Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/mm/swap.c | |
3 | * | |
4 | * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds | |
5 | */ | |
6 | ||
7 | /* | |
183ff22b | 8 | * This file contains the default values for the operation of the |
1da177e4 LT |
9 | * Linux VM subsystem. Fine-tuning documentation can be found in |
10 | * Documentation/sysctl/vm.txt. | |
11 | * Started 18.12.91 | |
12 | * Swap aging added 23.2.95, Stephen Tweedie. | |
13 | * Buffermem limits added 12.3.98, Rik van Riel. | |
14 | */ | |
15 | ||
16 | #include <linux/mm.h> | |
17 | #include <linux/sched.h> | |
18 | #include <linux/kernel_stat.h> | |
19 | #include <linux/swap.h> | |
20 | #include <linux/mman.h> | |
21 | #include <linux/pagemap.h> | |
22 | #include <linux/pagevec.h> | |
23 | #include <linux/init.h> | |
b95f1b31 | 24 | #include <linux/export.h> |
1da177e4 | 25 | #include <linux/mm_inline.h> |
1da177e4 LT |
26 | #include <linux/percpu_counter.h> |
27 | #include <linux/percpu.h> | |
28 | #include <linux/cpu.h> | |
29 | #include <linux/notifier.h> | |
e0bf68dd | 30 | #include <linux/backing-dev.h> |
66e1707b | 31 | #include <linux/memcontrol.h> |
5a0e3ad6 | 32 | #include <linux/gfp.h> |
1da177e4 | 33 | |
64d6519d LS |
34 | #include "internal.h" |
35 | ||
1da177e4 LT |
36 | /* How many pages do we try to swap or page in/out together? */ |
37 | int page_cluster; | |
38 | ||
f04e9ebb | 39 | static DEFINE_PER_CPU(struct pagevec[NR_LRU_LISTS], lru_add_pvecs); |
f84f9504 | 40 | static DEFINE_PER_CPU(struct pagevec, lru_rotate_pvecs); |
31560180 | 41 | static DEFINE_PER_CPU(struct pagevec, lru_deactivate_pvecs); |
902aaed0 | 42 | |
b221385b AB |
43 | /* |
44 | * This path almost never happens for VM activity - pages are normally | |
45 | * freed via pagevecs. But it gets used by networking. | |
46 | */ | |
920c7a5d | 47 | static void __page_cache_release(struct page *page) |
b221385b AB |
48 | { |
49 | if (PageLRU(page)) { | |
50 | unsigned long flags; | |
51 | struct zone *zone = page_zone(page); | |
52 | ||
53 | spin_lock_irqsave(&zone->lru_lock, flags); | |
54 | VM_BUG_ON(!PageLRU(page)); | |
55 | __ClearPageLRU(page); | |
1c1c53d4 | 56 | del_page_from_lru_list(zone, page, page_off_lru(page)); |
b221385b AB |
57 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
58 | } | |
91807063 AA |
59 | } |
60 | ||
61 | static void __put_single_page(struct page *page) | |
62 | { | |
63 | __page_cache_release(page); | |
fc91668e | 64 | free_hot_cold_page(page, 0); |
b221385b AB |
65 | } |
66 | ||
91807063 | 67 | static void __put_compound_page(struct page *page) |
1da177e4 | 68 | { |
91807063 | 69 | compound_page_dtor *dtor; |
1da177e4 | 70 | |
91807063 AA |
71 | __page_cache_release(page); |
72 | dtor = get_compound_page_dtor(page); | |
73 | (*dtor)(page); | |
74 | } | |
75 | ||
76 | static void put_compound_page(struct page *page) | |
77 | { | |
78 | if (unlikely(PageTail(page))) { | |
79 | /* __split_huge_page_refcount can run under us */ | |
70b50f94 AA |
80 | struct page *page_head = compound_trans_head(page); |
81 | ||
82 | if (likely(page != page_head && | |
83 | get_page_unless_zero(page_head))) { | |
91807063 | 84 | unsigned long flags; |
5bf5f03c PS |
85 | |
86 | /* | |
87 | * THP can not break up slab pages so avoid taking | |
88 | * compound_lock(). Slab performs non-atomic bit ops | |
89 | * on page->flags for better performance. In particular | |
90 | * slab_unlock() in slub used to be a hot path. It is | |
91 | * still hot on arches that do not support | |
92 | * this_cpu_cmpxchg_double(). | |
93 | */ | |
94 | if (PageSlab(page_head)) { | |
95 | if (PageTail(page)) { | |
96 | if (put_page_testzero(page_head)) | |
97 | VM_BUG_ON(1); | |
98 | ||
99 | atomic_dec(&page->_mapcount); | |
100 | goto skip_lock_tail; | |
101 | } else | |
102 | goto skip_lock; | |
103 | } | |
91807063 | 104 | /* |
70b50f94 AA |
105 | * page_head wasn't a dangling pointer but it |
