4 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
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.
12 * Swap aging added 23.2.95, Stephen Tweedie.
13 * Buffermem limits added 12.3.98, Rik van Riel.
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>
34 /* How many pages do we try to swap or page in/out together? */
37 void put_page(struct page *page)
39 if (unlikely(PageCompound(page))) {
40 page = (struct page *)page_private(page);
41 if (put_page_testzero(page)) {
42 void (*dtor)(struct page *page);
44 dtor = (void (*)(struct page *))page[1].mapping;
49 if (put_page_testzero(page))
50 __page_cache_release(page);
52 EXPORT_SYMBOL(put_page);
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.
59 * We don't expect many pages to come through here, so don't bother batching
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.
67 * Returns zero if it cleared PG_writeback.
69 int rotate_reclaimable_page(struct page *page)
83 zone = page_zone(page);
84 spin_lock_irqsave(&zone->lru_lock, flags);
85 if (PageLRU(page) && !PageActive(page)) {
87 list_add_tail(&page->lru, &zone->inactive_list);
88 inc_page_state(pgrotated);
90 if (!test_clear_page_writeback(page))
92 spin_unlock_irqrestore(&zone->lru_lock, flags);
97 * FIXME: speed this up?
99 void fastcall activate_page(struct page *page)
101 struct zone *zone = page_zone(page);
103 spin_lock_irq(&zone->lru_lock);
104 if (PageLRU(page) && !PageActive(page)) {
105 del_page_from_inactive_list(zone, page);
107 add_page_to_active_list(zone, page);
108 inc_page_state(pgactivate);
110 spin_unlock_irq(&zone->lru_lock);
114 * Mark a page as having seen activity.
116 * inactive,unreferenced -> inactive,referenced
117 * inactive,referenced -> active,unreferenced
118 * active,unreferenced -> active,referenced
120 void fastcall mark_page_accessed(struct page *page)
122 if (!PageActive(page) && PageReferenced(page) && PageLRU(page)) {
124 ClearPageReferenced(page);
125 } else if (!PageReferenced(page)) {
126 SetPageReferenced(page);
130 EXPORT_SYMBOL(mark_page_accessed);
133 * lru_cache_add: add a page to the page lists
134 * @page: the page to add
136 static DEFINE_PER_CPU(struct pagevec, lru_add_pvecs) = { 0, };
137 static DEFINE_PER_CPU(struct pagevec, lru_add_active_pvecs) = { 0, };
139 void fastcall lru_cache_add(struct page *page)
141 struct pagevec *pvec = &get_cpu_var(lru_add_pvecs);
143 page_cache_get(page);
144 if (!pagevec_add(pvec, page))
145 __pagevec_lru_add(pvec);
146 put_cpu_var(lru_add_pvecs);
149 void fastcall lru_cache_add_active(struct page *page)
151 struct pagevec *pvec = &get_cpu_var(lru_add_active_pvecs);
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);
159 static void __lru_add_drain(int cpu)
161 struct pagevec *pvec = &per_cpu(lru_add_pvecs, cpu);
163 /* CPU is dead, so no locking needed. */
164 if (pagevec_count(pvec))
165 __pagevec_lru_add(pvec);
166 pvec = &per_cpu(lru_add_active_pvecs, cpu);
167 if (pagevec_count(pvec))
168 __pagevec_lru_add_active(pvec);
171 void lru_add_drain(void)
173 __lru_add_drain(get_cpu());
178 static void lru_add_drain_per_cpu(void *dummy)
184 * Returns 0 for success
186 int lru_add_drain_all(void)
188 return schedule_on_each_cpu(lru_add_drain_per_cpu, NULL);
194 * Returns 0 for success
196 int lru_add_drain_all(void)
204 * This path almost never happens for VM activity - pages are normally
205 * freed via pagevecs. But it gets used by networking.
