blob: 7f760cbc73f35a7b49944fe86f046e5269850b31 [file] [log] [blame]
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001/*
2 * linux/mm/vmstat.c
3 *
4 * Manages VM statistics
5 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
Christoph Lameter2244b952006-06-30 01:55:33 -07006 *
7 * zoned VM statistics
8 * Copyright (C) 2006 Silicon Graphics, Inc.,
9 * Christoph Lameter <christoph@lameter.com>
Christoph Lameterf6ac2352006-06-30 01:55:32 -070010 */
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +040011#include <linux/fs.h>
Christoph Lameterf6ac2352006-06-30 01:55:32 -070012#include <linux/mm.h>
Alexey Dobriyan4e950f62007-07-30 02:36:13 +040013#include <linux/err.h>
Christoph Lameter2244b952006-06-30 01:55:33 -070014#include <linux/module.h>
Christoph Lameterdf9ecab2006-08-31 21:27:35 -070015#include <linux/cpu.h>
Adrian Bunkc748e132008-07-23 21:27:03 -070016#include <linux/vmstat.h>
Alexey Dobriyane8edc6e2007-05-21 01:22:52 +040017#include <linux/sched.h>
Christoph Lameterf6ac2352006-06-30 01:55:32 -070018
Christoph Lameterf8891e52006-06-30 01:55:45 -070019#ifdef CONFIG_VM_EVENT_COUNTERS
20DEFINE_PER_CPU(struct vm_event_state, vm_event_states) = {{0}};
21EXPORT_PER_CPU_SYMBOL(vm_event_states);
22
Rusty Russell174596a2009-01-01 10:12:29 +103023static void sum_vm_events(unsigned long *ret, const struct cpumask *cpumask)
Christoph Lameterf8891e52006-06-30 01:55:45 -070024{
Christoph Lameter9eccf2a2008-02-04 22:29:22 -080025 int cpu;
Christoph Lameterf8891e52006-06-30 01:55:45 -070026 int i;
27
28 memset(ret, 0, NR_VM_EVENT_ITEMS * sizeof(unsigned long));
29
Rusty Russellaa85ea52009-03-30 22:05:15 -060030 for_each_cpu(cpu, cpumask) {
Christoph Lameterf8891e52006-06-30 01:55:45 -070031 struct vm_event_state *this = &per_cpu(vm_event_states, cpu);
32
Christoph Lameterf8891e52006-06-30 01:55:45 -070033 for (i = 0; i < NR_VM_EVENT_ITEMS; i++)
34 ret[i] += this->event[i];
35 }
36}
37
38/*
39 * Accumulate the vm event counters across all CPUs.
40 * The result is unavoidably approximate - it can change
41 * during and after execution of this function.
42*/
43void all_vm_events(unsigned long *ret)
44{
KOSAKI Motohirob5be1132008-05-12 14:02:06 -070045 get_online_cpus();
Rusty Russell174596a2009-01-01 10:12:29 +103046 sum_vm_events(ret, cpu_online_mask);
KOSAKI Motohirob5be1132008-05-12 14:02:06 -070047 put_online_cpus();
Christoph Lameterf8891e52006-06-30 01:55:45 -070048}
Heiko Carstens32dd66f2006-07-10 04:44:31 -070049EXPORT_SYMBOL_GPL(all_vm_events);
Christoph Lameterf8891e52006-06-30 01:55:45 -070050
51#ifdef CONFIG_HOTPLUG
52/*
53 * Fold the foreign cpu events into our own.
54 *
55 * This is adding to the events on one processor
56 * but keeps the global counts constant.
57 */
58void vm_events_fold_cpu(int cpu)
59{
60 struct vm_event_state *fold_state = &per_cpu(vm_event_states, cpu);
61 int i;
62
63 for (i = 0; i < NR_VM_EVENT_ITEMS; i++) {
64 count_vm_events(i, fold_state->event[i]);
65 fold_state->event[i] = 0;
66 }
67}
68#endif /* CONFIG_HOTPLUG */
69
70#endif /* CONFIG_VM_EVENT_COUNTERS */
71
Christoph Lameter2244b952006-06-30 01:55:33 -070072/*
73 * Manage combined zone based / global counters
74 *
75 * vm_stat contains the global counters
76 */
77atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
78EXPORT_SYMBOL(vm_stat);
79
80#ifdef CONFIG_SMP
81
Christoph Lameterdf9ecab2006-08-31 21:27:35 -070082static int calculate_threshold(struct zone *zone)
83{
84 int threshold;
85 int mem; /* memory in 128 MB units */
86
87 /*
88 * The threshold scales with the number of processors and the amount
89 * of memory per zone. More memory means that we can defer updates for
90 * longer, more processors could lead to more contention.
91 * fls() is used to have a cheap way of logarithmic scaling.
