vmstats: add cond_resched() to refresh_cpu_vm_stats()
[linux-2.6.git] / mm / vmstat.c
1 /*
2  *  linux/mm/vmstat.c
3  *
4  *  Manages VM statistics
5  *  Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
6  *
7  *  zoned VM statistics
8  *  Copyright (C) 2006 Silicon Graphics, Inc.,
9  *              Christoph Lameter <christoph@lameter.com>
10  */
11
12 #include <linux/mm.h>
13 #include <linux/err.h>
14 #include <linux/module.h>
15 #include <linux/cpu.h>
16 #include <linux/sched.h>
17
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);
21
22 static void sum_vm_events(unsigned long *ret, cpumask_t *cpumask)
23 {
24         int cpu;
25         int i;
26
27         memset(ret, 0, NR_VM_EVENT_ITEMS * sizeof(unsigned long));
28
29         for_each_cpu_mask(cpu, *cpumask) {
30                 struct vm_event_state *this = &per_cpu(vm_event_states, cpu);
31
32                 for (i = 0; i < NR_VM_EVENT_ITEMS; i++)
33                         ret[i] += this->event[i];
34         }
35 }
36
37 /*
38  * Accumulate the vm event counters across all CPUs.
39  * The result is unavoidably approximate - it can change
40  * during and after execution of this function.
41 */
42 void all_vm_events(unsigned long *ret)
43 {
44         sum_vm_events(ret, &cpu_online_map);
45 }
46 EXPORT_SYMBOL_GPL(all_vm_events);
47
48 #ifdef CONFIG_HOTPLUG
49 /*
50  * Fold the foreign cpu events into our own.
51  *
52  * This is adding to the events on one processor
53  * but keeps the global counts constant.
54  */
55 void vm_events_fold_cpu(int cpu)
56 {
57         struct vm_event_state *fold_state = &per_cpu(vm_event_states, cpu);
58         int i;
59
60         for (i = 0; i < NR_VM_EVENT_ITEMS; i++) {
61                 count_vm_events(i, fold_state->event[i]);
62                 fold_state->event[i] = 0;
63         }
64 }
65 #endif /* CONFIG_HOTPLUG */
66
67 #endif /* CONFIG_VM_EVENT_COUNTERS */
68
69 /*
70  * Manage combined zone based / global counters
71  *
72  * vm_stat contains the global counters
73  */
74 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
75 EXPORT_SYMBOL(vm_stat);
76
77 #ifdef CONFIG_SMP
78
79 static int calculate_threshold(struct zone *zone)
80 {
81         int threshold;
82         int mem;        /* memory in 128 MB units */
83
84         /*
85          * The threshold scales with the number of processors and the amount
86          * of memory per zone. More memory means that we can defer updates for
87          * longer, more processors could lead to more contention.
88          * fls() is used to have a cheap way of logarithmic scaling.
89          *
90          * Some sample thresholds:
91          *
92          * Threshold    Processors      (fls)   Zonesize        fls(mem+1)
93          * ------------------------------------------------------------------
94          * 8            1               1       0.9-1 GB        4
95          * 16           2               2       0.9-1 GB        4
96          * 20           2               2       1-2 GB          5
97          * 24           2               2       2-4 GB          6
98          * 28           2               2       4-8 GB          7
99          * 32           2               2       8-16 GB         8
100          * 4            2               2       <128M           1
101          * 30           4               3       2-4 GB          5
102          * 48           4               3       8-16 GB         8
103          * 32           8               4       1-2 GB          4
104          * 32           8               4       0.9-1GB         4
105          * 10           16              5       <128M           1
106          * 40           16              5       900M            4
107          * 70           64              7       2-4 GB          5
108          * 84           64              7       4-8 GB          6
109          * 108          512             9       4-8 GB          6
110          * 125          1024            10      8-16 GB         8
111          * 125          1024            10      16-32 GB        9
112          */
113
114         mem = zone->present_pages >> (27 - PAGE_SHIFT);
115
116         threshold = 2 * fls(num_online_cpus()) * (1 + fls(mem));
117
118         /*
119          * Maximum threshold is 125
120          */
121         threshold = min(125, threshold);
122
123         return threshold;
124 }
125
126 /*
127  * Refresh the thresholds for each zone.
