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(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_populated_zone(zone) {
139 unsigned long max_drift, tolerate_drift;
141 threshold = calculate_threshold(zone);
143 for_each_online_cpu(cpu)
144 zone_pcp(zone, cpu)->stat_threshold = threshold;
147 * Only set percpu_drift_mark if there is a danger that
148 * NR_FREE_PAGES reports the low watermark is ok when in fact
149 * the min watermark could be breached by an allocation
151 tolerate_drift = low_wmark_pages(zone) - min_wmark_pages(zone);
152 max_drift = num_online_cpus() * threshold;
153 if (max_drift > tolerate_drift)
154 zone->percpu_drift_mark = high_wmark_pages(zone) +
160 * For use when we know that interrupts are disabled.
162 void __mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
165 struct per_cpu_pageset *pcp = zone_pcp(zone, smp_processor_id());
166 s8 *p = pcp->vm_stat_diff + item;
171 if (unlikely(x > pcp->stat_threshold || x < -pcp->stat_threshold)) {
172 zone_page_state_add(x, zone, item);
177 EXPORT_SYMBOL(__mod_zone_page_state);
180 * For an unknown interrupt state
182 void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
187 local_irq_save(flags);
188 __mod_zone_page_state(zone, item, delta);
189 local_irq_restore(flags);
191 EXPORT_SYMBOL(mod_zone_page_state);
194 * Optimized increment and decrement functions.
196 * These are only for a single page and therefore can take a struct page *
197 * argument instead of struct zone *. This allows the inclusion of the code
198 * generated for page_zone(page) into the optimized functions.
200 * No overflow check is necessary and therefore the differential can be
201 * incremented or decremented in place which may allow the compilers to
202 * generate better code.
203 * The increment or decrement is known and therefore one boundary check can
206 * NOTE: These functions are very performance sensitive. Change only
209 * Some processors have inc/dec instructions that are atomic vs an interrupt.
210 * However, the code must first determine the differential location in a zone
211 * based on the processor number and then inc/dec the counter. There is no
212 * guarantee without disabling preemption that the processor will not change
213 * in between and therefore the atomicity vs. interrupt cannot be exploited
214 * in a useful way here.
216 void __inc_zone_state(struct zone *zone, enum zone_stat_item item)
218 struct per_cpu_pageset *pcp = zone_pcp(zone, smp_processor_id());
219 s8 *p = pcp->vm_stat_diff + item;
223 if (unlikely(*p > pcp->stat_threshold)) {
224 int overstep = pcp->stat_threshold / 2;
226 zone_page_state_add(*p + overstep, zone, item);
231 void __inc_zone_page_state(struct page *page, enum zone_stat_item item)
233 __inc_zone_state(page_zone(page), item);
235 EXPORT_SYMBOL(__inc_zone_page_state);
237 void __dec_zone_state(struct zone *zone, enum zone_stat_item item)
239 struct per_cpu_pageset *pcp = zone_pcp(zone, smp_processor_id());
240 s8 *p = pcp->vm_stat_diff + item;
244 if (unlikely(*p < - pcp->stat_threshold)) {
245 int overstep = pcp->stat_threshold / 2;
247 zone_page_state_add(*p - overstep, zone, item);
252 void __dec_zone_page_state(struct page *page, enum zone_stat_item item)
254 __dec_zone_state(page_zone(page), item);
256 EXPORT_SYMBOL(__dec_zone_page_state);
258 void inc_zone_state(struct zone *zone, enum zone_stat_item item)
262 local_irq_save(flags);
263 __inc_zone_state(zone, item);
264 local_irq_restore(flags);
267 void inc_zone_page_state(struct page *page, enum zone_stat_item item)
272 zone = page_zone(page);
273 local_irq_save(flags);
274 __inc_zone_state(zone, item);
275 local_irq_restore(flags);
277 EXPORT_SYMBOL(inc_zone_page_state);
279 void dec_zone_page_state(struct page *page, enum zone_stat_item item)
283 local_irq_save(flags);
284 __dec_zone_page_state(page, item);
285 local_irq_restore(flags);
287 EXPORT_SYMBOL(dec_zone_page_state);
290 * Update the zone counters for one cpu.
292 * The cpu specified must be either the current cpu or a processor that
293 * is not online. If it is the current cpu then the execution thread must
294 * be pinned to the current cpu.
296 * Note that refresh_cpu_vm_stats strives to only access
297 * node local memory. The per cpu pagesets on remote zones are placed
298 * in the memory local to the processor using that pageset. So the
299 * loop over all zones will access a series of cachelines local to
302 * The call to zone_page_state_add updates the cachelines with the
303 * statistics in the remote zone struct as well as the global cachelines
304 * with the global counters. These could cause remote node cache line
305 * bouncing and will have to be only done when necessary.