92 *
93 * Some sample thresholds:
94 *
95 * Threshold Processors (fls) Zonesize fls(mem+1)
96 * ------------------------------------------------------------------
97 * 8 1 1 0.9-1 GB 4
98 * 16 2 2 0.9-1 GB 4
99 * 20 2 2 1-2 GB 5
100 * 24 2 2 2-4 GB 6
101 * 28 2 2 4-8 GB 7
102 * 32 2 2 8-16 GB 8
103 * 4 2 2 <128M 1
104 * 30 4 3 2-4 GB 5
105 * 48 4 3 8-16 GB 8
106 * 32 8 4 1-2 GB 4
107 * 32 8 4 0.9-1GB 4
108 * 10 16 5 <128M 1
109 * 40 16 5 900M 4
110 * 70 64 7 2-4 GB 5
111 * 84 64 7 4-8 GB 6
112 * 108 512 9 4-8 GB 6
113 * 125 1024 10 8-16 GB 8
114 * 125 1024 10 16-32 GB 9
115 */
116
117 mem = zone->present_pages >> (27 - PAGE_SHIFT);
118
119 threshold = 2 * fls(num_online_cpus()) * (1 + fls(mem));
120
121 /*
122 * Maximum threshold is 125
123 */
124 threshold = min(125, threshold);
125
126 return threshold;
127}
Christoph Lameter2244b952006-06-30 01:55:33 -0700128
129/*
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700130 * Refresh the thresholds for each zone.
Christoph Lameter2244b952006-06-30 01:55:33 -0700131 */
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700132static void refresh_zone_stat_thresholds(void)
Christoph Lameter2244b952006-06-30 01:55:33 -0700133{
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700134 struct zone *zone;
135 int cpu;
136 int threshold;
137
KOSAKI Motohiroee99c712009-03-31 15:19:31 -0700138 for_each_populated_zone(zone) {
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700139 threshold = calculate_threshold(zone);
140
141 for_each_online_cpu(cpu)
Christoph Lameter99dcc3e2010-01-05 15:34:51 +0900142 per_cpu_ptr(zone->pageset, cpu)->stat_threshold
143 = threshold;
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700144 }
Christoph Lameter2244b952006-06-30 01:55:33 -0700145}
146
147/*
148 * For use when we know that interrupts are disabled.
149 */
150void __mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
151 int delta)
152{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +0900153 struct per_cpu_pageset *pcp = this_cpu_ptr(zone->pageset);
154
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700155 s8 *p = pcp->vm_stat_diff + item;
Christoph Lameter2244b952006-06-30 01:55:33 -0700156 long x;
157
Christoph Lameter2244b952006-06-30 01:55:33 -0700158 x = delta + *p;
159
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700160 if (unlikely(x > pcp->stat_threshold || x < -pcp->stat_threshold)) {
Christoph Lameter2244b952006-06-30 01:55:33 -0700161 zone_page_state_add(x, zone, item);
162 x = 0;
163 }
Christoph Lameter2244b952006-06-30 01:55:33 -0700164 *p = x;
165}
166EXPORT_SYMBOL(__mod_zone_page_state);
167
168/*
169 * For an unknown interrupt state
170 */
171void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
172 int delta)
173{
174 unsigned long flags;
175
176 local_irq_save(flags);
177 __mod_zone_page_state(zone, item, delta);
178 local_irq_restore(flags);
179}
180EXPORT_SYMBOL(mod_zone_page_state);
181
182/*
183 * Optimized increment and decrement functions.
184 *
185 * These are only for a single page and therefore can take a struct page *
186 * argument instead of struct zone *. This allows the inclusion of the code
187 * generated for page_zone(page) into the optimized functions.
188 *
189 * No overflow check is necessary and therefore the differential can be
190 * incremented or decremented in place which may allow the compilers to
191 * generate better code.
Christoph Lameter2244b952006-06-30 01:55:33 -0700192 * The increment or decrement is known and therefore one boundary check can
193 * be omitted.
194 *
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700195 * NOTE: These functions are very performance sensitive. Change only
196 * with care.
197 *
Christoph Lameter2244b952006-06-30 01:55:33 -0700198 * Some processors have inc/dec instructions that are atomic vs an interrupt.
199 * However, the code must first determine the differential location in a zone
200 * based on the processor number and then inc/dec the counter. There is no
201 * guarantee without disabling preemption that the processor will not change
202 * in between and therefore the atomicity vs. interrupt cannot be exploited
203 * in a useful way here.
