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/sched.h>
18 #ifdef CONFIG_VM_EVENT_COUNTERS
19 DEFINE_PER_CPU(struct vm_event_state, vm_event_states) = {{0}};
20 EXPORT_PER_CPU_SYMBOL(vm_event_states);
22 static void sum_vm_events(unsigned long *ret, cpumask_t *cpumask)
27 memset(ret, 0, NR_VM_EVENT_ITEMS * sizeof(unsigned long));
29 cpu = first_cpu(*cpumask);
30 while (cpu < NR_CPUS) {
31 struct vm_event_state *this = &per_cpu(vm_event_states, cpu);
33 cpu = next_cpu(cpu, *cpumask);
36 prefetch(&per_cpu(vm_event_states, cpu));
39 for (i = 0; i < NR_VM_EVENT_ITEMS; i++)
40 ret[i] += this->event[i];
45 * Accumulate the vm event counters across all CPUs.
46 * The result is unavoidably approximate - it can change
47 * during and after execution of this function.
49 void all_vm_events(unsigned long *ret)
51 sum_vm_events(ret, &cpu_online_map);
53 EXPORT_SYMBOL_GPL(all_vm_events);
57 * Fold the foreign cpu events into our own.
59 * This is adding to the events on one processor
60 * but keeps the global counts constant.
62 void vm_events_fold_cpu(int cpu)
64 struct vm_event_state *fold_state = &per_cpu(vm_event_states, cpu);
67 for (i = 0; i < NR_VM_EVENT_ITEMS; i++) {
68 count_vm_events(i, fold_state->event[i]);
69 fold_state->event[i] = 0;
72 #endif /* CONFIG_HOTPLUG */
74 #endif /* CONFIG_VM_EVENT_COUNTERS */
77 * Manage combined zone based / global counters
79 * vm_stat contains the global counters
81 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
82 EXPORT_SYMBOL(vm_stat);
86 static int calculate_threshold(struct zone *zone)
89 int mem; /* memory in 128 MB units */
92 * The threshold scales with the number of processors and the amount
93 * of memory per zone. More memory means that we can defer updates for
94 * longer, more processors could lead to more contention.
95 * fls() is used to have a cheap way of logarithmic scaling.
97 * Some sample thresholds:
99 * Threshold Processors (fls) Zonesize fls(mem+1)
100 * ------------------------------------------------------------------
117 * 125 1024 10 8-16 GB 8
118 * 125 1024 10 16-32 GB 9
121 mem = zone->present_pages >> (27 - PAGE_SHIFT);
123 threshold = 2 * fls(num_online_cpus()) * (1 + fls(mem));
126 * Maximum threshold is 125
128 threshold = min(125, threshold);
134 * Refresh the thresholds for each zone.
136 static void refresh_zone_stat_thresholds(void)
142 for_each_zone(zone) {
144 if (!zone->present_pages)
147 threshold = calculate_threshold(zone);
149 for_each_online_cpu(cpu)
150 zone_pcp(zone, cpu)->stat_threshold = threshold;
155 * For use when we know that interrupts are disabled.
157 void __mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
160 struct per_cpu_pageset *pcp = zone_pcp(zone, smp_processor_id());
161 s8 *p = pcp->vm_stat_diff + item;
166 if (unlikely(x > pcp->stat_threshold || x < -pcp->stat_threshold)) {
167 zone_page_state_add(x, zone, item);
172 EXPORT_SYMBOL(__mod_zone_page_state);
175 * For an unknown interrupt state
177 void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
182 local_irq_save(flags);
183 __mod_zone_page_state(zone, item, delta);
184 local_irq_restore(flags);
186 EXPORT_SYMBOL(mod_zone_page_state);
189 * Optimized increment and decrement functions.
191 * These are only for a single page and therefore can take a struct page *
192 * argument instead of struct zone *. This allows the inclusion of the code
193 * generated for page_zone(page) into the optimized functions.
195 * No overflow check is necessary and therefore the differential can be
196 * incremented or decremented in place which may allow the compilers to
197 * generate better code.
198 * The increment or decrement is known and therefore one boundary check can
201 * NOTE: These functions are very performance sensitive. Change only
204 * Some processors have inc/dec instructions that are atomic vs an interrupt.
205 * However, the code must first determine the differential location in a zone
206 * based on the processor number and then inc/dec the counter. There is no
207 * guarantee without disabling preemption that the processor will not change
208 * in between and therefore the atomicity vs. interrupt cannot be exploited
209 * in a useful way here.
