2 * linux/mm/page_alloc.c
4 * Manages the free list, the system allocates free pages here.
5 * Note that kmalloc() lives in slab.c
7 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * Swap reorganised 29.12.95, Stephen Tweedie
9 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
10 * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
11 * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
12 * Zone balancing, Kanoj Sarcar, SGI, Jan 2000
13 * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
14 * (lots of bits borrowed from Ingo Molnar & Andrew Morton)
17 #include <linux/config.h>
18 #include <linux/stddef.h>
20 #include <linux/swap.h>
21 #include <linux/interrupt.h>
22 #include <linux/pagemap.h>
23 #include <linux/bootmem.h>
24 #include <linux/compiler.h>
25 #include <linux/module.h>
26 #include <linux/suspend.h>
27 #include <linux/pagevec.h>
28 #include <linux/blkdev.h>
29 #include <linux/slab.h>
30 #include <linux/notifier.h>
31 #include <linux/topology.h>
32 #include <linux/sysctl.h>
33 #include <linux/cpu.h>
35 #include <asm/tlbflush.h>
37 DECLARE_BITMAP(node_online_map, MAX_NUMNODES);
38 struct pglist_data *pgdat_list;
39 unsigned long totalram_pages;
40 unsigned long totalhigh_pages;
43 int sysctl_lower_zone_protection = 0;
45 EXPORT_SYMBOL(totalram_pages);
46 EXPORT_SYMBOL(nr_swap_pages);
49 * Used by page_zone() to look up the address of the struct zone whose
50 * id is encoded in the upper bits of page->flags
52 struct zone *zone_table[1 << (ZONES_SHIFT + NODES_SHIFT)];
53 EXPORT_SYMBOL(zone_table);
55 static char *zone_names[MAX_NR_ZONES] = { "DMA", "Normal", "HighMem" };
56 int min_free_kbytes = 1024;
59 * Temporary debugging check for pages not lying within a given zone.
61 static int bad_range(struct zone *zone, struct page *page)
63 if (page_to_pfn(page) >= zone->zone_start_pfn + zone->spanned_pages)
65 if (page_to_pfn(page) < zone->zone_start_pfn)
67 if (zone != page_zone(page))
72 static void bad_page(const char *function, struct page *page)
74 printk(KERN_EMERG "Bad page state at %s (in process '%s', page %p)\n",
75 function, current->comm, page);
76 printk(KERN_EMERG "flags:0x%08lx mapping:%p mapcount:%d count:%d\n",
77 (unsigned long)page->flags, page->mapping,
78 (int)page->mapcount, page_count(page));
79 printk(KERN_EMERG "Backtrace:\n");
81 printk(KERN_EMERG "Trying to fix it up, but a reboot is needed\n");
82 page->flags &= ~(1 << PG_private |
91 set_page_count(page, 0);
96 #ifndef CONFIG_HUGETLB_PAGE
97 #define prep_compound_page(page, order) do { } while (0)
98 #define destroy_compound_page(page, order) do { } while (0)
101 * Higher-order pages are called "compound pages". They are structured thusly:
103 * The first PAGE_SIZE page is called the "head page".
105 * The remaining PAGE_SIZE pages are called "tail pages".
107 * All pages have PG_compound set. All pages have their ->private pointing at
108 * the head page (even the head page has this).
110 * The first tail page's ->mapping, if non-zero, holds the address of the
111 * compound page's put_page() function.
113 * The order of the allocation is stored in the first tail page's ->index
114 * This is only for debug at present. This usage means that zero-order pages
115 * may not be compound.
117 static void prep_compound_page(struct page *page, unsigned long order)
120 int nr_pages = 1 << order;
123 page[1].index = order;
124 for (i = 0; i < nr_pages; i++) {
125 struct page *p = page + i;
128 p->private = (unsigned long)page;
132 static void destroy_compound_page(struct page *page, unsigned long order)
135 int nr_pages = 1 << order;
137 if (!PageCompound(page))
140 if (page[1].index != order)
141 bad_page(__FUNCTION__, page);
143 for (i = 0; i < nr_pages; i++) {
144 struct page *p = page + i;
146 if (!PageCompound(p))
147 bad_page(__FUNCTION__, page);
148 if (p->private != (unsigned long)page)
149 bad_page(__FUNCTION__, page);
150 ClearPageCompound(p);
153 #endif /* CONFIG_HUGETLB_PAGE */
156 * Freeing function for a buddy system allocator.
158 * The concept of a buddy system is to maintain direct-mapped table
159 * (containing bit values) for memory blocks of various "orders".
160 * The bottom level table contains the map for the smallest allocatable
161 * units of memory (here, pages), and each level above it describes
162 * pairs of units from the levels below, hence, "buddies".
163 * At a high level, all that happens here is marking the table entry
164 * at the bottom level available, and propagating the changes upward
165 * as necessary, plus some accounting needed to play nicely with other
166 * parts of the VM system.
167 * At each level, we keep one bit for each pair of blocks, which
168 * is set to 1 iff only one of the pair is allocated. So when we
169 * are allocating or freeing one, we can derive the state of the
170 * other. That is, if we allocate a small block, and both were
171 * free, the remainder of the region must be split into blocks.
172 * If a block is freed, and its buddy is also free, then this
173 * triggers coalescing into a block of larger size.
178 static inline void __free_pages_bulk (struct page *page, struct page *base,
179 struct zone *zone, struct free_area *area, unsigned long mask,
182 unsigned long page_idx, index;
185 destroy_compound_page(page, order);
186 page_idx = page - base;
187 if (page_idx & ~mask)
189 index = page_idx >> (1 + order);
191 zone->free_pages -= mask;
192 while (mask + (1 << (MAX_ORDER-1))) {
193 struct page *buddy1, *buddy2;
195 BUG_ON(area >= zone->free_area + MAX_ORDER);
196 if (!__test_and_change_bit(index, area->map))
198 * the buddy page is still allocated.
202 * Move the buddy up one level.
203 * This code is taking advantage of the identity:
206 buddy1 = base + (page_idx ^ -mask);
207 buddy2 = base + page_idx;
208 BUG_ON(bad_range(zone, buddy1));
209 BUG_ON(bad_range(zone, buddy2));
210 list_del(&buddy1->lru);
216 list_add(&(base + page_idx)->lru, &area->free_list);
219 static inline void free_pages_check(const char *function, struct page *page)
221 if ( page_mapped(page) ||
222 page->mapping != NULL ||
223 page_count(page) != 0 ||
234 1 << PG_writeback )))
235 bad_page(function, page);
237 ClearPageDirty(page);
241 * Frees a list of pages.
