31:27 -07001828
Rik van Riela79311c2009-01-06 14:40:01 -08001829 shrink_zone(priority, zone, sc);
KOSAKI Motohirobb21c7c2010-06-04 14:15:05 -07001830 all_unreclaimable = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001831 }
KOSAKI Motohirobb21c7c2010-06-04 14:15:05 -07001832 return all_unreclaimable;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001833}
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001834
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835/*
1836 * This is the main entry point to direct page reclaim.
1837 *
1838 * If a full scan of the inactive list fails to free enough memory then we
1839 * are "out of memory" and something needs to be killed.
1840 *
1841 * If the caller is !__GFP_FS then the probability of a failure is reasonably
1842 * high - the zone may be full of dirty or under-writeback pages, which this
Jens Axboe5b0830c2009-09-23 19:37:09 +02001843 * caller can't do much about. We kick the writeback threads and take explicit
1844 * naps in the hope that some of these pages can be written. But if the
1845 * allocating task holds filesystem locks which prevent writeout this might not
1846 * work, and the allocation attempt will fail.
Nishanth Aravamudana41f24e2008-04-29 00:58:25 -07001847 *
1848 * returns: 0, if no pages reclaimed
1849 * else, the number of pages reclaimed
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850 */
Mel Gormandac1d272008-04-28 02:12:12 -07001851static unsigned long do_try_to_free_pages(struct zonelist *zonelist,
Mel Gormandd1a2392008-04-28 02:12:17 -07001852 struct scan_control *sc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001853{
1854 int priority;
KOSAKI Motohirobb21c7c2010-06-04 14:15:05 -07001855 bool all_unreclaimable;
Andrew Morton69e05942006-03-22 00:08:19 -08001856 unsigned long total_scanned = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001857 struct reclaim_state *reclaim_state = current->reclaim_state;
Mel Gormandd1a2392008-04-28 02:12:17 -07001858 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001859 struct zone *zone;
KOSAKI Motohiro22fba332009-12-14 17:59:10 -08001860 unsigned long writeback_threshold;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861
Miao Xiec0ff7452010-05-24 14:32:08 -07001862 get_mems_allowed();
Keika Kobayashi873b4772008-07-25 01:48:52 -07001863 delayacct_freepages_start();
1864
KAMEZAWA Hiroyukie72e2bd2009-01-07 18:08:23 -08001865 if (scanning_global_lru(sc))
KAMEZAWA Hiroyuki1cfb4192008-02-07 00:14:37 -08001866 count_vm_event(ALLOCSTALL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001867
1868 for (priority = DEF_PRIORITY; priority >= 0; priority--) {
Balbir Singh66e17072008-02-07 00:13:56 -08001869 sc->nr_scanned = 0;
Rik van Rielf7b7fd82005-11-28 13:44:07 -08001870 if (!priority)
1871 disable_swap_token();
KOSAKI Motohirobb21c7c2010-06-04 14:15:05 -07001872 all_unreclaimable = shrink_zones(priority, zonelist, sc);
Balbir Singh66e17072008-02-07 00:13:56 -08001873 /*
1874 * Don't shrink slabs when reclaiming memory from
1875 * over limit cgroups
1876 */
KAMEZAWA Hiroyukie72e2bd2009-01-07 18:08:23 -08001877 if (scanning_global_lru(sc)) {
KOSAKI Motohiroc6a8a8c2010-08-09 17:19:14 -07001878 unsigned long lru_pages = 0;
Mel Gormand4debc62010-08-09 17:19:29 -07001879 for_each_zone_zonelist(zone, z, zonelist,
1880 gfp_zone(sc->gfp_mask)) {
KOSAKI Motohiroc6a8a8c2010-08-09 17:19:14 -07001881 if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL))
1882 continue;
1883
1884 lru_pages += zone_reclaimable_pages(zone);
1885 }
1886
Mel Gormandd1a2392008-04-28 02:12:17 -07001887 shrink_slab(sc->nr_scanned, sc->gfp_mask, lru_pages);
KAMEZAWA Hiroyuki91a45472008-02-07 00:14:29 -08001888 if (reclaim_state) {
Rik van Riela79311c2009-01-06 14:40:01 -08001889 sc->nr_reclaimed += reclaim_state->reclaimed_slab;
KAMEZAWA Hiroyuki91a45472008-02-07 00:14:29 -08001890 reclaim_state->reclaimed_slab = 0;
1891 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892 }
Balbir Singh66e17072008-02-07 00:13:56 -08001893 total_scanned += sc->nr_scanned;
KOSAKI Motohirobb21c7c2010-06-04 14:15:05 -07001894 if (sc->nr_reclaimed >= sc->nr_to_reclaim)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001895 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001896
1897 /*
1898 * Try to write back as many pages as we just scanned. This
1899 * tends to cause slow streaming writers to write data to the
1900 * disk smoothly, at the dirtying rate, which is nice. But
1901 * that's undesirable in laptop mode, where we *want* lumpy
1902 * writeout. So in laptop mode, write out the whole world.
1903 */
KOSAKI Motohiro22fba332009-12-14 17:59:10 -08001904 writeback_threshold = sc->nr_to_reclaim + sc->nr_to_reclaim / 2;
1905 if (total_scanned > writeback_threshold) {
Jens Axboe03ba3782009-09-09 09:08:54 +02001906 wakeup_flusher_threads(laptop_mode ? 0 : total_scanned);
Balbir Singh66e17072008-02-07 00:13:56 -08001907 sc->may_writepage = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001908 }
1909
1910 /* Take a nap, wait for some writeback to complete */
KOSAKI Motohiro7b517552009-12-14 17:59:12 -08001911 if (!sc->hibernation_mode && sc->nr_scanned &&
1912 priority < DEF_PRIORITY - 2)
Jens Axboe8aa7e842009-07-09 14:52:32 +02001913 congestion_wait(BLK_RW_ASYNC, HZ/10);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914 }
KOSAKI Motohirobb21c7c2010-06-04 14:15:05 -07001915
Linus Torvalds1da177e2005-04-16 15:20:36 -07001916out:
Martin Bligh3bb1a8522006-10-28 10:38:24 -07001917 /*
1918 * Now that we've scanned all the zones at this priority level, note
1919 * that level within the zone so that the next thread which performs
1920 * scanning of this zone will immediately start out at this priority
1921 * level. This affects only the decision whether or not to bring
1922 * mapped pages onto the inactive list.