128  */
129 static void refresh_zone_stat_thresholds(void)
130 {
131         struct zone *zone;
132         int cpu;
133         int threshold;
134
135         for_each_zone(zone) {
136
137                 if (!zone->present_pages)
138                         continue;
139
140                 threshold = calculate_threshold(zone);
141
142                 for_each_online_cpu(cpu)
143                         zone_pcp(zone, cpu)->stat_threshold = threshold;
144         }
145 }
146
147 /*
148  * For use when we know that interrupts are disabled.
149  */
150 void __mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
151                                 int delta)
152 {
153         struct per_cpu_pageset *pcp = zone_pcp(zone, smp_processor_id());
154         s8 *p = pcp->vm_stat_diff + item;
155         long x;
156
157         x = delta + *p;
158
159         if (unlikely(x > pcp->stat_threshold || x < -pcp->stat_threshold)) {
160                 zone_page_state_add(x, zone, item);
161                 x = 0;
162         }
163         *p = x;
164 }
165 EXPORT_SYMBOL(__mod_zone_page_state);
166
167 /*
168  * For an unknown interrupt state
169  */
170 void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
171                                         int delta)
172 {
173         unsigned long flags;
174
175         local_irq_save(flags);
176         __mod_zone_page_state(zone, item, delta);
177         local_irq_restore(flags);
178 }
179 EXPORT_SYMBOL(mod_zone_page_state);
180
181 /*
182  * Optimized increment and decrement functions.
183  *
184  * These are only for a single page and therefore can take a struct page *
185  * argument instead of struct zone *. This allows the inclusion of the code
186  * generated for page_zone(page) into the optimized functions.
187  *
188  * No overflow check is necessary and therefore the differential can be
189  * incremented or decremented in place which may allow the compilers to
190  * generate better code.
191  * The increment or decrement is known and therefore one boundary check can
192  * be omitted.
193  *
194  * NOTE: These functions are very performance sensitive. Change only
195  * with care.
196  *
197  * Some processors have inc/dec instructions that are atomic vs an interrupt.
198  * However, the code must first determine the differential location in a zone
199  * based on the processor number and then inc/dec the counter. There is no
200  * guarantee without disabling preemption that the processor will not change
201  * in between and therefore the atomicity vs. interrupt cannot be exploited
202  * in a useful way here.
203  */
204 void __inc_zone_state(struct zone *zone, enum zone_stat_item item)
205 {
206         struct per_cpu_pageset *pcp = zone_pcp(zone, smp_processor_id());
207         s8 *p = pcp->vm_stat_diff + item;
208
209         (*p)++;
210
211         if (unlikely(*p > pcp->stat_threshold)) {
212                 int overstep = pcp->stat_threshold / 2;
213
214                 zone_page_state_add(*p + overstep, zone, item);
215                 *p = -overstep;
216         }
217 }
218
219 void __inc_zone_page_state(struct page *page, enum zone_stat_item item)
220 {
221         __inc_zone_state(page_zone(page), item);
222 }
223 EXPORT_SYMBOL(__inc_zone_page_state);
224
225 void __dec_zone_state(struct zone *zone, enum zone_stat_item item)
226 {
227         struct per_cpu_pageset *pcp = zone_pcp(zone, smp_processor_id());
228         s8 *p = pcp->vm_stat_diff + item;
229
230         (*p)--;
231
232         if (unlikely(*p < - pcp->stat_threshold)) {
233                 int overstep = pcp->stat_threshold / 2;
234
235                 zone_page_state_add(*p - overstep, zone, item);
236                 *p = overstep;
237         }
238 }
239
240 void __dec_zone_page_state(struct page *page, enum zone_stat_item item)
241 {
242         __dec_zone_state(page_zone(page), item);
243 }
244 EXPORT_SYMBOL(__dec_zone_page_state);
245
246 void inc_zone_state(struct zone *zone, enum zone_stat_item item)
247 {
248         unsigned long flags;
249
250         local_irq_save(flags);
251         __inc_zone_state(zone, item);
252         local_irq_restore(flags);
253 }
254
255 void inc_zone_page_state(struct page *page, enum zone_stat_item item)
256 {
257         unsigned long flags;
258         struct zone *zone;
259
260         zone = page_zone(page);
261         local_irq_save(flags);
262         __inc_zone_state(zone, item);
263         local_irq_restore(flags);
264 }
265 EXPORT_SYMBOL(inc_zone_page_state);
266
267 void dec_zone_page_state(struct page *page, enum zone_stat_item item)
268 {
269         unsigned long flags;
270
271         local_irq_save(flags);
272         __dec_zone_page_state(page, item);
273         local_irq_restore(flags);
274 }
275 EXPORT_SYMBOL(dec_zone_page_state);
276
277 /*
278  * Update the zone counters for one cpu.
