]>
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 | ||
18022c5d MG |
239 | /* |
240 | * get_kernel_pages() - pin kernel pages in memory | |
241 | * @kiov: An array of struct kvec structures | |
242 | * @nr_segs: number of segments to pin | |
243 | * @write: pinning for read/write, currently ignored | |
244 | * @pages: array that receives pointers to the pages pinned. | |
245 | * Should be at least nr_segs long. | |
246 | * | |
247 | * Returns number of pages pinned. This may be fewer than the number | |
248 | * requested. If nr_pages is 0 or negative, returns 0. If no pages | |
249 | * were pinned, returns -errno. Each page returned must be released | |
250 | * with a put_page() call when it is finished with. | |
251 | */ | |
252 | int get_kernel_pages(const struct kvec *kiov, int nr_segs, int write, | |
253 | struct page **pages) | |
254 | { | |
255 | int seg; | |
256 | ||
257 | for (seg = 0; seg < nr_segs; seg++) { | |
258 | if (WARN_ON(kiov[seg].iov_len != PAGE_SIZE)) | |
259 | return seg; | |
260 | ||
261 | /* virt_to_page sanity checks the PFN */ | |
262 | pages[seg] = virt_to_page(kiov[seg].iov_base); | |
263 | page_cache_get(pages[seg]); | |
264 | } | |
265 | ||
266 | return seg; | |
267 | } | |
268 | EXPORT_SYMBOL_GPL(get_kernel_pages); | |
269 | ||
270 | /* | |
271 | * get_kernel_page() - pin a kernel page in memory | |
272 | * @start: starting kernel address | |
273 | * @write: pinning for read/write, currently ignored | |
274 | * @pages: array that receives pointer to the page pinned. | |
275 | * Must be at least nr_segs long. | |
276 | * | |
277 | * Returns 1 if page is pinned. If the page was not pinned, returns | |
278 | * -errno. The page returned must be released with a put_page() call | |
279 | * when it is finished with. | |
280 | */ | |
281 | int get_kernel_page(unsigned long start, int write, struct page **pages) | |
282 | { | |
283 | const struct kvec kiov = { | |
284 | .iov_base = (void *)start, | |
285 | .iov_len = PAGE_SIZE | |
286 | }; | |
287 | ||
288 | return get_kernel_pages(&kiov, 1, write, pages); | |
289 | } | |
290 | EXPORT_SYMBOL_GPL(get_kernel_page); | |
291 | ||
3dd7ae8e | 292 | static void pagevec_lru_move_fn(struct pagevec *pvec, |
fa9add64 HD |
293 | void (*move_fn)(struct page *page, struct lruvec *lruvec, void *arg), |
294 | void *arg) | |
902aaed0 HH |
295 | { |
296 | int i; | |
902aaed0 | 297 | struct zone *zone = NULL; |
fa9add64 | 298 | struct lruvec *lruvec; |
3dd7ae8e | 299 | unsigned long flags = 0; |
902aaed0 HH |
300 | |
301 | for (i = 0; i < pagevec_count(pvec); i++) { | |
302 | struct page *page = pvec->pages[i]; | |
303 | struct zone *pagezone = page_zone(page); | |
304 | ||
305 | if (pagezone != zone) { | |
306 | if (zone) | |
3dd7ae8e | 307 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
902aaed0 | 308 | zone = pagezone; |
3dd7ae8e | 309 | spin_lock_irqsave(&zone->lru_lock, flags); |
902aaed0 | 310 | } |
3dd7ae8e | 311 | |
fa9add64 HD |
312 | lruvec = mem_cgroup_page_lruvec(page, zone); |
313 | (*move_fn)(page, lruvec, arg); | |
902aaed0 HH |
314 | } |
315 | if (zone) | |
3dd7ae8e | 316 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
83896fb5 LT |
317 | release_pages(pvec->pages, pvec->nr, pvec->cold); |
318 | pagevec_reinit(pvec); | |
d8505dee SL |
319 | } |
320 | ||
fa9add64 HD |
321 | static void pagevec_move_tail_fn(struct page *page, struct lruvec *lruvec, |
322 | void *arg) | |
3dd7ae8e SL |
323 | { |
324 | int *pgmoved = arg; | |
3dd7ae8e SL |
325 | |