242 * Assumes all pages on list are in same zone, and of same order.
243 * count is the number of pages to free, or 0 for all on the list.
245 * If the zone was previously in an "all pages pinned" state then look to
246 * see if this freeing clears that state.
248 * And clear the zone's pages_scanned counter, to hold off the "all pages are
249 * pinned" detection logic.
252 free_pages_bulk(struct zone *zone, int count,
253 struct list_head *list, unsigned int order)
255 unsigned long mask, flags;
256 struct free_area *area;
257 struct page *base, *page = NULL;
260 mask = (~0UL) << order;
261 base = zone->zone_mem_map;
262 area = zone->free_area + order;
263 spin_lock_irqsave(&zone->lock, flags);
264 zone->all_unreclaimable = 0;
265 zone->pages_scanned = 0;
266 while (!list_empty(list) && count--) {
267 page = list_entry(list->prev, struct page, lru);
268 /* have to delete it as __free_pages_bulk list manipulates */
269 list_del(&page->lru);
270 __free_pages_bulk(page, base, zone, area, mask, order);
273 spin_unlock_irqrestore(&zone->lock, flags);
277 void __free_pages_ok(struct page *page, unsigned int order)
282 arch_free_page(page, order);
284 mod_page_state(pgfree, 1 << order);
285 for (i = 0 ; i < (1 << order) ; ++i)
286 free_pages_check(__FUNCTION__, page + i);
287 list_add(&page->lru, &list);
288 kernel_map_pages(page, 1<<order, 0);
289 free_pages_bulk(page_zone(page), 1, &list, order);
292 #define MARK_USED(index, order, area) \
293 __change_bit((index) >> (1+(order)), (area)->map)
295 static inline struct page *
296 expand(struct zone *zone, struct page *page,
297 unsigned long index, int low, int high, struct free_area *area)
299 unsigned long size = 1 << high;
302 BUG_ON(bad_range(zone, page));
306 list_add(&page->lru, &area->free_list);
307 MARK_USED(index, high, area);
314 static inline void set_page_refs(struct page *page, int order)
317 set_page_count(page, 1);
322 * We need to reference all the pages for this order, otherwise if
323 * anyone accesses one of the pages with (get/put) it will be freed.
325 for (i = 0; i < (1 << order); i++)
326 set_page_count(page+i, 1);
327 #endif /* CONFIG_MMU */
331 * This page is about to be returned from the page allocator
333 static void prep_new_page(struct page *page, int order)
335 if (page->mapping || page_mapped(page) ||
346 1 << PG_writeback )))
347 bad_page(__FUNCTION__, page);
349 page->flags &= ~(1 << PG_uptodate | 1 << PG_error |
350 1 << PG_referenced | 1 << PG_arch_1 |
351 1 << PG_checked | 1 << PG_mappedtodisk);
353 set_page_refs(page, order);
357 * Do the hard work of removing an element from the buddy allocator.
358 * Call me with the zone->lock already held.
360 static struct page *__rmqueue(struct zone *zone, unsigned int order)
362 struct free_area * area;
363 unsigned int current_order;
367 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
368 area = zone->free_area + current_order;
369 if (list_empty(&area->free_list))
372 page = list_entry(area->free_list.next, struct page, lru);
373 list_del(&page->lru);
374 index = page - zone->zone_mem_map;
375 if (current_order != MAX_ORDER-1)
376 MARK_USED(index, current_order, area);
377 zone->free_pages -= 1UL << order;
378 return expand(zone, page, index, order, current_order, area);
385 * Obtain a specified number of elements from the buddy allocator, all under
386 * a single hold of the lock, for efficiency. Add them to the supplied list.
387 * Returns the number of new pages which were placed at *list.
389 static int rmqueue_bulk(struct zone *zone, unsigned int order,
390 unsigned long count, struct list_head *list)
397 spin_lock_irqsave(&zone->lock, flags);
398 for (i = 0; i < count; ++i) {
399 page = __rmqueue(zone, order);
403 list_add_tail(&page->lru, list);
405 spin_unlock_irqrestore(&zone->lock, flags);
409 #if defined(CONFIG_PM) || defined(CONFIG_HOTPLUG_CPU)
410 static void __drain_pages(unsigned int cpu)
415 for_each_zone(zone) {
416 struct per_cpu_pageset *pset;
418 pset = &zone->pageset[cpu];
419 for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
420 struct per_cpu_pages *pcp;
423 pcp->count -= free_pages_bulk(zone, pcp->count,
428 #endif /* CONFIG_PM || CONFIG_HOTPLUG_CPU */
431 int is_head_of_free_region(struct page *page)
433 struct zone *zone = page_zone(page);
436 struct list_head *curr;
439 * Should not matter as we need quiescent system for
440 * suspend anyway, but...