1923 */
1924 if (priority < 0)
1925 priority = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926
Keika Kobayashi873b4772008-07-25 01:48:52 -07001927 delayacct_freepages_end();
Miao Xiec0ff7452010-05-24 14:32:08 -07001928 put_mems_allowed();
Keika Kobayashi873b4772008-07-25 01:48:52 -07001929
KOSAKI Motohirobb21c7c2010-06-04 14:15:05 -07001930 if (sc->nr_reclaimed)
1931 return sc->nr_reclaimed;
1932
1933 /* top priority shrink_zones still had more to do? don't OOM, then */
1934 if (scanning_global_lru(sc) && !all_unreclaimable)
1935 return 1;
1936
1937 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938}
1939
Mel Gormandac1d272008-04-28 02:12:12 -07001940unsigned long try_to_free_pages(struct zonelist *zonelist, int order,
KAMEZAWA Hiroyuki327c0e92009-03-31 15:23:31 -07001941 gfp_t gfp_mask, nodemask_t *nodemask)
Balbir Singh66e17072008-02-07 00:13:56 -08001942{
Mel Gorman33906bc2010-08-09 17:19:16 -07001943 unsigned long nr_reclaimed;
Balbir Singh66e17072008-02-07 00:13:56 -08001944 struct scan_control sc = {
1945 .gfp_mask = gfp_mask,
1946 .may_writepage = !laptop_mode,
KOSAKI Motohiro22fba332009-12-14 17:59:10 -08001947 .nr_to_reclaim = SWAP_CLUSTER_MAX,
Johannes Weinera6dc60f82009-03-31 15:19:30 -07001948 .may_unmap = 1,
KOSAKI Motohiro2e2e4252009-04-21 12:24:57 -07001949 .may_swap = 1,
Balbir Singh66e17072008-02-07 00:13:56 -08001950 .swappiness = vm_swappiness,
1951 .order = order,
1952 .mem_cgroup = NULL,
KAMEZAWA Hiroyuki327c0e92009-03-31 15:23:31 -07001953 .nodemask = nodemask,
Balbir Singh66e17072008-02-07 00:13:56 -08001954 };
1955
Mel Gorman33906bc2010-08-09 17:19:16 -07001956 trace_mm_vmscan_direct_reclaim_begin(order,
1957 sc.may_writepage,
1958 gfp_mask);
1959
1960 nr_reclaimed = do_try_to_free_pages(zonelist, &sc);
1961
1962 trace_mm_vmscan_direct_reclaim_end(nr_reclaimed);
1963
1964 return nr_reclaimed;
Balbir Singh66e17072008-02-07 00:13:56 -08001965}
1966
Balbir Singh00f0b822008-03-04 14:28:39 -08001967#ifdef CONFIG_CGROUP_MEM_RES_CTLR
Balbir Singh66e17072008-02-07 00:13:56 -08001968
Balbir Singh4e416952009-09-23 15:56:39 -07001969unsigned long mem_cgroup_shrink_node_zone(struct mem_cgroup *mem,
1970 gfp_t gfp_mask, bool noswap,
1971 unsigned int swappiness,
1972 struct zone *zone, int nid)
1973{
1974 struct scan_control sc = {
1975 .may_writepage = !laptop_mode,
1976 .may_unmap = 1,
1977 .may_swap = !noswap,
Balbir Singh4e416952009-09-23 15:56:39 -07001978 .swappiness = swappiness,
1979 .order = 0,
1980 .mem_cgroup = mem,
Balbir Singh4e416952009-09-23 15:56:39 -07001981 };
1982 nodemask_t nm = nodemask_of_node(nid);
1983
1984 sc.gfp_mask = (gfp_mask & GFP_RECLAIM_MASK) |
1985 (GFP_HIGHUSER_MOVABLE & ~GFP_RECLAIM_MASK);
1986 sc.nodemask = &nm;
1987 sc.nr_reclaimed = 0;
1988 sc.nr_scanned = 0;
KOSAKI Motohirobdce6d92010-08-09 17:19:56 -07001989
1990 trace_mm_vmscan_memcg_softlimit_reclaim_begin(0,
1991 sc.may_writepage,
1992 sc.gfp_mask);
1993
Balbir Singh4e416952009-09-23 15:56:39 -07001994 /*
1995 * NOTE: Although we can get the priority field, using it
1996 * here is not a good idea, since it limits the pages we can scan.
1997 * if we don't reclaim here, the shrink_zone from balance_pgdat
1998 * will pick up pages from other mem cgroup's as well. We hack
1999 * the priority and make it zero.
2000 */
2001 shrink_zone(0, zone, &sc);
KOSAKI Motohirobdce6d92010-08-09 17:19:56 -07002002
2003 trace_mm_vmscan_memcg_softlimit_reclaim_end(sc.nr_reclaimed);
2004
Balbir Singh4e416952009-09-23 15:56:39 -07002005 return sc.nr_reclaimed;
2006}
2007
Balbir Singhe1a1cd52008-02-07 00:14:02 -08002008unsigned long try_to_free_mem_cgroup_pages(struct mem_cgroup *mem_cont,
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08002009 gfp_t gfp_mask,
2010 bool noswap,
2011 unsigned int swappiness)
Balbir Singh66e17072008-02-07 00:13:56 -08002012{
Balbir Singh4e416952009-09-23 15:56:39 -07002013 struct zonelist *zonelist;
KOSAKI Motohirobdce6d92010-08-09 17:19:56 -07002014 unsigned long nr_reclaimed;
Balbir Singh66e17072008-02-07 00:13:56 -08002015 struct scan_control sc = {
Balbir Singh66e17072008-02-07 00:13:56 -08002016 .may_writepage = !laptop_mode,
Johannes Weinera6dc60f82009-03-31 15:19:30 -07002017 .may_unmap = 1,
KOSAKI Motohiro2e2e4252009-04-21 12:24:57 -07002018 .may_swap = !noswap,
KOSAKI Motohiro22fba332009-12-14 17:59:10 -08002019 .nr_to_reclaim = SWAP_CLUSTER_MAX,
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08002020 .swappiness = swappiness,
Balbir Singh66e17072008-02-07 00:13:56 -08002021 .order = 0,
2022 .mem_cgroup = mem_cont,
KAMEZAWA Hiroyuki327c0e92009-03-31 15:23:31 -07002023 .nodemask = NULL, /* we don't care the placement */
Balbir Singh66e17072008-02-07 00:13:56 -08002024 };
Balbir Singh66e17072008-02-07 00:13:56 -08002025
Mel Gormandd1a2392008-04-28 02:12:17 -07002026 sc.gfp_mask = (gfp_mask & GFP_RECLAIM_MASK) |
2027 (GFP_HIGHUSER_MOVABLE & ~GFP_RECLAIM_MASK);
2028 zonelist = NODE_DATA(numa_node_id())->node_zonelists;
KOSAKI Motohirobdce6d92010-08-09 17:19:56 -07002029
2030 trace_mm_vmscan_memcg_reclaim_begin(0,
2031 sc.may_writepage,
2032 sc.gfp_mask);
2033
2034 nr_reclaimed = do_try_to_free_pages(zonelist, &sc);
2035
2036 trace_mm_vmscan_memcg_reclaim_end(nr_reclaimed);
2037
2038 return nr_reclaimed;
Balbir Singh66e17072008-02-07 00:13:56 -08002039}
2040#endif
2041
Mel Gormanf50de2d2009-12-14 17:58:53 -08002042/* is kswapd sleeping prematurely? */
KOSAKI Motohirobb3ab592009-12-14 17:58:55 -08002043static int sleeping_prematurely(pg_data_t *pgdat, int order, long remaining)
Mel Gormanf50de2d2009-12-14 17:58:53 -08002044{
KOSAKI Motohirobb3ab592009-12-14 17:58:55 -08002045 int i;
Mel Gormanf50de2d2009-12-14 17:58:53 -08002046
2047 /* If a direct reclaimer woke kswapd within HZ/10, it's premature */
2048 if (remaining)
2049 return 1;
2050
2051 /* If after HZ/10, a zone is below the high mark, it's premature */
KOSAKI Motohirobb3ab592009-12-14 17:58:55 -08002052 for (i = 0; i < pgdat->nr_zones; i++) {
2053 struct zone *zone = pgdat->node_zones + i;
2054
2055 if (!populated_zone(zone))
2056 continue;
2057
KOSAKI Motohiro93e4a892010-03-05 13:41:55 -08002058 if (zone->all_unreclaimable)
KOSAKI Motohirode3fab32010-01-15 17:01:25 -08002059 continue;
2060
Mel Gormanf50de2d2009-12-14 17:58:53 -08002061 if (!zone_watermark_ok(zone, order, high_wmark_pages(zone),
2062 0, 0))
2063 return 1;
KOSAKI Motohirobb3ab592009-12-14 17:58:55 -08002064 }
Mel Gormanf50de2d2009-12-14 17:58:53 -08002065
2066 return 0;
2067}
2068
Linus Torvalds1da177e2005-04-16 15:20:36 -07002069/*
2070 * For kswapd, balance_pgdat() will work across all this node's zones until
Mel Gorman41858962009-06-16 15:32:12 -07002071 * they are all at high_wmark_pages(zone).
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002073 * Returns the number of pages which were actually freed.
2074 *
2075 * There is special handling here for zones which are full of pinned pages.
2076 * This can happen if the pages are all mlocked, or if they are all used by
2077 * device drivers (say, ZONE_DMA). Or if they are all in use by hugetlb.
2078 * What we do is to detect the case where all pages in the zone have been
2079 * scanned twice and there has been zero successful reclaim. Mark the zone as
2080 * dead and from now on, only perform a short scan. Basically we're polling
2081 * the zone for when the problem goes away.
