1
2
3
4
5
6
7
8
9
10
11
12
13
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>
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
34#include "internal.h"
35
36
37int page_cluster;
38
39static DEFINE_PER_CPU(struct pagevec[NR_LRU_LISTS], lru_add_pvecs);
40static DEFINE_PER_CPU(struct pagevec, lru_rotate_pvecs);
41
42
43
44
45
46static void __page_cache_release(struct page *page)
47{
48 if (PageLRU(page)) {
49 unsigned long flags;
50 struct zone *zone = page_zone(page);
51
52 spin_lock_irqsave(&zone->lru_lock, flags);
53 VM_BUG_ON(!PageLRU(page));
54 __ClearPageLRU(page);
55 del_page_from_lru(zone, page);
56 spin_unlock_irqrestore(&zone->lru_lock, flags);
57 }
58 free_hot_page(page);
59}
60
61static void put_compound_page(struct page *page)
62{
63 page = compound_head(page);
64 if (put_page_testzero(page)) {
65 compound_page_dtor *dtor;
66
67 dtor = get_compound_page_dtor(page);
68 (*dtor)(page);
69 }
70}
71
72void put_page(struct page *page)
73{
74 if (unlikely(PageCompound(page)))
75 put_compound_page(page);
76 else if (put_page_testzero(page))
77 __page_cache_release(page);
78}
79EXPORT_SYMBOL(put_page);
80
81
82
83
84
85
86
87
88void put_pages_list(struct list_head *pages)
89{
90 while (!list_empty(pages)) {
91 struct page *victim;
92
93 victim = list_entry(pages->prev, struct page, lru);
94 list_del(&victim->lru);
95 page_cache_release(victim);
96 }
97}
98EXPORT_SYMBOL(put_pages_list);
99
100
101
102
103
104static void pagevec_move_tail(struct pagevec *pvec)
105{
106 int i;
107 int pgmoved = 0;
108 struct zone *zone = NULL;
109
110 for (i = 0; i < pagevec_count(pvec); i++) {
111 struct page *page = pvec->pages[i];
112 struct zone *pagezone = page_zone(page);
113
114 if (pagezone != zone) {
115 if (zone)
116 spin_unlock(&zone->lru_lock);
117 zone = pagezone;
118 spin_lock(&zone->lru_lock);
119 }
120 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
121 int lru = page_is_file_cache(page);
122 list_move_tail(&page->lru, &zone->lru[lru].list);
123 pgmoved++;
124 }
125 }
126 if (zone)
127 spin_unlock(&zone->lru_lock);
128 __count_vm_events(PGROTATED, pgmoved);
129 release_pages(pvec->pages, pvec->nr, pvec->cold);
130 pagevec_reinit(pvec);
131}
132
133
134
135
136
137
138void rotate_reclaimable_page(struct page *page)
139{
140 if (!PageLocked(page) && !PageDirty(page) && !PageActive(page) &&
141 !PageUnevictable(page) && PageLRU(page)) {
142 struct pagevec *pvec;
143 unsigned long flags;
144
145 page_cache_get(page);
146 local_irq_save(flags);
147 pvec = &__get_cpu_var(lru_rotate_pvecs);
148 if (!pagevec_add(pvec, page))
149 pagevec_move_tail(pvec);
150 local_irq_restore(flags);
151 }
152}
153
154
155
156
157void activate_page(struct page *page)
158{
159 struct zone *zone = page_zone(page);
160
161 spin_lock_irq(&zone->lru_lock);
162 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
163 int file = page_is_file_cache(page);
164 int lru = LRU_BASE + file;
165 del_page_from_lru_list(zone, page, lru);
166
167 SetPageActive(page);
168 lru += LRU_ACTIVE;
169 add_page_to_lru_list(zone, page, lru);
170 __count_vm_event(PGACTIVATE);
171 mem_cgroup_move_lists(page, lru);
172
173 zone->recent_rotated[!!file]++;
174 zone->recent_scanned[!!file]++;
175 }
176 spin_unlock_irq(&zone->lru_lock);
177}
178
179
180
181
182
183
184
185
186void mark_page_accessed(struct page *page)
187{
188 if (!PageActive(page) && !PageUnevictable(page) &&
189 PageReferenced(page) && PageLRU(page)) {
190 activate_page(page);
191 ClearPageReferenced(page);
192 } else if (!PageReferenced(page)) {
193 SetPageReferenced(page);
194 }
195}
196
197EXPORT_SYMBOL(mark_page_accessed);
198
199void __lru_cache_add(struct page *page, enum lru_list lru)
200{
201 struct pagevec *pvec = &get_cpu_var(lru_add_pvecs)[lru];
