tty: moxa, fix refcounting in moxa_poll_port
[linux-2.6/mini2440.git] / mm / vmstat.c
blob66f6130976cb65c569f41108507479a68ec0779c
1 /*
2 * linux/mm/vmstat.c
4 * Manages VM statistics
5 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
7 * zoned VM statistics
8 * Copyright (C) 2006 Silicon Graphics, Inc.,
9 * Christoph Lameter <christoph@lameter.com>
11 #include <linux/fs.h>
12 #include <linux/mm.h>
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)
25 int cpu;
26 int i;
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)
45 get_online_cpus();
46 sum_vm_events(ret, cpu_online_mask);
47 put_online_cpus();
49 EXPORT_SYMBOL_GPL(all_vm_events);
51 #ifdef CONFIG_HOTPLUG
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);
61 int i;
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);
80 #ifdef CONFIG_SMP
82 static int calculate_threshold(struct zone *zone)
84 int threshold;
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 * ------------------------------------------------------------------
97 * 8 1 1 0.9-1 GB 4
98 * 16 2 2 0.9-1 GB 4
99 * 20 2 2 1-2 GB 5
100 * 24 2 2 2-4 GB 6
101 * 28 2 2 4-8 GB 7
102 * 32 2 2 8-16 GB 8
103 * 4 2 2 <128M 1
104 * 30 4 3 2-4 GB 5
105 * 48 4 3 8-16 GB 8
106 * 32 8 4 1-2 GB 4
107 * 32 8 4 0.9-1GB 4
108 * 10 16 5 <128M 1
109 * 40 16 5 900M 4
110 * 70 64 7 2-4 GB 5
111 * 84 64 7 4-8 GB 6
112 * 108 512 9 4-8 GB 6
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);
126 return threshold;
130 * Refresh the thresholds for each zone.
132 static void refresh_zone_stat_thresholds(void)
134 struct zone *zone;
135 int cpu;
136 int threshold;
138 for_each_populated_zone(zone) {
139 threshold = calculate_threshold(zone);
141 for_each_online_cpu(cpu)
142 zone_pcp(zone, cpu)->stat_threshold = threshold;
147 * For use when we know that interrupts are disabled.
149 void __mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
150 int delta)
152 struct per_cpu_pageset *pcp = zone_pcp(zone, smp_processor_id());
153 s8 *p = pcp->vm_stat_diff + item;
154 long x;
156 x = delta + *p;
158 if (unlikely(x > pcp->stat_threshold || x < -pcp->stat_threshold)) {
159 zone_page_state_add(x, zone, item);
160 x = 0;
162 *p = x;
164 EXPORT_SYMBOL(__mod_zone_page_state);
167 * For an unknown interrupt state
169 void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
170 int delta)
172 unsigned long flags;
174 local_irq_save(flags);
175 __mod_zone_page_state(zone, item, delta);
176 local_irq_restore(flags);
178 EXPORT_SYMBOL(mod_zone_page_state);
181 * Optimized increment and decrement functions.
183 * These are only for a single page and therefore can take a struct page *
184 * argument instead of struct zone *. This allows the inclusion of the code
185 * generated for page_zone(page) into the optimized functions.
187 * No overflow check is necessary and therefore the differential can be
188 * incremented or decremented in place which may allow the compilers to
189 * generate better code.
190 * The increment or decrement is known and therefore one boundary check can
191 * be omitted.
193 * NOTE: These functions are very performance sensitive. Change only
194 * with care.
196 * Some processors have inc/dec instructions that are atomic vs an interrupt.
197 * However, the code must first determine the differential location in a zone
198 * based on the processor number and then inc/dec the counter. There is no
199 * guarantee without disabling preemption that the processor will not change
200 * in between and therefore the atomicity vs. interrupt cannot be exploited
201 * in a useful way here.
