]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/mm/vmscan.c | |
3 | * | |
4 | * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds | |
5 | * | |
6 | * Swap reorganised 29.12.95, Stephen Tweedie. | |
7 | * kswapd added: 7.1.96 sct | |
8 | * Removed kswapd_ctl limits, and swap out as many pages as needed | |
9 | * to bring the system back to freepages.high: 2.4.97, Rik van Riel. | |
10 | * Zone aware kswapd started 02/00, Kanoj Sarcar (kanoj@sgi.com). | |
11 | * Multiqueue VM started 5.8.00, Rik van Riel. | |
12 | */ | |
13 | ||
14 | #include <linux/mm.h> | |
15 | #include <linux/module.h> | |
16 | #include <linux/slab.h> | |
17 | #include <linux/kernel_stat.h> | |
18 | #include <linux/swap.h> | |
19 | #include <linux/pagemap.h> | |
20 | #include <linux/init.h> | |
21 | #include <linux/highmem.h> | |
e129b5c2 | 22 | #include <linux/vmstat.h> |
1da177e4 LT |
23 | #include <linux/file.h> |
24 | #include <linux/writeback.h> | |
25 | #include <linux/blkdev.h> | |
26 | #include <linux/buffer_head.h> /* for try_to_release_page(), | |
27 | buffer_heads_over_limit */ | |
28 | #include <linux/mm_inline.h> | |
29 | #include <linux/pagevec.h> | |
30 | #include <linux/backing-dev.h> | |
31 | #include <linux/rmap.h> | |
32 | #include <linux/topology.h> | |
33 | #include <linux/cpu.h> | |
34 | #include <linux/cpuset.h> | |
35 | #include <linux/notifier.h> | |
36 | #include <linux/rwsem.h> | |
248a0301 | 37 | #include <linux/delay.h> |
3218ae14 | 38 | #include <linux/kthread.h> |
7dfb7103 | 39 | #include <linux/freezer.h> |
66e1707b | 40 | #include <linux/memcontrol.h> |
873b4771 | 41 | #include <linux/delayacct.h> |
af936a16 | 42 | #include <linux/sysctl.h> |
1da177e4 LT |
43 | |
44 | #include <asm/tlbflush.h> | |
45 | #include <asm/div64.h> | |
46 | ||
47 | #include <linux/swapops.h> | |
48 | ||
0f8053a5 NP |
49 | #include "internal.h" |
50 | ||
1da177e4 | 51 | struct scan_control { |
1da177e4 LT |
52 | /* Incremented by the number of inactive pages that were scanned */ |
53 | unsigned long nr_scanned; | |
54 | ||
a79311c1 RR |
55 | /* Number of pages freed so far during a call to shrink_zones() */ |
56 | unsigned long nr_reclaimed; | |
57 | ||
1da177e4 | 58 | /* This context's GFP mask */ |
6daa0e28 | 59 | gfp_t gfp_mask; |
1da177e4 LT |
60 | |
61 | int may_writepage; | |
62 | ||
f1fd1067 CL |
63 | /* Can pages be swapped as part of reclaim? */ |
64 | int may_swap; | |
65 | ||
1da177e4 LT |
66 | /* This context's SWAP_CLUSTER_MAX. If freeing memory for |
67 | * suspend, we effectively ignore SWAP_CLUSTER_MAX. | |
68 | * In this context, it doesn't matter that we scan the | |
69 | * whole list at once. */ | |
70 | int swap_cluster_max; | |
d6277db4 RW |
71 | |
72 | int swappiness; | |
408d8544 NP |
73 | |
74 | int all_unreclaimable; | |
5ad333eb AW |
75 | |
76 | int order; | |
66e1707b BS |
77 | |
78 | /* Which cgroup do we reclaim from */ | |
79 | struct mem_cgroup *mem_cgroup; | |
80 | ||
81 | /* Pluggable isolate pages callback */ | |
82 | unsigned long (*isolate_pages)(unsigned long nr, struct list_head *dst, | |
83 | unsigned long *scanned, int order, int mode, | |
84 | struct zone *z, struct mem_cgroup *mem_cont, | |
4f98a2fe | 85 | int active, int file); |
1da177e4 LT |
86 | }; |
87 | ||
1da177e4 LT |
88 | #define lru_to_page(_head) (list_entry((_head)->prev, struct page, lru)) |
89 | ||
90 | #ifdef ARCH_HAS_PREFETCH | |
91 | #define prefetch_prev_lru_page(_page, _base, _field) \ | |
92 | do { \ | |
93 | if ((_page)->lru.prev != _base) { \ | |
94 | struct page *prev; \ | |
95 | \ | |
96 | prev = lru_to_page(&(_page->lru)); \ | |
97 | prefetch(&prev->_field); \ | |
98 | } \ | |
99 | } while (0) | |
100 | #else | |
101 | #define prefetch_prev_lru_page(_page, _base, _field) do { } while (0) | |
102 | #endif | |
103 | ||
104 | #ifdef ARCH_HAS_PREFETCHW | |
105 | #define prefetchw_prev_lru_page(_page, _base, _field) \ | |
106 | do { \ | |
107 | if ((_page)->lru.prev != _base) { \ | |
108 | struct page *prev; \ | |
109 | \ | |
110 | prev = lru_to_page(&(_page->lru)); \ | |
111 | prefetchw(&prev->_field); \ | |
112 | } \ | |
113 | } while (0) | |
114 | #else | |
115 | #define prefetchw_prev_lru_page(_page, _base, _field) do { } while (0) | |
116 | #endif | |
117 | ||
118 | /* | |
119 | * From 0 .. 100. Higher means more swappy. | |
120 | */ | |
121 | int vm_swappiness = 60; | |
bd1e22b8 | 122 | long vm_total_pages; /* The total number of pages which the VM controls */ |
1da177e4 LT |
123 | |
124 | static LIST_HEAD(shrinker_list); | |
125 | static DECLARE_RWSEM(shrinker_rwsem); | |
126 | ||
00f0b825 | 127 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR |
91a45470 KH |
128 | #define scan_global_lru(sc) (!(sc)->mem_cgroup) |
129 | #else | |
130 | #define scan_global_lru(sc) (1) | |
131 | #endif | |
132 | ||
1da177e4 LT |
133 | /* |
134 | * Add a shrinker callback to be called from the vm | |
135 | */ | |
8e1f936b | 136 | void register_shrinker(struct shrinker *shrinker) |
1da177e4 | 137 | { |
8e1f936b RR |
138 | shrinker->nr = 0; |
139 | down_write(&shrinker_rwsem); | |
140 | list_add_tail(&shrinker->list, &shrinker_list); | |
141 | up_write(&shrinker_rwsem); | |
1da177e4 | 142 | } |
8e1f936b | 143 | EXPORT_SYMBOL(register_shrinker); |
1da177e4 LT |
144 | |
145 | /* | |
146 | * Remove one | |
147 | */ | |
8e1f936b | 148 | void unregister_shrinker(struct shrinker *shrinker) |
1da177e4 LT |
149 | { |
150 | down_write(&shrinker_rwsem); | |
151 | list_del(&shrinker->list); | |
152 | up_write(&shrinker_rwsem); | |
1da177e4 | 153 | } |
8e1f936b | 154 | EXPORT_SYMBOL(unregister_shrinker); |
1da177e4 LT |
155 | |
156 | #define SHRINK_BATCH 128 | |
157 | /* | |
158 | * Call the shrink functions to age shrinkable caches | |
159 | * | |
160 | * Here we assume it costs one seek to replace a lru page and that it also | |
161 | * takes a seek to recreate a cache object. With this in mind we age equal | |
162 | * percentages of the lru and ageable caches. This should balance the seeks | |
163 | * generated by these structures. | |
164 | * | |
183ff22b | 165 | * If the vm encountered mapped pages on the LRU it increase the pressure on |
1da177e4 LT |
166 | * slab to avoid swapping. |
167 | * | |
168 | * We do weird things to avoid (scanned*seeks*entries) overflowing 32 bits. | |
169 | * | |
170 | * `lru_pages' represents the number of on-LRU pages in all the zones which | |
171 | * are eligible for the caller's allocation attempt. It is used for balancing | |
172 | * slab reclaim versus page reclaim. | |
b15e0905 | 173 | * |
174 | * Returns the number of slab objects which we shrunk. | |
1da177e4 | 175 | */ |
69e05944 AM |
176 | unsigned long shrink_slab(unsigned long scanned, gfp_t gfp_mask, |
177 | unsigned long lru_pages) | |
1da177e4 LT |
178 | { |
179 | struct shrinker *shrinker; | |
69e05944 | 180 | unsigned long ret = 0; |
1da177e4 LT |
181 | |
182 | if (scanned == 0) | |
183 | scanned = SWAP_CLUSTER_MAX; | |
184 | ||
185 | if (!down_read_trylock(&shrinker_rwsem)) | |
b15e0905 | 186 | return 1; /* Assume we'll be able to shrink next time */ |
1da177e4 LT |
187 | |
188 | list_for_each_entry(shrinker, &shrinker_list, list) { | |
189 | unsigned long long delta; | |
190 | unsigned long total_scan; | |
8e1f936b | 191 | unsigned long max_pass = (*shrinker->shrink)(0, gfp_mask); |
1da177e4 LT |
192 | |
193 | delta = (4 * scanned) / shrinker->seeks; | |
ea164d73 | 194 | delta *= max_pass; |
1da177e4 LT |
195 | do_div(delta, lru_pages + 1); |
196 | shrinker->nr += delta; | |
ea164d73 AA |
197 | if (shrinker->nr < 0) { |
198 | printk(KERN_ERR "%s: nr=%ld\n", | |
d40cee24 | 199 | __func__, shrinker->nr); |
ea164d73 AA |
200 | shrinker->nr = max_pass; |
201 | } | |
202 | ||
203 | /* | |
204 | * Avoid risking looping forever due to too large nr value: | |
205 | * never try to free more than twice the estimate number of | |
206 | * freeable entries. | |
207 | */ | |
208 | if (shrinker->nr > max_pass * 2) | |
209 | shrinker->nr = max_pass * 2; | |
1da177e4 LT |
210 | |
211 | total_scan = shrinker->nr; | |
212 | shrinker->nr = 0; | |
213 | ||
214 | while (total_scan >= SHRINK_BATCH) { | |
215 | long this_scan = SHRINK_BATCH; | |
216 | int shrink_ret; | |
b15e0905 | 217 | int nr_before; |
1da177e4 | 218 | |
8e1f936b RR |
219 | nr_before = (*shrinker->shrink)(0, gfp_mask); |
220 | shrink_ret = (*shrinker->shrink)(this_scan, gfp_mask); | |
1da177e4 LT |
221 | if (shrink_ret == -1) |
222 | break; | |
b15e0905 | 223 | if (shrink_ret < nr_before) |
224 | ret += nr_before - shrink_ret; | |
f8891e5e | 225 | count_vm_events(SLABS_SCANNED, this_scan); |
1da177e4 LT |
226 | total_scan -= this_scan; |
227 | ||
228 | cond_resched(); | |
229 | } | |
230 | ||
231 | shrinker->nr += total_scan; | |
232 | } | |
233 | up_read(&shrinker_rwsem); | |
b15e0905 | 234 | return ret; |
1da177e4 LT |
235 | } |
236 | ||
237 | /* Called without lock on whether page is mapped, so answer is unstable */ | |
238 | static inline int page_mapping_inuse(struct page *page) | |
239 | { | |
240 | struct address_space *mapping; | |
241 | ||
242 | /* Page is in somebody's page tables. */ | |
243 | if (page_mapped(page)) | |
244 | return 1; | |
245 | ||
246 | /* Be more reluctant to reclaim swapcache than pagecache */ | |
247 | if (PageSwapCache(page)) | |
248 | return 1; | |
249 | ||
250 | mapping = page_mapping(page); | |
251 | if (!mapping) | |
252 | return 0; | |
253 | ||
254 | /* File is mmap'd by somebody? */ | |
255 | return mapping_mapped(mapping); | |
256 | } | |
257 | ||
258 | static inline int is_page_cache_freeable(struct page *page) | |
259 | { | |
260 | return page_count(page) - !!PagePrivate(page) == 2; | |
261 | } | |
262 | ||
263 | static int may_write_to_queue(struct backing_dev_info *bdi) | |
264 | { | |
930d9152 | 265 | if (current->flags & PF_SWAPWRITE) |
1da177e4 LT |
266 | return 1; |
267 | if (!bdi_write_congested(bdi)) | |
268 | return 1; | |
269 | if (bdi == current->backing_dev_info) | |
270 | return 1; | |
271 | return 0; | |
272 | } | |
273 | ||
274 | /* | |
275 | * We detected a synchronous write error writing a page out. Probably | |
276 | * -ENOSPC. We need to propagate that into the address_space for a subsequent | |
277 | * fsync(), msync() or close(). | |
278 | * | |
279 | * The tricky part is that after writepage we cannot touch the mapping: nothing | |
280 | * prevents it from being freed up. But we have a ref on the page and once | |
281 | * that page is locked, the mapping is pinned. | |
282 | * | |
283 | * We're allowed to run sleeping lock_page() here because we know the caller has | |
284 | * __GFP_FS. | |
285 | */ | |
286 | static void handle_write_error(struct address_space *mapping, | |
287 | struct page *page, int error) | |
288 | { | |
289 | lock_page(page); | |
3e9f45bd GC |
290 | if (page_mapping(page) == mapping) |
291 | mapping_set_error(mapping, error); | |
1da177e4 LT |
292 | unlock_page(page); |
293 | } | |
294 | ||
c661b078 AW |
295 | /* Request for sync pageout. */ |
296 | enum pageout_io { | |
297 | PAGEOUT_IO_ASYNC, | |
298 | PAGEOUT_IO_SYNC, | |
299 | }; | |
300 | ||
04e62a29 CL |
301 | /* possible outcome of pageout() */ |
302 | typedef enum { | |
303 | /* failed to write page out, page is locked */ | |
304 | PAGE_KEEP, | |
305 | /* move page to the active list, page is locked */ | |
306 | PAGE_ACTIVATE, | |
307 | /* page has been sent to the disk successfully, page is unlocked */ | |
308 | PAGE_SUCCESS, | |
309 | /* page is clean and locked */ | |
310 | PAGE_CLEAN, | |
311 | } pageout_t; | |
312 | ||
1da177e4 | 313 | /* |
1742f19f AM |
314 | * pageout is called by shrink_page_list() for each dirty page. |
315 | * Calls ->writepage(). | |
1da177e4 | 316 | */ |
c661b078 AW |
317 | static pageout_t pageout(struct page *page, struct address_space *mapping, |
318 | enum pageout_io sync_writeback) | |
1da177e4 LT |
319 | { |
320 | /* | |
321 | * If the page is dirty, only perform writeback if that write | |
322 | * will be non-blocking. To prevent this allocation from being | |
323 | * stalled by pagecache activity. But note that there may be | |
324 | * stalls if we need to run get_block(). We could test | |
325 | * PagePrivate for that. | |
326 | * | |
327 | * If this process is currently in generic_file_write() against | |
328 | * this page's queue, we can perform writeback even if that | |
329 | * will block. | |
330 | * | |
331 | * If the page is swapcache, write it back even if that would | |
332 | * block, for some throttling. This happens by accident, because | |
333 | * swap_backing_dev_info is bust: it doesn't reflect the | |
334 | * congestion state of the swapdevs. Easy to fix, if needed. | |
335 | * See swapfile.c:page_queue_congested(). | |
336 | */ | |
337 | if (!is_page_cache_freeable(page)) | |
338 | return PAGE_KEEP; | |
339 | if (!mapping) { | |
340 | /* | |
341 | * Some data journaling orphaned pages can have | |
342 | * page->mapping == NULL while being dirty with clean buffers. | |
343 | */ | |
323aca6c | 344 | if (PagePrivate(page)) { |
1da177e4 LT |
345 | if (try_to_free_buffers(page)) { |
346 | ClearPageDirty(page); | |