2082 *
2083 * kswapd scans the zones in the highmem->normal->dma direction. It skips
Mel Gorman41858962009-06-16 15:32:12 -07002084 * zones which have free_pages > high_wmark_pages(zone), but once a zone is
2085 * found to have free_pages <= high_wmark_pages(zone), we scan that zone and the
2086 * lower zones regardless of the number of free pages in the lower zones. This
2087 * interoperates with the page allocator fallback scheme to ensure that aging
2088 * of pages is balanced across the zones.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002089 */
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002090static unsigned long balance_pgdat(pg_data_t *pgdat, int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002091{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002092 int all_zones_ok;
2093 int priority;
2094 int i;
Andrew Morton69e05942006-03-22 00:08:19 -08002095 unsigned long total_scanned;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002096 struct reclaim_state *reclaim_state = current->reclaim_state;
Andrew Morton179e9632006-03-22 00:08:18 -08002097 struct scan_control sc = {
2098 .gfp_mask = GFP_KERNEL,
Johannes Weinera6dc60f82009-03-31 15:19:30 -07002099 .may_unmap = 1,
KOSAKI Motohiro2e2e4252009-04-21 12:24:57 -07002100 .may_swap = 1,
KOSAKI Motohiro22fba332009-12-14 17:59:10 -08002101 /*
2102 * kswapd doesn't want to be bailed out while reclaim. because
2103 * we want to put equal scanning pressure on each zone.
2104 */
2105 .nr_to_reclaim = ULONG_MAX,
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002106 .swappiness = vm_swappiness,
Andy Whitcroft5ad333e2007-07-17 04:03:16 -07002107 .order = order,
Balbir Singh66e17072008-02-07 00:13:56 -08002108 .mem_cgroup = NULL,
Andrew Morton179e9632006-03-22 00:08:18 -08002109 };
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110loop_again:
2111 total_scanned = 0;
Rik van Riela79311c2009-01-06 14:40:01 -08002112 sc.nr_reclaimed = 0;
Christoph Lameterc0bbbc72006-06-11 15:22:26 -07002113 sc.may_writepage = !laptop_mode;
Christoph Lameterf8891e52006-06-30 01:55:45 -07002114 count_vm_event(PAGEOUTRUN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116 for (priority = DEF_PRIORITY; priority >= 0; priority--) {
2117 int end_zone = 0; /* Inclusive. 0 = ZONE_DMA */
2118 unsigned long lru_pages = 0;
KOSAKI Motohirobb3ab592009-12-14 17:58:55 -08002119 int has_under_min_watermark_zone = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002120
Rik van Rielf7b7fd82005-11-28 13:44:07 -08002121 /* The swap token gets in the way of swapout... */
2122 if (!priority)
2123 disable_swap_token();
2124
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125 all_zones_ok = 1;
2126
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002127 /*
2128 * Scan in the highmem->dma direction for the highest
2129 * zone which needs scanning
2130 */
2131 for (i = pgdat->nr_zones - 1; i >= 0; i--) {
2132 struct zone *zone = pgdat->node_zones + i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002133
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002134 if (!populated_zone(zone))
2135 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002136
KOSAKI Motohiro93e4a892010-03-05 13:41:55 -08002137 if (zone->all_unreclaimable && priority != DEF_PRIORITY)
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002138 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002139
Rik van Riel556adec2008-10-18 20:26:34 -07002140 /*
2141 * Do some background aging of the anon list, to give
2142 * pages a chance to be referenced before reclaiming.
2143 */
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08002144 if (inactive_anon_is_low(zone, &sc))
Rik van Riel556adec2008-10-18 20:26:34 -07002145 shrink_active_list(SWAP_CLUSTER_MAX, zone,
2146 &sc, priority, 0);
2147
Mel Gorman41858962009-06-16 15:32:12 -07002148 if (!zone_watermark_ok(zone, order,
2149 high_wmark_pages(zone), 0, 0)) {
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002150 end_zone = i;
Andrew Mortone1dbeda2006-12-06 20:32:01 -08002151 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002153 }
Andrew Mortone1dbeda2006-12-06 20:32:01 -08002154 if (i < 0)
2155 goto out;
2156
Linus Torvalds1da177e2005-04-16 15:20:36 -07002157 for (i = 0; i <= end_zone; i++) {
2158 struct zone *zone = pgdat->node_zones + i;
2159
Wu Fengguangadea02a2009-09-21 17:01:42 -07002160 lru_pages += zone_reclaimable_pages(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002161 }
2162
2163 /*
2164 * Now scan the zone in the dma->highmem direction, stopping
2165 * at the last zone which needs scanning.
2166 *
2167 * We do this because the page allocator works in the opposite
2168 * direction. This prevents the page allocator from allocating
2169 * pages behind kswapd's direction of progress, which would
2170 * cause too much scanning of the lower zones.
2171 */
2172 for (i = 0; i <= end_zone; i++) {
2173 struct zone *zone = pgdat->node_zones + i;
akpm@osdl.orgb15e0902005-06-21 17:14:35 -07002174 int nr_slab;
Balbir Singh4e416952009-09-23 15:56:39 -07002175 int nid, zid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002176
Con Kolivasf3fe6512006-01-06 00:11:15 -08002177 if (!populated_zone(zone))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002178 continue;
2179
KOSAKI Motohiro93e4a892010-03-05 13:41:55 -08002180 if (zone->all_unreclaimable && priority != DEF_PRIORITY)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002181 continue;
2182
Linus Torvalds1da177e2005-04-16 15:20:36 -07002183 sc.nr_scanned = 0;
Balbir Singh4e416952009-09-23 15:56:39 -07002184
2185 nid = pgdat->node_id;
2186 zid = zone_idx(zone);
2187 /*
2188 * Call soft limit reclaim before calling shrink_zone.
2189 * For now we ignore the return value
2190 */
2191 mem_cgroup_soft_limit_reclaim(zone, order, sc.gfp_mask,
2192 nid, zid);
Rik van Riel32a43302007-10-16 01:24:50 -07002193 /*
2194 * We put equal pressure on every zone, unless one
2195 * zone has way too many pages free already.
2196 */
Mel Gorman41858962009-06-16 15:32:12 -07002197 if (!zone_watermark_ok(zone, order,
2198 8*high_wmark_pages(zone), end_zone, 0))
Rik van Riela79311c2009-01-06 14:40:01 -08002199 shrink_zone(priority, zone, &sc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200 reclaim_state->reclaimed_slab = 0;
akpm@osdl.orgb15e0902005-06-21 17:14:35 -07002201 nr_slab = shrink_slab(sc.nr_scanned, GFP_KERNEL,
2202 lru_pages);
Rik van Riela79311c2009-01-06 14:40:01 -08002203 sc.nr_reclaimed += reclaim_state->reclaimed_slab;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204 total_scanned += sc.nr_scanned;
KOSAKI Motohiro93e4a892010-03-05 13:41:55 -08002205 if (zone->all_unreclaimable)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002206 continue;
KOSAKI Motohiro93e4a892010-03-05 13:41:55 -08002207 if (nr_slab == 0 &&
2208 zone->pages_scanned >= (zone_reclaimable_pages(zone) * 6))
2209 zone->all_unreclaimable = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002210 /*
2211 * If we've done a decent amount of scanning and
2212 * the reclaim ratio is low, start doing writepage
2213 * even in laptop mode
2214 */
2215 if (total_scanned > SWAP_CLUSTER_MAX * 2 &&
Rik van Riela79311c2009-01-06 14:40:01 -08002216 total_scanned > sc.nr_reclaimed + sc.nr_reclaimed / 2)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002217 sc.may_writepage = 1;
KOSAKI Motohirobb3ab592009-12-14 17:58:55 -08002218
Minchan Kim45973d72010-03-05 13:41:45 -08002219 if (!zone_watermark_ok(zone, order,
2220 high_wmark_pages(zone), end_zone, 0)) {
2221 all_zones_ok = 0;
2222 /*
2223 * We are still under min water mark. This
2224 * means that we have a GFP_ATOMIC allocation
2225 * failure risk. Hurry up!
2226 */
2227 if (!zone_watermark_ok(zone, order,
2228 min_wmark_pages(zone), end_zone, 0))
2229 has_under_min_watermark_zone = 1;
2230 }
KOSAKI Motohirobb3ab592009-12-14 17:58:55 -08002231
Linus Torvalds1da177e2005-04-16 15:20:36 -07002232 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002233 if (all_zones_ok)
2234 break; /* kswapd: all done */
2235 /*
2236 * OK, kswapd is getting into trouble. Take a nap, then take
2237 * another pass across the zones.