279  *
280  * The cpu specified must be either the current cpu or a processor that
281  * is not online. If it is the current cpu then the execution thread must
282  * be pinned to the current cpu.
283  *
284  * Note that refresh_cpu_vm_stats strives to only access
285  * node local memory. The per cpu pagesets on remote zones are placed
286  * in the memory local to the processor using that pageset. So the
287  * loop over all zones will access a series of cachelines local to
288  * the processor.
289  *
290  * The call to zone_page_state_add updates the cachelines with the
291  * statistics in the remote zone struct as well as the global cachelines
292  * with the global counters. These could cause remote node cache line
293  * bouncing and will have to be only done when necessary.
294  */
295 void refresh_cpu_vm_stats(int cpu)
296 {
297         struct zone *zone;
298         int i;
299         int global_diff[NR_VM_ZONE_STAT_ITEMS] = { 0, };
300
301         for_each_zone(zone) {
302                 struct per_cpu_pageset *p;
303
304                 if (!populated_zone(zone))
305                         continue;
306
307                 p = zone_pcp(zone, cpu);
308
309                 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
310                         if (p->vm_stat_diff[i]) {
311                                 unsigned long flags;
312                                 int v;
313
314                                 local_irq_save(flags);
315                                 v = p->vm_stat_diff[i];
316                                 p->vm_stat_diff[i] = 0;
317                                 local_irq_restore(flags);
318                                 atomic_long_add(v, &zone->vm_stat[i]);
319                                 global_diff[i] += v;
320 #ifdef CONFIG_NUMA
321                                 /* 3 seconds idle till flush */
322                                 p->expire = 3;
323 #endif
324                         }
325                 cond_resched();
326 #ifdef CONFIG_NUMA
327                 /*
328                  * Deal with draining the remote pageset of this
329                  * processor
330                  *
331                  * Check if there are pages remaining in this pageset
332                  * if not then there is nothing to expire.
333                  */
334                 if (!p->expire || !p->pcp.count)
335                         continue;
336
337                 /*
338                  * We never drain zones local to this processor.
339                  */
340                 if (zone_to_nid(zone) == numa_node_id()) {
341                         p->expire = 0;
342                         continue;
343                 }
344
345                 p->expire--;
346                 if (p->expire)
347                         continue;
348
349                 if (p->pcp.count)
350                         drain_zone_pages(zone, &p->pcp);
351 #endif
352         }
353
354         for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
355                 if (global_diff[i])
356                         atomic_long_add(global_diff[i], &vm_stat[i]);
357 }
358
359 #endif
360
361 #ifdef CONFIG_NUMA
362 /*
363  * zonelist = the list of zones passed to the allocator
364  * z        = the zone from which the allocation occurred.
365  *
366  * Must be called with interrupts disabled.
