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> |
a27bb332 | 33 | #include <linux/uio.h> |
09642082 | 34 | #include <linux/hugetlb.h> |
1da177e4 | 35 | |
64d6519d LS |
36 | #include "internal.h" |
37 | ||
1da177e4 LT |
38 | /* How many pages do we try to swap or page in/out together? */ |
39 | int page_cluster; | |
40 | ||
f04e9ebb | 41 | static DEFINE_PER_CPU(struct pagevec[NR_LRU_LISTS], lru_add_pvecs); |
f84f9504 | 42 | static DEFINE_PER_CPU(struct pagevec, lru_rotate_pvecs); |
31560180 | 43 | static DEFINE_PER_CPU(struct pagevec, lru_deactivate_pvecs); |
902aaed0 | 44 | |
b221385b AB |
45 | /* |
46 | * This path almost never happens for VM activity - pages are normally | |
47 | * freed via pagevecs. But it gets used by networking. | |
48 | */ | |
920c7a5d | 49 | static void __page_cache_release(struct page *page) |
b221385b AB |
50 | { |
51 | if (PageLRU(page)) { | |
b221385b | 52 | struct zone *zone = page_zone(page); |
fa9add64 HD |
53 | struct lruvec *lruvec; |
54 | unsigned long flags; | |
b221385b AB |
55 | |
56 | spin_lock_irqsave(&zone->lru_lock, flags); | |
fa9add64 | 57 | lruvec = mem_cgroup_page_lruvec(page, zone); |
b221385b AB |
58 | VM_BUG_ON(!PageLRU(page)); |
59 | __ClearPageLRU(page); | |
fa9add64 | 60 | del_page_from_lru_list(page, lruvec, page_off_lru(page)); |
b221385b AB |
61 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
62 | } | |
91807063 AA |
63 | } |
64 | ||
65 | static void __put_single_page(struct page *page) | |
66 | { | |
67 | __page_cache_release(page); | |
fc91668e | 68 | free_hot_cold_page(page, 0); |
b221385b AB |
69 | } |
70 | ||
91807063 | 71 | static void __put_compound_page(struct page *page) |
1da177e4 | 72 | { |
91807063 | 73 | compound_page_dtor *dtor; |
1da177e4 | 74 | |
91807063 AA |
75 | __page_cache_release(page); |
76 | dtor = get_compound_page_dtor(page); | |
77 | (*dtor)(page); | |
78 | } | |
79 | ||
80 | static void put_compound_page(struct page *page) | |
81 | { | |
82 | if (unlikely(PageTail(page))) { | |
83 | /* __split_huge_page_refcount can run under us */ | |
def52acc | 84 | struct page *page_head = compound_head(page); |
70b50f94 AA |
85 | |
86 | if (likely(page != page_head && | |
87 | get_page_unless_zero(page_head))) { | |
91807063 | 88 | unsigned long flags; |
5bf5f03c PS |
89 | |
90 | /* | |
91 | * THP can not break up slab pages so avoid taking | |
92 | * compound_lock(). Slab performs non-atomic bit ops | |
93 | * on page->flags for better performance. In particular | |
94 | * slab_unlock() in slub used to be a hot path. It is | |
95 | * still hot on arches that do not support | |
96 | * this_cpu_cmpxchg_double(). | |
97 | */ | |
17b6ada0 AA |
98 | if (PageSlab(page_head) || PageHeadHuge(page_head)) { |
99 | if (likely(PageTail(page))) { | |
100 | /* | |
101 | * __split_huge_page_refcount | |
102 | * cannot race here. | |
103 | */ | |
104 | VM_BUG_ON(!PageHead(page_head)); | |
105 | atomic_dec(&page->_mapcount); | |
5bf5f03c PS |
106 | if (put_page_testzero(page_head)) |
107 | VM_BUG_ON(1); | |
17b6ada0 AA |
108 | if (put_page_testzero(page_head)) |
109 | __put_compound_page(page_head); | |
110 | return; | |
5bf5f03c | 111 | } else |
17b6ada0 AA |
112 | /* |
113 | * __split_huge_page_refcount | |
114 | * run before us, "page" was a | |
115 | * THP tail. The split | |
116 | * page_head has been freed | |
117 | * and reallocated as slab or | |
118 | * hugetlbfs page of smaller | |
119 | * order (only possible if | |
120 | * reallocated as slab on | |
121 | * x86). | |
122 | */ | |
5bf5f03c PS |
123 | goto skip_lock; |
124 | } | |
91807063 | 125 | /* |
70b50f94 AA |
126 | * page_head wasn't a dangling pointer but it |
127 | * may not be a head page anymore by the time | |
128 | * we obtain the lock. That is ok as long as it | |
129 | * can't be freed from under us. | |
91807063 | 130 | */ |
91807063 AA |
131 | flags = compound_lock_irqsave(page_head); |
132 | if (unlikely(!PageTail(page))) { | |
133 | /* __split_huge_page_refcount run before us */ | |
134 | compound_unlock_irqrestore(page_head, flags); | |
5bf5f03c | 135 | skip_lock: |
17b6ada0 AA |
136 | if (put_page_testzero(page_head)) { |
137 | /* | |
138 | * The head page may have been | |
139 | * freed and reallocated as a | |
140 | * compound page of smaller | |
141 | * order and then freed again. | |
142 | * All we know is that it | |
143 | * cannot have become: a THP | |
144 | * page, a compound page of | |