442 spin_lock_irqsave(&zone->lock, flags);
443 for (order = MAX_ORDER - 1; order >= 0; --order)
444 list_for_each(curr, &zone->free_area[order].free_list)
445 if (page == list_entry(curr, struct page, lru)) {
446 spin_unlock_irqrestore(&zone->lock, flags);
449 spin_unlock_irqrestore(&zone->lock, flags);
454 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
456 void drain_local_pages(void)
460 local_irq_save(flags);
461 __drain_pages(smp_processor_id());
462 local_irq_restore(flags);
464 #endif /* CONFIG_PM */
466 static void zone_statistics(struct zonelist *zonelist, struct zone *z)
471 pg_data_t *pg = z->zone_pgdat;
472 pg_data_t *orig = zonelist->zones[0]->zone_pgdat;
473 struct per_cpu_pageset *p;
475 local_irq_save(flags);
476 cpu = smp_processor_id();
477 p = &z->pageset[cpu];
479 z->pageset[cpu].numa_hit++;
482 zonelist->zones[0]->pageset[cpu].numa_foreign++;
484 if (pg == NODE_DATA(numa_node_id()))
488 local_irq_restore(flags);
493 * Free a 0-order page
495 static void FASTCALL(free_hot_cold_page(struct page *page, int cold));
496 static void fastcall free_hot_cold_page(struct page *page, int cold)
498 struct zone *zone = page_zone(page);
499 struct per_cpu_pages *pcp;
502 arch_free_page(page, 0);
504 kernel_map_pages(page, 1, 0);
505 inc_page_state(pgfree);
506 free_pages_check(__FUNCTION__, page);
507 pcp = &zone->pageset[get_cpu()].pcp[cold];
508 local_irq_save(flags);
509 if (pcp->count >= pcp->high)
510 pcp->count -= free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
511 list_add(&page->lru, &pcp->list);
513 local_irq_restore(flags);
517 void fastcall free_hot_page(struct page *page)
519 free_hot_cold_page(page, 0);
522 void fastcall free_cold_page(struct page *page)
524 free_hot_cold_page(page, 1);
528 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
529 * we cheat by calling it from here, in the order > 0 path. Saves a branch
534 buffered_rmqueue(struct zone *zone, int order, int gfp_flags)
537 struct page *page = NULL;
538 int cold = !!(gfp_flags & __GFP_COLD);
541 struct per_cpu_pages *pcp;
543 pcp = &zone->pageset[get_cpu()].pcp[cold];
544 local_irq_save(flags);
545 if (pcp->count <= pcp->low)
546 pcp->count += rmqueue_bulk(zone, 0,
547 pcp->batch, &pcp->list);
549 page = list_entry(pcp->list.next, struct page, lru);
550 list_del(&page->lru);
553 local_irq_restore(flags);
558 spin_lock_irqsave(&zone->lock, flags);
559 page = __rmqueue(zone, order);
560 spin_unlock_irqrestore(&zone->lock, flags);
564 BUG_ON(bad_range(zone, page));
565 mod_page_state_zone(zone, pgalloc, 1 << order);
566 prep_new_page(page, order);
567 if (order && (gfp_flags & __GFP_COMP))
568 prep_compound_page(page, order);
574 * This is the 'heart' of the zoned buddy allocator.
576 * Herein lies the mysterious "incremental min". That's the
578 * local_low = z->pages_low;
581 * thing. The intent here is to provide additional protection to low zones for
582 * allocation requests which _could_ use higher zones. So a GFP_HIGHMEM
583 * request is not allowed to dip as deeply into the normal zone as a GFP_KERNEL
584 * request. This preserves additional space in those lower zones for requests
585 * which really do need memory from those zones. It means that on a decent
586 * sized machine, GFP_HIGHMEM and GFP_KERNEL requests basically leave the DMA
589 struct page * fastcall
590 __alloc_pages(unsigned int gfp_mask, unsigned int order,
591 struct zonelist *zonelist)
593 const int wait = gfp_mask & __GFP_WAIT;
597 struct reclaim_state reclaim_state;
598 struct task_struct *p = current;
603 might_sleep_if(wait);
605 zones = zonelist->zones; /* the list of zones suitable for gfp_mask */
606 if (zones[0] == NULL) /* no zones in the zonelist */
609 alloc_type = zone_idx(zones[0]);
611 /* Go through the zonelist once, looking for a zone with enough free */
612 for (i = 0; zones[i] != NULL; i++) {
613 struct zone *z = zones[i];
615 min = (1<<order) + z->protection[alloc_type];
618 * We let real-time tasks dip their real-time paws a little
619 * deeper into reserves.
622 min -= z->pages_low >> 1;
624 if (z->free_pages >= min ||
625 (!wait && z->free_pages >= z->pages_high)) {
626 page = buffered_rmqueue(z, order, gfp_mask);
628 zone_statistics(zonelist, z);
634 /* we're somewhat low on memory, failed to find what we needed */
635 for (i = 0; zones[i] != NULL; i++)
636 wakeup_kswapd(zones[i]);
638 /* Go through the zonelist again, taking __GFP_HIGH into account */
639 for (i = 0; zones[i] != NULL; i++) {
640 struct zone *z = zones[i];
642 min = (1<<order) + z->protection[alloc_type];
644 if (gfp_mask & __GFP_HIGH)
645 min -= z->pages_low >> 2;
647 min -= z->pages_low >> 1;
649 if (z->free_pages >= min ||
650 (!wait && z->free_pages >= z->pages_high)) {
651 page = buffered_rmqueue(z, order, gfp_mask);
653 zone_statistics(zonelist, z);
659 /* here we're in the low on memory slow path */
662 if ((p->flags & (PF_MEMALLOC | PF_MEMDIE)) && !in_interrupt()) {
663 /* go through the zonelist yet again, ignoring mins */
664 for (i = 0; zones[i] != NULL; i++) {
665 struct zone *z = zones[i];
667 page = buffered_rmqueue(z, order, gfp_mask);
669 zone_statistics(zonelist, z);
676 /* Atomic allocations - we can't balance anything */
680 p->flags |= PF_MEMALLOC;
681 reclaim_state.reclaimed_slab = 0;
682 p->reclaim_state = &reclaim_state;
684 try_to_free_pages(zones, gfp_mask, order);
686 p->reclaim_state = NULL;
687 p->flags &= ~PF_MEMALLOC;
689 /* go through the zonelist yet one more time */
690 for (i = 0; zones[i] != NULL; i++) {
691 struct zone *z = zones[i];
693 min = (1UL << order) + z->protection[alloc_type];
695 if (z->free_pages >= min ||
696 (!wait && z->free_pages >= z->pages_high)) {
697 page = buffered_rmqueue(z, order, gfp_mask);
699 zone_statistics(zonelist, z);
706 * Don't let big-order allocations loop unless the caller explicitly
707 * requests that. Wait for some write requests to complete then retry.
709 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL, but that
710 * may not be true in other implementations.
713 if (!(gfp_mask & __GFP_NORETRY)) {
714 if ((order <= 3) || (gfp_mask & __GFP_REPEAT))
716 if (gfp_mask & __GFP_NOFAIL)
720 blk_congestion_wait(WRITE, HZ/50);
725 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
726 printk(KERN_WARNING "%s: page allocation failure."
727 " order:%d, mode:0x%x\n",
728 p->comm, order, gfp_mask);
733 kernel_map_pages(page, 1 << order, 1);
737 EXPORT_SYMBOL(__alloc_pages);
740 /* Early boot: Everything is done by one cpu, but the data structures will be
741 * used by all cpus - spread them on all nodes.
743 static __init unsigned long get_boot_pages(unsigned int gfp_mask, unsigned int order)
750 if (i > nodenr + numnodes)
752 if (node_present_pages(i%numnodes)) {
754 /* The node contains memory. Check that there is
755 * memory in the intended zonelist.