106 | * may not be a head page anymore by the time | |
107 | * we obtain the lock. That is ok as long as it | |
108 | * can't be freed from under us. | |
91807063 | 109 | */ |
91807063 AA |
110 | flags = compound_lock_irqsave(page_head); |
111 | if (unlikely(!PageTail(page))) { | |
112 | /* __split_huge_page_refcount run before us */ | |
113 | compound_unlock_irqrestore(page_head, flags); | |
5bf5f03c | 114 | skip_lock: |
91807063 AA |
115 | if (put_page_testzero(page_head)) |
116 | __put_single_page(page_head); | |
5bf5f03c | 117 | out_put_single: |
91807063 AA |
118 | if (put_page_testzero(page)) |
119 | __put_single_page(page); | |
120 | return; | |
121 | } | |
122 | VM_BUG_ON(page_head != page->first_page); | |
123 | /* | |
124 | * We can release the refcount taken by | |
70b50f94 AA |
125 | * get_page_unless_zero() now that |
126 | * __split_huge_page_refcount() is blocked on | |
127 | * the compound_lock. | |
91807063 AA |
128 | */ |
129 | if (put_page_testzero(page_head)) | |
130 | VM_BUG_ON(1); | |
131 | /* __split_huge_page_refcount will wait now */ | |
70b50f94 AA |
132 | VM_BUG_ON(page_mapcount(page) <= 0); |
133 | atomic_dec(&page->_mapcount); | |
91807063 | 134 | VM_BUG_ON(atomic_read(&page_head->_count) <= 0); |
70b50f94 | 135 | VM_BUG_ON(atomic_read(&page->_count) != 0); |
91807063 | 136 | compound_unlock_irqrestore(page_head, flags); |
5bf5f03c PS |
137 | |
138 | skip_lock_tail: | |
a95a82e9 AA |
139 | if (put_page_testzero(page_head)) { |
140 | if (PageHead(page_head)) | |
141 | __put_compound_page(page_head); | |
142 | else | |
143 | __put_single_page(page_head); | |
144 | } | |
91807063 AA |
145 | } else { |
146 | /* page_head is a dangling pointer */ | |
147 | VM_BUG_ON(PageTail(page)); | |
148 | goto out_put_single; | |
149 | } | |
150 | } else if (put_page_testzero(page)) { | |
151 | if (PageHead(page)) | |
152 | __put_compound_page(page); | |
153 | else | |
154 | __put_single_page(page); | |
1da177e4 | 155 | } |
8519fb30 NP |
156 | } |
157 | ||
158 | void put_page(struct page *page) | |
159 | { | |
160 | if (unlikely(PageCompound(page))) | |
161 | put_compound_page(page); | |
162 | else if (put_page_testzero(page)) | |
91807063 | 163 | __put_single_page(page); |
1da177e4 LT |
164 | } |
165 | EXPORT_SYMBOL(put_page); | |
1da177e4 | 166 | |
70b50f94 AA |
167 | /* |
168 | * This function is exported but must not be called by anything other | |
169 | * than get_page(). It implements the slow path of get_page(). | |
170 | */ | |
171 | bool __get_page_tail(struct page *page) | |
172 | { | |
173 | /* | |
174 | * This takes care of get_page() if run on a tail page | |
175 | * returned by one of the get_user_pages/follow_page variants. | |
176 | * get_user_pages/follow_page itself doesn't need the compound | |
177 | * lock because it runs __get_page_tail_foll() under the | |
178 | * proper PT lock that already serializes against | |
179 | * split_huge_page(). | |
180 | */ | |
181 | unsigned long flags; | |
182 | bool got = false; | |
183 | struct page *page_head = compound_trans_head(page); | |
184 | ||
185 | if (likely(page != page_head && get_page_unless_zero(page_head))) { | |
5bf5f03c PS |
186 | |
187 | /* Ref to put_compound_page() comment. */ | |
188 | if (PageSlab(page_head)) { | |
189 | if (likely(PageTail(page))) { | |
190 | __get_page_tail_foll(page, false); | |
191 | return true; | |
192 | } else { | |
193 | put_page(page_head); | |
194 | return false; | |
195 | } | |
196 | } | |
197 | ||
70b50f94 AA |
198 | /* |
199 | * page_head wasn't a dangling pointer but it | |
200 | * may not be a head page anymore by the time | |
201 | * we obtain the lock. That is ok as long as it | |
202 | * can't be freed from under us. | |
203 | */ | |
204 | flags = compound_lock_irqsave(page_head); | |
205 | /* here __split_huge_page_refcount won't run anymore */ | |
206 | if (likely(PageTail(page))) { | |
