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>
12 #include <linux/config.h>
14 #include <linux/module.h>
15 #include <linux/cpu.h>
17 void __get_zone_counts(unsigned long *active, unsigned long *inactive,
18 unsigned long *free, struct pglist_data *pgdat)
20 struct zone *zones = pgdat->node_zones;
26 for (i = 0; i < MAX_NR_ZONES; i++) {
27 *active += zones[i].nr_active;
28 *inactive += zones[i].nr_inactive;
29 *free += zones[i].free_pages;
33 void get_zone_counts(unsigned long *active,
34 unsigned long *inactive, unsigned long *free)
36 struct pglist_data *pgdat;
41 for_each_online_pgdat(pgdat) {
42 unsigned long l, m, n;
43 __get_zone_counts(&l, &m, &n, pgdat);
50 #ifdef CONFIG_VM_EVENT_COUNTERS
51 DEFINE_PER_CPU(struct vm_event_state, vm_event_states) = {{0}};
52 EXPORT_PER_CPU_SYMBOL(vm_event_states);
54 static void sum_vm_events(unsigned long *ret, cpumask_t *cpumask)
59 memset(ret, 0, NR_VM_EVENT_ITEMS * sizeof(unsigned long));
61 cpu = first_cpu(*cpumask);
62 while (cpu < NR_CPUS) {
63 struct vm_event_state *this = &per_cpu(vm_event_states, cpu);
65 cpu = next_cpu(cpu, *cpumask);
68 prefetch(&per_cpu(vm_event_states, cpu));
71 for (i = 0; i < NR_VM_EVENT_ITEMS; i++)
72 ret[i] += this->event[i];
77 * Accumulate the vm event counters across all CPUs.
78 * The result is unavoidably approximate - it can change
79 * during and after execution of this function.
81 void all_vm_events(unsigned long *ret)
83 sum_vm_events(ret, &cpu_online_map);
85 EXPORT_SYMBOL_GPL(all_vm_events);
89 * Fold the foreign cpu events into our own.
91 * This is adding to the events on one processor
92 * but keeps the global counts constant.
94 void vm_events_fold_cpu(int cpu)
96 struct vm_event_state *fold_state = &per_cpu(vm_event_states, cpu);
99 for (i = 0; i < NR_VM_EVENT_ITEMS; i++) {
100 count_vm_events(i, fold_state->event[i]);
101 fold_state->event[i] = 0;
104 #endif /* CONFIG_HOTPLUG */
106 #endif /* CONFIG_VM_EVENT_COUNTERS */
109 * Manage combined zone based / global counters
111 * vm_stat contains the global counters
113 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
114 EXPORT_SYMBOL(vm_stat);
118 static int calculate_threshold(struct zone *zone)
121 int mem; /* memory in 128 MB units */
124 * The threshold scales with the number of processors and the amount
125 * of memory per zone. More memory means that we can defer updates for
126 * longer, more processors could lead to more contention.
127 * fls() is used to have a cheap way of logarithmic scaling.
129 * Some sample thresholds:
131 * Threshold Processors (fls) Zonesize fls(mem+1)
132 * ------------------------------------------------------------------
149 * 125 1024 10 8-16 GB 8
150 * 125 1024 10 16-32 GB 9
153 mem = zone->present_pages >> (27 - PAGE_SHIFT);
155 threshold = 2 * fls(num_online_cpus()) * (1 + fls(mem));
158 * Maximum threshold is 125
160 threshold = min(125, threshold);
166 * Refresh the thresholds for each zone.
168 static void refresh_zone_stat_thresholds(void)
174 for_each_zone(zone) {
176 if (!zone->present_pages)
179 threshold = calculate_threshold(zone);
181 for_each_online_cpu(cpu)
182 zone_pcp(zone, cpu)->stat_threshold = threshold;
187 * For use when we know that interrupts are disabled.
189 void __mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
192 struct per_cpu_pageset *pcp = zone_pcp(zone, smp_processor_id());
193 s8 *p = pcp->vm_stat_diff + item;
198 if (unlikely(x > pcp->stat_threshold || x < -pcp->stat_threshold)) {
199 zone_page_state_add(x, zone, item);
204 EXPORT_SYMBOL(__mod_zone_page_state);
207 * For an unknown interrupt state
209 void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
214 local_irq_save(flags);
215 __mod_zone_page_state(zone, item, delta);
216 local_irq_restore(flags);
218 EXPORT_SYMBOL(mod_zone_page_state);
221 * Optimized increment and decrement functions.
223 * These are only for a single page and therefore can take a struct page *
224 * argument instead of struct zone *. This allows the inclusion of the code
225 * generated for page_zone(page) into the optimized functions.
227 * No overflow check is necessary and therefore the differential can be
228 * incremented or decremented in place which may allow the compilers to
229 * generate better code.