757 z = NODE_DATA(i%numnodes)->node_zonelists[gfp_mask & GFP_ZONEMASK].zones;
759 if ( (*z)->free_pages > (1UL<<order))
768 page = alloc_pages_node(i%numnodes, gfp_mask, order);
771 return (unsigned long) page_address(page);
776 * Common helper functions.
778 fastcall unsigned long __get_free_pages(unsigned int gfp_mask, unsigned int order)
783 if (unlikely(system_state == SYSTEM_BOOTING))
784 return get_boot_pages(gfp_mask, order);
786 page = alloc_pages(gfp_mask, order);
789 return (unsigned long) page_address(page);
792 EXPORT_SYMBOL(__get_free_pages);
794 fastcall unsigned long get_zeroed_page(unsigned int gfp_mask)
799 * get_zeroed_page() returns a 32-bit address, which cannot represent
802 BUG_ON(gfp_mask & __GFP_HIGHMEM);
804 page = alloc_pages(gfp_mask, 0);
806 void *address = page_address(page);
808 return (unsigned long) address;
813 EXPORT_SYMBOL(get_zeroed_page);
815 void __pagevec_free(struct pagevec *pvec)
817 int i = pagevec_count(pvec);
820 free_hot_cold_page(pvec->pages[i], pvec->cold);
823 fastcall void __free_pages(struct page *page, unsigned int order)
825 if (!PageReserved(page) && put_page_testzero(page)) {
829 __free_pages_ok(page, order);
833 EXPORT_SYMBOL(__free_pages);
835 fastcall void free_pages(unsigned long addr, unsigned int order)
838 BUG_ON(!virt_addr_valid(addr));
839 __free_pages(virt_to_page(addr), order);
843 EXPORT_SYMBOL(free_pages);
846 * Total amount of free (allocatable) RAM:
848 unsigned int nr_free_pages(void)
850 unsigned int sum = 0;
854 sum += zone->free_pages;
859 EXPORT_SYMBOL(nr_free_pages);
861 unsigned int nr_used_zone_pages(void)
863 unsigned int pages = 0;
867 pages += zone->nr_active + zone->nr_inactive;
873 unsigned int nr_free_pages_pgdat(pg_data_t *pgdat)
875 unsigned int i, sum = 0;
877 for (i = 0; i < MAX_NR_ZONES; i++)
878 sum += pgdat->node_zones[i].free_pages;
884 static unsigned int nr_free_zone_pages(int offset)
887 unsigned int sum = 0;
889 for_each_pgdat(pgdat) {
890 struct zonelist *zonelist = pgdat->node_zonelists + offset;
891 struct zone **zonep = zonelist->zones;
894 for (zone = *zonep++; zone; zone = *zonep++) {
895 unsigned long size = zone->present_pages;
896 unsigned long high = zone->pages_high;
906 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
908 unsigned int nr_free_buffer_pages(void)
910 return nr_free_zone_pages(GFP_USER & GFP_ZONEMASK);
914 * Amount of free RAM allocatable within all zones
916 unsigned int nr_free_pagecache_pages(void)
918 return nr_free_zone_pages(GFP_HIGHUSER & GFP_ZONEMASK);
921 #ifdef CONFIG_HIGHMEM
922 unsigned int nr_free_highpages (void)
925 unsigned int pages = 0;
927 for_each_pgdat(pgdat)
928 pages += pgdat->node_zones[ZONE_HIGHMEM].free_pages;
935 static void show_node(struct zone *zone)
937 printk("Node %d ", zone->zone_pgdat->node_id);
940 #define show_node(zone) do { } while (0)
944 * Accumulate the page_state information across all CPUs.
945 * The result is unavoidably approximate - it can change
946 * during and after execution of this function.
948 DEFINE_PER_CPU(struct page_state, page_states) = {0};
949 EXPORT_PER_CPU_SYMBOL(page_states);
951 atomic_t nr_pagecache = ATOMIC_INIT(0);
952 EXPORT_SYMBOL(nr_pagecache);
954 DEFINE_PER_CPU(long, nr_pagecache_local) = 0;
957 void __get_page_state(struct page_state *ret, int nr)
961 memset(ret, 0, sizeof(*ret));
962 while (cpu < NR_CPUS) {
963 unsigned long *in, *out, off;
965 if (!cpu_possible(cpu)) {
970 in = (unsigned long *)&per_cpu(page_states, cpu);
972 if (cpu < NR_CPUS && cpu_possible(cpu))
973 prefetch(&per_cpu(page_states, cpu));
974 out = (unsigned long *)ret;
975 for (off = 0; off < nr; off++)
980 void get_page_state(struct page_state *ret)
984 nr = offsetof(struct page_state, GET_PAGE_STATE_LAST);
985 nr /= sizeof(unsigned long);
987 __get_page_state(ret, nr + 1);
990 void get_full_page_state(struct page_state *ret)
992 __get_page_state(ret, sizeof(*ret) / sizeof(unsigned long));
995 void get_zone_counts(unsigned long *active,
996 unsigned long *inactive, unsigned long *free)
1003 for_each_zone(zone) {
1004 *active += zone->nr_active;
1005 *inactive += zone->nr_inactive;
1006 *free += zone->free_pages;
1010 void si_meminfo(struct sysinfo *val)
1012 val->totalram = totalram_pages;
1014 val->freeram = nr_free_pages();
1015 val->bufferram = nr_blockdev_pages();
1016 #ifdef CONFIG_HIGHMEM
1017 val->totalhigh = totalhigh_pages;
1018 val->freehigh = nr_free_highpages();
1023 val->mem_unit = PAGE_SIZE;
1024 if (vx_flags(VXF_VIRT_MEM, 0))
1025 vx_vsi_meminfo(val);
1028 EXPORT_SYMBOL(si_meminfo);
1031 void si_meminfo_node(struct sysinfo *val, int nid)
1033 pg_data_t *pgdat = NODE_DATA(nid);
1035 val->totalram = pgdat->node_present_pages;
1036 val->freeram = nr_free_pages_pgdat(pgdat);
1037 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1038 val->freehigh = pgdat->node_zones[ZONE_HIGHMEM].free_pages;
1039 val->mem_unit = PAGE_SIZE;
1043 #define K(x) ((x) << (PAGE_SHIFT-10))
1046 * Show free area list (used inside shift_scroll-lock stuff)
1047 * We also calculate the percentage fragmentation. We do this by counting the
1048 * memory on each free list with the exception of the first item on the list.
