4 * Manages VM statistics
5 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * Copyright (C) 2006 Silicon Graphics, Inc.,
9 * Christoph Lameter <christoph@lameter.com>
13 #include <linux/err.h>
14 #include <linux/module.h>
15 #include <linux/cpu.h>
16 #include <linux/vmstat.h>
17 #include <linux/sched.h>
19 #ifdef CONFIG_VM_EVENT_COUNTERS
20 DEFINE_PER_CPU(struct vm_event_state, vm_event_states) = {{0}};
21 EXPORT_PER_CPU_SYMBOL(vm_event_states);
23 static void sum_vm_events(unsigned long *ret, const struct cpumask *cpumask)
28 memset(ret, 0, NR_VM_EVENT_ITEMS * sizeof(unsigned long));
30 for_each_cpu_mask_nr(cpu, *cpumask) {
31 struct vm_event_state *this = &per_cpu(vm_event_states, cpu);
33 for (i = 0; i < NR_VM_EVENT_ITEMS; i++)
34 ret[i] += this->event[i];
39 * Accumulate the vm event counters across all CPUs.
40 * The result is unavoidably approximate - it can change
41 * during and after execution of this function.
43 void all_vm_events(unsigned long *ret)
46 sum_vm_events(ret, cpu_online_mask);
49 EXPORT_SYMBOL_GPL(all_vm_events);
53 * Fold the foreign cpu events into our own.
55 * This is adding to the events on one processor
56 * but keeps the global counts constant.
58 void vm_events_fold_cpu(int cpu)
60 struct vm_event_state *fold_state = &per_cpu(vm_event_states, cpu);
63 for (i = 0; i < NR_VM_EVENT_ITEMS; i++) {
64 count_vm_events(i, fold_state->event[i]);
65 fold_state->event[i] = 0;
68 #endif /* CONFIG_HOTPLUG */
70 #endif /* CONFIG_VM_EVENT_COUNTERS */
73 * Manage combined zone based / global counters
75 * vm_stat contains the global counters
77 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
78 EXPORT_SYMBOL(vm_stat);
82 static int calculate_threshold(struct zone *zone)
85 int mem; /* memory in 128 MB units */
88 * The threshold scales with the number of processors and the amount
89 * of memory per zone. More memory means that we can defer updates for
90 * longer, more processors could lead to more contention.
91 * fls() is used to have a cheap way of logarithmic scaling.
93 * Some sample thresholds:
95 * Threshold Processors (fls) Zonesize fls(mem+1)
96 * ------------------------------------------------------------------
113 * 125 1024 10 8-16 GB 8
114 * 125 1024 10 16-32 GB 9
117 mem = zone->present_pages >> (27 - PAGE_SHIFT);
119 threshold = 2 * fls(num_online_cpus()) * (1 + fls(mem));
122 * Maximum threshold is 125
124 threshold = min(125, threshold);
130 * Refresh the thresholds for each zone.
132 static void refresh_zone_stat_thresholds(void)
138 for_each_zone(zone) {
140 if (!zone->present_pages)
143 threshold = calculate_threshold(zone);
145 for_each_online_cpu(cpu)
146 zone_pcp(zone, cpu)->stat_threshold = threshold;
151 * For use when we know that interrupts are disabled.
153 void __mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
156 struct per_cpu_pageset *pcp = zone_pcp(zone, smp_processor_id());
157 s8 *p = pcp->vm_stat_diff + item;
162 if (unlikely(x > pcp->stat_threshold || x < -pcp->stat_threshold)) {
163 zone_page_state_add(x, zone, item);
168 EXPORT_SYMBOL(__mod_zone_page_state);
171 * For an unknown interrupt state
173 void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
178 local_irq_save(flags);
179 __mod_zone_page_state(zone, item, delta);
180 local_irq_restore(flags);
182 EXPORT_SYMBOL(mod_zone_page_state);
185 * Optimized increment and decrement functions.
187 * These are only for a single page and therefore can take a struct page *
188 * argument instead of struct zone *. This allows the inclusion of the code
189 * generated for page_zone(page) into the optimized functions.
191 * No overflow check is necessary and therefore the differential can be
192 * incremented or decremented in place which may allow the compilers to
193 * generate better code.
194 * The increment or decrement is known and therefore one boundary check can
197 * NOTE: These functions are very performance sensitive. Change only
200 * Some processors have inc/dec instructions that are atomic vs an interrupt.