207 | __get_page_tail_foll(page, false); | |
208 | got = true; | |
209 | } | |
210 | compound_unlock_irqrestore(page_head, flags); | |
211 | if (unlikely(!got)) | |
212 | put_page(page_head); | |
213 | } | |
214 | return got; | |
215 | } | |
216 | EXPORT_SYMBOL(__get_page_tail); | |
217 | ||
1d7ea732 | 218 | /** |
7682486b RD |
219 | * put_pages_list() - release a list of pages |
220 | * @pages: list of pages threaded on page->lru | |
1d7ea732 AZ |
221 | * |
222 | * Release a list of pages which are strung together on page.lru. Currently | |
223 | * used by read_cache_pages() and related error recovery code. | |
1d7ea732 AZ |
224 | */ |
225 | void put_pages_list(struct list_head *pages) | |
226 | { | |
227 | while (!list_empty(pages)) { | |
228 | struct page *victim; | |
229 | ||
230 | victim = list_entry(pages->prev, struct page, lru); | |
231 | list_del(&victim->lru); | |
232 | page_cache_release(victim); | |
233 | } | |
234 | } | |
235 | EXPORT_SYMBOL(put_pages_list); | |
236 | ||
3dd7ae8e SL |
237 | static void pagevec_lru_move_fn(struct pagevec *pvec, |
238 | void (*move_fn)(struct page *page, void *arg), | |
239 | void *arg) | |
902aaed0 HH |
240 | { |
241 | int i; | |
902aaed0 | 242 | struct zone *zone = NULL; |
3dd7ae8e | 243 | unsigned long flags = 0; |
902aaed0 HH |
244 | |
245 | for (i = 0; i < pagevec_count(pvec); i++) { | |
246 | struct page *page = pvec->pages[i]; | |
247 | struct zone *pagezone = page_zone(page); | |
248 | ||
249 | if (pagezone != zone) { | |
250 | if (zone) | |
3dd7ae8e | 251 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
902aaed0 | 252 | zone = pagezone; |
3dd7ae8e | 253 | spin_lock_irqsave(&zone->lru_lock, flags); |
902aaed0 | 254 | } |
3dd7ae8e SL |
255 | |
256 | (*move_fn)(page, arg); | |
902aaed0 HH |
257 | } |
258 | if (zone) | |
3dd7ae8e | 259 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
83896fb5 LT |
260 | release_pages(pvec->pages, pvec->nr, pvec->cold); |
261 | pagevec_reinit(pvec); | |
d8505dee SL |
262 | } |
263 | ||
3dd7ae8e SL |
264 | static void pagevec_move_tail_fn(struct page *page, void *arg) |
265 | { | |
266 | int *pgmoved = arg; | |
3dd7ae8e SL |
267 | |
268 | if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) { | |
269 | enum lru_list lru = page_lru_base_type(page); | |
925b7673 JW |
270 | struct lruvec *lruvec; |
271 | ||
272 | lruvec = mem_cgroup_lru_move_lists(page_zone(page), | |
273 | page, lru, lru); | |
274 | list_move_tail(&page->lru, &lruvec->lists[lru]); | |
3dd7ae8e SL |
275 | (*pgmoved)++; |
276 | } | |
277 | } | |
278 | ||
279 | /* | |
280 | * pagevec_move_tail() must be called with IRQ disabled. | |
281 | * Otherwise this may cause nasty races. | |
282 | */ | |
283 | static void pagevec_move_tail(struct pagevec *pvec) | |
284 | { | |
285 | int pgmoved = 0; | |
286 | ||
287 | pagevec_lru_move_fn(pvec, pagevec_move_tail_fn, &pgmoved); | |
288 | __count_vm_events(PGROTATED, pgmoved); | |
289 | } | |
290 | ||
1da177e4 LT |
291 | /* |
292 | * Writeback is about to end against a page which has been marked for immediate | |
293 | * reclaim. If it still appears to be reclaimable, move it to the tail of the | |
902aaed0 | 294 | * inactive list. |
1da177e4 | 295 | */ |
3dd7ae8e | 296 | void rotate_reclaimable_page(struct page *page) |
1da177e4 | 297 | { |
ac6aadb2 | 298 | if (!PageLocked(page) && !PageDirty(page) && !PageActive(page) && |
894bc310 | 299 | !PageUnevictable(page) && PageLRU(page)) { |
ac6aadb2 MS |
300 | struct pagevec *pvec; |
301 | unsigned long flags; | |
302 | ||
303 | page_cache_get(page); | |
304 | local_irq_save(flags); | |
305 | pvec = &__get_cpu_var(lru_rotate_pvecs); | |
306 | if (!pagevec_add(pvec, page)) | |
307 | pagevec_move_tail(pvec); | |
308 | local_irq_restore(flags); | |
309 | } | |
1da177e4 LT |
310 | } |
311 | ||
3e2f41f1 KM |
312 | static void update_page_reclaim_stat(struct zone *zone, struct page *page, |
