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>
17 * Accumulate the page_state information across all CPUs.
18 * The result is unavoidably approximate - it can change
19 * during and after execution of this function.
21 DEFINE_PER_CPU(struct page_state, page_states) = {0};
23 atomic_t nr_pagecache = ATOMIC_INIT(0);
24 EXPORT_SYMBOL(nr_pagecache);
26 DEFINE_PER_CPU(long, nr_pagecache_local) = 0;
29 static void __get_page_state(struct page_state *ret, int nr, cpumask_t *cpumask)
33 memset(ret, 0, nr * sizeof(unsigned long));
34 cpus_and(*cpumask, *cpumask, cpu_online_map);
36 for_each_cpu_mask(cpu, *cpumask) {
42 in = (unsigned long *)&per_cpu(page_states, cpu);
44 next_cpu = next_cpu(cpu, *cpumask);
45 if (likely(next_cpu < NR_CPUS))
46 prefetch(&per_cpu(page_states, next_cpu));
48 out = (unsigned long *)ret;
49 for (off = 0; off < nr; off++)
54 void get_page_state_node(struct page_state *ret, int node)
57 cpumask_t mask = node_to_cpumask(node);
59 nr = offsetof(struct page_state, GET_PAGE_STATE_LAST);
60 nr /= sizeof(unsigned long);
62 __get_page_state(ret, nr+1, &mask);
65 void get_page_state(struct page_state *ret)
68 cpumask_t mask = CPU_MASK_ALL;
70 nr = offsetof(struct page_state, GET_PAGE_STATE_LAST);
71 nr /= sizeof(unsigned long);
73 __get_page_state(ret, nr + 1, &mask);
76 void get_full_page_state(struct page_state *ret)
78 cpumask_t mask = CPU_MASK_ALL;
80 __get_page_state(ret, sizeof(*ret) / sizeof(unsigned long), &mask);
83 unsigned long read_page_state_offset(unsigned long offset)
85 unsigned long ret = 0;
88 for_each_online_cpu(cpu) {
91 in = (unsigned long)&per_cpu(page_states, cpu) + offset;
92 ret += *((unsigned long *)in);
97 void __mod_page_state_offset(unsigned long offset, unsigned long delta)
101 ptr = &__get_cpu_var(page_states);
102 *(unsigned long *)(ptr + offset) += delta;
104 EXPORT_SYMBOL(__mod_page_state_offset);
106 void mod_page_state_offset(unsigned long offset, unsigned long delta)
111 local_irq_save(flags);
112 ptr = &__get_cpu_var(page_states);
113 *(unsigned long *)(ptr + offset) += delta;
114 local_irq_restore(flags);
116 EXPORT_SYMBOL(mod_page_state_offset);
118 void __get_zone_counts(unsigned long *active, unsigned long *inactive,
119 unsigned long *free, struct pglist_data *pgdat)
121 struct zone *zones = pgdat->node_zones;
127 for (i = 0; i < MAX_NR_ZONES; i++) {
128 *active += zones[i].nr_active;
129 *inactive += zones[i].nr_inactive;
130 *free += zones[i].free_pages;
134 void get_zone_counts(unsigned long *active,
135 unsigned long *inactive, unsigned long *free)
137 struct pglist_data *pgdat;
142 for_each_online_pgdat(pgdat) {
143 unsigned long l, m, n;
144 __get_zone_counts(&l, &m, &n, pgdat);
152 * Manage combined zone based / global counters
154 * vm_stat contains the global counters
156 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
157 EXPORT_SYMBOL(vm_stat);
161 #define STAT_THRESHOLD 32
164 * Determine pointer to currently valid differential byte given a zone and
167 * Preemption must be off
169 static inline s8 *diff_pointer(struct zone *zone, enum zone_stat_item item)
171 return &zone_pcp(zone, smp_processor_id())->vm_stat_diff[item];
175 * For use when we know that interrupts are disabled.