201 * However, the code must first determine the differential location in a zone
202 * based on the processor number and then inc/dec the counter. There is no
203 * guarantee without disabling preemption that the processor will not change
204 * in between and therefore the atomicity vs. interrupt cannot be exploited
205 * in a useful way here.
207 void __inc_zone_state(struct zone *zone, enum zone_stat_item item)
209 struct per_cpu_pageset *pcp = zone_pcp(zone, smp_processor_id());
210 s8 *p = pcp->vm_stat_diff + item;
214 if (unlikely(*p > pcp->stat_threshold)) {
215 int overstep = pcp->stat_threshold / 2;
217 zone_page_state_add(*p + overstep, zone, item);
222 void __inc_zone_page_state(struct page *page, enum zone_stat_item item)
224 __inc_zone_state(page_zone(page), item);
226 EXPORT_SYMBOL(__inc_zone_page_state);
228 void __dec_zone_state(struct zone *zone, enum zone_stat_item item)
230 struct per_cpu_pageset *pcp = zone_pcp(zone, smp_processor_id());
231 s8 *p = pcp->vm_stat_diff + item;
235 if (unlikely(*p < - pcp->stat_threshold)) {
236 int overstep = pcp->stat_threshold / 2;
238 zone_page_state_add(*p - overstep, zone, item);
243 void __dec_zone_page_state(struct page *page, enum zone_stat_item item)
245 __dec_zone_state(page_zone(page), item);
247 EXPORT_SYMBOL(__dec_zone_page_state);
249 void inc_zone_state(struct zone *zone, enum zone_stat_item item)
253 local_irq_save(flags);
254 __inc_zone_state(zone, item);
255 local_irq_restore(flags);
258 void inc_zone_page_state(struct page *page, enum zone_stat_item item)
263 zone = page_zone(page);
264 local_irq_save(flags);
265 __inc_zone_state(zone, item);
266 local_irq_restore(flags);
268 EXPORT_SYMBOL(inc_zone_page_state);
270 void dec_zone_page_state(struct page *page, enum zone_stat_item item)
274 local_irq_save(flags);
275 __dec_zone_page_state(page, item);
276 local_irq_restore(flags);
278 EXPORT_SYMBOL(dec_zone_page_state);
281 * Update the zone counters for one cpu.
283 * The cpu specified must be either the current cpu or a processor that
284 * is not online. If it is the current cpu then the execution thread must
285 * be pinned to the current cpu.
287 * Note that refresh_cpu_vm_stats strives to only access
288 * node local memory. The per cpu pagesets on remote zones are placed
289 * in the memory local to the processor using that pageset. So the
290 * loop over all zones will access a series of cachelines local to
293 * The call to zone_page_state_add updates the cachelines with the
294 * statistics in the remote zone struct as well as the global cachelines
295 * with the global counters. These could cause remote node cache line
296 * bouncing and will have to be only done when necessary.
298 void refresh_cpu_vm_stats(int cpu)
302 int global_diff[NR_VM_ZONE_STAT_ITEMS] = { 0, };
304 for_each_zone(zone) {
305 struct per_cpu_pageset *p;
307 if (!populated_zone(zone))
310 p = zone_pcp(zone, cpu);
312 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
313 if (p->vm_stat_diff[i]) {
317 local_irq_save(flags);
318 v = p->vm_stat_diff[i];
319 p->vm_stat_diff[i] = 0;
320 local_irq_restore(flags);
321 atomic_long_add(v, &zone->vm_stat[i]);
324 /* 3 seconds idle till flush */
331 * Deal with draining the remote pageset of this
334 * Check if there are pages remaining in this pageset
335 * if not then there is nothing to expire.
337 if (!p->expire || !p->pcp.count)
341 * We never drain zones local to this processor.
343 if (zone_to_nid(zone) == numa_node_id()) {
353 drain_zone_pages(zone, &p->pcp);
357 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
359 atomic_long_add(global_diff[i], &vm_stat[i]);
366 * zonelist = the list of zones passed to the allocator
367 * z = the zone from which the allocation occurred.
369 * Must be called with interrupts disabled.