2238 */
KOSAKI Motohirobb3ab592009-12-14 17:58:55 -08002239 if (total_scanned && (priority < DEF_PRIORITY - 2)) {
2240 if (has_under_min_watermark_zone)
2241 count_vm_event(KSWAPD_SKIP_CONGESTION_WAIT);
2242 else
2243 congestion_wait(BLK_RW_ASYNC, HZ/10);
2244 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002245
2246 /*
2247 * We do this so kswapd doesn't build up large priorities for
2248 * example when it is freeing in parallel with allocators. It
2249 * matches the direct reclaim path behaviour in terms of impact
2250 * on zone->*_priority.
2251 */
Rik van Riela79311c2009-01-06 14:40:01 -08002252 if (sc.nr_reclaimed >= SWAP_CLUSTER_MAX)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002253 break;
2254 }
2255out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002256 if (!all_zones_ok) {
2257 cond_resched();
Rafael J. Wysocki83573762006-12-06 20:34:18 -08002258
2259 try_to_freeze();
2260
KOSAKI Motohiro73ce02e2009-01-06 14:40:33 -08002261 /*
2262 * Fragmentation may mean that the system cannot be
2263 * rebalanced for high-order allocations in all zones.
2264 * At this point, if nr_reclaimed < SWAP_CLUSTER_MAX,
2265 * it means the zones have been fully scanned and are still
2266 * not balanced. For high-order allocations, there is
2267 * little point trying all over again as kswapd may
2268 * infinite loop.
2269 *
2270 * Instead, recheck all watermarks at order-0 as they
2271 * are the most important. If watermarks are ok, kswapd will go
2272 * back to sleep. High-order users can still perform direct
2273 * reclaim if they wish.
2274 */
2275 if (sc.nr_reclaimed < SWAP_CLUSTER_MAX)
2276 order = sc.order = 0;
2277
Linus Torvalds1da177e2005-04-16 15:20:36 -07002278 goto loop_again;
2279 }
2280
Rik van Riela79311c2009-01-06 14:40:01 -08002281 return sc.nr_reclaimed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002282}
2283
2284/*
2285 * The background pageout daemon, started as a kernel thread
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002286 * from the init process.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002287 *
2288 * This basically trickles out pages so that we have _some_
2289 * free memory available even if there is no other activity
2290 * that frees anything up. This is needed for things like routing
2291 * etc, where we otherwise might have all activity going on in
2292 * asynchronous contexts that cannot page things out.
2293 *
2294 * If there are applications that are active memory-allocators
2295 * (most normal use), this basically shouldn't matter.
2296 */
2297static int kswapd(void *p)
2298{
2299 unsigned long order;
2300 pg_data_t *pgdat = (pg_data_t*)p;
2301 struct task_struct *tsk = current;
2302 DEFINE_WAIT(wait);
2303 struct reclaim_state reclaim_state = {
2304 .reclaimed_slab = 0,
2305 };
Rusty Russella70f7302009-03-13 14:49:46 +10302306 const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002307
Nick Piggincf40bd12009-01-21 08:12:39 +01002308 lockdep_set_current_reclaim_state(GFP_KERNEL);
2309
Rusty Russell174596a2009-01-01 10:12:29 +10302310 if (!cpumask_empty(cpumask))
Mike Travisc5f59f02008-04-04 18:11:10 -07002311 set_cpus_allowed_ptr(tsk, cpumask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002312 current->reclaim_state = &reclaim_state;
2313
2314 /*
2315 * Tell the memory management that we're a "memory allocator",
2316 * and that if we need more memory we should get access to it
2317 * regardless (see "__alloc_pages()"). "kswapd" should
2318 * never get caught in the normal page freeing logic.
2319 *
2320 * (Kswapd normally doesn't need memory anyway, but sometimes
2321 * you need a small amount of memory in order to be able to
2322 * page out something else, and this flag essentially protects
2323 * us from recursively trying to free more memory as we're
2324 * trying to free the first piece of memory in the first place).
2325 */
Christoph Lameter930d9152006-01-08 01:00:47 -08002326 tsk->flags |= PF_MEMALLOC | PF_SWAPWRITE | PF_KSWAPD;
Rafael J. Wysocki83144182007-07-17 04:03:35 -07002327 set_freezable();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002328
2329 order = 0;
2330 for ( ; ; ) {
2331 unsigned long new_order;
David Rientjes8fe23e02009-12-14 17:58:33 -08002332 int ret;
Christoph Lameter3e1d1d22005-06-24 23:13:50 -07002333
Linus Torvalds1da177e2005-04-16 15:20:36 -07002334 prepare_to_wait(&pgdat->kswapd_wait, &wait, TASK_INTERRUPTIBLE);
2335 new_order = pgdat->kswapd_max_order;
2336 pgdat->kswapd_max_order = 0;
2337 if (order < new_order) {
2338 /*
2339 * Don't sleep if someone wants a larger 'order'
2340 * allocation
2341 */
2342 order = new_order;
2343 } else {
Mel Gormanf50de2d2009-12-14 17:58:53 -08002344 if (!freezing(current) && !kthread_should_stop()) {
2345 long remaining = 0;
2346
2347 /* Try to sleep for a short interval */
KOSAKI Motohirobb3ab592009-12-14 17:58:55 -08002348 if (!sleeping_prematurely(pgdat, order, remaining)) {
Mel Gormanf50de2d2009-12-14 17:58:53 -08002349 remaining = schedule_timeout(HZ/10);
2350 finish_wait(&pgdat->kswapd_wait, &wait);
2351 prepare_to_wait(&pgdat->kswapd_wait, &wait, TASK_INTERRUPTIBLE);
2352 }
2353
2354 /*
2355 * After a short sleep, check if it was a
2356 * premature sleep. If not, then go fully
2357 * to sleep until explicitly woken up
2358 */
Mel Gorman33906bc2010-08-09 17:19:16 -07002359 if (!sleeping_prematurely(pgdat, order, remaining)) {
2360 trace_mm_vmscan_kswapd_sleep(pgdat->node_id);
Mel Gormanf50de2d2009-12-14 17:58:53 -08002361 schedule();
Mel Gorman33906bc2010-08-09 17:19:16 -07002362 } else {
Mel Gormanf50de2d2009-12-14 17:58:53 -08002363 if (remaining)
KOSAKI Motohirobb3ab592009-12-14 17:58:55 -08002364 count_vm_event(KSWAPD_LOW_WMARK_HIT_QUICKLY);
Mel Gormanf50de2d2009-12-14 17:58:53 -08002365 else
KOSAKI Motohirobb3ab592009-12-14 17:58:55 -08002366 count_vm_event(KSWAPD_HIGH_WMARK_HIT_QUICKLY);
Mel Gormanf50de2d2009-12-14 17:58:53 -08002367 }
2368 }
Rafael J. Wysockib1296cc2007-05-06 14:50:48 -07002369
Linus Torvalds1da177e2005-04-16 15:20:36 -07002370 order = pgdat->kswapd_max_order;
2371 }
2372 finish_wait(&pgdat->kswapd_wait, &wait);
2373
David Rientjes8fe23e02009-12-14 17:58:33 -08002374 ret = try_to_freeze();
2375 if (kthread_should_stop())
2376 break;
2377
2378 /*
2379 * We can speed up thawing tasks if we don't call balance_pgdat
2380 * after returning from the refrigerator
2381 */
Mel Gorman33906bc2010-08-09 17:19:16 -07002382 if (!ret) {
2383 trace_mm_vmscan_kswapd_wake(pgdat->node_id, order);
Rafael J. Wysockib1296cc2007-05-06 14:50:48 -07002384 balance_pgdat(pgdat, order);
Mel Gorman33906bc2010-08-09 17:19:16 -07002385 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002386 }
2387 return 0;
2388}
2389
2390/*
2391 * A zone is low on free memory, so wake its kswapd task to service it.
2392 */
2393void wakeup_kswapd(struct zone *zone, int order)
2394{
2395 pg_data_t *pgdat;
2396
Con Kolivasf3fe6512006-01-06 00:11:15 -08002397 if (!populated_zone(zone))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002398 return;
2399
2400 pgdat = zone->zone_pgdat;
Mel Gorman41858962009-06-16 15:32:12 -07002401 if (zone_watermark_ok(zone, order, low_wmark_pages(zone), 0, 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002402 return;
2403 if (pgdat->kswapd_max_order < order)
2404 pgdat->kswapd_max_order = order;
Mel Gorman33906bc2010-08-09 17:19:16 -07002405 trace_mm_vmscan_wakeup_kswapd(pgdat->node_id, zone_idx(zone), order);
Paul Jackson02a0e532006-12-13 00:34:25 -08002406 if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002407 return;
Con Kolivas8d0986e2005-09-13 01:25:07 -07002408 if (!waitqueue_active(&pgdat->kswapd_wait))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002409 return;
Con Kolivas8d0986e2005-09-13 01:25:07 -07002410 wake_up_interruptible(&pgdat->kswapd_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002411}
2412
Wu Fengguangadea02a2009-09-21 17:01:42 -07002413/*
2414 * The reclaimable count would be mostly accurate.
