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 | ||
13f7f789 | 43 | static DEFINE_PER_CPU(struct pagevec, lru_add_pvec); |
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); |
309381fe | 60 | VM_BUG_ON_PAGE(!PageLRU(page), page); |
b221385b | 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 | ||
c747ce79 JZ |
82 | /** |
83 | * Two special cases here: we could avoid taking compound_lock_irqsave | |
84 | * and could skip the tail refcounting(in _mapcount). | |
85 | * | |
86 | * 1. Hugetlbfs page: | |
87 | * | |
88 | * PageHeadHuge will remain true until the compound page | |
89 | * is released and enters the buddy allocator, and it could | |
90 | * not be split by __split_huge_page_refcount(). | |
91 | * | |
92 | * So if we see PageHeadHuge set, and we have the tail page pin, | |
93 | * then we could safely put head page. | |
94 | * | |
95 | * 2. Slab THP page: | |
96 | * | |
97 | * PG_slab is cleared before the slab frees the head page, and | |
98 | * tail pin cannot be the last reference left on the head page, | |
99 | * because the slab code is free to reuse the compound page | |
100 | * after a kfree/kmem_cache_free without having to check if | |
101 | * there's any tail pin left. In turn all tail pinsmust be always | |
102 | * released while the head is still pinned by the slab code | |
103 | * and so we know PG_slab will be still set too. | |
104 | * | |
105 | * So if we see PageSlab set, and we have the tail page pin, | |
106 | * then we could safely put head page. | |
107 | */ | |
108 | static __always_inline | |
109 | void put_unrefcounted_compound_page(struct page *page_head, struct page *page) | |
110 | { | |
111 | /* | |
112 | * If @page is a THP tail, we must read the tail page | |
113 | * flags after the head page flags. The | |
114 | * __split_huge_page_refcount side enforces write memory barriers | |
115 | * between clearing PageTail and before the head page | |
116 | * can be freed and reallocated. | |
117 | */ | |
118 | smp_rmb(); | |
119 | if (likely(PageTail(page))) { | |
120 | /* | |
121 | * __split_huge_page_refcount cannot race | |
122 | * here, see the comment above this function. | |
123 | */ | |
124 | VM_BUG_ON_PAGE(!PageHead(page_head), page_head); | |
125 | VM_BUG_ON_PAGE(page_mapcount(page) != 0, page); | |
126 | if (put_page_testzero(page_head)) { | |
127 | /* | |
128 | * If this is the tail of a slab THP page, | |
129 | * the tail pin must not be the last reference | |
130 | * held on the page, because the PG_slab cannot | |
131 | * be cleared before all tail pins (which skips | |
132 | * the _mapcount tail refcounting) have been | |
133 | * released. | |
134 | * | |
135 | * If this is the tail of a hugetlbfs page, | |
136 | * the tail pin may be the last reference on | |
137 | * the page instead, because PageHeadHuge will | |
138 | * not go away until the compound page enters | |
139 | * the buddy allocator. | |
140 | */ | |
141 | VM_BUG_ON_PAGE(PageSlab(page_head), page_head); | |
142 | __put_compound_page(page_head); | |
143 | } | |
144 | } else | |
145 | /* | |
146 | * __split_huge_page_refcount run before us, | |
147 | * @page was a THP tail. The split @page_head | |
148 | * has been freed and reallocated as slab or | |
149 | * hugetlbfs page of smaller order (only | |
150 | * possible if reallocated as slab on x86). | |
151 | */ | |
152 | if (put_page_testzero(page)) | |
153 | __put_single_page(page); | |
154 | } | |
155 | ||
156 | static __always_inline | |
157 | void put_refcounted_compound_page(struct page *page_head, struct page *page) | |
158 | { | |
159 | if (likely(page != page_head && get_page_unless_zero(page_head))) { | |
160 | unsigned long flags; | |
161 | ||
162 | /* | |
163 | * @page_head wasn't a dangling pointer but it may not | |
164 | * be a head page anymore by the time we obtain the | |
165 | * lock. That is ok as long as it can't be freed from | |
166 | * under us. | |
167 | */ | |
168 | flags = compound_lock_irqsave(page_head); | |
169 | if (unlikely(!PageTail(page))) { | |
170 | /* __split_huge_page_refcount run before us */ | |
171 | compound_unlock_irqrestore(page_head, flags); | |
172 | if (put_page_testzero(page_head)) { | |
173 | /* | |
174 | * The @page_head may have been freed | |
175 | * and reallocated as a compound page | |
176 | * of smaller order and then freed | |
177 | * again. All we know is that it | |
178 | * cannot have become: a THP page, a | |
179 | * compound page of higher order, a | |
180 | * tail page. That is because we | |
181 | * still hold the refcount of the | |
182 | * split THP tail and page_head was | |
183 | * the THP head before the split. | |
184 | */ | |
185 | if (PageHead(page_head)) | |
186 | __put_compound_page(page_head); | |
187 | else | |
188 | __put_single_page(page_head); | |
189 | } | |
190 | out_put_single: | |
191 | if (put_page_testzero(page)) | |
192 | __put_single_page(page); | |
193 | return; | |
194 | } | |
195 | VM_BUG_ON_PAGE(page_head != page->first_page, page); | |
196 | /* | |
197 | * We can release the refcount taken by | |
198 | * get_page_unless_zero() now that | |
199 | * __split_huge_page_refcount() is blocked on the | |
200 | * compound_lock. | |
201 | */ | |
202 | if (put_page_testzero(page_head)) | |
203 | VM_BUG_ON_PAGE(1, page_head); | |
204 | /* __split_huge_page_refcount will wait now */ | |
205 | VM_BUG_ON_PAGE(page_mapcount(page) <= 0, page); | |
206 | atomic_dec(&page->_mapcount); | |
207 | VM_BUG_ON_PAGE(atomic_read(&page_head->_count) <= 0, page_head); | |
208 | VM_BUG_ON_PAGE(atomic_read(&page->_count) != 0, page); | |
209 | compound_unlock_irqrestore(page_head, flags); | |
210 | ||
211 | if (put_page_testzero(page_head)) { | |
212 | if (PageHead(page_head)) | |