371 void zone_statistics(struct zone *preferred_zone, struct zone *z)
373 if (z->zone_pgdat == preferred_zone->zone_pgdat) {
374 __inc_zone_state(z, NUMA_HIT);
376 __inc_zone_state(z, NUMA_MISS);
377 __inc_zone_state(preferred_zone, NUMA_FOREIGN);
379 if (z->node == numa_node_id())
380 __inc_zone_state(z, NUMA_LOCAL);
382 __inc_zone_state(z, NUMA_OTHER);
386 #ifdef CONFIG_PROC_FS
387 #include <linux/proc_fs.h>
388 #include <linux/seq_file.h>
390 static char * const migratetype_names[MIGRATE_TYPES] = {
398 static void *frag_start(struct seq_file *m, loff_t *pos)
402 for (pgdat = first_online_pgdat();
404 pgdat = next_online_pgdat(pgdat))
410 static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
412 pg_data_t *pgdat = (pg_data_t *)arg;
415 return next_online_pgdat(pgdat);
418 static void frag_stop(struct seq_file *m, void *arg)
422 /* Walk all the zones in a node and print using a callback */
423 static void walk_zones_in_node(struct seq_file *m, pg_data_t *pgdat,
424 void (*print)(struct seq_file *m, pg_data_t *, struct zone *))
427 struct zone *node_zones = pgdat->node_zones;
430 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
431 if (!populated_zone(zone))
434 spin_lock_irqsave(&zone->lock, flags);
435 print(m, pgdat, zone);
436 spin_unlock_irqrestore(&zone->lock, flags);
440 static void frag_show_print(struct seq_file *m, pg_data_t *pgdat,
445 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
446 for (order = 0; order < MAX_ORDER; ++order)
447 seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
452 * This walks the free areas for each zone.
454 static int frag_show(struct seq_file *m, void *arg)
456 pg_data_t *pgdat = (pg_data_t *)arg;
457 walk_zones_in_node(m, pgdat, frag_show_print);
461 static void pagetypeinfo_showfree_print(struct seq_file *m,
462 pg_data_t *pgdat, struct zone *zone)
466 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++) {
467 seq_printf(m, "Node %4d, zone %8s, type %12s ",
470 migratetype_names[mtype]);
471 for (order = 0; order < MAX_ORDER; ++order) {
472 unsigned long freecount = 0;
473 struct free_area *area;
474 struct list_head *curr;
476 area = &(zone->free_area[order]);
478 list_for_each(curr, &area->free_list[mtype])
480 seq_printf(m, "%6lu ", freecount);
486 /* Print out the free pages at each order for each migatetype */
487 static int pagetypeinfo_showfree(struct seq_file *m, void *arg)
490 pg_data_t *pgdat = (pg_data_t *)arg;
493 seq_printf(m, "%-43s ", "Free pages count per migrate type at order");
494 for (order = 0; order < MAX_ORDER; ++order)
495 seq_printf(m, "%6d ", order);
498 walk_zones_in_node(m, pgdat, pagetypeinfo_showfree_print);
503 static void pagetypeinfo_showblockcount_print(struct seq_file *m,
504 pg_data_t *pgdat, struct zone *zone)
508 unsigned long start_pfn = zone->zone_start_pfn;
509 unsigned long end_pfn = start_pfn + zone->spanned_pages;
510 unsigned long count[MIGRATE_TYPES] = { 0, };
512 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
518 page = pfn_to_page(pfn);
519 #ifdef CONFIG_ARCH_FLATMEM_HAS_HOLES
521 * Ordinarily, memory holes in flatmem still have a valid
522 * memmap for the PFN range. However, an architecture for
523 * embedded systems (e.g. ARM) can free up the memmap backing
524 * holes to save memory on the assumption the memmap is
525 * never used. The page_zone linkages are then broken even
526 * though pfn_valid() returns true. Skip the page if the
527 * linkages are broken. Even if this test passed, the impact
528 * is that the counters for the movable type are off but
529 * fragmentation monitoring is likely meaningless on small
532 if (page_zone(page) != zone)
535 mtype = get_pageblock_migratetype(page);
537 if (mtype < MIGRATE_TYPES)
542 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
543 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
544 seq_printf(m, "%12lu ", count[mtype]);
548 /* Print out the free pages at each order for each migratetype */
549 static int pagetypeinfo_showblockcount(struct seq_file *m, void *arg)
552 pg_data_t *pgdat = (pg_data_t *)arg;
554 seq_printf(m, "\n%-23s", "Number of blocks type ");
555 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
556 seq_printf(m, "%12s ", migratetype_names[mtype]);
558 walk_zones_in_node(m, pgdat, pagetypeinfo_showblockcount_print);
564 * This prints out statistics in relation to grouping pages by mobility.
565 * It is expensive to collect so do not constantly read the file.
