]>
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)) { | |
b221385b | 50 | struct zone *zone = page_zone(page); |
fa9add64 HD |
51 | struct lruvec *lruvec; |
52 | unsigned long flags; | |
b221385b AB |
53 | |
54 | spin_lock_irqsave(&zone->lru_lock, flags); | |
fa9add64 | 55 | lruvec = mem_cgroup_page_lruvec(page, zone); |
b221385b AB |
56 | VM_BUG_ON(!PageLRU(page)); |
57 | __ClearPageLRU(page); | |
fa9add64 | 58 | del_page_from_lru_list(page, lruvec, page_off_lru(page)); |
b221385b AB |
59 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
60 | } | |
91807063 AA |
61 | } |
62 | ||
63 | static void __put_single_page(struct page *page) | |
64 | { | |
65 | __page_cache_release(page); | |
fc91668e | 66 | free_hot_cold_page(page, 0); |
b221385b AB |
67 | } |
68 | ||
91807063 | 69 | static void __put_compound_page(struct page *page) |
1da177e4 | 70 | { |
91807063 | 71 | compound_page_dtor *dtor; |
1da177e4 | 72 | |
91807063 AA |
73 | __page_cache_release(page); |
74 | dtor = get_compound_page_dtor(page); | |
75 | (*dtor)(page); | |
76 | } | |
77 | ||
78 | static void put_compound_page(struct page *page) | |
79 | { | |
80 | if (unlikely(PageTail(page))) { | |
81 | /* __split_huge_page_refcount can run under us */ | |
70b50f94 AA |
82 | struct page *page_head = compound_trans_head(page); |
83 | ||
84 | if (likely(page != page_head && | |
85 | get_page_unless_zero(page_head))) { | |
91807063 | 86 | unsigned long flags; |
5bf5f03c PS |
87 | |
88 | /* | |
89 | * THP can not break up slab pages so avoid taking | |
90 | * compound_lock(). Slab performs non-atomic bit ops | |
91 | * on page->flags for better performance. In particular | |
92 | * slab_unlock() in slub used to be a hot path. It is | |
93 | * still hot on arches that do not support | |
94 | * this_cpu_cmpxchg_double(). | |
95 | */ | |
96 | if (PageSlab(page_head)) { | |
97 | if (PageTail(page)) { | |
98 | if (put_page_testzero(page_head)) | |
99 | VM_BUG_ON(1); | |
100 | ||
101 | atomic_dec(&page->_mapcount); | |
102 | goto skip_lock_tail; | |
103 | } else | |
104 | goto skip_lock; | |
105 | } | |
91807063 | 106 | /* |
70b50f94 AA |
107 | * page_head wasn't a dangling pointer but it |
108 | * may not be a head page anymore by the time | |
109 | * we obtain the lock. That is ok as long as it | |
110 | * can't be freed from under us. | |
91807063 | 111 | */ |
91807063 AA |
112 | flags = compound_lock_irqsave(page_head); |
113 | if (unlikely(!PageTail(page))) { | |
114 | /* __split_huge_page_refcount run before us */ | |
115 | compound_unlock_irqrestore(page_head, flags); | |
5bf5f03c | 116 | skip_lock: |
91807063 AA |
117 | if (put_page_testzero(page_head)) |
118 | __put_single_page(page_head); | |
5bf5f03c | 119 | out_put_single: |
91807063 AA |
120 | if (put_page_testzero(page)) |
121 | __put_single_page(page); | |
122 | return; | |
123 | } | |
124 | VM_BUG_ON(page_head != page->first_page); | |
125 | /* | |
126 | * We can release the refcount taken by | |
70b50f94 AA |
127 | * get_page_unless_zero() now that |
128 | * __split_huge_page_refcount() is blocked on | |
129 | * the compound_lock. | |
91807063 AA |
130 | */ |
131 | if (put_page_testzero(page_head)) | |
132 | VM_BUG_ON(1); | |
133 | /* __split_huge_page_refcount will wait now */ | |
70b50f94 AA |
134 | VM_BUG_ON(page_mapcount(page) <= 0); |
135 | atomic_dec(&page->_mapcount); | |
91807063 | 136 | VM_BUG_ON(atomic_read(&page_head->_count) <= 0); |
70b50f94 | 137 | VM_BUG_ON(atomic_read(&page->_count) != 0); |
91807063 | 138 | compound_unlock_irqrestore(page_head, flags); |
5bf5f03c PS |
139 | |
140 | skip_lock_tail: | |
a95a82e9 AA |
141 | if (put_page_testzero(page_head)) { |
142 | if (PageHead(page_head)) | |