213 | __put_compound_page(page_head); | |
214 | else | |
215 | __put_single_page(page_head); | |
216 | } | |
217 | } else { | |
218 | /* @page_head is a dangling pointer */ | |
219 | VM_BUG_ON_PAGE(PageTail(page), page); | |
220 | goto out_put_single; | |
221 | } | |
222 | } | |
223 | ||
91807063 AA |
224 | static void put_compound_page(struct page *page) |
225 | { | |
26296ad2 | 226 | struct page *page_head; |
70b50f94 | 227 | |
26296ad2 AM |
228 | if (likely(!PageTail(page))) { |
229 | if (put_page_testzero(page)) { | |
ebf360f9 | 230 | /* |
26296ad2 AM |
231 | * By the time all refcounts have been released |
232 | * split_huge_page cannot run anymore from under us. | |
ebf360f9 | 233 | */ |
26296ad2 AM |
234 | if (PageHead(page)) |
235 | __put_compound_page(page); | |
236 | else | |
237 | __put_single_page(page); | |
ebf360f9 | 238 | } |
26296ad2 AM |
239 | return; |
240 | } | |
ebf360f9 | 241 | |
26296ad2 | 242 | /* __split_huge_page_refcount can run under us */ |
668f9abb | 243 | page_head = compound_head(page); |
5bf5f03c | 244 | |
26296ad2 AM |
245 | /* |
246 | * THP can not break up slab pages so avoid taking | |
247 | * compound_lock() and skip the tail page refcounting (in | |
248 | * _mapcount) too. Slab performs non-atomic bit ops on | |
249 | * page->flags for better performance. In particular | |
250 | * slab_unlock() in slub used to be a hot path. It is still | |
251 | * hot on arches that do not support | |
252 | * this_cpu_cmpxchg_double(). | |
253 | * | |
254 | * If "page" is part of a slab or hugetlbfs page it cannot be | |
255 | * splitted and the head page cannot change from under us. And | |
256 | * if "page" is part of a THP page under splitting, if the | |
257 | * head page pointed by the THP tail isn't a THP head anymore, | |
258 | * we'll find PageTail clear after smp_rmb() and we'll treat | |
259 | * it as a single page. | |
260 | */ | |
261 | if (!__compound_tail_refcounted(page_head)) { | |
262 | /* | |
263 | * If "page" is a THP tail, we must read the tail page | |
264 | * flags after the head page flags. The | |
265 | * split_huge_page side enforces write memory barriers | |
266 | * between clearing PageTail and before the head page | |
267 | * can be freed and reallocated. | |
268 | */ | |
269 | smp_rmb(); | |
270 | if (likely(PageTail(page))) { | |
91807063 | 271 | /* |
26296ad2 AM |
272 | * __split_huge_page_refcount cannot race |
273 | * here. | |
91807063 | 274 | */ |
309381fe SL |
275 | VM_BUG_ON_PAGE(!PageHead(page_head), page_head); |
276 | VM_BUG_ON_PAGE(page_mapcount(page) != 0, page); | |
26296ad2 AM |
277 | if (put_page_testzero(page_head)) { |
278 | /* | |
279 | * If this is the tail of a slab | |
280 | * compound page, the tail pin must | |
281 | * not be the last reference held on | |
282 | * the page, because the PG_slab | |
283 | * cannot be cleared before all tail | |
284 | * pins (which skips the _mapcount | |
285 | * tail refcounting) have been | |
286 | * released. For hugetlbfs the tail | |
287 | * pin may be the last reference on | |
288 | * the page instead, because | |
289 | * PageHeadHuge will not go away until | |
290 | * the compound page enters the buddy | |
291 | * allocator. | |
292 | */ | |
309381fe | 293 | VM_BUG_ON_PAGE(PageSlab(page_head), page_head); |
26296ad2 | 294 | __put_compound_page(page_head); |
91807063 | 295 | } |
26296ad2 AM |
296 | return; |
297 | } else | |
91807063 | 298 | /* |
26296ad2 AM |
299 | * __split_huge_page_refcount run before us, |
300 | * "page" was a THP tail. The split page_head | |
301 | * has been freed and reallocated as slab or | |
302 | * hugetlbfs page of smaller order (only | |
303 | * possible if reallocated as slab on x86). | |
91807063 | 304 | */ |
26296ad2 AM |
305 | goto out_put_single; |
306 | } | |
5bf5f03c | 307 | |
26296ad2 AM |
308 | if (likely(page != page_head && get_page_unless_zero(page_head))) { |
309 | unsigned long flags; | |
310 | ||
311 | /* | |
312 | * page_head wasn't a dangling pointer but it may not | |
313 | * be a head page anymore by the time we obtain the | |
314 | * lock. That is ok as long as it can't be freed from | |
315 | * under us. | |
316 | */ | |
317 | flags = compound_lock_irqsave(page_head); | |
318 | if (unlikely(!PageTail(page))) { | |
319 | /* __split_huge_page_refcount run before us */ | |
320 | compound_unlock_irqrestore(page_head, flags); | |
a95a82e9 | 321 | if (put_page_testzero(page_head)) { |
26296ad2 AM |
322 | /* |
323 | * The head page may have been freed | |
324 | * and reallocated as a compound page | |
325 | * of smaller order and then freed | |
326 | * again. All we know is that it | |
327 | * cannot have become: a THP page, a | |
328 | * compound page of higher order, a | |
329 | * tail page. That is because we | |
330 | * still hold the refcount of the | |
331 | * split THP tail and page_head was | |
332 | * the THP head before the split. | |
333 | */ | |
a95a82e9 AA |
334 | if (PageHead(page_head)) |
335 | __put_compound_page(page_head); | |
336 | else | |
337 | __put_single_page(page_head); | |
338 | } | |
26296ad2 AM |
339 | out_put_single: |
340 | if (put_page_testzero(page)) | |
341 | __put_single_page(page); | |
342 | return; | |
343 | } | |
309381fe | 344 | VM_BUG_ON_PAGE(page_head != page->first_page, page); |
26296ad2 AM |
345 | /* |
346 | * We can release the refcount taken by | |
347 | * get_page_unless_zero() now that | |
348 | * __split_huge_page_refcount() is blocked on the | |
349 | * compound_lock. | |
350 | */ | |
351 | if (put_page_testzero(page_head)) | |
309381fe | 352 | VM_BUG_ON_PAGE(1, page_head); |
26296ad2 | 353 | /* __split_huge_page_refcount will wait now */ |
309381fe | 354 | VM_BUG_ON_PAGE(page_mapcount(page) <= 0, page); |
26296ad2 | 355 | atomic_dec(&page->_mapcount); |