307 void refresh_cpu_vm_stats(int cpu)
311 int global_diff[NR_VM_ZONE_STAT_ITEMS] = { 0, };
313 for_each_populated_zone(zone) {
314 struct per_cpu_pageset *p;
316 p = zone_pcp(zone, cpu);
318 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
319 if (p->vm_stat_diff[i]) {
323 local_irq_save(flags);
324 v = p->vm_stat_diff[i];
325 p->vm_stat_diff[i] = 0;
326 local_irq_restore(flags);
327 atomic_long_add(v, &zone->vm_stat[i]);
330 /* 3 seconds idle till flush */
337 * Deal with draining the remote pageset of this
340 * Check if there are pages remaining in this pageset
341 * if not then there is nothing to expire.
343 if (!p->expire || !p->pcp.count)
347 * We never drain zones local to this processor.
349 if (zone_to_nid(zone) == numa_node_id()) {
359 drain_zone_pages(zone, &p->pcp);
363 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
365 atomic_long_add(global_diff[i], &vm_stat[i]);
372 * zonelist = the list of zones passed to the allocator
373 * z = the zone from which the allocation occurred.
375 * Must be called with interrupts disabled.
377 void zone_statistics(struct zone *preferred_zone, struct zone *z)
379 if (z->zone_pgdat == preferred_zone->zone_pgdat) {
380 __inc_zone_state(z, NUMA_HIT);
382 __inc_zone_state(z, NUMA_MISS);
383 __inc_zone_state(preferred_zone, NUMA_FOREIGN);
385 if (z->node == numa_node_id())
386 __inc_zone_state(z, NUMA_LOCAL);
388 __inc_zone_state(z, NUMA_OTHER);
392 #ifdef CONFIG_PROC_FS
393 #include <linux/proc_fs.h>
394 #include <linux/seq_file.h>
396 static char * const migratetype_names[MIGRATE_TYPES] = {
404 static void *frag_start(struct seq_file *m, loff_t *pos)
408 for (pgdat = first_online_pgdat();
410 pgdat = next_online_pgdat(pgdat))
416 static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
418 pg_data_t *pgdat = (pg_data_t *)arg;
421 return next_online_pgdat(pgdat);
424 static void frag_stop(struct seq_file *m, void *arg)
428 /* Walk all the zones in a node and print using a callback */
429 static void walk_zones_in_node(struct seq_file *m, pg_data_t *pgdat,
430 void (*print)(struct seq_file *m, pg_data_t *, struct zone *))
433 struct zone *node_zones = pgdat->node_zones;
436 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
437 if (!populated_zone(zone))
440 spin_lock_irqsave(&zone->lock, flags);
441 print(m, pgdat, zone);
442 spin_unlock_irqrestore(&zone->lock, flags);
446 static void frag_show_print(struct seq_file *m, pg_data_t *pgdat,
451 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
452 for (order = 0; order < MAX_ORDER; ++order)
453 seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
458 * This walks the free areas for each zone.
460 static int frag_show(struct seq_file *m, void *arg)
462 pg_data_t *pgdat = (pg_data_t *)arg;
463 walk_zones_in_node(m, pgdat, frag_show_print);
467 static void pagetypeinfo_showfree_print(struct seq_file *m,
468 pg_data_t *pgdat, struct zone *zone)
472 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++) {
473 seq_printf(m, "Node %4d, zone %8s, type %12s ",
476 migratetype_names[mtype]);
477 for (order = 0; order < MAX_ORDER; ++order) {
478 unsigned long freecount = 0;
479 struct free_area *area;
480 struct list_head *curr;
482 area = &(zone->free_area[order]);
484 list_for_each(curr, &area->free_list[mtype])
486 seq_printf(m, "%6lu ", freecount);
492 /* Print out the free pages at each order for each migatetype */
493 static int pagetypeinfo_showfree(struct seq_file *m, void *arg)
496 pg_data_t *pgdat = (pg_data_t *)arg;
499 seq_printf(m, "%-43s ", "Free pages count per migrate type at order");
500 for (order = 0; order < MAX_ORDER; ++order)
501 seq_printf(m, "%6d ", order);
504 walk_zones_in_node(m, pgdat, pagetypeinfo_showfree_print);
509 static void pagetypeinfo_showblockcount_print(struct seq_file *m,
510 pg_data_t *pgdat, struct zone *zone)
514 unsigned long start_pfn = zone->zone_start_pfn;
515 unsigned long end_pfn = start_pfn + zone->spanned_pages;
516 unsigned long count[MIGRATE_TYPES] = { 0, };
518 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
524 page = pfn_to_page(pfn);
526 /* Watch for unexpected holes punched in the memmap */
527 if (!memmap_valid_within(pfn, page, zone))
530 mtype = get_pageblock_migratetype(page);
532 if (mtype < MIGRATE_TYPES)
537 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
538 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
539 seq_printf(m, "%12lu ", count[mtype]);
543 /* Print out the free pages at each order for each migratetype */
544 static int pagetypeinfo_showblockcount(struct seq_file *m, void *arg)
547 pg_data_t *pgdat = (pg_data_t *)arg;
549 seq_printf(m, "\n%-23s", "Number of blocks type ");
550 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
551 seq_printf(m, "%12s ", migratetype_names[mtype]);
553 walk_zones_in_node(m, pgdat, pagetypeinfo_showblockcount_print);
559 * This prints out statistics in relation to grouping pages by mobility.