202
203 page_cache_get(page);
204 if (!pagevec_add(pvec, page))
205 ____pagevec_lru_add(pvec, lru);
206 put_cpu_var(lru_add_pvecs);
207}
208
209
210
211
212
213
214void lru_cache_add_lru(struct page *page, enum lru_list lru)
215{
216 if (PageActive(page)) {
217 VM_BUG_ON(PageUnevictable(page));
218 ClearPageActive(page);
219 } else if (PageUnevictable(page)) {
220 VM_BUG_ON(PageActive(page));
221 ClearPageUnevictable(page);
222 }
223
224 VM_BUG_ON(PageLRU(page) || PageActive(page) || PageUnevictable(page));
225 __lru_cache_add(page, lru);
226}
227
228
229
230
231
232
233
234
235
236
237
238void add_page_to_unevictable_list(struct page *page)
239{
240 struct zone *zone = page_zone(page);
241
242 spin_lock_irq(&zone->lru_lock);
243 SetPageUnevictable(page);
244 SetPageLRU(page);
245 add_page_to_lru_list(zone, page, LRU_UNEVICTABLE);
246 spin_unlock_irq(&zone->lru_lock);
247}
248
249
250
251
252
253
254
255
256
257
258
259void lru_cache_add_active_or_unevictable(struct page *page,
260 struct vm_area_struct *vma)
261{
262 if (page_evictable(page, vma))
263 lru_cache_add_lru(page, LRU_ACTIVE + page_is_file_cache(page));
264 else
265 add_page_to_unevictable_list(page);
266}
267
268
269
270
271
272
273static void drain_cpu_pagevecs(int cpu)
274{
275 struct pagevec *pvecs = per_cpu(lru_add_pvecs, cpu);
276 struct pagevec *pvec;
277 int lru;
278
279 for_each_lru(lru) {
280 pvec = &pvecs[lru - LRU_BASE];
281 if (pagevec_count(pvec))
282 ____pagevec_lru_add(pvec, lru);
283 }
284
285 pvec = &per_cpu(lru_rotate_pvecs, cpu);
286 if (pagevec_count(pvec)) {
287 unsigned long flags;
288
289
290 local_irq_save(flags);
291 pagevec_move_tail(pvec);
292 local_irq_restore(flags);
293 }
294}
295
296void lru_add_drain(void)
297{
298 drain_cpu_pagevecs(get_cpu());
299 put_cpu();
300}
301
302static void lru_add_drain_per_cpu(struct work_struct *dummy)
303{
304 lru_add_drain();
305}
306
307
308
309
310int lru_add_drain_all(void)
311{
312 return schedule_on_each_cpu(lru_add_drain_per_cpu);
313}
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328void release_pages(struct page **pages, int nr, int cold)
329{
330 int i;
331 struct pagevec pages_to_free;
332 struct zone *zone = NULL;
333 unsigned long uninitialized_var(flags);
334
335 pagevec_init(&pages_to_free, cold);
336 for (i = 0; i < nr; i++) {
337 struct page *page = pages[i];
338
339 if (unlikely(PageCompound(page))) {
340 if (zone) {
341 spin_unlock_irqrestore(&zone->lru_lock, flags);
342 zone = NULL;
343 }
344 put_compound_page(page);
345 continue;
346 }
347
348 if (!put_page_testzero(page))
349 continue;
350
351 if (PageLRU(page)) {
352 struct zone *pagezone = page_zone(page);
353
354 if (pagezone != zone) {
355 if (zone)
356 spin_unlock_irqrestore(&zone->lru_lock,
357 flags);
358 zone = pagezone;
359 spin_lock_irqsave(&zone->lru_lock, flags);
360 }
361 VM_BUG_ON(!PageLRU(page));
362 __ClearPageLRU(page);
363 del_page_from_lru(zone, page);
364 }
365
366 if (!pagevec_add(&pages_to_free, page)) {
367 if (zone) {
368 spin_unlock_irqrestore(&zone->lru_lock, flags);
369 zone = NULL;
370 }
371 __pagevec_free(&pages_to_free);
372 pagevec_reinit(&pages_to_free);
373 }
374 }
375 if (zone)
376 spin_unlock_irqrestore(&zone->lru_lock, flags);
377
378 pagevec_free(&pages_to_free);
379}
380
381
382
383
384
385
386
387
388
389
390
391void __pagevec_release(struct pagevec *pvec)
392{
393 lru_add_drain();
394 release_pages(pvec->pages, pagevec_count(pvec), pvec->cold);
395 pagevec_reinit(pvec);
396}
397
398EXPORT_SYMBOL(__pagevec_release);
399
400
401
402
403
404
405void __pagevec_release_nonlru(struct pagevec *pvec)
406{
407 int i;
408 struct pagevec pages_to_free;
409
410 pagevec_init(&pages_to_free, pvec->cold);
411 for (i = 0; i < pagevec_count(pvec); i++) {