1050 void show_free_areas(void)
1052 struct page_state ps;
1053 int cpu, temperature;
1054 unsigned long active;
1055 unsigned long inactive;
1059 for_each_zone(zone) {
1061 printk("%s per-cpu:", zone->name);
1063 if (!zone->present_pages) {
1069 for (cpu = 0; cpu < NR_CPUS; ++cpu) {
1070 struct per_cpu_pageset *pageset;
1072 if (!cpu_possible(cpu))
1075 pageset = zone->pageset + cpu;
1077 for (temperature = 0; temperature < 2; temperature++)
1078 printk("cpu %d %s: low %d, high %d, batch %d\n",
1080 temperature ? "cold" : "hot",
1081 pageset->pcp[temperature].low,
1082 pageset->pcp[temperature].high,
1083 pageset->pcp[temperature].batch);
1087 get_page_state(&ps);
1088 get_zone_counts(&active, &inactive, &free);
1090 printk("\nFree pages: %11ukB (%ukB HighMem)\n",
1092 K(nr_free_highpages()));
1094 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu "
1095 "unstable:%lu free:%u slab:%lu mapped:%lu pagetables:%lu\n",
1104 ps.nr_page_table_pages);
1106 for_each_zone(zone) {
1120 K(zone->free_pages),
1123 K(zone->pages_high),
1125 K(zone->nr_inactive),
1126 K(zone->present_pages)
1128 printk("protections[]:");
1129 for (i = 0; i < MAX_NR_ZONES; i++)
1130 printk(" %lu", zone->protection[i]);
1134 for_each_zone(zone) {
1135 struct list_head *elem;
1136 unsigned long nr, flags, order, total = 0;
1139 printk("%s: ", zone->name);
1140 if (!zone->present_pages) {
1145 spin_lock_irqsave(&zone->lock, flags);
1146 for (order = 0; order < MAX_ORDER; order++) {
1148 list_for_each(elem, &zone->free_area[order].free_list)
1150 total += nr << order;
1151 printk("%lu*%lukB ", nr, K(1UL) << order);
1153 spin_unlock_irqrestore(&zone->lock, flags);
1154 printk("= %lukB\n", K(total));
1157 show_swap_cache_info();
1161 * Builds allocation fallback zone lists.
1163 static int __init build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist, int j, int k)
1170 zone = pgdat->node_zones + ZONE_HIGHMEM;
1171 if (zone->present_pages) {
1172 #ifndef CONFIG_HIGHMEM
1175 zonelist->zones[j++] = zone;
1178 zone = pgdat->node_zones + ZONE_NORMAL;
1179 if (zone->present_pages)
1180 zonelist->zones[j++] = zone;
1182 zone = pgdat->node_zones + ZONE_DMA;
1183 if (zone->present_pages)
1184 zonelist->zones[j++] = zone;
1191 #define MAX_NODE_LOAD (numnodes)
1192 static int __initdata node_load[MAX_NUMNODES];
1194 * find_next_best_node - find the next node that should appear in a given
1195 * node's fallback list
1196 * @node: node whose fallback list we're appending
1197 * @used_node_mask: pointer to the bitmap of already used nodes
1199 * We use a number of factors to determine which is the next node that should
1200 * appear on a given node's fallback list. The node should not have appeared
1201 * already in @node's fallback list, and it should be the next closest node
1202 * according to the distance array (which contains arbitrary distance values
1203 * from each node to each node in the system), and should also prefer nodes
1204 * with no CPUs, since presumably they'll have very little allocation pressure
1205 * on them otherwise.
1206 * It returns -1 if no node is found.
1208 static int __init find_next_best_node(int node, void *used_node_mask)
1211 int min_val = INT_MAX;
1214 for (i = 0; i < numnodes; i++) {
1217 /* Start from local node */
1218 n = (node+i)%numnodes;
1220 /* Don't want a node to appear more than once */
1221 if (test_bit(n, used_node_mask))
1224 /* Use the distance array to find the distance */
1225 val = node_distance(node, n);
1227 /* Give preference to headless and unused nodes */
1228 tmp = node_to_cpumask(n);
1229 if (!cpus_empty(tmp))
1230 val += PENALTY_FOR_NODE_WITH_CPUS;
1232 /* Slight preference for less loaded node */
1233 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
1234 val += node_load[n];
1236 if (val < min_val) {
1243 set_bit(best_node, used_node_mask);
1248 static void __init build_zonelists(pg_data_t *pgdat)
1250 int i, j, k, node, local_node;
1251 int prev_node, load;
1252 struct zonelist *zonelist;
1253 DECLARE_BITMAP(used_mask, MAX_NUMNODES);
1255 /* initialize zonelists */
1256 for (i = 0; i < MAX_NR_ZONES; i++) {
1257 zonelist = pgdat->node_zonelists + i;
1258 memset(zonelist, 0, sizeof(*zonelist));
1259 zonelist->zones[0] = NULL;
1262 /* NUMA-aware ordering of nodes */
1263 local_node = pgdat->node_id;
1265 prev_node = local_node;
1266 bitmap_zero(used_mask, MAX_NUMNODES);
1267 while ((node = find_next_best_node(local_node, used_mask)) >= 0) {
1269 * We don't want to pressure a particular node.
1270 * So adding penalty to the first node in same
1271 * distance group to make it round-robin.
1273 if (node_distance(local_node, node) !=
1274 node_distance(local_node, prev_node))
1275 node_load[node] += load;
1278 for (i = 0; i < MAX_NR_ZONES; i++) {
1279 zonelist = pgdat->node_zonelists + i;
1280 for (j = 0; zonelist->zones[j] != NULL; j++);
1283 if (i & __GFP_HIGHMEM)
1288 j = build_zonelists_node(NODE_DATA(node), zonelist, j, k);
1289 zonelist->zones[j] = NULL;
1294 #else /* CONFIG_NUMA */
1296 static void __init build_zonelists(pg_data_t *pgdat)
1298 int i, j, k, node, local_node;
1300 local_node = pgdat->node_id;
1301 for (i = 0; i < MAX_NR_ZONES; i++) {
1302 struct zonelist *zonelist;
1304 zonelist = pgdat->node_zonelists + i;
1305 memset(zonelist, 0, sizeof(*zonelist));
1309 if (i & __GFP_HIGHMEM)
1314 j = build_zonelists_node(pgdat, zonelist, j, k);
1316 * Now we build the zonelist so that it contains the zones
1317 * of all the other nodes.