d40cee24 | 347 | printk("%s: orphaned page\n", __func__); |
1da177e4 LT |
348 | return PAGE_CLEAN; |
349 | } | |
350 | } | |
351 | return PAGE_KEEP; | |
352 | } | |
353 | if (mapping->a_ops->writepage == NULL) | |
354 | return PAGE_ACTIVATE; | |
355 | if (!may_write_to_queue(mapping->backing_dev_info)) | |
356 | return PAGE_KEEP; | |
357 | ||
358 | if (clear_page_dirty_for_io(page)) { | |
359 | int res; | |
360 | struct writeback_control wbc = { | |
361 | .sync_mode = WB_SYNC_NONE, | |
362 | .nr_to_write = SWAP_CLUSTER_MAX, | |
111ebb6e OH |
363 | .range_start = 0, |
364 | .range_end = LLONG_MAX, | |
1da177e4 LT |
365 | .nonblocking = 1, |
366 | .for_reclaim = 1, | |
367 | }; | |
368 | ||
369 | SetPageReclaim(page); | |
370 | res = mapping->a_ops->writepage(page, &wbc); | |
371 | if (res < 0) | |
372 | handle_write_error(mapping, page, res); | |
994fc28c | 373 | if (res == AOP_WRITEPAGE_ACTIVATE) { |
1da177e4 LT |
374 | ClearPageReclaim(page); |
375 | return PAGE_ACTIVATE; | |
376 | } | |
c661b078 AW |
377 | |
378 | /* | |
379 | * Wait on writeback if requested to. This happens when | |
380 | * direct reclaiming a large contiguous area and the | |
381 | * first attempt to free a range of pages fails. | |
382 | */ | |
383 | if (PageWriteback(page) && sync_writeback == PAGEOUT_IO_SYNC) | |
384 | wait_on_page_writeback(page); | |
385 | ||
1da177e4 LT |
386 | if (!PageWriteback(page)) { |
387 | /* synchronous write or broken a_ops? */ | |
388 | ClearPageReclaim(page); | |
389 | } | |
e129b5c2 | 390 | inc_zone_page_state(page, NR_VMSCAN_WRITE); |
1da177e4 LT |
391 | return PAGE_SUCCESS; |
392 | } | |
393 | ||
394 | return PAGE_CLEAN; | |
395 | } | |
396 | ||
a649fd92 | 397 | /* |
e286781d NP |
398 | * Same as remove_mapping, but if the page is removed from the mapping, it |
399 | * gets returned with a refcount of 0. | |
a649fd92 | 400 | */ |
e286781d | 401 | static int __remove_mapping(struct address_space *mapping, struct page *page) |
49d2e9cc | 402 | { |
28e4d965 NP |
403 | BUG_ON(!PageLocked(page)); |
404 | BUG_ON(mapping != page_mapping(page)); | |
49d2e9cc | 405 | |
19fd6231 | 406 | spin_lock_irq(&mapping->tree_lock); |
49d2e9cc | 407 | /* |
0fd0e6b0 NP |
408 | * The non racy check for a busy page. |
409 | * | |
410 | * Must be careful with the order of the tests. When someone has | |
411 | * a ref to the page, it may be possible that they dirty it then | |
412 | * drop the reference. So if PageDirty is tested before page_count | |
413 | * here, then the following race may occur: | |
414 | * | |
415 | * get_user_pages(&page); | |
416 | * [user mapping goes away] | |
417 | * write_to(page); | |
418 | * !PageDirty(page) [good] | |
419 | * SetPageDirty(page); | |
420 | * put_page(page); | |
421 | * !page_count(page) [good, discard it] | |
422 | * | |
423 | * [oops, our write_to data is lost] | |
424 | * | |
425 | * Reversing the order of the tests ensures such a situation cannot | |
426 | * escape unnoticed. The smp_rmb is needed to ensure the page->flags | |
427 | * load is not satisfied before that of page->_count. | |
428 | * | |
429 | * Note that if SetPageDirty is always performed via set_page_dirty, | |
430 | * and thus under tree_lock, then this ordering is not required. | |
49d2e9cc | 431 | */ |
e286781d | 432 | if (!page_freeze_refs(page, 2)) |
49d2e9cc | 433 | goto cannot_free; |
e286781d NP |
434 | /* note: atomic_cmpxchg in page_freeze_refs provides the smp_rmb */ |
435 | if (unlikely(PageDirty(page))) { | |
436 | page_unfreeze_refs(page, 2); | |
49d2e9cc | 437 | goto cannot_free; |
e286781d | 438 | } |
49d2e9cc CL |
439 | |
440 | if (PageSwapCache(page)) { | |
441 | swp_entry_t swap = { .val = page_private(page) }; | |
442 | __delete_from_swap_cache(page); | |
19fd6231 | 443 | spin_unlock_irq(&mapping->tree_lock); |
49d2e9cc | 444 | swap_free(swap); |
e286781d NP |
445 | } else { |
446 | __remove_from_page_cache(page); | |
19fd6231 | 447 | spin_unlock_irq(&mapping->tree_lock); |
49d2e9cc CL |
448 | } |
449 | ||
49d2e9cc CL |
450 | return 1; |
451 | ||
452 | cannot_free: | |
19fd6231 | 453 | spin_unlock_irq(&mapping->tree_lock); |
49d2e9cc CL |
454 | return 0; |
455 | } | |
456 | ||
e286781d NP |
457 | /* |
458 | * Attempt to detach a locked page from its ->mapping. If it is dirty or if | |
459 | * someone else has a ref on the page, abort and return 0. If it was | |
460 | * successfully detached, return 1. Assumes the caller has a single ref on | |
461 | * this page. | |
462 | */ | |
463 | int remove_mapping(struct address_space *mapping, struct page *page) | |
464 | { | |
465 | if (__remove_mapping(mapping, page)) { | |
466 | /* | |
467 | * Unfreezing the refcount with 1 rather than 2 effectively | |
468 | * drops the pagecache ref for us without requiring another | |
469 | * atomic operation. | |
470 | */ | |
471 | page_unfreeze_refs(page, 1); | |
472 | return 1; | |
473 | } | |
474 | return 0; | |
475 | } | |
476 | ||
894bc310 LS |
477 | /** |
478 | * putback_lru_page - put previously isolated page onto appropriate LRU list | |
479 | * @page: page to be put back to appropriate lru list | |
480 | * | |
481 | * Add previously isolated @page to appropriate LRU list. | |
482 | * Page may still be unevictable for other reasons. | |
483 | * | |
484 | * lru_lock must not be held, interrupts must be enabled. | |
485 | */ | |
486 | #ifdef CONFIG_UNEVICTABLE_LRU | |
487 | void putback_lru_page(struct page *page) | |
488 | { | |
489 | int lru; | |
490 | int active = !!TestClearPageActive(page); | |
bbfd28ee | 491 | int was_unevictable = PageUnevictable(page); |
894bc310 LS |
492 | |
493 | VM_BUG_ON(PageLRU(page)); | |
494 | ||
495 | redo: | |
496 | ClearPageUnevictable(page); | |
497 | ||
498 | if (page_evictable(page, NULL)) { | |
499 | /* | |
500 | * For evictable pages, we can use the cache. | |
501 | * In event of a race, worst case is we end up with an | |
502 | * unevictable page on [in]active list. | |
503 | * We know how to handle that. | |
504 | */ | |
505 | lru = active + page_is_file_cache(page); | |
506 | lru_cache_add_lru(page, lru); | |
507 | } else { | |
508 | /* | |
509 | * Put unevictable pages directly on zone's unevictable | |
510 | * list. | |
511 | */ | |
512 | lru = LRU_UNEVICTABLE; | |
513 | add_page_to_unevictable_list(page); | |
514 | } | |
894bc310 LS |
515 | |
516 | /* | |
517 | * page's status can change while we move it among lru. If an evictable | |
518 | * page is on unevictable list, it never be freed. To avoid that, | |
519 | * check after we added it to the list, again. | |
520 | */ | |
521 | if (lru == LRU_UNEVICTABLE && page_evictable(page, NULL)) { | |
522 | if (!isolate_lru_page(page)) { | |
523 | put_page(page); | |
524 | goto redo; | |
525 | } | |
526 | /* This means someone else dropped this page from LRU | |
527 | * So, it will be freed or putback to LRU again. There is | |
528 | * nothing to do here. | |
529 | */ | |
530 | } | |
531 | ||
bbfd28ee LS |
532 | if (was_unevictable && lru != LRU_UNEVICTABLE) |
533 | count_vm_event(UNEVICTABLE_PGRESCUED); | |
534 | else if (!was_unevictable && lru == LRU_UNEVICTABLE) | |
535 | count_vm_event(UNEVICTABLE_PGCULLED); | |
536 | ||
894bc310 LS |
537 | put_page(page); /* drop ref from isolate */ |
538 | } | |
539 | ||
540 | #else /* CONFIG_UNEVICTABLE_LRU */ | |
541 | ||
542 | void putback_lru_page(struct page *page) | |
543 | { | |
544 | int lru; | |
545 | VM_BUG_ON(PageLRU(page)); | |
546 | ||
547 | lru = !!TestClearPageActive(page) + page_is_file_cache(page); | |
548 | lru_cache_add_lru(page, lru); | |
894bc310 LS |
549 | put_page(page); |
550 | } | |
551 | #endif /* CONFIG_UNEVICTABLE_LRU */ | |
552 | ||
553 | ||
1da177e4 | 554 | /* |
1742f19f | 555 | * shrink_page_list() returns the number of reclaimed pages |
1da177e4 | 556 | */ |
1742f19f | 557 | static unsigned long shrink_page_list(struct list_head *page_list, |
c661b078 AW |
558 | struct scan_control *sc, |
559 | enum pageout_io sync_writeback) | |
1da177e4 LT |
560 | { |
561 | LIST_HEAD(ret_pages); | |
562 | struct pagevec freed_pvec; | |
563 | int pgactivate = 0; | |
05ff5137 | 564 | unsigned long nr_reclaimed = 0; |
1da177e4 LT |
565 | |
566 | cond_resched(); | |
567 | ||
568 | pagevec_init(&freed_pvec, 1); | |
569 | while (!list_empty(page_list)) { | |
570 | struct address_space *mapping; | |
571 | struct page *page; | |
572 | int may_enter_fs; | |
573 | int referenced; | |
574 | ||
575 | cond_resched(); | |
576 | ||
577 | page = lru_to_page(page_list); | |
578 | list_del(&page->lru); | |
579 | ||
529ae9aa | 580 | if (!trylock_page(page)) |
1da177e4 LT |
581 | goto keep; |
582 | ||
725d704e | 583 | VM_BUG_ON(PageActive(page)); |
1da177e4 LT |
584 | |
585 | sc->nr_scanned++; | |
80e43426 | 586 | |
b291f000 NP |
587 | if (unlikely(!page_evictable(page, NULL))) |
588 | goto cull_mlocked; | |
894bc310 | 589 | |
80e43426 CL |
590 | if (!sc->may_swap && page_mapped(page)) |
591 | goto keep_locked; | |
592 | ||
1da177e4 LT |
593 | /* Double the slab pressure for mapped and swapcache pages */ |
594 | if (page_mapped(page) || PageSwapCache(page)) | |
595 | sc->nr_scanned++; | |
596 | ||
c661b078 AW |
597 | may_enter_fs = (sc->gfp_mask & __GFP_FS) || |
598 | (PageSwapCache(page) && (sc->gfp_mask & __GFP_IO)); | |
599 | ||
600 | if (PageWriteback(page)) { | |
601 | /* | |
602 | * Synchronous reclaim is performed in two passes, | |
603 | * first an asynchronous pass over the list to | |
604 | * start parallel writeback, and a second synchronous | |
605 | * pass to wait for the IO to complete. Wait here | |
606 | * for any page for which writeback has already | |
607 | * started. | |
608 | */ | |
609 | if (sync_writeback == PAGEOUT_IO_SYNC && may_enter_fs) | |
610 | wait_on_page_writeback(page); | |
4dd4b920 | 611 | else |
c661b078 AW |
612 | goto keep_locked; |
613 | } | |
1da177e4 | 614 | |
bed7161a | 615 | referenced = page_referenced(page, 1, sc->mem_cgroup); |
1da177e4 | 616 | /* In active use or really unfreeable? Activate it. */ |
5ad333eb AW |
617 | if (sc->order <= PAGE_ALLOC_COSTLY_ORDER && |
618 | referenced && page_mapping_inuse(page)) | |
1da177e4 LT |
619 | goto activate_locked; |
620 | ||
1da177e4 LT |
621 | /* |
622 | * Anonymous process memory has backing store? | |
623 | * Try to allocate it some swap space here. | |
624 | */ | |
b291f000 | 625 | if (PageAnon(page) && !PageSwapCache(page)) { |
63eb6b93 HD |
626 | if (!(sc->gfp_mask & __GFP_IO)) |
627 | goto keep_locked; | |
ac47b003 | 628 | if (!add_to_swap(page)) |
1da177e4 | 629 | goto activate_locked; |
63eb6b93 | 630 | may_enter_fs = 1; |
b291f000 | 631 | } |
1da177e4 LT |
632 | |
633 | mapping = page_mapping(page); | |
1da177e4 LT |
634 | |
635 | /* | |
636 | * The page is mapped into the page tables of one or more | |
637 | * processes. Try to unmap it here. | |
638 | */ | |
639 | if (page_mapped(page) && mapping) { | |
a48d07af | 640 | switch (try_to_unmap(page, 0)) { |
1da177e4 LT |
641 | case SWAP_FAIL: |
642 | goto activate_locked; | |
643 | case SWAP_AGAIN: | |
644 | goto keep_locked; | |
b291f000 NP |
645 | case SWAP_MLOCK: |
646 | goto cull_mlocked; | |
1da177e4 LT |
647 | case SWAP_SUCCESS: |
648 | ; /* try to free the page below */ | |
649 | } | |
650 | } | |
651 | ||
652 | if (PageDirty(page)) { | |
5ad333eb | 653 | if (sc->order <= PAGE_ALLOC_COSTLY_ORDER && referenced) |
1da177e4 | 654 | goto keep_locked; |
4dd4b920 | 655 | if (!may_enter_fs) |
1da177e4 | 656 | goto keep_locked; |
52a8363e | 657 | if (!sc->may_writepage) |
1da177e4 LT |
658 | goto keep_locked; |
659 | ||
660 | /* Page is dirty, try to write it out here */ | |
c661b078 | 661 | switch (pageout(page, mapping, sync_writeback)) { |
1da177e4 LT |
662 | case PAGE_KEEP: |
663 | goto keep_locked; | |
664 | case PAGE_ACTIVATE: | |
665 | goto activate_locked; | |
666 | case PAGE_SUCCESS: | |
4dd4b920 | 667 | if (PageWriteback(page) || PageDirty(page)) |
1da177e4 LT |
668 | goto keep; |
669 | /* | |
670 | * A synchronous write - probably a ramdisk. Go | |
671 | * ahead and try to reclaim the page. | |
672 | */ | |
529ae9aa | 673 | if (!trylock_page(page)) |
1da177e4 LT |
674 | goto keep; |
675 | if (PageDirty(page) || PageWriteback(page)) | |
676 | goto keep_locked; | |
677 | mapping = page_mapping(page); | |
678 | case PAGE_CLEAN: | |
679 | ; /* try to free the page below */ | |
680 | } | |
681 | } | |
682 | ||
683 | /* | |
684 | * If the page has buffers, try to free the buffer mappings | |
685 | * associated with this page. If we succeed we try to free | |
686 | * the page as well. | |
687 | * | |
688 | * We do this even if the page is PageDirty(). | |
689 | * try_to_release_page() does not perform I/O, but it is | |
690 | * possible for a page to have PageDirty set, but it is actually | |
691 | * clean (all its buffers are clean). This happens if the | |
692 | * buffers were written out directly, with submit_bh(). ext3 | |
894bc310 | 693 | * will do this, as well as the blockdev mapping. |
1da177e4 LT |
694 | * try_to_release_page() will discover that cleanness and will |
695 | * drop the buffers and mark the page clean - it can be freed. | |