211 void __inc_zone_state(struct zone *zone, enum zone_stat_item item)
213 struct per_cpu_pageset *pcp = zone_pcp(zone, smp_processor_id());
214 s8 *p = pcp->vm_stat_diff + item;
218 if (unlikely(*p > pcp->stat_threshold)) {
219 int overstep = pcp->stat_threshold / 2;
221 zone_page_state_add(*p + overstep, zone, item);
226 void __inc_zone_page_state(struct page *page, enum zone_stat_item item)
228 __inc_zone_state(page_zone(page), item);
230 EXPORT_SYMBOL(__inc_zone_page_state);
232 void __dec_zone_state(struct zone *zone, enum zone_stat_item item)
234 struct per_cpu_pageset *pcp = zone_pcp(zone, smp_processor_id());
235 s8 *p = pcp->vm_stat_diff + item;
239 if (unlikely(*p < - pcp->stat_threshold)) {
240 int overstep = pcp->stat_threshold / 2;
242 zone_page_state_add(*p - overstep, zone, item);
247 void __dec_zone_page_state(struct page *page, enum zone_stat_item item)
249 __dec_zone_state(page_zone(page), item);
251 EXPORT_SYMBOL(__dec_zone_page_state);
253 void inc_zone_state(struct zone *zone, enum zone_stat_item item)
257 local_irq_save(flags);
258 __inc_zone_state(zone, item);
259 local_irq_restore(flags);
262 void inc_zone_page_state(struct page *page, enum zone_stat_item item)
267 zone = page_zone(page);
268 local_irq_save(flags);
269 __inc_zone_state(zone, item);
270 local_irq_restore(flags);
272 EXPORT_SYMBOL(inc_zone_page_state);
274 void dec_zone_page_state(struct page *page, enum zone_stat_item item)
278 local_irq_save(flags);
279 __dec_zone_page_state(page, item);
280 local_irq_restore(flags);
282 EXPORT_SYMBOL(dec_zone_page_state);
285 * Update the zone counters for one cpu.
287 * The cpu specified must be either the current cpu or a processor that
288 * is not online. If it is the current cpu then the execution thread must
289 * be pinned to the current cpu.
291 * Note that refresh_cpu_vm_stats strives to only access
292 * node local memory. The per cpu pagesets on remote zones are placed
293 * in the memory local to the processor using that pageset. So the
294 * loop over all zones will access a series of cachelines local to
297 * The call to zone_page_state_add updates the cachelines with the
298 * statistics in the remote zone struct as well as the global cachelines
299 * with the global counters. These could cause remote node cache line
300 * bouncing and will have to be only done when necessary.
302 void refresh_cpu_vm_stats(int cpu)
306 int global_diff[NR_VM_ZONE_STAT_ITEMS] = { 0, };
308 for_each_zone(zone) {
309 struct per_cpu_pageset *p;
311 if (!populated_zone(zone))
314 p = zone_pcp(zone, cpu);
316 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
317 if (p->vm_stat_diff[i]) {
321 local_irq_save(flags);
322 v = p->vm_stat_diff[i];
323 p->vm_stat_diff[i] = 0;
324 local_irq_restore(flags);
325 atomic_long_add(v, &zone->vm_stat[i]);
328 /* 3 seconds idle till flush */
334 * Deal with draining the remote pageset of this
337 * Check if there are pages remaining in this pageset
338 * if not then there is nothing to expire.
340 if (!p->expire || !p->pcp.count)
344 * We never drain zones local to this processor.
346 if (zone_to_nid(zone) == numa_node_id()) {
356 drain_zone_pages(zone, &p->pcp);
360 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
362 atomic_long_add(global_diff[i], &vm_stat[i]);
369 * zonelist = the list of zones passed to the allocator
370 * z = the zone from which the allocation occurred.
372 * Must be called with interrupts disabled.
374 void zone_statistics(struct zonelist *zonelist, struct zone *z)
376 if (z->zone_pgdat == zonelist->zones[0]->zone_pgdat) {
377 __inc_zone_state(z, NUMA_HIT);
379 __inc_zone_state(z, NUMA_MISS);
380 __inc_zone_state(zonelist->zones[0], NUMA_FOREIGN);
382 if (z->node == numa_node_id())
383 __inc_zone_state(z, NUMA_LOCAL);
385 __inc_zone_state(z, NUMA_OTHER);
389 #ifdef CONFIG_PROC_FS
391 #include <linux/seq_file.h>
393 static char * const migratetype_names[MIGRATE_TYPES] = {
400 static void *frag_start(struct seq_file *m, loff_t *pos)
404 for (pgdat = first_online_pgdat();
406 pgdat = next_online_pgdat(pgdat))
412 static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
414 pg_data_t *pgdat = (pg_data_t *)arg;
417 return next_online_pgdat(pgdat);
420 static void frag_stop(struct seq_file *m, void *arg)
424 /* Walk all the zones in a node and print using a callback */
425 static void walk_zones_in_node(struct seq_file *m, pg_data_t *pgdat,
426 void (*print)(struct seq_file *m, pg_data_t *, struct zone *))
429 struct zone *node_zones = pgdat->node_zones;
432 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
433 if (!populated_zone(zone))
436 spin_lock_irqsave(&zone->lock, flags);
437 print(m, pgdat, zone);
438 spin_unlock_irqrestore(&zone->lock, flags);
442 static void frag_show_print(struct seq_file *m, pg_data_t *pgdat,
447 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
448 for (order = 0; order < MAX_ORDER; ++order)
449 seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
454 * This walks the free areas for each zone.