230 * The increment or decrement is known and therefore one boundary check can
233 * NOTE: These functions are very performance sensitive. Change only
236 * Some processors have inc/dec instructions that are atomic vs an interrupt.
237 * However, the code must first determine the differential location in a zone
238 * based on the processor number and then inc/dec the counter. There is no
239 * guarantee without disabling preemption that the processor will not change
240 * in between and therefore the atomicity vs. interrupt cannot be exploited
241 * in a useful way here.
243 static void __inc_zone_state(struct zone *zone, enum zone_stat_item item)
245 struct per_cpu_pageset *pcp = zone_pcp(zone, smp_processor_id());
246 s8 *p = pcp->vm_stat_diff + item;
250 if (unlikely(*p > pcp->stat_threshold)) {
251 int overstep = pcp->stat_threshold / 2;
253 zone_page_state_add(*p + overstep, zone, item);
258 void __inc_zone_page_state(struct page *page, enum zone_stat_item item)
260 __inc_zone_state(page_zone(page), item);
262 EXPORT_SYMBOL(__inc_zone_page_state);
264 void __dec_zone_page_state(struct page *page, enum zone_stat_item item)
266 struct zone *zone = page_zone(page);
267 struct per_cpu_pageset *pcp = zone_pcp(zone, smp_processor_id());
268 s8 *p = pcp->vm_stat_diff + item;
272 if (unlikely(*p < - pcp->stat_threshold)) {
273 int overstep = pcp->stat_threshold / 2;
275 zone_page_state_add(*p - overstep, zone, item);
279 EXPORT_SYMBOL(__dec_zone_page_state);
281 void inc_zone_state(struct zone *zone, enum zone_stat_item item)
285 local_irq_save(flags);
286 __inc_zone_state(zone, item);
287 local_irq_restore(flags);
290 void inc_zone_page_state(struct page *page, enum zone_stat_item item)
295 zone = page_zone(page);
296 local_irq_save(flags);
297 __inc_zone_state(zone, item);
298 local_irq_restore(flags);
300 EXPORT_SYMBOL(inc_zone_page_state);
302 void dec_zone_page_state(struct page *page, enum zone_stat_item item)
306 local_irq_save(flags);
307 __dec_zone_page_state(page, item);
308 local_irq_restore(flags);
310 EXPORT_SYMBOL(dec_zone_page_state);
313 * Update the zone counters for one cpu.
315 void refresh_cpu_vm_stats(int cpu)
321 for_each_zone(zone) {
322 struct per_cpu_pageset *pcp;
324 pcp = zone_pcp(zone, cpu);
326 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
327 if (pcp->vm_stat_diff[i]) {
328 local_irq_save(flags);
329 zone_page_state_add(pcp->vm_stat_diff[i],
331 pcp->vm_stat_diff[i] = 0;
332 local_irq_restore(flags);
337 static void __refresh_cpu_vm_stats(void *dummy)
339 refresh_cpu_vm_stats(smp_processor_id());
343 * Consolidate all counters.
345 * Note that the result is less inaccurate but still inaccurate
346 * if concurrent processes are allowed to run.
348 void refresh_vm_stats(void)
350 on_each_cpu(__refresh_cpu_vm_stats, NULL, 0, 1);
352 EXPORT_SYMBOL(refresh_vm_stats);
358 * zonelist = the list of zones passed to the allocator
359 * z = the zone from which the allocation occurred.
361 * Must be called with interrupts disabled.
363 void zone_statistics(struct zonelist *zonelist, struct zone *z)
365 if (z->zone_pgdat == zonelist->zones[0]->zone_pgdat) {
366 __inc_zone_state(z, NUMA_HIT);
368 __inc_zone_state(z, NUMA_MISS);
369 __inc_zone_state(zonelist->zones[0], NUMA_FOREIGN);
371 if (z->zone_pgdat == NODE_DATA(numa_node_id()))
372 __inc_zone_state(z, NUMA_LOCAL);
374 __inc_zone_state(z, NUMA_OTHER);
378 #ifdef CONFIG_PROC_FS
380 #include <linux/seq_file.h>
382 static void *frag_start(struct seq_file *m, loff_t *pos)
386 for (pgdat = first_online_pgdat();
388 pgdat = next_online_pgdat(pgdat))
394 static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
396 pg_data_t *pgdat = (pg_data_t *)arg;
399 return next_online_pgdat(pgdat);
402 static void frag_stop(struct seq_file *m, void *arg)
407 * This walks the free areas for each zone.