309381fe SL |
356 | VM_BUG_ON_PAGE(atomic_read(&page_head->_count) <= 0, page_head); |
357 | VM_BUG_ON_PAGE(atomic_read(&page->_count) != 0, page); | |
26296ad2 AM |
358 | compound_unlock_irqrestore(page_head, flags); |
359 | ||
360 | if (put_page_testzero(page_head)) { | |
361 | if (PageHead(page_head)) | |
362 | __put_compound_page(page_head); | |
363 | else | |
364 | __put_single_page(page_head); | |
91807063 | 365 | } |
26296ad2 AM |
366 | } else { |
367 | /* page_head is a dangling pointer */ | |
309381fe | 368 | VM_BUG_ON_PAGE(PageTail(page), page); |
26296ad2 | 369 | goto out_put_single; |
1da177e4 | 370 | } |
8519fb30 NP |
371 | } |
372 | ||
373 | void put_page(struct page *page) | |
374 | { | |
375 | if (unlikely(PageCompound(page))) | |
376 | put_compound_page(page); | |
377 | else if (put_page_testzero(page)) | |
91807063 | 378 | __put_single_page(page); |
1da177e4 LT |
379 | } |
380 | EXPORT_SYMBOL(put_page); | |
1da177e4 | 381 | |
70b50f94 AA |
382 | /* |
383 | * This function is exported but must not be called by anything other | |
384 | * than get_page(). It implements the slow path of get_page(). | |
385 | */ | |
386 | bool __get_page_tail(struct page *page) | |
387 | { | |
388 | /* | |
389 | * This takes care of get_page() if run on a tail page | |
390 | * returned by one of the get_user_pages/follow_page variants. | |
391 | * get_user_pages/follow_page itself doesn't need the compound | |
392 | * lock because it runs __get_page_tail_foll() under the | |
393 | * proper PT lock that already serializes against | |
394 | * split_huge_page(). | |
395 | */ | |
27c73ae7 | 396 | unsigned long flags; |
ebf360f9 | 397 | bool got; |
668f9abb | 398 | struct page *page_head = compound_head(page); |
70b50f94 | 399 | |
ebf360f9 | 400 | /* Ref to put_compound_page() comment. */ |
3bfcd13e | 401 | if (!__compound_tail_refcounted(page_head)) { |
ebf360f9 AA |
402 | smp_rmb(); |
403 | if (likely(PageTail(page))) { | |
404 | /* | |
405 | * This is a hugetlbfs page or a slab | |
406 | * page. __split_huge_page_refcount | |
407 | * cannot race here. | |
408 | */ | |
309381fe | 409 | VM_BUG_ON_PAGE(!PageHead(page_head), page_head); |
ebf360f9 AA |
410 | __get_page_tail_foll(page, true); |
411 | return true; | |
412 | } else { | |
413 | /* | |
414 | * __split_huge_page_refcount run | |
415 | * before us, "page" was a THP | |
416 | * tail. The split page_head has been | |
417 | * freed and reallocated as slab or | |
418 | * hugetlbfs page of smaller order | |
419 | * (only possible if reallocated as | |
420 | * slab on x86). | |
421 | */ | |
422 | return false; | |
27c73ae7 | 423 | } |
ebf360f9 | 424 | } |
27c73ae7 | 425 | |
ebf360f9 AA |
426 | got = false; |
427 | if (likely(page != page_head && get_page_unless_zero(page_head))) { | |
27c73ae7 AA |
428 | /* |
429 | * page_head wasn't a dangling pointer but it | |
430 | * may not be a head page anymore by the time | |
431 | * we obtain the lock. That is ok as long as it | |
432 | * can't be freed from under us. | |
433 | */ | |
434 | flags = compound_lock_irqsave(page_head); | |
435 | /* here __split_huge_page_refcount won't run anymore */ | |
436 | if (likely(PageTail(page))) { | |
437 | __get_page_tail_foll(page, false); | |
438 | got = true; | |
5bf5f03c | 439 | } |
27c73ae7 AA |
440 | compound_unlock_irqrestore(page_head, flags); |
441 | if (unlikely(!got)) | |
442 | put_page(page_head); | |
70b50f94 AA |
443 | } |
444 | return got; | |
445 | } | |
446 | EXPORT_SYMBOL(__get_page_tail); | |
447 | ||
1d7ea732 | 448 | /** |
7682486b RD |
449 | * put_pages_list() - release a list of pages |
450 | * @pages: list of pages threaded on page->lru | |
1d7ea732 AZ |
451 | * |
452 | * Release a list of pages which are strung together on page.lru. Currently | |
453 | * used by read_cache_pages() and related error recovery code. | |
1d7ea732 AZ |
454 | */ |
455 | void put_pages_list(struct list_head *pages) | |
456 | { | |
457 | while (!list_empty(pages)) { | |
458 | struct page *victim; | |
459 | ||
460 | victim = list_entry(pages->prev, struct page, lru); | |
461 | list_del(&victim->lru); | |
462 | page_cache_release(victim); | |
463 | } | |
464 | } | |
465 | EXPORT_SYMBOL(put_pages_list); | |
466 | ||
18022c5d MG |
467 | /* |
468 | * get_kernel_pages() - pin kernel pages in memory | |
469 | * @kiov: An array of struct kvec structures | |
470 | * @nr_segs: number of segments to pin | |
471 | * @write: pinning for read/write, currently ignored | |
472 | * @pages: array that receives pointers to the pages pinned. | |
473 | * Should be at least nr_segs long. | |
474 | * | |
475 | * Returns number of pages pinned. This may be fewer than the number | |
476 | * requested. If nr_pages is 0 or negative, returns 0. If no pages | |
477 | * were pinned, returns -errno. Each page returned must be released | |
478 | * with a put_page() call when it is finished with. | |
479 | */ | |
480 | int get_kernel_pages(const struct kvec *kiov, int nr_segs, int write, | |
481 | struct page **pages) | |
482 | { | |
483 | int seg; | |
484 | ||
485 | for (seg = 0; seg < nr_segs; seg++) { | |
486 | if (WARN_ON(kiov[seg].iov_len != PAGE_SIZE)) | |
487 | return seg; | |
488 | ||
5a178119 | 489 | pages[seg] = kmap_to_page(kiov[seg].iov_base); |
18022c5d MG |
490 | page_cache_get(pages[seg]); |
491 | } | |
492 | ||
493 | return seg; | |
494 | } | |
495 | EXPORT_SYMBOL_GPL(get_kernel_pages); | |
496 | ||
497 | /* | |
498 | * get_kernel_page() - pin a kernel page in memory | |
499 | * @start: starting kernel address | |
500 | * @write: pinning for read/write, currently ignored | |
501 | * @pages: array that receives pointer to the page pinned. | |
502 | * Must be at least nr_segs long. | |
503 | * | |
