]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * linux/mm/swap.c | |
3 | * | |
4 | * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds | |
5 | */ | |
6 | ||
7 | /* | |
8 | * This file contains the default values for the operation of the | |
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> | |
24 | #include <linux/module.h> | |
25 | #include <linux/mm_inline.h> | |
26 | #include <linux/buffer_head.h> /* for try_to_release_page() */ | |
27 | #include <linux/percpu_counter.h> | |
28 | #include <linux/percpu.h> | |
29 | #include <linux/cpu.h> | |
30 | #include <linux/notifier.h> | |
31 | #include <linux/backing-dev.h> | |
32 | #include <linux/memcontrol.h> | |
33 | #include <linux/gfp.h> | |
34 | ||
35 | #include "internal.h" | |
36 | ||
37 | /* How many pages do we try to swap or page in/out together? */ | |
38 | int page_cluster; | |
39 | ||
40 | static DEFINE_PER_CPU(struct pagevec[NR_LRU_LISTS], lru_add_pvecs); | |
41 | static DEFINE_PER_CPU(struct pagevec, lru_rotate_pvecs); | |
42 | ||
43 | /* | |
44 | * This path almost never happens for VM activity - pages are normally | |
45 | * freed via pagevecs. But it gets used by networking. | |
46 | */ | |
47 | static void __page_cache_release(struct page *page) | |
48 | { | |
49 | if (PageLRU(page)) { | |
50 | unsigned long flags; | |
51 | struct zone *zone = page_zone(page); | |
52 | ||
53 | spin_lock_irqsave(&zone->lru_lock, flags); | |
54 | VM_BUG_ON(!PageLRU(page)); | |
55 | __ClearPageLRU(page); | |
56 | del_page_from_lru(zone, page); | |
57 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
58 | } | |
59 | } | |
60 | ||
61 | static void __put_single_page(struct page *page) | |
62 | { | |
63 | __page_cache_release(page); | |
64 | free_hot_cold_page(page, 0); | |
65 | } | |
66 | ||
67 | static void __put_compound_page(struct page *page) | |
68 | { | |
69 | compound_page_dtor *dtor; | |
70 | ||
71 | __page_cache_release(page); | |
72 | dtor = get_compound_page_dtor(page); | |
73 | (*dtor)(page); | |
74 | } | |
75 | ||
76 | static void put_compound_page(struct page *page) | |
77 | { | |
78 | if (unlikely(PageTail(page))) { | |
79 | /* __split_huge_page_refcount can run under us */ | |
80 | struct page *page_head = page->first_page; | |
81 | smp_rmb(); | |
82 | /* | |
83 | * If PageTail is still set after smp_rmb() we can be sure | |
84 | * that the page->first_page we read wasn't a dangling pointer. | |
85 | * See __split_huge_page_refcount() smp_wmb(). | |
86 | */ | |
87 | if (likely(PageTail(page) && get_page_unless_zero(page_head))) { | |
88 | unsigned long flags; | |
89 | /* | |
90 | * Verify that our page_head wasn't converted | |
91 | * to a a regular page before we got a | |
92 | * reference on it. | |
93 | */ | |
94 | if (unlikely(!PageHead(page_head))) { | |
95 | /* PageHead is cleared after PageTail */ | |
96 | smp_rmb(); | |
97 | VM_BUG_ON(PageTail(page)); | |
98 | goto out_put_head; | |
99 | } | |
100 | /* | |
101 | * Only run compound_lock on a valid PageHead, | |
102 | * after having it pinned with | |
103 | * get_page_unless_zero() above. | |
104 | */ | |
105 | smp_mb(); | |
106 | /* page_head wasn't a dangling pointer */ | |
107 | flags = compound_lock_irqsave(page_head); | |
108 | if (unlikely(!PageTail(page))) { | |
109 | /* __split_huge_page_refcount run before us */ | |
110 | compound_unlock_irqrestore(page_head, flags); | |
111 | VM_BUG_ON(PageHead(page_head)); | |