145 | * higher order, a tail page. | |
146 | * That is because we still | |
147 | * hold the refcount of the | |
148 | * split THP tail and | |
149 | * page_head was the THP head | |
150 | * before the split. | |
151 | */ | |
152 | if (PageHead(page_head)) | |
153 | __put_compound_page(page_head); | |
154 | else | |
155 | __put_single_page(page_head); | |
156 | } | |
5bf5f03c | 157 | out_put_single: |
91807063 AA |
158 | if (put_page_testzero(page)) |
159 | __put_single_page(page); | |
160 | return; | |
161 | } | |
162 | VM_BUG_ON(page_head != page->first_page); | |
163 | /* | |
164 | * We can release the refcount taken by | |
70b50f94 AA |
165 | * get_page_unless_zero() now that |
166 | * __split_huge_page_refcount() is blocked on | |
167 | * the compound_lock. | |
91807063 AA |
168 | */ |
169 | if (put_page_testzero(page_head)) | |
170 | VM_BUG_ON(1); | |
171 | /* __split_huge_page_refcount will wait now */ | |
70b50f94 AA |
172 | VM_BUG_ON(page_mapcount(page) <= 0); |
173 | atomic_dec(&page->_mapcount); | |
91807063 | 174 | VM_BUG_ON(atomic_read(&page_head->_count) <= 0); |
70b50f94 | 175 | VM_BUG_ON(atomic_read(&page->_count) != 0); |
91807063 | 176 | compound_unlock_irqrestore(page_head, flags); |
5bf5f03c | 177 | |
a95a82e9 AA |
178 | if (put_page_testzero(page_head)) { |
179 | if (PageHead(page_head)) | |
180 | __put_compound_page(page_head); | |
181 | else | |
182 | __put_single_page(page_head); | |
183 | } | |
91807063 AA |
184 | } else { |
185 | /* page_head is a dangling pointer */ | |
186 | VM_BUG_ON(PageTail(page)); | |
187 | goto out_put_single; | |
188 | } | |
189 | } else if (put_page_testzero(page)) { | |
190 | if (PageHead(page)) | |
191 | __put_compound_page(page); | |
192 | else | |
193 | __put_single_page(page); | |
1da177e4 | 194 | } |
8519fb30 NP |
195 | } |
196 | ||
197 | void put_page(struct page *page) | |
198 | { | |
199 | if (unlikely(PageCompound(page))) | |
200 | put_compound_page(page); | |
201 | else if (put_page_testzero(page)) | |
91807063 | 202 | __put_single_page(page); |
1da177e4 LT |
203 | } |
204 | EXPORT_SYMBOL(put_page); | |
1da177e4 | 205 | |
70b50f94 AA |
206 | /* |
207 | * This function is exported but must not be called by anything other | |
208 | * than get_page(). It implements the slow path of get_page(). | |
209 | */ | |
210 | bool __get_page_tail(struct page *page) | |
211 | { | |
212 | /* | |
213 | * This takes care of get_page() if run on a tail page | |
214 | * returned by one of the get_user_pages/follow_page variants. | |
215 | * get_user_pages/follow_page itself doesn't need the compound | |
216 | * lock because it runs __get_page_tail_foll() under the | |
217 | * proper PT lock that already serializes against | |
218 | * split_huge_page(). | |
219 | */ | |
17b6ada0 | 220 | unsigned long flags; |
70b50f94 | 221 | bool got = false; |
def52acc | 222 | struct page *page_head = compound_head(page); |
70b50f94 | 223 | |
17b6ada0 AA |
224 | if (likely(page != page_head && get_page_unless_zero(page_head))) { |
225 | /* Ref to put_compound_page() comment. */ | |
226 | if (PageSlab(page_head) || PageHeadHuge(page_head)) { | |
5bf5f03c | 227 | if (likely(PageTail(page))) { |
17b6ada0 AA |
228 | /* |
229 | * This is a hugetlbfs page or a slab | |
230 | * page. __split_huge_page_refcount | |
231 | * cannot race here. | |
232 | */ | |
233 | VM_BUG_ON(!PageHead(page_head)); | |
5bf5f03c | 234 | __get_page_tail_foll(page, false); |
17b6ada0 AA |
235 | return true; |
236 | } else { | |
237 | /* | |
238 | * __split_huge_page_refcount run | |
239 | * before us, "page" was a THP | |
240 | * tail. The split page_head has been | |
241 | * freed and reallocated as slab or | |
242 | * hugetlbfs page of smaller order | |
243 | * (only possible if reallocated as | |
244 | * slab on x86). | |
245 | */ | |
09642082 | 246 | put_page(page_head); |
17b6ada0 AA |
247 | return false; |
248 | } | |
249 | } | |
250 | ||
251 | /* | |
252 | * page_head wasn't a dangling pointer but it | |
253 | * may not be a head page anymore by the time | |
254 | * we obtain the lock. That is ok as long as it | |
255 | * can't be freed from under us. | |
256 | */ | |
257 | flags = compound_lock_irqsave(page_head); | |
258 | /* here __split_huge_page_refcount won't run anymore */ | |
259 | if (likely(PageTail(page))) { | |
260 | __get_page_tail_foll(page, false); | |
261 | got = true; | |
5bf5f03c | 262 | } |
17b6ada0 AA |
