blob: ee6d71ccfa56fe77173bc7806797b5fc01ec1112 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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 opereation 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/module.h>
28#include <linux/percpu_counter.h>
29#include <linux/percpu.h>
30#include <linux/cpu.h>
31#include <linux/notifier.h>
32#include <linux/init.h>
33
34/* How many pages do we try to swap or page in/out together? */
35int page_cluster;
36
Linus Torvalds1da177e2005-04-16 15:20:36 -070037void put_page(struct page *page)
38{
39 if (unlikely(PageCompound(page))) {
Hugh Dickins4c21e2f2005-10-29 18:16:40 -070040 page = (struct page *)page_private(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -070041 if (put_page_testzero(page)) {
42 void (*dtor)(struct page *page);
43
44 dtor = (void (*)(struct page *))page[1].mapping;
45 (*dtor)(page);
46 }
47 return;
48 }
Nick Pigginb5810032005-10-29 18:16:12 -070049 if (put_page_testzero(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -070050 __page_cache_release(page);
51}
52EXPORT_SYMBOL(put_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -070053
54/*
55 * Writeback is about to end against a page which has been marked for immediate
56 * reclaim. If it still appears to be reclaimable, move it to the tail of the
57 * inactive list. The page still has PageWriteback set, which will pin it.
58 *
59 * We don't expect many pages to come through here, so don't bother batching
60 * things up.
61 *
62 * To avoid placing the page at the tail of the LRU while PG_writeback is still
63 * set, this function will clear PG_writeback before performing the page
64 * motion. Do that inside the lru lock because once PG_writeback is cleared
65 * we may not touch the page.
66 *
67 * Returns zero if it cleared PG_writeback.
68 */
69int rotate_reclaimable_page(struct page *page)
70{
71 struct zone *zone;
72 unsigned long flags;
73
74 if (PageLocked(page))
75 return 1;
76 if (PageDirty(page))
77 return 1;
78 if (PageActive(page))
79 return 1;
80 if (!PageLRU(page))
81 return 1;
82
83 zone = page_zone(page);
84 spin_lock_irqsave(&zone->lru_lock, flags);
85 if (PageLRU(page) && !PageActive(page)) {
86 list_del(&page->lru);
87 list_add_tail(&page->lru, &zone->inactive_list);
88 inc_page_state(pgrotated);
89 }
90 if (!test_clear_page_writeback(page))
91 BUG();
92 spin_unlock_irqrestore(&zone->lru_lock, flags);
93 return 0;
94}
95
96/*
97 * FIXME: speed this up?
98 */
99void fastcall activate_page(struct page *page)
100{
101 struct zone *zone = page_zone(page);
102
103 spin_lock_irq(&zone->lru_lock);
104 if (PageLRU(page) && !PageActive(page)) {
105 del_page_from_inactive_list(zone, page);
106 SetPageActive(page);
107 add_page_to_active_list(zone, page);
108 inc_page_state(pgactivate);
109 }
110 spin_unlock_irq(&zone->lru_lock);
111}
112
113/*
114 * Mark a page as having seen activity.