143 | __put_compound_page(page_head); | |
144 | else | |
145 | __put_single_page(page_head); | |
146 | } | |
91807063 AA |
147 | } else { |
148 | /* page_head is a dangling pointer */ | |
149 | VM_BUG_ON(PageTail(page)); | |
150 | goto out_put_single; | |
151 | } | |
152 | } else if (put_page_testzero(page)) { | |
153 | if (PageHead(page)) | |
154 | __put_compound_page(page); | |
155 | else | |
156 | __put_single_page(page); | |
1da177e4 | 157 | } |
8519fb30 NP |
158 | } |
159 | ||
160 | void put_page(struct page *page) | |
161 | { | |
162 | if (unlikely(PageCompound(page))) | |
163 | put_compound_page(page); | |
164 | else if (put_page_testzero(page)) | |
91807063 | 165 | __put_single_page(page); |
1da177e4 LT |
166 | } |
167 | EXPORT_SYMBOL(put_page); | |
1da177e4 | 168 | |
70b50f94 AA |
169 | /* |
170 | * This function is exported but must not be called by anything other | |
171 | * than get_page(). It implements the slow path of get_page(). | |
172 | */ | |
173 | bool __get_page_tail(struct page *page) | |
174 | { | |
175 | /* | |
176 | * This takes care of get_page() if run on a tail page | |
177 | * returned by one of the get_user_pages/follow_page variants. | |
178 | * get_user_pages/follow_page itself doesn't need the compound | |
179 | * lock because it runs __get_page_tail_foll() under the | |
180 | * proper PT lock that already serializes against | |
181 | * split_huge_page(). | |
182 | */ | |
183 | unsigned long flags; | |
184 | bool got = false; | |
185 | struct page *page_head = compound_trans_head(page); | |
186 | ||
187 | if (likely(page != page_head && get_page_unless_zero(page_head))) { | |
5bf5f03c PS |
188 | |
189 | /* Ref to put_compound_page() comment. */ | |
190 | if (PageSlab(page_head)) { | |
191 | if (likely(PageTail(page))) { | |
192 | __get_page_tail_foll(page, false); | |
193 | return true; | |
194 | } else { | |
195 | put_page(page_head); | |
196 | return false; | |
197 | } | |
198 | } | |
199 | ||
70b50f94 AA |
200 | /* |
201 | * page_head wasn't a dangling pointer but it | |
202 | * may not be a head page anymore by the time | |
203 | * we obtain the lock. That is ok as long as it | |
204 | * can't be freed from under us. | |
205 | */ | |
206 | flags = compound_lock_irqsave(page_head); | |
207 | /* here __split_huge_page_refcount won't run anymore */ | |
208 | if (likely(PageTail(page))) { | |
209 | __get_page_tail_foll(page, false); | |
210 | got = true; | |
211 | } | |
212 | compound_unlock_irqrestore(page_head, flags); | |
213 | if (unlikely(!got)) | |
214 | put_page(page_head); | |
215 | } | |
216 | return got; | |
217 | } | |
218 | EXPORT_SYMBOL(__get_page_tail); | |
219 | ||
1d7ea732 | 220 | /** |
7682486b RD |
221 | * put_pages_list() - release a list of pages |
222 | * @pages: list of pages threaded on page->lru | |
1d7ea732 AZ |
223 | * |
224 | * Release a list of pages which are strung together on page.lru. Currently | |
225 | * used by read_cache_pages() and related error recovery code. | |
1d7ea732 AZ |
226 | */ |
227 | void put_pages_list(struct list_head *pages) | |
228 | { | |
229 | while (!list_empty(pages)) { | |
230 | struct page *victim; | |
231 | ||
232 | victim = list_entry(pages->prev, struct page, lru); | |
233 | list_del(&victim->lru); | |
234 | page_cache_release(victim); | |
235 | } | |
236 | } | |
237 | EXPORT_SYMBOL(put_pages_list); | |
238 | ||
3dd7ae8e | 239 | static void pagevec_lru_move_fn(struct pagevec *pvec, |
fa9add64 HD |
240 | void (*move_fn)(struct page *page, struct lruvec *lruvec, void *arg), |
241 | void *arg) | |
902aaed0 HH |
242 | { |
243 | int i; | |
902aaed0 | 244 | struct zone *zone = NULL; |
fa9add64 | 245 | struct lruvec *lruvec; |