326 | if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) { | |
327 | enum lru_list lru = page_lru_base_type(page); | |
925b7673 | 328 | list_move_tail(&page->lru, &lruvec->lists[lru]); |
3dd7ae8e SL |
329 | (*pgmoved)++; |
330 | } | |
331 | } | |
332 | ||
333 | /* | |
334 | * pagevec_move_tail() must be called with IRQ disabled. | |
335 | * Otherwise this may cause nasty races. | |
336 | */ | |
337 | static void pagevec_move_tail(struct pagevec *pvec) | |
338 | { | |
339 | int pgmoved = 0; | |
340 | ||
341 | pagevec_lru_move_fn(pvec, pagevec_move_tail_fn, &pgmoved); | |
342 | __count_vm_events(PGROTATED, pgmoved); | |
343 | } | |
344 | ||
1da177e4 LT |
345 | /* |
346 | * Writeback is about to end against a page which has been marked for immediate | |
347 | * reclaim. If it still appears to be reclaimable, move it to the tail of the | |
902aaed0 | 348 | * inactive list. |
1da177e4 | 349 | */ |
3dd7ae8e | 350 | void rotate_reclaimable_page(struct page *page) |
1da177e4 | 351 | { |
ac6aadb2 | 352 | if (!PageLocked(page) && !PageDirty(page) && !PageActive(page) && |
894bc310 | 353 | !PageUnevictable(page) && PageLRU(page)) { |
ac6aadb2 MS |
354 | struct pagevec *pvec; |
355 | unsigned long flags; | |
356 | ||
357 | page_cache_get(page); | |
358 | local_irq_save(flags); | |
359 | pvec = &__get_cpu_var(lru_rotate_pvecs); | |
360 | if (!pagevec_add(pvec, page)) | |
361 | pagevec_move_tail(pvec); | |
362 | local_irq_restore(flags); | |
363 | } | |
1da177e4 LT |
364 | } |
365 | ||
fa9add64 | 366 | static void update_page_reclaim_stat(struct lruvec *lruvec, |
3e2f41f1 KM |
367 | int file, int rotated) |
368 | { | |
fa9add64 | 369 | struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat; |
3e2f41f1 KM |
370 | |
371 | reclaim_stat->recent_scanned[file]++; | |
372 | if (rotated) | |
373 | reclaim_stat->recent_rotated[file]++; | |
3e2f41f1 KM |
374 | } |
375 | ||
fa9add64 HD |
376 | static void __activate_page(struct page *page, struct lruvec *lruvec, |
377 | void *arg) | |
1da177e4 | 378 | { |
744ed144 | 379 | if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) { |
7a608572 LT |
380 | int file = page_is_file_cache(page); |
381 | int lru = page_lru_base_type(page); | |
744ed144 | 382 | |
fa9add64 | 383 | del_page_from_lru_list(page, lruvec, lru); |
7a608572 LT |
384 | SetPageActive(page); |
385 | lru += LRU_ACTIVE; | |
fa9add64 | 386 | add_page_to_lru_list(page, lruvec, lru); |
4f98a2fe | 387 | |
fa9add64 HD |
388 | __count_vm_event(PGACTIVATE); |
389 | update_page_reclaim_stat(lruvec, file, 1); | |
1da177e4 | 390 | } |
eb709b0d SL |
391 | } |
392 | ||
393 | #ifdef CONFIG_SMP | |
394 | static DEFINE_PER_CPU(struct pagevec, activate_page_pvecs); | |
395 | ||
396 | static void activate_page_drain(int cpu) | |
397 | { | |
398 | struct pagevec *pvec = &per_cpu(activate_page_pvecs, cpu); | |
399 | ||
400 | if (pagevec_count(pvec)) | |
401 | pagevec_lru_move_fn(pvec, __activate_page, NULL); | |
402 | } | |
403 | ||
404 | void activate_page(struct page *page) | |
405 | { | |
406 | if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) { | |
407 | struct pagevec *pvec = &get_cpu_var(activate_page_pvecs); | |
408 | ||
409 | page_cache_get(page); | |
410 | if (!pagevec_add(pvec, page)) | |
411 | pagevec_lru_move_fn(pvec, __activate_page, NULL); | |
412 | put_cpu_var(activate_page_pvecs); | |
413 | } | |
414 | } | |
415 | ||
416 | #else | |
417 | static inline void activate_page_drain(int cpu) | |
418 | { | |
419 | } | |
420 | ||
421 | void activate_page(struct page *page) | |