1318 * We don't want to pressure a particular node, so when
1319 * building the zones for node N, we make sure that the
1320 * zones coming right after the local ones are those from
1321 * node N+1 (modulo N)
1323 for (node = local_node + 1; node < numnodes; node++)
1324 j = build_zonelists_node(NODE_DATA(node), zonelist, j, k);
1325 for (node = 0; node < local_node; node++)
1326 j = build_zonelists_node(NODE_DATA(node), zonelist, j, k);
1328 zonelist->zones[j] = NULL;
1332 #endif /* CONFIG_NUMA */
1334 void __init build_all_zonelists(void)
1338 for(i = 0 ; i < numnodes ; i++)
1339 build_zonelists(NODE_DATA(i));
1340 printk("Built %i zonelists\n", numnodes);
1344 * Helper functions to size the waitqueue hash table.
1345 * Essentially these want to choose hash table sizes sufficiently
1346 * large so that collisions trying to wait on pages are rare.
1347 * But in fact, the number of active page waitqueues on typical
1348 * systems is ridiculously low, less than 200. So this is even
1349 * conservative, even though it seems large.
1351 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
1352 * waitqueues, i.e. the size of the waitq table given the number of pages.
1354 #define PAGES_PER_WAITQUEUE 256
1356 static inline unsigned long wait_table_size(unsigned long pages)
1358 unsigned long size = 1;
1360 pages /= PAGES_PER_WAITQUEUE;
1362 while (size < pages)
1366 * Once we have dozens or even hundreds of threads sleeping
1367 * on IO we've got bigger problems than wait queue collision.
1368 * Limit the size of the wait table to a reasonable size.
1370 size = min(size, 4096UL);
1372 return max(size, 4UL);
1376 * This is an integer logarithm so that shifts can be used later
1377 * to extract the more random high bits from the multiplicative
1378 * hash function before the remainder is taken.
1380 static inline unsigned long wait_table_bits(unsigned long size)
1385 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
1387 static void __init calculate_zone_totalpages(struct pglist_data *pgdat,
1388 unsigned long *zones_size, unsigned long *zholes_size)
1390 unsigned long realtotalpages, totalpages = 0;
1393 for (i = 0; i < MAX_NR_ZONES; i++)
1394 totalpages += zones_size[i];
1395 pgdat->node_spanned_pages = totalpages;
1397 realtotalpages = totalpages;
1399 for (i = 0; i < MAX_NR_ZONES; i++)
1400 realtotalpages -= zholes_size[i];
1401 pgdat->node_present_pages = realtotalpages;
1402 printk("On node %d totalpages: %lu\n", pgdat->node_id, realtotalpages);
1407 * Initially all pages are reserved - free ones are freed
1408 * up by free_all_bootmem() once the early boot process is
1409 * done. Non-atomic initialization, single-pass.
1411 void __init memmap_init_zone(struct page *start, unsigned long size, int nid,
1412 unsigned long zone, unsigned long start_pfn)
1416 for (page = start; page < (start + size); page++) {
1417 set_page_zone(page, NODEZONE(nid, zone));
1418 set_page_count(page, 0);
1419 SetPageReserved(page);
1420 INIT_LIST_HEAD(&page->lru);
1421 #ifdef WANT_PAGE_VIRTUAL
1422 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
1423 if (zone != ZONE_HIGHMEM)
1424 set_page_address(page, __va(start_pfn << PAGE_SHIFT));
1430 #ifndef __HAVE_ARCH_MEMMAP_INIT
1431 #define memmap_init(start, size, nid, zone, start_pfn) \
1432 memmap_init_zone((start), (size), (nid), (zone), (start_pfn))
1436 * Set up the zone data structures:
1437 * - mark all pages reserved
1438 * - mark all memory queues empty
1439 * - clear the memory bitmaps
1441 static void __init free_area_init_core(struct pglist_data *pgdat,
1442 unsigned long *zones_size, unsigned long *zholes_size)
1445 const unsigned long zone_required_alignment = 1UL << (MAX_ORDER-1);
1446 int cpu, nid = pgdat->node_id;
1447 struct page *lmem_map = pgdat->node_mem_map;
1448 unsigned long zone_start_pfn = pgdat->node_start_pfn;
1450 pgdat->nr_zones = 0;
1451 init_waitqueue_head(&pgdat->kswapd_wait);
1453 for (j = 0; j < MAX_NR_ZONES; j++) {
1454 struct zone *zone = pgdat->node_zones + j;
1455 unsigned long size, realsize;
1456 unsigned long batch;
1458 zone_table[NODEZONE(nid, j)] = zone;
1459 realsize = size = zones_size[j];
1461 realsize -= zholes_size[j];
1463 zone->spanned_pages = size;
1464 zone->present_pages = realsize;
1465 zone->name = zone_names[j];
1466 spin_lock_init(&zone->lock);
1467 spin_lock_init(&zone->lru_lock);
1468 zone->zone_pgdat = pgdat;
1469 zone->free_pages = 0;
1471 zone->temp_priority = zone->prev_priority = DEF_PRIORITY;
1474 * The per-cpu-pages pools are set to around 1000th of the
1475 * size of the zone. But no more than 1/4 of a meg - there's
1476 * no point in going beyond the size of L2 cache.
1478 * OK, so we don't know how big the cache is. So guess.