504 | * Returns 1 if page is pinned. If the page was not pinned, returns | |
505 | * -errno. The page returned must be released with a put_page() call | |
506 | * when it is finished with. | |
507 | */ | |
508 | int get_kernel_page(unsigned long start, int write, struct page **pages) | |
509 | { | |
510 | const struct kvec kiov = { | |
511 | .iov_base = (void *)start, | |
512 | .iov_len = PAGE_SIZE | |
513 | }; | |
514 | ||
515 | return get_kernel_pages(&kiov, 1, write, pages); | |
516 | } | |
517 | EXPORT_SYMBOL_GPL(get_kernel_page); | |
518 | ||
3dd7ae8e | 519 | static void pagevec_lru_move_fn(struct pagevec *pvec, |
fa9add64 HD |
520 | void (*move_fn)(struct page *page, struct lruvec *lruvec, void *arg), |
521 | void *arg) | |
902aaed0 HH |
522 | { |
523 | int i; | |
902aaed0 | 524 | struct zone *zone = NULL; |
fa9add64 | 525 | struct lruvec *lruvec; |
3dd7ae8e | 526 | unsigned long flags = 0; |
902aaed0 HH |
527 | |
528 | for (i = 0; i < pagevec_count(pvec); i++) { | |
529 | struct page *page = pvec->pages[i]; | |
530 | struct zone *pagezone = page_zone(page); | |
531 | ||
532 | if (pagezone != zone) { | |
533 | if (zone) | |
3dd7ae8e | 534 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
902aaed0 | 535 | zone = pagezone; |
3dd7ae8e | 536 | spin_lock_irqsave(&zone->lru_lock, flags); |
902aaed0 | 537 | } |
3dd7ae8e | 538 | |
fa9add64 HD |
539 | lruvec = mem_cgroup_page_lruvec(page, zone); |
540 | (*move_fn)(page, lruvec, arg); | |
902aaed0 HH |
541 | } |
542 | if (zone) | |
3dd7ae8e | 543 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
83896fb5 LT |
544 | release_pages(pvec->pages, pvec->nr, pvec->cold); |
545 | pagevec_reinit(pvec); | |
d8505dee SL |
546 | } |
547 | ||
fa9add64 HD |
548 | static void pagevec_move_tail_fn(struct page *page, struct lruvec *lruvec, |
549 | void *arg) | |
3dd7ae8e SL |
550 | { |
551 | int *pgmoved = arg; | |
3dd7ae8e SL |
552 | |
553 | if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) { | |
554 | enum lru_list lru = page_lru_base_type(page); | |
925b7673 | 555 | list_move_tail(&page->lru, &lruvec->lists[lru]); |
3dd7ae8e SL |
556 | (*pgmoved)++; |
557 | } | |
558 | } | |
559 | ||
560 | /* | |
561 | * pagevec_move_tail() must be called with IRQ disabled. | |
562 | * Otherwise this may cause nasty races. | |
563 | */ | |
564 | static void pagevec_move_tail(struct pagevec *pvec) | |
565 | { | |
566 | int pgmoved = 0; | |
567 | ||
568 | pagevec_lru_move_fn(pvec, pagevec_move_tail_fn, &pgmoved); | |
569 | __count_vm_events(PGROTATED, pgmoved); | |
570 | } | |
571 | ||
1da177e4 LT |
572 | /* |
573 | * Writeback is about to end against a page which has been marked for immediate | |
574 | * reclaim. If it still appears to be reclaimable, move it to the tail of the | |
902aaed0 | 575 | * inactive list. |
1da177e4 | 576 | */ |
3dd7ae8e | 577 | void rotate_reclaimable_page(struct page *page) |
1da177e4 | 578 | { |
ac6aadb2 | 579 | if (!PageLocked(page) && !PageDirty(page) && !PageActive(page) && |
894bc310 | 580 | !PageUnevictable(page) && PageLRU(page)) { |
ac6aadb2 MS |
581 | struct pagevec *pvec; |
582 | unsigned long flags; | |
583 | ||
584 | page_cache_get(page); | |
585 | local_irq_save(flags); | |
7c8e0181 | 586 | pvec = this_cpu_ptr(&lru_rotate_pvecs); |
ac6aadb2 MS |
587 | if (!pagevec_add(pvec, page)) |
588 | pagevec_move_tail(pvec); | |
589 | local_irq_restore(flags); | |
590 | } | |
1da177e4 LT |
591 | } |
592 | ||
fa9add64 | 593 | static void update_page_reclaim_stat(struct lruvec *lruvec, |
3e2f41f1 KM |
594 | int file, int rotated) |
595 | { | |
fa9add64 | 596 | struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat; |
3e2f41f1 KM |
597 | |
598 | reclaim_stat->recent_scanned[file]++; | |
599 | if (rotated) | |
600 | reclaim_stat->recent_rotated[file]++; | |
3e2f41f1 KM |
601 | } |
602 | ||
fa9add64 HD |
603 | static void __activate_page(struct page *page, struct lruvec *lruvec, |
604 | void *arg) | |
1da177e4 | 605 | { |
744ed144 | 606 | if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) { |
7a608572 LT |
607 | int file = page_is_file_cache(page); |
608 | int lru = page_lru_base_type(page); | |
744ed144 | 609 | |
fa9add64 | 610 | del_page_from_lru_list(page, lruvec, lru); |
7a608572 LT |
611 | SetPageActive(page); |
612 | lru += LRU_ACTIVE; | |
fa9add64 | 613 | add_page_to_lru_list(page, lruvec, lru); |
c6286c98 | 614 | trace_mm_lru_activate(page, page_to_pfn(page)); |
4f98a2fe | 615 | |
fa9add64 HD |
616 | __count_vm_event(PGACTIVATE); |
617 | update_page_reclaim_stat(lruvec, file, 1); | |
1da177e4 | 618 | } |
eb709b0d SL |
619 | } |
620 | ||
621 | #ifdef CONFIG_SMP | |
622 | static DEFINE_PER_CPU(struct pagevec, activate_page_pvecs); | |
623 | ||
624 | static void activate_page_drain(int cpu) | |
625 | { | |
626 | struct pagevec *pvec = &per_cpu(activate_page_pvecs, cpu); | |
627 | ||
628 | if (pagevec_count(pvec)) | |
629 | pagevec_lru_move_fn(pvec, __activate_page, NULL); | |
630 | } | |
631 | ||
5fbc4616 CM |
632 | static bool need_activate_page_drain(int cpu) |
633 | { | |
634 | return pagevec_count(&per_cpu(activate_page_pvecs, cpu)) != 0; | |
635 | } | |
636 | ||
eb709b0d SL |
637 | void activate_page(struct page *page) |
638 | { | |
639 | if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) { | |
640 | struct pagevec *pvec = &get_cpu_var(activate_page_pvecs); | |
641 | ||
642 | page_cache_get(page); | |
643 | if (!pagevec_add(pvec, page)) | |
644 | pagevec_lru_move_fn(pvec, __activate_page, NULL); | |
645 | put_cpu_var(activate_page_pvecs); | |
646 | } | |
647 | } | |
648 | ||
649 | #else | |
650 | static inline void activate_page_drain(int cpu) | |
651 | { | |
652 | } | |
653 | ||
5fbc4616 CM |
654 | static bool need_activate_page_drain(int cpu) |
655 | { | |