560 * It is expensive to collect so do not constantly read the file.
562 static int pagetypeinfo_show(struct seq_file *m, void *arg)
564 pg_data_t *pgdat = (pg_data_t *)arg;
566 /* check memoryless node */
567 if (!node_state(pgdat->node_id, N_HIGH_MEMORY))
570 seq_printf(m, "Page block order: %d\n", pageblock_order);
571 seq_printf(m, "Pages per block: %lu\n", pageblock_nr_pages);
573 pagetypeinfo_showfree(m, pgdat);
574 pagetypeinfo_showblockcount(m, pgdat);
579 static const struct seq_operations fragmentation_op = {
586 static int fragmentation_open(struct inode *inode, struct file *file)
588 return seq_open(file, &fragmentation_op);
591 static const struct file_operations fragmentation_file_operations = {
592 .open = fragmentation_open,
595 .release = seq_release,
598 static const struct seq_operations pagetypeinfo_op = {
602 .show = pagetypeinfo_show,
605 static int pagetypeinfo_open(struct inode *inode, struct file *file)
607 return seq_open(file, &pagetypeinfo_op);
610 static const struct file_operations pagetypeinfo_file_ops = {
611 .open = pagetypeinfo_open,
614 .release = seq_release,
617 #ifdef CONFIG_ZONE_DMA
618 #define TEXT_FOR_DMA(xx) xx "_dma",
620 #define TEXT_FOR_DMA(xx)
623 #ifdef CONFIG_ZONE_DMA32
624 #define TEXT_FOR_DMA32(xx) xx "_dma32",
626 #define TEXT_FOR_DMA32(xx)
629 #ifdef CONFIG_HIGHMEM
630 #define TEXT_FOR_HIGHMEM(xx) xx "_high",
632 #define TEXT_FOR_HIGHMEM(xx)
635 #define TEXTS_FOR_ZONES(xx) TEXT_FOR_DMA(xx) TEXT_FOR_DMA32(xx) xx "_normal", \
636 TEXT_FOR_HIGHMEM(xx) xx "_movable",
638 static const char * const vmstat_text[] = {
639 /* Zoned VM counters */
652 "nr_slab_reclaimable",
653 "nr_slab_unreclaimable",
654 "nr_page_table_pages",
672 #ifdef CONFIG_VM_EVENT_COUNTERS
678 TEXTS_FOR_ZONES("pgalloc")
687 TEXTS_FOR_ZONES("pgrefill")
688 TEXTS_FOR_ZONES("pgsteal")
689 TEXTS_FOR_ZONES("pgscan_kswapd")
690 TEXTS_FOR_ZONES("pgscan_direct")
693 "zone_reclaim_failed",
703 #ifdef CONFIG_HUGETLB_PAGE
704 "htlb_buddy_alloc_success",
705 "htlb_buddy_alloc_fail",
707 "unevictable_pgs_culled",
708 "unevictable_pgs_scanned",
709 "unevictable_pgs_rescued",
710 "unevictable_pgs_mlocked",
711 "unevictable_pgs_munlocked",
712 "unevictable_pgs_cleared",
713 "unevictable_pgs_stranded",
714 "unevictable_pgs_mlockfreed",
718 static void zoneinfo_show_print(struct seq_file *m, pg_data_t *pgdat,
722 seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
731 zone_nr_free_pages(zone),
732 min_wmark_pages(zone),
733 low_wmark_pages(zone),
734 high_wmark_pages(zone),
737 zone->present_pages);
739 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
740 seq_printf(m, "\n %-12s %lu", vmstat_text[i],
741 zone_page_state(zone, i));
744 "\n protection: (%lu",
745 zone->lowmem_reserve[0]);
746 for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
747 seq_printf(m, ", %lu", zone->lowmem_reserve[i]);
751 for_each_online_cpu(i) {
752 struct per_cpu_pageset *pageset;
754 pageset = zone_pcp(zone, i);
765 seq_printf(m, "\n vm stats threshold: %d",
766 pageset->stat_threshold);
770 "\n all_unreclaimable: %u"
771 "\n prev_priority: %i"
773 "\n inactive_ratio: %u",
774 zone_is_all_unreclaimable(zone),
776 zone->zone_start_pfn,
777 zone->inactive_ratio);
782 * Output information about zones in @pgdat.