112 | out_put_head: | |
113 | if (put_page_testzero(page_head)) | |
114 | __put_single_page(page_head); | |
115 | out_put_single: | |
116 | if (put_page_testzero(page)) | |
117 | __put_single_page(page); | |
118 | return; | |
119 | } | |
120 | VM_BUG_ON(page_head != page->first_page); | |
121 | /* | |
122 | * We can release the refcount taken by | |
123 | * get_page_unless_zero now that | |
124 | * split_huge_page_refcount is blocked on the | |
125 | * compound_lock. | |
126 | */ | |
127 | if (put_page_testzero(page_head)) | |
128 | VM_BUG_ON(1); | |
129 | /* __split_huge_page_refcount will wait now */ | |
130 | VM_BUG_ON(atomic_read(&page->_count) <= 0); | |
131 | atomic_dec(&page->_count); | |
132 | VM_BUG_ON(atomic_read(&page_head->_count) <= 0); | |
133 | compound_unlock_irqrestore(page_head, flags); | |
134 | if (put_page_testzero(page_head)) { | |
135 | if (PageHead(page_head)) | |
136 | __put_compound_page(page_head); | |
137 | else | |
138 | __put_single_page(page_head); | |
139 | } | |
140 | } else { | |
141 | /* page_head is a dangling pointer */ | |
142 | VM_BUG_ON(PageTail(page)); | |
143 | goto out_put_single; | |
144 | } | |
145 | } else if (put_page_testzero(page)) { | |
146 | if (PageHead(page)) | |
147 | __put_compound_page(page); | |
148 | else | |
149 | __put_single_page(page); | |
150 | } | |
151 | } | |
152 | ||
153 | void put_page(struct page *page) | |
154 | { | |
155 | if (unlikely(PageCompound(page))) | |
156 | put_compound_page(page); | |
157 | else if (put_page_testzero(page)) | |
158 | __put_single_page(page); | |
159 | } | |
160 | EXPORT_SYMBOL(put_page); | |
161 | ||
162 | /** | |
163 | * put_pages_list() - release a list of pages | |
164 | * @pages: list of pages threaded on page->lru | |
165 | * | |
166 | * Release a list of pages which are strung together on page.lru. Currently | |
167 | * used by read_cache_pages() and related error recovery code. | |
168 | */ | |
169 | void put_pages_list(struct list_head *pages) | |
170 | { | |
171 | while (!list_empty(pages)) { | |
172 | struct page *victim; | |
173 | ||
174 | victim = list_entry(pages->prev, struct page, lru); | |
175 | list_del(&victim->lru); | |
176 | page_cache_release(victim); | |
177 | } | |
178 | } | |
179 | EXPORT_SYMBOL(put_pages_list); | |
180 | ||
181 | static void pagevec_lru_move_fn(struct pagevec *pvec, | |
182 | void (*move_fn)(struct page *page, void *arg), | |
183 | void *arg) | |
184 | { | |
185 | int i; | |
186 | struct zone *zone = NULL; | |
187 | unsigned long flags = 0; | |
188 | ||
189 | for (i = 0; i < pagevec_count(pvec); i++) { | |
190 | struct page *page = pvec->pages[i]; | |
191 | struct zone *pagezone = page_zone(page); | |
192 | ||
193 | if (pagezone != zone) { | |
194 | if (zone) | |
195 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
196 | zone = pagezone; | |
197 | spin_lock_irqsave(&zone->lru_lock, flags); | |
198 | } | |
199 | ||
200 | (*move_fn)(page, arg); | |
201 | } | |
202 | if (zone) | |
203 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
204 | release_pages(pvec->pages, pagevec_count(pvec), pvec->cold); | |
205 | pagevec_reinit(pvec); | |
206 | } | |
207 | ||
208 | static void pagevec_move_tail_fn(struct page *page, void *arg) | |
209 | { | |
210 | int *pgmoved = arg; | |
211 | struct zone *zone = page_zone(page); | |
212 | ||
213 | if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) { | |
214 | int lru = page_lru_base_type(page); | |
