[PATCH] zoned vm counters: conversion of nr_slab to per zone counter
[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/config.h>
13 #include <linux/mm.h>
14 #include <linux/module.h>
15
16 /*
17  * Accumulate the page_state information across all CPUs.
18  * The result is unavoidably approximate - it can change
19  * during and after execution of this function.
20  */
21 DEFINE_PER_CPU(struct page_state, page_states) = {0};
22
23 static void __get_page_state(struct page_state *ret, int nr, cpumask_t *cpumask)
24 {
25         unsigned cpu;
26
27         memset(ret, 0, nr * sizeof(unsigned long));
28         cpus_and(*cpumask, *cpumask, cpu_online_map);
29
30         for_each_cpu_mask(cpu, *cpumask) {
31                 unsigned long *in;
32                 unsigned long *out;
33                 unsigned off;
34                 unsigned next_cpu;
35
36                 in = (unsigned long *)&per_cpu(page_states, cpu);
37
38                 next_cpu = next_cpu(cpu, *cpumask);
39                 if (likely(next_cpu < NR_CPUS))
40                         prefetch(&per_cpu(page_states, next_cpu));
41
42                 out = (unsigned long *)ret;
43                 for (off = 0; off < nr; off++)
44                         *out++ += *in++;
45         }
46 }
47
48 void get_page_state_node(struct page_state *ret, int node)
49 {
50         int nr;
51         cpumask_t mask = node_to_cpumask(node);
52
53         nr = offsetof(struct page_state, GET_PAGE_STATE_LAST);
54         nr /= sizeof(unsigned long);
55
56         __get_page_state(ret, nr+1, &mask);
57 }
58
59 void get_page_state(struct page_state *ret)
60 {
61         int nr;
62         cpumask_t mask = CPU_MASK_ALL;
63
64         nr = offsetof(struct page_state, GET_PAGE_STATE_LAST);
65         nr /= sizeof(unsigned long);
66
67         __get_page_state(ret, nr + 1, &mask);
68 }
69
70 void get_full_page_state(struct page_state *ret)
71 {
72         cpumask_t mask = CPU_MASK_ALL;
73
74         __get_page_state(ret, sizeof(*ret) / sizeof(unsigned long), &mask);
75 }
76
77 unsigned long read_page_state_offset(unsigned long offset)
78 {
79         unsigned long ret = 0;
80         int cpu;
81
82         for_each_online_cpu(cpu) {
83                 unsigned long in;
84
85                 in = (unsigned long)&per_cpu(page_states, cpu) + offset;
86                 ret += *((unsigned long *)in);
87         }
88         return ret;
89 }
90
91 void __mod_page_state_offset(unsigned long offset, unsigned long delta)
92 {
93         void *ptr;
94
95         ptr = &__get_cpu_var(page_states);
96         *(unsigned long *)(ptr + offset) += delta;
97 }
98 EXPORT_SYMBOL(__mod_page_state_offset);
99
100 void mod_page_state_offset(unsigned long offset, unsigned long delta)
101 {
102         unsigned long flags;
103         void *ptr;
104
105         local_irq_save(flags);
106         ptr = &__get_cpu_var(page_states);
107         *(unsigned long *)(ptr + offset) += delta;
108         local_irq_restore(flags);
109 }
110 EXPORT_SYMBOL(mod_page_state_offset);
111
112 void __get_zone_counts(unsigned long *active, unsigned long *inactive,
113                         unsigned long *free, struct pglist_data *pgdat)
114 {
115         struct zone *zones = pgdat->node_zones;
116         int i;
117
118         *active = 0;
119         *inactive = 0;
120         *free = 0;
121         for (i = 0; i < MAX_NR_ZONES; i++) {
122                 *active += zones[i].nr_active;
123                 *inactive += zones[i].nr_inactive;
124                 *free += zones[i].free_pages;
125         }
126 }
127
128 void get_zone_counts(unsigned long *active,
129                 unsigned long *inactive, unsigned long *free)
130 {
131         struct pglist_data *pgdat;
132
133         *active = 0;
134         *inactive = 0;
135         *free = 0;
136         for_each_online_pgdat(pgdat) {
137                 unsigned long l, m, n;
138                 __get_zone_counts(&l, &m, &n, pgdat);
139                 *active += l;
140                 *inactive += m;
141                 *free += n;
142         }
143 }
144
145 /*
146  * Manage combined zone based / global counters
147  *
148  * vm_stat contains the global counters
149  */
150 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
151 EXPORT_SYMBOL(vm_stat);
152
153 #ifdef CONFIG_SMP
154
155 #define STAT_THRESHOLD 32
156
157 /*
158  * Determine pointer to currently valid differential byte given a zone and
159  * the item number.