177 void __mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
183 p = diff_pointer(zone, item);
186 if (unlikely(x > STAT_THRESHOLD || x < -STAT_THRESHOLD)) {
187 zone_page_state_add(x, zone, item);
193 EXPORT_SYMBOL(__mod_zone_page_state);
196 * For an unknown interrupt state
198 void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
203 local_irq_save(flags);
204 __mod_zone_page_state(zone, item, delta);
205 local_irq_restore(flags);
207 EXPORT_SYMBOL(mod_zone_page_state);
210 * Optimized increment and decrement functions.
212 * These are only for a single page and therefore can take a struct page *
213 * argument instead of struct zone *. This allows the inclusion of the code
214 * generated for page_zone(page) into the optimized functions.
216 * No overflow check is necessary and therefore the differential can be
217 * incremented or decremented in place which may allow the compilers to
218 * generate better code.
220 * The increment or decrement is known and therefore one boundary check can
223 * Some processors have inc/dec instructions that are atomic vs an interrupt.
224 * However, the code must first determine the differential location in a zone
225 * based on the processor number and then inc/dec the counter. There is no
226 * guarantee without disabling preemption that the processor will not change
227 * in between and therefore the atomicity vs. interrupt cannot be exploited
228 * in a useful way here.
230 void __inc_zone_page_state(struct page *page, enum zone_stat_item item)
232 struct zone *zone = page_zone(page);
233 s8 *p = diff_pointer(zone, item);
237 if (unlikely(*p > STAT_THRESHOLD)) {
238 zone_page_state_add(*p, zone, item);
242 EXPORT_SYMBOL(__inc_zone_page_state);
244 void __dec_zone_page_state(struct page *page, enum zone_stat_item item)
246 struct zone *zone = page_zone(page);
247 s8 *p = diff_pointer(zone, item);
251 if (unlikely(*p < -STAT_THRESHOLD)) {
252 zone_page_state_add(*p, zone, item);
256 EXPORT_SYMBOL(__dec_zone_page_state);
258 void inc_zone_page_state(struct page *page, enum zone_stat_item item)
264 zone = page_zone(page);
265 local_irq_save(flags);
266 p = diff_pointer(zone, item);
270 if (unlikely(*p > STAT_THRESHOLD)) {
271 zone_page_state_add(*p, zone, item);
274 local_irq_restore(flags);
276 EXPORT_SYMBOL(inc_zone_page_state);
278 void dec_zone_page_state(struct page *page, enum zone_stat_item item)
284 zone = page_zone(page);
285 local_irq_save(flags);
286 p = diff_pointer(zone, item);
290 if (unlikely(*p < -STAT_THRESHOLD)) {
291 zone_page_state_add(*p, zone, item);
294 local_irq_restore(flags);
296 EXPORT_SYMBOL(dec_zone_page_state);
299 * Update the zone counters for one cpu.
301 void refresh_cpu_vm_stats(int cpu)
307 for_each_zone(zone) {
308 struct per_cpu_pageset *pcp;
310 pcp = zone_pcp(zone, cpu);
312 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
313 if (pcp->vm_stat_diff[i]) {
314 local_irq_save(flags);
315 zone_page_state_add(pcp->vm_stat_diff[i],
317 pcp->vm_stat_diff[i] = 0;
318 local_irq_restore(flags);
323 static void __refresh_cpu_vm_stats(void *dummy)
325 refresh_cpu_vm_stats(smp_processor_id());
329 * Consolidate all counters.
331 * Note that the result is less inaccurate but still inaccurate
332 * if concurrent processes are allowed to run.
334 void refresh_vm_stats(void)
336 on_each_cpu(__refresh_cpu_vm_stats, NULL, 0, 1);
338 EXPORT_SYMBOL(refresh_vm_stats);
342 #ifdef CONFIG_PROC_FS
344 #include <linux/seq_file.h>
346 static void *frag_start(struct seq_file *m, loff_t *pos)
350 for (pgdat = first_online_pgdat();
352 pgdat = next_online_pgdat(pgdat))
358 static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
360 pg_data_t *pgdat = (pg_data_t *)arg;
363 return next_online_pgdat(pgdat);
366 static void frag_stop(struct seq_file *m, void *arg)
371 * This walks the free areas for each zone.