367  */
368 void zone_statistics(struct zone *preferred_zone, struct zone *z)
369 {
370         if (z->zone_pgdat == preferred_zone->zone_pgdat) {
371                 __inc_zone_state(z, NUMA_HIT);
372         } else {
373                 __inc_zone_state(z, NUMA_MISS);
374                 __inc_zone_state(preferred_zone, NUMA_FOREIGN);
375         }
376         if (z->node == numa_node_id())
377                 __inc_zone_state(z, NUMA_LOCAL);
378         else
379                 __inc_zone_state(z, NUMA_OTHER);
380 }
381 #endif
382
383 #ifdef CONFIG_PROC_FS
384
385 #include <linux/seq_file.h>
386
387 static char * const migratetype_names[MIGRATE_TYPES] = {
388         "Unmovable",
389         "Reclaimable",
390         "Movable",
391         "Reserve",
392         "Isolate",
393 };
394
395 static void *frag_start(struct seq_file *m, loff_t *pos)
396 {
397         pg_data_t *pgdat;
398         loff_t node = *pos;
399         for (pgdat = first_online_pgdat();
400              pgdat && node;
401              pgdat = next_online_pgdat(pgdat))
402                 --node;
403
404         return pgdat;
405 }
406
407 static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
408 {
409         pg_data_t *pgdat = (pg_data_t *)arg;
410
411         (*pos)++;
412         return next_online_pgdat(pgdat);
413 }
414
415 static void frag_stop(struct seq_file *m, void *arg)
416 {
417 }
418
419 /* Walk all the zones in a node and print using a callback */
420 static void walk_zones_in_node(struct seq_file *m, pg_data_t *pgdat,
421                 void (*print)(struct seq_file *m, pg_data_t *, struct zone *))
422 {
423         struct zone *zone;
424         struct zone *node_zones = pgdat->node_zones;
425         unsigned long flags;
426
427         for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
428                 if (!populated_zone(zone))
429                         continue;
430
431                 spin_lock_irqsave(&zone->lock, flags);
432                 print(m, pgdat, zone);
433                 spin_unlock_irqrestore(&zone->lock, flags);
434         }
435 }
436
437 static void frag_show_print(struct seq_file *m, pg_data_t *pgdat,
438                                                 struct zone *zone)
439 {
440         int order;
441
442         seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
443         for (order = 0; order < MAX_ORDER; ++order)
444                 seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
445         seq_putc(m, '\n');
446 }
447
448 /*
449  * This walks the free areas for each zone.
450  */
451 static int frag_show(struct seq_file *m, void *arg)
452 {
453         pg_data_t *pgdat = (pg_data_t *)arg;
454         walk_zones_in_node(m, pgdat, frag_show_print);
455         return 0;
456 }
457
458 static void pagetypeinfo_showfree_print(struct seq_file *m,
459                                         pg_data_t *pgdat, struct zone *zone)
460 {
461         int order, mtype;
462
463         for (mtype = 0; mtype < MIGRATE_TYPES; mtype++) {
464                 seq_printf(m, "Node %4d, zone %8s, type %12s ",
465                                         pgdat->node_id,
466                                         zone->name,
467                                         migratetype_names[mtype]);
468                 for (order = 0; order < MAX_ORDER; ++order) {
469                         unsigned long freecount = 0;
470                         struct free_area *area;
471                         struct list_head *curr;
472
473                         area = &(zone->free_area[order]);
474
475                         list_for_each(curr, &area->free_list[mtype])
476                                 freecount++;
477                         seq_printf(m, "%6lu ", freecount);
478                 }
479                 seq_putc(m, '\n');
480         }
481 }
482
483 /* Print out the free pages at each order for each migatetype */