313 | int file, int rotated) | |
314 | { | |
315 | struct zone_reclaim_stat *reclaim_stat = &zone->reclaim_stat; | |
316 | struct zone_reclaim_stat *memcg_reclaim_stat; | |
317 | ||
318 | memcg_reclaim_stat = mem_cgroup_get_reclaim_stat_from_page(page); | |
319 | ||
320 | reclaim_stat->recent_scanned[file]++; | |
321 | if (rotated) | |
322 | reclaim_stat->recent_rotated[file]++; | |
323 | ||
324 | if (!memcg_reclaim_stat) | |
325 | return; | |
326 | ||
327 | memcg_reclaim_stat->recent_scanned[file]++; | |
328 | if (rotated) | |
329 | memcg_reclaim_stat->recent_rotated[file]++; | |
330 | } | |
331 | ||
eb709b0d | 332 | static void __activate_page(struct page *page, void *arg) |
1da177e4 LT |
333 | { |
334 | struct zone *zone = page_zone(page); | |
744ed144 | 335 | |
744ed144 | 336 | if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) { |
7a608572 LT |
337 | int file = page_is_file_cache(page); |
338 | int lru = page_lru_base_type(page); | |
339 | del_page_from_lru_list(zone, page, lru); | |
744ed144 | 340 | |
7a608572 LT |
341 | SetPageActive(page); |
342 | lru += LRU_ACTIVE; | |
343 | add_page_to_lru_list(zone, page, lru); | |
344 | __count_vm_event(PGACTIVATE); | |
4f98a2fe | 345 | |
7a608572 | 346 | update_page_reclaim_stat(zone, page, file, 1); |
1da177e4 | 347 | } |
eb709b0d SL |
348 | } |
349 | ||
350 | #ifdef CONFIG_SMP | |
351 | static DEFINE_PER_CPU(struct pagevec, activate_page_pvecs); | |
352 | ||
353 | static void activate_page_drain(int cpu) | |
354 | { | |
355 | struct pagevec *pvec = &per_cpu(activate_page_pvecs, cpu); | |
356 | ||
357 | if (pagevec_count(pvec)) | |
358 | pagevec_lru_move_fn(pvec, __activate_page, NULL); | |
359 | } | |
360 | ||
361 | void activate_page(struct page *page) | |
362 | { | |
363 | if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) { | |
364 | struct pagevec *pvec = &get_cpu_var(activate_page_pvecs); | |
365 | ||
366 | page_cache_get(page); | |
367 | if (!pagevec_add(pvec, page)) | |
368 | pagevec_lru_move_fn(pvec, __activate_page, NULL); | |
369 | put_cpu_var(activate_page_pvecs); | |
370 | } | |
371 | } | |
372 | ||
373 | #else | |
374 | static inline void activate_page_drain(int cpu) | |
375 | { | |
376 | } | |
377 | ||
378 | void activate_page(struct page *page) | |
379 | { | |
380 | struct zone *zone = page_zone(page); | |
381 | ||
382 | spin_lock_irq(&zone->lru_lock); | |
383 | __activate_page(page, NULL); | |
1da177e4 LT |
384 | spin_unlock_irq(&zone->lru_lock); |
385 | } | |
eb709b0d | 386 | #endif |
1da177e4 LT |
387 | |
388 | /* | |
389 | * Mark a page as having seen activity. | |
390 | * | |
391 | * inactive,unreferenced -> inactive,referenced | |
392 | * inactive,referenced -> active,unreferenced | |
393 | * active,unreferenced -> active,referenced | |
394 | */ | |
920c7a5d | 395 | void mark_page_accessed(struct page *page) |
1da177e4 | 396 | { |
894bc310 LS |
397 | if (!PageActive(page) && !PageUnevictable(page) && |
398 | PageReferenced(page) && PageLRU(page)) { | |
1da177e4 LT |
399 | activate_page(page); |
400 | ClearPageReferenced(page); | |
401 | } else if (!PageReferenced(page)) { | |
402 | SetPageReferenced(page); | |
403 | } | |
404 | } | |
1da177e4 LT |
405 | EXPORT_SYMBOL(mark_page_accessed); |
406 | ||
f04e9ebb | 407 | void __lru_cache_add(struct page *page, enum lru_list lru) |
1da177e4 | 408 | { |
f04e9ebb | 409 | struct pagevec *pvec = &get_cpu_var(lru_add_pvecs)[lru]; |
1da177e4 LT |
410 | |
411 | page_cache_get(page); | |
412 | if (!pagevec_add(pvec, page)) | |
5095ae83 | 413 | __pagevec_lru_add(pvec, lru); |
1da177e4 LT |
414 | put_cpu_var(lru_add_pvecs); |
415 | } | |
47846b06 | 416 | EXPORT_SYMBOL(__lru_cache_add); |
1da177e4 | 417 | |
f04e9ebb KM |
418 | /** |
419 | * lru_cache_add_lru - add a page to a page list | |
420 | * @page: the page to be added to the LRU. | |
421 | * @lru: the LRU list to which the page is added. | |
422 | */ | |