3dd7ae8e | 246 | unsigned long flags = 0; |
902aaed0 HH |
247 | |
248 | for (i = 0; i < pagevec_count(pvec); i++) { | |
249 | struct page *page = pvec->pages[i]; | |
250 | struct zone *pagezone = page_zone(page); | |
251 | ||
252 | if (pagezone != zone) { | |
253 | if (zone) | |
3dd7ae8e | 254 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
902aaed0 | 255 | zone = pagezone; |
3dd7ae8e | 256 | spin_lock_irqsave(&zone->lru_lock, flags); |
902aaed0 | 257 | } |
3dd7ae8e | 258 | |
fa9add64 HD |
259 | lruvec = mem_cgroup_page_lruvec(page, zone); |
260 | (*move_fn)(page, lruvec, arg); | |
902aaed0 HH |
261 | } |
262 | if (zone) | |
3dd7ae8e | 263 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
83896fb5 LT |
264 | release_pages(pvec->pages, pvec->nr, pvec->cold); |
265 | pagevec_reinit(pvec); | |
d8505dee SL |
266 | } |
267 | ||
fa9add64 HD |
268 | static void pagevec_move_tail_fn(struct page *page, struct lruvec *lruvec, |
269 | void *arg) | |
3dd7ae8e SL |
270 | { |
271 | int *pgmoved = arg; | |
3dd7ae8e SL |
272 | |
273 | if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) { | |
274 | enum lru_list lru = page_lru_base_type(page); | |
925b7673 | 275 | list_move_tail(&page->lru, &lruvec->lists[lru]); |
3dd7ae8e SL |
276 | (*pgmoved)++; |
277 | } | |
278 | } | |
279 | ||
280 | /* | |
281 | * pagevec_move_tail() must be called with IRQ disabled. | |
282 | * Otherwise this may cause nasty races. | |
283 | */ | |
284 | static void pagevec_move_tail(struct pagevec *pvec) | |
285 | { | |
286 | int pgmoved = 0; | |
287 | ||
288 | pagevec_lru_move_fn(pvec, pagevec_move_tail_fn, &pgmoved); | |
289 | __count_vm_events(PGROTATED, pgmoved); | |
290 | } | |
291 | ||
1da177e4 LT |
292 | /* |
293 | * Writeback is about to end against a page which has been marked for immediate | |
294 | * reclaim. If it still appears to be reclaimable, move it to the tail of the | |
902aaed0 | 295 | * inactive list. |
1da177e4 | 296 | */ |
3dd7ae8e | 297 | void rotate_reclaimable_page(struct page *page) |
1da177e4 | 298 | { |
ac6aadb2 | 299 | if (!PageLocked(page) && !PageDirty(page) && !PageActive(page) && |
894bc310 | 300 | !PageUnevictable(page) && PageLRU(page)) { |
ac6aadb2 MS |
301 | struct pagevec *pvec; |
302 | unsigned long flags; | |
303 | ||
304 | page_cache_get(page); | |
305 | local_irq_save(flags); | |
306 | pvec = &__get_cpu_var(lru_rotate_pvecs); | |
307 | if (!pagevec_add(pvec, page)) | |
308 | pagevec_move_tail(pvec); | |
309 | local_irq_restore(flags); | |
310 | } | |
1da177e4 LT |
311 | } |
312 | ||
fa9add64 | 313 | static void update_page_reclaim_stat(struct lruvec *lruvec, |
3e2f41f1 KM |
314 | int file, int rotated) |
315 | { | |
fa9add64 | 316 | struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat; |
3e2f41f1 KM |
317 | |
318 | reclaim_stat->recent_scanned[file]++; | |
319 | if (rotated) | |
320 | reclaim_stat->recent_rotated[file]++; | |
3e2f41f1 KM |
321 | } |
322 | ||
fa9add64 HD |
323 | static void __activate_page(struct page *page, struct lruvec *lruvec, |
324 | void *arg) | |
1da177e4 | 325 | { |
744ed144 | 326 | if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) { |
7a608572 LT |
327 | int file = page_is_file_cache(page); |
328 | int lru = page_lru_base_type(page); | |
744ed144 | 329 | |
fa9add64 | 330 | del_page_from_lru_list(page, lruvec, lru); |
7a608572 LT |
331 | SetPageActive(page); |
332 | lru += LRU_ACTIVE; | |
fa9add64 | 333 | add_page_to_lru_list(page, lruvec, lru); |
4f98a2fe | 334 | |
fa9add64 HD |
335 | __count_vm_event(PGACTIVATE); |
336 | update_page_reclaim_stat(lruvec, file, 1); | |
1da177e4 | 337 | } |
eb709b0d SL |
338 | } |
339 | ||
340 | #ifdef CONFIG_SMP | |