484 static int pagetypeinfo_showfree(struct seq_file *m, void *arg)
485 {
486         int order;
487         pg_data_t *pgdat = (pg_data_t *)arg;
488
489         /* Print header */
490         seq_printf(m, "%-43s ", "Free pages count per migrate type at order");
491         for (order = 0; order < MAX_ORDER; ++order)
492                 seq_printf(m, "%6d ", order);
493         seq_putc(m, '\n');
494
495         walk_zones_in_node(m, pgdat, pagetypeinfo_showfree_print);
496
497         return 0;
498 }
499
500 static void pagetypeinfo_showblockcount_print(struct seq_file *m,
501                                         pg_data_t *pgdat, struct zone *zone)
502 {
503         int mtype;
504         unsigned long pfn;
505         unsigned long start_pfn = zone->zone_start_pfn;
506         unsigned long end_pfn = start_pfn + zone->spanned_pages;
507         unsigned long count[MIGRATE_TYPES] = { 0, };
508
509         for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
510                 struct page *page;
511
512                 if (!pfn_valid(pfn))
513                         continue;
514
515                 page = pfn_to_page(pfn);
516                 mtype = get_pageblock_migratetype(page);
517
518                 count[mtype]++;
519         }
520
521         /* Print counts */
522         seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
523         for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
524                 seq_printf(m, "%12lu ", count[mtype]);
525         seq_putc(m, '\n');
526 }
527
528 /* Print out the free pages at each order for each migratetype */
529 static int pagetypeinfo_showblockcount(struct seq_file *m, void *arg)
530 {
531         int mtype;
532         pg_data_t *pgdat = (pg_data_t *)arg;
533
534         seq_printf(m, "\n%-23s", "Number of blocks type ");
535         for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
536                 seq_printf(m, "%12s ", migratetype_names[mtype]);
537         seq_putc(m, '\n');
538         walk_zones_in_node(m, pgdat, pagetypeinfo_showblockcount_print);
539
540         return 0;
541 }
542
543 /*
544  * This prints out statistics in relation to grouping pages by mobility.
545  * It is expensive to collect so do not constantly read the file.
546  */
547 static int pagetypeinfo_show(struct seq_file *m, void *arg)
548 {
549         pg_data_t *pgdat = (pg_data_t *)arg;
550
551         seq_printf(m, "Page block order: %d\n", pageblock_order);
552         seq_printf(m, "Pages per block:  %lu\n", pageblock_nr_pages);
553         seq_putc(m, '\n');
554         pagetypeinfo_showfree(m, pgdat);
555         pagetypeinfo_showblockcount(m, pgdat);
556
557         return 0;
558 }
559
560 const struct seq_operations fragmentation_op = {
561         .start  = frag_start,
562         .next   = frag_next,
563         .stop   = frag_stop,
564         .show   = frag_show,
565 };
566
567 const struct seq_operations pagetypeinfo_op = {
568         .start  = frag_start,
569         .next   = frag_next,
570         .stop   = frag_stop,
571         .show   = pagetypeinfo_show,
572 };
573
574 #ifdef CONFIG_ZONE_DMA
575 #define TEXT_FOR_DMA(xx) xx "_dma",
576 #else
577 #define TEXT_FOR_DMA(xx)
578 #endif
579
580 #ifdef CONFIG_ZONE_DMA32
581 #define TEXT_FOR_DMA32(xx) xx "_dma32",
582 #else
583 #define TEXT_FOR_DMA32(xx)
584 #endif
585
586 #ifdef CONFIG_HIGHMEM
587 #define TEXT_FOR_HIGHMEM(xx) xx "_high",
588 #else
589 #define TEXT_FOR_HIGHMEM(xx)
590 #endif
591
592 #define TEXTS_FOR_ZONES(xx) TEXT_FOR_DMA(xx) TEXT_FOR_DMA32(xx) xx "_normal", \
593                                         TEXT_FOR_HIGHMEM(xx) xx "_movable",
594
595 static const char * const vmstat_text[] = {
596         /* Zoned VM counters */
597         "nr_free_pages",
598         "nr_inactive",
599         "nr_active",
600         "nr_anon_pages",
601         "nr_mapped",
602         "nr_file_pages",
603         "nr_dirty",
604         "nr_writeback",
605         "nr_slab_reclaimable",