567 static int pagetypeinfo_show(struct seq_file *m, void *arg)
569 pg_data_t *pgdat = (pg_data_t *)arg;
571 /* check memoryless node */
572 if (!node_state(pgdat->node_id, N_HIGH_MEMORY))
575 seq_printf(m, "Page block order: %d\n", pageblock_order);
576 seq_printf(m, "Pages per block: %lu\n", pageblock_nr_pages);
578 pagetypeinfo_showfree(m, pgdat);
579 pagetypeinfo_showblockcount(m, pgdat);
584 static const struct seq_operations fragmentation_op = {
591 static int fragmentation_open(struct inode *inode, struct file *file)
593 return seq_open(file, &fragmentation_op);
596 static const struct file_operations fragmentation_file_operations = {
597 .open = fragmentation_open,
600 .release = seq_release,
603 static const struct seq_operations pagetypeinfo_op = {
607 .show = pagetypeinfo_show,
610 static int pagetypeinfo_open(struct inode *inode, struct file *file)
612 return seq_open(file, &pagetypeinfo_op);
615 static const struct file_operations pagetypeinfo_file_ops = {
616 .open = pagetypeinfo_open,
619 .release = seq_release,
622 #ifdef CONFIG_ZONE_DMA
623 #define TEXT_FOR_DMA(xx) xx "_dma",
625 #define TEXT_FOR_DMA(xx)
628 #ifdef CONFIG_ZONE_DMA32
629 #define TEXT_FOR_DMA32(xx) xx "_dma32",
631 #define TEXT_FOR_DMA32(xx)
634 #ifdef CONFIG_HIGHMEM
635 #define TEXT_FOR_HIGHMEM(xx) xx "_high",
637 #define TEXT_FOR_HIGHMEM(xx)
640 #define TEXTS_FOR_ZONES(xx) TEXT_FOR_DMA(xx) TEXT_FOR_DMA32(xx) xx "_normal", \
641 TEXT_FOR_HIGHMEM(xx) xx "_movable",
643 static const char * const vmstat_text[] = {
644 /* Zoned VM counters */
650 #ifdef CONFIG_UNEVICTABLE_LRU
659 "nr_slab_reclaimable",
660 "nr_slab_unreclaimable",
661 "nr_page_table_pages",
676 #ifdef CONFIG_VM_EVENT_COUNTERS
682 TEXTS_FOR_ZONES("pgalloc")
691 TEXTS_FOR_ZONES("pgrefill")
692 TEXTS_FOR_ZONES("pgsteal")
693 TEXTS_FOR_ZONES("pgscan_kswapd")
694 TEXTS_FOR_ZONES("pgscan_direct")
704 #ifdef CONFIG_HUGETLB_PAGE
705 "htlb_buddy_alloc_success",
706 "htlb_buddy_alloc_fail",
708 #ifdef CONFIG_UNEVICTABLE_LRU
709 "unevictable_pgs_culled",
710 "unevictable_pgs_scanned",
711 "unevictable_pgs_rescued",
712 "unevictable_pgs_mlocked",
713 "unevictable_pgs_munlocked",
714 "unevictable_pgs_cleared",
715 "unevictable_pgs_stranded",
716 "unevictable_pgs_mlockfreed",
721 static void zoneinfo_show_print(struct seq_file *m, pg_data_t *pgdat,
725 seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
731 "\n scanned %lu (aa: %lu ia: %lu af: %lu if: %lu)"
734 zone_page_state(zone, NR_FREE_PAGES),
739 zone->lru[LRU_ACTIVE_ANON].nr_scan,
740 zone->lru[LRU_INACTIVE_ANON].nr_scan,
741 zone->lru[LRU_ACTIVE_FILE].nr_scan,
742 zone->lru[LRU_INACTIVE_FILE].nr_scan,
744 zone->present_pages);
746 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
747 seq_printf(m, "\n %-12s %lu", vmstat_text[i],
748 zone_page_state(zone, i));
751 "\n protection: (%lu",
752 zone->lowmem_reserve[0]);
753 for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
754 seq_printf(m, ", %lu", zone->lowmem_reserve[i]);
758 for_each_online_cpu(i) {
759 struct per_cpu_pageset *pageset;
761 pageset = zone_pcp(zone, i);
772 seq_printf(m, "\n vm stats threshold: %d",
773 pageset->stat_threshold);
777 "\n all_unreclaimable: %u"
778 "\n prev_priority: %i"
780 "\n inactive_ratio: %u",
781 zone_is_all_unreclaimable(zone),
783 zone->zone_start_pfn,
784 zone->inactive_ratio);
789 * Output information about zones in @pgdat.