160  *
161  * Preemption must be off
162  */
163 static inline s8 *diff_pointer(struct zone *zone, enum zone_stat_item item)
164 {
165         return &zone_pcp(zone, smp_processor_id())->vm_stat_diff[item];
166 }
167
168 /*
169  * For use when we know that interrupts are disabled.
170  */
171 void __mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
172                                 int delta)
173 {
174         s8 *p;
175         long x;
176
177         p = diff_pointer(zone, item);
178         x = delta + *p;
179
180         if (unlikely(x > STAT_THRESHOLD || x < -STAT_THRESHOLD)) {
181                 zone_page_state_add(x, zone, item);
182                 x = 0;
183         }
184
185         *p = x;
186 }
187 EXPORT_SYMBOL(__mod_zone_page_state);
188
189 /*
190  * For an unknown interrupt state
191  */
192 void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
193                                         int delta)
194 {
195         unsigned long flags;
196
197         local_irq_save(flags);
198         __mod_zone_page_state(zone, item, delta);
199         local_irq_restore(flags);
200 }
201 EXPORT_SYMBOL(mod_zone_page_state);
202
203 /*
204  * Optimized increment and decrement functions.
205  *
206  * These are only for a single page and therefore can take a struct page *
207  * argument instead of struct zone *. This allows the inclusion of the code
208  * generated for page_zone(page) into the optimized functions.
209  *
210  * No overflow check is necessary and therefore the differential can be
211  * incremented or decremented in place which may allow the compilers to
212  * generate better code.
213  *
214  * The increment or decrement is known and therefore one boundary check can
215  * be omitted.
216  *
217  * Some processors have inc/dec instructions that are atomic vs an interrupt.
218  * However, the code must first determine the differential location in a zone
219  * based on the processor number and then inc/dec the counter. There is no
220  * guarantee without disabling preemption that the processor will not change
221  * in between and therefore the atomicity vs. interrupt cannot be exploited
222  * in a useful way here.
223  */
224 void __inc_zone_page_state(struct page *page, enum zone_stat_item item)
225 {
226         struct zone *zone = page_zone(page);
227         s8 *p = diff_pointer(zone, item);
228
229         (*p)++;
230
231         if (unlikely(*p > STAT_THRESHOLD)) {
232                 zone_page_state_add(*p, zone, item);
233                 *p = 0;
234         }
235 }
236 EXPORT_SYMBOL(__inc_zone_page_state);
237
238 void __dec_zone_page_state(struct page *page, enum zone_stat_item item)
239 {
240         struct zone *zone = page_zone(page);
241         s8 *p = diff_pointer(zone, item);
242
243         (*p)--;
244
245         if (unlikely(*p < -STAT_THRESHOLD)) {
246                 zone_page_state_add(*p, zone, item);
247                 *p = 0;
248         }
249 }
250 EXPORT_SYMBOL(__dec_zone_page_state);
251
252 void inc_zone_page_state(struct page *page, enum zone_stat_item item)
253 {
254         unsigned long flags;
255         struct zone *zone;
256         s8 *p;
257
258         zone = page_zone(page);
259         local_irq_save(flags);
260         p = diff_pointer(zone, item);
261
262         (*p)++;
263
264         if (unlikely(*p > STAT_THRESHOLD)) {
265                 zone_page_state_add(*p, zone, item);
266                 *p = 0;
267         }
268         local_irq_restore(flags);
269 }
270 EXPORT_SYMBOL(inc_zone_page_state);
271
272 void dec_zone_page_state(struct page *page, enum zone_stat_item item)
273 {
274         unsigned long flags;
275         struct zone *zone;
276         s8 *p;
277
278         zone = page_zone(page);
279         local_irq_save(flags);
280         p = diff_pointer(zone, item);
281
282         (*p)--;
283
284         if (unlikely(*p < -STAT_THRESHOLD)) {
285                 zone_page_state_add(*p, zone, item);
286                 *p = 0;
287         }
288         local_irq_restore(flags);
289 }
290 EXPORT_SYMBOL(dec_zone_page_state);
291
292 /*
293  * Update the zone counters for one cpu.
294  */
295 void refresh_cpu_vm_stats(int cpu)
296 {
297         struct zone *zone;
298         int i;
299         unsigned long flags;
300
301         for_each_zone(zone) {
302                 struct per_cpu_pageset *pcp;
303
304                 pcp = zone_pcp(zone, cpu);
305
306                 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
307                         if (pcp->vm_stat_diff[i]) {
308                                 local_irq_save(flags);
309                                 zone_page_state_add(pcp->vm_stat_diff[i],
310                                         zone, i);
311                                 pcp->vm_stat_diff[i] = 0;
312                                 local_irq_restore(flags);
313                         }
314         }
315 }
316
317 static void __refresh_cpu_vm_stats(void *dummy)
318 {
319         refresh_cpu_vm_stats(smp_processor_id());
320 }
321
322 /*
323  * Consolidate all counters.