215 | list_move_tail(&page->lru, &zone->lru[lru].list); | |
216 | (*pgmoved)++; | |
217 | } | |
218 | } | |
219 | ||
220 | /* | |
221 | * pagevec_move_tail() must be called with IRQ disabled. | |
222 | * Otherwise this may cause nasty races. | |
223 | */ | |
224 | static void pagevec_move_tail(struct pagevec *pvec) | |
225 | { | |
226 | int pgmoved = 0; | |
227 | ||
228 | pagevec_lru_move_fn(pvec, pagevec_move_tail_fn, &pgmoved); | |
229 | __count_vm_events(PGROTATED, pgmoved); | |
230 | } | |
231 | ||
232 | /* | |
233 | * Writeback is about to end against a page which has been marked for immediate | |
234 | * reclaim. If it still appears to be reclaimable, move it to the tail of the | |
235 | * inactive list. | |
236 | */ | |
237 | void rotate_reclaimable_page(struct page *page) | |
238 | { | |
239 | if (!PageLocked(page) && !PageDirty(page) && !PageActive(page) && | |
240 | !PageUnevictable(page) && PageLRU(page)) { | |
241 | struct pagevec *pvec; | |
242 | unsigned long flags; | |
243 | ||
244 | page_cache_get(page); | |
245 | local_irq_save(flags); | |
246 | pvec = &__get_cpu_var(lru_rotate_pvecs); | |
247 | if (!pagevec_add(pvec, page)) | |
248 | pagevec_move_tail(pvec); | |
249 | local_irq_restore(flags); | |
250 | } | |
251 | } | |
252 | ||
253 | static void update_page_reclaim_stat(struct zone *zone, struct page *page, | |
254 | int file, int rotated) | |
255 | { | |
256 | struct zone_reclaim_stat *reclaim_stat = &zone->reclaim_stat; | |
257 | struct zone_reclaim_stat *memcg_reclaim_stat; | |
258 | ||
259 | memcg_reclaim_stat = mem_cgroup_get_reclaim_stat_from_page(page); | |
260 | ||
261 | reclaim_stat->recent_scanned[file]++; | |
262 | if (rotated) | |
263 | reclaim_stat->recent_rotated[file]++; | |
264 | ||
265 | if (!memcg_reclaim_stat) | |
266 | return; | |
267 | ||
268 | memcg_reclaim_stat->recent_scanned[file]++; | |
269 | if (rotated) | |
270 | memcg_reclaim_stat->recent_rotated[file]++; | |
271 | } | |
272 | ||
273 | /* | |
274 | * A page will go to active list either by activate_page or putback_lru_page. | |
275 | * In the activate_page case, the page hasn't active bit set. The page might | |
276 | * not in LRU list because it's isolated before it gets a chance to be moved to | |
277 | * active list. The window is small because pagevec just stores several pages. | |
278 | * For such case, we do nothing for such page. | |
279 | * In the putback_lru_page case, the page isn't in lru list but has active | |
280 | * bit set | |
281 | */ | |
282 | static void __activate_page(struct page *page, void *arg) | |
283 | { | |
284 | struct zone *zone = page_zone(page); | |
285 | int file = page_is_file_cache(page); | |
286 | int lru = page_lru_base_type(page); | |
287 | bool putback = !PageLRU(page); | |
288 | ||
289 | /* The page is isolated before it's moved to active list */ | |
290 | if (!PageLRU(page) && !PageActive(page)) | |
291 | return; | |
292 | if ((PageLRU(page) && PageActive(page)) || PageUnevictable(page)) | |
293 | return; | |
294 | ||
295 | if (!putback) | |
296 | del_page_from_lru_list(zone, page, lru); | |
297 | else | |
298 | SetPageLRU(page); | |
299 | ||
300 | SetPageActive(page); | |
301 | lru += LRU_ACTIVE; | |
302 | add_page_to_lru_list(zone, page, lru); | |
303 | ||
304 | if (putback) | |
305 | return; | |
306 | __count_vm_event(PGACTIVATE); | |
307 | update_page_reclaim_stat(zone, page, file, 1); | |
308 | } | |