263 | compound_unlock_irqrestore(page_head, flags); |
264 | if (unlikely(!got)) | |
265 | put_page(page_head); | |
70b50f94 AA |
266 | } |
267 | return got; | |
268 | } | |
269 | EXPORT_SYMBOL(__get_page_tail); | |
270 | ||
1d7ea732 | 271 | /** |
7682486b RD |
272 | * put_pages_list() - release a list of pages |
273 | * @pages: list of pages threaded on page->lru | |
1d7ea732 AZ |
274 | * |
275 | * Release a list of pages which are strung together on page.lru. Currently | |
276 | * used by read_cache_pages() and related error recovery code. | |
1d7ea732 AZ |
277 | */ |
278 | void put_pages_list(struct list_head *pages) | |
279 | { | |
280 | while (!list_empty(pages)) { | |
281 | struct page *victim; | |
282 | ||
283 | victim = list_entry(pages->prev, struct page, lru); | |
284 | list_del(&victim->lru); | |
285 | page_cache_release(victim); | |
286 | } | |
287 | } | |
288 | EXPORT_SYMBOL(put_pages_list); | |
289 | ||
18022c5d MG |
290 | /* |
291 | * get_kernel_pages() - pin kernel pages in memory | |
292 | * @kiov: An array of struct kvec structures | |
293 | * @nr_segs: number of segments to pin | |
294 | * @write: pinning for read/write, currently ignored | |
295 | * @pages: array that receives pointers to the pages pinned. | |
296 | * Should be at least nr_segs long. | |
297 | * | |
298 | * Returns number of pages pinned. This may be fewer than the number | |
299 | * requested. If nr_pages is 0 or negative, returns 0. If no pages | |
300 | * were pinned, returns -errno. Each page returned must be released | |
301 | * with a put_page() call when it is finished with. | |
302 | */ | |
303 | int get_kernel_pages(const struct kvec *kiov, int nr_segs, int write, | |
304 | struct page **pages) | |
305 | { | |
306 | int seg; | |
307 | ||
308 | for (seg = 0; seg < nr_segs; seg++) { | |
309 | if (WARN_ON(kiov[seg].iov_len != PAGE_SIZE)) | |
310 | return seg; | |
311 | ||
5a178119 | 312 | pages[seg] = kmap_to_page(kiov[seg].iov_base); |
18022c5d MG |
313 | page_cache_get(pages[seg]); |
314 | } | |
315 | ||
316 | return seg; | |
317 | } | |
318 | EXPORT_SYMBOL_GPL(get_kernel_pages); | |
319 | ||
320 | /* | |
321 | * get_kernel_page() - pin a kernel page in memory | |
322 | * @start: starting kernel address | |
323 | * @write: pinning for read/write, currently ignored | |
324 | * @pages: array that receives pointer to the page pinned. | |
325 | * Must be at least nr_segs long. | |
326 | * | |
327 | * Returns 1 if page is pinned. If the page was not pinned, returns | |
328 | * -errno. The page returned must be released with a put_page() call | |
329 | * when it is finished with. | |
330 | */ | |
331 | int get_kernel_page(unsigned long start, int write, struct page **pages) | |
332 | { | |
333 | const struct kvec kiov = { | |
334 | .iov_base = (void *)start, | |
335 | .iov_len = PAGE_SIZE | |
336 | }; | |
337 | ||
338 | return get_kernel_pages(&kiov, 1, write, pages); | |
339 | } | |
340 | EXPORT_SYMBOL_GPL(get_kernel_page); | |
341 | ||
3dd7ae8e | 342 | static void pagevec_lru_move_fn(struct pagevec *pvec, |
fa9add64 HD |
343 | void (*move_fn)(struct page *page, struct lruvec *lruvec, void *arg), |
344 | void *arg) | |
902aaed0 HH |
345 | { |
346 | int i; | |
902aaed0 | 347 | struct zone *zone = NULL; |
fa9add64 | 348 | struct lruvec *lruvec; |
3dd7ae8e | 349 | unsigned long flags = 0; |
902aaed0 HH |
350 | |
351 | for (i = 0; i < pagevec_count(pvec); i++) { | |
352 | struct page *page = pvec->pages[i]; | |
353 | struct zone *pagezone = page_zone(page); | |
354 | ||
355 | if (pagezone != zone) { | |
356 | if (zone) | |
3dd7ae8e | 357 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
902aaed0 | 358 | zone = pagezone; |
3dd7ae8e | 359 | spin_lock_irqsave(&zone->lru_lock, flags); |
902aaed0 | 360 | } |
3dd7ae8e | 361 | |
fa9add64 HD |
362 | lruvec = mem_cgroup_page_lruvec(page, zone); |
363 | (*move_fn)(page, lruvec, arg); | |
902aaed0 HH |
364 | } |
365 | if (zone) | |
3dd7ae8e | 366 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
83896fb5 LT |
367 | release_pages(pvec->pages, pvec->nr, pvec->cold); |
368 | pagevec_reinit(pvec); | |
d8505dee SL |
369 | } |
370 | ||
fa9add64 HD |
371 | static void pagevec_move_tail_fn(struct page *page, struct lruvec *lruvec, |
372 | void *arg) | |
3dd7ae8e SL |
373 | { |
374 | int *pgmoved = arg; | |
3dd7ae8e SL |
375 | |
376 | if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) { | |
377 | enum lru_list lru = page_lru_base_type(page); | |