423 | void lru_cache_add_lru(struct page *page, enum lru_list lru) | |
1da177e4 | 424 | { |
f04e9ebb | 425 | if (PageActive(page)) { |
894bc310 | 426 | VM_BUG_ON(PageUnevictable(page)); |
f04e9ebb | 427 | ClearPageActive(page); |
894bc310 LS |
428 | } else if (PageUnevictable(page)) { |
429 | VM_BUG_ON(PageActive(page)); | |
430 | ClearPageUnevictable(page); | |
f04e9ebb | 431 | } |
1da177e4 | 432 | |
894bc310 | 433 | VM_BUG_ON(PageLRU(page) || PageActive(page) || PageUnevictable(page)); |
f04e9ebb | 434 | __lru_cache_add(page, lru); |
1da177e4 LT |
435 | } |
436 | ||
894bc310 LS |
437 | /** |
438 | * add_page_to_unevictable_list - add a page to the unevictable list | |
439 | * @page: the page to be added to the unevictable list | |
440 | * | |
441 | * Add page directly to its zone's unevictable list. To avoid races with | |
442 | * tasks that might be making the page evictable, through eg. munlock, | |
443 | * munmap or exit, while it's not on the lru, we want to add the page | |
444 | * while it's locked or otherwise "invisible" to other tasks. This is | |
445 | * difficult to do when using the pagevec cache, so bypass that. | |
446 | */ | |
447 | void add_page_to_unevictable_list(struct page *page) | |
448 | { | |
449 | struct zone *zone = page_zone(page); | |
450 | ||
451 | spin_lock_irq(&zone->lru_lock); | |
452 | SetPageUnevictable(page); | |
453 | SetPageLRU(page); | |
454 | add_page_to_lru_list(zone, page, LRU_UNEVICTABLE); | |
455 | spin_unlock_irq(&zone->lru_lock); | |
456 | } | |
457 | ||
31560180 MK |
458 | /* |
459 | * If the page can not be invalidated, it is moved to the | |
460 | * inactive list to speed up its reclaim. It is moved to the | |
461 | * head of the list, rather than the tail, to give the flusher | |
462 | * threads some time to write it out, as this is much more | |
463 | * effective than the single-page writeout from reclaim. | |
278df9f4 MK |
464 | * |
465 | * If the page isn't page_mapped and dirty/writeback, the page | |
466 | * could reclaim asap using PG_reclaim. | |
467 | * | |
468 | * 1. active, mapped page -> none | |
469 | * 2. active, dirty/writeback page -> inactive, head, PG_reclaim | |
470 | * 3. inactive, mapped page -> none | |
471 | * 4. inactive, dirty/writeback page -> inactive, head, PG_reclaim | |
472 | * 5. inactive, clean -> inactive, tail | |
473 | * 6. Others -> none | |
474 | * | |
475 | * In 4, why it moves inactive's head, the VM expects the page would | |
476 | * be write it out by flusher threads as this is much more effective | |
477 | * than the single-page writeout from reclaim. | |
31560180 | 478 | */ |
3dd7ae8e | 479 | static void lru_deactivate_fn(struct page *page, void *arg) |
31560180 MK |
480 | { |
481 | int lru, file; | |
278df9f4 | 482 | bool active; |
3dd7ae8e | 483 | struct zone *zone = page_zone(page); |
31560180 | 484 | |
278df9f4 | 485 | if (!PageLRU(page)) |
31560180 MK |
486 | return; |
487 | ||
bad49d9c MK |
488 | if (PageUnevictable(page)) |
489 | return; | |
490 | ||
31560180 MK |
491 | /* Some processes are using the page */ |
492 | if (page_mapped(page)) | |
493 | return; | |
494 | ||
278df9f4 MK |
495 | active = PageActive(page); |
496 | ||
31560180 MK |
497 | file = page_is_file_cache(page); |
498 | lru = page_lru_base_type(page); | |
278df9f4 | 499 | del_page_from_lru_list(zone, page, lru + active); |
31560180 MK |
500 | ClearPageActive(page); |
501 | ClearPageReferenced(page); | |
502 | add_page_to_lru_list(zone, page, lru); | |
31560180 | 503 | |
278df9f4 MK |
504 | if (PageWriteback(page) || PageDirty(page)) { |
505 | /* | |
506 | * PG_reclaim could be raced with end_page_writeback | |
507 | * It can make readahead confusing. But race window | |
508 | * is _really_ small and it's non-critical problem. | |
509 | */ | |
510 | SetPageReclaim(page); | |
511 | } else { | |
925b7673 | 512 | struct lruvec *lruvec; |
278df9f4 MK |
513 | /* |
514 | * The page's writeback ends up during pagevec | |