341 | static DEFINE_PER_CPU(struct pagevec, activate_page_pvecs); | |
342 | ||
343 | static void activate_page_drain(int cpu) | |
344 | { | |
345 | struct pagevec *pvec = &per_cpu(activate_page_pvecs, cpu); | |
346 | ||
347 | if (pagevec_count(pvec)) | |
348 | pagevec_lru_move_fn(pvec, __activate_page, NULL); | |
349 | } | |
350 | ||
351 | void activate_page(struct page *page) | |
352 | { | |
353 | if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) { | |
354 | struct pagevec *pvec = &get_cpu_var(activate_page_pvecs); | |
355 | ||
356 | page_cache_get(page); | |
357 | if (!pagevec_add(pvec, page)) | |
358 | pagevec_lru_move_fn(pvec, __activate_page, NULL); | |
359 | put_cpu_var(activate_page_pvecs); | |
360 | } | |
361 | } | |
362 | ||
363 | #else | |
364 | static inline void activate_page_drain(int cpu) | |
365 | { | |
366 | } | |
367 | ||
368 | void activate_page(struct page *page) | |
369 | { | |
370 | struct zone *zone = page_zone(page); | |
371 | ||
372 | spin_lock_irq(&zone->lru_lock); | |
fa9add64 | 373 | __activate_page(page, mem_cgroup_page_lruvec(page, zone), NULL); |
1da177e4 LT |
374 | spin_unlock_irq(&zone->lru_lock); |
375 | } | |
eb709b0d | 376 | #endif |
1da177e4 LT |
377 | |
378 | /* | |
379 | * Mark a page as having seen activity. | |
380 | * | |
381 | * inactive,unreferenced -> inactive,referenced | |
382 | * inactive,referenced -> active,unreferenced | |
383 | * active,unreferenced -> active,referenced | |
384 | */ | |
920c7a5d | 385 | void mark_page_accessed(struct page *page) |
1da177e4 | 386 | { |
894bc310 LS |
387 | if (!PageActive(page) && !PageUnevictable(page) && |
388 | PageReferenced(page) && PageLRU(page)) { | |
1da177e4 LT |
389 | activate_page(page); |
390 | ClearPageReferenced(page); | |
391 | } else if (!PageReferenced(page)) { | |
392 | SetPageReferenced(page); | |
393 | } | |
394 | } | |
1da177e4 LT |
395 | EXPORT_SYMBOL(mark_page_accessed); |
396 | ||
f04e9ebb | 397 | void __lru_cache_add(struct page *page, enum lru_list lru) |
1da177e4 | 398 | { |
f04e9ebb | 399 | struct pagevec *pvec = &get_cpu_var(lru_add_pvecs)[lru]; |
1da177e4 LT |
400 | |
401 | page_cache_get(page); | |
402 | if (!pagevec_add(pvec, page)) | |
5095ae83 | 403 | __pagevec_lru_add(pvec, lru); |
1da177e4 LT |
404 | put_cpu_var(lru_add_pvecs); |
405 | } | |
47846b06 | 406 | EXPORT_SYMBOL(__lru_cache_add); |
1da177e4 | 407 | |
f04e9ebb KM |
408 | /** |
409 | * lru_cache_add_lru - add a page to a page list | |
410 | * @page: the page to be added to the LRU. | |
411 | * @lru: the LRU list to which the page is added. | |
412 | */ | |
413 | void lru_cache_add_lru(struct page *page, enum lru_list lru) | |
1da177e4 | 414 | { |
f04e9ebb | 415 | if (PageActive(page)) { |
894bc310 | 416 | VM_BUG_ON(PageUnevictable(page)); |
f04e9ebb | 417 | ClearPageActive(page); |
894bc310 LS |
418 | } else if (PageUnevictable(page)) { |
419 | VM_BUG_ON(PageActive(page)); | |
420 | ClearPageUnevictable(page); | |
f04e9ebb | 421 | } |
1da177e4 | 422 | |
894bc310 | 423 | VM_BUG_ON(PageLRU(page) || PageActive(page) || PageUnevictable(page)); |
f04e9ebb | 424 | __lru_cache_add(page, lru); |
1da177e4 LT |
425 | } |
426 | ||
894bc310 LS |
427 | /** |
428 | * add_page_to_unevictable_list - add a page to the unevictable list | |
429 | * @page: the page to be added to the unevictable list | |
430 | * | |
431 | * Add page directly to its zone's unevictable list. To avoid races with | |
432 | * tasks that might be making the page evictable, through eg. munlock, | |
433 | * munmap or exit, while it's not on the lru, we want to add the page | |
434 | * while it's locked or otherwise "invisible" to other tasks. This is | |