456 static int frag_show(struct seq_file *m, void *arg)
458 pg_data_t *pgdat = (pg_data_t *)arg;
459 walk_zones_in_node(m, pgdat, frag_show_print);
463 static void pagetypeinfo_showfree_print(struct seq_file *m,
464 pg_data_t *pgdat, struct zone *zone)
468 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++) {
469 seq_printf(m, "Node %4d, zone %8s, type %12s ",
472 migratetype_names[mtype]);
473 for (order = 0; order < MAX_ORDER; ++order) {
474 unsigned long freecount = 0;
475 struct free_area *area;
476 struct list_head *curr;
478 area = &(zone->free_area[order]);
480 list_for_each(curr, &area->free_list[mtype])
482 seq_printf(m, "%6lu ", freecount);
488 /* Print out the free pages at each order for each migatetype */
489 static int pagetypeinfo_showfree(struct seq_file *m, void *arg)
492 pg_data_t *pgdat = (pg_data_t *)arg;
495 seq_printf(m, "%-43s ", "Free pages count per migrate type at order");
496 for (order = 0; order < MAX_ORDER; ++order)
497 seq_printf(m, "%6d ", order);
500 walk_zones_in_node(m, pgdat, pagetypeinfo_showfree_print);
505 static void pagetypeinfo_showblockcount_print(struct seq_file *m,
506 pg_data_t *pgdat, struct zone *zone)
510 unsigned long start_pfn = zone->zone_start_pfn;
511 unsigned long end_pfn = start_pfn + zone->spanned_pages;
512 unsigned long count[MIGRATE_TYPES] = { 0, };
514 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
520 page = pfn_to_page(pfn);
521 mtype = get_pageblock_migratetype(page);
527 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
528 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
529 seq_printf(m, "%12lu ", count[mtype]);
533 /* Print out the free pages at each order for each migratetype */
534 static int pagetypeinfo_showblockcount(struct seq_file *m, void *arg)
537 pg_data_t *pgdat = (pg_data_t *)arg;
539 seq_printf(m, "\n%-23s", "Number of blocks type ");
540 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
541 seq_printf(m, "%12s ", migratetype_names[mtype]);
543 walk_zones_in_node(m, pgdat, pagetypeinfo_showblockcount_print);
549 * This prints out statistics in relation to grouping pages by mobility.
550 * It is expensive to collect so do not constantly read the file.
552 static int pagetypeinfo_show(struct seq_file *m, void *arg)
554 pg_data_t *pgdat = (pg_data_t *)arg;
556 seq_printf(m, "Page block order: %d\n", pageblock_order);
557 seq_printf(m, "Pages per block: %lu\n", pageblock_nr_pages);
559 pagetypeinfo_showfree(m, pgdat);
560 pagetypeinfo_showblockcount(m, pgdat);
565 const struct seq_operations fragmentation_op = {
572 const struct seq_operations pagetypeinfo_op = {
576 .show = pagetypeinfo_show,
579 #ifdef CONFIG_ZONE_DMA
580 #define TEXT_FOR_DMA(xx) xx "_dma",
582 #define TEXT_FOR_DMA(xx)
585 #ifdef CONFIG_ZONE_DMA32
586 #define TEXT_FOR_DMA32(xx) xx "_dma32",
588 #define TEXT_FOR_DMA32(xx)
591 #ifdef CONFIG_HIGHMEM
592 #define TEXT_FOR_HIGHMEM(xx) xx "_high",
594 #define TEXT_FOR_HIGHMEM(xx)
597 #define TEXTS_FOR_ZONES(xx) TEXT_FOR_DMA(xx) TEXT_FOR_DMA32(xx) xx "_normal", \
598 TEXT_FOR_HIGHMEM(xx) xx "_movable",
600 static const char * const vmstat_text[] = {
601 /* Zoned VM counters */
610 "nr_slab_reclaimable",
611 "nr_slab_unreclaimable",
612 "nr_page_table_pages",
626 #ifdef CONFIG_VM_EVENT_COUNTERS
632 TEXTS_FOR_ZONES("pgalloc")
641 TEXTS_FOR_ZONES("pgrefill")
642 TEXTS_FOR_ZONES("pgsteal")
643 TEXTS_FOR_ZONES("pgscan_kswapd")
644 TEXTS_FOR_ZONES("pgscan_direct")
657 static void zoneinfo_show_print(struct seq_file *m, pg_data_t *pgdat,
661 seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
667 "\n scanned %lu (a: %lu i: %lu)"
670 zone_page_state(zone, NR_FREE_PAGES),
675 zone->nr_scan_active, zone->nr_scan_inactive,
677 zone->present_pages);
679 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
680 seq_printf(m, "\n %-12s %lu", vmstat_text[i],
681 zone_page_state(zone, i));
684 "\n protection: (%lu",
685 zone->lowmem_reserve[0]);
686 for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
687 seq_printf(m, ", %lu", zone->lowmem_reserve[i]);
691 for_each_online_cpu(i) {
692 struct per_cpu_pageset *pageset;
694 pageset = zone_pcp(zone, i);
705 seq_printf(m, "\n vm stats threshold: %d",
706 pageset->stat_threshold);
710 "\n all_unreclaimable: %u"
711 "\n prev_priority: %i"
713 zone_is_all_unreclaimable(zone),
715 zone->zone_start_pfn);
720 * Output information about zones in @pgdat.