203 void __inc_zone_state(struct zone *zone, enum zone_stat_item item)
205 struct per_cpu_pageset *pcp = zone_pcp(zone, smp_processor_id());
206 s8 *p = pcp->vm_stat_diff + item;
208 (*p)++;
210 if (unlikely(*p > pcp->stat_threshold)) {
211 int overstep = pcp->stat_threshold / 2;
213 zone_page_state_add(*p + overstep, zone, item);
214 *p = -overstep;
218 void __inc_zone_page_state(struct page *page, enum zone_stat_item item)
220 __inc_zone_state(page_zone(page), item);
222 EXPORT_SYMBOL(__inc_zone_page_state);
224 void __dec_zone_state(struct zone *zone, enum zone_stat_item item)
226 struct per_cpu_pageset *pcp = zone_pcp(zone, smp_processor_id());
227 s8 *p = pcp->vm_stat_diff + item;
229 (*p)--;
231 if (unlikely(*p < - pcp->stat_threshold)) {
232 int overstep = pcp->stat_threshold / 2;
234 zone_page_state_add(*p - overstep, zone, item);
235 *p = overstep;
239 void __dec_zone_page_state(struct page *page, enum zone_stat_item item)
241 __dec_zone_state(page_zone(page), item);
243 EXPORT_SYMBOL(__dec_zone_page_state);
245 void inc_zone_state(struct zone *zone, enum zone_stat_item item)
247 unsigned long flags;
249 local_irq_save(flags);
250 __inc_zone_state(zone, item);
251 local_irq_restore(flags);
254 void inc_zone_page_state(struct page *page, enum zone_stat_item item)
256 unsigned long flags;
257 struct zone *zone;
259 zone = page_zone(page);
260 local_irq_save(flags);
261 __inc_zone_state(zone, item);
262 local_irq_restore(flags);
264 EXPORT_SYMBOL(inc_zone_page_state);
266 void dec_zone_page_state(struct page *page, enum zone_stat_item item)
268 unsigned long flags;
270 local_irq_save(flags);
271 __dec_zone_page_state(page, item);
272 local_irq_restore(flags);
274 EXPORT_SYMBOL(dec_zone_page_state);
277 * Update the zone counters for one cpu.
279 * The cpu specified must be either the current cpu or a processor that
280 * is not online. If it is the current cpu then the execution thread must
281 * be pinned to the current cpu.
283 * Note that refresh_cpu_vm_stats strives to only access
284 * node local memory. The per cpu pagesets on remote zones are placed
285 * in the memory local to the processor using that pageset. So the
286 * loop over all zones will access a series of cachelines local to
287 * the processor.
289 * The call to zone_page_state_add updates the cachelines with the
290 * statistics in the remote zone struct as well as the global cachelines
291 * with the global counters. These could cause remote node cache line
292 * bouncing and will have to be only done when necessary.
294 void refresh_cpu_vm_stats(int cpu)
296 struct zone *zone;
297 int i;
298 int global_diff[NR_VM_ZONE_STAT_ITEMS] = { 0, };
300 for_each_populated_zone(zone) {
301 struct per_cpu_pageset *p;
303 p = zone_pcp(zone, cpu);
305 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
306 if (p->vm_stat_diff[i]) {
307 unsigned long flags;
308 int v;
310 local_irq_save(flags);
311 v = p->vm_stat_diff[i];
312 p->vm_stat_diff[i] = 0;
313 local_irq_restore(flags);
314 atomic_long_add(v, &zone->vm_stat[i]);
315 global_diff[i] += v;
316 #ifdef CONFIG_NUMA
317 /* 3 seconds idle till flush */
318 p->expire = 3;
319 #endif
321 cond_resched();
322 #ifdef CONFIG_NUMA
324 * Deal with draining the remote pageset of this
325 * processor
327 * Check if there are pages remaining in this pageset
328 * if not then there is nothing to expire.
330 if (!p->expire || !p->pcp.count)
331 continue;
334 * We never drain zones local to this processor.