207 void fastcall __page_cache_release(struct page *page)
210 struct zone *zone = page_zone(page);
212 spin_lock_irqsave(&zone->lru_lock, flags);
213 if (TestClearPageLRU(page))
214 del_page_from_lru(zone, page);
215 if (page_count(page) != 0)
217 spin_unlock_irqrestore(&zone->lru_lock, flags);
222 EXPORT_SYMBOL(__page_cache_release);
225 * Batched page_cache_release(). Decrement the reference count on all the
226 * passed pages. If it fell to zero then remove the page from the LRU and
229 * Avoid taking zone->lru_lock if possible, but if it is taken, retain it
230 * for the remainder of the operation.
232 * The locking in this function is against shrink_cache(): we recheck the
233 * page count inside the lock to see whether shrink_cache grabbed the page
234 * via the LRU. If it did, give up: shrink_cache will free it.
236 void release_pages(struct page **pages, int nr, int cold)
239 struct pagevec pages_to_free;
240 struct zone *zone = NULL;
242 pagevec_init(&pages_to_free, cold);
243 for (i = 0; i < nr; i++) {
244 struct page *page = pages[i];
245 struct zone *pagezone;
247 if (!put_page_testzero(page))
250 pagezone = page_zone(page);
251 if (pagezone != zone) {
253 spin_unlock_irq(&zone->lru_lock);
255 spin_lock_irq(&zone->lru_lock);
257 if (TestClearPageLRU(page))
258 del_page_from_lru(zone, page);
259 if (page_count(page) == 0) {
260 if (!pagevec_add(&pages_to_free, page)) {
261 spin_unlock_irq(&zone->lru_lock);
262 __pagevec_free(&pages_to_free);
263 pagevec_reinit(&pages_to_free);
264 zone = NULL; /* No lock is held */
269 spin_unlock_irq(&zone->lru_lock);
271 pagevec_free(&pages_to_free);
275 * The pages which we're about to release may be in the deferred lru-addition
276 * queues. That would prevent them from really being freed right now. That's
277 * OK from a correctness point of view but is inefficient - those pages may be
278 * cache-warm and we want to give them back to the page allocator ASAP.
280 * So __pagevec_release() will drain those queues here. __pagevec_lru_add()
281 * and __pagevec_lru_add_active() call release_pages() directly to avoid
284 void __pagevec_release(struct pagevec *pvec)
287 release_pages(pvec->pages, pagevec_count(pvec), pvec->cold);
288 pagevec_reinit(pvec);
291 EXPORT_SYMBOL(__pagevec_release);
294 * pagevec_release() for pages which are known to not be on the LRU
296 * This function reinitialises the caller's pagevec.
298 void __pagevec_release_nonlru(struct pagevec *pvec)
301 struct pagevec pages_to_free;
303 pagevec_init(&pages_to_free, pvec->cold);
304 for (i = 0; i < pagevec_count(pvec); i++) {
305 struct page *page = pvec->pages[i];
307 BUG_ON(PageLRU(page));
308 if (put_page_testzero(page))
309 pagevec_add(&pages_to_free, page);
311 pagevec_free(&pages_to_free);
312 pagevec_reinit(pvec);
316 * Add the passed pages to the LRU, then drop the caller's refcount
317 * on them. Reinitialises the caller's pagevec.
