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> |
1da177e4 | 34 | |
64d6519d LS |
35 | #include "internal.h" |
36 | ||
c6286c98 MG |
37 | #define CREATE_TRACE_POINTS |
38 | #include <trace/events/pagemap.h> | |
39 | ||
1da177e4 LT |
40 | /* How many pages do we try to swap or page in/out together? */ |
41 | int page_cluster; | |
42 | ||
f04e9ebb | 43 | static DEFINE_PER_CPU(struct pagevec[NR_LRU_LISTS], lru_add_pvecs); |
f84f9504 | 44 | static DEFINE_PER_CPU(struct pagevec, lru_rotate_pvecs); |
31560180 | 45 | static DEFINE_PER_CPU(struct pagevec, lru_deactivate_pvecs); |
902aaed0 | 46 | |
b221385b AB |
47 | /* |
48 | * This path almost never happens for VM activity - pages are normally | |
49 | * freed via pagevecs. But it gets used by networking. | |
50 | */ | |
920c7a5d | 51 | static void __page_cache_release(struct page *page) |
b221385b AB |
52 | { |
53 | if (PageLRU(page)) { | |
b221385b | 54 | struct zone *zone = page_zone(page); |
fa9add64 HD |
55 | struct lruvec *lruvec; |
56 | unsigned long flags; | |
b221385b AB |
57 | |
58 | spin_lock_irqsave(&zone->lru_lock, flags); | |
fa9add64 | 59 | lruvec = mem_cgroup_page_lruvec(page, zone); |
b221385b AB |
60 | VM_BUG_ON(!PageLRU(page)); |
61 | __ClearPageLRU(page); | |
fa9add64 | 62 | del_page_from_lru_list(page, lruvec, page_off_lru(page)); |
b221385b AB |
63 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
64 | } | |
91807063 AA |
65 | } |
66 | ||
67 | static void __put_single_page(struct page *page) | |
68 | { | |
69 | __page_cache_release(page); | |
fc91668e | 70 | free_hot_cold_page(page, 0); |
b221385b AB |
71 | } |
72 | ||
91807063 | 73 | static void __put_compound_page(struct page *page) |
1da177e4 | 74 | { |
91807063 | 75 | compound_page_dtor *dtor; |
1da177e4 | 76 | |
91807063 AA |
77 | __page_cache_release(page); |
78 | dtor = get_compound_page_dtor(page); | |
79 | (*dtor)(page); | |
80 | } | |
81 | ||
82 | static void put_compound_page(struct page *page) | |
83 | { | |
84 | if (unlikely(PageTail(page))) { | |
85 | /* __split_huge_page_refcount can run under us */ | |
70b50f94 AA |
86 | struct page *page_head = compound_trans_head(page); |
87 | ||
88 | if (likely(page != page_head && | |
89 | get_page_unless_zero(page_head))) { | |
91807063 | 90 | unsigned long flags; |
5bf5f03c PS |
91 | |
92 | /* | |
93 | * THP can not break up slab pages so avoid taking | |
94 | * compound_lock(). Slab performs non-atomic bit ops | |
95 | * on page->flags for better performance. In particular | |
96 | * slab_unlock() in slub used to be a hot path. It is | |
97 | * still hot on arches that do not support | |
98 | * this_cpu_cmpxchg_double(). | |
99 | */ | |
100 | if (PageSlab(page_head)) { | |
101 | if (PageTail(page)) { | |
102 | if (put_page_testzero(page_head)) | |
103 | VM_BUG_ON(1); | |
104 | ||
105 | atomic_dec(&page->_mapcount); | |
106 | goto skip_lock_tail; | |
107 | } else | |
108 | goto skip_lock; | |
109 | } | |
91807063 | 110 | /* |
70b50f94 AA |
111 | * page_head wasn't a dangling pointer but it |
112 | * may not be a head page anymore by the time | |
113 | * we obtain the lock. That is ok as long as it | |
114 | * can't be freed from under us. | |
91807063 | 115 | */ |
91807063 AA |
116 | flags = compound_lock_irqsave(page_head); |
117 | if (unlikely(!PageTail(page))) { | |
118 | /* __split_huge_page_refcount run before us */ | |
119 | compound_unlock_irqrestore(page_head, flags); | |
5bf5f03c | 120 | skip_lock: |
91807063 AA |
121 | if (put_page_testzero(page_head)) |
122 | __put_single_page(page_head); | |
5bf5f03c | 123 | out_put_single: |
91807063 AA |
124 | if (put_page_testzero(page)) |
125 | __put_single_page(page); | |
126 | return; | |
127 | } | |
128 | VM_BUG_ON(page_head != page->first_page); | |
129 | /* | |
130 | * We can release the refcount taken by | |
70b50f94 AA |
131 | * get_page_unless_zero() now that |
132 | * __split_huge_page_refcount() is blocked on | |
133 | * the compound_lock. | |
91807063 AA |
134 | */ |
135 | if (put_page_testzero(page_head)) | |
136 | VM_BUG_ON(1); | |
137 | /* __split_huge_page_refcount will wait now */ | |
70b50f94 AA |
138 | VM_BUG_ON(page_mapcount(page) <= 0); |
139 | atomic_dec(&page->_mapcount); | |