515 | * We moves tha page into tail of inactive. | |
516 | */ | |
925b7673 JW |
517 | lruvec = mem_cgroup_lru_move_lists(zone, page, lru, lru); |
518 | list_move_tail(&page->lru, &lruvec->lists[lru]); | |
278df9f4 MK |
519 | __count_vm_event(PGROTATED); |
520 | } | |
521 | ||
522 | if (active) | |
523 | __count_vm_event(PGDEACTIVATE); | |
31560180 MK |
524 | update_page_reclaim_stat(zone, page, file, 0); |
525 | } | |
526 | ||
902aaed0 HH |
527 | /* |
528 | * Drain pages out of the cpu's pagevecs. | |
529 | * Either "cpu" is the current CPU, and preemption has already been | |
530 | * disabled; or "cpu" is being hot-unplugged, and is already dead. | |
531 | */ | |
f0cb3c76 | 532 | void lru_add_drain_cpu(int cpu) |
1da177e4 | 533 | { |
f04e9ebb | 534 | struct pagevec *pvecs = per_cpu(lru_add_pvecs, cpu); |
902aaed0 | 535 | struct pagevec *pvec; |
f04e9ebb | 536 | int lru; |
1da177e4 | 537 | |
f04e9ebb KM |
538 | for_each_lru(lru) { |
539 | pvec = &pvecs[lru - LRU_BASE]; | |
540 | if (pagevec_count(pvec)) | |
5095ae83 | 541 | __pagevec_lru_add(pvec, lru); |
f04e9ebb | 542 | } |
902aaed0 HH |
543 | |
544 | pvec = &per_cpu(lru_rotate_pvecs, cpu); | |
545 | if (pagevec_count(pvec)) { | |
546 | unsigned long flags; | |
547 | ||
548 | /* No harm done if a racing interrupt already did this */ | |
549 | local_irq_save(flags); | |
550 | pagevec_move_tail(pvec); | |
551 | local_irq_restore(flags); | |
552 | } | |
31560180 MK |
553 | |
554 | pvec = &per_cpu(lru_deactivate_pvecs, cpu); | |
555 | if (pagevec_count(pvec)) | |
3dd7ae8e | 556 | pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL); |
eb709b0d SL |
557 | |
558 | activate_page_drain(cpu); | |
31560180 MK |
559 | } |
560 | ||
561 | /** | |
562 | * deactivate_page - forcefully deactivate a page | |
563 | * @page: page to deactivate | |
564 | * | |
565 | * This function hints the VM that @page is a good reclaim candidate, | |
566 | * for example if its invalidation fails due to the page being dirty | |
567 | * or under writeback. | |
568 | */ | |
569 | void deactivate_page(struct page *page) | |
570 | { | |
821ed6bb MK |
571 | /* |
572 | * In a workload with many unevictable page such as mprotect, unevictable | |
573 | * page deactivation for accelerating reclaim is pointless. | |
574 | */ | |
575 | if (PageUnevictable(page)) | |
576 | return; | |
577 | ||
31560180 MK |
578 | if (likely(get_page_unless_zero(page))) { |
579 | struct pagevec *pvec = &get_cpu_var(lru_deactivate_pvecs); | |
580 | ||
581 | if (!pagevec_add(pvec, page)) | |
3dd7ae8e | 582 | pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL); |
31560180 MK |
583 | put_cpu_var(lru_deactivate_pvecs); |
584 | } | |
80bfed90 AM |
585 | } |
586 | ||
587 | void lru_add_drain(void) | |
588 | { | |
f0cb3c76 | 589 | lru_add_drain_cpu(get_cpu()); |
80bfed90 | 590 | put_cpu(); |
1da177e4 LT |
591 | } |
592 | ||
c4028958 | 593 | static void lru_add_drain_per_cpu(struct work_struct *dummy) |
053837fc NP |
594 | { |
595 | lru_add_drain(); | |
596 | } | |
597 | ||
598 | /* | |
599 | * Returns 0 for success | |
600 | */ | |
601 | int lru_add_drain_all(void) | |
602 | { | |
c4028958 | 603 | return schedule_on_each_cpu(lru_add_drain_per_cpu); |
053837fc NP |
604 | } |
605 | ||
1da177e4 LT |
606 | /* |
607 | * Batched page_cache_release(). Decrement the reference count on all the | |
608 | * passed pages. If it fell to zero then remove the page from the LRU and | |
609 | * free it. | |
610 | * | |
611 | * Avoid taking zone->lru_lock if possible, but if it is taken, retain it | |
612 | * for the remainder of the operation. | |
613 | * | |
ab33dc09 FLVC |
614 | * The locking in this function is against shrink_inactive_list(): we recheck |
615 | * the page count inside the lock to see whether shrink_inactive_list() | |
616 | * grabbed the page via the LRU. If it did, give up: shrink_inactive_list() | |
617 | * will free it. | |