115 *
116 * inactive,unreferenced -> inactive,referenced
117 * inactive,referenced -> active,unreferenced
118 * active,unreferenced -> active,referenced
119 */
120void fastcall mark_page_accessed(struct page *page)
121{
122 if (!PageActive(page) && PageReferenced(page) && PageLRU(page)) {
123 activate_page(page);
124 ClearPageReferenced(page);
125 } else if (!PageReferenced(page)) {
126 SetPageReferenced(page);
127 }
128}
129
130EXPORT_SYMBOL(mark_page_accessed);
131
132/**
133 * lru_cache_add: add a page to the page lists
134 * @page: the page to add
135 */
136static DEFINE_PER_CPU(struct pagevec, lru_add_pvecs) = { 0, };
137static DEFINE_PER_CPU(struct pagevec, lru_add_active_pvecs) = { 0, };
138
139void fastcall lru_cache_add(struct page *page)
140{
141 struct pagevec *pvec = &get_cpu_var(lru_add_pvecs);
142
143 page_cache_get(page);
144 if (!pagevec_add(pvec, page))
145 __pagevec_lru_add(pvec);
146 put_cpu_var(lru_add_pvecs);
147}
148
149void fastcall lru_cache_add_active(struct page *page)
150{
151 struct pagevec *pvec = &get_cpu_var(lru_add_active_pvecs);
152
153 page_cache_get(page);
154 if (!pagevec_add(pvec, page))
155 __pagevec_lru_add_active(pvec);
156 put_cpu_var(lru_add_active_pvecs);
157}
158
Andrew Morton80bfed92006-01-06 00:11:14 -0800159static void __lru_add_drain(int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160{
Andrew Morton80bfed92006-01-06 00:11:14 -0800161 struct pagevec *pvec = &per_cpu(lru_add_pvecs, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162
Andrew Morton80bfed92006-01-06 00:11:14 -0800163 /* CPU is dead, so no locking needed. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164 if (pagevec_count(pvec))
165 __pagevec_lru_add(pvec);
Andrew Morton80bfed92006-01-06 00:11:14 -0800166 pvec = &per_cpu(lru_add_active_pvecs, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167 if (pagevec_count(pvec))
168 __pagevec_lru_add_active(pvec);
Andrew Morton80bfed92006-01-06 00:11:14 -0800169}
170
171void lru_add_drain(void)
172{
173 __lru_add_drain(get_cpu());
174 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175}
176
177/*
178 * This path almost never happens for VM activity - pages are normally
179 * freed via pagevecs. But it gets used by networking.
180 */
181void fastcall __page_cache_release(struct page *page)
182{
183 unsigned long flags;
184 struct zone *zone = page_zone(page);
185
186 spin_lock_irqsave(&zone->lru_lock, flags);
187 if (TestClearPageLRU(page))
188 del_page_from_lru(zone, page);
189 if (page_count(page) != 0)
190 page = NULL;
191 spin_unlock_irqrestore(&zone->lru_lock, flags);
192 if (page)
193 free_hot_page(page);
194}
195
196EXPORT_SYMBOL(__page_cache_release);
197
198/*
199 * Batched page_cache_release(). Decrement the reference count on all the
200 * passed pages. If it fell to zero then remove the page from the LRU and
201 * free it.
202 *
203 * Avoid taking zone->lru_lock if possible, but if it is taken, retain it
204 * for the remainder of the operation.
205 *
206 * The locking in this function is against shrink_cache(): we recheck the
207 * page count inside the lock to see whether shrink_cache grabbed the page
208 * via the LRU. If it did, give up: shrink_cache will free it.
209 */
210void release_pages(struct page **pages, int nr, int cold)
211{
212 int i;
213 struct pagevec pages_to_free;
214 struct zone *zone = NULL;
215
216 pagevec_init(&pages_to_free, cold);
217 for (i = 0; i < nr; i++) {
218 struct page *page = pages[i];
219 struct zone *pagezone;
220
Nick Pigginb5810032005-10-29 18:16:12 -0700221 if (!put_page_testzero(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222 continue;
223
224 pagezone = page_zone(page);
225 if (pagezone != zone) {
226 if (zone)
227 spin_unlock_irq(&zone->lru_lock);
228 zone = pagezone;
229 spin_lock_irq(&zone->lru_lock);
230 }
231 if (TestClearPageLRU(page))
232 del_page_from_lru(zone, page);
233 if (page_count(page) == 0) {
234 if (!pagevec_add(&pages_to_free, page)) {
235 spin_unlock_irq(&zone->lru_lock);
236 __pagevec_free(&pages_to_free);
237 pagevec_reinit(&pages_to_free);
238 zone = NULL; /* No lock is held */
239 }
240 }
241 }
242 if (zone)
243 spin_unlock_irq(&zone->lru_lock);
244
245 pagevec_free(&pages_to_free);
246}
247
248/*
249 * The pages which we're about to release may be in the deferred lru-addition
250 * queues. That would prevent them from really being freed right now. That's
251 * OK from a correctness point of view but is inefficient - those pages may be
252 * cache-warm and we want to give them back to the page allocator ASAP.