422 | { | |
423 | struct zone *zone = page_zone(page); | |
424 | ||
425 | spin_lock_irq(&zone->lru_lock); | |
fa9add64 | 426 | __activate_page(page, mem_cgroup_page_lruvec(page, zone), NULL); |
1da177e4 LT |
427 | spin_unlock_irq(&zone->lru_lock); |
428 | } | |
eb709b0d | 429 | #endif |
1da177e4 LT |
430 | |
431 | /* | |
432 | * Mark a page as having seen activity. | |
433 | * | |
434 | * inactive,unreferenced -> inactive,referenced | |
435 | * inactive,referenced -> active,unreferenced | |
436 | * active,unreferenced -> active,referenced | |
437 | */ | |
920c7a5d | 438 | void mark_page_accessed(struct page *page) |
1da177e4 | 439 | { |
894bc310 LS |
440 | if (!PageActive(page) && !PageUnevictable(page) && |
441 | PageReferenced(page) && PageLRU(page)) { | |
1da177e4 LT |
442 | activate_page(page); |
443 | ClearPageReferenced(page); | |
444 | } else if (!PageReferenced(page)) { | |
445 | SetPageReferenced(page); | |
446 | } | |
447 | } | |
1da177e4 LT |
448 | EXPORT_SYMBOL(mark_page_accessed); |
449 | ||
f04e9ebb | 450 | void __lru_cache_add(struct page *page, enum lru_list lru) |
1da177e4 | 451 | { |
f04e9ebb | 452 | struct pagevec *pvec = &get_cpu_var(lru_add_pvecs)[lru]; |
1da177e4 LT |
453 | |
454 | page_cache_get(page); | |
455 | if (!pagevec_add(pvec, page)) | |
5095ae83 | 456 | __pagevec_lru_add(pvec, lru); |
1da177e4 LT |
457 | put_cpu_var(lru_add_pvecs); |
458 | } | |
47846b06 | 459 | EXPORT_SYMBOL(__lru_cache_add); |
1da177e4 | 460 | |
f04e9ebb KM |
461 | /** |
462 | * lru_cache_add_lru - add a page to a page list | |
463 | * @page: the page to be added to the LRU. | |
464 | * @lru: the LRU list to which the page is added. | |
465 | */ | |
466 | void lru_cache_add_lru(struct page *page, enum lru_list lru) | |
1da177e4 | 467 | { |
f04e9ebb | 468 | if (PageActive(page)) { |
894bc310 | 469 | VM_BUG_ON(PageUnevictable(page)); |
f04e9ebb | 470 | ClearPageActive(page); |
894bc310 LS |
471 | } else if (PageUnevictable(page)) { |
472 | VM_BUG_ON(PageActive(page)); | |
473 | ClearPageUnevictable(page); | |
f04e9ebb | 474 | } |
1da177e4 | 475 | |
894bc310 | 476 | VM_BUG_ON(PageLRU(page) || PageActive(page) || PageUnevictable(page)); |
f04e9ebb | 477 | __lru_cache_add(page, lru); |
1da177e4 LT |
478 | } |
479 | ||
894bc310 LS |
480 | /** |
481 | * add_page_to_unevictable_list - add a page to the unevictable list | |
482 | * @page: the page to be added to the unevictable list | |
483 | * | |
484 | * Add page directly to its zone's unevictable list. To avoid races with | |
485 | * tasks that might be making the page evictable, through eg. munlock, | |
486 | * munmap or exit, while it's not on the lru, we want to add the page | |
487 | * while it's locked or otherwise "invisible" to other tasks. This is | |
488 | * difficult to do when using the pagevec cache, so bypass that. | |
489 | */ | |
490 | void add_page_to_unevictable_list(struct page *page) | |
491 | { | |
492 | struct zone *zone = page_zone(page); | |
fa9add64 | 493 | struct lruvec *lruvec; |
894bc310 LS |
494 | |
495 | spin_lock_irq(&zone->lru_lock); | |
fa9add64 | 496 | lruvec = mem_cgroup_page_lruvec(page, zone); |
894bc310 LS |
497 | SetPageUnevictable(page); |
498 | SetPageLRU(page); | |
fa9add64 | 499 | add_page_to_lru_list(page, lruvec, LRU_UNEVICTABLE); |
894bc310 LS |
500 | spin_unlock_irq(&zone->lru_lock); |
501 | } | |
502 | ||
31560180 MK |
503 | /* |
504 | * If the page can not be invalidated, it is moved to the | |
505 | * inactive list to speed up its reclaim. It is moved to the | |