435 | * difficult to do when using the pagevec cache, so bypass that. | |
436 | */ | |
437 | void add_page_to_unevictable_list(struct page *page) | |
438 | { | |
439 | struct zone *zone = page_zone(page); | |
fa9add64 | 440 | struct lruvec *lruvec; |
894bc310 LS |
441 | |
442 | spin_lock_irq(&zone->lru_lock); | |
fa9add64 | 443 | lruvec = mem_cgroup_page_lruvec(page, zone); |
894bc310 LS |
444 | SetPageUnevictable(page); |
445 | SetPageLRU(page); | |
fa9add64 | 446 | add_page_to_lru_list(page, lruvec, LRU_UNEVICTABLE); |
894bc310 LS |
447 | spin_unlock_irq(&zone->lru_lock); |
448 | } | |
449 | ||
31560180 MK |
450 | /* |
451 | * If the page can not be invalidated, it is moved to the | |
452 | * inactive list to speed up its reclaim. It is moved to the | |
453 | * head of the list, rather than the tail, to give the flusher | |
454 | * threads some time to write it out, as this is much more | |
455 | * effective than the single-page writeout from reclaim. | |
278df9f4 MK |
456 | * |
457 | * If the page isn't page_mapped and dirty/writeback, the page | |
458 | * could reclaim asap using PG_reclaim. | |
459 | * | |
460 | * 1. active, mapped page -> none | |
461 | * 2. active, dirty/writeback page -> inactive, head, PG_reclaim | |
462 | * 3. inactive, mapped page -> none | |
463 | * 4. inactive, dirty/writeback page -> inactive, head, PG_reclaim | |
464 | * 5. inactive, clean -> inactive, tail | |
465 | * 6. Others -> none | |
466 | * | |
467 | * In 4, why it moves inactive's head, the VM expects the page would | |
468 | * be write it out by flusher threads as this is much more effective | |
469 | * than the single-page writeout from reclaim. | |
31560180 | 470 | */ |
fa9add64 HD |
471 | static void lru_deactivate_fn(struct page *page, struct lruvec *lruvec, |
472 | void *arg) | |
31560180 MK |
473 | { |
474 | int lru, file; | |
278df9f4 | 475 | bool active; |
31560180 | 476 | |
278df9f4 | 477 | if (!PageLRU(page)) |
31560180 MK |
478 | return; |
479 | ||
bad49d9c MK |
480 | if (PageUnevictable(page)) |
481 | return; | |
482 | ||
31560180 MK |
483 | /* Some processes are using the page */ |
484 | if (page_mapped(page)) | |
485 | return; | |
486 | ||
278df9f4 | 487 | active = PageActive(page); |
31560180 MK |
488 | file = page_is_file_cache(page); |
489 | lru = page_lru_base_type(page); | |
fa9add64 HD |
490 | |
491 | del_page_from_lru_list(page, lruvec, lru + active); | |
31560180 MK |
492 | ClearPageActive(page); |
493 | ClearPageReferenced(page); | |
fa9add64 | 494 | add_page_to_lru_list(page, lruvec, lru); |
31560180 | 495 | |
278df9f4 MK |
496 | if (PageWriteback(page) || PageDirty(page)) { |
497 | /* | |
498 | * PG_reclaim could be raced with end_page_writeback | |
499 | * It can make readahead confusing. But race window | |
500 | * is _really_ small and it's non-critical problem. | |
501 | */ | |
502 | SetPageReclaim(page); | |
503 | } else { | |
504 | /* | |
505 | * The page's writeback ends up during pagevec | |
506 | * We moves tha page into tail of inactive. | |
507 | */ | |
925b7673 | 508 | list_move_tail(&page->lru, &lruvec->lists[lru]); |
278df9f4 MK |
509 | __count_vm_event(PGROTATED); |
510 | } | |
511 | ||
512 | if (active) | |
513 | __count_vm_event(PGDEACTIVATE); | |
fa9add64 | 514 | update_page_reclaim_stat(lruvec, file, 0); |
31560180 MK |
515 | } |
516 | ||
902aaed0 HH |
517 | /* |
518 | * Drain pages out of the cpu's pagevecs. | |
519 | * Either "cpu" is the current CPU, and preemption has already been | |
520 | * disabled; or "cpu" is being hot-unplugged, and is already dead. | |
521 | */ | |
f0cb3c76 | 522 | void lru_add_drain_cpu(int cpu) |
1da177e4 | 523 | { |
f04e9ebb | 524 | struct pagevec *pvecs = per_cpu(lru_add_pvecs, cpu); |