412 struct page *page = pvec->pages[i];
413
414 VM_BUG_ON(PageLRU(page));
415 if (put_page_testzero(page))
416 pagevec_add(&pages_to_free, page);
417 }
418 pagevec_free(&pages_to_free);
419 pagevec_reinit(pvec);
420}
421
422
423
424
425
426void ____pagevec_lru_add(struct pagevec *pvec, enum lru_list lru)
427{
428 int i;
429 struct zone *zone = NULL;
430 VM_BUG_ON(is_unevictable_lru(lru));
431
432 for (i = 0; i < pagevec_count(pvec); i++) {
433 struct page *page = pvec->pages[i];
434 struct zone *pagezone = page_zone(page);
435 int file;
436
437 if (pagezone != zone) {
438 if (zone)
439 spin_unlock_irq(&zone->lru_lock);
440 zone = pagezone;
441 spin_lock_irq(&zone->lru_lock);
442 }
443 VM_BUG_ON(PageActive(page));
444 VM_BUG_ON(PageUnevictable(page));
445 VM_BUG_ON(PageLRU(page));
446 SetPageLRU(page);
447 file = is_file_lru(lru);
448 zone->recent_scanned[file]++;
449 if (is_active_lru(lru)) {
450 SetPageActive(page);
451 zone->recent_rotated[file]++;
452 }
453 add_page_to_lru_list(zone, page, lru);
454 }
455 if (zone)
456 spin_unlock_irq(&zone->lru_lock);
457 release_pages(pvec->pages, pvec->nr, pvec->cold);
458 pagevec_reinit(pvec);
459}
460
461EXPORT_SYMBOL(____pagevec_lru_add);
462
463
464
465
466void pagevec_strip(struct pagevec *pvec)
467{
468 int i;
469
470 for (i = 0; i < pagevec_count(pvec); i++) {
471 struct page *page = pvec->pages[i];
472
473 if (PagePrivate(page) && trylock_page(page)) {
474 if (PagePrivate(page))
475 try_to_release_page(page, 0);
476 unlock_page(page);
477 }
478 }
479}
480
481
482
483
484
485
486
487
488
489
490void pagevec_swap_free(struct pagevec *pvec)
491{
492 int i;
493
494 for (i = 0; i < pagevec_count(pvec); i++) {
495 struct page *page = pvec->pages[i];
496
497 if (PageSwapCache(page) && trylock_page(page)) {
498 if (PageSwapCache(page))
499 remove_exclusive_swap_page_ref(page);
500 unlock_page(page);
501 }
502 }
503}
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping,
522 pgoff_t start, unsigned nr_pages)
523{
524 pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages);
525 return pagevec_count(pvec);
526}
527
528EXPORT_SYMBOL(pagevec_lookup);
529
530unsigned pagevec_lookup_tag(struct pagevec *pvec, struct address_space *mapping,
531 pgoff_t *index, int tag, unsigned nr_pages)
532{
533 pvec->nr = find_get_pages_tag(mapping, index, tag,
534 nr_pages, pvec->pages);
535 return pagevec_count(pvec);
536}
537
538EXPORT_SYMBOL(pagevec_lookup_tag);
539
540#ifdef CONFIG_SMP
541
542
543
544
545#define ACCT_THRESHOLD max(16, NR_CPUS * 2)
546
547static DEFINE_PER_CPU(long, committed_space);
548
549void vm_acct_memory(long pages)
550{
551 long *local;
552
553 preempt_disable();
554 local = &__get_cpu_var(committed_space);
555 *local += pages;
556 if (*local > ACCT_THRESHOLD || *local < -ACCT_THRESHOLD) {
557 atomic_long_add(*local, &vm_committed_space);
558 *local = 0;
559 }
560 preempt_enable();
561}
562
563#ifdef CONFIG_HOTPLUG_CPU
564
565
566static int cpu_swap_callback(struct notifier_block *nfb,
567 unsigned long action,
568 void *hcpu)
569{
570 long *committed;
571
572 committed = &per_cpu(committed_space, (long)hcpu);
573 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
574 atomic_long_add(*committed, &vm_committed_space);
575 *committed = 0;
576 drain_cpu_pagevecs((long)hcpu);
577 }
578 return NOTIFY_OK;
579}
580#endif
581#endif
582
583
584
585
586void __init swap_setup(void)
587{
588 unsigned long megs = num_physpages >> (20 - PAGE_SHIFT);
589
590#ifdef CONFIG_SWAP
591 bdi_init(swapper_space.backing_dev_info);
592#endif
593
594
595 if (megs < 16)
596 page_cluster = 2;
597 else
598 page_cluster = 3;
599
600
601
602
603#ifdef CONFIG_HOTPLUG_CPU
604 hotcpu_notifier(cpu_swap_callback, 0);
605#endif
606}