336 if (zone_to_nid(zone) == numa_node_id()) {
337 p->expire = 0;
338 continue;
341 p->expire--;
342 if (p->expire)
343 continue;
345 if (p->pcp.count)
346 drain_zone_pages(zone, &p->pcp);
347 #endif
350 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
351 if (global_diff[i])
352 atomic_long_add(global_diff[i], &vm_stat[i]);
355 #endif
357 #ifdef CONFIG_NUMA
359 * zonelist = the list of zones passed to the allocator
360 * z = the zone from which the allocation occurred.
362 * Must be called with interrupts disabled.
364 void zone_statistics(struct zone *preferred_zone, struct zone *z)
366 if (z->zone_pgdat == preferred_zone->zone_pgdat) {
367 __inc_zone_state(z, NUMA_HIT);
368 } else {
369 __inc_zone_state(z, NUMA_MISS);
370 __inc_zone_state(preferred_zone, NUMA_FOREIGN);
372 if (z->node == numa_node_id())
373 __inc_zone_state(z, NUMA_LOCAL);
374 else
375 __inc_zone_state(z, NUMA_OTHER);
377 #endif
379 #ifdef CONFIG_PROC_FS
380 #include <linux/proc_fs.h>
381 #include <linux/seq_file.h>
383 static char * const migratetype_names[MIGRATE_TYPES] = {
384 "Unmovable",
385 "Reclaimable",
386 "Movable",
387 "Reserve",
388 "Isolate",
391 static void *frag_start(struct seq_file *m, loff_t *pos)
393 pg_data_t *pgdat;
394 loff_t node = *pos;
395 for (pgdat = first_online_pgdat();
396 pgdat && node;
397 pgdat = next_online_pgdat(pgdat))
398 --node;
400 return pgdat;
403 static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
405 pg_data_t *pgdat = (pg_data_t *)arg;
407 (*pos)++;
408 return next_online_pgdat(pgdat);
411 static void frag_stop(struct seq_file *m, void *arg)
415 /* Walk all the zones in a node and print using a callback */
416 static void walk_zones_in_node(struct seq_file *m, pg_data_t *pgdat,
417 void (*print)(struct seq_file *m, pg_data_t *, struct zone *))
419 struct zone *zone;
420 struct zone *node_zones = pgdat->node_zones;
421 unsigned long flags;
423 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
424 if (!populated_zone(zone))
425 continue;
427 spin_lock_irqsave(&zone->lock, flags);
428 print(m, pgdat, zone);
429 spin_unlock_irqrestore(&zone->lock, flags);
433 static void frag_show_print(struct seq_file *m, pg_data_t *pgdat,
434 struct zone *zone)
436 int order;
438 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
439 for (order = 0; order < MAX_ORDER; ++order)
440 seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
441 seq_putc(m, '\n');
445 * This walks the free areas for each zone.
447 static int frag_show(struct seq_file *m, void *arg)
449 pg_data_t *pgdat = (pg_data_t *)arg;
450 walk_zones_in_node(m, pgdat, frag_show_print);
451 return 0;
454 static void pagetypeinfo_showfree_print(struct seq_file *m,
455 pg_data_t *pgdat, struct zone *zone)
457 int order, mtype;
459 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++) {
460 seq_printf(m, "Node %4d, zone %8s, type %12s ",
461 pgdat->node_id,
462 zone->name,
463 migratetype_names[mtype]);
464 for (order = 0; order < MAX_ORDER; ++order) {
465 unsigned long freecount = 0;
466 struct free_area *area;
467 struct list_head *curr;
469 area = &(zone->free_area[order]);
471 list_for_each(curr, &area->free_list[mtype])
472 freecount++;
473 seq_printf(m, "%6lu ", freecount);
475 seq_putc(m, '\n');
479 /* Print out the free pages at each order for each migatetype */
480 static int pagetypeinfo_showfree(struct seq_file *m, void *arg)
482 int order;
483 pg_data_t *pgdat = (pg_data_t *)arg;
485 /* Print header */
486 seq_printf(m, "%-43s ", "Free pages count per migrate type at order");
487 for (order = 0; order < MAX_ORDER; ++order)
488 seq_printf(m, "%6d ", order);
489 seq_putc(m, '\n');
491 walk_zones_in_node(m, pgdat, pagetypeinfo_showfree_print);
493 return 0;
496 static void pagetypeinfo_showblockcount_print(struct seq_file *m,
497 pg_data_t *pgdat, struct zone *zone)
499 int mtype;
500 unsigned long pfn;
501 unsigned long start_pfn = zone->zone_start_pfn;
502 unsigned long end_pfn = start_pfn + zone->spanned_pages;
503 unsigned long count[MIGRATE_TYPES] = { 0, };
505 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
506 struct page *page;
508 if (!pfn_valid(pfn))
509 continue;
511 page = pfn_to_page(pfn);
512 #ifdef CONFIG_ARCH_FLATMEM_HAS_HOLES
514 * Ordinarily, memory holes in flatmem still have a valid
515 * memmap for the PFN range. However, an architecture for
516 * embedded systems (e.g. ARM) can free up the memmap backing
517 * holes to save memory on the assumption the memmap is
518 * never used. The page_zone linkages are then broken even
519 * though pfn_valid() returns true. Skip the page if the
520 * linkages are broken. Even if this test passed, the impact
521 * is that the counters for the movable type are off but
522 * fragmentation monitoring is likely meaningless on small
523 * systems.