2415 * The less reclaimable pages may be
2416 * - mlocked pages, which will be moved to unevictable list when encountered
2417 * - mapped pages, which may require several travels to be reclaimed
2418 * - dirty pages, which is not "instantly" reclaimable
2419 */
2420unsigned long global_reclaimable_pages(void)
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002421{
Wu Fengguangadea02a2009-09-21 17:01:42 -07002422 int nr;
2423
2424 nr = global_page_state(NR_ACTIVE_FILE) +
2425 global_page_state(NR_INACTIVE_FILE);
2426
2427 if (nr_swap_pages > 0)
2428 nr += global_page_state(NR_ACTIVE_ANON) +
2429 global_page_state(NR_INACTIVE_ANON);
2430
2431 return nr;
2432}
2433
2434unsigned long zone_reclaimable_pages(struct zone *zone)
2435{
2436 int nr;
2437
2438 nr = zone_page_state(zone, NR_ACTIVE_FILE) +
2439 zone_page_state(zone, NR_INACTIVE_FILE);
2440
2441 if (nr_swap_pages > 0)
2442 nr += zone_page_state(zone, NR_ACTIVE_ANON) +
2443 zone_page_state(zone, NR_INACTIVE_ANON);
2444
2445 return nr;
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002446}
2447
Rafael J. Wysockic6f37f12009-05-24 22:16:31 +02002448#ifdef CONFIG_HIBERNATION
Linus Torvalds1da177e2005-04-16 15:20:36 -07002449/*
KOSAKI Motohiro7b517552009-12-14 17:59:12 -08002450 * Try to free `nr_to_reclaim' of memory, system-wide, and return the number of
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002451 * freed pages.
2452 *
2453 * Rather than trying to age LRUs the aim is to preserve the overall
2454 * LRU order by reclaiming preferentially
2455 * inactive > active > active referenced > active mapped
Linus Torvalds1da177e2005-04-16 15:20:36 -07002456 */
KOSAKI Motohiro7b517552009-12-14 17:59:12 -08002457unsigned long shrink_all_memory(unsigned long nr_to_reclaim)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002458{
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002459 struct reclaim_state reclaim_state;
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002460 struct scan_control sc = {
KOSAKI Motohiro7b517552009-12-14 17:59:12 -08002461 .gfp_mask = GFP_HIGHUSER_MOVABLE,
2462 .may_swap = 1,
2463 .may_unmap = 1,
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002464 .may_writepage = 1,
KOSAKI Motohiro7b517552009-12-14 17:59:12 -08002465 .nr_to_reclaim = nr_to_reclaim,
2466 .hibernation_mode = 1,
2467 .swappiness = vm_swappiness,
2468 .order = 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002469 };
KOSAKI Motohiro7b517552009-12-14 17:59:12 -08002470 struct zonelist * zonelist = node_zonelist(numa_node_id(), sc.gfp_mask);
2471 struct task_struct *p = current;
2472 unsigned long nr_reclaimed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002473
KOSAKI Motohiro7b517552009-12-14 17:59:12 -08002474 p->flags |= PF_MEMALLOC;
2475 lockdep_set_current_reclaim_state(sc.gfp_mask);
2476 reclaim_state.reclaimed_slab = 0;
2477 p->reclaim_state = &reclaim_state;
Andrew Morton69e05942006-03-22 00:08:19 -08002478
KOSAKI Motohiro7b517552009-12-14 17:59:12 -08002479 nr_reclaimed = do_try_to_free_pages(zonelist, &sc);
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002480
KOSAKI Motohiro7b517552009-12-14 17:59:12 -08002481 p->reclaim_state = NULL;
2482 lockdep_clear_current_reclaim_state();
2483 p->flags &= ~PF_MEMALLOC;
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002484
KOSAKI Motohiro7b517552009-12-14 17:59:12 -08002485 return nr_reclaimed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002486}
Rafael J. Wysockic6f37f12009-05-24 22:16:31 +02002487#endif /* CONFIG_HIBERNATION */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002488
Linus Torvalds1da177e2005-04-16 15:20:36 -07002489/* It's optimal to keep kswapds on the same CPUs as their memory, but
2490 not required for correctness. So if the last cpu in a node goes
2491 away, we get changed to run anywhere: as the first one comes back,
2492 restore their cpu bindings. */
Chandra Seetharaman9c7b2162006-06-27 02:54:07 -07002493static int __devinit cpu_callback(struct notifier_block *nfb,
Andrew Morton69e05942006-03-22 00:08:19 -08002494 unsigned long action, void *hcpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002495{
Yasunori Goto58c0a4a2007-10-16 01:25:40 -07002496 int nid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002497
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07002498 if (action == CPU_ONLINE || action == CPU_ONLINE_FROZEN) {
Yasunori Goto58c0a4a2007-10-16 01:25:40 -07002499 for_each_node_state(nid, N_HIGH_MEMORY) {
Mike Travisc5f59f02008-04-04 18:11:10 -07002500 pg_data_t *pgdat = NODE_DATA(nid);
Rusty Russella70f7302009-03-13 14:49:46 +10302501 const struct cpumask *mask;
2502
2503 mask = cpumask_of_node(pgdat->node_id);
Mike Travisc5f59f02008-04-04 18:11:10 -07002504
Rusty Russell3e597942009-01-01 10:12:24 +10302505 if (cpumask_any_and(cpu_online_mask, mask) < nr_cpu_ids)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002506 /* One of our CPUs online: restore mask */
Mike Travisc5f59f02008-04-04 18:11:10 -07002507 set_cpus_allowed_ptr(pgdat->kswapd, mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002508 }
2509 }
2510 return NOTIFY_OK;
2511}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002512
Yasunori Goto3218ae12006-06-27 02:53:33 -07002513/*
2514 * This kswapd start function will be called by init and node-hot-add.
2515 * On node-hot-add, kswapd will moved to proper cpus if cpus are hot-added.
2516 */
2517int kswapd_run(int nid)
2518{
2519 pg_data_t *pgdat = NODE_DATA(nid);
2520 int ret = 0;
2521
2522 if (pgdat->kswapd)
2523 return 0;
2524
2525 pgdat->kswapd = kthread_run(kswapd, pgdat, "kswapd%d", nid);
2526 if (IS_ERR(pgdat->kswapd)) {
2527 /* failure at boot is fatal */
2528 BUG_ON(system_state == SYSTEM_BOOTING);
2529 printk("Failed to start kswapd on node %d\n",nid);
2530 ret = -1;
2531 }
2532 return ret;
2533}
2534
David Rientjes8fe23e02009-12-14 17:58:33 -08002535/*
2536 * Called by memory hotplug when all memory in a node is offlined.
2537 */
2538void kswapd_stop(int nid)
2539{
2540 struct task_struct *kswapd = NODE_DATA(nid)->kswapd;
2541
2542 if (kswapd)
2543 kthread_stop(kswapd);
2544}
2545
Linus Torvalds1da177e2005-04-16 15:20:36 -07002546static int __init kswapd_init(void)
2547{
Yasunori Goto3218ae12006-06-27 02:53:33 -07002548 int nid;
Andrew Morton69e05942006-03-22 00:08:19 -08002549
Linus Torvalds1da177e2005-04-16 15:20:36 -07002550 swap_setup();
Christoph Lameter9422ffb2007-10-16 01:25:31 -07002551 for_each_node_state(nid, N_HIGH_MEMORY)
Yasunori Goto3218ae12006-06-27 02:53:33 -07002552 kswapd_run(nid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002553 hotcpu_notifier(cpu_callback, 0);
2554 return 0;
2555}
2556
2557module_init(kswapd_init)
Christoph Lameter9eeff232006-01-18 17:42:31 -08002558
2559#ifdef CONFIG_NUMA
2560/*
2561 * Zone reclaim mode
2562 *
2563 * If non-zero call zone_reclaim when the number of free pages falls below
2564 * the watermarks.