656 | return false; | |
657 | } | |
658 | ||
eb709b0d SL |
659 | void activate_page(struct page *page) |
660 | { | |
661 | struct zone *zone = page_zone(page); | |
662 | ||
663 | spin_lock_irq(&zone->lru_lock); | |
fa9add64 | 664 | __activate_page(page, mem_cgroup_page_lruvec(page, zone), NULL); |
1da177e4 LT |
665 | spin_unlock_irq(&zone->lru_lock); |
666 | } | |
eb709b0d | 667 | #endif |
1da177e4 | 668 | |
059285a2 MG |
669 | static void __lru_cache_activate_page(struct page *page) |
670 | { | |
671 | struct pagevec *pvec = &get_cpu_var(lru_add_pvec); | |
672 | int i; | |
673 | ||
674 | /* | |
675 | * Search backwards on the optimistic assumption that the page being | |
676 | * activated has just been added to this pagevec. Note that only | |
677 | * the local pagevec is examined as a !PageLRU page could be in the | |
678 | * process of being released, reclaimed, migrated or on a remote | |
679 | * pagevec that is currently being drained. Furthermore, marking | |
680 | * a remote pagevec's page PageActive potentially hits a race where | |
681 | * a page is marked PageActive just after it is added to the inactive | |
682 | * list causing accounting errors and BUG_ON checks to trigger. | |
683 | */ | |
684 | for (i = pagevec_count(pvec) - 1; i >= 0; i--) { | |
685 | struct page *pagevec_page = pvec->pages[i]; | |
686 | ||
687 | if (pagevec_page == page) { | |
688 | SetPageActive(page); | |
689 | break; | |
690 | } | |
691 | } | |
692 | ||
693 | put_cpu_var(lru_add_pvec); | |
694 | } | |
695 | ||
1da177e4 LT |
696 | /* |
697 | * Mark a page as having seen activity. | |
698 | * | |
699 | * inactive,unreferenced -> inactive,referenced | |
700 | * inactive,referenced -> active,unreferenced | |
701 | * active,unreferenced -> active,referenced | |
702 | */ | |
920c7a5d | 703 | void mark_page_accessed(struct page *page) |
1da177e4 | 704 | { |
894bc310 | 705 | if (!PageActive(page) && !PageUnevictable(page) && |
059285a2 MG |
706 | PageReferenced(page)) { |
707 | ||
708 | /* | |
709 | * If the page is on the LRU, queue it for activation via | |
710 | * activate_page_pvecs. Otherwise, assume the page is on a | |
711 | * pagevec, mark it active and it'll be moved to the active | |
712 | * LRU on the next drain. | |
713 | */ | |
714 | if (PageLRU(page)) | |
715 | activate_page(page); | |
716 | else | |
717 | __lru_cache_activate_page(page); | |
1da177e4 | 718 | ClearPageReferenced(page); |
a528910e JW |
719 | if (page_is_file_cache(page)) |
720 | workingset_activation(page); | |
1da177e4 LT |
721 | } else if (!PageReferenced(page)) { |
722 | SetPageReferenced(page); | |
723 | } | |
724 | } | |
1da177e4 LT |
725 | EXPORT_SYMBOL(mark_page_accessed); |
726 | ||
2329d375 | 727 | static void __lru_cache_add(struct page *page) |
1da177e4 | 728 | { |
13f7f789 MG |
729 | struct pagevec *pvec = &get_cpu_var(lru_add_pvec); |
730 | ||
1da177e4 | 731 | page_cache_get(page); |
d741c9cd | 732 | if (!pagevec_space(pvec)) |
a0b8cab3 | 733 | __pagevec_lru_add(pvec); |
d741c9cd | 734 | pagevec_add(pvec, page); |
13f7f789 | 735 | put_cpu_var(lru_add_pvec); |
1da177e4 | 736 | } |
2329d375 JZ |
737 | |
738 | /** | |
739 | * lru_cache_add: add a page to the page lists | |
740 | * @page: the page to add | |
741 | */ | |
742 | void lru_cache_add_anon(struct page *page) | |
743 | { | |
744 | ClearPageActive(page); | |
745 | __lru_cache_add(page); | |
746 | } | |
747 | ||
748 | void lru_cache_add_file(struct page *page) | |
749 | { | |
750 | ClearPageActive(page); | |
751 | __lru_cache_add(page); | |
752 | } | |
753 | EXPORT_SYMBOL(lru_cache_add_file); | |
1da177e4 | 754 | |
f04e9ebb | 755 | /** |
c53954a0 | 756 | * lru_cache_add - add a page to a page list |
f04e9ebb | 757 | * @page: the page to be added to the LRU. |
2329d375 JZ |
758 | * |
759 | * Queue the page for addition to the LRU via pagevec. The decision on whether | |
760 | * to add the page to the [in]active [file|anon] list is deferred until the | |
761 | * pagevec is drained. This gives a chance for the caller of lru_cache_add() | |
762 | * have the page added to the active list using mark_page_accessed(). | |
f04e9ebb | 763 | */ |
c53954a0 | 764 | void lru_cache_add(struct page *page) |
1da177e4 | 765 | { |
309381fe SL |
766 | VM_BUG_ON_PAGE(PageActive(page) && PageUnevictable(page), page); |
767 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
c53954a0 | 768 | __lru_cache_add(page); |
1da177e4 LT |
769 | } |
770 | ||
894bc310 LS |
771 | /** |
772 | * add_page_to_unevictable_list - add a page to the unevictable list | |
773 | * @page: the page to be added to the unevictable list | |
774 | * | |
775 | * Add page directly to its zone's unevictable list. To avoid races with | |
776 | * tasks that might be making the page evictable, through eg. munlock, | |
777 | * munmap or exit, while it's not on the lru, we want to add the page | |
778 | * while it's locked or otherwise "invisible" to other tasks. This is | |
779 | * difficult to do when using the pagevec cache, so bypass that. | |
780 | */ | |
781 | void add_page_to_unevictable_list(struct page *page) | |
782 | { | |
783 | struct zone *zone = page_zone(page); | |
fa9add64 | 784 | struct lruvec *lruvec; |
894bc310 LS |
785 | |
786 | spin_lock_irq(&zone->lru_lock); | |
fa9add64 | 787 | lruvec = mem_cgroup_page_lruvec(page, zone); |
ef2a2cbd | 788 | ClearPageActive(page); |
894bc310 LS |
789 | SetPageUnevictable(page); |
790 | SetPageLRU(page); | |
fa9add64 | 791 | add_page_to_lru_list(page, lruvec, LRU_UNEVICTABLE); |
894bc310 LS |
792 | spin_unlock_irq(&zone->lru_lock); |
793 | } | |
794 | ||
31560180 MK |
795 | /* |
796 | * If the page can not be invalidated, it is moved to the | |
797 | * inactive list to speed up its reclaim. It is moved to the | |