525 if (page_zone(page) != zone)
526 continue;
527 #endif
528 mtype = get_pageblock_migratetype(page);
530 if (mtype < MIGRATE_TYPES)
531 count[mtype]++;
534 /* Print counts */
535 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
536 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
537 seq_printf(m, "%12lu ", count[mtype]);
538 seq_putc(m, '\n');
541 /* Print out the free pages at each order for each migratetype */
542 static int pagetypeinfo_showblockcount(struct seq_file *m, void *arg)
544 int mtype;
545 pg_data_t *pgdat = (pg_data_t *)arg;
547 seq_printf(m, "\n%-23s", "Number of blocks type ");
548 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
549 seq_printf(m, "%12s ", migratetype_names[mtype]);
550 seq_putc(m, '\n');
551 walk_zones_in_node(m, pgdat, pagetypeinfo_showblockcount_print);
553 return 0;
557 * This prints out statistics in relation to grouping pages by mobility.
558 * It is expensive to collect so do not constantly read the file.
560 static int pagetypeinfo_show(struct seq_file *m, void *arg)
562 pg_data_t *pgdat = (pg_data_t *)arg;
564 /* check memoryless node */
565 if (!node_state(pgdat->node_id, N_HIGH_MEMORY))
566 return 0;
568 seq_printf(m, "Page block order: %d\n", pageblock_order);
569 seq_printf(m, "Pages per block: %lu\n", pageblock_nr_pages);
570 seq_putc(m, '\n');
571 pagetypeinfo_showfree(m, pgdat);
572 pagetypeinfo_showblockcount(m, pgdat);
574 return 0;
577 static const struct seq_operations fragmentation_op = {
578 .start = frag_start,
579 .next = frag_next,
580 .stop = frag_stop,
581 .show = frag_show,
584 static int fragmentation_open(struct inode *inode, struct file *file)
586 return seq_open(file, &fragmentation_op);
589 static const struct file_operations fragmentation_file_operations = {
590 .open = fragmentation_open,
591 .read = seq_read,
592 .llseek = seq_lseek,
593 .release = seq_release,
596 static const struct seq_operations pagetypeinfo_op = {
597 .start = frag_start,
598 .next = frag_next,
599 .stop = frag_stop,
600 .show = pagetypeinfo_show,
603 static int pagetypeinfo_open(struct inode *inode, struct file *file)
605 return seq_open(file, &pagetypeinfo_op);
608 static const struct file_operations pagetypeinfo_file_ops = {
609 .open = pagetypeinfo_open,
610 .read = seq_read,
611 .llseek = seq_lseek,
612 .release = seq_release,
615 #ifdef CONFIG_ZONE_DMA
616 #define TEXT_FOR_DMA(xx) xx "_dma",
617 #else
618 #define TEXT_FOR_DMA(xx)
619 #endif
621 #ifdef CONFIG_ZONE_DMA32
622 #define TEXT_FOR_DMA32(xx) xx "_dma32",
623 #else
624 #define TEXT_FOR_DMA32(xx)
625 #endif
627 #ifdef CONFIG_HIGHMEM
628 #define TEXT_FOR_HIGHMEM(xx) xx "_high",
629 #else
630 #define TEXT_FOR_HIGHMEM(xx)
631 #endif
633 #define TEXTS_FOR_ZONES(xx) TEXT_FOR_DMA(xx) TEXT_FOR_DMA32(xx) xx "_normal", \
634 TEXT_FOR_HIGHMEM(xx) xx "_movable",
636 static const char * const vmstat_text[] = {