319 void __pagevec_lru_add(struct pagevec *pvec)
322 struct zone *zone = NULL;
324 for (i = 0; i < pagevec_count(pvec); i++) {
325 struct page *page = pvec->pages[i];
326 struct zone *pagezone = page_zone(page);
328 if (pagezone != zone) {
330 spin_unlock_irq(&zone->lru_lock);
332 spin_lock_irq(&zone->lru_lock);
334 if (TestSetPageLRU(page))
336 add_page_to_inactive_list(zone, page);
339 spin_unlock_irq(&zone->lru_lock);
340 release_pages(pvec->pages, pvec->nr, pvec->cold);
341 pagevec_reinit(pvec);
344 EXPORT_SYMBOL(__pagevec_lru_add);
346 void __pagevec_lru_add_active(struct pagevec *pvec)
349 struct zone *zone = NULL;
351 for (i = 0; i < pagevec_count(pvec); i++) {
352 struct page *page = pvec->pages[i];
353 struct zone *pagezone = page_zone(page);
355 if (pagezone != zone) {
357 spin_unlock_irq(&zone->lru_lock);
359 spin_lock_irq(&zone->lru_lock);
361 if (TestSetPageLRU(page))
363 if (TestSetPageActive(page))
365 add_page_to_active_list(zone, page);
368 spin_unlock_irq(&zone->lru_lock);
369 release_pages(pvec->pages, pvec->nr, pvec->cold);
370 pagevec_reinit(pvec);
374 * Try to drop buffers from the pages in a pagevec
376 void pagevec_strip(struct pagevec *pvec)
380 for (i = 0; i < pagevec_count(pvec); i++) {
381 struct page *page = pvec->pages[i];
383 if (PagePrivate(page) && !TestSetPageLocked(page)) {
384 try_to_release_page(page, 0);
391 * pagevec_lookup - gang pagecache lookup
392 * @pvec: Where the resulting pages are placed
393 * @mapping: The address_space to search
394 * @start: The starting page index
395 * @nr_pages: The maximum number of pages
397 * pagevec_lookup() will search for and return a group of up to @nr_pages pages
398 * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a
399 * reference against the pages in @pvec.
401 * The search returns a group of mapping-contiguous pages with ascending
402 * indexes. There may be holes in the indices due to not-present pages.
404 * pagevec_lookup() returns the number of pages which were found.
406 unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping,
407 pgoff_t start, unsigned nr_pages)
409 pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages);
410 return pagevec_count(pvec);
413 EXPORT_SYMBOL(pagevec_lookup);
415 unsigned pagevec_lookup_tag(struct pagevec *pvec, struct address_space *mapping,
416 pgoff_t *index, int tag, unsigned nr_pages)
418 pvec->nr = find_get_pages_tag(mapping, index, tag,
419 nr_pages, pvec->pages);
420 return pagevec_count(pvec);
423 EXPORT_SYMBOL(pagevec_lookup_tag);
427 * We tolerate a little inaccuracy to avoid ping-ponging the counter between
430 #define ACCT_THRESHOLD max(16, NR_CPUS * 2)
432 static DEFINE_PER_CPU(long, committed_space) = 0;
434 void vm_acct_memory(long pages)
439 local = &__get_cpu_var(committed_space);
441 if (*local > ACCT_THRESHOLD || *local < -ACCT_THRESHOLD) {
442 atomic_add(*local, &vm_committed_space);
448 #ifdef CONFIG_HOTPLUG_CPU
450 /* Drop the CPU's cached committed space back into the central pool. */
451 static int cpu_swap_callback(struct notifier_block *nfb,
452 unsigned long action,
457 committed = &per_cpu(committed_space, (long)hcpu);
458 if (action == CPU_DEAD) {
459 atomic_add(*committed, &vm_committed_space);
461 __lru_add_drain((long)hcpu);
465 #endif /* CONFIG_HOTPLUG_CPU */
466 #endif /* CONFIG_SMP */
469 void percpu_counter_mod(struct percpu_counter *fbc, long amount)
475 pcount = per_cpu_ptr(fbc->counters, cpu);
476 count = *pcount + amount;
477 if (count >= FBC_BATCH || count <= -FBC_BATCH) {
478 spin_lock(&fbc->lock);
480 spin_unlock(&fbc->lock);
486 EXPORT_SYMBOL(percpu_counter_mod);
490 * Perform any setup for the swap system
492 void __init swap_setup(void)
494 unsigned long megs = num_physpages >> (20 - PAGE_SHIFT);
496 /* Use a smaller cluster for small-memory machines */
502 * Right now other parts of the system means that we
503 * _really_ don't want to cluster much more
505 hotcpu_notifier(cpu_swap_callback, 0);