1da177e4 LT |
618 | */ |
619 | void release_pages(struct page **pages, int nr, int cold) | |
620 | { | |
621 | int i; | |
cc59850e | 622 | LIST_HEAD(pages_to_free); |
1da177e4 | 623 | struct zone *zone = NULL; |
902aaed0 | 624 | unsigned long uninitialized_var(flags); |
1da177e4 | 625 | |
1da177e4 LT |
626 | for (i = 0; i < nr; i++) { |
627 | struct page *page = pages[i]; | |
1da177e4 | 628 | |
8519fb30 NP |
629 | if (unlikely(PageCompound(page))) { |
630 | if (zone) { | |
902aaed0 | 631 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
8519fb30 NP |
632 | zone = NULL; |
633 | } | |
634 | put_compound_page(page); | |
635 | continue; | |
636 | } | |
637 | ||
b5810039 | 638 | if (!put_page_testzero(page)) |
1da177e4 LT |
639 | continue; |
640 | ||
46453a6e NP |
641 | if (PageLRU(page)) { |
642 | struct zone *pagezone = page_zone(page); | |
894bc310 | 643 | |
46453a6e NP |
644 | if (pagezone != zone) { |
645 | if (zone) | |
902aaed0 HH |
646 | spin_unlock_irqrestore(&zone->lru_lock, |
647 | flags); | |
46453a6e | 648 | zone = pagezone; |
902aaed0 | 649 | spin_lock_irqsave(&zone->lru_lock, flags); |
46453a6e | 650 | } |
725d704e | 651 | VM_BUG_ON(!PageLRU(page)); |
67453911 | 652 | __ClearPageLRU(page); |
1c1c53d4 | 653 | del_page_from_lru_list(zone, page, page_off_lru(page)); |
46453a6e NP |
654 | } |
655 | ||
cc59850e | 656 | list_add(&page->lru, &pages_to_free); |
1da177e4 LT |
657 | } |
658 | if (zone) | |
902aaed0 | 659 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
1da177e4 | 660 | |
cc59850e | 661 | free_hot_cold_page_list(&pages_to_free, cold); |
1da177e4 | 662 | } |
0be8557b | 663 | EXPORT_SYMBOL(release_pages); |
1da177e4 LT |
664 | |
665 | /* | |
666 | * The pages which we're about to release may be in the deferred lru-addition | |
667 | * queues. That would prevent them from really being freed right now. That's | |
668 | * OK from a correctness point of view but is inefficient - those pages may be | |
669 | * cache-warm and we want to give them back to the page allocator ASAP. | |
670 | * | |
671 | * So __pagevec_release() will drain those queues here. __pagevec_lru_add() | |
672 | * and __pagevec_lru_add_active() call release_pages() directly to avoid | |
673 | * mutual recursion. | |
674 | */ | |
675 | void __pagevec_release(struct pagevec *pvec) | |
676 | { | |
677 | lru_add_drain(); | |
678 | release_pages(pvec->pages, pagevec_count(pvec), pvec->cold); | |
679 | pagevec_reinit(pvec); | |
680 | } | |
7f285701 SF |
681 | EXPORT_SYMBOL(__pagevec_release); |
682 | ||
12d27107 | 683 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
71e3aac0 AA |
684 | /* used by __split_huge_page_refcount() */ |
685 | void lru_add_page_tail(struct zone* zone, | |
686 | struct page *page, struct page *page_tail) | |
687 | { | |
7512102c | 688 | int uninitialized_var(active); |
71e3aac0 AA |
689 | enum lru_list lru; |
690 | const int file = 0; | |
71e3aac0 AA |
691 | |
692 | VM_BUG_ON(!PageHead(page)); | |
693 | VM_BUG_ON(PageCompound(page_tail)); | |
694 | VM_BUG_ON(PageLRU(page_tail)); | |
b9980cdc | 695 | VM_BUG_ON(NR_CPUS != 1 && !spin_is_locked(&zone->lru_lock)); |
71e3aac0 AA |
696 | |
697 | SetPageLRU(page_tail); | |
698 | ||
699 | if (page_evictable(page_tail, NULL)) { | |
700 | if (PageActive(page)) { | |
701 | SetPageActive(page_tail); | |
702 | active = 1; | |
703 | lru = LRU_ACTIVE_ANON; | |
704 | } else { | |
705 | active = 0; | |
706 | lru = LRU_INACTIVE_ANON; | |
707 | } | |
71e3aac0 AA |
708 | } else { |
709 | SetPageUnevictable(page_tail); | |
12d27107 HD |
710 | lru = LRU_UNEVICTABLE; |
711 | } | |
712 | ||
713 | if (likely(PageLRU(page))) | |
714 | list_add_tail(&page_tail->lru, &page->lru); | |
715 | else { | |
716 | struct list_head *list_head; | |
717 | /* | |
718 | * Head page has not yet been counted, as an hpage, | |
719 | * so we must account for each subpage individually. | |
720 | * | |