902aaed0 | 525 | struct pagevec *pvec; |
f04e9ebb | 526 | int lru; |
1da177e4 | 527 | |
f04e9ebb KM |
528 | for_each_lru(lru) { |
529 | pvec = &pvecs[lru - LRU_BASE]; | |
530 | if (pagevec_count(pvec)) | |
5095ae83 | 531 | __pagevec_lru_add(pvec, lru); |
f04e9ebb | 532 | } |
902aaed0 HH |
533 | |
534 | pvec = &per_cpu(lru_rotate_pvecs, cpu); | |
535 | if (pagevec_count(pvec)) { | |
536 | unsigned long flags; | |
537 | ||
538 | /* No harm done if a racing interrupt already did this */ | |
539 | local_irq_save(flags); | |
540 | pagevec_move_tail(pvec); | |
541 | local_irq_restore(flags); | |
542 | } | |
31560180 MK |
543 | |
544 | pvec = &per_cpu(lru_deactivate_pvecs, cpu); | |
545 | if (pagevec_count(pvec)) | |
3dd7ae8e | 546 | pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL); |
eb709b0d SL |
547 | |
548 | activate_page_drain(cpu); | |
31560180 MK |
549 | } |
550 | ||
551 | /** | |
552 | * deactivate_page - forcefully deactivate a page | |
553 | * @page: page to deactivate | |
554 | * | |
555 | * This function hints the VM that @page is a good reclaim candidate, | |
556 | * for example if its invalidation fails due to the page being dirty | |
557 | * or under writeback. | |
558 | */ | |
559 | void deactivate_page(struct page *page) | |
560 | { | |
821ed6bb MK |
561 | /* |
562 | * In a workload with many unevictable page such as mprotect, unevictable | |
563 | * page deactivation for accelerating reclaim is pointless. | |
564 | */ | |
565 | if (PageUnevictable(page)) | |
566 | return; | |
567 | ||
31560180 MK |
568 | if (likely(get_page_unless_zero(page))) { |
569 | struct pagevec *pvec = &get_cpu_var(lru_deactivate_pvecs); | |
570 | ||
571 | if (!pagevec_add(pvec, page)) | |
3dd7ae8e | 572 | pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL); |
31560180 MK |
573 | put_cpu_var(lru_deactivate_pvecs); |
574 | } | |
80bfed90 AM |
575 | } |
576 | ||
577 | void lru_add_drain(void) | |
578 | { | |
f0cb3c76 | 579 | lru_add_drain_cpu(get_cpu()); |
80bfed90 | 580 | put_cpu(); |
1da177e4 LT |
581 | } |
582 | ||
c4028958 | 583 | static void lru_add_drain_per_cpu(struct work_struct *dummy) |
053837fc NP |
584 | { |
585 | lru_add_drain(); | |
586 | } | |
587 | ||
588 | /* | |
589 | * Returns 0 for success | |
590 | */ | |
591 | int lru_add_drain_all(void) | |
592 | { | |
c4028958 | 593 | return schedule_on_each_cpu(lru_add_drain_per_cpu); |
053837fc NP |
594 | } |
595 | ||
1da177e4 LT |
596 | /* |
597 | * Batched page_cache_release(). Decrement the reference count on all the | |
598 | * passed pages. If it fell to zero then remove the page from the LRU and | |
599 | * free it. | |
600 | * | |
601 | * Avoid taking zone->lru_lock if possible, but if it is taken, retain it | |
602 | * for the remainder of the operation. | |
603 | * | |
ab33dc09 FLVC |
604 | * The locking in this function is against shrink_inactive_list(): we recheck |
605 | * the page count inside the lock to see whether shrink_inactive_list() | |
606 | * grabbed the page via the LRU. If it did, give up: shrink_inactive_list() | |
607 | * will free it. | |
1da177e4 LT |
608 | */ |
609 | void release_pages(struct page **pages, int nr, int cold) | |
610 | { | |
611 | int i; | |
cc59850e | 612 | LIST_HEAD(pages_to_free); |
1da177e4 | 613 | struct zone *zone = NULL; |
fa9add64 | 614 | struct lruvec *lruvec; |
902aaed0 | 615 | unsigned long uninitialized_var(flags); |
1da177e4 | 616 | |
1da177e4 LT |
617 | for (i = 0; i < nr; i++) { |
618 | struct page *page = pages[i]; | |
1da177e4 | 619 | |
8519fb30 NP |
620 | if (unlikely(PageCompound(page))) { |
621 | if (zone) { | |
902aaed0 | 622 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
8519fb30 NP |