309 | ||
310 | #ifdef CONFIG_SMP | |
311 | static DEFINE_PER_CPU(struct pagevec, activate_page_pvecs); | |
312 | ||
313 | static void activate_page_drain(int cpu) | |
314 | { | |
315 | struct pagevec *pvec = &per_cpu(activate_page_pvecs, cpu); | |
316 | ||
317 | if (pagevec_count(pvec)) | |
318 | pagevec_lru_move_fn(pvec, __activate_page, NULL); | |
319 | } | |
320 | ||
321 | void activate_page(struct page *page) | |
322 | { | |
323 | if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) { | |
324 | struct pagevec *pvec = &get_cpu_var(activate_page_pvecs); | |
325 | ||
326 | page_cache_get(page); | |
327 | if (!pagevec_add(pvec, page)) | |
328 | pagevec_lru_move_fn(pvec, __activate_page, NULL); | |
329 | put_cpu_var(activate_page_pvecs); | |
330 | } | |
331 | } | |
332 | ||
333 | /* Caller should hold zone->lru_lock */ | |
334 | int putback_active_lru_page(struct zone *zone, struct page *page) | |
335 | { | |
336 | struct pagevec *pvec = &get_cpu_var(activate_page_pvecs); | |
337 | ||
338 | if (!pagevec_add(pvec, page)) { | |
339 | spin_unlock_irq(&zone->lru_lock); | |
340 | pagevec_lru_move_fn(pvec, __activate_page, NULL); | |
341 | spin_lock_irq(&zone->lru_lock); | |
342 | } | |
343 | put_cpu_var(activate_page_pvecs); | |
344 | return 1; | |
345 | } | |
346 | ||
347 | #else | |
348 | static inline void activate_page_drain(int cpu) | |
349 | { | |
350 | } | |
351 | ||
352 | void activate_page(struct page *page) | |
353 | { | |
354 | struct zone *zone = page_zone(page); | |
355 | ||
356 | spin_lock_irq(&zone->lru_lock); | |
357 | if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) | |
358 | __activate_page(page, NULL); | |
359 | spin_unlock_irq(&zone->lru_lock); | |
360 | } | |
361 | #endif | |
362 | ||
363 | /* | |
364 | * Mark a page as having seen activity. | |
365 | * | |
366 | * inactive,unreferenced -> inactive,referenced | |
367 | * inactive,referenced -> active,unreferenced | |
368 | * active,unreferenced -> active,referenced | |
369 | */ | |
370 | void mark_page_accessed(struct page *page) | |
371 | { | |
372 | if (!PageActive(page) && !PageUnevictable(page) && | |
373 | PageReferenced(page) && PageLRU(page)) { | |
374 | activate_page(page); | |
375 | ClearPageReferenced(page); | |
376 | } else if (!PageReferenced(page)) { | |
377 | SetPageReferenced(page); | |
378 | } | |
379 | } | |
380 | ||
381 | EXPORT_SYMBOL(mark_page_accessed); | |
382 | ||
383 | void __lru_cache_add(struct page *page, enum lru_list lru) | |
384 | { | |
385 | struct pagevec *pvec = &get_cpu_var(lru_add_pvecs)[lru]; | |
386 | ||
387 | page_cache_get(page); | |
388 | if (!pagevec_add(pvec, page)) | |
389 | ____pagevec_lru_add(pvec, lru); | |
390 | put_cpu_var(lru_add_pvecs); | |
391 | } | |
392 | EXPORT_SYMBOL(__lru_cache_add); | |
393 | ||
394 | /** | |
395 | * lru_cache_add_lru - add a page to a page list | |
396 | * @page: the page to be added to the LRU. | |
397 | * @lru: the LRU list to which the page is added. | |
398 | */ | |
399 | void lru_cache_add_lru(struct page *page, enum lru_list lru) | |
400 | { | |
401 | if (PageActive(page)) { | |
402 | VM_BUG_ON(PageUnevictable(page)); | |
403 | ClearPageActive(page); | |
404 | } else if (PageUnevictable(page)) { | |
405 | VM_BUG_ON(PageActive(page)); | |
406 | ClearPageUnevictable(page); | |
407 | } | |
408 | ||
409 | VM_BUG_ON(PageLRU(page) || PageActive(page) || PageUnevictable(page)); | |
410 | __lru_cache_add(page, lru); | |