Christoph Lameter9eeff232006-01-18 17:42:31 -08002565 */
2566int zone_reclaim_mode __read_mostly;
2567
Christoph Lameter1b2ffb72006-02-01 03:05:34 -08002568#define RECLAIM_OFF 0
Fernando Luis Vazquez Cao7d034312008-07-29 22:33:41 -07002569#define RECLAIM_ZONE (1<<0) /* Run shrink_inactive_list on the zone */
Christoph Lameter1b2ffb72006-02-01 03:05:34 -08002570#define RECLAIM_WRITE (1<<1) /* Writeout pages during reclaim */
2571#define RECLAIM_SWAP (1<<2) /* Swap pages out during reclaim */
2572
Christoph Lameter9eeff232006-01-18 17:42:31 -08002573/*
Christoph Lametera92f7122006-02-01 03:05:32 -08002574 * Priority for ZONE_RECLAIM. This determines the fraction of pages
2575 * of a node considered for each zone_reclaim. 4 scans 1/16th of
2576 * a zone.
2577 */
2578#define ZONE_RECLAIM_PRIORITY 4
2579
Christoph Lameter9eeff232006-01-18 17:42:31 -08002580/*
Christoph Lameter96146342006-07-03 00:24:13 -07002581 * Percentage of pages in a zone that must be unmapped for zone_reclaim to
2582 * occur.
2583 */
2584int sysctl_min_unmapped_ratio = 1;
2585
2586/*
Christoph Lameter0ff38492006-09-25 23:31:52 -07002587 * If the number of slab pages in a zone grows beyond this percentage then
2588 * slab reclaim needs to occur.
2589 */
2590int sysctl_min_slab_ratio = 5;
2591
Mel Gorman90afa5d2009-06-16 15:33:20 -07002592static inline unsigned long zone_unmapped_file_pages(struct zone *zone)
2593{
2594 unsigned long file_mapped = zone_page_state(zone, NR_FILE_MAPPED);
2595 unsigned long file_lru = zone_page_state(zone, NR_INACTIVE_FILE) +
2596 zone_page_state(zone, NR_ACTIVE_FILE);
2597
2598 /*
2599 * It's possible for there to be more file mapped pages than
2600 * accounted for by the pages on the file LRU lists because
2601 * tmpfs pages accounted for as ANON can also be FILE_MAPPED
2602 */
2603 return (file_lru > file_mapped) ? (file_lru - file_mapped) : 0;
2604}
2605
2606/* Work out how many page cache pages we can reclaim in this reclaim_mode */
2607static long zone_pagecache_reclaimable(struct zone *zone)
2608{
2609 long nr_pagecache_reclaimable;
2610 long delta = 0;
2611
2612 /*
2613 * If RECLAIM_SWAP is set, then all file pages are considered
2614 * potentially reclaimable. Otherwise, we have to worry about
2615 * pages like swapcache and zone_unmapped_file_pages() provides
2616 * a better estimate
2617 */
2618 if (zone_reclaim_mode & RECLAIM_SWAP)
2619 nr_pagecache_reclaimable = zone_page_state(zone, NR_FILE_PAGES);
2620 else
2621 nr_pagecache_reclaimable = zone_unmapped_file_pages(zone);
2622
2623 /* If we can't clean pages, remove dirty pages from consideration */
2624 if (!(zone_reclaim_mode & RECLAIM_WRITE))
2625 delta += zone_page_state(zone, NR_FILE_DIRTY);
2626
2627 /* Watch for any possible underflows due to delta */
2628 if (unlikely(delta > nr_pagecache_reclaimable))
2629 delta = nr_pagecache_reclaimable;
2630
2631 return nr_pagecache_reclaimable - delta;
2632}
2633
Christoph Lameter0ff38492006-09-25 23:31:52 -07002634/*
Christoph Lameter9eeff232006-01-18 17:42:31 -08002635 * Try to free up some pages from this zone through reclaim.
2636 */
Andrew Morton179e9632006-03-22 00:08:18 -08002637static int __zone_reclaim(struct zone *zone, gfp_t gfp_mask, unsigned int order)
Christoph Lameter9eeff232006-01-18 17:42:31 -08002638{
Christoph Lameter7fb2d462006-03-22 00:08:22 -08002639 /* Minimum pages needed in order to stay on node */
Andrew Morton69e05942006-03-22 00:08:19 -08002640 const unsigned long nr_pages = 1 << order;
Christoph Lameter9eeff232006-01-18 17:42:31 -08002641 struct task_struct *p = current;
2642 struct reclaim_state reclaim_state;
Christoph Lameter86959492006-03-22 00:08:18 -08002643 int priority;
Andrew Morton179e9632006-03-22 00:08:18 -08002644 struct scan_control sc = {
2645 .may_writepage = !!(zone_reclaim_mode & RECLAIM_WRITE),
Johannes Weinera6dc60f82009-03-31 15:19:30 -07002646 .may_unmap = !!(zone_reclaim_mode & RECLAIM_SWAP),
KOSAKI Motohiro2e2e4252009-04-21 12:24:57 -07002647 .may_swap = 1,
KOSAKI Motohiro22fba332009-12-14 17:59:10 -08002648 .nr_to_reclaim = max_t(unsigned long, nr_pages,
2649 SWAP_CLUSTER_MAX),
Andrew Morton179e9632006-03-22 00:08:18 -08002650 .gfp_mask = gfp_mask,
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002651 .swappiness = vm_swappiness,
Johannes Weinerbd2f6192009-03-31 15:19:38 -07002652 .order = order,
Andrew Morton179e9632006-03-22 00:08:18 -08002653 };
KOSAKI Motohiro15748042010-08-09 17:19:50 -07002654 unsigned long nr_slab_pages0, nr_slab_pages1;
Christoph Lameter9eeff232006-01-18 17:42:31 -08002655
Christoph Lameter9eeff232006-01-18 17:42:31 -08002656 cond_resched();
Christoph Lameterd4f77962006-02-24 13:04:22 -08002657 /*
2658 * We need to be able to allocate from the reserves for RECLAIM_SWAP
2659 * and we also need to be able to write out pages for RECLAIM_WRITE
2660 * and RECLAIM_SWAP.
2661 */
2662 p->flags |= PF_MEMALLOC | PF_SWAPWRITE;
KOSAKI Motohiro76ca5422010-03-05 13:41:47 -08002663 lockdep_set_current_reclaim_state(gfp_mask);
Christoph Lameter9eeff232006-01-18 17:42:31 -08002664 reclaim_state.reclaimed_slab = 0;
2665 p->reclaim_state = &reclaim_state;
Christoph Lameterc84db232006-02-01 03:05:29 -08002666
Mel Gorman90afa5d2009-06-16 15:33:20 -07002667 if (zone_pagecache_reclaimable(zone) > zone->min_unmapped_pages) {
Christoph Lameter0ff38492006-09-25 23:31:52 -07002668 /*
2669 * Free memory by calling shrink zone with increasing
2670 * priorities until we have enough memory freed.
2671 */
2672 priority = ZONE_RECLAIM_PRIORITY;
2673 do {
Rik van Riela79311c2009-01-06 14:40:01 -08002674 shrink_zone(priority, zone, &sc);
Christoph Lameter0ff38492006-09-25 23:31:52 -07002675 priority--;
Rik van Riela79311c2009-01-06 14:40:01 -08002676 } while (priority >= 0 && sc.nr_reclaimed < nr_pages);
Christoph Lameter0ff38492006-09-25 23:31:52 -07002677 }
Christoph Lameterc84db232006-02-01 03:05:29 -08002678
KOSAKI Motohiro15748042010-08-09 17:19:50 -07002679 nr_slab_pages0 = zone_page_state(zone, NR_SLAB_RECLAIMABLE);
2680 if (nr_slab_pages0 > zone->min_slab_pages) {
Christoph Lameter2a16e3f2006-02-01 03:05:35 -08002681 /*
Christoph Lameter7fb2d462006-03-22 00:08:22 -08002682 * shrink_slab() does not currently allow us to determine how
Christoph Lameter0ff38492006-09-25 23:31:52 -07002683 * many pages were freed in this zone. So we take the current
2684 * number of slab pages and shake the slab until it is reduced
2685 * by the same nr_pages that we used for reclaiming unmapped
2686 * pages.