637 /* Zoned VM counters */
638 "nr_free_pages",
639 "nr_inactive_anon",
640 "nr_active_anon",
641 "nr_inactive_file",
642 "nr_active_file",
643 #ifdef CONFIG_UNEVICTABLE_LRU
644 "nr_unevictable",
645 "nr_mlock",
646 #endif
647 "nr_anon_pages",
648 "nr_mapped",
649 "nr_file_pages",
650 "nr_dirty",
651 "nr_writeback",
652 "nr_slab_reclaimable",
653 "nr_slab_unreclaimable",
654 "nr_page_table_pages",
655 "nr_unstable",
656 "nr_bounce",
657 "nr_vmscan_write",
658 "nr_writeback_temp",
660 #ifdef CONFIG_NUMA
661 "numa_hit",
662 "numa_miss",
663 "numa_foreign",
664 "numa_interleave",
665 "numa_local",
666 "numa_other",
667 #endif
669 #ifdef CONFIG_VM_EVENT_COUNTERS
670 "pgpgin",
671 "pgpgout",
672 "pswpin",
673 "pswpout",
675 TEXTS_FOR_ZONES("pgalloc")
677 "pgfree",
678 "pgactivate",
679 "pgdeactivate",
681 "pgfault",
682 "pgmajfault",
684 TEXTS_FOR_ZONES("pgrefill")
685 TEXTS_FOR_ZONES("pgsteal")
686 TEXTS_FOR_ZONES("pgscan_kswapd")
687 TEXTS_FOR_ZONES("pgscan_direct")
689 "pginodesteal",
690 "slabs_scanned",
691 "kswapd_steal",
692 "kswapd_inodesteal",
693 "pageoutrun",
694 "allocstall",
696 "pgrotated",
697 #ifdef CONFIG_HUGETLB_PAGE
698 "htlb_buddy_alloc_success",
699 "htlb_buddy_alloc_fail",
700 #endif
701 #ifdef CONFIG_UNEVICTABLE_LRU
702 "unevictable_pgs_culled",
703 "unevictable_pgs_scanned",
704 "unevictable_pgs_rescued",
705 "unevictable_pgs_mlocked",
706 "unevictable_pgs_munlocked",
707 "unevictable_pgs_cleared",
708 "unevictable_pgs_stranded",
709 "unevictable_pgs_mlockfreed",
710 #endif
711 #endif
714 static void zoneinfo_show_print(struct seq_file *m, pg_data_t *pgdat,
715 struct zone *zone)
717 int i;
718 seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
719 seq_printf(m,
720 "\n pages free %lu"
721 "\n min %lu"
722 "\n low %lu"
723 "\n high %lu"
724 "\n scanned %lu (aa: %lu ia: %lu af: %lu if: %lu)"
725 "\n spanned %lu"
726 "\n present %lu",
727 zone_page_state(zone, NR_FREE_PAGES),
728 zone->pages_min,
729 zone->pages_low,
730 zone->pages_high,
731 zone->pages_scanned,
732 zone->lru[LRU_ACTIVE_ANON].nr_scan,
733 zone->lru[LRU_INACTIVE_ANON].nr_scan,
734 zone->lru[LRU_ACTIVE_FILE].nr_scan,
735 zone->lru[LRU_INACTIVE_FILE].nr_scan,
736 zone->spanned_pages,
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));
743 seq_printf(m,
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]);
748 seq_printf(m,
750 "\n pagesets");
751 for_each_online_cpu(i) {
752 struct per_cpu_pageset *pageset;
754 pageset = zone_pcp(zone, i);
755 seq_printf(m,
756 "\n cpu: %i"
757 "\n count: %i"
758 "\n high: %i"
759 "\n batch: %i",
761 pageset->pcp.count,
762 pageset->pcp.high,
763 pageset->pcp.batch);
764 #ifdef CONFIG_SMP
765 seq_printf(m, "\n vm stats threshold: %d",
766 pageset->stat_threshold);
767 #endif
769 seq_printf(m,
770 "\n all_unreclaimable: %u"
771 "\n prev_priority: %i"
772 "\n start_pfn: %lu"
773 "\n inactive_ratio: %u",