91807063 | 140 | VM_BUG_ON(atomic_read(&page_head->_count) <= 0); |
70b50f94 | 141 | VM_BUG_ON(atomic_read(&page->_count) != 0); |
91807063 | 142 | compound_unlock_irqrestore(page_head, flags); |
5bf5f03c PS |
143 | |
144 | skip_lock_tail: | |
a95a82e9 AA |
145 | if (put_page_testzero(page_head)) { |
146 | if (PageHead(page_head)) | |
147 | __put_compound_page(page_head); | |
148 | else | |
149 | __put_single_page(page_head); | |
150 | } | |
91807063 AA |
151 | } else { |
152 | /* page_head is a dangling pointer */ | |
153 | VM_BUG_ON(PageTail(page)); | |
154 | goto out_put_single; | |
155 | } | |
156 | } else if (put_page_testzero(page)) { | |
157 | if (PageHead(page)) | |
158 | __put_compound_page(page); | |
159 | else | |
160 | __put_single_page(page); | |
1da177e4 | 161 | } |
8519fb30 NP |
162 | } |
163 | ||
164 | void put_page(struct page *page) | |
165 | { | |
166 | if (unlikely(PageCompound(page))) | |
167 | put_compound_page(page); | |
168 | else if (put_page_testzero(page)) | |
91807063 | 169 | __put_single_page(page); |
1da177e4 LT |
170 | } |
171 | EXPORT_SYMBOL(put_page); | |
1da177e4 | 172 | |
70b50f94 AA |
173 | /* |
174 | * This function is exported but must not be called by anything other | |
175 | * than get_page(). It implements the slow path of get_page(). | |
176 | */ | |
177 | bool __get_page_tail(struct page *page) | |
178 | { | |
179 | /* | |
180 | * This takes care of get_page() if run on a tail page | |
181 | * returned by one of the get_user_pages/follow_page variants. | |
182 | * get_user_pages/follow_page itself doesn't need the compound | |
183 | * lock because it runs __get_page_tail_foll() under the | |
184 | * proper PT lock that already serializes against | |
185 | * split_huge_page(). | |
186 | */ | |
187 | unsigned long flags; | |
188 | bool got = false; | |
189 | struct page *page_head = compound_trans_head(page); | |
190 | ||
191 | if (likely(page != page_head && get_page_unless_zero(page_head))) { | |
5bf5f03c PS |
192 | |
193 | /* Ref to put_compound_page() comment. */ | |
194 | if (PageSlab(page_head)) { | |
195 | if (likely(PageTail(page))) { | |
196 | __get_page_tail_foll(page, false); | |
197 | return true; | |
198 | } else { | |
199 | put_page(page_head); | |
200 | return false; | |
201 | } | |
202 | } | |
203 | ||
70b50f94 AA |
204 | /* |
205 | * page_head wasn't a dangling pointer but it | |
206 | * may not be a head page anymore by the time | |
207 | * we obtain the lock. That is ok as long as it | |
208 | * can't be freed from under us. | |
209 | */ | |
210 | flags = compound_lock_irqsave(page_head); | |
211 | /* here __split_huge_page_refcount won't run anymore */ | |
212 | if (likely(PageTail(page))) { | |
213 | __get_page_tail_foll(page, false); | |
214 | got = true; | |
215 | } | |
216 | compound_unlock_irqrestore(page_head, flags); | |
217 | if (unlikely(!got)) | |
218 | put_page(page_head); | |
219 | } | |
220 | return got; | |
221 | } | |
222 | EXPORT_SYMBOL(__get_page_tail); | |
223 | ||
1d7ea732 | 224 | /** |
7682486b RD |
225 | * put_pages_list() - release a list of pages |
226 | * @pages: list of pages threaded on page->lru | |
1d7ea732 AZ |
227 | * |
228 | * Release a list of pages which are strung together on page.lru. Currently | |
229 | * used by read_cache_pages() and related error recovery code. | |
1d7ea732 AZ |
230 | */ |
231 | void put_pages_list(struct list_head *pages) | |
232 | { | |
233 | while (!list_empty(pages)) { | |
234 | struct page *victim; | |
235 | ||
236 | victim = list_entry(pages->prev, struct page, lru); | |
237 | list_del(&victim->lru); | |
238 | page_cache_release(victim); | |
239 | } | |
240 | } | |
241 | EXPORT_SYMBOL(put_pages_list); | |
242 | ||
18022c5d MG |
243 | /* |
244 | * get_kernel_pages() - pin kernel pages in memory | |
245 | * @kiov: An array of struct kvec structures | |
246 | * @nr_segs: number of segments to pin | |
247 | * @write: pinning for read/write, currently ignored | |
248 | * @pages: array that receives pointers to the pages pinned. | |
249 | * Should be at least nr_segs long. | |
250 | * | |
251 | * Returns number of pages pinned. This may be fewer than the number | |