696 | * | |
697 | * Rarely, pages can have buffers and no ->mapping. These are | |
698 | * the pages which were not successfully invalidated in | |
699 | * truncate_complete_page(). We try to drop those buffers here | |
700 | * and if that worked, and the page is no longer mapped into | |
701 | * process address space (page_count == 1) it can be freed. | |
702 | * Otherwise, leave the page on the LRU so it is swappable. | |
703 | */ | |
704 | if (PagePrivate(page)) { | |
705 | if (!try_to_release_page(page, sc->gfp_mask)) | |
706 | goto activate_locked; | |
e286781d NP |
707 | if (!mapping && page_count(page) == 1) { |
708 | unlock_page(page); | |
709 | if (put_page_testzero(page)) | |
710 | goto free_it; | |
711 | else { | |
712 | /* | |
713 | * rare race with speculative reference. | |
714 | * the speculative reference will free | |
715 | * this page shortly, so we may | |
716 | * increment nr_reclaimed here (and | |
717 | * leave it off the LRU). | |
718 | */ | |
719 | nr_reclaimed++; | |
720 | continue; | |
721 | } | |
722 | } | |
1da177e4 LT |
723 | } |
724 | ||
e286781d | 725 | if (!mapping || !__remove_mapping(mapping, page)) |
49d2e9cc | 726 | goto keep_locked; |
1da177e4 | 727 | |
a978d6f5 NP |
728 | /* |
729 | * At this point, we have no other references and there is | |
730 | * no way to pick any more up (removed from LRU, removed | |
731 | * from pagecache). Can use non-atomic bitops now (and | |
732 | * we obviously don't have to worry about waking up a process | |
733 | * waiting on the page lock, because there are no references. | |
734 | */ | |
735 | __clear_page_locked(page); | |
e286781d | 736 | free_it: |
05ff5137 | 737 | nr_reclaimed++; |
e286781d NP |
738 | if (!pagevec_add(&freed_pvec, page)) { |
739 | __pagevec_free(&freed_pvec); | |
740 | pagevec_reinit(&freed_pvec); | |
741 | } | |
1da177e4 LT |
742 | continue; |
743 | ||
b291f000 | 744 | cull_mlocked: |
63d6c5ad HD |
745 | if (PageSwapCache(page)) |
746 | try_to_free_swap(page); | |
b291f000 NP |
747 | unlock_page(page); |
748 | putback_lru_page(page); | |
749 | continue; | |
750 | ||
1da177e4 | 751 | activate_locked: |
68a22394 RR |
752 | /* Not a candidate for swapping, so reclaim swap space. */ |
753 | if (PageSwapCache(page) && vm_swap_full()) | |
a2c43eed | 754 | try_to_free_swap(page); |
894bc310 | 755 | VM_BUG_ON(PageActive(page)); |
1da177e4 LT |
756 | SetPageActive(page); |
757 | pgactivate++; | |
758 | keep_locked: | |
759 | unlock_page(page); | |
760 | keep: | |
761 | list_add(&page->lru, &ret_pages); | |
b291f000 | 762 | VM_BUG_ON(PageLRU(page) || PageUnevictable(page)); |
1da177e4 LT |
763 | } |
764 | list_splice(&ret_pages, page_list); | |
765 | if (pagevec_count(&freed_pvec)) | |
e286781d | 766 | __pagevec_free(&freed_pvec); |
f8891e5e | 767 | count_vm_events(PGACTIVATE, pgactivate); |
05ff5137 | 768 | return nr_reclaimed; |
1da177e4 LT |
769 | } |
770 | ||
5ad333eb AW |
771 | /* LRU Isolation modes. */ |
772 | #define ISOLATE_INACTIVE 0 /* Isolate inactive pages. */ | |
773 | #define ISOLATE_ACTIVE 1 /* Isolate active pages. */ | |
774 | #define ISOLATE_BOTH 2 /* Isolate both active and inactive pages. */ | |
775 | ||
776 | /* | |
777 | * Attempt to remove the specified page from its LRU. Only take this page | |
778 | * if it is of the appropriate PageActive status. Pages which are being | |
779 | * freed elsewhere are also ignored. | |
780 | * | |
781 | * page: page to consider | |
782 | * mode: one of the LRU isolation modes defined above | |
783 | * | |
784 | * returns 0 on success, -ve errno on failure. | |
785 | */ | |
4f98a2fe | 786 | int __isolate_lru_page(struct page *page, int mode, int file) |
5ad333eb AW |
787 | { |
788 | int ret = -EINVAL; | |
789 | ||
790 | /* Only take pages on the LRU. */ | |
791 | if (!PageLRU(page)) | |
792 | return ret; | |
793 | ||
794 | /* | |
795 | * When checking the active state, we need to be sure we are | |
796 | * dealing with comparible boolean values. Take the logical not | |
797 | * of each. | |
798 | */ | |
799 | if (mode != ISOLATE_BOTH && (!PageActive(page) != !mode)) | |
800 | return ret; | |
801 | ||
4f98a2fe RR |
802 | if (mode != ISOLATE_BOTH && (!page_is_file_cache(page) != !file)) |
803 | return ret; | |
804 | ||
894bc310 LS |
805 | /* |
806 | * When this function is being called for lumpy reclaim, we | |
807 | * initially look into all LRU pages, active, inactive and | |
808 | * unevictable; only give shrink_page_list evictable pages. | |
809 | */ | |
810 | if (PageUnevictable(page)) | |
811 | return ret; | |
812 | ||
5ad333eb | 813 | ret = -EBUSY; |
08e552c6 | 814 | |
5ad333eb AW |
815 | if (likely(get_page_unless_zero(page))) { |
816 | /* | |
817 | * Be careful not to clear PageLRU until after we're | |
818 | * sure the page is not being freed elsewhere -- the | |
819 | * page release code relies on it. | |
820 | */ | |
821 | ClearPageLRU(page); | |
822 | ret = 0; | |
08e552c6 | 823 | mem_cgroup_del_lru(page); |
5ad333eb AW |
824 | } |
825 | ||
826 | return ret; | |
827 | } | |
828 | ||
1da177e4 LT |
829 | /* |
830 | * zone->lru_lock is heavily contended. Some of the functions that | |
831 | * shrink the lists perform better by taking out a batch of pages | |
832 | * and working on them outside the LRU lock. | |
833 | * | |
834 | * For pagecache intensive workloads, this function is the hottest | |
835 | * spot in the kernel (apart from copy_*_user functions). | |
836 | * | |
837 | * Appropriate locks must be held before calling this function. | |
838 | * | |
839 | * @nr_to_scan: The number of pages to look through on the list. | |
840 | * @src: The LRU list to pull pages off. | |
841 | * @dst: The temp list to put pages on to. | |
842 | * @scanned: The number of pages that were scanned. | |
5ad333eb AW |
843 | * @order: The caller's attempted allocation order |
844 | * @mode: One of the LRU isolation modes | |
4f98a2fe | 845 | * @file: True [1] if isolating file [!anon] pages |
1da177e4 LT |
846 | * |
847 | * returns how many pages were moved onto *@dst. | |
848 | */ | |
69e05944 AM |
849 | static unsigned long isolate_lru_pages(unsigned long nr_to_scan, |
850 | struct list_head *src, struct list_head *dst, | |
4f98a2fe | 851 | unsigned long *scanned, int order, int mode, int file) |
1da177e4 | 852 | { |
69e05944 | 853 | unsigned long nr_taken = 0; |
c9b02d97 | 854 | unsigned long scan; |
1da177e4 | 855 | |
c9b02d97 | 856 | for (scan = 0; scan < nr_to_scan && !list_empty(src); scan++) { |
5ad333eb AW |
857 | struct page *page; |
858 | unsigned long pfn; | |
859 | unsigned long end_pfn; | |
860 | unsigned long page_pfn; | |
861 | int zone_id; | |
862 | ||
1da177e4 LT |
863 | page = lru_to_page(src); |
864 | prefetchw_prev_lru_page(page, src, flags); | |
865 | ||
725d704e | 866 | VM_BUG_ON(!PageLRU(page)); |
8d438f96 | 867 | |
4f98a2fe | 868 | switch (__isolate_lru_page(page, mode, file)) { |
5ad333eb AW |
869 | case 0: |
870 | list_move(&page->lru, dst); | |
7c8ee9a8 | 871 | nr_taken++; |
5ad333eb AW |
872 | break; |
873 | ||
874 | case -EBUSY: | |
875 | /* else it is being freed elsewhere */ | |
876 | list_move(&page->lru, src); | |
877 | continue; | |
46453a6e | 878 | |
5ad333eb AW |
879 | default: |
880 | BUG(); | |
881 | } | |
882 | ||
883 | if (!order) | |
884 | continue; | |
885 | ||
886 | /* | |
887 | * Attempt to take all pages in the order aligned region | |
888 | * surrounding the tag page. Only take those pages of | |
889 | * the same active state as that tag page. We may safely | |
890 | * round the target page pfn down to the requested order | |
891 | * as the mem_map is guarenteed valid out to MAX_ORDER, | |
892 | * where that page is in a different zone we will detect | |
893 | * it from its zone id and abort this block scan. | |
894 | */ | |
895 | zone_id = page_zone_id(page); | |
896 | page_pfn = page_to_pfn(page); | |
897 | pfn = page_pfn & ~((1 << order) - 1); | |
898 | end_pfn = pfn + (1 << order); | |
899 | for (; pfn < end_pfn; pfn++) { | |
900 | struct page *cursor_page; | |
901 | ||
902 | /* The target page is in the block, ignore it. */ | |
903 | if (unlikely(pfn == page_pfn)) | |
904 | continue; | |
905 | ||
906 | /* Avoid holes within the zone. */ | |
907 | if (unlikely(!pfn_valid_within(pfn))) | |
908 | break; | |
909 | ||
910 | cursor_page = pfn_to_page(pfn); | |
4f98a2fe | 911 | |
5ad333eb AW |
912 | /* Check that we have not crossed a zone boundary. */ |
913 | if (unlikely(page_zone_id(cursor_page) != zone_id)) | |
914 | continue; | |
4f98a2fe | 915 | switch (__isolate_lru_page(cursor_page, mode, file)) { |
5ad333eb AW |
916 | case 0: |
917 | list_move(&cursor_page->lru, dst); | |
918 | nr_taken++; | |
919 | scan++; | |
920 | break; | |
921 | ||
922 | case -EBUSY: | |
923 | /* else it is being freed elsewhere */ | |
924 | list_move(&cursor_page->lru, src); | |
925 | default: | |
894bc310 | 926 | break; /* ! on LRU or wrong list */ |
5ad333eb AW |
927 | } |
928 | } | |
1da177e4 LT |
929 | } |
930 | ||
931 | *scanned = scan; | |
932 | return nr_taken; | |
933 | } | |
934 | ||
66e1707b BS |
935 | static unsigned long isolate_pages_global(unsigned long nr, |
936 | struct list_head *dst, | |
937 | unsigned long *scanned, int order, | |
938 | int mode, struct zone *z, | |
939 | struct mem_cgroup *mem_cont, | |
4f98a2fe | 940 | int active, int file) |
66e1707b | 941 | { |
4f98a2fe | 942 | int lru = LRU_BASE; |
66e1707b | 943 | if (active) |
4f98a2fe RR |
944 | lru += LRU_ACTIVE; |
945 | if (file) | |
946 | lru += LRU_FILE; | |
947 | return isolate_lru_pages(nr, &z->lru[lru].list, dst, scanned, order, | |
948 | mode, !!file); | |
66e1707b BS |
949 | } |
950 | ||
5ad333eb AW |
951 | /* |
952 | * clear_active_flags() is a helper for shrink_active_list(), clearing | |
953 | * any active bits from the pages in the list. | |
954 | */ | |
4f98a2fe RR |
955 | static unsigned long clear_active_flags(struct list_head *page_list, |
956 | unsigned int *count) | |
5ad333eb AW |
957 | { |
958 | int nr_active = 0; | |
4f98a2fe | 959 | int lru; |
5ad333eb AW |
960 | struct page *page; |
961 | ||
4f98a2fe RR |
962 | list_for_each_entry(page, page_list, lru) { |
963 | lru = page_is_file_cache(page); | |
5ad333eb | 964 | if (PageActive(page)) { |
4f98a2fe | 965 | lru += LRU_ACTIVE; |
5ad333eb AW |
966 | ClearPageActive(page); |
967 | nr_active++; | |
968 | } | |
4f98a2fe RR |
969 | count[lru]++; |
970 | } | |
5ad333eb AW |
971 | |
972 | return nr_active; | |
973 | } | |
974 | ||
62695a84 NP |
975 | /** |
976 | * isolate_lru_page - tries to isolate a page from its LRU list | |
977 | * @page: page to isolate from its LRU list | |
978 | * | |
979 | * Isolates a @page from an LRU list, clears PageLRU and adjusts the | |
980 | * vmstat statistic corresponding to whatever LRU list the page was on. | |
981 | * | |
982 | * Returns 0 if the page was removed from an LRU list. | |
983 | * Returns -EBUSY if the page was not on an LRU list. | |
984 | * | |
985 | * The returned page will have PageLRU() cleared. If it was found on | |
894bc310 LS |
986 | * the active list, it will have PageActive set. If it was found on |
987 | * the unevictable list, it will have the PageUnevictable bit set. That flag | |
988 | * may need to be cleared by the caller before letting the page go. | |
62695a84 NP |
989 | * |
990 | * The vmstat statistic corresponding to the list on which the page was | |
991 | * found will be decremented. | |
992 | * | |
993 | * Restrictions: | |
994 | * (1) Must be called with an elevated refcount on the page. This is a | |
995 | * fundamentnal difference from isolate_lru_pages (which is called | |
996 | * without a stable reference). | |
997 | * (2) the lru_lock must not be held. | |
998 | * (3) interrupts must be enabled. | |
999 | */ | |
1000 | int isolate_lru_page(struct page *page) | |
1001 | { | |
1002 | int ret = -EBUSY; | |
1003 | ||
1004 | if (PageLRU(page)) { | |
1005 | struct zone *zone = page_zone(page); | |
1006 | ||
1007 | spin_lock_irq(&zone->lru_lock); | |
1008 | if (PageLRU(page) && get_page_unless_zero(page)) { | |
894bc310 | 1009 | int lru = page_lru(page); |
62695a84 NP |
1010 | ret = 0; |
1011 | ClearPageLRU(page); | |
4f98a2fe | 1012 | |
4f98a2fe | 1013 | del_page_from_lru_list(zone, page, lru); |
62695a84 NP |
1014 | } |
1015 | spin_unlock_irq(&zone->lru_lock); | |
1016 | } | |
1017 | return ret; | |
1018 | } | |
1019 | ||
1da177e4 | 1020 | /* |
1742f19f AM |
1021 | * shrink_inactive_list() is a helper for shrink_zone(). It returns the number |
1022 | * of reclaimed pages | |
1da177e4 | 1023 | */ |
1742f19f | 1024 | static unsigned long shrink_inactive_list(unsigned long max_scan, |
33c120ed RR |
1025 | struct zone *zone, struct scan_control *sc, |
1026 | int priority, int file) | |
1da177e4 LT |
1027 | { |
1028 | LIST_HEAD(page_list); | |
1029 | struct pagevec pvec; | |
69e05944 | 1030 | unsigned long nr_scanned = 0; |
05ff5137 | 1031 | unsigned long nr_reclaimed = 0; |
1da177e4 LT |
1032 | |
1033 | pagevec_init(&pvec, 1); | |
1034 | ||
1035 | lru_add_drain(); | |
1036 | spin_lock_irq(&zone->lru_lock); | |
69e05944 | 1037 | do { |
1da177e4 | 1038 | struct page *page; |