798 | * head of the list, rather than the tail, to give the flusher | |
799 | * threads some time to write it out, as this is much more | |
800 | * effective than the single-page writeout from reclaim. | |
278df9f4 MK |
801 | * |
802 | * If the page isn't page_mapped and dirty/writeback, the page | |
803 | * could reclaim asap using PG_reclaim. | |
804 | * | |
805 | * 1. active, mapped page -> none | |
806 | * 2. active, dirty/writeback page -> inactive, head, PG_reclaim | |
807 | * 3. inactive, mapped page -> none | |
808 | * 4. inactive, dirty/writeback page -> inactive, head, PG_reclaim | |
809 | * 5. inactive, clean -> inactive, tail | |
810 | * 6. Others -> none | |
811 | * | |
812 | * In 4, why it moves inactive's head, the VM expects the page would | |
813 | * be write it out by flusher threads as this is much more effective | |
814 | * than the single-page writeout from reclaim. | |
31560180 | 815 | */ |
fa9add64 HD |
816 | static void lru_deactivate_fn(struct page *page, struct lruvec *lruvec, |
817 | void *arg) | |
31560180 MK |
818 | { |
819 | int lru, file; | |
278df9f4 | 820 | bool active; |
31560180 | 821 | |
278df9f4 | 822 | if (!PageLRU(page)) |
31560180 MK |
823 | return; |
824 | ||
bad49d9c MK |
825 | if (PageUnevictable(page)) |
826 | return; | |
827 | ||
31560180 MK |
828 | /* Some processes are using the page */ |
829 | if (page_mapped(page)) | |
830 | return; | |
831 | ||
278df9f4 | 832 | active = PageActive(page); |
31560180 MK |
833 | file = page_is_file_cache(page); |
834 | lru = page_lru_base_type(page); | |
fa9add64 HD |
835 | |
836 | del_page_from_lru_list(page, lruvec, lru + active); | |
31560180 MK |
837 | ClearPageActive(page); |
838 | ClearPageReferenced(page); | |
fa9add64 | 839 | add_page_to_lru_list(page, lruvec, lru); |
31560180 | 840 | |
278df9f4 MK |
841 | if (PageWriteback(page) || PageDirty(page)) { |
842 | /* | |
843 | * PG_reclaim could be raced with end_page_writeback | |
844 | * It can make readahead confusing. But race window | |
845 | * is _really_ small and it's non-critical problem. | |
846 | */ | |
847 | SetPageReclaim(page); | |
848 | } else { | |
849 | /* | |
850 | * The page's writeback ends up during pagevec | |
851 | * We moves tha page into tail of inactive. | |
852 | */ | |
925b7673 | 853 | list_move_tail(&page->lru, &lruvec->lists[lru]); |
278df9f4 MK |
854 | __count_vm_event(PGROTATED); |
855 | } | |
856 | ||
857 | if (active) | |
858 | __count_vm_event(PGDEACTIVATE); | |
fa9add64 | 859 | update_page_reclaim_stat(lruvec, file, 0); |
31560180 MK |
860 | } |
861 | ||
902aaed0 HH |
862 | /* |
863 | * Drain pages out of the cpu's pagevecs. | |
864 | * Either "cpu" is the current CPU, and preemption has already been | |
865 | * disabled; or "cpu" is being hot-unplugged, and is already dead. | |
866 | */ | |
f0cb3c76 | 867 | void lru_add_drain_cpu(int cpu) |
1da177e4 | 868 | { |
13f7f789 | 869 | struct pagevec *pvec = &per_cpu(lru_add_pvec, cpu); |
1da177e4 | 870 | |
13f7f789 | 871 | if (pagevec_count(pvec)) |
a0b8cab3 | 872 | __pagevec_lru_add(pvec); |
902aaed0 HH |
873 | |
874 | pvec = &per_cpu(lru_rotate_pvecs, cpu); | |
875 | if (pagevec_count(pvec)) { | |
876 | unsigned long flags; | |
877 | ||
878 | /* No harm done if a racing interrupt already did this */ | |
879 | local_irq_save(flags); | |
880 | pagevec_move_tail(pvec); | |
881 | local_irq_restore(flags); | |
882 | } | |
31560180 MK |
883 | |
884 | pvec = &per_cpu(lru_deactivate_pvecs, cpu); | |
885 | if (pagevec_count(pvec)) | |
3dd7ae8e | 886 | pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL); |
eb709b0d SL |
887 | |
888 | activate_page_drain(cpu); | |
31560180 MK |
889 | } |
890 | ||
891 | /** | |
892 | * deactivate_page - forcefully deactivate a page | |
893 | * @page: page to deactivate | |
894 | * | |
895 | * This function hints the VM that @page is a good reclaim candidate, | |
896 | * for example if its invalidation fails due to the page being dirty | |
897 | * or under writeback. | |
898 | */ | |
899 | void deactivate_page(struct page *page) | |
900 | { | |
821ed6bb MK |
901 | /* |
902 | * In a workload with many unevictable page such as mprotect, unevictable | |
903 | * page deactivation for accelerating reclaim is pointless. | |
904 | */ | |
905 | if (PageUnevictable(page)) | |
906 | return; | |
907 | ||
31560180 MK |
908 | if (likely(get_page_unless_zero(page))) { |
909 | struct pagevec *pvec = &get_cpu_var(lru_deactivate_pvecs); | |
910 | ||
911 | if (!pagevec_add(pvec, page)) | |
3dd7ae8e | 912 | pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL); |
31560180 MK |
913 | put_cpu_var(lru_deactivate_pvecs); |
914 | } | |
80bfed90 AM |
915 | } |
916 | ||
917 | void lru_add_drain(void) | |
918 | { | |
f0cb3c76 | 919 | lru_add_drain_cpu(get_cpu()); |
80bfed90 | 920 | put_cpu(); |
1da177e4 LT |
921 | } |
922 | ||
c4028958 | 923 | static void lru_add_drain_per_cpu(struct work_struct *dummy) |
053837fc NP |
924 | { |
925 | lru_add_drain(); | |
926 | } | |
927 | ||
5fbc4616 CM |
928 | static DEFINE_PER_CPU(struct work_struct, lru_add_drain_work); |
929 | ||
930 | void lru_add_drain_all(void) | |
053837fc | 931 | { |
5fbc4616 CM |
932 | static DEFINE_MUTEX(lock); |
933 | static struct cpumask has_work; | |
934 | int cpu; | |
935 | ||
936 | mutex_lock(&lock); | |
937 | get_online_cpus(); | |
938 | cpumask_clear(&has_work); | |
939 | ||
940 | for_each_online_cpu(cpu) { | |
941 | struct work_struct *work = &per_cpu(lru_add_drain_work, cpu); | |
942 | ||
943 | if (pagevec_count(&per_cpu(lru_add_pvec, cpu)) || | |
944 | pagevec_count(&per_cpu(lru_rotate_pvecs, cpu)) || | |
945 | pagevec_count(&per_cpu(lru_deactivate_pvecs, cpu)) || | |
946 | need_activate_page_drain(cpu)) { | |
947 | INIT_WORK(work, lru_add_drain_per_cpu); | |