252 | * requested. If nr_pages is 0 or negative, returns 0. If no pages | |
253 | * were pinned, returns -errno. Each page returned must be released | |
254 | * with a put_page() call when it is finished with. | |
255 | */ | |
256 | int get_kernel_pages(const struct kvec *kiov, int nr_segs, int write, | |
257 | struct page **pages) | |
258 | { | |
259 | int seg; | |
260 | ||
261 | for (seg = 0; seg < nr_segs; seg++) { | |
262 | if (WARN_ON(kiov[seg].iov_len != PAGE_SIZE)) | |
263 | return seg; | |
264 | ||
5a178119 | 265 | pages[seg] = kmap_to_page(kiov[seg].iov_base); |
18022c5d MG |
266 | page_cache_get(pages[seg]); |
267 | } | |
268 | ||
269 | return seg; | |
270 | } | |
271 | EXPORT_SYMBOL_GPL(get_kernel_pages); | |
272 | ||
273 | /* | |
274 | * get_kernel_page() - pin a kernel page in memory | |
275 | * @start: starting kernel address | |
276 | * @write: pinning for read/write, currently ignored | |
277 | * @pages: array that receives pointer to the page pinned. | |
278 | * Must be at least nr_segs long. | |
279 | * | |
280 | * Returns 1 if page is pinned. If the page was not pinned, returns | |
281 | * -errno. The page returned must be released with a put_page() call | |
282 | * when it is finished with. | |
283 | */ | |
284 | int get_kernel_page(unsigned long start, int write, struct page **pages) | |
285 | { | |
286 | const struct kvec kiov = { | |
287 | .iov_base = (void *)start, | |
288 | .iov_len = PAGE_SIZE | |
289 | }; | |
290 | ||
291 | return get_kernel_pages(&kiov, 1, write, pages); | |
292 | } | |
293 | EXPORT_SYMBOL_GPL(get_kernel_page); | |
294 | ||
3dd7ae8e | 295 | static void pagevec_lru_move_fn(struct pagevec *pvec, |
fa9add64 HD |
296 | void (*move_fn)(struct page *page, struct lruvec *lruvec, void *arg), |
297 | void *arg) | |
902aaed0 HH |
298 | { |
299 | int i; | |
902aaed0 | 300 | struct zone *zone = NULL; |
fa9add64 | 301 | struct lruvec *lruvec; |
3dd7ae8e | 302 | unsigned long flags = 0; |
902aaed0 HH |
303 | |
304 | for (i = 0; i < pagevec_count(pvec); i++) { | |
305 | struct page *page = pvec->pages[i]; | |
306 | struct zone *pagezone = page_zone(page); | |
307 | ||
308 | if (pagezone != zone) { | |
309 | if (zone) | |
3dd7ae8e | 310 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
902aaed0 | 311 | zone = pagezone; |
3dd7ae8e | 312 | spin_lock_irqsave(&zone->lru_lock, flags); |
902aaed0 | 313 | } |
3dd7ae8e | 314 | |
fa9add64 HD |
315 | lruvec = mem_cgroup_page_lruvec(page, zone); |
316 | (*move_fn)(page, lruvec, arg); | |
902aaed0 HH |
317 | } |
318 | if (zone) | |
3dd7ae8e | 319 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
83896fb5 LT |
320 | release_pages(pvec->pages, pvec->nr, pvec->cold); |
321 | pagevec_reinit(pvec); | |
d8505dee SL |
322 | } |
323 | ||
fa9add64 HD |
324 | static void pagevec_move_tail_fn(struct page *page, struct lruvec *lruvec, |
325 | void *arg) | |
3dd7ae8e SL |
326 | { |
327 | int *pgmoved = arg; | |
3dd7ae8e SL |
328 | |
329 | if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) { | |
330 | enum lru_list lru = page_lru_base_type(page); | |
925b7673 | 331 | list_move_tail(&page->lru, &lruvec->lists[lru]); |
3dd7ae8e SL |
332 | (*pgmoved)++; |
333 | } | |
334 | } | |
335 | ||
336 | /* | |
337 | * pagevec_move_tail() must be called with IRQ disabled. | |
338 | * Otherwise this may cause nasty races. | |
339 | */ | |
340 | static void pagevec_move_tail(struct pagevec *pvec) | |
341 | { | |
342 | int pgmoved = 0; | |
343 | ||
344 | pagevec_lru_move_fn(pvec, pagevec_move_tail_fn, &pgmoved); | |
345 | __count_vm_events(PGROTATED, pgmoved); | |
346 | } | |
347 | ||
1da177e4 LT |
348 | /* |
349 | * Writeback is about to end against a page which has been marked for immediate | |
350 | * reclaim. If it still appears to be reclaimable, move it to the tail of the | |
902aaed0 | 351 | * inactive list. |
1da177e4 | 352 | */ |
3dd7ae8e | 353 | void rotate_reclaimable_page(struct page *page) |
1da177e4 | 354 | { |
ac6aadb2 | 355 | if (!PageLocked(page) && !PageDirty(page) && !PageActive(page) && |
894bc310 | 356 | !PageUnevictable(page) && PageLRU(page)) { |
ac6aadb2 MS |
357 | struct pagevec *pvec; |
358 | unsigned long flags; | |
359 | ||
360 | page_cache_get(page); | |
361 | local_irq_save(flags); | |