69e05944 AM |
1039 | unsigned long nr_taken; |
1040 | unsigned long nr_scan; | |
1041 | unsigned long nr_freed; | |
5ad333eb | 1042 | unsigned long nr_active; |
4f98a2fe | 1043 | unsigned int count[NR_LRU_LISTS] = { 0, }; |
33c120ed RR |
1044 | int mode = ISOLATE_INACTIVE; |
1045 | ||
1046 | /* | |
1047 | * If we need a large contiguous chunk of memory, or have | |
1048 | * trouble getting a small set of contiguous pages, we | |
1049 | * will reclaim both active and inactive pages. | |
1050 | * | |
1051 | * We use the same threshold as pageout congestion_wait below. | |
1052 | */ | |
1053 | if (sc->order > PAGE_ALLOC_COSTLY_ORDER) | |
1054 | mode = ISOLATE_BOTH; | |
1055 | else if (sc->order && priority < DEF_PRIORITY - 2) | |
1056 | mode = ISOLATE_BOTH; | |
1da177e4 | 1057 | |
66e1707b | 1058 | nr_taken = sc->isolate_pages(sc->swap_cluster_max, |
4f98a2fe RR |
1059 | &page_list, &nr_scan, sc->order, mode, |
1060 | zone, sc->mem_cgroup, 0, file); | |
1061 | nr_active = clear_active_flags(&page_list, count); | |
e9187bdc | 1062 | __count_vm_events(PGDEACTIVATE, nr_active); |
5ad333eb | 1063 | |
4f98a2fe RR |
1064 | __mod_zone_page_state(zone, NR_ACTIVE_FILE, |
1065 | -count[LRU_ACTIVE_FILE]); | |
1066 | __mod_zone_page_state(zone, NR_INACTIVE_FILE, | |
1067 | -count[LRU_INACTIVE_FILE]); | |
1068 | __mod_zone_page_state(zone, NR_ACTIVE_ANON, | |
1069 | -count[LRU_ACTIVE_ANON]); | |
1070 | __mod_zone_page_state(zone, NR_INACTIVE_ANON, | |
1071 | -count[LRU_INACTIVE_ANON]); | |
1072 | ||
1073 | if (scan_global_lru(sc)) { | |
1cfb419b | 1074 | zone->pages_scanned += nr_scan; |
4f98a2fe RR |
1075 | zone->recent_scanned[0] += count[LRU_INACTIVE_ANON]; |
1076 | zone->recent_scanned[0] += count[LRU_ACTIVE_ANON]; | |
1077 | zone->recent_scanned[1] += count[LRU_INACTIVE_FILE]; | |
1078 | zone->recent_scanned[1] += count[LRU_ACTIVE_FILE]; | |
1079 | } | |
1da177e4 LT |
1080 | spin_unlock_irq(&zone->lru_lock); |
1081 | ||
69e05944 | 1082 | nr_scanned += nr_scan; |
c661b078 AW |
1083 | nr_freed = shrink_page_list(&page_list, sc, PAGEOUT_IO_ASYNC); |
1084 | ||
1085 | /* | |
1086 | * If we are direct reclaiming for contiguous pages and we do | |
1087 | * not reclaim everything in the list, try again and wait | |
1088 | * for IO to complete. This will stall high-order allocations | |
1089 | * but that should be acceptable to the caller | |
1090 | */ | |
1091 | if (nr_freed < nr_taken && !current_is_kswapd() && | |
1092 | sc->order > PAGE_ALLOC_COSTLY_ORDER) { | |
1093 | congestion_wait(WRITE, HZ/10); | |
1094 | ||
1095 | /* | |
1096 | * The attempt at page out may have made some | |
1097 | * of the pages active, mark them inactive again. | |
1098 | */ | |
4f98a2fe | 1099 | nr_active = clear_active_flags(&page_list, count); |
c661b078 AW |
1100 | count_vm_events(PGDEACTIVATE, nr_active); |
1101 | ||
1102 | nr_freed += shrink_page_list(&page_list, sc, | |
1103 | PAGEOUT_IO_SYNC); | |
1104 | } | |
1105 | ||
05ff5137 | 1106 | nr_reclaimed += nr_freed; |
a74609fa NP |
1107 | local_irq_disable(); |
1108 | if (current_is_kswapd()) { | |
f8891e5e CL |
1109 | __count_zone_vm_events(PGSCAN_KSWAPD, zone, nr_scan); |
1110 | __count_vm_events(KSWAPD_STEAL, nr_freed); | |
1cfb419b | 1111 | } else if (scan_global_lru(sc)) |
f8891e5e | 1112 | __count_zone_vm_events(PGSCAN_DIRECT, zone, nr_scan); |
1cfb419b | 1113 | |
918d3f90 | 1114 | __count_zone_vm_events(PGSTEAL, zone, nr_freed); |
a74609fa | 1115 | |
fb8d14e1 WF |
1116 | if (nr_taken == 0) |
1117 | goto done; | |
1118 | ||
a74609fa | 1119 | spin_lock(&zone->lru_lock); |
1da177e4 LT |
1120 | /* |
1121 | * Put back any unfreeable pages. | |
1122 | */ | |
1123 | while (!list_empty(&page_list)) { | |
894bc310 | 1124 | int lru; |
1da177e4 | 1125 | page = lru_to_page(&page_list); |
725d704e | 1126 | VM_BUG_ON(PageLRU(page)); |
1da177e4 | 1127 | list_del(&page->lru); |
894bc310 LS |
1128 | if (unlikely(!page_evictable(page, NULL))) { |
1129 | spin_unlock_irq(&zone->lru_lock); | |
1130 | putback_lru_page(page); | |
1131 | spin_lock_irq(&zone->lru_lock); | |
1132 | continue; | |
1133 | } | |
1134 | SetPageLRU(page); | |
1135 | lru = page_lru(page); | |
1136 | add_page_to_lru_list(zone, page, lru); | |
4f98a2fe RR |
1137 | if (PageActive(page) && scan_global_lru(sc)) { |
1138 | int file = !!page_is_file_cache(page); | |
1139 | zone->recent_rotated[file]++; | |
1140 | } | |
1da177e4 LT |
1141 | if (!pagevec_add(&pvec, page)) { |
1142 | spin_unlock_irq(&zone->lru_lock); | |
1143 | __pagevec_release(&pvec); | |
1144 | spin_lock_irq(&zone->lru_lock); | |
1145 | } | |
1146 | } | |
69e05944 | 1147 | } while (nr_scanned < max_scan); |
fb8d14e1 | 1148 | spin_unlock(&zone->lru_lock); |
1da177e4 | 1149 | done: |
fb8d14e1 | 1150 | local_irq_enable(); |
1da177e4 | 1151 | pagevec_release(&pvec); |
05ff5137 | 1152 | return nr_reclaimed; |
1da177e4 LT |
1153 | } |
1154 | ||
3bb1a852 MB |
1155 | /* |
1156 | * We are about to scan this zone at a certain priority level. If that priority | |
1157 | * level is smaller (ie: more urgent) than the previous priority, then note | |
1158 | * that priority level within the zone. This is done so that when the next | |
1159 | * process comes in to scan this zone, it will immediately start out at this | |
1160 | * priority level rather than having to build up its own scanning priority. | |
1161 | * Here, this priority affects only the reclaim-mapped threshold. | |
1162 | */ | |
1163 | static inline void note_zone_scanning_priority(struct zone *zone, int priority) | |
1164 | { | |
1165 | if (priority < zone->prev_priority) | |
1166 | zone->prev_priority = priority; | |
1167 | } | |
1168 | ||
1da177e4 LT |
1169 | /* |
1170 | * This moves pages from the active list to the inactive list. | |
1171 | * | |
1172 | * We move them the other way if the page is referenced by one or more | |
1173 | * processes, from rmap. | |
1174 | * | |
1175 | * If the pages are mostly unmapped, the processing is fast and it is | |
1176 | * appropriate to hold zone->lru_lock across the whole operation. But if | |
1177 | * the pages are mapped, the processing is slow (page_referenced()) so we | |
1178 | * should drop zone->lru_lock around each page. It's impossible to balance | |
1179 | * this, so instead we remove the pages from the LRU while processing them. | |
1180 | * It is safe to rely on PG_active against the non-LRU pages in here because | |
1181 | * nobody will play with that bit on a non-LRU page. | |
1182 | * | |
1183 | * The downside is that we have to touch page->_count against each page. | |
1184 | * But we had to alter page->flags anyway. | |
1185 | */ | |
1cfb419b KH |
1186 | |
1187 | ||
1742f19f | 1188 | static void shrink_active_list(unsigned long nr_pages, struct zone *zone, |
4f98a2fe | 1189 | struct scan_control *sc, int priority, int file) |
1da177e4 | 1190 | { |
69e05944 | 1191 | unsigned long pgmoved; |
1da177e4 | 1192 | int pgdeactivate = 0; |
69e05944 | 1193 | unsigned long pgscanned; |
1da177e4 | 1194 | LIST_HEAD(l_hold); /* The pages which were snipped off */ |
b69408e8 | 1195 | LIST_HEAD(l_inactive); |
1da177e4 LT |
1196 | struct page *page; |
1197 | struct pagevec pvec; | |
4f98a2fe | 1198 | enum lru_list lru; |
1da177e4 LT |
1199 | |
1200 | lru_add_drain(); | |
1201 | spin_lock_irq(&zone->lru_lock); | |
66e1707b BS |
1202 | pgmoved = sc->isolate_pages(nr_pages, &l_hold, &pgscanned, sc->order, |
1203 | ISOLATE_ACTIVE, zone, | |
4f98a2fe | 1204 | sc->mem_cgroup, 1, file); |
1cfb419b KH |
1205 | /* |
1206 | * zone->pages_scanned is used for detect zone's oom | |
1207 | * mem_cgroup remembers nr_scan by itself. | |
1208 | */ | |
4f98a2fe | 1209 | if (scan_global_lru(sc)) { |
1cfb419b | 1210 | zone->pages_scanned += pgscanned; |
4f98a2fe RR |
1211 | zone->recent_scanned[!!file] += pgmoved; |
1212 | } | |
1cfb419b | 1213 | |
4f98a2fe RR |
1214 | if (file) |
1215 | __mod_zone_page_state(zone, NR_ACTIVE_FILE, -pgmoved); | |
1216 | else | |
1217 | __mod_zone_page_state(zone, NR_ACTIVE_ANON, -pgmoved); | |
1da177e4 LT |
1218 | spin_unlock_irq(&zone->lru_lock); |
1219 | ||
556adecb | 1220 | pgmoved = 0; |
1da177e4 LT |
1221 | while (!list_empty(&l_hold)) { |
1222 | cond_resched(); | |
1223 | page = lru_to_page(&l_hold); | |
1224 | list_del(&page->lru); | |
7e9cd484 | 1225 | |
894bc310 LS |
1226 | if (unlikely(!page_evictable(page, NULL))) { |
1227 | putback_lru_page(page); | |
1228 | continue; | |
1229 | } | |
1230 | ||
7e9cd484 RR |
1231 | /* page_referenced clears PageReferenced */ |
1232 | if (page_mapping_inuse(page) && | |
1233 | page_referenced(page, 0, sc->mem_cgroup)) | |
1234 | pgmoved++; | |
1235 | ||
1da177e4 LT |
1236 | list_add(&page->lru, &l_inactive); |
1237 | } | |
1238 | ||
b555749a AM |
1239 | /* |
1240 | * Move the pages to the [file or anon] inactive list. | |
1241 | */ | |
1242 | pagevec_init(&pvec, 1); | |
1243 | pgmoved = 0; | |
1244 | lru = LRU_BASE + file * LRU_FILE; | |
1245 | ||
2a1dc509 | 1246 | spin_lock_irq(&zone->lru_lock); |
556adecb | 1247 | /* |
7e9cd484 RR |
1248 | * Count referenced pages from currently used mappings as |
1249 | * rotated, even though they are moved to the inactive list. | |
1250 | * This helps balance scan pressure between file and anonymous | |
1251 | * pages in get_scan_ratio. | |
1252 | */ | |
077cbc58 KM |
1253 | if (scan_global_lru(sc)) |
1254 | zone->recent_rotated[!!file] += pgmoved; | |
556adecb | 1255 | |
1da177e4 LT |
1256 | while (!list_empty(&l_inactive)) { |
1257 | page = lru_to_page(&l_inactive); | |
1258 | prefetchw_prev_lru_page(page, &l_inactive, flags); | |
725d704e | 1259 | VM_BUG_ON(PageLRU(page)); |
8d438f96 | 1260 | SetPageLRU(page); |
725d704e | 1261 | VM_BUG_ON(!PageActive(page)); |
4c84cacf NP |
1262 | ClearPageActive(page); |
1263 | ||
4f98a2fe | 1264 | list_move(&page->lru, &zone->lru[lru].list); |
08e552c6 | 1265 | mem_cgroup_add_lru_list(page, lru); |
1da177e4 LT |
1266 | pgmoved++; |
1267 | if (!pagevec_add(&pvec, page)) { | |
4f98a2fe | 1268 | __mod_zone_page_state(zone, NR_LRU_BASE + lru, pgmoved); |
1da177e4 LT |
1269 | spin_unlock_irq(&zone->lru_lock); |
1270 | pgdeactivate += pgmoved; | |
1271 | pgmoved = 0; | |
1272 | if (buffer_heads_over_limit) | |
1273 | pagevec_strip(&pvec); | |
1274 | __pagevec_release(&pvec); | |
1275 | spin_lock_irq(&zone->lru_lock); | |
1276 | } | |
1277 | } | |
4f98a2fe | 1278 | __mod_zone_page_state(zone, NR_LRU_BASE + lru, pgmoved); |
1da177e4 LT |
1279 | pgdeactivate += pgmoved; |
1280 | if (buffer_heads_over_limit) { | |
1281 | spin_unlock_irq(&zone->lru_lock); | |
1282 | pagevec_strip(&pvec); | |
1283 | spin_lock_irq(&zone->lru_lock); | |
1284 | } | |
f8891e5e CL |
1285 | __count_zone_vm_events(PGREFILL, zone, pgscanned); |
1286 | __count_vm_events(PGDEACTIVATE, pgdeactivate); | |
1287 | spin_unlock_irq(&zone->lru_lock); | |
68a22394 RR |
1288 | if (vm_swap_full()) |
1289 | pagevec_swap_free(&pvec); | |
1da177e4 | 1290 | |
a74609fa | 1291 | pagevec_release(&pvec); |
1da177e4 LT |
1292 | } |
1293 | ||
f89eb90e KM |
1294 | /** |
1295 | * inactive_anon_is_low - check if anonymous pages need to be deactivated | |
1296 | * @zone: zone to check | |
1297 | * | |
1298 | * Returns true if the zone does not have enough inactive anon pages, | |
1299 | * meaning some active anon pages need to be deactivated. | |
1300 | */ | |
1301 | static int inactive_anon_is_low(struct zone *zone) | |
1302 | { | |
1303 | unsigned long active, inactive; | |
1304 | ||
1305 | active = zone_page_state(zone, NR_ACTIVE_ANON); | |
1306 | inactive = zone_page_state(zone, NR_INACTIVE_ANON); | |
1307 | ||
1308 | if (inactive * zone->inactive_ratio < active) | |
1309 | return 1; | |
1310 | ||
1311 | return 0; | |
1312 | } | |
1313 | ||
4f98a2fe | 1314 | static unsigned long shrink_list(enum lru_list lru, unsigned long nr_to_scan, |
b69408e8 CL |
1315 | struct zone *zone, struct scan_control *sc, int priority) |
1316 | { | |
4f98a2fe RR |
1317 | int file = is_file_lru(lru); |
1318 | ||
556adecb RR |
1319 | if (lru == LRU_ACTIVE_FILE) { |
1320 | shrink_active_list(nr_to_scan, zone, sc, priority, file); | |
1321 | return 0; | |
1322 | } | |
1323 | ||
1324 | if (lru == LRU_ACTIVE_ANON && | |
1325 | (!scan_global_lru(sc) || inactive_anon_is_low(zone))) { | |
4f98a2fe | 1326 | shrink_active_list(nr_to_scan, zone, sc, priority, file); |
b69408e8 CL |
1327 | return 0; |
1328 | } | |
33c120ed | 1329 | return shrink_inactive_list(nr_to_scan, zone, sc, priority, file); |
4f98a2fe RR |
1330 | } |
1331 | ||
1332 | /* | |
1333 | * Determine how aggressively the anon and file LRU lists should be | |
1334 | * scanned. The relative value of each set of LRU lists is determined | |
1335 | * by looking at the fraction of the pages scanned we did rotate back | |
1336 | * onto the active list instead of evict. | |
1337 | * | |
1338 | * percent[0] specifies how much pressure to put on ram/swap backed | |
1339 | * memory, while percent[1] determines pressure on the file LRUs. | |
1340 | */ | |
1341 | static void get_scan_ratio(struct zone *zone, struct scan_control *sc, | |
1342 | unsigned long *percent) | |
1343 | { | |