506 | * head of the list, rather than the tail, to give the flusher | |
507 | * threads some time to write it out, as this is much more | |
508 | * effective than the single-page writeout from reclaim. | |
278df9f4 MK |
509 | * |
510 | * If the page isn't page_mapped and dirty/writeback, the page | |
511 | * could reclaim asap using PG_reclaim. | |
512 | * | |
513 | * 1. active, mapped page -> none | |
514 | * 2. active, dirty/writeback page -> inactive, head, PG_reclaim | |
515 | * 3. inactive, mapped page -> none | |
516 | * 4. inactive, dirty/writeback page -> inactive, head, PG_reclaim | |
517 | * 5. inactive, clean -> inactive, tail | |
518 | * 6. Others -> none | |
519 | * | |
520 | * In 4, why it moves inactive's head, the VM expects the page would | |
521 | * be write it out by flusher threads as this is much more effective | |
522 | * than the single-page writeout from reclaim. | |
31560180 | 523 | */ |
fa9add64 HD |
524 | static void lru_deactivate_fn(struct page *page, struct lruvec *lruvec, |
525 | void *arg) | |
31560180 MK |
526 | { |
527 | int lru, file; | |
278df9f4 | 528 | bool active; |
31560180 | 529 | |
278df9f4 | 530 | if (!PageLRU(page)) |
31560180 MK |
531 | return; |
532 | ||
bad49d9c MK |
533 | if (PageUnevictable(page)) |
534 | return; | |
535 | ||
31560180 MK |
536 | /* Some processes are using the page */ |
537 | if (page_mapped(page)) | |
538 | return; | |
539 | ||
278df9f4 | 540 | active = PageActive(page); |
31560180 MK |
541 | file = page_is_file_cache(page); |
542 | lru = page_lru_base_type(page); | |
fa9add64 HD |
543 | |
544 | del_page_from_lru_list(page, lruvec, lru + active); | |
31560180 MK |
545 | ClearPageActive(page); |
546 | ClearPageReferenced(page); | |
fa9add64 | 547 | add_page_to_lru_list(page, lruvec, lru); |
31560180 | 548 | |
278df9f4 MK |
549 | if (PageWriteback(page) || PageDirty(page)) { |
550 | /* | |
551 | * PG_reclaim could be raced with end_page_writeback | |
552 | * It can make readahead confusing. But race window | |
553 | * is _really_ small and it's non-critical problem. | |
554 | */ | |
555 | SetPageReclaim(page); | |
556 | } else { | |
557 | /* | |
558 | * The page's writeback ends up during pagevec | |
559 | * We moves tha page into tail of inactive. | |
560 | */ | |
925b7673 | 561 | list_move_tail(&page->lru, &lruvec->lists[lru]); |
278df9f4 MK |
562 | __count_vm_event(PGROTATED); |
563 | } | |
564 | ||
565 | if (active) | |
566 | __count_vm_event(PGDEACTIVATE); | |
fa9add64 | 567 | update_page_reclaim_stat(lruvec, file, 0); |
31560180 MK |
568 | } |
569 | ||
902aaed0 HH |
570 | /* |
571 | * Drain pages out of the cpu's pagevecs. | |
572 | * Either "cpu" is the current CPU, and preemption has already been | |
573 | * disabled; or "cpu" is being hot-unplugged, and is already dead. | |
574 | */ | |
f0cb3c76 | 575 | void lru_add_drain_cpu(int cpu) |
1da177e4 | 576 | { |
f04e9ebb | 577 | struct pagevec *pvecs = per_cpu(lru_add_pvecs, cpu); |
902aaed0 | 578 | struct pagevec *pvec; |
f04e9ebb | 579 | int lru; |
1da177e4 | 580 | |
f04e9ebb KM |
581 | for_each_lru(lru) { |
582 | pvec = &pvecs[lru - LRU_BASE]; | |
583 | if (pagevec_count(pvec)) | |
5095ae83 | 584 | __pagevec_lru_add(pvec, lru); |
f04e9ebb | 585 | } |
902aaed0 HH |
586 | |
587 | pvec = &per_cpu(lru_rotate_pvecs, cpu); | |
588 | if (pagevec_count(pvec)) { | |
589 | unsigned long flags; | |
590 | ||
591 | /* No harm done if a racing interrupt already did this */ | |
592 | local_irq_save(flags); | |