925b7673 | 378 | list_move_tail(&page->lru, &lruvec->lists[lru]); |
3dd7ae8e SL |
379 | (*pgmoved)++; |
380 | } | |
381 | } | |
382 | ||
383 | /* | |
384 | * pagevec_move_tail() must be called with IRQ disabled. | |
385 | * Otherwise this may cause nasty races. | |
386 | */ | |
387 | static void pagevec_move_tail(struct pagevec *pvec) | |
388 | { | |
389 | int pgmoved = 0; | |
390 | ||
391 | pagevec_lru_move_fn(pvec, pagevec_move_tail_fn, &pgmoved); | |
392 | __count_vm_events(PGROTATED, pgmoved); | |
393 | } | |
394 | ||
1da177e4 LT |
395 | /* |
396 | * Writeback is about to end against a page which has been marked for immediate | |
397 | * reclaim. If it still appears to be reclaimable, move it to the tail of the | |
902aaed0 | 398 | * inactive list. |
1da177e4 | 399 | */ |
3dd7ae8e | 400 | void rotate_reclaimable_page(struct page *page) |
1da177e4 | 401 | { |
ac6aadb2 | 402 | if (!PageLocked(page) && !PageDirty(page) && !PageActive(page) && |
894bc310 | 403 | !PageUnevictable(page) && PageLRU(page)) { |
ac6aadb2 MS |
404 | struct pagevec *pvec; |
405 | unsigned long flags; | |
406 | ||
407 | page_cache_get(page); | |
408 | local_irq_save(flags); | |
409 | pvec = &__get_cpu_var(lru_rotate_pvecs); | |
410 | if (!pagevec_add(pvec, page)) | |
411 | pagevec_move_tail(pvec); | |
412 | local_irq_restore(flags); | |
413 | } | |
1da177e4 LT |
414 | } |
415 | ||
fa9add64 | 416 | static void update_page_reclaim_stat(struct lruvec *lruvec, |
3e2f41f1 KM |
417 | int file, int rotated) |
418 | { | |
fa9add64 | 419 | struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat; |
3e2f41f1 KM |
420 | |
421 | reclaim_stat->recent_scanned[file]++; | |
422 | if (rotated) | |
423 | reclaim_stat->recent_rotated[file]++; | |
3e2f41f1 KM |
424 | } |
425 | ||
fa9add64 HD |
426 | static void __activate_page(struct page *page, struct lruvec *lruvec, |
427 | void *arg) | |
1da177e4 | 428 | { |
744ed144 | 429 | if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) { |
7a608572 LT |
430 | int file = page_is_file_cache(page); |
431 | int lru = page_lru_base_type(page); | |
744ed144 | 432 | |
fa9add64 | 433 | del_page_from_lru_list(page, lruvec, lru); |
7a608572 LT |
434 | SetPageActive(page); |
435 | lru += LRU_ACTIVE; | |
fa9add64 | 436 | add_page_to_lru_list(page, lruvec, lru); |
4f98a2fe | 437 | |
fa9add64 HD |
438 | __count_vm_event(PGACTIVATE); |
439 | update_page_reclaim_stat(lruvec, file, 1); | |
1da177e4 | 440 | } |
eb709b0d SL |
441 | } |
442 | ||
443 | #ifdef CONFIG_SMP | |
444 | static DEFINE_PER_CPU(struct pagevec, activate_page_pvecs); | |
445 | ||
446 | static void activate_page_drain(int cpu) | |
447 | { | |
448 | struct pagevec *pvec = &per_cpu(activate_page_pvecs, cpu); | |
449 | ||
450 | if (pagevec_count(pvec)) | |
451 | pagevec_lru_move_fn(pvec, __activate_page, NULL); | |
452 | } | |
453 | ||
454 | void activate_page(struct page *page) | |
455 | { | |
456 | if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) { | |
457 | struct pagevec *pvec = &get_cpu_var(activate_page_pvecs); | |
458 | ||
459 | page_cache_get(page); | |
460 | if (!pagevec_add(pvec, page)) | |
461 | pagevec_lru_move_fn(pvec, __activate_page, NULL); | |
462 | put_cpu_var(activate_page_pvecs); | |
463 | } | |
464 | } | |
465 | ||
466 | #else | |
467 | static inline void activate_page_drain(int cpu) | |
468 | { | |
469 | } | |
470 | ||
471 | void activate_page(struct page *page) | |
472 | { | |
473 | struct zone *zone = page_zone(page); | |
474 | ||
475 | spin_lock_irq(&zone->lru_lock); | |
fa9add64 | 476 | __activate_page(page, mem_cgroup_page_lruvec(page, zone), NULL); |
1da177e4 LT |
477 | spin_unlock_irq(&zone->lru_lock); |
478 | } | |
eb709b0d | 479 | #endif |
1da177e4 LT |
480 | |
481 | /* | |
482 | * Mark a page as having seen activity. | |
483 | * | |
484 | * inactive,unreferenced -> inactive,referenced | |
485 | * inactive,referenced -> active,unreferenced | |
486 | * active,unreferenced -> active,referenced | |
487 | */ | |
920c7a5d | 488 | void mark_page_accessed(struct page *page) |
1da177e4 | 489 | { |
894bc310 LS |
490 | if (!PageActive(page) && !PageUnevictable(page) && |
491 | PageReferenced(page) && PageLRU(page)) { | |
1da177e4 LT |
492 | activate_page(page); |
493 | ClearPageReferenced(page); | |
494 | } else if (!PageReferenced(page)) { | |
495 | SetPageReferenced(page); | |
496 | } | |
497 | } | |
1da177e4 LT |
498 | EXPORT_SYMBOL(mark_page_accessed); |