253 *
254 * So __pagevec_release() will drain those queues here. __pagevec_lru_add()
255 * and __pagevec_lru_add_active() call release_pages() directly to avoid
256 * mutual recursion.
257 */
258void __pagevec_release(struct pagevec *pvec)
259{
260 lru_add_drain();
261 release_pages(pvec->pages, pagevec_count(pvec), pvec->cold);
262 pagevec_reinit(pvec);
263}
264
Steve French7f285702005-11-01 10:22:55 -0800265EXPORT_SYMBOL(__pagevec_release);
266
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267/*
268 * pagevec_release() for pages which are known to not be on the LRU
269 *
270 * This function reinitialises the caller's pagevec.
271 */
272void __pagevec_release_nonlru(struct pagevec *pvec)
273{
274 int i;
275 struct pagevec pages_to_free;
276
277 pagevec_init(&pages_to_free, pvec->cold);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278 for (i = 0; i < pagevec_count(pvec); i++) {
279 struct page *page = pvec->pages[i];
280
281 BUG_ON(PageLRU(page));
282 if (put_page_testzero(page))
283 pagevec_add(&pages_to_free, page);
284 }
285 pagevec_free(&pages_to_free);
286 pagevec_reinit(pvec);
287}
288
289/*
290 * Add the passed pages to the LRU, then drop the caller's refcount
291 * on them. Reinitialises the caller's pagevec.
292 */
293void __pagevec_lru_add(struct pagevec *pvec)
294{
295 int i;
296 struct zone *zone = NULL;
297
298 for (i = 0; i < pagevec_count(pvec); i++) {
299 struct page *page = pvec->pages[i];
300 struct zone *pagezone = page_zone(page);
301
302 if (pagezone != zone) {
303 if (zone)
304 spin_unlock_irq(&zone->lru_lock);
305 zone = pagezone;
306 spin_lock_irq(&zone->lru_lock);
307 }
308 if (TestSetPageLRU(page))
309 BUG();
310 add_page_to_inactive_list(zone, page);
311 }
312 if (zone)
313 spin_unlock_irq(&zone->lru_lock);
314 release_pages(pvec->pages, pvec->nr, pvec->cold);
315 pagevec_reinit(pvec);
316}
317
318EXPORT_SYMBOL(__pagevec_lru_add);
319
320void __pagevec_lru_add_active(struct pagevec *pvec)
321{
322 int i;
323 struct zone *zone = NULL;
324
325 for (i = 0; i < pagevec_count(pvec); i++) {
326 struct page *page = pvec->pages[i];
327 struct zone *pagezone = page_zone(page);
328
329 if (pagezone != zone) {
330 if (zone)
331 spin_unlock_irq(&zone->lru_lock);
332 zone = pagezone;
333 spin_lock_irq(&zone->lru_lock);
334 }
335 if (TestSetPageLRU(page))
336 BUG();
337 if (TestSetPageActive(page))
338 BUG();
339 add_page_to_active_list(zone, page);
340 }
341 if (zone)
342 spin_unlock_irq(&zone->lru_lock);
343 release_pages(pvec->pages, pvec->nr, pvec->cold);
344 pagevec_reinit(pvec);
345}
346
347/*
348 * Try to drop buffers from the pages in a pagevec
349 */
350void pagevec_strip(struct pagevec *pvec)
351{
352 int i;
353
354 for (i = 0; i < pagevec_count(pvec); i++) {
355 struct page *page = pvec->pages[i];
356
357 if (PagePrivate(page) && !TestSetPageLocked(page)) {
358 try_to_release_page(page, 0);
359 unlock_page(page);
360 }
361 }
362}
363
364/**
365 * pagevec_lookup - gang pagecache lookup
366 * @pvec: Where the resulting pages are placed
367 * @mapping: The address_space to search
368 * @start: The starting page index
369 * @nr_pages: The maximum number of pages
370 *
371 * pagevec_lookup() will search for and return a group of up to @nr_pages pages
372 * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a
373 * reference against the pages in @pvec.