593 | pagevec_move_tail(pvec); | |
594 | local_irq_restore(flags); | |
595 | } | |
31560180 MK |
596 | |
597 | pvec = &per_cpu(lru_deactivate_pvecs, cpu); | |
598 | if (pagevec_count(pvec)) | |
3dd7ae8e | 599 | pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL); |
eb709b0d SL |
600 | |
601 | activate_page_drain(cpu); | |
31560180 MK |
602 | } |
603 | ||
604 | /** | |
605 | * deactivate_page - forcefully deactivate a page | |
606 | * @page: page to deactivate | |
607 | * | |
608 | * This function hints the VM that @page is a good reclaim candidate, | |
609 | * for example if its invalidation fails due to the page being dirty | |
610 | * or under writeback. | |
611 | */ | |
612 | void deactivate_page(struct page *page) | |
613 | { | |
821ed6bb MK |
614 | /* |
615 | * In a workload with many unevictable page such as mprotect, unevictable | |
616 | * page deactivation for accelerating reclaim is pointless. | |
617 | */ | |
618 | if (PageUnevictable(page)) | |
619 | return; | |
620 | ||
31560180 MK |
621 | if (likely(get_page_unless_zero(page))) { |
622 | struct pagevec *pvec = &get_cpu_var(lru_deactivate_pvecs); | |
623 | ||
624 | if (!pagevec_add(pvec, page)) | |
3dd7ae8e | 625 | pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL); |
31560180 MK |
626 | put_cpu_var(lru_deactivate_pvecs); |
627 | } | |
80bfed90 AM |
628 | } |
629 | ||
630 | void lru_add_drain(void) | |
631 | { | |
f0cb3c76 | 632 | lru_add_drain_cpu(get_cpu()); |
80bfed90 | 633 | put_cpu(); |
1da177e4 LT |
634 | } |
635 | ||
c4028958 | 636 | static void lru_add_drain_per_cpu(struct work_struct *dummy) |
053837fc NP |
637 | { |
638 | lru_add_drain(); | |
639 | } | |
640 | ||
641 | /* | |
642 | * Returns 0 for success | |
643 | */ | |
644 | int lru_add_drain_all(void) | |
645 | { | |
c4028958 | 646 | return schedule_on_each_cpu(lru_add_drain_per_cpu); |
053837fc NP |
647 | } |
648 | ||
1da177e4 LT |
649 | /* |
650 | * Batched page_cache_release(). Decrement the reference count on all the | |
651 | * passed pages. If it fell to zero then remove the page from the LRU and | |
652 | * free it. | |
653 | * | |
654 | * Avoid taking zone->lru_lock if possible, but if it is taken, retain it | |
655 | * for the remainder of the operation. | |
656 | * | |
ab33dc09 FLVC |
657 | * The locking in this function is against shrink_inactive_list(): we recheck |
658 | * the page count inside the lock to see whether shrink_inactive_list() | |
659 | * grabbed the page via the LRU. If it did, give up: shrink_inactive_list() | |
660 | * will free it. | |
1da177e4 LT |
661 | */ |
662 | void release_pages(struct page **pages, int nr, int cold) | |
663 | { | |
664 | int i; | |
cc59850e | 665 | LIST_HEAD(pages_to_free); |
1da177e4 | 666 | struct zone *zone = NULL; |
fa9add64 | 667 | struct lruvec *lruvec; |
902aaed0 | 668 | unsigned long uninitialized_var(flags); |
1da177e4 | 669 | |
1da177e4 LT |
670 | for (i = 0; i < nr; i++) { |
671 | struct page *page = pages[i]; | |
1da177e4 | 672 | |
8519fb30 NP |
673 | if (unlikely(PageCompound(page))) { |
674 | if (zone) { | |
902aaed0 | 675 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
8519fb30 NP |
676 | zone = NULL; |
677 | } | |
678 | put_compound_page(page); | |
679 | continue; | |
680 | } | |
681 | ||
b5810039 | 682 | if (!put_page_testzero(page)) |
1da177e4 LT |
683 | continue; |
684 | ||
46453a6e NP |
685 | if (PageLRU(page)) { |
686 | struct zone *pagezone = page_zone(page); | |
894bc310 | 687 | |
46453a6e NP |
688 | if (pagezone != zone) { |