499 | ||
d741c9cd RD |
500 | /* |
501 | * Order of operations is important: flush the pagevec when it's already | |
502 | * full, not when adding the last page, to make sure that last page is | |
503 | * not added to the LRU directly when passed to this function. Because | |
504 | * mark_page_accessed() (called after this when writing) only activates | |
505 | * pages that are on the LRU, linear writes in subpage chunks would see | |
506 | * every PAGEVEC_SIZE page activated, which is unexpected. | |
507 | */ | |
f04e9ebb | 508 | void __lru_cache_add(struct page *page, enum lru_list lru) |
1da177e4 | 509 | { |
f04e9ebb | 510 | struct pagevec *pvec = &get_cpu_var(lru_add_pvecs)[lru]; |
1da177e4 LT |
511 | |
512 | page_cache_get(page); | |
d741c9cd | 513 | if (!pagevec_space(pvec)) |
5095ae83 | 514 | __pagevec_lru_add(pvec, lru); |
d741c9cd | 515 | pagevec_add(pvec, page); |
1da177e4 LT |
516 | put_cpu_var(lru_add_pvecs); |
517 | } | |
47846b06 | 518 | EXPORT_SYMBOL(__lru_cache_add); |
1da177e4 | 519 | |
f04e9ebb KM |
520 | /** |
521 | * lru_cache_add_lru - add a page to a page list | |
522 | * @page: the page to be added to the LRU. | |
523 | * @lru: the LRU list to which the page is added. | |
524 | */ | |
525 | void lru_cache_add_lru(struct page *page, enum lru_list lru) | |
1da177e4 | 526 | { |
f04e9ebb | 527 | if (PageActive(page)) { |
894bc310 | 528 | VM_BUG_ON(PageUnevictable(page)); |
f04e9ebb | 529 | ClearPageActive(page); |
894bc310 LS |
530 | } else if (PageUnevictable(page)) { |
531 | VM_BUG_ON(PageActive(page)); | |
532 | ClearPageUnevictable(page); | |
f04e9ebb | 533 | } |
1da177e4 | 534 | |
894bc310 | 535 | VM_BUG_ON(PageLRU(page) || PageActive(page) || PageUnevictable(page)); |
f04e9ebb | 536 | __lru_cache_add(page, lru); |
1da177e4 LT |
537 | } |
538 | ||
894bc310 LS |
539 | /** |
540 | * add_page_to_unevictable_list - add a page to the unevictable list | |
541 | * @page: the page to be added to the unevictable list | |
542 | * | |
543 | * Add page directly to its zone's unevictable list. To avoid races with | |
544 | * tasks that might be making the page evictable, through eg. munlock, | |
545 | * munmap or exit, while it's not on the lru, we want to add the page | |
546 | * while it's locked or otherwise "invisible" to other tasks. This is | |
547 | * difficult to do when using the pagevec cache, so bypass that. | |
548 | */ | |
549 | void add_page_to_unevictable_list(struct page *page) | |
550 | { | |
551 | struct zone *zone = page_zone(page); | |
fa9add64 | 552 | struct lruvec *lruvec; |
894bc310 LS |
553 | |
554 | spin_lock_irq(&zone->lru_lock); | |
fa9add64 | 555 | lruvec = mem_cgroup_page_lruvec(page, zone); |
894bc310 LS |
556 | SetPageUnevictable(page); |
557 | SetPageLRU(page); | |
fa9add64 | 558 | add_page_to_lru_list(page, lruvec, LRU_UNEVICTABLE); |
894bc310 LS |
559 | spin_unlock_irq(&zone->lru_lock); |
560 | } | |
561 | ||
31560180 MK |
562 | /* |
563 | * If the page can not be invalidated, it is moved to the | |
564 | * inactive list to speed up its reclaim. It is moved to the | |
565 | * head of the list, rather than the tail, to give the flusher | |
566 | * threads some time to write it out, as this is much more | |
567 | * effective than the single-page writeout from reclaim. | |
278df9f4 MK |
568 | * |
569 | * If the page isn't page_mapped and dirty/writeback, the page | |
570 | * could reclaim asap using PG_reclaim. | |
571 | * | |
572 | * 1. active, mapped page -> none | |
573 | * 2. active, dirty/writeback page -> inactive, head, PG_reclaim | |
574 | * 3. inactive, mapped page -> none | |
575 | * 4. inactive, dirty/writeback page -> inactive, head, PG_reclaim | |
576 | * 5. inactive, clean -> inactive, tail | |
577 | * 6. Others -> none | |
578 | * | |
579 | * In 4, why it moves inactive's head, the VM expects the page would | |
580 | * be write it out by flusher threads as this is much more effective | |
581 | * than the single-page writeout from reclaim. | |
31560180 | 582 | */ |
fa9add64 HD |
583 | static void lru_deactivate_fn(struct page *page, struct lruvec *lruvec, |
584 | void *arg) | |
31560180 MK |
585 | { |
586 | int lru, file; | |
278df9f4 | 587 | bool active; |
31560180 | 588 | |
278df9f4 | 589 | if (!PageLRU(page)) |
31560180 MK |
590 | return; |
591 | ||
bad49d9c MK |
592 | if (PageUnevictable(page)) |
593 | return; | |
594 | ||
31560180 MK |
595 | /* Some processes are using the page */ |