373 static int frag_show(struct seq_file *m, void *arg)
375 pg_data_t *pgdat = (pg_data_t *)arg;
377 struct zone *node_zones = pgdat->node_zones;
381 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
382 if (!populated_zone(zone))
385 spin_lock_irqsave(&zone->lock, flags);
386 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
387 for (order = 0; order < MAX_ORDER; ++order)
388 seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
389 spin_unlock_irqrestore(&zone->lock, flags);
395 struct seq_operations fragmentation_op = {
402 static char *vmstat_text[] = {
403 /* Zoned VM counters */
409 "nr_page_table_pages",
440 "pgscan_kswapd_high",
441 "pgscan_kswapd_normal",
442 "pgscan_kswapd_dma32",
445 "pgscan_direct_high",
446 "pgscan_direct_normal",
447 "pgscan_direct_dma32",
462 * Output information about zones in @pgdat.
464 static int zoneinfo_show(struct seq_file *m, void *arg)
466 pg_data_t *pgdat = arg;
468 struct zone *node_zones = pgdat->node_zones;
471 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; zone++) {
474 if (!populated_zone(zone))
477 spin_lock_irqsave(&zone->lock, flags);
478 seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
486 "\n scanned %lu (a: %lu i: %lu)"
496 zone->nr_scan_active, zone->nr_scan_inactive,
498 zone->present_pages);
500 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
501 seq_printf(m, "\n %-12s %lu", vmstat_text[i],
502 zone_page_state(zone, i));
505 "\n protection: (%lu",
506 zone->lowmem_reserve[0]);
507 for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
508 seq_printf(m, ", %lu", zone->lowmem_reserve[i]);
512 for_each_online_cpu(i) {
513 struct per_cpu_pageset *pageset;
516 pageset = zone_pcp(zone, i);
517 for (j = 0; j < ARRAY_SIZE(pageset->pcp); j++) {
518 if (pageset->pcp[j].count)
521 if (j == ARRAY_SIZE(pageset->pcp))
523 for (j = 0; j < ARRAY_SIZE(pageset->pcp); j++) {
530 pageset->pcp[j].count,
531 pageset->pcp[j].high,
532 pageset->pcp[j].batch);
538 "\n numa_foreign: %lu"
539 "\n interleave_hit: %lu"
541 "\n other_node: %lu",
544 pageset->numa_foreign,
545 pageset->interleave_hit,
547 pageset->other_node);
551 "\n all_unreclaimable: %u"
552 "\n prev_priority: %i"
553 "\n temp_priority: %i"
555 zone->all_unreclaimable,
558 zone->zone_start_pfn);
559 spin_unlock_irqrestore(&zone->lock, flags);
565 struct seq_operations zoneinfo_op = {
566 .start = frag_start, /* iterate over all zones. The same as in
570 .show = zoneinfo_show,
573 static void *vmstat_start(struct seq_file *m, loff_t *pos)
576 struct page_state *ps;
579 if (*pos >= ARRAY_SIZE(vmstat_text))
582 v = kmalloc(NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long)
583 + sizeof(*ps), GFP_KERNEL);
586 return ERR_PTR(-ENOMEM);
587 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
588 v[i] = global_page_state(i);
589 ps = (struct page_state *)(v + NR_VM_ZONE_STAT_ITEMS);
590 get_full_page_state(ps);
591 ps->pgpgin /= 2; /* sectors -> kbytes */
596 static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
599 if (*pos >= ARRAY_SIZE(vmstat_text))
601 return (unsigned long *)m->private + *pos;
604 static int vmstat_show(struct seq_file *m, void *arg)
606 unsigned long *l = arg;
607 unsigned long off = l - (unsigned long *)m->private;
609 seq_printf(m, "%s %lu\n", vmstat_text[off], *l);
613 static void vmstat_stop(struct seq_file *m, void *arg)
619 struct seq_operations vmstat_op = {
620 .start = vmstat_start,
626 #endif /* CONFIG_PROC_FS */