374 *
375 * The search returns a group of mapping-contiguous pages with ascending
376 * indexes. There may be holes in the indices due to not-present pages.
377 *
378 * pagevec_lookup() returns the number of pages which were found.
379 */
380unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping,
381 pgoff_t start, unsigned nr_pages)
382{
383 pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages);
384 return pagevec_count(pvec);
385}
386
387unsigned pagevec_lookup_tag(struct pagevec *pvec, struct address_space *mapping,
388 pgoff_t *index, int tag, unsigned nr_pages)
389{
390 pvec->nr = find_get_pages_tag(mapping, index, tag,
391 nr_pages, pvec->pages);
392 return pagevec_count(pvec);
393}
394
Steve French7f285702005-11-01 10:22:55 -0800395EXPORT_SYMBOL(pagevec_lookup_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396
397#ifdef CONFIG_SMP
398/*
399 * We tolerate a little inaccuracy to avoid ping-ponging the counter between
400 * CPUs
401 */
402#define ACCT_THRESHOLD max(16, NR_CPUS * 2)
403
404static DEFINE_PER_CPU(long, committed_space) = 0;
405
406void vm_acct_memory(long pages)
407{
408 long *local;
409
410 preempt_disable();
411 local = &__get_cpu_var(committed_space);
412 *local += pages;
413 if (*local > ACCT_THRESHOLD || *local < -ACCT_THRESHOLD) {
414 atomic_add(*local, &vm_committed_space);
415 *local = 0;
416 }
417 preempt_enable();
418}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419
420#ifdef CONFIG_HOTPLUG_CPU
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421
422/* Drop the CPU's cached committed space back into the central pool. */
423static int cpu_swap_callback(struct notifier_block *nfb,
424 unsigned long action,
425 void *hcpu)
426{
427 long *committed;
428
429 committed = &per_cpu(committed_space, (long)hcpu);
430 if (action == CPU_DEAD) {
431 atomic_add(*committed, &vm_committed_space);
432 *committed = 0;
Andrew Morton80bfed92006-01-06 00:11:14 -0800433 __lru_add_drain((long)hcpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434 }
435 return NOTIFY_OK;
436}
437#endif /* CONFIG_HOTPLUG_CPU */
438#endif /* CONFIG_SMP */
439
440#ifdef CONFIG_SMP
441void percpu_counter_mod(struct percpu_counter *fbc, long amount)
442{
443 long count;
444 long *pcount;
445 int cpu = get_cpu();
446
447 pcount = per_cpu_ptr(fbc->counters, cpu);
448 count = *pcount + amount;
449 if (count >= FBC_BATCH || count <= -FBC_BATCH) {
450 spin_lock(&fbc->lock);
451 fbc->count += count;
452 spin_unlock(&fbc->lock);
453 count = 0;
454 }
455 *pcount = count;
456 put_cpu();
457}
458EXPORT_SYMBOL(percpu_counter_mod);
459#endif
460
461/*
462 * Perform any setup for the swap system
463 */
464void __init swap_setup(void)
465{
466 unsigned long megs = num_physpages >> (20 - PAGE_SHIFT);
467
468 /* Use a smaller cluster for small-memory machines */
469 if (megs < 16)
470 page_cluster = 2;
471 else
472 page_cluster = 3;
473 /*
474 * Right now other parts of the system means that we
475 * _really_ don't want to cluster much more
476 */
477 hotcpu_notifier(cpu_swap_callback, 0);
478}