596 | if (page_mapped(page)) | |
597 | return; | |
598 | ||
278df9f4 | 599 | active = PageActive(page); |
31560180 MK |
600 | file = page_is_file_cache(page); |
601 | lru = page_lru_base_type(page); | |
fa9add64 HD |
602 | |
603 | del_page_from_lru_list(page, lruvec, lru + active); | |
31560180 MK |
604 | ClearPageActive(page); |
605 | ClearPageReferenced(page); | |
fa9add64 | 606 | add_page_to_lru_list(page, lruvec, lru); |
31560180 | 607 | |
278df9f4 MK |
608 | if (PageWriteback(page) || PageDirty(page)) { |
609 | /* | |
610 | * PG_reclaim could be raced with end_page_writeback | |
611 | * It can make readahead confusing. But race window | |
612 | * is _really_ small and it's non-critical problem. | |
613 | */ | |
614 | SetPageReclaim(page); | |
615 | } else { | |
616 | /* | |
617 | * The page's writeback ends up during pagevec | |
618 | * We moves tha page into tail of inactive. | |
619 | */ | |
925b7673 | 620 | list_move_tail(&page->lru, &lruvec->lists[lru]); |
278df9f4 MK |
621 | __count_vm_event(PGROTATED); |
622 | } | |
623 | ||
624 | if (active) | |
625 | __count_vm_event(PGDEACTIVATE); | |
fa9add64 | 626 | update_page_reclaim_stat(lruvec, file, 0); |
31560180 MK |
627 | } |
628 | ||
902aaed0 HH |
629 | /* |
630 | * Drain pages out of the cpu's pagevecs. | |
631 | * Either "cpu" is the current CPU, and preemption has already been | |
632 | * disabled; or "cpu" is being hot-unplugged, and is already dead. | |
633 | */ | |
f0cb3c76 | 634 | void lru_add_drain_cpu(int cpu) |
1da177e4 | 635 | { |
f04e9ebb | 636 | struct pagevec *pvecs = per_cpu(lru_add_pvecs, cpu); |
902aaed0 | 637 | struct pagevec *pvec; |
f04e9ebb | 638 | int lru; |
1da177e4 | 639 | |
f04e9ebb KM |
640 | for_each_lru(lru) { |
641 | pvec = &pvecs[lru - LRU_BASE]; | |
642 | if (pagevec_count(pvec)) | |
5095ae83 | 643 | __pagevec_lru_add(pvec, lru); |
f04e9ebb | 644 | } |
902aaed0 HH |
645 | |
646 | pvec = &per_cpu(lru_rotate_pvecs, cpu); | |
647 | if (pagevec_count(pvec)) { | |
648 | unsigned long flags; | |
649 | ||
650 | /* No harm done if a racing interrupt already did this */ | |
651 | local_irq_save(flags); | |
652 | pagevec_move_tail(pvec); | |
653 | local_irq_restore(flags); | |
654 | } | |
31560180 MK |
655 | |
656 | pvec = &per_cpu(lru_deactivate_pvecs, cpu); | |
657 | if (pagevec_count(pvec)) | |
3dd7ae8e | 658 | pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL); |
eb709b0d SL |
659 | |
660 | activate_page_drain(cpu); | |
31560180 MK |
661 | } |
662 | ||
663 | /** | |
664 | * deactivate_page - forcefully deactivate a page | |
665 | * @page: page to deactivate | |
666 | * | |
667 | * This function hints the VM that @page is a good reclaim candidate, | |
668 | * for example if its invalidation fails due to the page being dirty | |
669 | * or under writeback. | |
670 | */ | |
671 | void deactivate_page(struct page *page) | |
672 | { | |
821ed6bb MK |
673 | /* |
674 | * In a workload with many unevictable page such as mprotect, unevictable | |
675 | * page deactivation for accelerating reclaim is pointless. | |
676 | */ | |
677 | if (PageUnevictable(page)) | |
678 | return; | |
679 | ||
31560180 MK |
680 | if (likely(get_page_unless_zero(page))) { |
681 | struct pagevec *pvec = &get_cpu_var(lru_deactivate_pvecs); | |
682 | ||
683 | if (!pagevec_add(pvec, page)) | |
3dd7ae8e | 684 | pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL); |
31560180 MK |
685 | put_cpu_var(lru_deactivate_pvecs); |
686 | } | |
80bfed90 AM |
687 | } |
688 | ||
689 | void lru_add_drain(void) | |
690 | { | |
f0cb3c76 | 691 | lru_add_drain_cpu(get_cpu()); |
80bfed90 | 692 | put_cpu(); |
1da177e4 LT |
693 | } |
694 | ||
c4028958 | 695 | static void lru_add_drain_per_cpu(struct work_struct *dummy) |
053837fc NP |
696 | { |
697 | lru_add_drain(); | |
698 | } | |
699 | ||
700 | /* | |
701 | * Returns 0 for success | |
702 | */ | |
703 | int lru_add_drain_all(void) | |
704 | { | |
c4028958 | 705 | return schedule_on_each_cpu(lru_add_drain_per_cpu); |
053837fc NP |
706 | } |
707 | ||
1da177e4 LT |
708 | /* |
709 | * Batched page_cache_release(). Decrement the reference count on all the | |
710 | * passed pages. If it fell to zero then remove the page from the LRU and | |
711 | * free it. | |
712 | * | |
713 | * Avoid taking zone->lru_lock if possible, but if it is taken, retain it | |
714 | * for the remainder of the operation. | |
715 | * | |
ab33dc09 FLVC |
716 | * The locking in this function is against shrink_inactive_list(): we recheck |