606         "nr_slab_unreclaimable",
607         "nr_page_table_pages",
608         "nr_unstable",
609         "nr_bounce",
610         "nr_vmscan_write",
611
612 #ifdef CONFIG_NUMA
613         "numa_hit",
614         "numa_miss",
615         "numa_foreign",
616         "numa_interleave",
617         "numa_local",
618         "numa_other",
619 #endif
620
621 #ifdef CONFIG_VM_EVENT_COUNTERS
622         "pgpgin",
623         "pgpgout",
624         "pswpin",
625         "pswpout",
626
627         TEXTS_FOR_ZONES("pgalloc")
628
629         "pgfree",
630         "pgactivate",
631         "pgdeactivate",
632
633         "pgfault",
634         "pgmajfault",
635
636         TEXTS_FOR_ZONES("pgrefill")
637         TEXTS_FOR_ZONES("pgsteal")
638         TEXTS_FOR_ZONES("pgscan_kswapd")
639         TEXTS_FOR_ZONES("pgscan_direct")
640
641         "pginodesteal",
642         "slabs_scanned",
643         "kswapd_steal",
644         "kswapd_inodesteal",
645         "pageoutrun",
646         "allocstall",
647
648         "pgrotated",
649 #ifdef CONFIG_HUGETLB_PAGE
650         "htlb_buddy_alloc_success",
651         "htlb_buddy_alloc_fail",
652 #endif
653 #endif
654 };
655
656 static void zoneinfo_show_print(struct seq_file *m, pg_data_t *pgdat,
657                                                         struct zone *zone)
658 {
659         int i;
660         seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
661         seq_printf(m,
662                    "\n  pages free     %lu"
663                    "\n        min      %lu"
664                    "\n        low      %lu"
665                    "\n        high     %lu"
666                    "\n        scanned  %lu (a: %lu i: %lu)"
667                    "\n        spanned  %lu"
668                    "\n        present  %lu",
669                    zone_page_state(zone, NR_FREE_PAGES),
670                    zone->pages_min,
671                    zone->pages_low,
672                    zone->pages_high,
673                    zone->pages_scanned,
674                    zone->nr_scan_active, zone->nr_scan_inactive,
675                    zone->spanned_pages,
676                    zone->present_pages);
677
678         for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
679                 seq_printf(m, "\n    %-12s %lu", vmstat_text[i],
680                                 zone_page_state(zone, i));
681
682         seq_printf(m,
683                    "\n        protection: (%lu",
684                    zone->lowmem_reserve[0]);
685         for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
686                 seq_printf(m, ", %lu", zone->lowmem_reserve[i]);
687         seq_printf(m,
688                    ")"
689                    "\n  pagesets");
690         for_each_online_cpu(i) {
691                 struct per_cpu_pageset *pageset;
692
693                 pageset = zone_pcp(zone, i);
694                 seq_printf(m,
695                            "\n    cpu: %i"
696                            "\n              count: %i"
697                            "\n              high:  %i"
698                            "\n              batch: %i",
699                            i,
700                            pageset->pcp.count,
701                            pageset->pcp.high,
702                            pageset->pcp.batch);
703 #ifdef CONFIG_SMP
704                 seq_printf(m, "\n  vm stats threshold: %d",
705                                 pageset->stat_threshold);
706 #endif
707         }
708         seq_printf(m,
709                    "\n  all_unreclaimable: %u"
710                    "\n  prev_priority:     %i"
711                    "\n  start_pfn:         %lu",
712                            zone_is_all_unreclaimable(zone),
713                    zone->prev_priority,
714                    zone->zone_start_pfn);
715         seq_putc(m, '\n');
716 }
717
718 /*
719  * Output information about zones in @pgdat.