409 static int frag_show(struct seq_file *m, void *arg)
411 pg_data_t *pgdat = (pg_data_t *)arg;
413 struct zone *node_zones = pgdat->node_zones;
417 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
418 if (!populated_zone(zone))
421 spin_lock_irqsave(&zone->lock, flags);
422 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
423 for (order = 0; order < MAX_ORDER; ++order)
424 seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
425 spin_unlock_irqrestore(&zone->lock, flags);
431 struct seq_operations fragmentation_op = {
438 #ifdef CONFIG_ZONE_DMA32
439 #define TEXT_FOR_DMA32(xx) xx "_dma32",
441 #define TEXT_FOR_DMA32(xx)
444 #ifdef CONFIG_HIGHMEM
445 #define TEXT_FOR_HIGHMEM(xx) xx "_high",
447 #define TEXT_FOR_HIGHMEM(xx)
450 #define TEXTS_FOR_ZONES(xx) xx "_dma", TEXT_FOR_DMA32(xx) xx "_normal", \
453 static char *vmstat_text[] = {
454 /* Zoned VM counters */
459 "nr_page_table_pages",
474 #ifdef CONFIG_VM_EVENT_COUNTERS
480 TEXTS_FOR_ZONES("pgalloc")
489 TEXTS_FOR_ZONES("pgrefill")
490 TEXTS_FOR_ZONES("pgsteal")
491 TEXTS_FOR_ZONES("pgscan_kswapd")
492 TEXTS_FOR_ZONES("pgscan_direct")
506 * Output information about zones in @pgdat.
508 static int zoneinfo_show(struct seq_file *m, void *arg)
510 pg_data_t *pgdat = arg;
512 struct zone *node_zones = pgdat->node_zones;
515 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; zone++) {
518 if (!populated_zone(zone))
521 spin_lock_irqsave(&zone->lock, flags);
522 seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
530 "\n scanned %lu (a: %lu i: %lu)"
540 zone->nr_scan_active, zone->nr_scan_inactive,
542 zone->present_pages);
544 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
545 seq_printf(m, "\n %-12s %lu", vmstat_text[i],
546 zone_page_state(zone, i));
549 "\n protection: (%lu",
550 zone->lowmem_reserve[0]);
551 for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
552 seq_printf(m, ", %lu", zone->lowmem_reserve[i]);
556 for_each_online_cpu(i) {
557 struct per_cpu_pageset *pageset;
560 pageset = zone_pcp(zone, i);
561 for (j = 0; j < ARRAY_SIZE(pageset->pcp); j++) {
562 if (pageset->pcp[j].count)
565 if (j == ARRAY_SIZE(pageset->pcp))
567 for (j = 0; j < ARRAY_SIZE(pageset->pcp); j++) {
574 pageset->pcp[j].count,
575 pageset->pcp[j].high,
576 pageset->pcp[j].batch);
579 seq_printf(m, "\n vm stats threshold: %d",
580 pageset->stat_threshold);
584 "\n all_unreclaimable: %u"
585 "\n prev_priority: %i"
586 "\n temp_priority: %i"
588 zone->all_unreclaimable,
591 zone->zone_start_pfn);
592 spin_unlock_irqrestore(&zone->lock, flags);
598 struct seq_operations zoneinfo_op = {
599 .start = frag_start, /* iterate over all zones. The same as in
603 .show = zoneinfo_show,
606 static void *vmstat_start(struct seq_file *m, loff_t *pos)
609 #ifdef CONFIG_VM_EVENT_COUNTERS
614 if (*pos >= ARRAY_SIZE(vmstat_text))
617 #ifdef CONFIG_VM_EVENT_COUNTERS
618 v = kmalloc(NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long)
619 + sizeof(struct vm_event_state), GFP_KERNEL);
621 v = kmalloc(NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long),
626 return ERR_PTR(-ENOMEM);
627 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
628 v[i] = global_page_state(i);
629 #ifdef CONFIG_VM_EVENT_COUNTERS
630 e = v + NR_VM_ZONE_STAT_ITEMS;
632 e[PGPGIN] /= 2; /* sectors -> kbytes */
638 static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
641 if (*pos >= ARRAY_SIZE(vmstat_text))
643 return (unsigned long *)m->private + *pos;
646 static int vmstat_show(struct seq_file *m, void *arg)
648 unsigned long *l = arg;
649 unsigned long off = l - (unsigned long *)m->private;
651 seq_printf(m, "%s %lu\n", vmstat_text[off], *l);
655 static void vmstat_stop(struct seq_file *m, void *arg)
661 struct seq_operations vmstat_op = {
662 .start = vmstat_start,
668 #endif /* CONFIG_PROC_FS */
672 * Use the cpu notifier to insure that the thresholds are recalculated
675 static int __cpuinit vmstat_cpuup_callback(struct notifier_block *nfb,
676 unsigned long action,
681 case CPU_UP_CANCELED:
683 refresh_zone_stat_thresholds();
691 static struct notifier_block __cpuinitdata vmstat_notifier =
692 { &vmstat_cpuup_callback, NULL, 0 };
694 int __init setup_vmstat(void)
696 refresh_zone_stat_thresholds();
697 register_cpu_notifier(&vmstat_notifier);
700 module_init(setup_vmstat)