Christoph Lameter2a16e3f2006-02-01 03:05:35 -08002687 *
Christoph Lameter0ff38492006-09-25 23:31:52 -07002688 * Note that shrink_slab will free memory on all zones and may
2689 * take a long time.
Christoph Lameter2a16e3f2006-02-01 03:05:35 -08002690 */
KOSAKI Motohiro4dc4b3d2010-08-09 17:19:54 -07002691 for (;;) {
2692 unsigned long lru_pages = zone_reclaimable_pages(zone);
2693
2694 /* No reclaimable slab or very low memory pressure */
2695 if (!shrink_slab(sc.nr_scanned, gfp_mask, lru_pages))
2696 break;
2697
2698 /* Freed enough memory */
2699 nr_slab_pages1 = zone_page_state(zone,
2700 NR_SLAB_RECLAIMABLE);
2701 if (nr_slab_pages1 + nr_pages <= nr_slab_pages0)
2702 break;
2703 }
Christoph Lameter83e33a42006-09-25 23:31:53 -07002704
2705 /*
2706 * Update nr_reclaimed by the number of slab pages we
2707 * reclaimed from this zone.
2708 */
KOSAKI Motohiro15748042010-08-09 17:19:50 -07002709 nr_slab_pages1 = zone_page_state(zone, NR_SLAB_RECLAIMABLE);
2710 if (nr_slab_pages1 < nr_slab_pages0)
2711 sc.nr_reclaimed += nr_slab_pages0 - nr_slab_pages1;
Christoph Lameter2a16e3f2006-02-01 03:05:35 -08002712 }
2713
Christoph Lameter9eeff232006-01-18 17:42:31 -08002714 p->reclaim_state = NULL;
Christoph Lameterd4f77962006-02-24 13:04:22 -08002715 current->flags &= ~(PF_MEMALLOC | PF_SWAPWRITE);
KOSAKI Motohiro76ca5422010-03-05 13:41:47 -08002716 lockdep_clear_current_reclaim_state();
Rik van Riela79311c2009-01-06 14:40:01 -08002717 return sc.nr_reclaimed >= nr_pages;
Christoph Lameter9eeff232006-01-18 17:42:31 -08002718}
Andrew Morton179e9632006-03-22 00:08:18 -08002719
2720int zone_reclaim(struct zone *zone, gfp_t gfp_mask, unsigned int order)
2721{
Andrew Morton179e9632006-03-22 00:08:18 -08002722 int node_id;
David Rientjesd773ed62007-10-16 23:26:01 -07002723 int ret;
Andrew Morton179e9632006-03-22 00:08:18 -08002724
2725 /*
Christoph Lameter0ff38492006-09-25 23:31:52 -07002726 * Zone reclaim reclaims unmapped file backed pages and
2727 * slab pages if we are over the defined limits.
Christoph Lameter34aa1332006-06-30 01:55:37 -07002728 *
Christoph Lameter96146342006-07-03 00:24:13 -07002729 * A small portion of unmapped file backed pages is needed for
2730 * file I/O otherwise pages read by file I/O will be immediately
2731 * thrown out if the zone is overallocated. So we do not reclaim
2732 * if less than a specified percentage of the zone is used by
2733 * unmapped file backed pages.
Andrew Morton179e9632006-03-22 00:08:18 -08002734 */
Mel Gorman90afa5d2009-06-16 15:33:20 -07002735 if (zone_pagecache_reclaimable(zone) <= zone->min_unmapped_pages &&
2736 zone_page_state(zone, NR_SLAB_RECLAIMABLE) <= zone->min_slab_pages)
Mel Gormanfa5e0842009-06-16 15:33:22 -07002737 return ZONE_RECLAIM_FULL;
Andrew Morton179e9632006-03-22 00:08:18 -08002738
KOSAKI Motohiro93e4a892010-03-05 13:41:55 -08002739 if (zone->all_unreclaimable)
Mel Gormanfa5e0842009-06-16 15:33:22 -07002740 return ZONE_RECLAIM_FULL;
David Rientjesd773ed62007-10-16 23:26:01 -07002741
Andrew Morton179e9632006-03-22 00:08:18 -08002742 /*
David Rientjesd773ed62007-10-16 23:26:01 -07002743 * Do not scan if the allocation should not be delayed.
Andrew Morton179e9632006-03-22 00:08:18 -08002744 */
David Rientjesd773ed62007-10-16 23:26:01 -07002745 if (!(gfp_mask & __GFP_WAIT) || (current->flags & PF_MEMALLOC))
Mel Gormanfa5e0842009-06-16 15:33:22 -07002746 return ZONE_RECLAIM_NOSCAN;
Andrew Morton179e9632006-03-22 00:08:18 -08002747
2748 /*
2749 * Only run zone reclaim on the local zone or on zones that do not
2750 * have associated processors. This will favor the local processor
2751 * over remote processors and spread off node memory allocations
2752 * as wide as possible.
2753 */
Christoph Lameter89fa3022006-09-25 23:31:55 -07002754 node_id = zone_to_nid(zone);
Christoph Lameter37c07082007-10-16 01:25:36 -07002755 if (node_state(node_id, N_CPU) && node_id != numa_node_id())
Mel Gormanfa5e0842009-06-16 15:33:22 -07002756 return ZONE_RECLAIM_NOSCAN;
David Rientjesd773ed62007-10-16 23:26:01 -07002757
2758 if (zone_test_and_set_flag(zone, ZONE_RECLAIM_LOCKED))
Mel Gormanfa5e0842009-06-16 15:33:22 -07002759 return ZONE_RECLAIM_NOSCAN;
2760
David Rientjesd773ed62007-10-16 23:26:01 -07002761 ret = __zone_reclaim(zone, gfp_mask, order);
2762 zone_clear_flag(zone, ZONE_RECLAIM_LOCKED);
2763
Mel Gorman24cf725182009-06-16 15:33:23 -07002764 if (!ret)
2765 count_vm_event(PGSCAN_ZONE_RECLAIM_FAILED);
2766
David Rientjesd773ed62007-10-16 23:26:01 -07002767 return ret;
Andrew Morton179e9632006-03-22 00:08:18 -08002768}
Christoph Lameter9eeff232006-01-18 17:42:31 -08002769#endif
Lee Schermerhorn894bc312008-10-18 20:26:39 -07002770
Lee Schermerhorn894bc312008-10-18 20:26:39 -07002771/*
2772 * page_evictable - test whether a page is evictable
2773 * @page: the page to test
2774 * @vma: the VMA in which the page is or will be mapped, may be NULL
2775 *
2776 * Test whether page is evictable--i.e., should be placed on active/inactive
Nick Pigginb291f002008-10-18 20:26:44 -07002777 * lists vs unevictable list. The vma argument is !NULL when called from the
2778 * fault path to determine how to instantate a new page.
Lee Schermerhorn894bc312008-10-18 20:26:39 -07002779 *
2780 * Reasons page might not be evictable:
Lee Schermerhornba9ddf42008-10-18 20:26:42 -07002781 * (1) page's mapping marked unevictable
Nick Pigginb291f002008-10-18 20:26:44 -07002782 * (2) page is part of an mlocked VMA
Lee Schermerhornba9ddf42008-10-18 20:26:42 -07002783 *
Lee Schermerhorn894bc312008-10-18 20:26:39 -07002784 */
2785int page_evictable(struct page *page, struct vm_area_struct *vma)
2786{
2787
Lee Schermerhornba9ddf42008-10-18 20:26:42 -07002788 if (mapping_unevictable(page_mapping(page)))
2789 return 0;
2790
Nick Pigginb291f002008-10-18 20:26:44 -07002791 if (PageMlocked(page) || (vma && is_mlocked_vma(vma, page)))
2792 return 0;
Lee Schermerhorn894bc312008-10-18 20:26:39 -07002793
2794 return 1;
2795}
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07002796
2797/**
2798 * check_move_unevictable_page - check page for evictability and move to appropriate zone lru list
2799 * @page: page to check evictability and move to appropriate lru list
2800 * @zone: zone page is in
2801 *
2802 * Checks a page for evictability and moves the page to the appropriate
2803 * zone lru list.
2804 *
2805 * Restrictions: zone->lru_lock must be held, page must be on LRU and must
2806 * have PageUnevictable set.