948 | schedule_work_on(cpu, work); | |
949 | cpumask_set_cpu(cpu, &has_work); | |
950 | } | |
951 | } | |
952 | ||
953 | for_each_cpu(cpu, &has_work) | |
954 | flush_work(&per_cpu(lru_add_drain_work, cpu)); | |
955 | ||
956 | put_online_cpus(); | |
957 | mutex_unlock(&lock); | |
053837fc NP |
958 | } |
959 | ||
1da177e4 LT |
960 | /* |
961 | * Batched page_cache_release(). Decrement the reference count on all the | |
962 | * passed pages. If it fell to zero then remove the page from the LRU and | |
963 | * free it. | |
964 | * | |
965 | * Avoid taking zone->lru_lock if possible, but if it is taken, retain it | |
966 | * for the remainder of the operation. | |
967 | * | |
ab33dc09 FLVC |
968 | * The locking in this function is against shrink_inactive_list(): we recheck |
969 | * the page count inside the lock to see whether shrink_inactive_list() | |
970 | * grabbed the page via the LRU. If it did, give up: shrink_inactive_list() | |
971 | * will free it. | |
1da177e4 LT |
972 | */ |
973 | void release_pages(struct page **pages, int nr, int cold) | |
974 | { | |
975 | int i; | |
cc59850e | 976 | LIST_HEAD(pages_to_free); |
1da177e4 | 977 | struct zone *zone = NULL; |
fa9add64 | 978 | struct lruvec *lruvec; |
902aaed0 | 979 | unsigned long uninitialized_var(flags); |
1da177e4 | 980 | |
1da177e4 LT |
981 | for (i = 0; i < nr; i++) { |
982 | struct page *page = pages[i]; | |
1da177e4 | 983 | |
8519fb30 NP |
984 | if (unlikely(PageCompound(page))) { |
985 | if (zone) { | |
902aaed0 | 986 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
8519fb30 NP |
987 | zone = NULL; |
988 | } | |
989 | put_compound_page(page); | |
990 | continue; | |
991 | } | |
992 | ||
b5810039 | 993 | if (!put_page_testzero(page)) |
1da177e4 LT |
994 | continue; |
995 | ||
46453a6e NP |
996 | if (PageLRU(page)) { |
997 | struct zone *pagezone = page_zone(page); | |
894bc310 | 998 | |
46453a6e NP |
999 | if (pagezone != zone) { |
1000 | if (zone) | |
902aaed0 HH |
1001 | spin_unlock_irqrestore(&zone->lru_lock, |
1002 | flags); | |
46453a6e | 1003 | zone = pagezone; |
902aaed0 | 1004 | spin_lock_irqsave(&zone->lru_lock, flags); |
46453a6e | 1005 | } |
fa9add64 HD |
1006 | |
1007 | lruvec = mem_cgroup_page_lruvec(page, zone); | |
309381fe | 1008 | VM_BUG_ON_PAGE(!PageLRU(page), page); |
67453911 | 1009 | __ClearPageLRU(page); |
fa9add64 | 1010 | del_page_from_lru_list(page, lruvec, page_off_lru(page)); |
46453a6e NP |
1011 | } |
1012 | ||
c53954a0 MG |
1013 | /* Clear Active bit in case of parallel mark_page_accessed */ |
1014 | ClearPageActive(page); | |
1015 | ||
cc59850e | 1016 | list_add(&page->lru, &pages_to_free); |
1da177e4 LT |
1017 | } |
1018 | if (zone) | |
902aaed0 | 1019 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
1da177e4 | 1020 | |
cc59850e | 1021 | free_hot_cold_page_list(&pages_to_free, cold); |
1da177e4 | 1022 | } |
0be8557b | 1023 | EXPORT_SYMBOL(release_pages); |
1da177e4 LT |
1024 | |
1025 | /* | |
1026 | * The pages which we're about to release may be in the deferred lru-addition | |
1027 | * queues. That would prevent them from really being freed right now. That's | |
1028 | * OK from a correctness point of view but is inefficient - those pages may be | |
1029 | * cache-warm and we want to give them back to the page allocator ASAP. | |
1030 | * | |
1031 | * So __pagevec_release() will drain those queues here. __pagevec_lru_add() | |
1032 | * and __pagevec_lru_add_active() call release_pages() directly to avoid | |
1033 | * mutual recursion. | |
1034 | */ | |
1035 | void __pagevec_release(struct pagevec *pvec) | |
1036 | { | |
1037 | lru_add_drain(); | |
1038 | release_pages(pvec->pages, pagevec_count(pvec), pvec->cold); | |
1039 | pagevec_reinit(pvec); | |
1040 | } | |
7f285701 SF |
1041 | EXPORT_SYMBOL(__pagevec_release); |
1042 | ||
12d27107 | 1043 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
71e3aac0 | 1044 | /* used by __split_huge_page_refcount() */ |
fa9add64 | 1045 | void lru_add_page_tail(struct page *page, struct page *page_tail, |
5bc7b8ac | 1046 | struct lruvec *lruvec, struct list_head *list) |
71e3aac0 | 1047 | { |
71e3aac0 | 1048 | const int file = 0; |
71e3aac0 | 1049 | |
309381fe SL |
1050 | VM_BUG_ON_PAGE(!PageHead(page), page); |
1051 | VM_BUG_ON_PAGE(PageCompound(page_tail), page); | |
1052 | VM_BUG_ON_PAGE(PageLRU(page_tail), page); | |
fa9add64 HD |
1053 | VM_BUG_ON(NR_CPUS != 1 && |
1054 | !spin_is_locked(&lruvec_zone(lruvec)->lru_lock)); | |
71e3aac0 | 1055 | |
5bc7b8ac SL |
1056 | if (!list) |
1057 | SetPageLRU(page_tail); | |
71e3aac0 | 1058 | |
12d27107 HD |
1059 | if (likely(PageLRU(page))) |
1060 | list_add_tail(&page_tail->lru, &page->lru); | |
5bc7b8ac SL |
1061 | else if (list) { |
1062 | /* page reclaim is reclaiming a huge page */ | |
1063 | get_page(page_tail); | |
1064 | list_add_tail(&page_tail->lru, list); | |
1065 | } else { | |
12d27107 HD |
1066 | struct list_head *list_head; |
1067 | /* | |
1068 | * Head page has not yet been counted, as an hpage, | |
1069 | * so we must account for each subpage individually. | |
1070 | * | |
1071 | * Use the standard add function to put page_tail on the list, | |
1072 | * but then correct its position so they all end up in order. | |
1073 | */ | |
e180cf80 | 1074 | add_page_to_lru_list(page_tail, lruvec, page_lru(page_tail)); |
12d27107 HD |
1075 | list_head = page_tail->lru.prev; |
1076 | list_move_tail(&page_tail->lru, list_head); | |
71e3aac0 | 1077 | } |
7512102c HD |
1078 | |
1079 | if (!PageUnevictable(page)) | |
e180cf80 | 1080 | update_page_reclaim_stat(lruvec, file, PageActive(page_tail)); |
71e3aac0 | 1081 | } |
12d27107 | 1082 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
71e3aac0 | 1083 | |
fa9add64 HD |