411 | } | |
412 | ||
413 | /** | |
414 | * add_page_to_unevictable_list - add a page to the unevictable list | |
415 | * @page: the page to be added to the unevictable list | |
416 | * | |
417 | * Add page directly to its zone's unevictable list. To avoid races with | |
418 | * tasks that might be making the page evictable, through eg. munlock, | |
419 | * munmap or exit, while it's not on the lru, we want to add the page | |
420 | * while it's locked or otherwise "invisible" to other tasks. This is | |
421 | * difficult to do when using the pagevec cache, so bypass that. | |
422 | */ | |
423 | void add_page_to_unevictable_list(struct page *page) | |
424 | { | |
425 | struct zone *zone = page_zone(page); | |
426 | ||
427 | spin_lock_irq(&zone->lru_lock); | |
428 | SetPageUnevictable(page); | |
429 | SetPageLRU(page); | |
430 | add_page_to_lru_list(zone, page, LRU_UNEVICTABLE); | |
431 | spin_unlock_irq(&zone->lru_lock); | |
432 | } | |
433 | ||
434 | /* | |
435 | * Drain pages out of the cpu's pagevecs. | |
436 | * Either "cpu" is the current CPU, and preemption has already been | |
437 | * disabled; or "cpu" is being hot-unplugged, and is already dead. | |
438 | */ | |
439 | static void drain_cpu_pagevecs(int cpu) | |
440 | { | |
441 | struct pagevec *pvecs = per_cpu(lru_add_pvecs, cpu); | |
442 | struct pagevec *pvec; | |
443 | int lru; | |
444 | ||
445 | for_each_lru(lru) { | |
446 | pvec = &pvecs[lru - LRU_BASE]; | |
447 | if (pagevec_count(pvec)) | |
448 | ____pagevec_lru_add(pvec, lru); | |
449 | } | |
450 | ||
451 | pvec = &per_cpu(lru_rotate_pvecs, cpu); | |
452 | if (pagevec_count(pvec)) { | |
453 | unsigned long flags; | |
454 | ||
455 | /* No harm done if a racing interrupt already did this */ | |
456 | local_irq_save(flags); | |
457 | pagevec_move_tail(pvec); | |
458 | local_irq_restore(flags); | |
459 | } | |
460 | activate_page_drain(cpu); | |
461 | } | |
462 | ||
463 | void lru_add_drain(void) | |
464 | { | |
465 | drain_cpu_pagevecs(get_cpu()); | |
466 | put_cpu(); | |
467 | } | |
468 | ||
469 | static void lru_add_drain_per_cpu(struct work_struct *dummy) | |
470 | { | |
471 | lru_add_drain(); | |
472 | } | |
473 | ||
474 | /* | |
475 | * Returns 0 for success | |
476 | */ | |
477 | int lru_add_drain_all(void) | |
478 | { | |
479 | return schedule_on_each_cpu(lru_add_drain_per_cpu); | |
480 | } | |
481 | ||
482 | /* | |
483 | * Batched page_cache_release(). Decrement the reference count on all the | |
484 | * passed pages. If it fell to zero then remove the page from the LRU and | |
485 | * free it. | |
486 | * | |
487 | * Avoid taking zone->lru_lock if possible, but if it is taken, retain it | |
488 | * for the remainder of the operation. | |
489 | * | |
490 | * The locking in this function is against shrink_inactive_list(): we recheck | |
491 | * the page count inside the lock to see whether shrink_inactive_list() | |
492 | * grabbed the page via the LRU. If it did, give up: shrink_inactive_list() | |
493 | * will free it. | |
494 | */ | |
495 | void release_pages(struct page **pages, int nr, int cold) | |
496 | { | |
497 | int i; | |
498 | struct pagevec pages_to_free; | |
499 | struct zone *zone = NULL; | |
500 | unsigned long uninitialized_var(flags); | |
501 | ||
502 | pagevec_init(&pages_to_free, cold); | |
503 | for (i = 0; i < nr; i++) { | |
504 | struct page *page = pages[i]; | |
505 | ||
506 | if (unlikely(PageCompound(page))) { | |