623 | zone = NULL; |
624 | } | |
625 | put_compound_page(page); | |
626 | continue; | |
627 | } | |
628 | ||
b5810039 | 629 | if (!put_page_testzero(page)) |
1da177e4 LT |
630 | continue; |
631 | ||
46453a6e NP |
632 | if (PageLRU(page)) { |
633 | struct zone *pagezone = page_zone(page); | |
894bc310 | 634 | |
46453a6e NP |
635 | if (pagezone != zone) { |
636 | if (zone) | |
902aaed0 HH |
637 | spin_unlock_irqrestore(&zone->lru_lock, |
638 | flags); | |
46453a6e | 639 | zone = pagezone; |
902aaed0 | 640 | spin_lock_irqsave(&zone->lru_lock, flags); |
46453a6e | 641 | } |
fa9add64 HD |
642 | |
643 | lruvec = mem_cgroup_page_lruvec(page, zone); | |
725d704e | 644 | VM_BUG_ON(!PageLRU(page)); |
67453911 | 645 | __ClearPageLRU(page); |
fa9add64 | 646 | del_page_from_lru_list(page, lruvec, page_off_lru(page)); |
46453a6e NP |
647 | } |
648 | ||
cc59850e | 649 | list_add(&page->lru, &pages_to_free); |
1da177e4 LT |
650 | } |
651 | if (zone) | |
902aaed0 | 652 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
1da177e4 | 653 | |
cc59850e | 654 | free_hot_cold_page_list(&pages_to_free, cold); |
1da177e4 | 655 | } |
0be8557b | 656 | EXPORT_SYMBOL(release_pages); |
1da177e4 LT |
657 | |
658 | /* | |
659 | * The pages which we're about to release may be in the deferred lru-addition | |
660 | * queues. That would prevent them from really being freed right now. That's | |
661 | * OK from a correctness point of view but is inefficient - those pages may be | |
662 | * cache-warm and we want to give them back to the page allocator ASAP. | |
663 | * | |
664 | * So __pagevec_release() will drain those queues here. __pagevec_lru_add() | |
665 | * and __pagevec_lru_add_active() call release_pages() directly to avoid | |
666 | * mutual recursion. | |
667 | */ | |
668 | void __pagevec_release(struct pagevec *pvec) | |
669 | { | |
670 | lru_add_drain(); | |
671 | release_pages(pvec->pages, pagevec_count(pvec), pvec->cold); | |
672 | pagevec_reinit(pvec); | |
673 | } | |
7f285701 SF |
674 | EXPORT_SYMBOL(__pagevec_release); |
675 | ||
12d27107 | 676 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
71e3aac0 | 677 | /* used by __split_huge_page_refcount() */ |
fa9add64 HD |
678 | void lru_add_page_tail(struct page *page, struct page *page_tail, |
679 | struct lruvec *lruvec) | |
71e3aac0 | 680 | { |
7512102c | 681 | int uninitialized_var(active); |
71e3aac0 AA |
682 | enum lru_list lru; |
683 | const int file = 0; | |
71e3aac0 AA |
684 | |
685 | VM_BUG_ON(!PageHead(page)); | |
686 | VM_BUG_ON(PageCompound(page_tail)); | |
687 | VM_BUG_ON(PageLRU(page_tail)); | |
fa9add64 HD |
688 | VM_BUG_ON(NR_CPUS != 1 && |
689 | !spin_is_locked(&lruvec_zone(lruvec)->lru_lock)); | |
71e3aac0 AA |
690 | |
691 | SetPageLRU(page_tail); | |
692 | ||
693 | if (page_evictable(page_tail, NULL)) { | |
694 | if (PageActive(page)) { | |
695 | SetPageActive(page_tail); | |
696 | active = 1; | |
697 | lru = LRU_ACTIVE_ANON; | |
698 | } else { | |
699 | active = 0; | |
700 | lru = LRU_INACTIVE_ANON; | |
701 | } | |
71e3aac0 AA |
702 | } else { |
703 | SetPageUnevictable(page_tail); | |
12d27107 HD |
704 | lru = LRU_UNEVICTABLE; |
705 | } | |
706 | ||
707 | if (likely(PageLRU(page))) | |
708 | list_add_tail(&page_tail->lru, &page->lru); | |
709 | else { | |
710 | struct list_head *list_head; | |
711 | /* | |
712 | * Head page has not yet been counted, as an hpage, | |
713 | * so we must account for each subpage individually. | |
714 | * | |
715 | * Use the standard add function to put page_tail on the list, | |
716 | * but then correct its position so they all end up in order. | |
717 | */ | |
fa9add64 | 718 | add_page_to_lru_list(page_tail, lruvec, lru); |
12d27107 HD |