784 static int zoneinfo_show(struct seq_file *m, void *arg)
786 pg_data_t *pgdat = (pg_data_t *)arg;
787 walk_zones_in_node(m, pgdat, zoneinfo_show_print);
791 static const struct seq_operations zoneinfo_op = {
792 .start = frag_start, /* iterate over all zones. The same as in
796 .show = zoneinfo_show,
799 static int zoneinfo_open(struct inode *inode, struct file *file)
801 return seq_open(file, &zoneinfo_op);
804 static const struct file_operations proc_zoneinfo_file_operations = {
805 .open = zoneinfo_open,
808 .release = seq_release,
811 static void *vmstat_start(struct seq_file *m, loff_t *pos)
814 #ifdef CONFIG_VM_EVENT_COUNTERS
819 if (*pos >= ARRAY_SIZE(vmstat_text))
822 #ifdef CONFIG_VM_EVENT_COUNTERS
823 v = kmalloc(NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long)
824 + sizeof(struct vm_event_state), GFP_KERNEL);
826 v = kmalloc(NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long),
831 return ERR_PTR(-ENOMEM);
832 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
833 v[i] = global_page_state(i);
834 #ifdef CONFIG_VM_EVENT_COUNTERS
835 e = v + NR_VM_ZONE_STAT_ITEMS;
837 e[PGPGIN] /= 2; /* sectors -> kbytes */
843 static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
846 if (*pos >= ARRAY_SIZE(vmstat_text))
848 return (unsigned long *)m->private + *pos;
851 static int vmstat_show(struct seq_file *m, void *arg)
853 unsigned long *l = arg;
854 unsigned long off = l - (unsigned long *)m->private;
856 seq_printf(m, "%s %lu\n", vmstat_text[off], *l);
860 static void vmstat_stop(struct seq_file *m, void *arg)
866 static const struct seq_operations vmstat_op = {
867 .start = vmstat_start,
873 static int vmstat_open(struct inode *inode, struct file *file)
875 return seq_open(file, &vmstat_op);
878 static const struct file_operations proc_vmstat_file_operations = {
882 .release = seq_release,
884 #endif /* CONFIG_PROC_FS */
887 static DEFINE_PER_CPU(struct delayed_work, vmstat_work);
888 int sysctl_stat_interval __read_mostly = HZ;
890 static void vmstat_update(struct work_struct *w)
892 refresh_cpu_vm_stats(smp_processor_id());
893 schedule_delayed_work(&__get_cpu_var(vmstat_work),
894 round_jiffies_relative(sysctl_stat_interval));
897 static void __cpuinit start_cpu_timer(int cpu)
899 struct delayed_work *vmstat_work = &per_cpu(vmstat_work, cpu);
901 INIT_DELAYED_WORK_DEFERRABLE(vmstat_work, vmstat_update);
902 schedule_delayed_work_on(cpu, vmstat_work,
903 __round_jiffies_relative(HZ, cpu));
907 * Use the cpu notifier to insure that the thresholds are recalculated
910 static int __cpuinit vmstat_cpuup_callback(struct notifier_block *nfb,
911 unsigned long action,
914 long cpu = (long)hcpu;
918 case CPU_ONLINE_FROZEN:
919 start_cpu_timer(cpu);
921 case CPU_DOWN_PREPARE:
922 case CPU_DOWN_PREPARE_FROZEN:
923 cancel_rearming_delayed_work(&per_cpu(vmstat_work, cpu));
924 per_cpu(vmstat_work, cpu).work.func = NULL;
926 case CPU_DOWN_FAILED:
927 case CPU_DOWN_FAILED_FROZEN:
928 start_cpu_timer(cpu);
931 case CPU_DEAD_FROZEN:
932 refresh_zone_stat_thresholds();
940 static struct notifier_block __cpuinitdata vmstat_notifier =
941 { &vmstat_cpuup_callback, NULL, 0 };
944 static int __init setup_vmstat(void)
949 refresh_zone_stat_thresholds();
950 register_cpu_notifier(&vmstat_notifier);
952 for_each_online_cpu(cpu)
953 start_cpu_timer(cpu);
955 #ifdef CONFIG_PROC_FS
956 proc_create("buddyinfo", S_IRUGO, NULL, &fragmentation_file_operations);
957 proc_create("pagetypeinfo", S_IRUGO, NULL, &pagetypeinfo_file_ops);
958 proc_create("vmstat", S_IRUGO, NULL, &proc_vmstat_file_operations);
959 proc_create("zoneinfo", S_IRUGO, NULL, &proc_zoneinfo_file_operations);
963 module_init(setup_vmstat)