362 | pvec = &__get_cpu_var(lru_rotate_pvecs); | |
363 | if (!pagevec_add(pvec, page)) | |
364 | pagevec_move_tail(pvec); | |
365 | local_irq_restore(flags); | |
366 | } | |
1da177e4 LT |
367 | } |
368 | ||
fa9add64 | 369 | static void update_page_reclaim_stat(struct lruvec *lruvec, |
3e2f41f1 KM |
370 | int file, int rotated) |
371 | { | |
fa9add64 | 372 | struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat; |
3e2f41f1 KM |
373 | |
374 | reclaim_stat->recent_scanned[file]++; | |
375 | if (rotated) | |
376 | reclaim_stat->recent_rotated[file]++; | |
3e2f41f1 KM |
377 | } |
378 | ||
fa9add64 HD |
379 | static void __activate_page(struct page *page, struct lruvec *lruvec, |
380 | void *arg) | |
1da177e4 | 381 | { |
744ed144 | 382 | if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) { |
7a608572 LT |
383 | int file = page_is_file_cache(page); |
384 | int lru = page_lru_base_type(page); | |
744ed144 | 385 | |
fa9add64 | 386 | del_page_from_lru_list(page, lruvec, lru); |
7a608572 LT |
387 | SetPageActive(page); |
388 | lru += LRU_ACTIVE; | |
fa9add64 | 389 | add_page_to_lru_list(page, lruvec, lru); |
c6286c98 | 390 | trace_mm_lru_activate(page, page_to_pfn(page)); |
4f98a2fe | 391 | |
fa9add64 HD |
392 | __count_vm_event(PGACTIVATE); |
393 | update_page_reclaim_stat(lruvec, file, 1); | |
1da177e4 | 394 | } |
eb709b0d SL |
395 | } |
396 | ||
397 | #ifdef CONFIG_SMP | |
398 | static DEFINE_PER_CPU(struct pagevec, activate_page_pvecs); | |
399 | ||
400 | static void activate_page_drain(int cpu) | |
401 | { | |
402 | struct pagevec *pvec = &per_cpu(activate_page_pvecs, cpu); | |
403 | ||
404 | if (pagevec_count(pvec)) | |
405 | pagevec_lru_move_fn(pvec, __activate_page, NULL); | |
406 | } | |
407 | ||
408 | void activate_page(struct page *page) | |
409 | { | |
410 | if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) { | |
411 | struct pagevec *pvec = &get_cpu_var(activate_page_pvecs); | |
412 | ||
413 | page_cache_get(page); | |
414 | if (!pagevec_add(pvec, page)) | |
415 | pagevec_lru_move_fn(pvec, __activate_page, NULL); | |
416 | put_cpu_var(activate_page_pvecs); | |
417 | } | |
418 | } | |
419 | ||
420 | #else | |
421 | static inline void activate_page_drain(int cpu) | |
422 | { | |
423 | } | |
424 | ||
425 | void activate_page(struct page *page) | |
426 | { | |
427 | struct zone *zone = page_zone(page); | |
428 | ||
429 | spin_lock_irq(&zone->lru_lock); | |
fa9add64 | 430 | __activate_page(page, mem_cgroup_page_lruvec(page, zone), NULL); |
1da177e4 LT |
431 | spin_unlock_irq(&zone->lru_lock); |
432 | } | |
eb709b0d | 433 | #endif |
1da177e4 LT |
434 | |
435 | /* | |
436 | * Mark a page as having seen activity. | |
437 | * | |
438 | * inactive,unreferenced -> inactive,referenced | |
439 | * inactive,referenced -> active,unreferenced | |
440 | * active,unreferenced -> active,referenced | |
441 | */ | |
920c7a5d | 442 | void mark_page_accessed(struct page *page) |
1da177e4 | 443 | { |
894bc310 LS |
444 | if (!PageActive(page) && !PageUnevictable(page) && |
445 | PageReferenced(page) && PageLRU(page)) { | |
1da177e4 LT |
446 | activate_page(page); |
447 | ClearPageReferenced(page); | |
448 | } else if (!PageReferenced(page)) { | |
449 | SetPageReferenced(page); | |
450 | } | |
451 | } | |
1da177e4 LT |
452 | EXPORT_SYMBOL(mark_page_accessed); |
453 | ||
d741c9cd RD |
454 | /* |
455 | * Order of operations is important: flush the pagevec when it's already | |
456 | * full, not when adding the last page, to make sure that last page is | |
457 | * not added to the LRU directly when passed to this function. Because | |
458 | * mark_page_accessed() (called after this when writing) only activates | |
459 | * pages that are on the LRU, linear writes in subpage chunks would see | |
460 | * every PAGEVEC_SIZE page activated, which is unexpected. | |
461 | */ | |
f04e9ebb | 462 | void __lru_cache_add(struct page *page, enum lru_list lru) |
1da177e4 | 463 | { |
f04e9ebb | 464 | struct pagevec *pvec = &get_cpu_var(lru_add_pvecs)[lru]; |
1da177e4 LT |
465 | |
466 | page_cache_get(page); | |
d741c9cd | 467 | if (!pagevec_space(pvec)) |
5095ae83 | 468 | __pagevec_lru_add(pvec, lru); |
d741c9cd | 469 | pagevec_add(pvec, page); |