689 | if (zone) | |
902aaed0 HH |
690 | spin_unlock_irqrestore(&zone->lru_lock, |
691 | flags); | |
46453a6e | 692 | zone = pagezone; |
902aaed0 | 693 | spin_lock_irqsave(&zone->lru_lock, flags); |
46453a6e | 694 | } |
fa9add64 HD |
695 | |
696 | lruvec = mem_cgroup_page_lruvec(page, zone); | |
725d704e | 697 | VM_BUG_ON(!PageLRU(page)); |
67453911 | 698 | __ClearPageLRU(page); |
fa9add64 | 699 | del_page_from_lru_list(page, lruvec, page_off_lru(page)); |
46453a6e NP |
700 | } |
701 | ||
cc59850e | 702 | list_add(&page->lru, &pages_to_free); |
1da177e4 LT |
703 | } |
704 | if (zone) | |
902aaed0 | 705 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
1da177e4 | 706 | |
cc59850e | 707 | free_hot_cold_page_list(&pages_to_free, cold); |
1da177e4 | 708 | } |
0be8557b | 709 | EXPORT_SYMBOL(release_pages); |
1da177e4 LT |
710 | |
711 | /* | |
712 | * The pages which we're about to release may be in the deferred lru-addition | |
713 | * queues. That would prevent them from really being freed right now. That's | |
714 | * OK from a correctness point of view but is inefficient - those pages may be | |
715 | * cache-warm and we want to give them back to the page allocator ASAP. | |
716 | * | |
717 | * So __pagevec_release() will drain those queues here. __pagevec_lru_add() | |
718 | * and __pagevec_lru_add_active() call release_pages() directly to avoid | |
719 | * mutual recursion. | |
720 | */ | |
721 | void __pagevec_release(struct pagevec *pvec) | |
722 | { | |
723 | lru_add_drain(); | |
724 | release_pages(pvec->pages, pagevec_count(pvec), pvec->cold); | |
725 | pagevec_reinit(pvec); | |
726 | } | |
7f285701 SF |
727 | EXPORT_SYMBOL(__pagevec_release); |
728 | ||
12d27107 | 729 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
71e3aac0 | 730 | /* used by __split_huge_page_refcount() */ |
fa9add64 HD |
731 | void lru_add_page_tail(struct page *page, struct page *page_tail, |
732 | struct lruvec *lruvec) | |
71e3aac0 | 733 | { |
7512102c | 734 | int uninitialized_var(active); |
71e3aac0 AA |
735 | enum lru_list lru; |
736 | const int file = 0; | |
71e3aac0 AA |
737 | |
738 | VM_BUG_ON(!PageHead(page)); | |
739 | VM_BUG_ON(PageCompound(page_tail)); | |
740 | VM_BUG_ON(PageLRU(page_tail)); | |
fa9add64 HD |
741 | VM_BUG_ON(NR_CPUS != 1 && |
742 | !spin_is_locked(&lruvec_zone(lruvec)->lru_lock)); | |
71e3aac0 AA |
743 | |
744 | SetPageLRU(page_tail); | |
745 | ||
746 | if (page_evictable(page_tail, NULL)) { | |
747 | if (PageActive(page)) { | |
748 | SetPageActive(page_tail); | |
749 | active = 1; | |
750 | lru = LRU_ACTIVE_ANON; | |
751 | } else { | |
752 | active = 0; | |
753 | lru = LRU_INACTIVE_ANON; | |
754 | } | |
71e3aac0 AA |
755 | } else { |
756 | SetPageUnevictable(page_tail); | |
12d27107 HD |
757 | lru = LRU_UNEVICTABLE; |
758 | } | |
759 | ||
760 | if (likely(PageLRU(page))) | |
761 | list_add_tail(&page_tail->lru, &page->lru); | |
762 | else { | |
763 | struct list_head *list_head; | |
764 | /* | |
765 | * Head page has not yet been counted, as an hpage, | |
766 | * so we must account for each subpage individually. | |
767 | * | |
768 | * Use the standard add function to put page_tail on the list, | |
769 | * but then correct its position so they all end up in order. | |
770 | */ | |
fa9add64 | 771 | add_page_to_lru_list(page_tail, lruvec, lru); |
12d27107 HD |
772 | list_head = page_tail->lru.prev; |
773 | list_move_tail(&page_tail->lru, list_head); | |
71e3aac0 | 774 | } |
7512102c HD |
775 | |
776 | if (!PageUnevictable(page)) | |