721 | * Use the standard add function to put page_tail on the list, | |
722 | * but then correct its position so they all end up in order. | |
723 | */ | |
724 | add_page_to_lru_list(zone, page_tail, lru); | |
725 | list_head = page_tail->lru.prev; | |
726 | list_move_tail(&page_tail->lru, list_head); | |
71e3aac0 | 727 | } |
7512102c HD |
728 | |
729 | if (!PageUnevictable(page)) | |
730 | update_page_reclaim_stat(zone, page_tail, file, active); | |
71e3aac0 | 731 | } |
12d27107 | 732 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
71e3aac0 | 733 | |
5095ae83 | 734 | static void __pagevec_lru_add_fn(struct page *page, void *arg) |
3dd7ae8e SL |
735 | { |
736 | enum lru_list lru = (enum lru_list)arg; | |
737 | struct zone *zone = page_zone(page); | |
738 | int file = is_file_lru(lru); | |
739 | int active = is_active_lru(lru); | |
740 | ||
741 | VM_BUG_ON(PageActive(page)); | |
742 | VM_BUG_ON(PageUnevictable(page)); | |
743 | VM_BUG_ON(PageLRU(page)); | |
744 | ||
745 | SetPageLRU(page); | |
746 | if (active) | |
747 | SetPageActive(page); | |
3dd7ae8e | 748 | add_page_to_lru_list(zone, page, lru); |
7512102c | 749 | update_page_reclaim_stat(zone, page, file, active); |
3dd7ae8e SL |
750 | } |
751 | ||
1da177e4 LT |
752 | /* |
753 | * Add the passed pages to the LRU, then drop the caller's refcount | |
754 | * on them. Reinitialises the caller's pagevec. | |
755 | */ | |
5095ae83 | 756 | void __pagevec_lru_add(struct pagevec *pvec, enum lru_list lru) |
1da177e4 | 757 | { |
894bc310 | 758 | VM_BUG_ON(is_unevictable_lru(lru)); |
1da177e4 | 759 | |
5095ae83 | 760 | pagevec_lru_move_fn(pvec, __pagevec_lru_add_fn, (void *)lru); |
1da177e4 | 761 | } |
5095ae83 | 762 | EXPORT_SYMBOL(__pagevec_lru_add); |
1da177e4 | 763 | |
1da177e4 LT |
764 | /** |
765 | * pagevec_lookup - gang pagecache lookup | |
766 | * @pvec: Where the resulting pages are placed | |
767 | * @mapping: The address_space to search | |
768 | * @start: The starting page index | |
769 | * @nr_pages: The maximum number of pages | |
770 | * | |
771 | * pagevec_lookup() will search for and return a group of up to @nr_pages pages | |
772 | * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a | |
773 | * reference against the pages in @pvec. | |
774 | * | |
775 | * The search returns a group of mapping-contiguous pages with ascending | |
776 | * indexes. There may be holes in the indices due to not-present pages. | |
777 | * | |
778 | * pagevec_lookup() returns the number of pages which were found. | |
779 | */ | |
780 | unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping, | |
781 | pgoff_t start, unsigned nr_pages) | |
782 | { | |
783 | pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages); | |
784 | return pagevec_count(pvec); | |
785 | } | |
78539fdf CH |
786 | EXPORT_SYMBOL(pagevec_lookup); |
787 | ||
1da177e4 LT |
788 | unsigned pagevec_lookup_tag(struct pagevec *pvec, struct address_space *mapping, |
789 | pgoff_t *index, int tag, unsigned nr_pages) | |
790 | { | |
791 | pvec->nr = find_get_pages_tag(mapping, index, tag, | |
792 | nr_pages, pvec->pages); | |
793 | return pagevec_count(pvec); | |
794 | } | |
7f285701 | 795 | EXPORT_SYMBOL(pagevec_lookup_tag); |
1da177e4 | 796 | |
1da177e4 LT |
797 | /* |
798 | * Perform any setup for the swap system | |
799 | */ | |
800 | void __init swap_setup(void) | |
801 | { | |
4481374c | 802 | unsigned long megs = totalram_pages >> (20 - PAGE_SHIFT); |
1da177e4 | 803 | |
e0bf68dd PZ |
804 | #ifdef CONFIG_SWAP |
805 | bdi_init(swapper_space.backing_dev_info); | |
806 | #endif | |
807 | ||
1da177e4 LT |
808 | /* Use a smaller cluster for small-memory machines */ |
809 | if (megs < 16) | |
810 | page_cluster = 2; | |
811 | else | |
812 | page_cluster = 3; | |
813 | /* | |
814 | * Right now other parts of the system means that we | |
815 | * _really_ don't want to cluster much more | |
816 | */ | |
1da177e4 | 817 | } |