507 | if (zone) { | |
508 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
509 | zone = NULL; | |
510 | } | |
511 | put_compound_page(page); | |
512 | continue; | |
513 | } | |
514 | ||
515 | if (!put_page_testzero(page)) | |
516 | continue; | |
517 | ||
518 | if (PageLRU(page)) { | |
519 | struct zone *pagezone = page_zone(page); | |
520 | ||
521 | if (pagezone != zone) { | |
522 | if (zone) | |
523 | spin_unlock_irqrestore(&zone->lru_lock, | |
524 | flags); | |
525 | zone = pagezone; | |
526 | spin_lock_irqsave(&zone->lru_lock, flags); | |
527 | } | |
528 | VM_BUG_ON(!PageLRU(page)); | |
529 | __ClearPageLRU(page); | |
530 | del_page_from_lru(zone, page); | |
531 | } | |
532 | ||
533 | if (!pagevec_add(&pages_to_free, page)) { | |
534 | if (zone) { | |
535 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
536 | zone = NULL; | |
537 | } | |
538 | __pagevec_free(&pages_to_free); | |
539 | pagevec_reinit(&pages_to_free); | |
540 | } | |
541 | } | |
542 | if (zone) | |
543 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
544 | ||
545 | pagevec_free(&pages_to_free); | |
546 | } | |
547 | EXPORT_SYMBOL(release_pages); | |
548 | ||
549 | /* | |
550 | * The pages which we're about to release may be in the deferred lru-addition | |
551 | * queues. That would prevent them from really being freed right now. That's | |
552 | * OK from a correctness point of view but is inefficient - those pages may be | |
553 | * cache-warm and we want to give them back to the page allocator ASAP. | |
554 | * | |
555 | * So __pagevec_release() will drain those queues here. __pagevec_lru_add() | |
556 | * and __pagevec_lru_add_active() call release_pages() directly to avoid | |
557 | * mutual recursion. | |
558 | */ | |
559 | void __pagevec_release(struct pagevec *pvec) | |
560 | { | |
561 | lru_add_drain(); | |
562 | release_pages(pvec->pages, pagevec_count(pvec), pvec->cold); | |
563 | pagevec_reinit(pvec); | |
564 | } | |
565 | ||
566 | EXPORT_SYMBOL(__pagevec_release); | |
567 | ||
568 | /* used by __split_huge_page_refcount() */ | |
569 | void lru_add_page_tail(struct zone* zone, | |
570 | struct page *page, struct page *page_tail) | |
571 | { | |
572 | int active; | |
573 | enum lru_list lru; | |
574 | const int file = 0; | |
575 | struct list_head *head; | |
576 | ||
577 | VM_BUG_ON(!PageHead(page)); | |
578 | VM_BUG_ON(PageCompound(page_tail)); | |
579 | VM_BUG_ON(PageLRU(page_tail)); | |
580 | VM_BUG_ON(!spin_is_locked(&zone->lru_lock)); | |
581 | ||
582 | SetPageLRU(page_tail); | |
583 | ||
584 | if (page_evictable(page_tail, NULL)) { | |
585 | if (PageActive(page)) { | |
586 | SetPageActive(page_tail); | |
587 | active = 1; | |
588 | lru = LRU_ACTIVE_ANON; | |
589 | } else { | |
590 | active = 0; | |
591 | lru = LRU_INACTIVE_ANON; | |
592 | } | |
593 | update_page_reclaim_stat(zone, page_tail, file, active); | |
594 | if (likely(PageLRU(page))) | |
595 | head = page->lru.prev; | |
596 | else | |
597 | head = &zone->lru[lru].list; | |
598 | __add_page_to_lru_list(zone, page_tail, lru, head); | |
599 | } else { | |
600 | SetPageUnevictable(page_tail); | |
601 | add_page_to_lru_list(zone, page_tail, LRU_UNEVICTABLE); | |
602 | } | |
603 | } | |
604 | ||
605 | static void ____pagevec_lru_add_fn(struct page *page, void *arg) | |
606 | { | |
607 | enum lru_list lru = (enum lru_list)arg; | |
608 | struct zone *zone = page_zone(page); | |