719 | list_head = page_tail->lru.prev; |
720 | list_move_tail(&page_tail->lru, list_head); | |
71e3aac0 | 721 | } |
7512102c HD |
722 | |
723 | if (!PageUnevictable(page)) | |
fa9add64 | 724 | update_page_reclaim_stat(lruvec, file, active); |
71e3aac0 | 725 | } |
12d27107 | 726 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
71e3aac0 | 727 | |
fa9add64 HD |
728 | static void __pagevec_lru_add_fn(struct page *page, struct lruvec *lruvec, |
729 | void *arg) | |
3dd7ae8e SL |
730 | { |
731 | enum lru_list lru = (enum lru_list)arg; | |
3dd7ae8e SL |
732 | int file = is_file_lru(lru); |
733 | int active = is_active_lru(lru); | |
734 | ||
735 | VM_BUG_ON(PageActive(page)); | |
736 | VM_BUG_ON(PageUnevictable(page)); | |
737 | VM_BUG_ON(PageLRU(page)); | |
738 | ||
739 | SetPageLRU(page); | |
740 | if (active) | |
741 | SetPageActive(page); | |
fa9add64 HD |
742 | add_page_to_lru_list(page, lruvec, lru); |
743 | update_page_reclaim_stat(lruvec, file, active); | |
3dd7ae8e SL |
744 | } |
745 | ||
1da177e4 LT |
746 | /* |
747 | * Add the passed pages to the LRU, then drop the caller's refcount | |
748 | * on them. Reinitialises the caller's pagevec. | |
749 | */ | |
5095ae83 | 750 | void __pagevec_lru_add(struct pagevec *pvec, enum lru_list lru) |
1da177e4 | 751 | { |
894bc310 | 752 | VM_BUG_ON(is_unevictable_lru(lru)); |
1da177e4 | 753 | |
5095ae83 | 754 | pagevec_lru_move_fn(pvec, __pagevec_lru_add_fn, (void *)lru); |
1da177e4 | 755 | } |
5095ae83 | 756 | EXPORT_SYMBOL(__pagevec_lru_add); |
1da177e4 | 757 | |
1da177e4 LT |
758 | /** |
759 | * pagevec_lookup - gang pagecache lookup | |
760 | * @pvec: Where the resulting pages are placed | |
761 | * @mapping: The address_space to search | |
762 | * @start: The starting page index | |
763 | * @nr_pages: The maximum number of pages | |
764 | * | |
765 | * pagevec_lookup() will search for and return a group of up to @nr_pages pages | |
766 | * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a | |
767 | * reference against the pages in @pvec. | |
768 | * | |
769 | * The search returns a group of mapping-contiguous pages with ascending | |
770 | * indexes. There may be holes in the indices due to not-present pages. | |
771 | * | |
772 | * pagevec_lookup() returns the number of pages which were found. | |
773 | */ | |
774 | unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping, | |
775 | pgoff_t start, unsigned nr_pages) | |
776 | { | |
777 | pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages); | |
778 | return pagevec_count(pvec); | |
779 | } | |
78539fdf CH |
780 | EXPORT_SYMBOL(pagevec_lookup); |
781 | ||
1da177e4 LT |
782 | unsigned pagevec_lookup_tag(struct pagevec *pvec, struct address_space *mapping, |
783 | pgoff_t *index, int tag, unsigned nr_pages) | |
784 | { | |
785 | pvec->nr = find_get_pages_tag(mapping, index, tag, | |
786 | nr_pages, pvec->pages); | |
787 | return pagevec_count(pvec); | |
788 | } | |
7f285701 | 789 | EXPORT_SYMBOL(pagevec_lookup_tag); |
1da177e4 | 790 | |
1da177e4 LT |
791 | /* |
792 | * Perform any setup for the swap system | |
793 | */ | |
794 | void __init swap_setup(void) | |
795 | { | |
4481374c | 796 | unsigned long megs = totalram_pages >> (20 - PAGE_SHIFT); |
1da177e4 | 797 | |
e0bf68dd PZ |
798 | #ifdef CONFIG_SWAP |
799 | bdi_init(swapper_space.backing_dev_info); | |
800 | #endif | |
801 | ||
1da177e4 LT |
802 | /* Use a smaller cluster for small-memory machines */ |
803 | if (megs < 16) | |
804 | page_cluster = 2; | |
805 | else | |
806 | page_cluster = 3; | |
807 | /* | |
808 | * Right now other parts of the system means that we | |
809 | * _really_ don't want to cluster much more | |
810 | */ | |
1da177e4 | 811 | } |