[PATCH] md: remove the working_disks and failed_disks from raid5 state data.
[linux-2.6] / 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/config.h>
13 #include <linux/mm.h>
14 #include <linux/module.h>
15 #include <linux/cpu.h>
16
17 void __get_zone_counts(unsigned long *active, unsigned long *inactive,
18                         unsigned long *free, struct pglist_data *pgdat)
19 {
20         struct zone *zones = pgdat->node_zones;
21         int i;
22
23         *active = 0;
24         *inactive = 0;
25         *free = 0;
26         for (i = 0; i < MAX_NR_ZONES; i++) {
27                 *active += zones[i].nr_active;
28                 *inactive += zones[i].nr_inactive;
29                 *free += zones[i].free_pages;
30         }
31 }
32
33 void get_zone_counts(unsigned long *active,
34                 unsigned long *inactive, unsigned long *free)
35 {
36         struct pglist_data *pgdat;
37
38         *active = 0;
39         *inactive = 0;
40         *free = 0;
41         for_each_online_pgdat(pgdat) {
42                 unsigned long l, m, n;
43                 __get_zone_counts(&l, &m, &n, pgdat);
44                 *active += l;
45                 *inactive += m;
46                 *free += n;
47         }
48 }
49
50 #ifdef CONFIG_VM_EVENT_COUNTERS
51 DEFINE_PER_CPU(struct vm_event_state, vm_event_states) = {{0}};
52 EXPORT_PER_CPU_SYMBOL(vm_event_states);
53
54 static void sum_vm_events(unsigned long *ret, cpumask_t *cpumask)
55 {
56         int cpu = 0;
57         int i;
58
59         memset(ret, 0, NR_VM_EVENT_ITEMS * sizeof(unsigned long));
60
61         cpu = first_cpu(*cpumask);
62         while (cpu < NR_CPUS) {
63                 struct vm_event_state *this = &per_cpu(vm_event_states, cpu);
64
65                 cpu = next_cpu(cpu, *cpumask);
66
67                 if (cpu < NR_CPUS)
68                         prefetch(&per_cpu(vm_event_states, cpu));
69
70
71                 for (i = 0; i < NR_VM_EVENT_ITEMS; i++)
72                         ret[i] += this->event[i];
73         }
74 }
75
76 /*
77  * Accumulate the vm event counters across all CPUs.
78  * The result is unavoidably approximate - it can change
79  * during and after execution of this function.
80 */
81 void all_vm_events(unsigned long *ret)
82 {
83         sum_vm_events(ret, &cpu_online_map);
84 }
85 EXPORT_SYMBOL_GPL(all_vm_events);
86
87 #ifdef CONFIG_HOTPLUG
88 /*
89  * Fold the foreign cpu events into our own.
90  *
91  * This is adding to the events on one processor
92  * but keeps the global counts constant.
93  */
94 void vm_events_fold_cpu(int cpu)
95 {
96         struct vm_event_state *fold_state = &per_cpu(vm_event_states, cpu);
97         int i;
98
99         for (i = 0; i < NR_VM_EVENT_ITEMS; i++) {
100                 count_vm_events(i, fold_state->event[i]);
101                 fold_state->event[i] = 0;
102         }
103 }
104 #endif /* CONFIG_HOTPLUG */
105
106 #endif /* CONFIG_VM_EVENT_COUNTERS */
107
108 /*
109  * Manage combined zone based / global counters
110  *
111  * vm_stat contains the global counters
112  */
113 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
114 EXPORT_SYMBOL(vm_stat);
115
116 #ifdef CONFIG_SMP
117
118 static int calculate_threshold(struct zone *zone)
119 {
120         int threshold;
121         int mem;        /* memory in 128 MB units */
122
123         /*
124          * The threshold scales with the number of processors and the amount
125          * of memory per zone. More memory means that we can defer updates for
126          * longer, more processors could lead to more contention.
127          * fls() is used to have a cheap way of logarithmic scaling.