791 static int zoneinfo_show(struct seq_file *m, void *arg)
793 pg_data_t *pgdat = (pg_data_t *)arg;
794 walk_zones_in_node(m, pgdat, zoneinfo_show_print);
798 static const struct seq_operations zoneinfo_op = {
799 .start = frag_start, /* iterate over all zones. The same as in
803 .show = zoneinfo_show,
806 static int zoneinfo_open(struct inode *inode, struct file *file)
808 return seq_open(file, &zoneinfo_op);
811 static const struct file_operations proc_zoneinfo_file_operations = {
812 .open = zoneinfo_open,
815 .release = seq_release,
818 static void *vmstat_start(struct seq_file *m, loff_t *pos)
821 #ifdef CONFIG_VM_EVENT_COUNTERS
826 if (*pos >= ARRAY_SIZE(vmstat_text))
829 #ifdef CONFIG_VM_EVENT_COUNTERS
830 v = kmalloc(NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long)
831 + sizeof(struct vm_event_state), GFP_KERNEL);
833 v = kmalloc(NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long),
838 return ERR_PTR(-ENOMEM);
839 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
840 v[i] = global_page_state(i);
841 #ifdef CONFIG_VM_EVENT_COUNTERS
842 e = v + NR_VM_ZONE_STAT_ITEMS;
844 e[PGPGIN] /= 2; /* sectors -> kbytes */
850 static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
853 if (*pos >= ARRAY_SIZE(vmstat_text))
855 return (unsigned long *)m->private + *pos;
858 static int vmstat_show(struct seq_file *m, void *arg)
860 unsigned long *l = arg;
861 unsigned long off = l - (unsigned long *)m->private;
863 seq_printf(m, "%s %lu\n", vmstat_text[off], *l);
867 static void vmstat_stop(struct seq_file *m, void *arg)
873 static const struct seq_operations vmstat_op = {
874 .start = vmstat_start,
880 static int vmstat_open(struct inode *inode, struct file *file)
882 return seq_open(file, &vmstat_op);
885 static const struct file_operations proc_vmstat_file_operations = {
889 .release = seq_release,
891 #endif /* CONFIG_PROC_FS */
894 static DEFINE_PER_CPU(struct delayed_work, vmstat_work);
895 int sysctl_stat_interval __read_mostly = HZ;
897 static void vmstat_update(struct work_struct *w)
899 refresh_cpu_vm_stats(smp_processor_id());
900 schedule_delayed_work(&__get_cpu_var(vmstat_work),
901 sysctl_stat_interval);
904 static void __cpuinit start_cpu_timer(int cpu)
906 struct delayed_work *vmstat_work = &per_cpu(vmstat_work, cpu);
908 INIT_DELAYED_WORK_DEFERRABLE(vmstat_work, vmstat_update);
909 schedule_delayed_work_on(cpu, vmstat_work, HZ + cpu);
913 * Use the cpu notifier to insure that the thresholds are recalculated
916 static int __cpuinit vmstat_cpuup_callback(struct notifier_block *nfb,
917 unsigned long action,
920 long cpu = (long)hcpu;
924 case CPU_ONLINE_FROZEN:
925 start_cpu_timer(cpu);
927 case CPU_DOWN_PREPARE:
928 case CPU_DOWN_PREPARE_FROZEN:
929 cancel_rearming_delayed_work(&per_cpu(vmstat_work, cpu));
930 per_cpu(vmstat_work, cpu).work.func = NULL;
932 case CPU_DOWN_FAILED:
933 case CPU_DOWN_FAILED_FROZEN:
934 start_cpu_timer(cpu);
937 case CPU_DEAD_FROZEN:
938 refresh_zone_stat_thresholds();
946 static struct notifier_block __cpuinitdata vmstat_notifier =
947 { &vmstat_cpuup_callback, NULL, 0 };
950 static int __init setup_vmstat(void)
955 refresh_zone_stat_thresholds();
956 register_cpu_notifier(&vmstat_notifier);
958 for_each_online_cpu(cpu)
959 start_cpu_timer(cpu);
961 #ifdef CONFIG_PROC_FS
962 proc_create("buddyinfo", S_IRUGO, NULL, &fragmentation_file_operations);
963 proc_create("pagetypeinfo", S_IRUGO, NULL, &pagetypeinfo_file_ops);
964 proc_create("vmstat", S_IRUGO, NULL, &proc_vmstat_file_operations);
965 proc_create("zoneinfo", S_IRUGO, NULL, &proc_zoneinfo_file_operations);
969 module_init(setup_vmstat)