717 | * the page count inside the lock to see whether shrink_inactive_list() | |
718 | * grabbed the page via the LRU. If it did, give up: shrink_inactive_list() | |
719 | * will free it. | |
1da177e4 LT |
720 | */ |
721 | void release_pages(struct page **pages, int nr, int cold) | |
722 | { | |
723 | int i; | |
cc59850e | 724 | LIST_HEAD(pages_to_free); |
1da177e4 | 725 | struct zone *zone = NULL; |
fa9add64 | 726 | struct lruvec *lruvec; |
902aaed0 | 727 | unsigned long uninitialized_var(flags); |
1da177e4 | 728 | |
1da177e4 LT |
729 | for (i = 0; i < nr; i++) { |
730 | struct page *page = pages[i]; | |
1da177e4 | 731 | |
8519fb30 NP |
732 | if (unlikely(PageCompound(page))) { |
733 | if (zone) { | |
902aaed0 | 734 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
8519fb30 NP |
735 | zone = NULL; |
736 | } | |
737 | put_compound_page(page); | |
738 | continue; | |
739 | } | |
740 | ||
b5810039 | 741 | if (!put_page_testzero(page)) |
1da177e4 LT |
742 | continue; |
743 | ||
46453a6e NP |
744 | if (PageLRU(page)) { |
745 | struct zone *pagezone = page_zone(page); | |
894bc310 | 746 | |
46453a6e NP |
747 | if (pagezone != zone) { |
748 | if (zone) | |
902aaed0 HH |
749 | spin_unlock_irqrestore(&zone->lru_lock, |
750 | flags); | |
46453a6e | 751 | zone = pagezone; |
902aaed0 | 752 | spin_lock_irqsave(&zone->lru_lock, flags); |
46453a6e | 753 | } |
fa9add64 HD |
754 | |
755 | lruvec = mem_cgroup_page_lruvec(page, zone); | |
725d704e | 756 | VM_BUG_ON(!PageLRU(page)); |
67453911 | 757 | __ClearPageLRU(page); |
fa9add64 | 758 | del_page_from_lru_list(page, lruvec, page_off_lru(page)); |
46453a6e NP |
759 | } |
760 | ||
cc59850e | 761 | list_add(&page->lru, &pages_to_free); |
1da177e4 LT |
762 | } |
763 | if (zone) | |
902aaed0 | 764 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
1da177e4 | 765 | |
cc59850e | 766 | free_hot_cold_page_list(&pages_to_free, cold); |
1da177e4 | 767 | } |
0be8557b | 768 | EXPORT_SYMBOL(release_pages); |
1da177e4 LT |
769 | |
770 | /* | |
771 | * The pages which we're about to release may be in the deferred lru-addition | |
772 | * queues. That would prevent them from really being freed right now. That's | |
773 | * OK from a correctness point of view but is inefficient - those pages may be | |
774 | * cache-warm and we want to give them back to the page allocator ASAP. | |
775 | * | |
776 | * So __pagevec_release() will drain those queues here. __pagevec_lru_add() | |
777 | * and __pagevec_lru_add_active() call release_pages() directly to avoid | |
778 | * mutual recursion. | |
779 | */ | |
780 | void __pagevec_release(struct pagevec *pvec) | |
781 | { | |
782 | lru_add_drain(); | |
783 | release_pages(pvec->pages, pagevec_count(pvec), pvec->cold); | |
784 | pagevec_reinit(pvec); | |
785 | } | |
7f285701 SF |
786 | EXPORT_SYMBOL(__pagevec_release); |
787 | ||
12d27107 | 788 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
71e3aac0 | 789 | /* used by __split_huge_page_refcount() */ |
fa9add64 | 790 | void lru_add_page_tail(struct page *page, struct page *page_tail, |
5bc7b8ac | 791 | struct lruvec *lruvec, struct list_head *list) |
71e3aac0 | 792 | { |
7512102c | 793 | int uninitialized_var(active); |
71e3aac0 AA |
794 | enum lru_list lru; |
795 | const int file = 0; | |
71e3aac0 AA |
796 | |
797 | VM_BUG_ON(!PageHead(page)); | |
798 | VM_BUG_ON(PageCompound(page_tail)); | |
799 | VM_BUG_ON(PageLRU(page_tail)); | |
fa9add64 HD |
800 | VM_BUG_ON(NR_CPUS != 1 && |
801 | !spin_is_locked(&lruvec_zone(lruvec)->lru_lock)); | |
71e3aac0 | 802 | |
5bc7b8ac SL |
803 | if (!list) |
804 | SetPageLRU(page_tail); | |
71e3aac0 | 805 | |
39b5f29a | 806 | if (page_evictable(page_tail)) { |
71e3aac0 AA |
807 | if (PageActive(page)) { |
808 | SetPageActive(page_tail); | |
809 | active = 1; | |
810 | lru = LRU_ACTIVE_ANON; | |
811 | } else { | |
812 | active = 0; | |
813 | lru = LRU_INACTIVE_ANON; | |
814 | } | |
71e3aac0 AA |
815 | } else { |
816 | SetPageUnevictable(page_tail); | |
12d27107 HD |
817 | lru = LRU_UNEVICTABLE; |
818 | } | |
819 | ||
820 | if (likely(PageLRU(page))) | |
821 | list_add_tail(&page_tail->lru, &page->lru); | |
5bc7b8ac SL |
822 | else if (list) { |
823 | /* page reclaim is reclaiming a huge page */ | |
824 | get_page(page_tail); | |
825 | list_add_tail(&page_tail->lru, list); | |
826 | } else { | |
12d27107 HD |
827 | struct list_head *list_head; |
828 | /* | |