128          *
129          * Some sample thresholds:
130          *
131          * Threshold    Processors      (fls)   Zonesize        fls(mem+1)
132          * ------------------------------------------------------------------
133          * 8            1               1       0.9-1 GB        4
134          * 16           2               2       0.9-1 GB        4
135          * 20           2               2       1-2 GB          5
136          * 24           2               2       2-4 GB          6
137          * 28           2               2       4-8 GB          7
138          * 32           2               2       8-16 GB         8
139          * 4            2               2       <128M           1
140          * 30           4               3       2-4 GB          5
141          * 48           4               3       8-16 GB         8
142          * 32           8               4       1-2 GB          4
143          * 32           8               4       0.9-1GB         4
144          * 10           16              5       <128M           1
145          * 40           16              5       900M            4
146          * 70           64              7       2-4 GB          5
147          * 84           64              7       4-8 GB          6
148          * 108          512             9       4-8 GB          6
149          * 125          1024            10      8-16 GB         8
150          * 125          1024            10      16-32 GB        9
151          */
152
153         mem = zone->present_pages >> (27 - PAGE_SHIFT);
154
155         threshold = 2 * fls(num_online_cpus()) * (1 + fls(mem));
156
157         /*
158          * Maximum threshold is 125
159          */
160         threshold = min(125, threshold);
161
162         return threshold;
163 }
164
165 /*
166  * Refresh the thresholds for each zone.
167  */
168 static void refresh_zone_stat_thresholds(void)
169 {
170         struct zone *zone;
171         int cpu;
172         int threshold;
173
174         for_each_zone(zone) {
175
176                 if (!zone->present_pages)
177                         continue;
178
179                 threshold = calculate_threshold(zone);
180
181                 for_each_online_cpu(cpu)
182                         zone_pcp(zone, cpu)->stat_threshold = threshold;
183         }
184 }
185
186 /*
187  * For use when we know that interrupts are disabled.
188  */
189 void __mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
190                                 int delta)
191 {
192         struct per_cpu_pageset *pcp = zone_pcp(zone, smp_processor_id());
193         s8 *p = pcp->vm_stat_diff + item;
194         long x;
195
196         x = delta + *p;
197
198         if (unlikely(x > pcp->stat_threshold || x < -pcp->stat_threshold)) {
199                 zone_page_state_add(x, zone, item);
200                 x = 0;
201         }
202         *p = x;
203 }
204 EXPORT_SYMBOL(__mod_zone_page_state);
205
206 /*
207  * For an unknown interrupt state
208  */
209 void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
210                                         int delta)
211 {
212         unsigned long flags;
213
214         local_irq_save(flags);
215         __mod_zone_page_state(zone, item, delta);
216         local_irq_restore(flags);
217 }
218 EXPORT_SYMBOL(mod_zone_page_state);
219
220 /*
221  * Optimized increment and decrement functions.
222  *
223  * These are only for a single page and therefore can take a struct page *
224  * argument instead of struct zone *. This allows the inclusion of the code
225  * generated for page_zone(page) into the optimized functions.
226  *
227  * No overflow check is necessary and therefore the differential can be
228  * incremented or decremented in place which may allow the compilers to
229  * generate better code.
230  * The increment or decrement is known and therefore one boundary check can
231  * be omitted.
232  *
233  * NOTE: These functions are very performance sensitive. Change only
234  * with care.
235  *
236  * Some processors have inc/dec instructions that are atomic vs an interrupt.
237  * However, the code must first determine the differential location in a zone
238  * based on the processor number and then inc/dec the counter. There is no
239  * guarantee without disabling preemption that the processor will not change
240  * in between and therefore the atomicity vs. interrupt cannot be exploited
241  * in a useful way here.
242  */
243 static void __inc_zone_state(struct zone *zone, enum zone_stat_item item)
244 {
245         struct per_cpu_pageset *pcp = zone_pcp(zone, smp_processor_id());
246         s8 *p = pcp->vm_stat_diff + item;
247
248         (*p)++;
249
250         if (unlikely(*p > pcp->stat_threshold)) {
251                 int overstep = pcp->stat_threshold / 2;
252
253                 zone_page_state_add(*p + overstep, zone, item);
254                 *p = -overstep;
255         }
256 }
257
258 void __inc_zone_page_state(struct page *page, enum zone_stat_item item)
259 {
260         __inc_zone_state(page_zone(page), item);
261 }
262 EXPORT_SYMBOL(__inc_zone_page_state);
263
264 void __dec_zone_page_state(struct page *page, enum zone_stat_item item)
265 {
266         struct zone *zone = page_zone(page);
267         struct per_cpu_pageset *pcp = zone_pcp(zone, smp_processor_id());
268         s8 *p = pcp->vm_stat_diff + item;
269
270         (*p)--;
271
272         if (unlikely(*p < - pcp->stat_threshold)) {
273                 int overstep = pcp->stat_threshold / 2;
274
275                 zone_page_state_add(*p - overstep, zone, item);
276                 *p = overstep;
277         }
278 }
279 EXPORT_SYMBOL(__dec_zone_page_state);
280
281 void inc_zone_state(struct zone *zone, enum zone_stat_item item)
282 {
283         unsigned long flags;
284
285         local_irq_save(flags);
286         __inc_zone_state(zone, item);
287         local_irq_restore(flags);
288 }
289
290 void inc_zone_page_state(struct page *page, enum zone_stat_item item)
291 {
292         unsigned long flags;
293         struct zone *zone;
294
295         zone = page_zone(page);
296         local_irq_save(flags);
297         __inc_zone_state(zone, item);
298         local_irq_restore(flags);
299 }
300 EXPORT_SYMBOL(inc_zone_page_state);
301
302 void dec_zone_page_state(struct page *page, enum zone_stat_item item)
303 {
304         unsigned long flags;
305
306         local_irq_save(flags);
307         __dec_zone_page_state(page, item);
308         local_irq_restore(flags);
309 }
310 EXPORT_SYMBOL(dec_zone_page_state);
311
312 /*
313  * Update the zone counters for one cpu.
