Root/mm/swap.c

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? */
38int page_cluster;
39
40static DEFINE_PER_CPU(struct pagevec[NR_LRU_LISTS], lru_add_pvecs);
41static 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 */
47static 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    free_hot_cold_page(page, 0);
60}
61
62static void put_compound_page(struct page *page)
63{
64    page = compound_head(page);
65    if (put_page_testzero(page)) {
66        compound_page_dtor *dtor;
67
68        dtor = get_compound_page_dtor(page);
69        (*dtor)(page);
70    }
71}
72
73void put_page(struct page *page)
74{
75    if (unlikely(PageCompound(page)))
76        put_compound_page(page);
77    else if (put_page_testzero(page))
78        __page_cache_release(page);
79}
80EXPORT_SYMBOL(put_page);
81
82/**
83 * put_pages_list() - release a list of pages
84 * @pages: list of pages threaded on page->lru
85 *
86 * Release a list of pages which are strung together on page.lru. Currently
87 * used by read_cache_pages() and related error recovery code.
88 */
89void put_pages_list(struct list_head *pages)
90{
91    while (!list_empty(pages)) {
92        struct page *victim;
93
94        victim = list_entry(pages->prev, struct page, lru);
95        list_del(&victim->lru);
96        page_cache_release(victim);
97    }
98}
99EXPORT_SYMBOL(put_pages_list);
100
101/*
102 * pagevec_move_tail() must be called with IRQ disabled.
103 * Otherwise this may cause nasty races.
104 */
105static void pagevec_move_tail(struct pagevec *pvec)
106{
107    int i;
108    int pgmoved = 0;
109    struct zone *zone = NULL;
110
111    for (i = 0; i < pagevec_count(pvec); i++) {
112        struct page *page = pvec->pages[i];
113        struct zone *pagezone = page_zone(page);
114
115        if (pagezone != zone) {
116            if (zone)
117                spin_unlock(&zone->lru_lock);
118            zone = pagezone;
119            spin_lock(&zone->lru_lock);
120        }
121        if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
122            int lru = page_lru_base_type(page);
123            list_move_tail(&page->lru, &zone->lru[lru].list);
124            pgmoved++;
125        }
126    }
127    if (zone)
128        spin_unlock(&zone->lru_lock);
129    __count_vm_events(PGROTATED, pgmoved);
130    release_pages(pvec->pages, pvec->nr, pvec->cold);
131    pagevec_reinit(pvec);
132}
133
134/*
135 * Writeback is about to end against a page which has been marked for immediate
136 * reclaim. If it still appears to be reclaimable, move it to the tail of the
137 * inactive list.
138 */
139void rotate_reclaimable_page(struct page *page)
140{
141    if (!PageLocked(page) && !PageDirty(page) && !PageActive(page) &&
142        !PageUnevictable(page) && PageLRU(page)) {
143        struct pagevec *pvec;
144        unsigned long flags;
145
146        page_cache_get(page);
147        local_irq_save(flags);
148        pvec = &__get_cpu_var(lru_rotate_pvecs);
149        if (!pagevec_add(pvec, page))
150            pagevec_move_tail(pvec);
151        local_irq_restore(flags);
152    }
153}
154
155static void update_page_reclaim_stat(struct zone *zone, struct page *page,
156                     int file, int rotated)
157{
158    struct zone_reclaim_stat *reclaim_stat = &zone->reclaim_stat;
159    struct zone_reclaim_stat *memcg_reclaim_stat;
160
161    memcg_reclaim_stat = mem_cgroup_get_reclaim_stat_from_page(page);
162
163    reclaim_stat->recent_scanned[file]++;
164    if (rotated)
165        reclaim_stat->recent_rotated[file]++;
166
167    if (!memcg_reclaim_stat)
168        return;
169
170    memcg_reclaim_stat->recent_scanned[file]++;
171    if (rotated)
172        memcg_reclaim_stat->recent_rotated[file]++;
173}
174
175/*
176 * FIXME: speed this up?
177 */
178void activate_page(struct page *page)
179{
180    struct zone *zone = page_zone(page);
181
182    spin_lock_irq(&zone->lru_lock);
183    if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
184        int file = page_is_file_cache(page);
185        int lru = page_lru_base_type(page);
186        del_page_from_lru_list(zone, page, lru);
187
188        SetPageActive(page);
189        lru += LRU_ACTIVE;
190        add_page_to_lru_list(zone, page, lru);
191        __count_vm_event(PGACTIVATE);
192
193        update_page_reclaim_stat(zone, page, file, 1);
194    }
195    spin_unlock_irq(&zone->lru_lock);
196}
197
198/*
199 * Mark a page as having seen activity.
