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>
33 #include <linux/ckrm_mem_inline.h>
35 /* How many pages do we try to swap or page in/out together? */
38 #ifdef CONFIG_HUGETLB_PAGE
40 void put_page(struct page *page)
42 if (unlikely(PageCompound(page))) {
43 page = (struct page *)page->private;
44 if (put_page_testzero(page)) {
45 void (*dtor)(struct page *page);
47 dtor = (void (*)(struct page *))page[1].mapping;
52 if (!PageReserved(page) && put_page_testzero(page))
53 __page_cache_release(page);
55 EXPORT_SYMBOL(put_page);
59 * Writeback is about to end against a page which has been marked for immediate
60 * reclaim. If it still appears to be reclaimable, move it to the tail of the
61 * inactive list. The page still has PageWriteback set, which will pin it.
63 * We don't expect many pages to come through here, so don't bother batching
66 * To avoid placing the page at the tail of the LRU while PG_writeback is still
67 * set, this function will clear PG_writeback before performing the page
68 * motion. Do that inside the lru lock because once PG_writeback is cleared
69 * we may not touch the page.
71 * Returns zero if it cleared PG_writeback.
73 int rotate_reclaimable_page(struct page *page)
87 zone = page_zone(page);
88 spin_lock_irqsave(&zone->lru_lock, flags);
89 if (PageLRU(page) && !PageActive(page)) {
91 ckrm_add_tail_inactive(page);
92 inc_page_state(pgrotated);
94 if (!test_clear_page_writeback(page))
96 spin_unlock_irqrestore(&zone->lru_lock, flags);
101 * FIXME: speed this up?
103 void fastcall activate_page(struct page *page)
105 struct zone *zone = page_zone(page);
107 spin_lock_irq(&zone->lru_lock);
108 if (PageLRU(page) && !PageActive(page)) {
109 del_page_from_inactive_list(zone, page);
111 add_page_to_active_list(zone, page);
112 inc_page_state(pgactivate);
114 spin_unlock_irq(&zone->lru_lock);
118 * Mark a page as having seen activity.
120 * inactive,unreferenced -> inactive,referenced
121 * inactive,referenced -> active,unreferenced
122 * active,unreferenced -> active,referenced
124 void fastcall mark_page_accessed(struct page *page)
126 if (!PageActive(page) && PageReferenced(page) && PageLRU(page)) {
128 ClearPageReferenced(page);
129 } else if (!PageReferenced(page)) {
130 SetPageReferenced(page);
134 EXPORT_SYMBOL(mark_page_accessed);
137 * lru_cache_add: add a page to the page lists
138 * @page: the page to add
140 static DEFINE_PER_CPU(struct pagevec, lru_add_pvecs) = { 0, };
141 static DEFINE_PER_CPU(struct pagevec, lru_add_active_pvecs) = { 0, };
143 void fastcall lru_cache_add(struct page *page)
145 struct pagevec *pvec = &get_cpu_var(lru_add_pvecs);
147 page_cache_get(page);
148 if (!pagevec_add(pvec, page))
149 __pagevec_lru_add(pvec);
150 put_cpu_var(lru_add_pvecs);
153 void fastcall lru_cache_add_active(struct page *page)
155 struct pagevec *pvec = &get_cpu_var(lru_add_active_pvecs);
157 page_cache_get(page);
158 if (!pagevec_add(pvec, page))
159 __pagevec_lru_add_active(pvec);
160 put_cpu_var(lru_add_active_pvecs);
163 void lru_add_drain(void)
165 struct pagevec *pvec = &get_cpu_var(lru_add_pvecs);
167 if (pagevec_count(pvec))
168 __pagevec_lru_add(pvec);
169 pvec = &__get_cpu_var(lru_add_active_pvecs);
170 if (pagevec_count(pvec))
171 __pagevec_lru_add_active(pvec);
172 put_cpu_var(lru_add_pvecs);
176 * This path almost never happens for VM activity - pages are normally
177 * freed via pagevecs. But it gets used by networking.