1da177e4 LT |
470 | put_cpu_var(lru_add_pvecs); |
471 | } | |
47846b06 | 472 | EXPORT_SYMBOL(__lru_cache_add); |
1da177e4 | 473 | |
f04e9ebb KM |
474 | /** |
475 | * lru_cache_add_lru - add a page to a page list | |
476 | * @page: the page to be added to the LRU. | |
477 | * @lru: the LRU list to which the page is added. | |
478 | */ | |
479 | void lru_cache_add_lru(struct page *page, enum lru_list lru) | |
1da177e4 | 480 | { |
f04e9ebb | 481 | if (PageActive(page)) { |
894bc310 | 482 | VM_BUG_ON(PageUnevictable(page)); |
f04e9ebb | 483 | ClearPageActive(page); |
894bc310 LS |
484 | } else if (PageUnevictable(page)) { |
485 | VM_BUG_ON(PageActive(page)); | |
486 | ClearPageUnevictable(page); | |
f04e9ebb | 487 | } |
1da177e4 | 488 | |
894bc310 | 489 | VM_BUG_ON(PageLRU(page) || PageActive(page) || PageUnevictable(page)); |
f04e9ebb | 490 | __lru_cache_add(page, lru); |
1da177e4 LT |
491 | } |
492 | ||
894bc310 LS |
493 | /** |
494 | * add_page_to_unevictable_list - add a page to the unevictable list | |
495 | * @page: the page to be added to the unevictable list | |
496 | * | |
497 | * Add page directly to its zone's unevictable list. To avoid races with | |
498 | * tasks that might be making the page evictable, through eg. munlock, | |
499 | * munmap or exit, while it's not on the lru, we want to add the page | |
500 | * while it's locked or otherwise "invisible" to other tasks. This is | |
501 | * difficult to do when using the pagevec cache, so bypass that. | |
502 | */ | |
503 | void add_page_to_unevictable_list(struct page *page) | |
504 | { | |
505 | struct zone *zone = page_zone(page); | |
fa9add64 | 506 | struct lruvec *lruvec; |
894bc310 LS |
507 | |
508 | spin_lock_irq(&zone->lru_lock); | |
fa9add64 | 509 | lruvec = mem_cgroup_page_lruvec(page, zone); |
894bc310 LS |
510 | SetPageUnevictable(page); |
511 | SetPageLRU(page); | |
fa9add64 | 512 | add_page_to_lru_list(page, lruvec, LRU_UNEVICTABLE); |
894bc310 LS |
513 | spin_unlock_irq(&zone->lru_lock); |
514 | } | |
515 | ||
31560180 MK |
516 | /* |
517 | * If the page can not be invalidated, it is moved to the | |
518 | * inactive list to speed up its reclaim. It is moved to the | |
519 | * head of the list, rather than the tail, to give the flusher | |
520 | * threads some time to write it out, as this is much more | |
521 | * effective than the single-page writeout from reclaim. | |
278df9f4 MK |
522 | * |
523 | * If the page isn't page_mapped and dirty/writeback, the page | |
524 | * could reclaim asap using PG_reclaim. | |
525 | * | |
526 | * 1. active, mapped page -> none | |
527 | * 2. active, dirty/writeback page -> inactive, head, PG_reclaim | |
528 | * 3. inactive, mapped page -> none | |
529 | * 4. inactive, dirty/writeback page -> inactive, head, PG_reclaim | |
530 | * 5. inactive, clean -> inactive, tail | |
531 | * 6. Others -> none | |
532 | * | |
533 | * In 4, why it moves inactive's head, the VM expects the page would | |
534 | * be write it out by flusher threads as this is much more effective | |
535 | * than the single-page writeout from reclaim. | |
31560180 | 536 | */ |
fa9add64 HD |
537 | static void lru_deactivate_fn(struct page *page, struct lruvec *lruvec, |
538 | void *arg) | |
31560180 MK |
539 | { |
540 | int lru, file; | |
278df9f4 | 541 | bool active; |
31560180 | 542 | |
278df9f4 | 543 | if (!PageLRU(page)) |
31560180 MK |
544 | return; |
545 | ||
bad49d9c MK |
546 | if (PageUnevictable(page)) |
547 | return; | |
548 | ||
31560180 MK |
549 | /* Some processes are using the page */ |
550 | if (page_mapped(page)) | |
551 | return; | |
552 | ||
278df9f4 | 553 | active = PageActive(page); |
31560180 MK |
554 | file = page_is_file_cache(page); |
555 | lru = page_lru_base_type(page); | |
fa9add64 HD |
556 | |
557 | del_page_from_lru_list(page, lruvec, lru + active); | |
31560180 MK |
558 | ClearPageActive(page); |
559 | ClearPageReferenced(page); | |
fa9add64 | 560 | add_page_to_lru_list(page, lruvec, lru); |
31560180 | 561 | |
278df9f4 MK |
562 | if (PageWriteback(page) || PageDirty(page)) { |
563 | /* | |
564 | * PG_reclaim could be raced with end_page_writeback | |
565 | * It can make readahead confusing. But race window | |
