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