314  */
315 void refresh_cpu_vm_stats(int cpu)
316 {
317         struct zone *zone;
318         int i;
319         unsigned long flags;
320
321         for_each_zone(zone) {
322                 struct per_cpu_pageset *pcp;
323
324                 if (!populated_zone(zone))
325                         continue;
326
327                 pcp = zone_pcp(zone, cpu);
328
329                 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
330                         if (pcp->vm_stat_diff[i]) {
331                                 local_irq_save(flags);
332                                 zone_page_state_add(pcp->vm_stat_diff[i],
333                                         zone, i);
334                                 pcp->vm_stat_diff[i] = 0;
335                                 local_irq_restore(flags);
336                         }
337         }
338 }
339
340 static void __refresh_cpu_vm_stats(void *dummy)
341 {
342         refresh_cpu_vm_stats(smp_processor_id());
343 }
344
345 /*
346  * Consolidate all counters.
347  *
348  * Note that the result is less inaccurate but still inaccurate
349  * if concurrent processes are allowed to run.
350  */
351 void refresh_vm_stats(void)
352 {
353         on_each_cpu(__refresh_cpu_vm_stats, NULL, 0, 1);
354 }
355 EXPORT_SYMBOL(refresh_vm_stats);
356
357 #endif
358
359 #ifdef CONFIG_NUMA
360 /*
361  * zonelist = the list of zones passed to the allocator
362  * z        = the zone from which the allocation occurred.
363  *
364  * Must be called with interrupts disabled.
365  */
366 void zone_statistics(struct zonelist *zonelist, struct zone *z)
367 {
368         if (z->zone_pgdat == zonelist->zones[0]->zone_pgdat) {
369                 __inc_zone_state(z, NUMA_HIT);
370         } else {
371                 __inc_zone_state(z, NUMA_MISS);
372                 __inc_zone_state(zonelist->zones[0], NUMA_FOREIGN);
373         }
374         if (z->node == numa_node_id())
375                 __inc_zone_state(z, NUMA_LOCAL);
376         else
377                 __inc_zone_state(z, NUMA_OTHER);
378 }
379 #endif
380
381 #ifdef CONFIG_PROC_FS
382
383 #include <linux/seq_file.h>
384
385 static void *frag_start(struct seq_file *m, loff_t *pos)
386 {
387         pg_data_t *pgdat;
388         loff_t node = *pos;
389         for (pgdat = first_online_pgdat();
390              pgdat && node;
391              pgdat = next_online_pgdat(pgdat))
392                 --node;
393
394         return pgdat;
395 }
396
397 static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
398 {
399         pg_data_t *pgdat = (pg_data_t *)arg;
400
401         (*pos)++;
402         return next_online_pgdat(pgdat);
403 }
404
405 static void frag_stop(struct seq_file *m, void *arg)
406 {
407 }
408
409 /*
410  * This walks the free areas for each zone.