829 | * Head page has not yet been counted, as an hpage, | |
830 | * so we must account for each subpage individually. | |
831 | * | |
832 | * Use the standard add function to put page_tail on the list, | |
833 | * but then correct its position so they all end up in order. | |
834 | */ | |
fa9add64 | 835 | add_page_to_lru_list(page_tail, lruvec, lru); |
12d27107 HD |
836 | list_head = page_tail->lru.prev; |
837 | list_move_tail(&page_tail->lru, list_head); | |
71e3aac0 | 838 | } |
7512102c HD |
839 | |
840 | if (!PageUnevictable(page)) | |
fa9add64 | 841 | update_page_reclaim_stat(lruvec, file, active); |
71e3aac0 | 842 | } |
12d27107 | 843 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
71e3aac0 | 844 | |
fa9add64 HD |
845 | static void __pagevec_lru_add_fn(struct page *page, struct lruvec *lruvec, |
846 | void *arg) | |
3dd7ae8e SL |
847 | { |
848 | enum lru_list lru = (enum lru_list)arg; | |
3dd7ae8e SL |
849 | int file = is_file_lru(lru); |
850 | int active = is_active_lru(lru); | |
851 | ||
852 | VM_BUG_ON(PageActive(page)); | |
853 | VM_BUG_ON(PageUnevictable(page)); | |
854 | VM_BUG_ON(PageLRU(page)); | |
855 | ||
856 | SetPageLRU(page); | |
857 | if (active) | |
858 | SetPageActive(page); | |
fa9add64 HD |
859 | add_page_to_lru_list(page, lruvec, lru); |
860 | update_page_reclaim_stat(lruvec, file, active); | |
3dd7ae8e SL |
861 | } |
862 | ||
1da177e4 LT |
863 | /* |
864 | * Add the passed pages to the LRU, then drop the caller's refcount | |
865 | * on them. Reinitialises the caller's pagevec. | |
866 | */ | |
5095ae83 | 867 | void __pagevec_lru_add(struct pagevec *pvec, enum lru_list lru) |
1da177e4 | 868 | { |
894bc310 | 869 | VM_BUG_ON(is_unevictable_lru(lru)); |
1da177e4 | 870 | |
5095ae83 | 871 | pagevec_lru_move_fn(pvec, __pagevec_lru_add_fn, (void *)lru); |
1da177e4 | 872 | } |
5095ae83 | 873 | EXPORT_SYMBOL(__pagevec_lru_add); |
1da177e4 | 874 | |
1da177e4 LT |
875 | /** |
876 | * pagevec_lookup - gang pagecache lookup | |
877 | * @pvec: Where the resulting pages are placed | |
878 | * @mapping: The address_space to search | |
879 | * @start: The starting page index | |
880 | * @nr_pages: The maximum number of pages | |
881 | * | |
882 | * pagevec_lookup() will search for and return a group of up to @nr_pages pages | |
883 | * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a | |
884 | * reference against the pages in @pvec. | |
885 | * | |
886 | * The search returns a group of mapping-contiguous pages with ascending | |
887 | * indexes. There may be holes in the indices due to not-present pages. | |
888 | * | |
889 | * pagevec_lookup() returns the number of pages which were found. | |
890 | */ | |
891 | unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping, | |
892 | pgoff_t start, unsigned nr_pages) | |
893 | { | |
894 | pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages); | |
895 | return pagevec_count(pvec); | |
896 | } | |
78539fdf CH |
897 | EXPORT_SYMBOL(pagevec_lookup); |
898 | ||
1da177e4 LT |
899 | unsigned pagevec_lookup_tag(struct pagevec *pvec, struct address_space *mapping, |
900 | pgoff_t *index, int tag, unsigned nr_pages) | |
901 | { | |
902 | pvec->nr = find_get_pages_tag(mapping, index, tag, | |
903 | nr_pages, pvec->pages); | |
904 | return pagevec_count(pvec); | |
905 | } | |
7f285701 | 906 | EXPORT_SYMBOL(pagevec_lookup_tag); |
1da177e4 | 907 | |
1da177e4 LT |
908 | /* |
909 | * Perform any setup for the swap system | |
910 | */ | |
911 | void __init swap_setup(void) | |
912 | { | |
4481374c | 913 | unsigned long megs = totalram_pages >> (20 - PAGE_SHIFT); |
e0bf68dd | 914 | #ifdef CONFIG_SWAP |
33806f06 SL |
915 | int i; |
916 | ||
917 | bdi_init(swapper_spaces[0].backing_dev_info); | |
918 | for (i = 0; i < MAX_SWAPFILES; i++) { | |
919 | spin_lock_init(&swapper_spaces[i].tree_lock); | |
920 | INIT_LIST_HEAD(&swapper_spaces[i].i_mmap_nonlinear); | |
921 | } | |
e0bf68dd PZ |
922 | #endif |
923 | ||
1da177e4 | 924 | /* Use a smaller cluster for small-memory machines */ |
6fa3eb70 S |
925 | #ifdef CONFIG_ZRAM |
926 | page_cluster = 0; // disable swap read-ahead | |
927 | #else | |
1da177e4 LT |
928 | if (megs < 16) |
929 | page_cluster = 2; | |
930 | else | |
931 | page_cluster = 3; | |
6fa3eb70 | 932 | #endif |
1da177e4 LT |
933 | /* |
934 | * Right now other parts of the system means that we | |
935 | * _really_ don't want to cluster much more | |
936 | */ | |
1da177e4 | 937 | } |