722 static int zoneinfo_show(struct seq_file *m, void *arg)
724 pg_data_t *pgdat = (pg_data_t *)arg;
725 walk_zones_in_node(m, pgdat, zoneinfo_show_print);
729 const struct seq_operations zoneinfo_op = {
730 .start = frag_start, /* iterate over all zones. The same as in
734 .show = zoneinfo_show,
737 static void *vmstat_start(struct seq_file *m, loff_t *pos)
740 #ifdef CONFIG_VM_EVENT_COUNTERS
745 if (*pos >= ARRAY_SIZE(vmstat_text))
748 #ifdef CONFIG_VM_EVENT_COUNTERS
749 v = kmalloc(NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long)
750 + sizeof(struct vm_event_state), GFP_KERNEL);
752 v = kmalloc(NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long),
757 return ERR_PTR(-ENOMEM);
758 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
759 v[i] = global_page_state(i);
760 #ifdef CONFIG_VM_EVENT_COUNTERS
761 e = v + NR_VM_ZONE_STAT_ITEMS;
763 e[PGPGIN] /= 2; /* sectors -> kbytes */
769 static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
772 if (*pos >= ARRAY_SIZE(vmstat_text))
774 return (unsigned long *)m->private + *pos;
777 static int vmstat_show(struct seq_file *m, void *arg)
779 unsigned long *l = arg;
780 unsigned long off = l - (unsigned long *)m->private;
782 seq_printf(m, "%s %lu\n", vmstat_text[off], *l);
786 static void vmstat_stop(struct seq_file *m, void *arg)
792 const struct seq_operations vmstat_op = {
793 .start = vmstat_start,
799 #endif /* CONFIG_PROC_FS */
802 static DEFINE_PER_CPU(struct delayed_work, vmstat_work);
803 int sysctl_stat_interval __read_mostly = HZ;
805 static void vmstat_update(struct work_struct *w)
807 refresh_cpu_vm_stats(smp_processor_id());
808 schedule_delayed_work(&__get_cpu_var(vmstat_work),
809 sysctl_stat_interval);
812 static void __cpuinit start_cpu_timer(int cpu)
814 struct delayed_work *vmstat_work = &per_cpu(vmstat_work, cpu);
816 INIT_DELAYED_WORK_DEFERRABLE(vmstat_work, vmstat_update);
817 schedule_delayed_work_on(cpu, vmstat_work, HZ + cpu);
821 * Use the cpu notifier to insure that the thresholds are recalculated
824 static int __cpuinit vmstat_cpuup_callback(struct notifier_block *nfb,
825 unsigned long action,
828 long cpu = (long)hcpu;
832 case CPU_ONLINE_FROZEN:
833 start_cpu_timer(cpu);
835 case CPU_DOWN_PREPARE:
836 case CPU_DOWN_PREPARE_FROZEN:
837 cancel_rearming_delayed_work(&per_cpu(vmstat_work, cpu));
838 per_cpu(vmstat_work, cpu).work.func = NULL;
840 case CPU_DOWN_FAILED:
841 case CPU_DOWN_FAILED_FROZEN:
842 start_cpu_timer(cpu);
845 case CPU_DEAD_FROZEN:
846 refresh_zone_stat_thresholds();
854 static struct notifier_block __cpuinitdata vmstat_notifier =
855 { &vmstat_cpuup_callback, NULL, 0 };
857 static int __init setup_vmstat(void)
861 refresh_zone_stat_thresholds();
862 register_cpu_notifier(&vmstat_notifier);
864 for_each_online_cpu(cpu)
865 start_cpu_timer(cpu);
868 module_init(setup_vmstat)