1344 | unsigned long anon, file, free; | |
1345 | unsigned long anon_prio, file_prio; | |
1346 | unsigned long ap, fp; | |
1347 | ||
4f98a2fe RR |
1348 | /* If we have no swap space, do not bother scanning anon pages. */ |
1349 | if (nr_swap_pages <= 0) { | |
1350 | percent[0] = 0; | |
1351 | percent[1] = 100; | |
1352 | return; | |
1353 | } | |
1354 | ||
b962716b HD |
1355 | anon = zone_page_state(zone, NR_ACTIVE_ANON) + |
1356 | zone_page_state(zone, NR_INACTIVE_ANON); | |
1357 | file = zone_page_state(zone, NR_ACTIVE_FILE) + | |
1358 | zone_page_state(zone, NR_INACTIVE_FILE); | |
1359 | free = zone_page_state(zone, NR_FREE_PAGES); | |
1360 | ||
4f98a2fe RR |
1361 | /* If we have very few page cache pages, force-scan anon pages. */ |
1362 | if (unlikely(file + free <= zone->pages_high)) { | |
1363 | percent[0] = 100; | |
1364 | percent[1] = 0; | |
1365 | return; | |
1366 | } | |
1367 | ||
1368 | /* | |
1369 | * OK, so we have swap space and a fair amount of page cache | |
1370 | * pages. We use the recently rotated / recently scanned | |
1371 | * ratios to determine how valuable each cache is. | |
1372 | * | |
1373 | * Because workloads change over time (and to avoid overflow) | |
1374 | * we keep these statistics as a floating average, which ends | |
1375 | * up weighing recent references more than old ones. | |
1376 | * | |
1377 | * anon in [0], file in [1] | |
1378 | */ | |
1379 | if (unlikely(zone->recent_scanned[0] > anon / 4)) { | |
1380 | spin_lock_irq(&zone->lru_lock); | |
1381 | zone->recent_scanned[0] /= 2; | |
1382 | zone->recent_rotated[0] /= 2; | |
1383 | spin_unlock_irq(&zone->lru_lock); | |
1384 | } | |
1385 | ||
1386 | if (unlikely(zone->recent_scanned[1] > file / 4)) { | |
1387 | spin_lock_irq(&zone->lru_lock); | |
1388 | zone->recent_scanned[1] /= 2; | |
1389 | zone->recent_rotated[1] /= 2; | |
1390 | spin_unlock_irq(&zone->lru_lock); | |
1391 | } | |
1392 | ||
1393 | /* | |
1394 | * With swappiness at 100, anonymous and file have the same priority. | |
1395 | * This scanning priority is essentially the inverse of IO cost. | |
1396 | */ | |
1397 | anon_prio = sc->swappiness; | |
1398 | file_prio = 200 - sc->swappiness; | |
1399 | ||
1400 | /* | |
00d8089c RR |
1401 | * The amount of pressure on anon vs file pages is inversely |
1402 | * proportional to the fraction of recently scanned pages on | |
1403 | * each list that were recently referenced and in active use. | |
4f98a2fe RR |
1404 | */ |
1405 | ap = (anon_prio + 1) * (zone->recent_scanned[0] + 1); | |
1406 | ap /= zone->recent_rotated[0] + 1; | |
1407 | ||
1408 | fp = (file_prio + 1) * (zone->recent_scanned[1] + 1); | |
1409 | fp /= zone->recent_rotated[1] + 1; | |
1410 | ||
1411 | /* Normalize to percentages */ | |
1412 | percent[0] = 100 * ap / (ap + fp + 1); | |
1413 | percent[1] = 100 - percent[0]; | |
b69408e8 CL |
1414 | } |
1415 | ||
4f98a2fe | 1416 | |
1da177e4 LT |
1417 | /* |
1418 | * This is a basic per-zone page freer. Used by both kswapd and direct reclaim. | |
1419 | */ | |
a79311c1 | 1420 | static void shrink_zone(int priority, struct zone *zone, |
05ff5137 | 1421 | struct scan_control *sc) |
1da177e4 | 1422 | { |
b69408e8 | 1423 | unsigned long nr[NR_LRU_LISTS]; |
8695949a | 1424 | unsigned long nr_to_scan; |
4f98a2fe | 1425 | unsigned long percent[2]; /* anon @ 0; file @ 1 */ |
b69408e8 | 1426 | enum lru_list l; |
01dbe5c9 KM |
1427 | unsigned long nr_reclaimed = sc->nr_reclaimed; |
1428 | unsigned long swap_cluster_max = sc->swap_cluster_max; | |
1da177e4 | 1429 | |
4f98a2fe RR |
1430 | get_scan_ratio(zone, sc, percent); |
1431 | ||
894bc310 | 1432 | for_each_evictable_lru(l) { |
4f98a2fe RR |
1433 | if (scan_global_lru(sc)) { |
1434 | int file = is_file_lru(l); | |
1435 | int scan; | |
e0f79b8f | 1436 | |
4f98a2fe RR |
1437 | scan = zone_page_state(zone, NR_LRU_BASE + l); |
1438 | if (priority) { | |
1439 | scan >>= priority; | |
1440 | scan = (scan * percent[file]) / 100; | |
1441 | } | |
e0f79b8f | 1442 | zone->lru[l].nr_scan += scan; |
b69408e8 | 1443 | nr[l] = zone->lru[l].nr_scan; |
01dbe5c9 | 1444 | if (nr[l] >= swap_cluster_max) |
b69408e8 CL |
1445 | zone->lru[l].nr_scan = 0; |
1446 | else | |
1447 | nr[l] = 0; | |
4f98a2fe RR |
1448 | } else { |
1449 | /* | |
1450 | * This reclaim occurs not because zone memory shortage | |
1451 | * but because memory controller hits its limit. | |
1452 | * Don't modify zone reclaim related data. | |
1453 | */ | |
1454 | nr[l] = mem_cgroup_calc_reclaim(sc->mem_cgroup, zone, | |
1455 | priority, l); | |
b69408e8 | 1456 | } |
1cfb419b | 1457 | } |
1da177e4 | 1458 | |
556adecb RR |
1459 | while (nr[LRU_INACTIVE_ANON] || nr[LRU_ACTIVE_FILE] || |
1460 | nr[LRU_INACTIVE_FILE]) { | |
894bc310 | 1461 | for_each_evictable_lru(l) { |
b69408e8 | 1462 | if (nr[l]) { |
01dbe5c9 | 1463 | nr_to_scan = min(nr[l], swap_cluster_max); |
b69408e8 | 1464 | nr[l] -= nr_to_scan; |
1da177e4 | 1465 | |
01dbe5c9 KM |
1466 | nr_reclaimed += shrink_list(l, nr_to_scan, |
1467 | zone, sc, priority); | |
b69408e8 | 1468 | } |
1da177e4 | 1469 | } |
a79311c1 RR |
1470 | /* |
1471 | * On large memory systems, scan >> priority can become | |
1472 | * really large. This is fine for the starting priority; | |
1473 | * we want to put equal scanning pressure on each zone. | |
1474 | * However, if the VM has a harder time of freeing pages, | |
1475 | * with multiple processes reclaiming pages, the total | |
1476 | * freeing target can get unreasonably large. | |
1477 | */ | |
01dbe5c9 | 1478 | if (nr_reclaimed > swap_cluster_max && |
a79311c1 RR |
1479 | priority < DEF_PRIORITY && !current_is_kswapd()) |
1480 | break; | |
1da177e4 LT |
1481 | } |
1482 | ||
01dbe5c9 KM |
1483 | sc->nr_reclaimed = nr_reclaimed; |
1484 | ||
556adecb RR |
1485 | /* |
1486 | * Even if we did not try to evict anon pages at all, we want to | |
1487 | * rebalance the anon lru active/inactive ratio. | |
1488 | */ | |
1489 | if (!scan_global_lru(sc) || inactive_anon_is_low(zone)) | |
1490 | shrink_active_list(SWAP_CLUSTER_MAX, zone, sc, priority, 0); | |
1491 | else if (!scan_global_lru(sc)) | |
1492 | shrink_active_list(SWAP_CLUSTER_MAX, zone, sc, priority, 0); | |
1493 | ||
232ea4d6 | 1494 | throttle_vm_writeout(sc->gfp_mask); |
1da177e4 LT |
1495 | } |
1496 | ||
1497 | /* | |
1498 | * This is the direct reclaim path, for page-allocating processes. We only | |
1499 | * try to reclaim pages from zones which will satisfy the caller's allocation | |
1500 | * request. | |
1501 | * | |
1502 | * We reclaim from a zone even if that zone is over pages_high. Because: | |
1503 | * a) The caller may be trying to free *extra* pages to satisfy a higher-order | |
1504 | * allocation or | |
1505 | * b) The zones may be over pages_high but they must go *over* pages_high to | |
1506 | * satisfy the `incremental min' zone defense algorithm. | |
1507 | * | |
1da177e4 LT |
1508 | * If a zone is deemed to be full of pinned pages then just give it a light |
1509 | * scan then give up on it. | |
1510 | */ | |
a79311c1 | 1511 | static void shrink_zones(int priority, struct zonelist *zonelist, |
05ff5137 | 1512 | struct scan_control *sc) |
1da177e4 | 1513 | { |
54a6eb5c | 1514 | enum zone_type high_zoneidx = gfp_zone(sc->gfp_mask); |
dd1a239f | 1515 | struct zoneref *z; |
54a6eb5c | 1516 | struct zone *zone; |
1cfb419b | 1517 | |
408d8544 | 1518 | sc->all_unreclaimable = 1; |
54a6eb5c | 1519 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
f3fe6512 | 1520 | if (!populated_zone(zone)) |
1da177e4 | 1521 | continue; |
1cfb419b KH |
1522 | /* |
1523 | * Take care memory controller reclaiming has small influence | |
1524 | * to global LRU. | |
1525 | */ | |
1526 | if (scan_global_lru(sc)) { | |
1527 | if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL)) | |
1528 | continue; | |
1529 | note_zone_scanning_priority(zone, priority); | |
1da177e4 | 1530 | |
1cfb419b KH |
1531 | if (zone_is_all_unreclaimable(zone) && |
1532 | priority != DEF_PRIORITY) | |
1533 | continue; /* Let kswapd poll it */ | |
1534 | sc->all_unreclaimable = 0; | |
1535 | } else { | |
1536 | /* | |
1537 | * Ignore cpuset limitation here. We just want to reduce | |
1538 | * # of used pages by us regardless of memory shortage. | |
1539 | */ | |
1540 | sc->all_unreclaimable = 0; | |
1541 | mem_cgroup_note_reclaim_priority(sc->mem_cgroup, | |
1542 | priority); | |
1543 | } | |
408d8544 | 1544 | |
a79311c1 | 1545 | shrink_zone(priority, zone, sc); |
1da177e4 LT |
1546 | } |
1547 | } | |
4f98a2fe | 1548 | |
1da177e4 LT |
1549 | /* |
1550 | * This is the main entry point to direct page reclaim. | |
1551 | * | |
1552 | * If a full scan of the inactive list fails to free enough memory then we | |
1553 | * are "out of memory" and something needs to be killed. | |
1554 | * | |
1555 | * If the caller is !__GFP_FS then the probability of a failure is reasonably | |
1556 | * high - the zone may be full of dirty or under-writeback pages, which this | |
1557 | * caller can't do much about. We kick pdflush and take explicit naps in the | |
1558 | * hope that some of these pages can be written. But if the allocating task | |
1559 | * holds filesystem locks which prevent writeout this might not work, and the | |
1560 | * allocation attempt will fail. | |
a41f24ea NA |
1561 | * |
1562 | * returns: 0, if no pages reclaimed | |
1563 | * else, the number of pages reclaimed | |
1da177e4 | 1564 | */ |
dac1d27b | 1565 | static unsigned long do_try_to_free_pages(struct zonelist *zonelist, |
dd1a239f | 1566 | struct scan_control *sc) |
1da177e4 LT |
1567 | { |
1568 | int priority; | |
c700be3d | 1569 | unsigned long ret = 0; |
69e05944 | 1570 | unsigned long total_scanned = 0; |
1da177e4 | 1571 | struct reclaim_state *reclaim_state = current->reclaim_state; |
1da177e4 | 1572 | unsigned long lru_pages = 0; |
dd1a239f | 1573 | struct zoneref *z; |
54a6eb5c | 1574 | struct zone *zone; |
dd1a239f | 1575 | enum zone_type high_zoneidx = gfp_zone(sc->gfp_mask); |
1da177e4 | 1576 | |
873b4771 KK |
1577 | delayacct_freepages_start(); |
1578 | ||
1cfb419b KH |
1579 | if (scan_global_lru(sc)) |
1580 | count_vm_event(ALLOCSTALL); | |
1581 | /* | |
1582 | * mem_cgroup will not do shrink_slab. | |
1583 | */ | |
1584 | if (scan_global_lru(sc)) { | |
54a6eb5c | 1585 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
1da177e4 | 1586 | |
1cfb419b KH |
1587 | if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL)) |
1588 | continue; | |
1da177e4 | 1589 | |
4f98a2fe | 1590 | lru_pages += zone_lru_pages(zone); |
1cfb419b | 1591 | } |
1da177e4 LT |
1592 | } |
1593 | ||
1594 | for (priority = DEF_PRIORITY; priority >= 0; priority--) { | |
66e1707b | 1595 | sc->nr_scanned = 0; |
f7b7fd8f RR |
1596 | if (!priority) |
1597 | disable_swap_token(); | |
a79311c1 | 1598 | shrink_zones(priority, zonelist, sc); |
66e1707b BS |
1599 | /* |
1600 | * Don't shrink slabs when reclaiming memory from | |
1601 | * over limit cgroups | |
1602 | */ | |
91a45470 | 1603 | if (scan_global_lru(sc)) { |
dd1a239f | 1604 | shrink_slab(sc->nr_scanned, sc->gfp_mask, lru_pages); |
91a45470 | 1605 | if (reclaim_state) { |
a79311c1 | 1606 | sc->nr_reclaimed += reclaim_state->reclaimed_slab; |
91a45470 KH |
1607 | reclaim_state->reclaimed_slab = 0; |
1608 | } | |
1da177e4 | 1609 | } |
66e1707b | 1610 | total_scanned += sc->nr_scanned; |
a79311c1 RR |
1611 | if (sc->nr_reclaimed >= sc->swap_cluster_max) { |
1612 | ret = sc->nr_reclaimed; | |
1da177e4 LT |
1613 | goto out; |
1614 | } | |
1615 | ||
1616 | /* | |
1617 | * Try to write back as many pages as we just scanned. This | |
1618 | * tends to cause slow streaming writers to write data to the | |
1619 | * disk smoothly, at the dirtying rate, which is nice. But | |
1620 | * that's undesirable in laptop mode, where we *want* lumpy | |
1621 | * writeout. So in laptop mode, write out the whole world. | |
1622 | */ | |
66e1707b BS |
1623 | if (total_scanned > sc->swap_cluster_max + |
1624 | sc->swap_cluster_max / 2) { | |
687a21ce | 1625 | wakeup_pdflush(laptop_mode ? 0 : total_scanned); |
66e1707b | 1626 | sc->may_writepage = 1; |
1da177e4 LT |
1627 | } |
1628 | ||
1629 | /* Take a nap, wait for some writeback to complete */ | |
4dd4b920 | 1630 | if (sc->nr_scanned && priority < DEF_PRIORITY - 2) |
3fcfab16 | 1631 | congestion_wait(WRITE, HZ/10); |
1da177e4 | 1632 | } |
87547ee9 | 1633 | /* top priority shrink_zones still had more to do? don't OOM, then */ |
91a45470 | 1634 | if (!sc->all_unreclaimable && scan_global_lru(sc)) |
a79311c1 | 1635 | ret = sc->nr_reclaimed; |
1da177e4 | 1636 | out: |
3bb1a852 MB |
1637 | /* |
1638 | * Now that we've scanned all the zones at this priority level, note | |
1639 | * that level within the zone so that the next thread which performs | |
1640 | * scanning of this zone will immediately start out at this priority | |
1641 | * level. This affects only the decision whether or not to bring | |
1642 | * mapped pages onto the inactive list. | |
1643 | */ | |
1644 | if (priority < 0) | |
1645 | priority = 0; | |
1da177e4 | 1646 | |
1cfb419b | 1647 | if (scan_global_lru(sc)) { |
54a6eb5c | 1648 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