fa9add64 | 777 | update_page_reclaim_stat(lruvec, file, active); |
71e3aac0 | 778 | } |
12d27107 | 779 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
71e3aac0 | 780 | |
fa9add64 HD |
781 | static void __pagevec_lru_add_fn(struct page *page, struct lruvec *lruvec, |
782 | void *arg) | |
3dd7ae8e SL |
783 | { |
784 | enum lru_list lru = (enum lru_list)arg; | |
3dd7ae8e SL |
785 | int file = is_file_lru(lru); |
786 | int active = is_active_lru(lru); | |
787 | ||
788 | VM_BUG_ON(PageActive(page)); | |
789 | VM_BUG_ON(PageUnevictable(page)); | |
790 | VM_BUG_ON(PageLRU(page)); | |
791 | ||
792 | SetPageLRU(page); | |
793 | if (active) | |
794 | SetPageActive(page); | |
fa9add64 HD |
795 | add_page_to_lru_list(page, lruvec, lru); |
796 | update_page_reclaim_stat(lruvec, file, active); | |
3dd7ae8e SL |
797 | } |
798 | ||
1da177e4 LT |
799 | /* |
800 | * Add the passed pages to the LRU, then drop the caller's refcount | |
801 | * on them. Reinitialises the caller's pagevec. | |
802 | */ | |
5095ae83 | 803 | void __pagevec_lru_add(struct pagevec *pvec, enum lru_list lru) |
1da177e4 | 804 | { |
894bc310 | 805 | VM_BUG_ON(is_unevictable_lru(lru)); |
1da177e4 | 806 | |
5095ae83 | 807 | pagevec_lru_move_fn(pvec, __pagevec_lru_add_fn, (void *)lru); |
1da177e4 | 808 | } |
5095ae83 | 809 | EXPORT_SYMBOL(__pagevec_lru_add); |
1da177e4 | 810 | |
1da177e4 LT |
811 | /** |
812 | * pagevec_lookup - gang pagecache lookup | |
813 | * @pvec: Where the resulting pages are placed | |
814 | * @mapping: The address_space to search | |
815 | * @start: The starting page index | |
816 | * @nr_pages: The maximum number of pages | |
817 | * | |
818 | * pagevec_lookup() will search for and return a group of up to @nr_pages pages | |
819 | * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a | |
820 | * reference against the pages in @pvec. | |
821 | * | |
822 | * The search returns a group of mapping-contiguous pages with ascending | |
823 | * indexes. There may be holes in the indices due to not-present pages. | |
824 | * | |
825 | * pagevec_lookup() returns the number of pages which were found. | |
826 | */ | |
827 | unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping, | |
828 | pgoff_t start, unsigned nr_pages) | |
829 | { | |
830 | pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages); | |
831 | return pagevec_count(pvec); | |
832 | } | |
78539fdf CH |
833 | EXPORT_SYMBOL(pagevec_lookup); |
834 | ||
1da177e4 LT |
835 | unsigned pagevec_lookup_tag(struct pagevec *pvec, struct address_space *mapping, |
836 | pgoff_t *index, int tag, unsigned nr_pages) | |
837 | { | |
838 | pvec->nr = find_get_pages_tag(mapping, index, tag, | |
839 | nr_pages, pvec->pages); | |
840 | return pagevec_count(pvec); | |
841 | } | |
7f285701 | 842 | EXPORT_SYMBOL(pagevec_lookup_tag); |
1da177e4 | 843 | |
1da177e4 LT |
844 | /* |
845 | * Perform any setup for the swap system | |
846 | */ | |
847 | void __init swap_setup(void) | |
848 | { | |
4481374c | 849 | unsigned long megs = totalram_pages >> (20 - PAGE_SHIFT); |
1da177e4 | 850 | |
e0bf68dd PZ |
851 | #ifdef CONFIG_SWAP |
852 | bdi_init(swapper_space.backing_dev_info); | |
853 | #endif | |
854 | ||
1da177e4 LT |
855 | /* Use a smaller cluster for small-memory machines */ |
856 | if (megs < 16) | |
857 | page_cluster = 2; | |
858 | else | |
859 | page_cluster = 3; | |
860 | /* | |
861 | * Right now other parts of the system means that we | |
862 | * _really_ don't want to cluster much more | |
863 | */ | |
1da177e4 | 864 | } |