566 | * is _really_ small and it's non-critical problem. | |
567 | */ | |
568 | SetPageReclaim(page); | |
569 | } else { | |
570 | /* | |
571 | * The page's writeback ends up during pagevec | |
572 | * We moves tha page into tail of inactive. | |
573 | */ | |
925b7673 | 574 | list_move_tail(&page->lru, &lruvec->lists[lru]); |
278df9f4 MK |
575 | __count_vm_event(PGROTATED); |
576 | } | |
577 | ||
578 | if (active) | |
579 | __count_vm_event(PGDEACTIVATE); | |
fa9add64 | 580 | update_page_reclaim_stat(lruvec, file, 0); |
31560180 MK |
581 | } |
582 | ||
902aaed0 HH |
583 | /* |
584 | * Drain pages out of the cpu's pagevecs. | |
585 | * Either "cpu" is the current CPU, and preemption has already been | |
586 | * disabled; or "cpu" is being hot-unplugged, and is already dead. | |
587 | */ | |
f0cb3c76 | 588 | void lru_add_drain_cpu(int cpu) |
1da177e4 | 589 | { |
f04e9ebb | 590 | struct pagevec *pvecs = per_cpu(lru_add_pvecs, cpu); |
902aaed0 | 591 | struct pagevec *pvec; |
f04e9ebb | 592 | int lru; |
1da177e4 | 593 | |
f04e9ebb KM |
594 | for_each_lru(lru) { |
595 | pvec = &pvecs[lru - LRU_BASE]; | |
596 | if (pagevec_count(pvec)) | |
5095ae83 | 597 | __pagevec_lru_add(pvec, lru); |
f04e9ebb | 598 | } |
902aaed0 HH |
599 | |
600 | pvec = &per_cpu(lru_rotate_pvecs, cpu); | |
601 | if (pagevec_count(pvec)) { | |
602 | unsigned long flags; | |
603 | ||
604 | /* No harm done if a racing interrupt already did this */ | |
605 | local_irq_save(flags); | |
606 | pagevec_move_tail(pvec); | |
607 | local_irq_restore(flags); | |
608 | } | |
31560180 MK |
609 | |
610 | pvec = &per_cpu(lru_deactivate_pvecs, cpu); | |
611 | if (pagevec_count(pvec)) | |
3dd7ae8e | 612 | pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL); |
eb709b0d SL |
613 | |
614 | activate_page_drain(cpu); | |
31560180 MK |
615 | } |
616 | ||
617 | /** | |
618 | * deactivate_page - forcefully deactivate a page | |
619 | * @page: page to deactivate | |
620 | * | |
621 | * This function hints the VM that @page is a good reclaim candidate, | |
622 | * for example if its invalidation fails due to the page being dirty | |
623 | * or under writeback. | |
624 | */ | |
625 | void deactivate_page(struct page *page) | |
626 | { | |
821ed6bb MK |
627 | /* |
628 | * In a workload with many unevictable page such as mprotect, unevictable | |
629 | * page deactivation for accelerating reclaim is pointless. | |
630 | */ | |
631 | if (PageUnevictable(page)) | |
632 | return; | |
633 | ||
31560180 MK |
634 | if (likely(get_page_unless_zero(page))) { |
635 | struct pagevec *pvec = &get_cpu_var(lru_deactivate_pvecs); | |
636 | ||
637 | if (!pagevec_add(pvec, page)) | |
3dd7ae8e | 638 | pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL); |
31560180 MK |
639 | put_cpu_var(lru_deactivate_pvecs); |
640 | } | |
80bfed90 AM |
641 | } |
642 | ||
643 | void lru_add_drain(void) | |
644 | { | |
f0cb3c76 | 645 | lru_add_drain_cpu(get_cpu()); |
80bfed90 | 646 | put_cpu(); |
1da177e4 LT |
647 | } |
648 | ||
c4028958 | 649 | static void lru_add_drain_per_cpu(struct work_struct *dummy) |
053837fc NP |
650 | { |
651 | lru_add_drain(); | |
652 | } | |
653 | ||
654 | /* | |
655 | * Returns 0 for success | |
656 | */ | |
657 | int lru_add_drain_all(void) | |
658 | { | |
c4028958 | 659 | return schedule_on_each_cpu(lru_add_drain_per_cpu); |
053837fc NP |
660 | } |
661 | ||
1da177e4 LT |
662 | /* |
663 | * Batched page_cache_release(). Decrement the reference count on all the | |
664 | * passed pages. If it fell to zero then remove the page from the LRU and | |
665 | * free it. | |
666 | * | |
667 | * Avoid taking zone->lru_lock if possible, but if it is taken, retain it | |
668 | * for the remainder of the operation. | |
669 | * | |
ab33dc09 FLVC |
670 | * The locking in this function is against shrink_inactive_list(): we recheck |
671 | * the page count inside the lock to see whether shrink_inactive_list() | |
672 | * grabbed the page via the LRU. If it did, give up: shrink_inactive_list() | |
673 | * will free it. | |
1da177e4 LT |
674 | */ |
675 | void release_pages(struct page **pages, int nr, int cold) | |
676 | { | |
677 | int i; | |
cc59850e | 678 | LIST_HEAD(pages_to_free); |
