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 * This path almost never happens for VM activity - pages are normally
179 * freed via pagevecs. But it gets used by networking.
181 void fastcall __page_cache_release(struct page *page)
184 struct zone *zone = page_zone(page);
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)
191 spin_unlock_irqrestore(&zone->lru_lock, flags);
196 EXPORT_SYMBOL(__page_cache_release);
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
203 * Avoid taking zone->lru_lock if possible, but if it is taken, retain it
204 * for the remainder of the operation.
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.
210 void release_pages(struct page **pages, int nr, int cold)
213 struct pagevec pages_to_free;
214 struct zone *zone = NULL;
216 pagevec_init(&pages_to_free, cold);
217 for (i = 0; i < nr; i++) {
218 struct page *page = pages[i];
219 struct zone *pagezone;
221 if (!put_page_testzero(page))
224 pagezone = page_zone(page);
225 if (pagezone != zone) {
227 spin_unlock_irq(&zone->lru_lock);
229 spin_lock_irq(&zone->lru_lock);
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 */
243 spin_unlock_irq(&zone->lru_lock);
245 pagevec_free(&pages_to_free);
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.
254 * So __pagevec_release() will drain those queues here. __pagevec_lru_add()
255 * and __pagevec_lru_add_active() call release_pages() directly to avoid
258 void __pagevec_release(struct pagevec *pvec)
261 release_pages(pvec->pages, pagevec_count(pvec), pvec->cold);
262 pagevec_reinit(pvec);
265 EXPORT_SYMBOL(__pagevec_release);
268 * pagevec_release() for pages which are known to not be on the LRU
270 * This function reinitialises the caller's pagevec.
272 void __pagevec_release_nonlru(struct pagevec *pvec)
275 struct pagevec pages_to_free;
277 pagevec_init(&pages_to_free, pvec->cold);
278 for (i = 0; i < pagevec_count(pvec); i++) {
279 struct page *page = pvec->pages[i];
281 BUG_ON(PageLRU(page));
282 if (put_page_testzero(page))
283 pagevec_add(&pages_to_free, page);
285 pagevec_free(&pages_to_free);
286 pagevec_reinit(pvec);
290 * Add the passed pages to the LRU, then drop the caller's refcount
291 * on them. Reinitialises the caller's pagevec.
293 void __pagevec_lru_add(struct pagevec *pvec)
296 struct zone *zone = NULL;
298 for (i = 0; i < pagevec_count(pvec); i++) {
299 struct page *page = pvec->pages[i];
300 struct zone *pagezone = page_zone(page);
302 if (pagezone != zone) {
304 spin_unlock_irq(&zone->lru_lock);
306 spin_lock_irq(&zone->lru_lock);
308 if (TestSetPageLRU(page))
310 add_page_to_inactive_list(zone, page);
313 spin_unlock_irq(&zone->lru_lock);
314 release_pages(pvec->pages, pvec->nr, pvec->cold);
315 pagevec_reinit(pvec);
318 EXPORT_SYMBOL(__pagevec_lru_add);
320 void __pagevec_lru_add_active(struct pagevec *pvec)
323 struct zone *zone = NULL;
325 for (i = 0; i < pagevec_count(pvec); i++) {
326 struct page *page = pvec->pages[i];
327 struct zone *pagezone = page_zone(page);
329 if (pagezone != zone) {
331 spin_unlock_irq(&zone->lru_lock);
333 spin_lock_irq(&zone->lru_lock);
335 if (TestSetPageLRU(page))
337 if (TestSetPageActive(page))
339 add_page_to_active_list(zone, page);
342 spin_unlock_irq(&zone->lru_lock);
343 release_pages(pvec->pages, pvec->nr, pvec->cold);
344 pagevec_reinit(pvec);
348 * Try to drop buffers from the pages in a pagevec
350 void pagevec_strip(struct pagevec *pvec)
354 for (i = 0; i < pagevec_count(pvec); i++) {
355 struct page *page = pvec->pages[i];
357 if (PagePrivate(page) && !TestSetPageLocked(page)) {
358 try_to_release_page(page, 0);
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
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.
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.
378 * pagevec_lookup() returns the number of pages which were found.
380 unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping,
381 pgoff_t start, unsigned nr_pages)
383 pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages);
384 return pagevec_count(pvec);
387 EXPORT_SYMBOL(pagevec_lookup);
389 unsigned pagevec_lookup_tag(struct pagevec *pvec, struct address_space *mapping,
390 pgoff_t *index, int tag, unsigned nr_pages)
392 pvec->nr = find_get_pages_tag(mapping, index, tag,
393 nr_pages, pvec->pages);
394 return pagevec_count(pvec);
397 EXPORT_SYMBOL(pagevec_lookup_tag);
401 * We tolerate a little inaccuracy to avoid ping-ponging the counter between
404 #define ACCT_THRESHOLD max(16, NR_CPUS * 2)
406 static DEFINE_PER_CPU(long, committed_space) = 0;
408 void vm_acct_memory(long pages)
413 local = &__get_cpu_var(committed_space);
415 if (*local > ACCT_THRESHOLD || *local < -ACCT_THRESHOLD) {
416 atomic_add(*local, &vm_committed_space);
422 #ifdef CONFIG_HOTPLUG_CPU
424 /* Drop the CPU's cached committed space back into the central pool. */
425 static int cpu_swap_callback(struct notifier_block *nfb,
426 unsigned long action,
431 committed = &per_cpu(committed_space, (long)hcpu);
432 if (action == CPU_DEAD) {
433 atomic_add(*committed, &vm_committed_space);
435 __lru_add_drain((long)hcpu);
439 #endif /* CONFIG_HOTPLUG_CPU */
440 #endif /* CONFIG_SMP */
443 void percpu_counter_mod(struct percpu_counter *fbc, long amount)
449 pcount = per_cpu_ptr(fbc->counters, cpu);
450 count = *pcount + amount;
451 if (count >= FBC_BATCH || count <= -FBC_BATCH) {
452 spin_lock(&fbc->lock);
454 spin_unlock(&fbc->lock);
460 EXPORT_SYMBOL(percpu_counter_mod);
464 * Perform any setup for the swap system
466 void __init swap_setup(void)
468 unsigned long megs = num_physpages >> (20 - PAGE_SHIFT);
470 /* Use a smaller cluster for small-memory machines */
476 * Right now other parts of the system means that we
477 * _really_ don't want to cluster much more
479 hotcpu_notifier(cpu_swap_callback, 0);