324  *
325  * Note that the result is less inaccurate but still inaccurate
326  * if concurrent processes are allowed to run.
327  */
328 void refresh_vm_stats(void)
329 {
330         on_each_cpu(__refresh_cpu_vm_stats, NULL, 0, 1);
331 }
332 EXPORT_SYMBOL(refresh_vm_stats);
333
334 #endif
335
336 #ifdef CONFIG_PROC_FS
337
338 #include <linux/seq_file.h>
339
340 static void *frag_start(struct seq_file *m, loff_t *pos)
341 {
342         pg_data_t *pgdat;
343         loff_t node = *pos;
344         for (pgdat = first_online_pgdat();
345              pgdat && node;
346              pgdat = next_online_pgdat(pgdat))
347                 --node;
348
349         return pgdat;
350 }
351
352 static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
353 {
354         pg_data_t *pgdat = (pg_data_t *)arg;
355
356         (*pos)++;
357         return next_online_pgdat(pgdat);
358 }
359
360 static void frag_stop(struct seq_file *m, void *arg)
361 {
362 }
363
364 /*
365  * This walks the free areas for each zone.
366  */
367 static int frag_show(struct seq_file *m, void *arg)
368 {
369         pg_data_t *pgdat = (pg_data_t *)arg;
370         struct zone *zone;
371         struct zone *node_zones = pgdat->node_zones;
372         unsigned long flags;
373         int order;
374
375         for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
376                 if (!populated_zone(zone))
377                         continue;
378
379                 spin_lock_irqsave(&zone->lock, flags);
380                 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
381                 for (order = 0; order < MAX_ORDER; ++order)
382                         seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
383                 spin_unlock_irqrestore(&zone->lock, flags);
384                 seq_putc(m, '\n');
385         }
386         return 0;
387 }
388
389 struct seq_operations fragmentation_op = {
390         .start  = frag_start,
391         .next   = frag_next,
392         .stop   = frag_stop,
393         .show   = frag_show,
394 };
395
396 static char *vmstat_text[] = {
397         /* Zoned VM counters */
398         "nr_anon_pages",
399         "nr_mapped",
400         "nr_file_pages",
401         "nr_slab",
402
403         /* Page state */
404         "nr_dirty",
405         "nr_writeback",
406         "nr_unstable",
407         "nr_page_table_pages",
408
409         "pgpgin",
410         "pgpgout",
411         "pswpin",
412         "pswpout",
413
414         "pgalloc_high",
415         "pgalloc_normal",
416         "pgalloc_dma32",
417         "pgalloc_dma",
418
419         "pgfree",
420         "pgactivate",
421         "pgdeactivate",
422
423         "pgfault",
424         "pgmajfault",
425
426         "pgrefill_high",
427         "pgrefill_normal",
428         "pgrefill_dma32",
429         "pgrefill_dma",
430
431         "pgsteal_high",
432         "pgsteal_normal",
433         "pgsteal_dma32",
434         "pgsteal_dma",
435
436         "pgscan_kswapd_high",
437         "pgscan_kswapd_normal",
438         "pgscan_kswapd_dma32",
439         "pgscan_kswapd_dma",
440
441         "pgscan_direct_high",
442         "pgscan_direct_normal",
443         "pgscan_direct_dma32",
444         "pgscan_direct_dma",
445
446         "pginodesteal",
447         "slabs_scanned",
448         "kswapd_steal",
449         "kswapd_inodesteal",
450         "pageoutrun",
451         "allocstall",
452
453         "pgrotated",
454         "nr_bounce",
455 };
456
457 /*
458  * Output information about zones in @pgdat.