1da177e4 | 679 | struct zone *zone = NULL; |
fa9add64 | 680 | struct lruvec *lruvec; |
902aaed0 | 681 | unsigned long uninitialized_var(flags); |
1da177e4 | 682 | |
1da177e4 LT |
683 | for (i = 0; i < nr; i++) { |
684 | struct page *page = pages[i]; | |
1da177e4 | 685 | |
8519fb30 NP |
686 | if (unlikely(PageCompound(page))) { |
687 | if (zone) { | |
902aaed0 | 688 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
8519fb30 NP |
689 | zone = NULL; |
690 | } | |
691 | put_compound_page(page); | |
692 | continue; | |
693 | } | |
694 | ||
b5810039 | 695 | if (!put_page_testzero(page)) |
1da177e4 LT |
696 | continue; |
697 | ||
46453a6e NP |
698 | if (PageLRU(page)) { |
699 | struct zone *pagezone = page_zone(page); | |
894bc310 | 700 | |
46453a6e NP |
701 | if (pagezone != zone) { |
702 | if (zone) | |
902aaed0 HH |
703 | spin_unlock_irqrestore(&zone->lru_lock, |
704 | flags); | |
46453a6e | 705 | zone = pagezone; |
902aaed0 | 706 | spin_lock_irqsave(&zone->lru_lock, flags); |
46453a6e | 707 | } |
fa9add64 HD |
708 | |
709 | lruvec = mem_cgroup_page_lruvec(page, zone); | |
725d704e | 710 | VM_BUG_ON(!PageLRU(page)); |
67453911 | 711 | __ClearPageLRU(page); |
fa9add64 | 712 | del_page_from_lru_list(page, lruvec, page_off_lru(page)); |
46453a6e NP |
713 | } |
714 | ||
cc59850e | 715 | list_add(&page->lru, &pages_to_free); |
1da177e4 LT |
716 | } |
717 | if (zone) | |
902aaed0 | 718 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
1da177e4 | 719 | |
cc59850e | 720 | free_hot_cold_page_list(&pages_to_free, cold); |
1da177e4 | 721 | } |
0be8557b | 722 | EXPORT_SYMBOL(release_pages); |
1da177e4 LT |
723 | |
724 | /* | |
725 | * The pages which we're about to release may be in the deferred lru-addition | |
726 | * queues. That would prevent them from really being freed right now. That's | |
727 | * OK from a correctness point of view but is inefficient - those pages may be | |
728 | * cache-warm and we want to give them back to the page allocator ASAP. | |
729 | * | |
730 | * So __pagevec_release() will drain those queues here. __pagevec_lru_add() | |
731 | * and __pagevec_lru_add_active() call release_pages() directly to avoid | |
732 | * mutual recursion. | |
733 | */ | |
734 | void __pagevec_release(struct pagevec *pvec) | |
735 | { | |
736 | lru_add_drain(); | |
737 | release_pages(pvec->pages, pagevec_count(pvec), pvec->cold); | |
738 | pagevec_reinit(pvec); | |
739 | } | |
7f285701 SF |
740 | EXPORT_SYMBOL(__pagevec_release); |
741 | ||
12d27107 | 742 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
71e3aac0 | 743 | /* used by __split_huge_page_refcount() */ |
fa9add64 | 744 | void lru_add_page_tail(struct page *page, struct page *page_tail, |
5bc7b8ac | 745 | struct lruvec *lruvec, struct list_head *list) |
71e3aac0 | 746 | { |
7512102c | 747 | int uninitialized_var(active); |
71e3aac0 AA |
748 | enum lru_list lru; |
749 | const int file = 0; | |
71e3aac0 AA |
750 | |
751 | VM_BUG_ON(!PageHead(page)); | |
752 | VM_BUG_ON(PageCompound(page_tail)); | |
753 | VM_BUG_ON(PageLRU(page_tail)); | |
fa9add64 HD |
754 | VM_BUG_ON(NR_CPUS != 1 && |
755 | !spin_is_locked(&lruvec_zone(lruvec)->lru_lock)); | |
71e3aac0 | 756 | |
5bc7b8ac SL |
757 | if (!list) |
758 | SetPageLRU(page_tail); | |
71e3aac0 | 759 | |
39b5f29a | 760 | if (page_evictable(page_tail)) { |
71e3aac0 AA |
761 | if (PageActive(page)) { |
762 | SetPageActive(page_tail); | |
763 | active = 1; | |
764 | lru = LRU_ACTIVE_ANON; | |
765 | } else { | |
766 | active = 0; | |
767 | lru = LRU_INACTIVE_ANON; | |
768 | } | |
71e3aac0 AA |
769 | } else { |
770 | SetPageUnevictable(page_tail); | |
12d27107 HD |
771 | lru = LRU_UNEVICTABLE; |
772 | } | |
773 | ||
774 | if (likely(PageLRU(page))) | |
775 | list_add_tail(&page_tail->lru, &page->lru); | |
5bc7b8ac SL |
776 | else if (list) { |
777 | /* page reclaim is reclaiming a huge page */ | |
778 | get_page(page_tail); | |
779 | list_add_tail(&page_tail->lru, list); | |
780 | } else { | |
12d27107 HD |
781 | struct list_head *list_head; |
782 | /* | |
783 | * Head page has not yet been counted, as an hpage, | |
784 | * so we must account for each subpage individually. | |
785 | * | |