720  */
721 static int zoneinfo_show(struct seq_file *m, void *arg)
722 {
723         pg_data_t *pgdat = (pg_data_t *)arg;
724         walk_zones_in_node(m, pgdat, zoneinfo_show_print);
725         return 0;
726 }
727
728 const struct seq_operations zoneinfo_op = {
729         .start  = frag_start, /* iterate over all zones. The same as in
730                                * fragmentation. */
731         .next   = frag_next,
732         .stop   = frag_stop,
733         .show   = zoneinfo_show,
734 };
735
736 static void *vmstat_start(struct seq_file *m, loff_t *pos)
737 {
738         unsigned long *v;
739 #ifdef CONFIG_VM_EVENT_COUNTERS
740         unsigned long *e;
741 #endif
742         int i;
743
744         if (*pos >= ARRAY_SIZE(vmstat_text))
745                 return NULL;
746
747 #ifdef CONFIG_VM_EVENT_COUNTERS
748         v = kmalloc(NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long)
749                         + sizeof(struct vm_event_state), GFP_KERNEL);
750 #else
751         v = kmalloc(NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long),
752                         GFP_KERNEL);
753 #endif
754         m->private = v;
755         if (!v)
756                 return ERR_PTR(-ENOMEM);
757         for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
758                 v[i] = global_page_state(i);
759 #ifdef CONFIG_VM_EVENT_COUNTERS
760         e = v + NR_VM_ZONE_STAT_ITEMS;
761         all_vm_events(e);
762         e[PGPGIN] /= 2;         /* sectors -> kbytes */
763         e[PGPGOUT] /= 2;
764 #endif
765         return v + *pos;
766 }
767
768 static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
769 {
770         (*pos)++;
771         if (*pos >= ARRAY_SIZE(vmstat_text))
772                 return NULL;
773         return (unsigned long *)m->private + *pos;
774 }
775
776 static int vmstat_show(struct seq_file *m, void *arg)
777 {
778         unsigned long *l = arg;
779         unsigned long off = l - (unsigned long *)m->private;
780
781         seq_printf(m, "%s %lu\n", vmstat_text[off], *l);
782         return 0;
783 }
784
785 static void vmstat_stop(struct seq_file *m, void *arg)
786 {
787         kfree(m->private);
788         m->private = NULL;
789 }
790
791 const struct seq_operations vmstat_op = {
792         .start  = vmstat_start,
793         .next   = vmstat_next,
794         .stop   = vmstat_stop,
795         .show   = vmstat_show,
796 };
797
798 #endif /* CONFIG_PROC_FS */
799
800 #ifdef CONFIG_SMP
801 static DEFINE_PER_CPU(struct delayed_work, vmstat_work);
802 int sysctl_stat_interval __read_mostly = HZ;
803
804 static void vmstat_update(struct work_struct *w)
805 {
806         refresh_cpu_vm_stats(smp_processor_id());
807         schedule_delayed_work(&__get_cpu_var(vmstat_work),
808                 sysctl_stat_interval);
809 }
810
811 static void __cpuinit start_cpu_timer(int cpu)
812 {
813         struct delayed_work *vmstat_work = &per_cpu(vmstat_work, cpu);
814
815         INIT_DELAYED_WORK_DEFERRABLE(vmstat_work, vmstat_update);
816         schedule_delayed_work_on(cpu, vmstat_work, HZ + cpu);
817 }
818
819 /*
820  * Use the cpu notifier to insure that the thresholds are recalculated
821  * when necessary.
822  */
823 static int __cpuinit vmstat_cpuup_callback(struct notifier_block *nfb,
824                 unsigned long action,
825                 void *hcpu)
826 {
827         long cpu = (long)hcpu;
828
829         switch (action) {
830         case CPU_ONLINE:
831         case CPU_ONLINE_FROZEN:
832                 start_cpu_timer(cpu);
833                 break;
834         case CPU_DOWN_PREPARE:
835         case CPU_DOWN_PREPARE_FROZEN:
836                 cancel_rearming_delayed_work(&per_cpu(vmstat_work, cpu));
837                 per_cpu(vmstat_work, cpu).work.func = NULL;
838                 break;
839         case CPU_DOWN_FAILED:
840         case CPU_DOWN_FAILED_FROZEN:
841                 start_cpu_timer(cpu);
842                 break;
843         case CPU_DEAD:
844         case CPU_DEAD_FROZEN:
845                 refresh_zone_stat_thresholds();
846                 break;
847         default:
848                 break;
849         }
850         return NOTIFY_OK;
851 }
852
853 static struct notifier_block __cpuinitdata vmstat_notifier =
854         { &vmstat_cpuup_callback, NULL, 0 };
855
856 static int __init setup_vmstat(void)
857 {
858         int cpu;
859
860         refresh_zone_stat_thresholds();
861         register_cpu_notifier(&vmstat_notifier);
862
863         for_each_online_cpu(cpu)
864                 start_cpu_timer(cpu);
865         return 0;
866 }
867 module_init(setup_vmstat)
868 #endif