1cfb419b KH |
1649 | |
1650 | if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL)) | |
1651 | continue; | |
1652 | ||
1653 | zone->prev_priority = priority; | |
1654 | } | |
1655 | } else | |
1656 | mem_cgroup_record_reclaim_priority(sc->mem_cgroup, priority); | |
1da177e4 | 1657 | |
873b4771 KK |
1658 | delayacct_freepages_end(); |
1659 | ||
1da177e4 LT |
1660 | return ret; |
1661 | } | |
1662 | ||
dac1d27b MG |
1663 | unsigned long try_to_free_pages(struct zonelist *zonelist, int order, |
1664 | gfp_t gfp_mask) | |
66e1707b BS |
1665 | { |
1666 | struct scan_control sc = { | |
1667 | .gfp_mask = gfp_mask, | |
1668 | .may_writepage = !laptop_mode, | |
1669 | .swap_cluster_max = SWAP_CLUSTER_MAX, | |
1670 | .may_swap = 1, | |
1671 | .swappiness = vm_swappiness, | |
1672 | .order = order, | |
1673 | .mem_cgroup = NULL, | |
1674 | .isolate_pages = isolate_pages_global, | |
1675 | }; | |
1676 | ||
dd1a239f | 1677 | return do_try_to_free_pages(zonelist, &sc); |
66e1707b BS |
1678 | } |
1679 | ||
00f0b825 | 1680 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR |
66e1707b | 1681 | |
e1a1cd59 | 1682 | unsigned long try_to_free_mem_cgroup_pages(struct mem_cgroup *mem_cont, |
8c7c6e34 KH |
1683 | gfp_t gfp_mask, |
1684 | bool noswap) | |
66e1707b BS |
1685 | { |
1686 | struct scan_control sc = { | |
66e1707b BS |
1687 | .may_writepage = !laptop_mode, |
1688 | .may_swap = 1, | |
1689 | .swap_cluster_max = SWAP_CLUSTER_MAX, | |
1690 | .swappiness = vm_swappiness, | |
1691 | .order = 0, | |
1692 | .mem_cgroup = mem_cont, | |
1693 | .isolate_pages = mem_cgroup_isolate_pages, | |
1694 | }; | |
dac1d27b | 1695 | struct zonelist *zonelist; |
66e1707b | 1696 | |
8c7c6e34 KH |
1697 | if (noswap) |
1698 | sc.may_swap = 0; | |
1699 | ||
dd1a239f MG |
1700 | sc.gfp_mask = (gfp_mask & GFP_RECLAIM_MASK) | |
1701 | (GFP_HIGHUSER_MOVABLE & ~GFP_RECLAIM_MASK); | |
1702 | zonelist = NODE_DATA(numa_node_id())->node_zonelists; | |
1703 | return do_try_to_free_pages(zonelist, &sc); | |
66e1707b BS |
1704 | } |
1705 | #endif | |
1706 | ||
1da177e4 LT |
1707 | /* |
1708 | * For kswapd, balance_pgdat() will work across all this node's zones until | |
1709 | * they are all at pages_high. | |
1710 | * | |
1da177e4 LT |
1711 | * Returns the number of pages which were actually freed. |
1712 | * | |
1713 | * There is special handling here for zones which are full of pinned pages. | |
1714 | * This can happen if the pages are all mlocked, or if they are all used by | |
1715 | * device drivers (say, ZONE_DMA). Or if they are all in use by hugetlb. | |
1716 | * What we do is to detect the case where all pages in the zone have been | |
1717 | * scanned twice and there has been zero successful reclaim. Mark the zone as | |
1718 | * dead and from now on, only perform a short scan. Basically we're polling | |
1719 | * the zone for when the problem goes away. | |
1720 | * | |
1721 | * kswapd scans the zones in the highmem->normal->dma direction. It skips | |
1722 | * zones which have free_pages > pages_high, but once a zone is found to have | |
1723 | * free_pages <= pages_high, we scan that zone and the lower zones regardless | |
1724 | * of the number of free pages in the lower zones. This interoperates with | |
1725 | * the page allocator fallback scheme to ensure that aging of pages is balanced | |
1726 | * across the zones. | |
1727 | */ | |
d6277db4 | 1728 | static unsigned long balance_pgdat(pg_data_t *pgdat, int order) |
1da177e4 | 1729 | { |
1da177e4 LT |
1730 | int all_zones_ok; |
1731 | int priority; | |
1732 | int i; | |
69e05944 | 1733 | unsigned long total_scanned; |
1da177e4 | 1734 | struct reclaim_state *reclaim_state = current->reclaim_state; |
179e9639 AM |
1735 | struct scan_control sc = { |
1736 | .gfp_mask = GFP_KERNEL, | |
1737 | .may_swap = 1, | |
d6277db4 RW |
1738 | .swap_cluster_max = SWAP_CLUSTER_MAX, |
1739 | .swappiness = vm_swappiness, | |
5ad333eb | 1740 | .order = order, |
66e1707b BS |
1741 | .mem_cgroup = NULL, |
1742 | .isolate_pages = isolate_pages_global, | |
179e9639 | 1743 | }; |
3bb1a852 MB |
1744 | /* |
1745 | * temp_priority is used to remember the scanning priority at which | |
1746 | * this zone was successfully refilled to free_pages == pages_high. | |
1747 | */ | |
1748 | int temp_priority[MAX_NR_ZONES]; | |
1da177e4 LT |
1749 | |
1750 | loop_again: | |
1751 | total_scanned = 0; | |
a79311c1 | 1752 | sc.nr_reclaimed = 0; |
c0bbbc73 | 1753 | sc.may_writepage = !laptop_mode; |
f8891e5e | 1754 | count_vm_event(PAGEOUTRUN); |
1da177e4 | 1755 | |
3bb1a852 MB |
1756 | for (i = 0; i < pgdat->nr_zones; i++) |
1757 | temp_priority[i] = DEF_PRIORITY; | |
1da177e4 LT |
1758 | |
1759 | for (priority = DEF_PRIORITY; priority >= 0; priority--) { | |
1760 | int end_zone = 0; /* Inclusive. 0 = ZONE_DMA */ | |
1761 | unsigned long lru_pages = 0; | |
1762 | ||
f7b7fd8f RR |
1763 | /* The swap token gets in the way of swapout... */ |
1764 | if (!priority) | |
1765 | disable_swap_token(); | |
1766 | ||
1da177e4 LT |
1767 | all_zones_ok = 1; |
1768 | ||
d6277db4 RW |
1769 | /* |
1770 | * Scan in the highmem->dma direction for the highest | |
1771 | * zone which needs scanning | |
1772 | */ | |
1773 | for (i = pgdat->nr_zones - 1; i >= 0; i--) { | |
1774 | struct zone *zone = pgdat->node_zones + i; | |
1da177e4 | 1775 | |
d6277db4 RW |
1776 | if (!populated_zone(zone)) |
1777 | continue; | |
1da177e4 | 1778 | |
e815af95 DR |
1779 | if (zone_is_all_unreclaimable(zone) && |
1780 | priority != DEF_PRIORITY) | |
d6277db4 | 1781 | continue; |
1da177e4 | 1782 | |
556adecb RR |
1783 | /* |
1784 | * Do some background aging of the anon list, to give | |
1785 | * pages a chance to be referenced before reclaiming. | |
1786 | */ | |
1787 | if (inactive_anon_is_low(zone)) | |
1788 | shrink_active_list(SWAP_CLUSTER_MAX, zone, | |
1789 | &sc, priority, 0); | |
1790 | ||
d6277db4 RW |
1791 | if (!zone_watermark_ok(zone, order, zone->pages_high, |
1792 | 0, 0)) { | |
1793 | end_zone = i; | |
e1dbeda6 | 1794 | break; |
1da177e4 | 1795 | } |
1da177e4 | 1796 | } |
e1dbeda6 AM |
1797 | if (i < 0) |
1798 | goto out; | |
1799 | ||
1da177e4 LT |
1800 | for (i = 0; i <= end_zone; i++) { |
1801 | struct zone *zone = pgdat->node_zones + i; | |
1802 | ||
4f98a2fe | 1803 | lru_pages += zone_lru_pages(zone); |
1da177e4 LT |
1804 | } |
1805 | ||
1806 | /* | |
1807 | * Now scan the zone in the dma->highmem direction, stopping | |
1808 | * at the last zone which needs scanning. | |
1809 | * | |
1810 | * We do this because the page allocator works in the opposite | |
1811 | * direction. This prevents the page allocator from allocating | |
1812 | * pages behind kswapd's direction of progress, which would | |
1813 | * cause too much scanning of the lower zones. | |
1814 | */ | |
1815 | for (i = 0; i <= end_zone; i++) { | |
1816 | struct zone *zone = pgdat->node_zones + i; | |
b15e0905 | 1817 | int nr_slab; |
1da177e4 | 1818 | |
f3fe6512 | 1819 | if (!populated_zone(zone)) |
1da177e4 LT |
1820 | continue; |
1821 | ||
e815af95 DR |
1822 | if (zone_is_all_unreclaimable(zone) && |
1823 | priority != DEF_PRIORITY) | |
1da177e4 LT |
1824 | continue; |
1825 | ||
d6277db4 RW |
1826 | if (!zone_watermark_ok(zone, order, zone->pages_high, |
1827 | end_zone, 0)) | |
1828 | all_zones_ok = 0; | |
3bb1a852 | 1829 | temp_priority[i] = priority; |
1da177e4 | 1830 | sc.nr_scanned = 0; |
3bb1a852 | 1831 | note_zone_scanning_priority(zone, priority); |
32a4330d RR |
1832 | /* |
1833 | * We put equal pressure on every zone, unless one | |
1834 | * zone has way too many pages free already. | |
1835 | */ | |
1836 | if (!zone_watermark_ok(zone, order, 8*zone->pages_high, | |
1837 | end_zone, 0)) | |
a79311c1 | 1838 | shrink_zone(priority, zone, &sc); |
1da177e4 | 1839 | reclaim_state->reclaimed_slab = 0; |
b15e0905 | 1840 | nr_slab = shrink_slab(sc.nr_scanned, GFP_KERNEL, |
1841 | lru_pages); | |
a79311c1 | 1842 | sc.nr_reclaimed += reclaim_state->reclaimed_slab; |
1da177e4 | 1843 | total_scanned += sc.nr_scanned; |
e815af95 | 1844 | if (zone_is_all_unreclaimable(zone)) |
1da177e4 | 1845 | continue; |
b15e0905 | 1846 | if (nr_slab == 0 && zone->pages_scanned >= |
4f98a2fe | 1847 | (zone_lru_pages(zone) * 6)) |
e815af95 DR |
1848 | zone_set_flag(zone, |
1849 | ZONE_ALL_UNRECLAIMABLE); | |
1da177e4 LT |
1850 | /* |
1851 | * If we've done a decent amount of scanning and | |
1852 | * the reclaim ratio is low, start doing writepage | |
1853 | * even in laptop mode | |
1854 | */ | |
1855 | if (total_scanned > SWAP_CLUSTER_MAX * 2 && | |
a79311c1 | 1856 | total_scanned > sc.nr_reclaimed + sc.nr_reclaimed / 2) |
1da177e4 LT |
1857 | sc.may_writepage = 1; |
1858 | } | |
1da177e4 LT |
1859 | if (all_zones_ok) |
1860 | break; /* kswapd: all done */ | |
1861 | /* | |
1862 | * OK, kswapd is getting into trouble. Take a nap, then take | |
1863 | * another pass across the zones. | |
1864 | */ | |
4dd4b920 | 1865 | if (total_scanned && priority < DEF_PRIORITY - 2) |
3fcfab16 | 1866 | congestion_wait(WRITE, HZ/10); |
1da177e4 LT |
1867 | |
1868 | /* | |
1869 | * We do this so kswapd doesn't build up large priorities for | |
1870 | * example when it is freeing in parallel with allocators. It | |
1871 | * matches the direct reclaim path behaviour in terms of impact | |
1872 | * on zone->*_priority. | |
1873 | */ | |
a79311c1 | 1874 | if (sc.nr_reclaimed >= SWAP_CLUSTER_MAX) |
1da177e4 LT |
1875 | break; |
1876 | } | |
1877 | out: | |
3bb1a852 MB |
1878 | /* |
1879 | * Note within each zone the priority level at which this zone was | |
1880 | * brought into a happy state. So that the next thread which scans this | |
1881 | * zone will start out at that priority level. | |
1882 | */ | |
1da177e4 LT |
1883 | for (i = 0; i < pgdat->nr_zones; i++) { |
1884 | struct zone *zone = pgdat->node_zones + i; | |
1885 | ||
3bb1a852 | 1886 | zone->prev_priority = temp_priority[i]; |
1da177e4 LT |
1887 | } |
1888 | if (!all_zones_ok) { | |
1889 | cond_resched(); | |
8357376d RW |
1890 | |
1891 | try_to_freeze(); | |
1892 | ||
73ce02e9 KM |
1893 | /* |
1894 | * Fragmentation may mean that the system cannot be | |
1895 | * rebalanced for high-order allocations in all zones. | |
1896 | * At this point, if nr_reclaimed < SWAP_CLUSTER_MAX, | |
1897 | * it means the zones have been fully scanned and are still | |
1898 | * not balanced. For high-order allocations, there is | |
1899 | * little point trying all over again as kswapd may | |
1900 | * infinite loop. | |
1901 | * | |
1902 | * Instead, recheck all watermarks at order-0 as they | |
1903 | * are the most important. If watermarks are ok, kswapd will go | |
1904 | * back to sleep. High-order users can still perform direct | |
1905 | * reclaim if they wish. | |
1906 | */ | |
1907 | if (sc.nr_reclaimed < SWAP_CLUSTER_MAX) | |
1908 | order = sc.order = 0; | |
1909 | ||
1da177e4 LT |
1910 | goto loop_again; |
1911 | } | |
1912 | ||
a79311c1 | 1913 | return sc.nr_reclaimed; |
1da177e4 LT |
1914 | } |
1915 | ||
1916 | /* | |
1917 | * The background pageout daemon, started as a kernel thread | |
4f98a2fe | 1918 | * from the init process. |
1da177e4 LT |
1919 | * |
1920 | * This basically trickles out pages so that we have _some_ | |
1921 | * free memory available even if there is no other activity | |
1922 | * that frees anything up. This is needed for things like routing | |
1923 | * etc, where we otherwise might have all activity going on in | |
1924 | * asynchronous contexts that cannot page things out. | |
1925 | * | |
1926 | * If there are applications that are active memory-allocators | |
1927 | * (most normal use), this basically shouldn't matter. | |
1928 | */ | |
1929 | static int kswapd(void *p) | |
1930 | { | |
1931 | unsigned long order; | |
1932 | pg_data_t *pgdat = (pg_data_t*)p; | |
1933 | struct task_struct *tsk = current; | |
1934 | DEFINE_WAIT(wait); | |
1935 | struct reclaim_state reclaim_state = { | |
1936 | .reclaimed_slab = 0, | |
1937 | }; | |
c5f59f08 | 1938 | node_to_cpumask_ptr(cpumask, pgdat->node_id); |
1da177e4 | 1939 | |
174596a0 | 1940 | if (!cpumask_empty(cpumask)) |
c5f59f08 | 1941 | set_cpus_allowed_ptr(tsk, cpumask); |
1da177e4 LT |
1942 | current->reclaim_state = &reclaim_state; |
1943 | ||
1944 | /* | |
1945 | * Tell the memory management that we're a "memory allocator", | |
1946 | * and that if we need more memory we should get access to it | |
1947 | * regardless (see "__alloc_pages()"). "kswapd" should | |
1948 | * never get caught in the normal page freeing logic. | |
1949 | * | |
1950 | * (Kswapd normally doesn't need memory anyway, but sometimes | |
1951 | * you need a small amount of memory in order to be able to | |
1952 | * page out something else, and this flag essentially protects | |
1953 | * us from recursively trying to free more memory as we're | |
1954 | * trying to free the first piece of memory in the first place). | |
1955 | */ | |
930d9152 | 1956 | tsk->flags |= PF_MEMALLOC | PF_SWAPWRITE | PF_KSWAPD; |
83144186 | 1957 | set_freezable(); |
1da177e4 LT |
1958 | |
1959 | order = 0; | |
1960 | for ( ; ; ) { | |