179 void fastcall __page_cache_release(struct page *page)
182 struct zone *zone = page_zone(page);
184 spin_lock_irqsave(&zone->lru_lock, flags);
185 if (TestClearPageLRU(page))
186 del_page_from_lru(zone, page);
187 if (page_count(page) != 0)
189 spin_unlock_irqrestore(&zone->lru_lock, flags);
194 EXPORT_SYMBOL(__page_cache_release);
197 * Batched page_cache_release(). Decrement the reference count on all the
198 * passed pages. If it fell to zero then remove the page from the LRU and
201 * Avoid taking zone->lru_lock if possible, but if it is taken, retain it
202 * for the remainder of the operation.
204 * The locking in this function is against shrink_cache(): we recheck the
205 * page count inside the lock to see whether shrink_cache grabbed the page
206 * via the LRU. If it did, give up: shrink_cache will free it.
208 void release_pages(struct page **pages, int nr, int cold)
211 struct pagevec pages_to_free;
212 struct zone *zone = NULL;
214 pagevec_init(&pages_to_free, cold);
215 for (i = 0; i < nr; i++) {
216 struct page *page = pages[i];
217 struct zone *pagezone;
219 if (PageReserved(page) || !put_page_testzero(page))
222 pagezone = page_zone(page);
223 if (pagezone != zone) {
225 spin_unlock_irq(&zone->lru_lock);
227 spin_lock_irq(&zone->lru_lock);
229 if (TestClearPageLRU(page))
230 del_page_from_lru(zone, page);
231 if (page_count(page) == 0) {
232 if (!pagevec_add(&pages_to_free, page)) {
233 spin_unlock_irq(&zone->lru_lock);
234 __pagevec_free(&pages_to_free);
235 pagevec_reinit(&pages_to_free);
236 zone = NULL; /* No lock is held */
241 spin_unlock_irq(&zone->lru_lock);
243 pagevec_free(&pages_to_free);
247 * The pages which we're about to release may be in the deferred lru-addition
248 * queues. That would prevent them from really being freed right now. That's
249 * OK from a correctness point of view but is inefficient - those pages may be
250 * cache-warm and we want to give them back to the page allocator ASAP.
252 * So __pagevec_release() will drain those queues here. __pagevec_lru_add()
253 * and __pagevec_lru_add_active() call release_pages() directly to avoid
256 void __pagevec_release(struct pagevec *pvec)
259 release_pages(pvec->pages, pagevec_count(pvec), pvec->cold);
260 pagevec_reinit(pvec);
264 * pagevec_release() for pages which are known to not be on the LRU
266 * This function reinitialises the caller's pagevec.
268 void __pagevec_release_nonlru(struct pagevec *pvec)
271 struct pagevec pages_to_free;
273 pagevec_init(&pages_to_free, pvec->cold);
274 pages_to_free.cold = pvec->cold;
275 for (i = 0; i < pagevec_count(pvec); i++) {
276 struct page *page = pvec->pages[i];
278 BUG_ON(PageLRU(page));
279 if (put_page_testzero(page))
280 pagevec_add(&pages_to_free, page);
282 pagevec_free(&pages_to_free);
283 pagevec_reinit(pvec);
287 * Add the passed pages to the LRU, then drop the caller's refcount
288 * on them. Reinitialises the caller's pagevec.