1480 batch = zone->present_pages / 1024;
1481 if (batch * PAGE_SIZE > 256 * 1024)
1482 batch = (256 * 1024) / PAGE_SIZE;
1483 batch /= 4; /* We effectively *= 4 below */
1487 for (cpu = 0; cpu < NR_CPUS; cpu++) {
1488 struct per_cpu_pages *pcp;
1490 pcp = &zone->pageset[cpu].pcp[0]; /* hot */
1492 pcp->low = 2 * batch;
1493 pcp->high = 6 * batch;
1494 pcp->batch = 1 * batch;
1495 INIT_LIST_HEAD(&pcp->list);
1497 pcp = &zone->pageset[cpu].pcp[1]; /* cold */
1500 pcp->high = 2 * batch;
1501 pcp->batch = 1 * batch;
1502 INIT_LIST_HEAD(&pcp->list);
1504 printk(" %s zone: %lu pages, LIFO batch:%lu\n",
1505 zone_names[j], realsize, batch);
1506 INIT_LIST_HEAD(&zone->active_list);
1507 INIT_LIST_HEAD(&zone->inactive_list);
1508 atomic_set(&zone->nr_scan_active, 0);
1509 atomic_set(&zone->nr_scan_inactive, 0);
1510 zone->nr_active = 0;
1511 zone->nr_inactive = 0;
1516 * The per-page waitqueue mechanism uses hashed waitqueues
1519 zone->wait_table_size = wait_table_size(size);
1520 zone->wait_table_bits =
1521 wait_table_bits(zone->wait_table_size);
1522 zone->wait_table = (wait_queue_head_t *)
1523 alloc_bootmem_node(pgdat, zone->wait_table_size
1524 * sizeof(wait_queue_head_t));
1526 for(i = 0; i < zone->wait_table_size; ++i)
1527 init_waitqueue_head(zone->wait_table + i);
1529 pgdat->nr_zones = j+1;
1531 zone->zone_mem_map = lmem_map;
1532 zone->zone_start_pfn = zone_start_pfn;
1534 if ((zone_start_pfn) & (zone_required_alignment-1))
1535 printk("BUG: wrong zone alignment, it will crash\n");
1537 memmap_init(lmem_map, size, nid, j, zone_start_pfn);
1539 zone_start_pfn += size;
1542 for (i = 0; ; i++) {
1543 unsigned long bitmap_size;
1545 INIT_LIST_HEAD(&zone->free_area[i].free_list);
1546 if (i == MAX_ORDER-1) {
1547 zone->free_area[i].map = NULL;
1552 * Page buddy system uses "index >> (i+1)",
1553 * where "index" is at most "size-1".
1555 * The extra "+3" is to round down to byte
1556 * size (8 bits per byte assumption). Thus
1557 * we get "(size-1) >> (i+4)" as the last byte
1560 * The "+1" is because we want to round the
1561 * byte allocation up rather than down. So
1562 * we should have had a "+7" before we shifted
1563 * down by three. Also, we have to add one as
1564 * we actually _use_ the last bit (it's [0,n]
1565 * inclusive, not [0,n[).
1567 * So we actually had +7+1 before we shift
1568 * down by 3. But (n+8) >> 3 == (n >> 3) + 1
1569 * (modulo overflows, which we do not have).
1571 * Finally, we LONG_ALIGN because all bitmap
1572 * operations are on longs.
1574 bitmap_size = (size-1) >> (i+4);
1575 bitmap_size = LONG_ALIGN(bitmap_size+1);
1576 zone->free_area[i].map =
1577 (unsigned long *) alloc_bootmem_node(pgdat, bitmap_size);
1582 void __init free_area_init_node(int nid, struct pglist_data *pgdat,
1583 struct page *node_mem_map, unsigned long *zones_size,
1584 unsigned long node_start_pfn, unsigned long *zholes_size)
1588 pgdat->node_id = nid;
1589 pgdat->node_start_pfn = node_start_pfn;
1590 calculate_zone_totalpages(pgdat, zones_size, zholes_size);
1591 if (!node_mem_map) {
1592 size = (pgdat->node_spanned_pages + 1) * sizeof(struct page);
1593 node_mem_map = alloc_bootmem_node(pgdat, size);
1595 pgdat->node_mem_map = node_mem_map;
1597 free_area_init_core(pgdat, zones_size, zholes_size);
1600 #ifndef CONFIG_DISCONTIGMEM
1601 static bootmem_data_t contig_bootmem_data;
1602 struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data };
1604 EXPORT_SYMBOL(contig_page_data);
1606 void __init free_area_init(unsigned long *zones_size)
1608 free_area_init_node(0, &contig_page_data, NULL, zones_size,
1609 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
1610 mem_map = contig_page_data.node_mem_map;
1614 #ifdef CONFIG_PROC_FS
1616 #include <linux/seq_file.h>
1618 static void *frag_start(struct seq_file *m, loff_t *pos)
1623 for (pgdat = pgdat_list; pgdat && node; pgdat = pgdat->pgdat_next)
1629 static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
1631 pg_data_t *pgdat = (pg_data_t *)arg;
1634 return pgdat->pgdat_next;
1637 static void frag_stop(struct seq_file *m, void *arg)
1642 * This walks the freelist for each zone. Whilst this is slow, I'd rather
1643 * be slow here than slow down the fast path by keeping stats - mjbligh
1645 static int frag_show(struct seq_file *m, void *arg)
1647 pg_data_t *pgdat = (pg_data_t *)arg;
1649 struct zone *node_zones = pgdat->node_zones;
1650 unsigned long flags;
1653 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
1654 if (!zone->present_pages)
1657 spin_lock_irqsave(&zone->lock, flags);
1658 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
1659 for (order = 0; order < MAX_ORDER; ++order) {
1660 unsigned long nr_bufs = 0;
1661 struct list_head *elem;
1663 list_for_each(elem, &(zone->free_area[order].free_list))
1665 seq_printf(m, "%6lu ", nr_bufs);
1667 spin_unlock_irqrestore(&zone->lock, flags);
1673 struct seq_operations fragmentation_op = {
1674 .start = frag_start,
1680 static char *vmstat_text[] = {
1684 "nr_page_table_pages",
1709 "pgscan_kswapd_high",
1710 "pgscan_kswapd_normal",
1712 "pgscan_kswapd_dma",
1713 "pgscan_direct_high",
1714 "pgscan_direct_normal",
1715 "pgscan_direct_dma",
1720 "kswapd_inodesteal",
1727 static void *vmstat_start(struct seq_file *m, loff_t *pos)
1729 struct page_state *ps;
1731 if (*pos >= ARRAY_SIZE(vmstat_text))
1734 ps = kmalloc(sizeof(*ps), GFP_KERNEL);
1737 return ERR_PTR(-ENOMEM);
1738 get_full_page_state(ps);
1739 ps->pgpgin /= 2; /* sectors -> kbytes */
1741 return (unsigned long *)ps + *pos;
1744 static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
1747 if (*pos >= ARRAY_SIZE(vmstat_text))
1749 return (unsigned long *)m->private + *pos;
1752 static int vmstat_show(struct seq_file *m, void *arg)
1754 unsigned long *l = arg;
1755 unsigned long off = l - (unsigned long *)m->private;
1757 seq_printf(m, "%s %lu\n", vmstat_text[off], *l);
1761 static void vmstat_stop(struct seq_file *m, void *arg)
1767 struct seq_operations vmstat_op = {
1768 .start = vmstat_start,
1769 .next = vmstat_next,
1770 .stop = vmstat_stop,
1771 .show = vmstat_show,
1774 #endif /* CONFIG_PROC_FS */
1776 #ifdef CONFIG_HOTPLUG_CPU
1777 static int page_alloc_cpu_notify(struct notifier_block *self,
1778 unsigned long action, void *hcpu)
1780 int cpu = (unsigned long)hcpu;
1783 if (action == CPU_DEAD) {
1784 /* Drain local pagecache count. */
1785 count = &per_cpu(nr_pagecache_local, cpu);
1786 atomic_add(*count, &nr_pagecache);
1788 local_irq_disable();
1794 #endif /* CONFIG_HOTPLUG_CPU */
1796 void __init page_alloc_init(void)
1798 hotcpu_notifier(page_alloc_cpu_notify, 0);
1801 static unsigned long higherzone_val(struct zone *z, int max_zone,
1804 int z_idx = zone_idx(z);
1805 struct zone *higherzone;
1806 unsigned long pages;
1808 /* there is no higher zone to get a contribution from */
1809 if (z_idx == MAX_NR_ZONES-1)
1812 higherzone = &z->zone_pgdat->node_zones[z_idx+1];
1814 /* We always start with the higher zone's protection value */
1815 pages = higherzone->protection[alloc_type];
1818 * We get a lower-zone-protection contribution only if there are
1819 * pages in the higher zone and if we're not the highest zone
1820 * in the current zonelist. e.g., never happens for GFP_DMA. Happens
1821 * only for ZONE_DMA in a GFP_KERNEL allocation and happens for ZONE_DMA
1822 * and ZONE_NORMAL for a GFP_HIGHMEM allocation.