1961 | unsigned long new_order; | |
3e1d1d28 | 1962 | |
1da177e4 LT |
1963 | prepare_to_wait(&pgdat->kswapd_wait, &wait, TASK_INTERRUPTIBLE); |
1964 | new_order = pgdat->kswapd_max_order; | |
1965 | pgdat->kswapd_max_order = 0; | |
1966 | if (order < new_order) { | |
1967 | /* | |
1968 | * Don't sleep if someone wants a larger 'order' | |
1969 | * allocation | |
1970 | */ | |
1971 | order = new_order; | |
1972 | } else { | |
b1296cc4 RW |
1973 | if (!freezing(current)) |
1974 | schedule(); | |
1975 | ||
1da177e4 LT |
1976 | order = pgdat->kswapd_max_order; |
1977 | } | |
1978 | finish_wait(&pgdat->kswapd_wait, &wait); | |
1979 | ||
b1296cc4 RW |
1980 | if (!try_to_freeze()) { |
1981 | /* We can speed up thawing tasks if we don't call | |
1982 | * balance_pgdat after returning from the refrigerator | |
1983 | */ | |
1984 | balance_pgdat(pgdat, order); | |
1985 | } | |
1da177e4 LT |
1986 | } |
1987 | return 0; | |
1988 | } | |
1989 | ||
1990 | /* | |
1991 | * A zone is low on free memory, so wake its kswapd task to service it. | |
1992 | */ | |
1993 | void wakeup_kswapd(struct zone *zone, int order) | |
1994 | { | |
1995 | pg_data_t *pgdat; | |
1996 | ||
f3fe6512 | 1997 | if (!populated_zone(zone)) |
1da177e4 LT |
1998 | return; |
1999 | ||
2000 | pgdat = zone->zone_pgdat; | |
7fb1d9fc | 2001 | if (zone_watermark_ok(zone, order, zone->pages_low, 0, 0)) |
1da177e4 LT |
2002 | return; |
2003 | if (pgdat->kswapd_max_order < order) | |
2004 | pgdat->kswapd_max_order = order; | |
02a0e53d | 2005 | if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL)) |
1da177e4 | 2006 | return; |
8d0986e2 | 2007 | if (!waitqueue_active(&pgdat->kswapd_wait)) |
1da177e4 | 2008 | return; |
8d0986e2 | 2009 | wake_up_interruptible(&pgdat->kswapd_wait); |
1da177e4 LT |
2010 | } |
2011 | ||
4f98a2fe RR |
2012 | unsigned long global_lru_pages(void) |
2013 | { | |
2014 | return global_page_state(NR_ACTIVE_ANON) | |
2015 | + global_page_state(NR_ACTIVE_FILE) | |
2016 | + global_page_state(NR_INACTIVE_ANON) | |
2017 | + global_page_state(NR_INACTIVE_FILE); | |
2018 | } | |
2019 | ||
1da177e4 LT |
2020 | #ifdef CONFIG_PM |
2021 | /* | |
d6277db4 RW |
2022 | * Helper function for shrink_all_memory(). Tries to reclaim 'nr_pages' pages |
2023 | * from LRU lists system-wide, for given pass and priority, and returns the | |
2024 | * number of reclaimed pages | |
2025 | * | |
2026 | * For pass > 3 we also try to shrink the LRU lists that contain a few pages | |
2027 | */ | |
e07aa05b NC |
2028 | static unsigned long shrink_all_zones(unsigned long nr_pages, int prio, |
2029 | int pass, struct scan_control *sc) | |
d6277db4 RW |
2030 | { |
2031 | struct zone *zone; | |
2032 | unsigned long nr_to_scan, ret = 0; | |
b69408e8 | 2033 | enum lru_list l; |
d6277db4 RW |
2034 | |
2035 | for_each_zone(zone) { | |
2036 | ||
2037 | if (!populated_zone(zone)) | |
2038 | continue; | |
2039 | ||
e815af95 | 2040 | if (zone_is_all_unreclaimable(zone) && prio != DEF_PRIORITY) |
d6277db4 RW |
2041 | continue; |
2042 | ||
894bc310 LS |
2043 | for_each_evictable_lru(l) { |
2044 | /* For pass = 0, we don't shrink the active list */ | |
4f98a2fe RR |
2045 | if (pass == 0 && |
2046 | (l == LRU_ACTIVE || l == LRU_ACTIVE_FILE)) | |
b69408e8 CL |
2047 | continue; |
2048 | ||
2049 | zone->lru[l].nr_scan += | |
2050 | (zone_page_state(zone, NR_LRU_BASE + l) | |
2051 | >> prio) + 1; | |
2052 | if (zone->lru[l].nr_scan >= nr_pages || pass > 3) { | |
2053 | zone->lru[l].nr_scan = 0; | |
c8785385 | 2054 | nr_to_scan = min(nr_pages, |
b69408e8 CL |
2055 | zone_page_state(zone, |
2056 | NR_LRU_BASE + l)); | |
2057 | ret += shrink_list(l, nr_to_scan, zone, | |
2058 | sc, prio); | |
2059 | if (ret >= nr_pages) | |
2060 | return ret; | |
d6277db4 RW |
2061 | } |
2062 | } | |
d6277db4 RW |
2063 | } |
2064 | ||
2065 | return ret; | |
2066 | } | |
2067 | ||
2068 | /* | |
2069 | * Try to free `nr_pages' of memory, system-wide, and return the number of | |
2070 | * freed pages. | |
2071 | * | |
2072 | * Rather than trying to age LRUs the aim is to preserve the overall | |
2073 | * LRU order by reclaiming preferentially | |
2074 | * inactive > active > active referenced > active mapped | |
1da177e4 | 2075 | */ |
69e05944 | 2076 | unsigned long shrink_all_memory(unsigned long nr_pages) |
1da177e4 | 2077 | { |
d6277db4 | 2078 | unsigned long lru_pages, nr_slab; |
69e05944 | 2079 | unsigned long ret = 0; |
d6277db4 RW |
2080 | int pass; |
2081 | struct reclaim_state reclaim_state; | |
d6277db4 RW |
2082 | struct scan_control sc = { |
2083 | .gfp_mask = GFP_KERNEL, | |
2084 | .may_swap = 0, | |
2085 | .swap_cluster_max = nr_pages, | |
2086 | .may_writepage = 1, | |
2087 | .swappiness = vm_swappiness, | |
66e1707b | 2088 | .isolate_pages = isolate_pages_global, |
1da177e4 LT |
2089 | }; |
2090 | ||
2091 | current->reclaim_state = &reclaim_state; | |
69e05944 | 2092 | |
4f98a2fe | 2093 | lru_pages = global_lru_pages(); |
972d1a7b | 2094 | nr_slab = global_page_state(NR_SLAB_RECLAIMABLE); |
d6277db4 RW |
2095 | /* If slab caches are huge, it's better to hit them first */ |
2096 | while (nr_slab >= lru_pages) { | |
2097 | reclaim_state.reclaimed_slab = 0; | |
2098 | shrink_slab(nr_pages, sc.gfp_mask, lru_pages); | |
2099 | if (!reclaim_state.reclaimed_slab) | |
1da177e4 | 2100 | break; |
d6277db4 RW |
2101 | |
2102 | ret += reclaim_state.reclaimed_slab; | |
2103 | if (ret >= nr_pages) | |
2104 | goto out; | |
2105 | ||
2106 | nr_slab -= reclaim_state.reclaimed_slab; | |
1da177e4 | 2107 | } |
d6277db4 RW |
2108 | |
2109 | /* | |
2110 | * We try to shrink LRUs in 5 passes: | |
2111 | * 0 = Reclaim from inactive_list only | |
2112 | * 1 = Reclaim from active list but don't reclaim mapped | |
2113 | * 2 = 2nd pass of type 1 | |
2114 | * 3 = Reclaim mapped (normal reclaim) | |
2115 | * 4 = 2nd pass of type 3 | |
2116 | */ | |
2117 | for (pass = 0; pass < 5; pass++) { | |
2118 | int prio; | |
2119 | ||
d6277db4 RW |
2120 | /* Force reclaiming mapped pages in the passes #3 and #4 */ |
2121 | if (pass > 2) { | |
2122 | sc.may_swap = 1; | |
2123 | sc.swappiness = 100; | |
2124 | } | |
2125 | ||
2126 | for (prio = DEF_PRIORITY; prio >= 0; prio--) { | |
2127 | unsigned long nr_to_scan = nr_pages - ret; | |
2128 | ||
d6277db4 | 2129 | sc.nr_scanned = 0; |
d6277db4 RW |
2130 | ret += shrink_all_zones(nr_to_scan, prio, pass, &sc); |
2131 | if (ret >= nr_pages) | |
2132 | goto out; | |
2133 | ||
2134 | reclaim_state.reclaimed_slab = 0; | |
76395d37 | 2135 | shrink_slab(sc.nr_scanned, sc.gfp_mask, |
4f98a2fe | 2136 | global_lru_pages()); |
d6277db4 RW |
2137 | ret += reclaim_state.reclaimed_slab; |
2138 | if (ret >= nr_pages) | |
2139 | goto out; | |
2140 | ||
2141 | if (sc.nr_scanned && prio < DEF_PRIORITY - 2) | |
3fcfab16 | 2142 | congestion_wait(WRITE, HZ / 10); |
d6277db4 | 2143 | } |
248a0301 | 2144 | } |
d6277db4 RW |
2145 | |
2146 | /* | |
2147 | * If ret = 0, we could not shrink LRUs, but there may be something | |
2148 | * in slab caches | |
2149 | */ | |
76395d37 | 2150 | if (!ret) { |
d6277db4 RW |
2151 | do { |
2152 | reclaim_state.reclaimed_slab = 0; | |
4f98a2fe | 2153 | shrink_slab(nr_pages, sc.gfp_mask, global_lru_pages()); |
d6277db4 RW |
2154 | ret += reclaim_state.reclaimed_slab; |
2155 | } while (ret < nr_pages && reclaim_state.reclaimed_slab > 0); | |
76395d37 | 2156 | } |
d6277db4 RW |
2157 | |
2158 | out: | |
1da177e4 | 2159 | current->reclaim_state = NULL; |
d6277db4 | 2160 | |
1da177e4 LT |
2161 | return ret; |
2162 | } | |
2163 | #endif | |
2164 | ||
1da177e4 LT |
2165 | /* It's optimal to keep kswapds on the same CPUs as their memory, but |
2166 | not required for correctness. So if the last cpu in a node goes | |
2167 | away, we get changed to run anywhere: as the first one comes back, | |
2168 | restore their cpu bindings. */ | |
9c7b216d | 2169 | static int __devinit cpu_callback(struct notifier_block *nfb, |
69e05944 | 2170 | unsigned long action, void *hcpu) |
1da177e4 | 2171 | { |
58c0a4a7 | 2172 | int nid; |
1da177e4 | 2173 | |
8bb78442 | 2174 | if (action == CPU_ONLINE || action == CPU_ONLINE_FROZEN) { |
58c0a4a7 | 2175 | for_each_node_state(nid, N_HIGH_MEMORY) { |
c5f59f08 MT |
2176 | pg_data_t *pgdat = NODE_DATA(nid); |
2177 | node_to_cpumask_ptr(mask, pgdat->node_id); | |
2178 | ||
3e597945 | 2179 | if (cpumask_any_and(cpu_online_mask, mask) < nr_cpu_ids) |
1da177e4 | 2180 | /* One of our CPUs online: restore mask */ |
c5f59f08 | 2181 | set_cpus_allowed_ptr(pgdat->kswapd, mask); |
1da177e4 LT |
2182 | } |
2183 | } | |
2184 | return NOTIFY_OK; | |
2185 | } | |
1da177e4 | 2186 | |
3218ae14 YG |
2187 | /* |
2188 | * This kswapd start function will be called by init and node-hot-add. | |
2189 | * On node-hot-add, kswapd will moved to proper cpus if cpus are hot-added. | |
2190 | */ | |
2191 | int kswapd_run(int nid) | |
2192 | { | |
2193 | pg_data_t *pgdat = NODE_DATA(nid); | |
2194 | int ret = 0; | |
2195 | ||
2196 | if (pgdat->kswapd) | |
2197 | return 0; | |
2198 | ||
2199 | pgdat->kswapd = kthread_run(kswapd, pgdat, "kswapd%d", nid); | |
2200 | if (IS_ERR(pgdat->kswapd)) { | |
2201 | /* failure at boot is fatal */ | |
2202 | BUG_ON(system_state == SYSTEM_BOOTING); | |
2203 | printk("Failed to start kswapd on node %d\n",nid); | |
2204 | ret = -1; | |
2205 | } | |
2206 | return ret; | |
2207 | } | |
2208 | ||
1da177e4 LT |
2209 | static int __init kswapd_init(void) |
2210 | { | |
3218ae14 | 2211 | int nid; |
69e05944 | 2212 | |
1da177e4 | 2213 | swap_setup(); |
9422ffba | 2214 | for_each_node_state(nid, N_HIGH_MEMORY) |
3218ae14 | 2215 | kswapd_run(nid); |
1da177e4 LT |
2216 | hotcpu_notifier(cpu_callback, 0); |
2217 | return 0; | |
2218 | } | |
2219 | ||
2220 | module_init(kswapd_init) | |
9eeff239 CL |
2221 | |
2222 | #ifdef CONFIG_NUMA | |
2223 | /* | |
2224 | * Zone reclaim mode | |
2225 | * | |
2226 | * If non-zero call zone_reclaim when the number of free pages falls below | |
2227 | * the watermarks. | |
9eeff239 CL |
2228 | */ |
2229 | int zone_reclaim_mode __read_mostly; | |
2230 | ||
1b2ffb78 | 2231 | #define RECLAIM_OFF 0 |
7d03431c | 2232 | #define RECLAIM_ZONE (1<<0) /* Run shrink_inactive_list on the zone */ |
1b2ffb78 CL |
2233 | #define RECLAIM_WRITE (1<<1) /* Writeout pages during reclaim */ |
2234 | #define RECLAIM_SWAP (1<<2) /* Swap pages out during reclaim */ | |
2235 | ||
a92f7126 CL |
2236 | /* |
2237 | * Priority for ZONE_RECLAIM. This determines the fraction of pages | |
2238 | * of a node considered for each zone_reclaim. 4 scans 1/16th of | |
2239 | * a zone. | |
2240 | */ | |
2241 | #define ZONE_RECLAIM_PRIORITY 4 | |
2242 | ||
9614634f CL |
2243 | /* |
2244 | * Percentage of pages in a zone that must be unmapped for zone_reclaim to | |
2245 | * occur. | |
2246 | */ | |
2247 | int sysctl_min_unmapped_ratio = 1; | |
2248 | ||
0ff38490 CL |
2249 | /* |
2250 | * If the number of slab pages in a zone grows beyond this percentage then | |
2251 | * slab reclaim needs to occur. | |
2252 | */ | |
2253 | int sysctl_min_slab_ratio = 5; | |
2254 | ||
9eeff239 CL |
2255 | /* |
2256 | * Try to free up some pages from this zone through reclaim. | |
2257 | */ | |
179e9639 | 2258 | static int __zone_reclaim(struct zone *zone, gfp_t gfp_mask, unsigned int order) |
9eeff239 | 2259 | { |
7fb2d46d | 2260 | /* Minimum pages needed in order to stay on node */ |
69e05944 | 2261 | const unsigned long nr_pages = 1 << order; |
9eeff239 CL |
2262 | struct task_struct *p = current; |
2263 | struct reclaim_state reclaim_state; | |
8695949a | 2264 | int priority; |
179e9639 AM |
2265 | struct scan_control sc = { |
2266 | .may_writepage = !!(zone_reclaim_mode & RECLAIM_WRITE), | |
2267 | .may_swap = !!(zone_reclaim_mode & RECLAIM_SWAP), | |
69e05944 AM |
2268 | .swap_cluster_max = max_t(unsigned long, nr_pages, |
2269 | SWAP_CLUSTER_MAX), | |
179e9639 | 2270 | .gfp_mask = gfp_mask, |
d6277db4 | 2271 | .swappiness = vm_swappiness, |
66e1707b | 2272 | .isolate_pages = isolate_pages_global, |
179e9639 | 2273 | }; |
83e33a47 | 2274 | unsigned long slab_reclaimable; |
9eeff239 CL |
2275 | |
2276 | disable_swap_token(); | |
9eeff239 | 2277 | cond_resched(); |
d4f7796e CL |
2278 | /* |
2279 | * We need to be able to allocate from the reserves for RECLAIM_SWAP | |
2280 | * and we also need to be able to write out pages for RECLAIM_WRITE | |
2281 | * and RECLAIM_SWAP. | |
2282 | */ | |
2283 | p->flags |= PF_MEMALLOC | PF_SWAPWRITE; | |
9eeff239 CL |
2284 | reclaim_state.reclaimed_slab = 0; |
2285 | p->reclaim_state = &reclaim_state; | |
c84db23c | 2286 | |
0ff38490 CL |
2287 | if (zone_page_state(zone, NR_FILE_PAGES) - |
2288 | zone_page_state(zone, NR_FILE_MAPPED) > | |
2289 | zone->min_unmapped_pages) { | |
2290 | /* | |
2291 | * Free memory by calling shrink zone with increasing | |
2292 | * priorities until we have enough memory freed. | |
2293 | */ | |
2294 | priority = ZONE_RECLAIM_PRIORITY; | |
2295 | do { | |
3bb1a852 | 2296 | note_zone_scanning_priority(zone, priority); |
a79311c1 | 2297 | shrink_zone(priority, zone, &sc); |
0ff38490 | 2298 | priority--; |