290 void __pagevec_lru_add(struct pagevec *pvec)
293 struct zone *zone = NULL;
295 for (i = 0; i < pagevec_count(pvec); i++) {
296 struct page *page = pvec->pages[i];
297 struct zone *pagezone = page_zone(page);
299 if (pagezone != zone) {
301 spin_unlock_irq(&zone->lru_lock);
303 spin_lock_irq(&zone->lru_lock);
305 if (TestSetPageLRU(page))
307 add_page_to_inactive_list(zone, page);
310 spin_unlock_irq(&zone->lru_lock);
311 release_pages(pvec->pages, pvec->nr, pvec->cold);
312 pagevec_reinit(pvec);
315 EXPORT_SYMBOL(__pagevec_lru_add);
317 void __pagevec_lru_add_active(struct pagevec *pvec)
320 struct zone *zone = NULL;
322 for (i = 0; i < pagevec_count(pvec); i++) {
323 struct page *page = pvec->pages[i];
324 struct zone *pagezone = page_zone(page);
326 if (pagezone != zone) {
328 spin_unlock_irq(&zone->lru_lock);
330 spin_lock_irq(&zone->lru_lock);
332 if (TestSetPageLRU(page))
334 if (TestSetPageActive(page))
336 add_page_to_active_list(zone, page);
339 spin_unlock_irq(&zone->lru_lock);
340 release_pages(pvec->pages, pvec->nr, pvec->cold);
341 pagevec_reinit(pvec);
345 * Try to drop buffers from the pages in a pagevec
347 void pagevec_strip(struct pagevec *pvec)
351 for (i = 0; i < pagevec_count(pvec); i++) {
352 struct page *page = pvec->pages[i];
354 if (PagePrivate(page) && !TestSetPageLocked(page)) {
355 try_to_release_page(page, 0);
362 * pagevec_lookup - gang pagecache lookup
363 * @pvec: Where the resulting pages are placed
364 * @mapping: The address_space to search
365 * @start: The starting page index
366 * @nr_pages: The maximum number of pages
368 * pagevec_lookup() will search for and return a group of up to @nr_pages pages
369 * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a
370 * reference against the pages in @pvec.
372 * The search returns a group of mapping-contiguous pages with ascending
373 * indexes. There may be holes in the indices due to not-present pages.
375 * pagevec_lookup() returns the number of pages which were found.
377 unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping,
378 pgoff_t start, unsigned nr_pages)
380 pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages);
381 return pagevec_count(pvec);
384 unsigned pagevec_lookup_tag(struct pagevec *pvec, struct address_space *mapping,
385 pgoff_t *index, int tag, unsigned nr_pages)
387 pvec->nr = find_get_pages_tag(mapping, index, tag,
388 nr_pages, pvec->pages);
389 return pagevec_count(pvec);
395 * We tolerate a little inaccuracy to avoid ping-ponging the counter between
398 #define ACCT_THRESHOLD max(16, NR_CPUS * 2)
400 static DEFINE_PER_CPU(long, committed_space) = 0;
402 void vm_acct_memory(long pages)
407 local = &__get_cpu_var(committed_space);
409 if (*local > ACCT_THRESHOLD || *local < -ACCT_THRESHOLD) {
410 atomic_add(*local, &vm_committed_space);
415 EXPORT_SYMBOL(vm_acct_memory);
417 #ifdef CONFIG_HOTPLUG_CPU
418 static void lru_drain_cache(unsigned int cpu)
420 struct pagevec *pvec = &per_cpu(lru_add_pvecs, cpu);
422 /* CPU is dead, so no locking needed. */
423 if (pagevec_count(pvec))
424 __pagevec_lru_add(pvec);
425 pvec = &per_cpu(lru_add_active_pvecs, cpu);
426 if (pagevec_count(pvec))
427 __pagevec_lru_add_active(pvec);
430 /* Drop the CPU's cached committed space back into the central pool. */
431 static int cpu_swap_callback(struct notifier_block *nfb,
432 unsigned long action,
437 committed = &per_cpu(committed_space, (long)hcpu);
438 if (action == CPU_DEAD) {
439 atomic_add(*committed, &vm_committed_space);
441 lru_drain_cache((long)hcpu);
445 #endif /* CONFIG_HOTPLUG_CPU */
446 #endif /* CONFIG_SMP */
449 void percpu_counter_mod(struct percpu_counter *fbc, long amount)
455 pcount = per_cpu_ptr(fbc->counters, cpu);
456 count = *pcount + amount;
457 if (count >= FBC_BATCH || count <= -FBC_BATCH) {
458 spin_lock(&fbc->lock);
460 spin_unlock(&fbc->lock);
466 EXPORT_SYMBOL(percpu_counter_mod);
470 * Perform any setup for the swap system
472 void __init swap_setup(void)
474 unsigned long megs = num_physpages >> (20 - PAGE_SHIFT);
476 /* Use a smaller cluster for small-memory machines */
482 * Right now other parts of the system means that we
483 * _really_ don't want to cluster much more
485 hotcpu_notifier(cpu_swap_callback, 0);