774 zone_is_all_unreclaimable(zone),
775 zone->prev_priority,
776 zone->zone_start_pfn,
777 zone->inactive_ratio);
778 seq_putc(m, '\n');
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);
788 return 0;
791 static const struct seq_operations zoneinfo_op = {
792 .start = frag_start, /* iterate over all zones. The same as in
793 * fragmentation. */
794 .next = frag_next,
795 .stop = frag_stop,
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,
806 .read = seq_read,
807 .llseek = seq_lseek,
808 .release = seq_release,
811 static void *vmstat_start(struct seq_file *m, loff_t *pos)
813 unsigned long *v;
814 #ifdef CONFIG_VM_EVENT_COUNTERS
815 unsigned long *e;
816 #endif
817 int i;
819 if (*pos >= ARRAY_SIZE(vmstat_text))
820 return NULL;
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);
825 #else
826 v = kmalloc(NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long),
827 GFP_KERNEL);
828 #endif
829 m->private = v;
830 if (!v)
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;
836 all_vm_events(e);
837 e[PGPGIN] /= 2; /* sectors -> kbytes */
838 e[PGPGOUT] /= 2;
839 #endif
840 return v + *pos;
843 static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
845 (*pos)++;
846 if (*pos >= ARRAY_SIZE(vmstat_text))
847 return NULL;
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);
857 return 0;
860 static void vmstat_stop(struct seq_file *m, void *arg)
862 kfree(m->private);
863 m->private = NULL;
866 static const struct seq_operations vmstat_op = {
867 .start = vmstat_start,
868 .next = vmstat_next,
869 .stop = vmstat_stop,
870 .show = vmstat_show,
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 = {
879 .open = vmstat_open,
880 .read = seq_read,
881 .llseek = seq_lseek,
882 .release = seq_release,
884 #endif /* CONFIG_PROC_FS */
886 #ifdef CONFIG_SMP
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
908 * when necessary.
910 static int __cpuinit vmstat_cpuup_callback(struct notifier_block *nfb,
911 unsigned long action,
912 void *hcpu)
914 long cpu = (long)hcpu;
916 switch (action) {
917 case CPU_ONLINE:
918 case CPU_ONLINE_FROZEN:
919 start_cpu_timer(cpu);
920 break;
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;
925 break;
926 case CPU_DOWN_FAILED:
927 case CPU_DOWN_FAILED_FROZEN:
928 start_cpu_timer(cpu);
929 break;
930 case CPU_DEAD:
931 case CPU_DEAD_FROZEN:
932 refresh_zone_stat_thresholds();
933 break;
934 default:
935 break;
937 return NOTIFY_OK;
940 static struct notifier_block __cpuinitdata vmstat_notifier =
941 { &vmstat_cpuup_callback, NULL, 0 };
942 #endif
944 static int __init setup_vmstat(void)
946 #ifdef CONFIG_SMP
947 int cpu;
949 refresh_zone_stat_thresholds();
950 register_cpu_notifier(&vmstat_notifier);
952 for_each_online_cpu(cpu)
953 start_cpu_timer(cpu);
954 #endif
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);
960 #endif
961 return 0;
963 module_init(setup_vmstat)