a79311c1 | 2299 | } while (priority >= 0 && sc.nr_reclaimed < nr_pages); |
0ff38490 | 2300 | } |
c84db23c | 2301 | |
83e33a47 CL |
2302 | slab_reclaimable = zone_page_state(zone, NR_SLAB_RECLAIMABLE); |
2303 | if (slab_reclaimable > zone->min_slab_pages) { | |
2a16e3f4 | 2304 | /* |
7fb2d46d | 2305 | * shrink_slab() does not currently allow us to determine how |
0ff38490 CL |
2306 | * many pages were freed in this zone. So we take the current |
2307 | * number of slab pages and shake the slab until it is reduced | |
2308 | * by the same nr_pages that we used for reclaiming unmapped | |
2309 | * pages. | |
2a16e3f4 | 2310 | * |
0ff38490 CL |
2311 | * Note that shrink_slab will free memory on all zones and may |
2312 | * take a long time. | |
2a16e3f4 | 2313 | */ |
0ff38490 | 2314 | while (shrink_slab(sc.nr_scanned, gfp_mask, order) && |
83e33a47 CL |
2315 | zone_page_state(zone, NR_SLAB_RECLAIMABLE) > |
2316 | slab_reclaimable - nr_pages) | |
0ff38490 | 2317 | ; |
83e33a47 CL |
2318 | |
2319 | /* | |
2320 | * Update nr_reclaimed by the number of slab pages we | |
2321 | * reclaimed from this zone. | |
2322 | */ | |
a79311c1 | 2323 | sc.nr_reclaimed += slab_reclaimable - |
83e33a47 | 2324 | zone_page_state(zone, NR_SLAB_RECLAIMABLE); |
2a16e3f4 CL |
2325 | } |
2326 | ||
9eeff239 | 2327 | p->reclaim_state = NULL; |
d4f7796e | 2328 | current->flags &= ~(PF_MEMALLOC | PF_SWAPWRITE); |
a79311c1 | 2329 | return sc.nr_reclaimed >= nr_pages; |
9eeff239 | 2330 | } |
179e9639 AM |
2331 | |
2332 | int zone_reclaim(struct zone *zone, gfp_t gfp_mask, unsigned int order) | |
2333 | { | |
179e9639 | 2334 | int node_id; |
d773ed6b | 2335 | int ret; |
179e9639 AM |
2336 | |
2337 | /* | |
0ff38490 CL |
2338 | * Zone reclaim reclaims unmapped file backed pages and |
2339 | * slab pages if we are over the defined limits. | |
34aa1330 | 2340 | * |
9614634f CL |
2341 | * A small portion of unmapped file backed pages is needed for |
2342 | * file I/O otherwise pages read by file I/O will be immediately | |
2343 | * thrown out if the zone is overallocated. So we do not reclaim | |
2344 | * if less than a specified percentage of the zone is used by | |
2345 | * unmapped file backed pages. | |
179e9639 | 2346 | */ |
34aa1330 | 2347 | if (zone_page_state(zone, NR_FILE_PAGES) - |
0ff38490 CL |
2348 | zone_page_state(zone, NR_FILE_MAPPED) <= zone->min_unmapped_pages |
2349 | && zone_page_state(zone, NR_SLAB_RECLAIMABLE) | |
2350 | <= zone->min_slab_pages) | |
9614634f | 2351 | return 0; |
179e9639 | 2352 | |
d773ed6b DR |
2353 | if (zone_is_all_unreclaimable(zone)) |
2354 | return 0; | |
2355 | ||
179e9639 | 2356 | /* |
d773ed6b | 2357 | * Do not scan if the allocation should not be delayed. |
179e9639 | 2358 | */ |
d773ed6b | 2359 | if (!(gfp_mask & __GFP_WAIT) || (current->flags & PF_MEMALLOC)) |
179e9639 AM |
2360 | return 0; |
2361 | ||
2362 | /* | |
2363 | * Only run zone reclaim on the local zone or on zones that do not | |
2364 | * have associated processors. This will favor the local processor | |
2365 | * over remote processors and spread off node memory allocations | |
2366 | * as wide as possible. | |
2367 | */ | |
89fa3024 | 2368 | node_id = zone_to_nid(zone); |
37c0708d | 2369 | if (node_state(node_id, N_CPU) && node_id != numa_node_id()) |
179e9639 | 2370 | return 0; |
d773ed6b DR |
2371 | |
2372 | if (zone_test_and_set_flag(zone, ZONE_RECLAIM_LOCKED)) | |
2373 | return 0; | |
2374 | ret = __zone_reclaim(zone, gfp_mask, order); | |
2375 | zone_clear_flag(zone, ZONE_RECLAIM_LOCKED); | |
2376 | ||
2377 | return ret; | |
179e9639 | 2378 | } |
9eeff239 | 2379 | #endif |
894bc310 LS |
2380 | |
2381 | #ifdef CONFIG_UNEVICTABLE_LRU | |
2382 | /* | |
2383 | * page_evictable - test whether a page is evictable | |
2384 | * @page: the page to test | |
2385 | * @vma: the VMA in which the page is or will be mapped, may be NULL | |
2386 | * | |
2387 | * Test whether page is evictable--i.e., should be placed on active/inactive | |
b291f000 NP |
2388 | * lists vs unevictable list. The vma argument is !NULL when called from the |
2389 | * fault path to determine how to instantate a new page. | |
894bc310 LS |
2390 | * |
2391 | * Reasons page might not be evictable: | |
ba9ddf49 | 2392 | * (1) page's mapping marked unevictable |
b291f000 | 2393 | * (2) page is part of an mlocked VMA |
ba9ddf49 | 2394 | * |
894bc310 LS |
2395 | */ |
2396 | int page_evictable(struct page *page, struct vm_area_struct *vma) | |
2397 | { | |
2398 | ||
ba9ddf49 LS |
2399 | if (mapping_unevictable(page_mapping(page))) |
2400 | return 0; | |
2401 | ||
b291f000 NP |
2402 | if (PageMlocked(page) || (vma && is_mlocked_vma(vma, page))) |
2403 | return 0; | |
894bc310 LS |
2404 | |
2405 | return 1; | |
2406 | } | |
89e004ea LS |
2407 | |
2408 | /** | |
2409 | * check_move_unevictable_page - check page for evictability and move to appropriate zone lru list | |
2410 | * @page: page to check evictability and move to appropriate lru list | |
2411 | * @zone: zone page is in | |
2412 | * | |
2413 | * Checks a page for evictability and moves the page to the appropriate | |
2414 | * zone lru list. | |
2415 | * | |
2416 | * Restrictions: zone->lru_lock must be held, page must be on LRU and must | |
2417 | * have PageUnevictable set. | |
2418 | */ | |
2419 | static void check_move_unevictable_page(struct page *page, struct zone *zone) | |
2420 | { | |
2421 | VM_BUG_ON(PageActive(page)); | |
2422 | ||
2423 | retry: | |
2424 | ClearPageUnevictable(page); | |
2425 | if (page_evictable(page, NULL)) { | |
2426 | enum lru_list l = LRU_INACTIVE_ANON + page_is_file_cache(page); | |
af936a16 | 2427 | |
89e004ea LS |
2428 | __dec_zone_state(zone, NR_UNEVICTABLE); |
2429 | list_move(&page->lru, &zone->lru[l].list); | |
08e552c6 | 2430 | mem_cgroup_move_lists(page, LRU_UNEVICTABLE, l); |
89e004ea LS |
2431 | __inc_zone_state(zone, NR_INACTIVE_ANON + l); |
2432 | __count_vm_event(UNEVICTABLE_PGRESCUED); | |
2433 | } else { | |
2434 | /* | |
2435 | * rotate unevictable list | |
2436 | */ | |
2437 | SetPageUnevictable(page); | |
2438 | list_move(&page->lru, &zone->lru[LRU_UNEVICTABLE].list); | |
08e552c6 | 2439 | mem_cgroup_rotate_lru_list(page, LRU_UNEVICTABLE); |
89e004ea LS |
2440 | if (page_evictable(page, NULL)) |
2441 | goto retry; | |
2442 | } | |
2443 | } | |
2444 | ||
2445 | /** | |
2446 | * scan_mapping_unevictable_pages - scan an address space for evictable pages | |
2447 | * @mapping: struct address_space to scan for evictable pages | |
2448 | * | |
2449 | * Scan all pages in mapping. Check unevictable pages for | |
2450 | * evictability and move them to the appropriate zone lru list. | |
2451 | */ | |
2452 | void scan_mapping_unevictable_pages(struct address_space *mapping) | |
2453 | { | |
2454 | pgoff_t next = 0; | |
2455 | pgoff_t end = (i_size_read(mapping->host) + PAGE_CACHE_SIZE - 1) >> | |
2456 | PAGE_CACHE_SHIFT; | |
2457 | struct zone *zone; | |
2458 | struct pagevec pvec; | |
2459 | ||
2460 | if (mapping->nrpages == 0) | |
2461 | return; | |
2462 | ||
2463 | pagevec_init(&pvec, 0); | |
2464 | while (next < end && | |
2465 | pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) { | |
2466 | int i; | |
2467 | int pg_scanned = 0; | |
2468 | ||
2469 | zone = NULL; | |
2470 | ||
2471 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
2472 | struct page *page = pvec.pages[i]; | |
2473 | pgoff_t page_index = page->index; | |
2474 | struct zone *pagezone = page_zone(page); | |
2475 | ||
2476 | pg_scanned++; | |
2477 | if (page_index > next) | |
2478 | next = page_index; | |
2479 | next++; | |
2480 | ||
2481 | if (pagezone != zone) { | |
2482 | if (zone) | |
2483 | spin_unlock_irq(&zone->lru_lock); | |
2484 | zone = pagezone; | |
2485 | spin_lock_irq(&zone->lru_lock); | |
2486 | } | |
2487 | ||
2488 | if (PageLRU(page) && PageUnevictable(page)) | |
2489 | check_move_unevictable_page(page, zone); | |
2490 | } | |
2491 | if (zone) | |
2492 | spin_unlock_irq(&zone->lru_lock); | |
2493 | pagevec_release(&pvec); | |
2494 | ||
2495 | count_vm_events(UNEVICTABLE_PGSCANNED, pg_scanned); | |
2496 | } | |
2497 | ||
2498 | } | |
af936a16 LS |
2499 | |
2500 | /** | |
2501 | * scan_zone_unevictable_pages - check unevictable list for evictable pages | |
2502 | * @zone - zone of which to scan the unevictable list | |
2503 | * | |
2504 | * Scan @zone's unevictable LRU lists to check for pages that have become | |
2505 | * evictable. Move those that have to @zone's inactive list where they | |
2506 | * become candidates for reclaim, unless shrink_inactive_zone() decides | |
2507 | * to reactivate them. Pages that are still unevictable are rotated | |
2508 | * back onto @zone's unevictable list. | |
2509 | */ | |
2510 | #define SCAN_UNEVICTABLE_BATCH_SIZE 16UL /* arbitrary lock hold batch size */ | |
14b90b22 | 2511 | static void scan_zone_unevictable_pages(struct zone *zone) |
af936a16 LS |
2512 | { |
2513 | struct list_head *l_unevictable = &zone->lru[LRU_UNEVICTABLE].list; | |
2514 | unsigned long scan; | |
2515 | unsigned long nr_to_scan = zone_page_state(zone, NR_UNEVICTABLE); | |
2516 | ||
2517 | while (nr_to_scan > 0) { | |
2518 | unsigned long batch_size = min(nr_to_scan, | |
2519 | SCAN_UNEVICTABLE_BATCH_SIZE); | |
2520 | ||
2521 | spin_lock_irq(&zone->lru_lock); | |
2522 | for (scan = 0; scan < batch_size; scan++) { | |
2523 | struct page *page = lru_to_page(l_unevictable); | |
2524 | ||
2525 | if (!trylock_page(page)) | |
2526 | continue; | |
2527 | ||
2528 | prefetchw_prev_lru_page(page, l_unevictable, flags); | |
2529 | ||
2530 | if (likely(PageLRU(page) && PageUnevictable(page))) | |
2531 | check_move_unevictable_page(page, zone); | |
2532 | ||
2533 | unlock_page(page); | |
2534 | } | |
2535 | spin_unlock_irq(&zone->lru_lock); | |
2536 | ||
2537 | nr_to_scan -= batch_size; | |
2538 | } | |
2539 | } | |
2540 | ||
2541 | ||
2542 | /** | |
2543 | * scan_all_zones_unevictable_pages - scan all unevictable lists for evictable pages | |
2544 | * | |
2545 | * A really big hammer: scan all zones' unevictable LRU lists to check for | |
2546 | * pages that have become evictable. Move those back to the zones' | |
2547 | * inactive list where they become candidates for reclaim. | |
2548 | * This occurs when, e.g., we have unswappable pages on the unevictable lists, | |
2549 | * and we add swap to the system. As such, it runs in the context of a task | |
2550 | * that has possibly/probably made some previously unevictable pages | |
2551 | * evictable. | |
2552 | */ | |
ff30153b | 2553 | static void scan_all_zones_unevictable_pages(void) |
af936a16 LS |
2554 | { |
2555 | struct zone *zone; | |
2556 | ||
2557 | for_each_zone(zone) { | |
2558 | scan_zone_unevictable_pages(zone); | |
2559 | } | |
2560 | } | |
2561 | ||
2562 | /* | |
2563 | * scan_unevictable_pages [vm] sysctl handler. On demand re-scan of | |
2564 | * all nodes' unevictable lists for evictable pages | |
2565 | */ | |
2566 | unsigned long scan_unevictable_pages; | |
2567 | ||
2568 | int scan_unevictable_handler(struct ctl_table *table, int write, | |
2569 | struct file *file, void __user *buffer, | |
2570 | size_t *length, loff_t *ppos) | |
2571 | { | |
2572 | proc_doulongvec_minmax(table, write, file, buffer, length, ppos); | |
2573 | ||
2574 | if (write && *(unsigned long *)table->data) | |
2575 | scan_all_zones_unevictable_pages(); | |
2576 | ||
2577 | scan_unevictable_pages = 0; | |
2578 | return 0; | |
2579 | } | |
2580 | ||
2581 | /* | |
2582 | * per node 'scan_unevictable_pages' attribute. On demand re-scan of | |
2583 | * a specified node's per zone unevictable lists for evictable pages. | |
2584 | */ | |
2585 | ||
2586 | static ssize_t read_scan_unevictable_node(struct sys_device *dev, | |
2587 | struct sysdev_attribute *attr, | |
2588 | char *buf) | |
2589 | { | |
2590 | return sprintf(buf, "0\n"); /* always zero; should fit... */ | |
2591 | } | |
2592 | ||
2593 | static ssize_t write_scan_unevictable_node(struct sys_device *dev, | |
2594 | struct sysdev_attribute *attr, | |
2595 | const char *buf, size_t count) | |
2596 | { | |
2597 | struct zone *node_zones = NODE_DATA(dev->id)->node_zones; | |
2598 | struct zone *zone; | |
2599 | unsigned long res; | |
2600 | unsigned long req = strict_strtoul(buf, 10, &res); | |
2601 | ||
2602 | if (!req) | |
2603 | return 1; /* zero is no-op */ | |
2604 | ||
2605 | for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) { | |
2606 | if (!populated_zone(zone)) | |
2607 | continue; | |
2608 | scan_zone_unevictable_pages(zone); | |
2609 | } | |
2610 | return 1; | |
2611 | } | |
2612 | ||
2613 | ||
2614 | static SYSDEV_ATTR(scan_unevictable_pages, S_IRUGO | S_IWUSR, | |
2615 | read_scan_unevictable_node, | |
2616 | write_scan_unevictable_node); | |
2617 | ||
2618 | int scan_unevictable_register_node(struct node *node) | |
2619 | { | |
2620 | return sysdev_create_file(&node->sysdev, &attr_scan_unevictable_pages); | |
2621 | } | |
2622 | ||
2623 | void scan_unevictable_unregister_node(struct node *node) | |
2624 | { | |
2625 | sysdev_remove_file(&node->sysdev, &attr_scan_unevictable_pages); | |
2626 | } | |
2627 | ||
894bc310 | 2628 | #endif |