1824 if (higherzone->present_pages && z_idx < alloc_type)
1825 pages += higherzone->pages_low * sysctl_lower_zone_protection;
1831 * setup_per_zone_protection - called whenver min_free_kbytes or
1832 * sysctl_lower_zone_protection changes. Ensures that each zone
1833 * has a correct pages_protected value, so an adequate number of
1834 * pages are left in the zone after a successful __alloc_pages().
1836 * This algorithm is way confusing. I tries to keep the same behavior
1837 * as we had with the incremental min iterative algorithm.
1839 static void setup_per_zone_protection(void)
1841 struct pglist_data *pgdat;
1842 struct zone *zones, *zone;
1846 for_each_pgdat(pgdat) {
1847 zones = pgdat->node_zones;
1849 for (i = 0, max_zone = 0; i < MAX_NR_ZONES; i++)
1850 if (zones[i].present_pages)
1854 * For each of the different allocation types:
1855 * GFP_DMA -> GFP_KERNEL -> GFP_HIGHMEM
1857 for (i = 0; i < MAX_NR_ZONES; i++) {
1859 * For each of the zones:
1860 * ZONE_HIGHMEM -> ZONE_NORMAL -> ZONE_DMA
1862 for (j = MAX_NR_ZONES-1; j >= 0; j--) {
1866 * We never protect zones that don't have memory
1867 * in them (j>max_zone) or zones that aren't in
1868 * the zonelists for a certain type of
1869 * allocation (j>i). We have to assign these to
1870 * zero because the lower zones take
1871 * contributions from the higher zones.
1873 if (j > max_zone || j > i) {
1874 zone->protection[i] = 0;
1878 * The contribution of the next higher zone
1880 zone->protection[i] = higherzone_val(zone,
1882 zone->protection[i] += zone->pages_low;
1889 * setup_per_zone_pages_min - called when min_free_kbytes changes. Ensures
1890 * that the pages_{min,low,high} values for each zone are set correctly
1891 * with respect to min_free_kbytes.
1893 static void setup_per_zone_pages_min(void)
1895 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
1896 unsigned long lowmem_pages = 0;
1898 unsigned long flags;
1900 /* Calculate total number of !ZONE_HIGHMEM pages */
1901 for_each_zone(zone) {
1902 if (!is_highmem(zone))
1903 lowmem_pages += zone->present_pages;
1906 for_each_zone(zone) {
1907 spin_lock_irqsave(&zone->lru_lock, flags);
1908 if (is_highmem(zone)) {
1910 * Often, highmem doesn't need to reserve any pages.
1911 * But the pages_min/low/high values are also used for
1912 * batching up page reclaim activity so we need a
1913 * decent value here.
1917 min_pages = zone->present_pages / 1024;
1918 if (min_pages < SWAP_CLUSTER_MAX)
1919 min_pages = SWAP_CLUSTER_MAX;
1920 if (min_pages > 128)
1922 zone->pages_min = min_pages;
1924 /* if it's a lowmem zone, reserve a number of pages
1925 * proportionate to the zone's size.
1927 zone->pages_min = (pages_min * zone->present_pages) /
1931 zone->pages_low = zone->pages_min * 2;
1932 zone->pages_high = zone->pages_min * 3;
1933 spin_unlock_irqrestore(&zone->lru_lock, flags);
1938 * Initialise min_free_kbytes.
1940 * For small machines we want it small (128k min). For large machines
1941 * we want it large (16MB max). But it is not linear, because network
1942 * bandwidth does not increase linearly with machine size. We use
1944 * min_free_kbytes = sqrt(lowmem_kbytes)
1960 static int __init init_per_zone_pages_min(void)
1962 unsigned long lowmem_kbytes;
1964 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
1966 min_free_kbytes = int_sqrt(lowmem_kbytes);
1967 if (min_free_kbytes < 128)
1968 min_free_kbytes = 128;
1969 if (min_free_kbytes > 16384)
1970 min_free_kbytes = 16384;
1971 setup_per_zone_pages_min();
1972 setup_per_zone_protection();
1975 module_init(init_per_zone_pages_min)
1978 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
1979 * that we can call two helper functions whenever min_free_kbytes
1982 int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
1983 struct file *file, void __user *buffer, size_t *length)
1985 proc_dointvec(table, write, file, buffer, length);
1986 setup_per_zone_pages_min();
1987 setup_per_zone_protection();
1992 * lower_zone_protection_sysctl_handler - just a wrapper around
1993 * proc_dointvec() so that we can call setup_per_zone_protection()
1994 * whenever sysctl_lower_zone_protection changes.
1996 int lower_zone_protection_sysctl_handler(ctl_table *table, int write,
1997 struct file *file, void __user *buffer, size_t *length)
1999 proc_dointvec_minmax(table, write, file, buffer, length);
2000 setup_per_zone_protection();