2807 */
2808static void check_move_unevictable_page(struct page *page, struct zone *zone)
2809{
2810 VM_BUG_ON(PageActive(page));
2811
2812retry:
2813 ClearPageUnevictable(page);
2814 if (page_evictable(page, NULL)) {
Johannes Weiner401a8e12009-09-21 17:02:58 -07002815 enum lru_list l = page_lru_base_type(page);
Lee Schermerhornaf936a12008-10-18 20:26:53 -07002816
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07002817 __dec_zone_state(zone, NR_UNEVICTABLE);
2818 list_move(&page->lru, &zone->lru[l].list);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002819 mem_cgroup_move_lists(page, LRU_UNEVICTABLE, l);
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07002820 __inc_zone_state(zone, NR_INACTIVE_ANON + l);
2821 __count_vm_event(UNEVICTABLE_PGRESCUED);
2822 } else {
2823 /*
2824 * rotate unevictable list
2825 */
2826 SetPageUnevictable(page);
2827 list_move(&page->lru, &zone->lru[LRU_UNEVICTABLE].list);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002828 mem_cgroup_rotate_lru_list(page, LRU_UNEVICTABLE);
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07002829 if (page_evictable(page, NULL))
2830 goto retry;
2831 }
2832}
2833
2834/**
2835 * scan_mapping_unevictable_pages - scan an address space for evictable pages
2836 * @mapping: struct address_space to scan for evictable pages
2837 *
2838 * Scan all pages in mapping. Check unevictable pages for
2839 * evictability and move them to the appropriate zone lru list.
2840 */
2841void scan_mapping_unevictable_pages(struct address_space *mapping)
2842{
2843 pgoff_t next = 0;
2844 pgoff_t end = (i_size_read(mapping->host) + PAGE_CACHE_SIZE - 1) >>
2845 PAGE_CACHE_SHIFT;
2846 struct zone *zone;
2847 struct pagevec pvec;
2848
2849 if (mapping->nrpages == 0)
2850 return;
2851
2852 pagevec_init(&pvec, 0);
2853 while (next < end &&
2854 pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
2855 int i;
2856 int pg_scanned = 0;
2857
2858 zone = NULL;
2859
2860 for (i = 0; i < pagevec_count(&pvec); i++) {
2861 struct page *page = pvec.pages[i];
2862 pgoff_t page_index = page->index;
2863 struct zone *pagezone = page_zone(page);
2864
2865 pg_scanned++;
2866 if (page_index > next)
2867 next = page_index;
2868 next++;
2869
2870 if (pagezone != zone) {
2871 if (zone)
2872 spin_unlock_irq(&zone->lru_lock);
2873 zone = pagezone;
2874 spin_lock_irq(&zone->lru_lock);
2875 }
2876
2877 if (PageLRU(page) && PageUnevictable(page))
2878 check_move_unevictable_page(page, zone);
2879 }
2880 if (zone)
2881 spin_unlock_irq(&zone->lru_lock);
2882 pagevec_release(&pvec);
2883
2884 count_vm_events(UNEVICTABLE_PGSCANNED, pg_scanned);
2885 }
2886
2887}
Lee Schermerhornaf936a12008-10-18 20:26:53 -07002888
2889/**
2890 * scan_zone_unevictable_pages - check unevictable list for evictable pages
2891 * @zone - zone of which to scan the unevictable list
2892 *
2893 * Scan @zone's unevictable LRU lists to check for pages that have become
2894 * evictable. Move those that have to @zone's inactive list where they
2895 * become candidates for reclaim, unless shrink_inactive_zone() decides
2896 * to reactivate them. Pages that are still unevictable are rotated
2897 * back onto @zone's unevictable list.
2898 */
2899#define SCAN_UNEVICTABLE_BATCH_SIZE 16UL /* arbitrary lock hold batch size */
KOSAKI Motohiro14b90b22009-01-06 14:39:45 -08002900static void scan_zone_unevictable_pages(struct zone *zone)
Lee Schermerhornaf936a12008-10-18 20:26:53 -07002901{
2902 struct list_head *l_unevictable = &zone->lru[LRU_UNEVICTABLE].list;
2903 unsigned long scan;
2904 unsigned long nr_to_scan = zone_page_state(zone, NR_UNEVICTABLE);
2905
2906 while (nr_to_scan > 0) {
2907 unsigned long batch_size = min(nr_to_scan,
2908 SCAN_UNEVICTABLE_BATCH_SIZE);
2909
2910 spin_lock_irq(&zone->lru_lock);
2911 for (scan = 0; scan < batch_size; scan++) {
2912 struct page *page = lru_to_page(l_unevictable);
2913
2914 if (!trylock_page(page))
2915 continue;
2916
2917 prefetchw_prev_lru_page(page, l_unevictable, flags);
2918
2919 if (likely(PageLRU(page) && PageUnevictable(page)))
2920 check_move_unevictable_page(page, zone);
2921
2922 unlock_page(page);
2923 }
2924 spin_unlock_irq(&zone->lru_lock);
2925
2926 nr_to_scan -= batch_size;
2927 }
2928}
2929
2930
2931/**
2932 * scan_all_zones_unevictable_pages - scan all unevictable lists for evictable pages
2933 *
2934 * A really big hammer: scan all zones' unevictable LRU lists to check for
2935 * pages that have become evictable. Move those back to the zones'
2936 * inactive list where they become candidates for reclaim.
2937 * This occurs when, e.g., we have unswappable pages on the unevictable lists,
2938 * and we add swap to the system. As such, it runs in the context of a task
2939 * that has possibly/probably made some previously unevictable pages
2940 * evictable.
2941 */
KOSAKI Motohiroff301532009-01-06 14:39:44 -08002942static void scan_all_zones_unevictable_pages(void)
Lee Schermerhornaf936a12008-10-18 20:26:53 -07002943{
2944 struct zone *zone;
2945
2946 for_each_zone(zone) {
2947 scan_zone_unevictable_pages(zone);
2948 }
2949}
2950
2951/*
2952 * scan_unevictable_pages [vm] sysctl handler. On demand re-scan of
2953 * all nodes' unevictable lists for evictable pages
2954 */
2955unsigned long scan_unevictable_pages;
2956
2957int scan_unevictable_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07002958 void __user *buffer,
Lee Schermerhornaf936a12008-10-18 20:26:53 -07002959 size_t *length, loff_t *ppos)
2960{
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07002961 proc_doulongvec_minmax(table, write, buffer, length, ppos);
Lee Schermerhornaf936a12008-10-18 20:26:53 -07002962
2963 if (write && *(unsigned long *)table->data)
2964 scan_all_zones_unevictable_pages();
2965
2966 scan_unevictable_pages = 0;
2967 return 0;
2968}
2969
2970/*
2971 * per node 'scan_unevictable_pages' attribute. On demand re-scan of
2972 * a specified node's per zone unevictable lists for evictable pages.
2973 */
2974
2975static ssize_t read_scan_unevictable_node(struct sys_device *dev,
2976 struct sysdev_attribute *attr,
2977 char *buf)
2978{
2979 return sprintf(buf, "0\n"); /* always zero; should fit... */
2980}
2981
2982static ssize_t write_scan_unevictable_node(struct sys_device *dev,
2983 struct sysdev_attribute *attr,
2984 const char *buf, size_t count)
2985{
2986 struct zone *node_zones = NODE_DATA(dev->id)->node_zones;
2987 struct zone *zone;
2988 unsigned long res;
2989 unsigned long req = strict_strtoul(buf, 10, &res);
2990
2991 if (!req)
2992 return 1; /* zero is no-op */
2993
2994 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
2995 if (!populated_zone(zone))
2996 continue;
2997 scan_zone_unevictable_pages(zone);
2998 }
2999 return 1;
3000}
3001
3002
3003static SYSDEV_ATTR(scan_unevictable_pages, S_IRUGO | S_IWUSR,
3004 read_scan_unevictable_node,
3005 write_scan_unevictable_node);
3006
3007int scan_unevictable_register_node(struct node *node)
3008{
3009 return sysdev_create_file(&node->sysdev, &attr_scan_unevictable_pages);
3010}
3011
3012void scan_unevictable_unregister_node(struct node *node)
3013{
3014 sysdev_remove_file(&node->sysdev, &attr_scan_unevictable_pages);
3015}
3016