200 *
201 * inactive,unreferenced -> inactive,referenced
202 * inactive,referenced -> active,unreferenced
203 * active,unreferenced -> active,referenced
204 */
205void mark_page_accessed(struct page *page)
206{
207    if (!PageActive(page) && !PageUnevictable(page) &&
208            PageReferenced(page) && PageLRU(page)) {
209        activate_page(page);
210        ClearPageReferenced(page);
211    } else if (!PageReferenced(page)) {
212        SetPageReferenced(page);
213    }
214}
215
216EXPORT_SYMBOL(mark_page_accessed);
217
218void __lru_cache_add(struct page *page, enum lru_list lru)
219{
220    struct pagevec *pvec = &get_cpu_var(lru_add_pvecs)[lru];
221
222    page_cache_get(page);
223    if (!pagevec_add(pvec, page))
224        ____pagevec_lru_add(pvec, lru);
225    put_cpu_var(lru_add_pvecs);
226}
227EXPORT_SYMBOL(__lru_cache_add);
228
229/**
230 * lru_cache_add_lru - add a page to a page list
231 * @page: the page to be added to the LRU.
232 * @lru: the LRU list to which the page is added.
233 */
234void lru_cache_add_lru(struct page *page, enum lru_list lru)
235{
236    if (PageActive(page)) {
237        VM_BUG_ON(PageUnevictable(page));
238        ClearPageActive(page);
239    } else if (PageUnevictable(page)) {
240        VM_BUG_ON(PageActive(page));
241        ClearPageUnevictable(page);
242    }
243
244    VM_BUG_ON(PageLRU(page) || PageActive(page) || PageUnevictable(page));
245    __lru_cache_add(page, lru);
246}
247
248/**
249 * add_page_to_unevictable_list - add a page to the unevictable list
250 * @page: the page to be added to the unevictable list
251 *
252 * Add page directly to its zone's unevictable list. To avoid races with
253 * tasks that might be making the page evictable, through eg. munlock,
254 * munmap or exit, while it's not on the lru, we want to add the page
255 * while it's locked or otherwise "invisible" to other tasks. This is
256 * difficult to do when using the pagevec cache, so bypass that.
257 */
258void add_page_to_unevictable_list(struct page *page)
259{
260    struct zone *zone = page_zone(page);
261
262    spin_lock_irq(&zone->lru_lock);
263    SetPageUnevictable(page);
264    SetPageLRU(page);
265    add_page_to_lru_list(zone, page, LRU_UNEVICTABLE);
266    spin_unlock_irq(&zone->lru_lock);
267}
268
269/*
270 * Drain pages out of the cpu's pagevecs.
271 * Either "cpu" is the current CPU, and preemption has already been
272 * disabled; or "cpu" is being hot-unplugged, and is already dead.
273 */
274static void drain_cpu_pagevecs(int cpu)
275{
276    struct pagevec *pvecs = per_cpu(lru_add_pvecs, cpu);
277    struct pagevec *pvec;
278    int lru;
279
280    for_each_lru(lru) {
281        pvec = &pvecs[lru - LRU_BASE];
282        if (pagevec_count(pvec))
283            ____pagevec_lru_add(pvec, lru);
284    }
285
286    pvec = &per_cpu(lru_rotate_pvecs, cpu);
287    if (pagevec_count(pvec)) {
288        unsigned long flags;
289
290        /* No harm done if a racing interrupt already did this */
291        local_irq_save(flags);
292        pagevec_move_tail(pvec);
293        local_irq_restore(flags);
294    }
295}
296
297void lru_add_drain(void)
298{
299    drain_cpu_pagevecs(get_cpu());
300    put_cpu();
301}
302
303static void lru_add_drain_per_cpu(struct work_struct *dummy)
304{
305    lru_add_drain();
306}
307
308/*
309 * Returns 0 for success
310 */
311int lru_add_drain_all(void)
312{
313    return schedule_on_each_cpu(lru_add_drain_per_cpu);
314}
315
316/*
317 * Batched page_cache_release(). Decrement the reference count on all the
318 * passed pages. If it fell to zero then remove the page from the LRU and
319 * free it.
320 *
321 * Avoid taking zone->lru_lock if possible, but if it is taken, retain it
322 * for the remainder of the operation.
323 *
324 * The locking in this function is against shrink_inactive_list(): we recheck
325 * the page count inside the lock to see whether shrink_inactive_list()
326 * grabbed the page via the LRU. If it did, give up: shrink_inactive_list()
327 * will free it.
328 */
329void release_pages(struct page **pages, int nr, int cold)
330{
331    int i;
332    struct pagevec pages_to_free;
333    struct zone *zone = NULL;
334    unsigned long uninitialized_var(flags);
335
336    pagevec_init(&pages_to_free, cold);
337    for (i = 0; i < nr; i++) {
338        struct page *page = pages[i];
339
340        if (unlikely(PageCompound(page))) {
341            if (zone) {
342                spin_unlock_irqrestore(&zone->lru_lock, flags);
343                zone = NULL;
344            }
345            put_compound_page(page);
346            continue;
347        }
348
349        if (!put_page_testzero(page))
350            continue;
351
352        if (PageLRU(page)) {
353            struct zone *pagezone = page_zone(page);
354
355            if (pagezone != zone) {
356                if (zone)
357                    spin_unlock_irqrestore(&zone->lru_lock,
358                                    flags);
359                zone = pagezone;
360                spin_lock_irqsave(&zone->lru_lock, flags);
361            }
362            VM_BUG_ON(!PageLRU(page));
363            __ClearPageLRU(page);
364            del_page_from_lru(zone, page);
365        }
366
367        if (!pagevec_add(&pages_to_free, page)) {
368            if (zone) {
369                spin_unlock_irqrestore(&zone->lru_lock, flags);
370                zone = NULL;
371            }
372            __pagevec_free(&pages_to_free);
373            pagevec_reinit(&pages_to_free);
374          }
375    }
376    if (zone)
377        spin_unlock_irqrestore(&zone->lru_lock, flags);
378
379    pagevec_free(&pages_to_free);
380}
381
382/*
383 * The pages which we're about to release may be in the deferred lru-addition
384 * queues. That would prevent them from really being freed right now. That's
385 * OK from a correctness point of view but is inefficient - those pages may be
386 * cache-warm and we want to give them back to the page allocator ASAP.