459  */
460 static int zoneinfo_show(struct seq_file *m, void *arg)
461 {
462         pg_data_t *pgdat = arg;
463         struct zone *zone;
464         struct zone *node_zones = pgdat->node_zones;
465         unsigned long flags;
466
467         for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; zone++) {
468                 int i;
469
470                 if (!populated_zone(zone))
471                         continue;
472
473                 spin_lock_irqsave(&zone->lock, flags);
474                 seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
475                 seq_printf(m,
476                            "\n  pages free     %lu"
477                            "\n        min      %lu"
478                            "\n        low      %lu"
479                            "\n        high     %lu"
480                            "\n        active   %lu"
481                            "\n        inactive %lu"
482                            "\n        scanned  %lu (a: %lu i: %lu)"
483                            "\n        spanned  %lu"
484                            "\n        present  %lu",
485                            zone->free_pages,
486                            zone->pages_min,
487                            zone->pages_low,
488                            zone->pages_high,
489                            zone->nr_active,
490                            zone->nr_inactive,
491                            zone->pages_scanned,
492                            zone->nr_scan_active, zone->nr_scan_inactive,
493                            zone->spanned_pages,
494                            zone->present_pages);
495
496                 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
497                         seq_printf(m, "\n    %-12s %lu", vmstat_text[i],
498                                         zone_page_state(zone, i));
499
500                 seq_printf(m,
501                            "\n        protection: (%lu",
502                            zone->lowmem_reserve[0]);
503                 for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
504                         seq_printf(m, ", %lu", zone->lowmem_reserve[i]);
505                 seq_printf(m,
506                            ")"
507                            "\n  pagesets");
508                 for_each_online_cpu(i) {
509                         struct per_cpu_pageset *pageset;
510                         int j;
511
512                         pageset = zone_pcp(zone, i);
513                         for (j = 0; j < ARRAY_SIZE(pageset->pcp); j++) {
514                                 if (pageset->pcp[j].count)
515                                         break;
516                         }
517                         if (j == ARRAY_SIZE(pageset->pcp))
518                                 continue;
519                         for (j = 0; j < ARRAY_SIZE(pageset->pcp); j++) {
520                                 seq_printf(m,
521                                            "\n    cpu: %i pcp: %i"
522                                            "\n              count: %i"
523                                            "\n              high:  %i"
524                                            "\n              batch: %i",
525                                            i, j,
526                                            pageset->pcp[j].count,
527                                            pageset->pcp[j].high,
528                                            pageset->pcp[j].batch);
529                         }
530 #ifdef CONFIG_NUMA
531                         seq_printf(m,
532                                    "\n            numa_hit:       %lu"
533                                    "\n            numa_miss:      %lu"
534                                    "\n            numa_foreign:   %lu"
535                                    "\n            interleave_hit: %lu"
536                                    "\n            local_node:     %lu"
537                                    "\n            other_node:     %lu",
538                                    pageset->numa_hit,
539                                    pageset->numa_miss,
540                                    pageset->numa_foreign,
541                                    pageset->interleave_hit,
542                                    pageset->local_node,
543                                    pageset->other_node);
544 #endif
545                 }
546                 seq_printf(m,
547                            "\n  all_unreclaimable: %u"
548                            "\n  prev_priority:     %i"
549                            "\n  temp_priority:     %i"
550                            "\n  start_pfn:         %lu",
551                            zone->all_unreclaimable,
552                            zone->prev_priority,
553                            zone->temp_priority,
554                            zone->zone_start_pfn);
555                 spin_unlock_irqrestore(&zone->lock, flags);
556                 seq_putc(m, '\n');
557         }
558         return 0;
559 }
560
561 struct seq_operations zoneinfo_op = {
562         .start  = frag_start, /* iterate over all zones. The same as in
563                                * fragmentation. */
564         .next   = frag_next,
565         .stop   = frag_stop,
566         .show   = zoneinfo_show,
567 };
568
569 static void *vmstat_start(struct seq_file *m, loff_t *pos)
570 {
571         unsigned long *v;
572         struct page_state *ps;
573         int i;
574
575         if (*pos >= ARRAY_SIZE(vmstat_text))
576                 return NULL;
577
578         v = kmalloc(NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long)
579                         + sizeof(*ps), GFP_KERNEL);
580         m->private = v;
581         if (!v)
582                 return ERR_PTR(-ENOMEM);
583         for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
584                 v[i] = global_page_state(i);
585         ps = (struct page_state *)(v + NR_VM_ZONE_STAT_ITEMS);
586         get_full_page_state(ps);
587         ps->pgpgin /= 2;                /* sectors -> kbytes */
588         ps->pgpgout /= 2;
589         return v + *pos;
590 }
591
592 static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
593 {
594         (*pos)++;
595         if (*pos >= ARRAY_SIZE(vmstat_text))
596                 return NULL;
597         return (unsigned long *)m->private + *pos;
598 }
599
600 static int vmstat_show(struct seq_file *m, void *arg)
601 {
602         unsigned long *l = arg;
603         unsigned long off = l - (unsigned long *)m->private;
604
605         seq_printf(m, "%s %lu\n", vmstat_text[off], *l);
606         return 0;
607 }
608
609 static void vmstat_stop(struct seq_file *m, void *arg)
610 {
611         kfree(m->private);
612         m->private = NULL;
613 }
614
615 struct seq_operations vmstat_op = {
616         .start  = vmstat_start,
617         .next   = vmstat_next,
618         .stop   = vmstat_stop,
619         .show   = vmstat_show,
620 };
621
622 #endif /* CONFIG_PROC_FS */
623