609 | int file = is_file_lru(lru); | |
610 | int active = is_active_lru(lru); | |
611 | ||
612 | VM_BUG_ON(PageActive(page)); | |
613 | VM_BUG_ON(PageUnevictable(page)); | |
614 | VM_BUG_ON(PageLRU(page)); | |
615 | ||
616 | SetPageLRU(page); | |
617 | if (active) | |
618 | SetPageActive(page); | |
619 | update_page_reclaim_stat(zone, page, file, active); | |
620 | add_page_to_lru_list(zone, page, lru); | |
621 | } | |
622 | ||
623 | /* | |
624 | * Add the passed pages to the LRU, then drop the caller's refcount | |
625 | * on them. Reinitialises the caller's pagevec. | |
626 | */ | |
627 | void ____pagevec_lru_add(struct pagevec *pvec, enum lru_list lru) | |
628 | { | |
629 | VM_BUG_ON(is_unevictable_lru(lru)); | |
630 | ||
631 | pagevec_lru_move_fn(pvec, ____pagevec_lru_add_fn, (void *)lru); | |
632 | } | |
633 | ||
634 | EXPORT_SYMBOL(____pagevec_lru_add); | |
635 | ||
636 | /* | |
637 | * Try to drop buffers from the pages in a pagevec | |
638 | */ | |
639 | void pagevec_strip(struct pagevec *pvec) | |
640 | { | |
641 | int i; | |
642 | ||
643 | for (i = 0; i < pagevec_count(pvec); i++) { | |
644 | struct page *page = pvec->pages[i]; | |
645 | ||
646 | if (page_has_private(page) && trylock_page(page)) { | |
647 | if (page_has_private(page)) | |
648 | try_to_release_page(page, 0); | |
649 | unlock_page(page); | |
650 | } | |
651 | } | |
652 | } | |
653 | ||
654 | /** | |
655 | * pagevec_lookup - gang pagecache lookup | |
656 | * @pvec: Where the resulting pages are placed | |
657 | * @mapping: The address_space to search | |
658 | * @start: The starting page index | |
659 | * @nr_pages: The maximum number of pages | |
660 | * | |
661 | * pagevec_lookup() will search for and return a group of up to @nr_pages pages | |
662 | * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a | |
663 | * reference against the pages in @pvec. | |
664 | * | |
665 | * The search returns a group of mapping-contiguous pages with ascending | |
666 | * indexes. There may be holes in the indices due to not-present pages. | |
667 | * | |
668 | * pagevec_lookup() returns the number of pages which were found. | |
669 | */ | |
670 | unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping, | |
671 | pgoff_t start, unsigned nr_pages) | |
672 | { | |
673 | pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages); | |
674 | return pagevec_count(pvec); | |
675 | } | |
676 | ||
677 | EXPORT_SYMBOL(pagevec_lookup); | |
678 | ||
679 | unsigned pagevec_lookup_tag(struct pagevec *pvec, struct address_space *mapping, | |
680 | pgoff_t *index, int tag, unsigned nr_pages) | |
681 | { | |
682 | pvec->nr = find_get_pages_tag(mapping, index, tag, | |
683 | nr_pages, pvec->pages); | |
684 | return pagevec_count(pvec); | |
685 | } | |
686 | ||
687 | EXPORT_SYMBOL(pagevec_lookup_tag); | |
688 | ||
689 | /* | |
690 | * Perform any setup for the swap system | |
691 | */ | |
692 | void __init swap_setup(void) | |
693 | { | |
694 | unsigned long megs = totalram_pages >> (20 - PAGE_SHIFT); | |
695 | ||
696 | #ifdef CONFIG_SWAP | |
697 | bdi_init(swapper_space.backing_dev_info); | |
698 | #endif | |
699 | ||
700 | /* Use a smaller cluster for small-memory machines */ | |
701 | if (megs < 16) | |
702 | page_cluster = 2; | |
703 | else | |
704 | page_cluster = 3; | |
705 | /* | |
706 | * Right now other parts of the system means that we | |
707 | * _really_ don't want to cluster much more | |
708 | */ | |
709 | } |