387 *
388 * So __pagevec_release() will drain those queues here. __pagevec_lru_add()
389 * and __pagevec_lru_add_active() call release_pages() directly to avoid
390 * mutual recursion.
391 */
392void __pagevec_release(struct pagevec *pvec)
393{
394    lru_add_drain();
395    release_pages(pvec->pages, pagevec_count(pvec), pvec->cold);
396    pagevec_reinit(pvec);
397}
398
399EXPORT_SYMBOL(__pagevec_release);
400
401/*
402 * Add the passed pages to the LRU, then drop the caller's refcount
403 * on them. Reinitialises the caller's pagevec.
404 */
405void ____pagevec_lru_add(struct pagevec *pvec, enum lru_list lru)
406{
407    int i;
408    struct zone *zone = NULL;
409
410    VM_BUG_ON(is_unevictable_lru(lru));
411
412    for (i = 0; i < pagevec_count(pvec); i++) {
413        struct page *page = pvec->pages[i];
414        struct zone *pagezone = page_zone(page);
415        int file;
416        int active;
417
418        if (pagezone != zone) {
419            if (zone)
420                spin_unlock_irq(&zone->lru_lock);
421            zone = pagezone;
422            spin_lock_irq(&zone->lru_lock);
423        }
424        VM_BUG_ON(PageActive(page));
425        VM_BUG_ON(PageUnevictable(page));
426        VM_BUG_ON(PageLRU(page));
427        SetPageLRU(page);
428        active = is_active_lru(lru);
429        file = is_file_lru(lru);
430        if (active)
431            SetPageActive(page);
432        update_page_reclaim_stat(zone, page, file, active);
433        add_page_to_lru_list(zone, page, lru);
434    }
435    if (zone)
436        spin_unlock_irq(&zone->lru_lock);
437    release_pages(pvec->pages, pvec->nr, pvec->cold);
438    pagevec_reinit(pvec);
439}
440
441EXPORT_SYMBOL(____pagevec_lru_add);
442
443/*
444 * Try to drop buffers from the pages in a pagevec
445 */
446void pagevec_strip(struct pagevec *pvec)
447{
448    int i;
449
450    for (i = 0; i < pagevec_count(pvec); i++) {
451        struct page *page = pvec->pages[i];
452
453        if (page_has_private(page) && trylock_page(page)) {
454            if (page_has_private(page))
455                try_to_release_page(page, 0);
456            unlock_page(page);
457        }
458    }
459}
460
461/**
462 * pagevec_lookup - gang pagecache lookup
463 * @pvec: Where the resulting pages are placed
464 * @mapping: The address_space to search
465 * @start: The starting page index
466 * @nr_pages: The maximum number of pages
467 *
468 * pagevec_lookup() will search for and return a group of up to @nr_pages pages
469 * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a
470 * reference against the pages in @pvec.
471 *
472 * The search returns a group of mapping-contiguous pages with ascending
473 * indexes. There may be holes in the indices due to not-present pages.
474 *
475 * pagevec_lookup() returns the number of pages which were found.
476 */
477unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping,
478        pgoff_t start, unsigned nr_pages)
479{
480    pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages);
481    return pagevec_count(pvec);
482}
483
484EXPORT_SYMBOL(pagevec_lookup);
485
486unsigned pagevec_lookup_tag(struct pagevec *pvec, struct address_space *mapping,
487        pgoff_t *index, int tag, unsigned nr_pages)
488{
489    pvec->nr = find_get_pages_tag(mapping, index, tag,
490                    nr_pages, pvec->pages);
491    return pagevec_count(pvec);
492}
493
494EXPORT_SYMBOL(pagevec_lookup_tag);
495
496/*
497 * Perform any setup for the swap system
498 */
499void __init swap_setup(void)
500{
501    unsigned long megs = totalram_pages >> (20 - PAGE_SHIFT);
502
503#ifdef CONFIG_SWAP
504    bdi_init(swapper_space.backing_dev_info);
505#endif
506
507    /* Use a smaller cluster for small-memory machines */
508    if (megs < 16)
509        page_cluster = 2;
510    else
511        page_cluster = 3;
512    /*
513     * Right now other parts of the system means that we
514     * _really_ don't want to cluster much more
515     */
516}
517

Archive Download this file



interactive