411  */
412 static int frag_show(struct seq_file *m, void *arg)
413 {
414         pg_data_t *pgdat = (pg_data_t *)arg;
415         struct zone *zone;
416         struct zone *node_zones = pgdat->node_zones;
417         unsigned long flags;
418         int order;
419
420         for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
421                 if (!populated_zone(zone))
422                         continue;
423
424                 spin_lock_irqsave(&zone->lock, flags);
425                 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
426                 for (order = 0; order < MAX_ORDER; ++order)
427                         seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
428                 spin_unlock_irqrestore(&zone->lock, flags);
429                 seq_putc(m, '\n');
430         }
431         return 0;
432 }
433
434 struct seq_operations fragmentation_op = {
435         .start  = frag_start,
436         .next   = frag_next,
437         .stop   = frag_stop,
438         .show   = frag_show,
439 };
440
441 #ifdef CONFIG_ZONE_DMA32
442 #define TEXT_FOR_DMA32(xx) xx "_dma32",
443 #else
444 #define TEXT_FOR_DMA32(xx)
445 #endif
446
447 #ifdef CONFIG_HIGHMEM
448 #define TEXT_FOR_HIGHMEM(xx) xx "_high",
449 #else
450 #define TEXT_FOR_HIGHMEM(xx)
451 #endif
452
453 #define TEXTS_FOR_ZONES(xx) xx "_dma", TEXT_FOR_DMA32(xx) xx "_normal", \
454                                         TEXT_FOR_HIGHMEM(xx)
455
456 static char *vmstat_text[] = {
457         /* Zoned VM counters */
458         "nr_anon_pages",
459         "nr_mapped",
460         "nr_file_pages",
461         "nr_slab_reclaimable",
462         "nr_slab_unreclaimable",
463         "nr_page_table_pages",
464         "nr_dirty",
465         "nr_writeback",
466         "nr_unstable",
467         "nr_bounce",
468         "nr_vmscan_write",
469
470 #ifdef CONFIG_NUMA
471         "numa_hit",
472         "numa_miss",
473         "numa_foreign",
474         "numa_interleave",
475         "numa_local",
476         "numa_other",
477 #endif
478
479 #ifdef CONFIG_VM_EVENT_COUNTERS
480         "pgpgin",
481         "pgpgout",
482         "pswpin",
483         "pswpout",
484
485         TEXTS_FOR_ZONES("pgalloc")
486
487         "pgfree",
488         "pgactivate",
489         "pgdeactivate",
490
491         "pgfault",
492         "pgmajfault",
493
494         TEXTS_FOR_ZONES("pgrefill")
495         TEXTS_FOR_ZONES("pgsteal")
496         TEXTS_FOR_ZONES("pgscan_kswapd")
497         TEXTS_FOR_ZONES("pgscan_direct")
498
499         "pginodesteal",
500         "slabs_scanned",
501         "kswapd_steal",
502         "kswapd_inodesteal",
503         "pageoutrun",
504         "allocstall",
505
506         "pgrotated",
507 #endif
508 };
509
510 /*
511  * Output information about zones in @pgdat.
512  */
513 static int zoneinfo_show(struct seq_file *m, void *arg)
514 {
515         pg_data_t *pgdat = arg;
516         struct zone *zone;
517         struct zone *node_zones = pgdat->node_zones;
518         unsigned long flags;
519
520         for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; zone++) {
521                 int i;
522
523                 if (!populated_zone(zone))
524                         continue;
525
526                 spin_lock_irqsave(&zone->lock, flags);
527                 seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
528                 seq_printf(m,
529                            "\n  pages free     %lu"
530                            "\n        min      %lu"
531                            "\n        low      %lu"
532                            "\n        high     %lu"
533                            "\n        active   %lu"
534                            "\n        inactive %lu"
535                            "\n        scanned  %lu (a: %lu i: %lu)"
536                            "\n        spanned  %lu"
537                            "\n        present  %lu",
538                            zone->free_pages,
539                            zone->pages_min,
540                            zone->pages_low,
541                            zone->pages_high,
542                            zone->nr_active,
543                            zone->nr_inactive,
544                            zone->pages_scanned,
545                            zone->nr_scan_active, zone->nr_scan_inactive,
546                            zone->spanned_pages,
547                            zone->present_pages);
548
549                 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
550                         seq_printf(m, "\n    %-12s %lu", vmstat_text[i],
551                                         zone_page_state(zone, i));
552
553                 seq_printf(m,
554                            "\n        protection: (%lu",
555                            zone->lowmem_reserve[0]);
556                 for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
557                         seq_printf(m, ", %lu", zone->lowmem_reserve[i]);
558                 seq_printf(m,
559                            ")"
560                            "\n  pagesets");
561                 for_each_online_cpu(i) {
562                         struct per_cpu_pageset *pageset;