1084 | static void __pagevec_lru_add_fn(struct page *page, struct lruvec *lruvec, |
1085 | void *arg) | |
3dd7ae8e | 1086 | { |
13f7f789 MG |
1087 | int file = page_is_file_cache(page); |
1088 | int active = PageActive(page); | |
1089 | enum lru_list lru = page_lru(page); | |
3dd7ae8e | 1090 | |
309381fe | 1091 | VM_BUG_ON_PAGE(PageLRU(page), page); |
3dd7ae8e SL |
1092 | |
1093 | SetPageLRU(page); | |
fa9add64 HD |
1094 | add_page_to_lru_list(page, lruvec, lru); |
1095 | update_page_reclaim_stat(lruvec, file, active); | |
c6286c98 | 1096 | trace_mm_lru_insertion(page, page_to_pfn(page), lru, trace_pagemap_flags(page)); |
3dd7ae8e SL |
1097 | } |
1098 | ||
1da177e4 LT |
1099 | /* |
1100 | * Add the passed pages to the LRU, then drop the caller's refcount | |
1101 | * on them. Reinitialises the caller's pagevec. | |
1102 | */ | |
a0b8cab3 | 1103 | void __pagevec_lru_add(struct pagevec *pvec) |
1da177e4 | 1104 | { |
a0b8cab3 | 1105 | pagevec_lru_move_fn(pvec, __pagevec_lru_add_fn, NULL); |
1da177e4 | 1106 | } |
5095ae83 | 1107 | EXPORT_SYMBOL(__pagevec_lru_add); |
1da177e4 | 1108 | |
0cd6144a JW |
1109 | /** |
1110 | * pagevec_lookup_entries - gang pagecache lookup | |
1111 | * @pvec: Where the resulting entries are placed | |
1112 | * @mapping: The address_space to search | |
1113 | * @start: The starting entry index | |
1114 | * @nr_entries: The maximum number of entries | |
1115 | * @indices: The cache indices corresponding to the entries in @pvec | |
1116 | * | |
1117 | * pagevec_lookup_entries() will search for and return a group of up | |
1118 | * to @nr_entries pages and shadow entries in the mapping. All | |
1119 | * entries are placed in @pvec. pagevec_lookup_entries() takes a | |
1120 | * reference against actual pages in @pvec. | |
1121 | * | |
1122 | * The search returns a group of mapping-contiguous entries with | |
1123 | * ascending indexes. There may be holes in the indices due to | |
1124 | * not-present entries. | |
1125 | * | |
1126 | * pagevec_lookup_entries() returns the number of entries which were | |
1127 | * found. | |
1128 | */ | |
1129 | unsigned pagevec_lookup_entries(struct pagevec *pvec, | |
1130 | struct address_space *mapping, | |
1131 | pgoff_t start, unsigned nr_pages, | |
1132 | pgoff_t *indices) | |
1133 | { | |
1134 | pvec->nr = find_get_entries(mapping, start, nr_pages, | |
1135 | pvec->pages, indices); | |
1136 | return pagevec_count(pvec); | |
1137 | } | |
1138 | ||
1139 | /** | |
1140 | * pagevec_remove_exceptionals - pagevec exceptionals pruning | |
1141 | * @pvec: The pagevec to prune | |
1142 | * | |
1143 | * pagevec_lookup_entries() fills both pages and exceptional radix | |
1144 | * tree entries into the pagevec. This function prunes all | |
1145 | * exceptionals from @pvec without leaving holes, so that it can be | |
1146 | * passed on to page-only pagevec operations. | |
1147 | */ | |
1148 | void pagevec_remove_exceptionals(struct pagevec *pvec) | |
1149 | { | |
1150 | int i, j; | |
1151 | ||
1152 | for (i = 0, j = 0; i < pagevec_count(pvec); i++) { | |
1153 | struct page *page = pvec->pages[i]; | |
1154 | if (!radix_tree_exceptional_entry(page)) | |
1155 | pvec->pages[j++] = page; | |
1156 | } | |
1157 | pvec->nr = j; | |
1158 | } | |
1159 | ||
1da177e4 LT |
1160 | /** |
1161 | * pagevec_lookup - gang pagecache lookup | |
1162 | * @pvec: Where the resulting pages are placed | |
1163 | * @mapping: The address_space to search | |
1164 | * @start: The starting page index | |
1165 | * @nr_pages: The maximum number of pages | |
1166 | * | |
1167 | * pagevec_lookup() will search for and return a group of up to @nr_pages pages | |
1168 | * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a | |
1169 | * reference against the pages in @pvec. | |
1170 | * | |
1171 | * The search returns a group of mapping-contiguous pages with ascending | |
1172 | * indexes. There may be holes in the indices due to not-present pages. | |
1173 | * | |
1174 | * pagevec_lookup() returns the number of pages which were found. | |
1175 | */ | |
1176 | unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping, | |
1177 | pgoff_t start, unsigned nr_pages) | |
1178 | { | |
1179 | pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages); | |
1180 | return pagevec_count(pvec); | |
1181 | } | |
78539fdf CH |
1182 | EXPORT_SYMBOL(pagevec_lookup); |
1183 | ||
1da177e4 LT |
1184 | unsigned pagevec_lookup_tag(struct pagevec *pvec, struct address_space *mapping, |
1185 | pgoff_t *index, int tag, unsigned nr_pages) | |
1186 | { | |
1187 | pvec->nr = find_get_pages_tag(mapping, index, tag, | |
1188 | nr_pages, pvec->pages); | |
1189 | return pagevec_count(pvec); | |
1190 | } | |
7f285701 | 1191 | EXPORT_SYMBOL(pagevec_lookup_tag); |
1da177e4 | 1192 | |
1da177e4 LT |
1193 | /* |
1194 | * Perform any setup for the swap system | |
1195 | */ | |
1196 | void __init swap_setup(void) | |
1197 | { | |
4481374c | 1198 | unsigned long megs = totalram_pages >> (20 - PAGE_SHIFT); |
e0bf68dd | 1199 | #ifdef CONFIG_SWAP |
33806f06 SL |
1200 | int i; |
1201 | ||
8077c0d9 MP |
1202 | if (bdi_init(swapper_spaces[0].backing_dev_info)) |
1203 | panic("Failed to init swap bdi"); | |
33806f06 SL |
1204 | for (i = 0; i < MAX_SWAPFILES; i++) { |
1205 | spin_lock_init(&swapper_spaces[i].tree_lock); | |
1206 | INIT_LIST_HEAD(&swapper_spaces[i].i_mmap_nonlinear); | |
1207 | } | |
e0bf68dd PZ |
1208 | #endif |
1209 | ||
1da177e4 LT |
1210 | /* Use a smaller cluster for small-memory machines */ |
1211 | if (megs < 16) | |
1212 | page_cluster = 2; | |
1213 | else | |
1214 | page_cluster = 3; | |
1215 | /* | |
1216 | * Right now other parts of the system means that we | |
1217 | * _really_ don't want to cluster much more | |
1218 | */ | |
1da177e4 | 1219 | } |