786 | * Use the standard add function to put page_tail on the list, | |
787 | * but then correct its position so they all end up in order. | |
788 | */ | |
fa9add64 | 789 | add_page_to_lru_list(page_tail, lruvec, lru); |
12d27107 HD |
790 | list_head = page_tail->lru.prev; |
791 | list_move_tail(&page_tail->lru, list_head); | |
71e3aac0 | 792 | } |
7512102c HD |
793 | |
794 | if (!PageUnevictable(page)) | |
fa9add64 | 795 | update_page_reclaim_stat(lruvec, file, active); |
71e3aac0 | 796 | } |
12d27107 | 797 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
71e3aac0 | 798 | |
fa9add64 HD |
799 | static void __pagevec_lru_add_fn(struct page *page, struct lruvec *lruvec, |
800 | void *arg) | |
3dd7ae8e SL |
801 | { |
802 | enum lru_list lru = (enum lru_list)arg; | |
3dd7ae8e SL |
803 | int file = is_file_lru(lru); |
804 | int active = is_active_lru(lru); | |
805 | ||
806 | VM_BUG_ON(PageActive(page)); | |
807 | VM_BUG_ON(PageUnevictable(page)); | |
808 | VM_BUG_ON(PageLRU(page)); | |
809 | ||
810 | SetPageLRU(page); | |
811 | if (active) | |
812 | SetPageActive(page); | |
fa9add64 HD |
813 | add_page_to_lru_list(page, lruvec, lru); |
814 | update_page_reclaim_stat(lruvec, file, active); | |
c6286c98 | 815 | trace_mm_lru_insertion(page, page_to_pfn(page), lru, trace_pagemap_flags(page)); |
3dd7ae8e SL |
816 | } |
817 | ||
1da177e4 LT |
818 | /* |
819 | * Add the passed pages to the LRU, then drop the caller's refcount | |
820 | * on them. Reinitialises the caller's pagevec. | |
821 | */ | |
5095ae83 | 822 | void __pagevec_lru_add(struct pagevec *pvec, enum lru_list lru) |
1da177e4 | 823 | { |
894bc310 | 824 | VM_BUG_ON(is_unevictable_lru(lru)); |
1da177e4 | 825 | |
5095ae83 | 826 | pagevec_lru_move_fn(pvec, __pagevec_lru_add_fn, (void *)lru); |
1da177e4 | 827 | } |
5095ae83 | 828 | EXPORT_SYMBOL(__pagevec_lru_add); |
1da177e4 | 829 | |
1da177e4 LT |
830 | /** |
831 | * pagevec_lookup - gang pagecache lookup | |
832 | * @pvec: Where the resulting pages are placed | |
833 | * @mapping: The address_space to search | |
834 | * @start: The starting page index | |
835 | * @nr_pages: The maximum number of pages | |
836 | * | |
837 | * pagevec_lookup() will search for and return a group of up to @nr_pages pages | |
838 | * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a | |
839 | * reference against the pages in @pvec. | |
840 | * | |
841 | * The search returns a group of mapping-contiguous pages with ascending | |
842 | * indexes. There may be holes in the indices due to not-present pages. | |
843 | * | |
844 | * pagevec_lookup() returns the number of pages which were found. | |
845 | */ | |
846 | unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping, | |
847 | pgoff_t start, unsigned nr_pages) | |
848 | { | |
849 | pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages); | |
850 | return pagevec_count(pvec); | |
851 | } | |
78539fdf CH |
852 | EXPORT_SYMBOL(pagevec_lookup); |
853 | ||
1da177e4 LT |
854 | unsigned pagevec_lookup_tag(struct pagevec *pvec, struct address_space *mapping, |
855 | pgoff_t *index, int tag, unsigned nr_pages) | |
856 | { | |
857 | pvec->nr = find_get_pages_tag(mapping, index, tag, | |
858 | nr_pages, pvec->pages); | |
859 | return pagevec_count(pvec); | |
860 | } | |
7f285701 | 861 | EXPORT_SYMBOL(pagevec_lookup_tag); |
1da177e4 | 862 | |
1da177e4 LT |
863 | /* |
864 | * Perform any setup for the swap system | |
865 | */ | |
866 | void __init swap_setup(void) | |
867 | { | |
4481374c | 868 | unsigned long megs = totalram_pages >> (20 - PAGE_SHIFT); |
e0bf68dd | 869 | #ifdef CONFIG_SWAP |
33806f06 SL |
870 | int i; |
871 | ||
872 | bdi_init(swapper_spaces[0].backing_dev_info); | |
873 | for (i = 0; i < MAX_SWAPFILES; i++) { | |
874 | spin_lock_init(&swapper_spaces[i].tree_lock); | |
875 | INIT_LIST_HEAD(&swapper_spaces[i].i_mmap_nonlinear); | |
876 | } | |
e0bf68dd PZ |
877 | #endif |
878 | ||
1da177e4 LT |
879 | /* Use a smaller cluster for small-memory machines */ |
880 | if (megs < 16) | |
881 | page_cluster = 2; | |
882 | else | |
883 | page_cluster = 3; | |
884 | /* | |
885 | * Right now other parts of the system means that we | |
886 | * _really_ don't want to cluster much more | |
887 | */ | |
1da177e4 | 888 | } |