563                         int j;
564
565                         pageset = zone_pcp(zone, i);
566                         for (j = 0; j < ARRAY_SIZE(pageset->pcp); j++) {
567                                 if (pageset->pcp[j].count)
568                                         break;
569                         }
570                         if (j == ARRAY_SIZE(pageset->pcp))
571                                 continue;
572                         for (j = 0; j < ARRAY_SIZE(pageset->pcp); j++) {
573                                 seq_printf(m,
574                                            "\n    cpu: %i pcp: %i"
575                                            "\n              count: %i"
576                                            "\n              high:  %i"
577                                            "\n              batch: %i",
578                                            i, j,
579                                            pageset->pcp[j].count,
580                                            pageset->pcp[j].high,
581                                            pageset->pcp[j].batch);
582                         }
583 #ifdef CONFIG_SMP
584                         seq_printf(m, "\n  vm stats threshold: %d",
585                                         pageset->stat_threshold);
586 #endif
587                 }
588                 seq_printf(m,
589                            "\n  all_unreclaimable: %u"
590                            "\n  prev_priority:     %i"
591                            "\n  temp_priority:     %i"
592                            "\n  start_pfn:         %lu",
593                            zone->all_unreclaimable,
594                            zone->prev_priority,
595                            zone->temp_priority,
596                            zone->zone_start_pfn);
597                 spin_unlock_irqrestore(&zone->lock, flags);
598                 seq_putc(m, '\n');
599         }
600         return 0;
601 }
602
603 struct seq_operations zoneinfo_op = {
604         .start  = frag_start, /* iterate over all zones. The same as in
605                                * fragmentation. */
606         .next   = frag_next,
607         .stop   = frag_stop,
608         .show   = zoneinfo_show,
609 };
610
611 static void *vmstat_start(struct seq_file *m, loff_t *pos)
612 {
613         unsigned long *v;
614 #ifdef CONFIG_VM_EVENT_COUNTERS
615         unsigned long *e;
616 #endif
617         int i;
618
619         if (*pos >= ARRAY_SIZE(vmstat_text))
620                 return NULL;
621
622 #ifdef CONFIG_VM_EVENT_COUNTERS
623         v = kmalloc(NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long)
624                         + sizeof(struct vm_event_state), GFP_KERNEL);
625 #else
626         v = kmalloc(NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long),
627                         GFP_KERNEL);
628 #endif
629         m->private = v;
630         if (!v)
631                 return ERR_PTR(-ENOMEM);
632         for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
633                 v[i] = global_page_state(i);
634 #ifdef CONFIG_VM_EVENT_COUNTERS
635         e = v + NR_VM_ZONE_STAT_ITEMS;
636         all_vm_events(e);
637         e[PGPGIN] /= 2;         /* sectors -> kbytes */
638         e[PGPGOUT] /= 2;
639 #endif
640         return v + *pos;
641 }
642
643 static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
644 {
645         (*pos)++;
646         if (*pos >= ARRAY_SIZE(vmstat_text))
647                 return NULL;
648         return (unsigned long *)m->private + *pos;
649 }
650
651 static int vmstat_show(struct seq_file *m, void *arg)
652 {
653         unsigned long *l = arg;
654         unsigned long off = l - (unsigned long *)m->private;
655
656         seq_printf(m, "%s %lu\n", vmstat_text[off], *l);
657         return 0;
658 }
659
660 static void vmstat_stop(struct seq_file *m, void *arg)
661 {
662         kfree(m->private);
663         m->private = NULL;
664 }
665
666 struct seq_operations vmstat_op = {
667         .start  = vmstat_start,
668         .next   = vmstat_next,
669         .stop   = vmstat_stop,
670         .show   = vmstat_show,
671 };
672
673 #endif /* CONFIG_PROC_FS */
674
675 #ifdef CONFIG_SMP
676 /*
677  * Use the cpu notifier to insure that the thresholds are recalculated
678  * when necessary.
679  */
680 static int __cpuinit vmstat_cpuup_callback(struct notifier_block *nfb,
681                 unsigned long action,
682                 void *hcpu)
683 {
684         switch (action) {
685                 case CPU_UP_PREPARE:
686                 case CPU_UP_CANCELED:
687                 case CPU_DEAD:
688                         refresh_zone_stat_thresholds();
689                         break;
690                 default:
691                         break;
692         }
693         return NOTIFY_OK;
694 }
695
696 static struct notifier_block __cpuinitdata vmstat_notifier =
697         { &vmstat_cpuup_callback, NULL, 0 };
698
699 int __init setup_vmstat(void)
700 {
701         refresh_zone_stat_thresholds();
702         register_cpu_notifier(&vmstat_notifier);
703         return 0;
704 }
705 module_init(setup_vmstat)
706 #endif