204 */
Christoph Lameterc8785382007-02-10 01:43:01 -0800205void __inc_zone_state(struct zone *zone, enum zone_stat_item item)
Christoph Lameter2244b952006-06-30 01:55:33 -0700206{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +0900207 struct per_cpu_pageset *pcp = this_cpu_ptr(zone->pageset);
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700208 s8 *p = pcp->vm_stat_diff + item;
Christoph Lameter2244b952006-06-30 01:55:33 -0700209
210 (*p)++;
211
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700212 if (unlikely(*p > pcp->stat_threshold)) {
213 int overstep = pcp->stat_threshold / 2;
214
215 zone_page_state_add(*p + overstep, zone, item);
216 *p = -overstep;
Christoph Lameter2244b952006-06-30 01:55:33 -0700217 }
218}
Christoph Lameterca889e62006-06-30 01:55:44 -0700219
220void __inc_zone_page_state(struct page *page, enum zone_stat_item item)
221{
222 __inc_zone_state(page_zone(page), item);
223}
Christoph Lameter2244b952006-06-30 01:55:33 -0700224EXPORT_SYMBOL(__inc_zone_page_state);
225
Christoph Lameterc8785382007-02-10 01:43:01 -0800226void __dec_zone_state(struct zone *zone, enum zone_stat_item item)
Christoph Lameter2244b952006-06-30 01:55:33 -0700227{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +0900228 struct per_cpu_pageset *pcp = this_cpu_ptr(zone->pageset);
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700229 s8 *p = pcp->vm_stat_diff + item;
Christoph Lameter2244b952006-06-30 01:55:33 -0700230
231 (*p)--;
232
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700233 if (unlikely(*p < - pcp->stat_threshold)) {
234 int overstep = pcp->stat_threshold / 2;
235
236 zone_page_state_add(*p - overstep, zone, item);
237 *p = overstep;
Christoph Lameter2244b952006-06-30 01:55:33 -0700238 }
239}
Christoph Lameterc8785382007-02-10 01:43:01 -0800240
241void __dec_zone_page_state(struct page *page, enum zone_stat_item item)
242{
243 __dec_zone_state(page_zone(page), item);
244}
Christoph Lameter2244b952006-06-30 01:55:33 -0700245EXPORT_SYMBOL(__dec_zone_page_state);
246
Christoph Lameterca889e62006-06-30 01:55:44 -0700247void inc_zone_state(struct zone *zone, enum zone_stat_item item)
248{
249 unsigned long flags;
250
251 local_irq_save(flags);
252 __inc_zone_state(zone, item);
253 local_irq_restore(flags);
254}
255
Christoph Lameter2244b952006-06-30 01:55:33 -0700256void inc_zone_page_state(struct page *page, enum zone_stat_item item)
257{
258 unsigned long flags;
259 struct zone *zone;
Christoph Lameter2244b952006-06-30 01:55:33 -0700260
261 zone = page_zone(page);
262 local_irq_save(flags);
Christoph Lameterca889e62006-06-30 01:55:44 -0700263 __inc_zone_state(zone, item);
Christoph Lameter2244b952006-06-30 01:55:33 -0700264 local_irq_restore(flags);
265}
266EXPORT_SYMBOL(inc_zone_page_state);
267
268void dec_zone_page_state(struct page *page, enum zone_stat_item item)
269{
270 unsigned long flags;
Christoph Lameter2244b952006-06-30 01:55:33 -0700271
Christoph Lameter2244b952006-06-30 01:55:33 -0700272 local_irq_save(flags);
Christoph Lametera302eb42006-08-31 21:27:34 -0700273 __dec_zone_page_state(page, item);
Christoph Lameter2244b952006-06-30 01:55:33 -0700274 local_irq_restore(flags);
275}
276EXPORT_SYMBOL(dec_zone_page_state);
277
278/*
279 * Update the zone counters for one cpu.
Christoph Lameter4037d452007-05-09 02:35:14 -0700280 *
Christoph Lametera7f75e22008-02-04 22:29:16 -0800281 * The cpu specified must be either the current cpu or a processor that
282 * is not online. If it is the current cpu then the execution thread must
283 * be pinned to the current cpu.
284 *
Christoph Lameter4037d452007-05-09 02:35:14 -0700285 * Note that refresh_cpu_vm_stats strives to only access
286 * node local memory. The per cpu pagesets on remote zones are placed
287 * in the memory local to the processor using that pageset. So the
288 * loop over all zones will access a series of cachelines local to
289 * the processor.
290 *
291 * The call to zone_page_state_add updates the cachelines with the
292 * statistics in the remote zone struct as well as the global cachelines
293 * with the global counters. These could cause remote node cache line
294 * bouncing and will have to be only done when necessary.
Christoph Lameter2244b952006-06-30 01:55:33 -0700295 */
296void refresh_cpu_vm_stats(int cpu)
297{
298 struct zone *zone;
299 int i;
Christoph Lametera7f75e22008-02-04 22:29:16 -0800300 int global_diff[NR_VM_ZONE_STAT_ITEMS] = { 0, };
Christoph Lameter2244b952006-06-30 01:55:33 -0700301
KOSAKI Motohiroee99c712009-03-31 15:19:31 -0700302 for_each_populated_zone(zone) {
Christoph Lameter4037d452007-05-09 02:35:14 -0700303 struct per_cpu_pageset *p;
Christoph Lameter2244b952006-06-30 01:55:33 -0700304
Christoph Lameter99dcc3e2010-01-05 15:34:51 +0900305 p = per_cpu_ptr(zone->pageset, cpu);
Christoph Lameter2244b952006-06-30 01:55:33 -0700306
307 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
Christoph Lameter4037d452007-05-09 02:35:14 -0700308 if (p->vm_stat_diff[i]) {
Christoph Lametera7f75e22008-02-04 22:29:16 -0800309 unsigned long flags;
310 int v;
311
Christoph Lameter2244b952006-06-30 01:55:33 -0700312 local_irq_save(flags);
Christoph Lametera7f75e22008-02-04 22:29:16 -0800313 v = p->vm_stat_diff[i];
Christoph Lameter4037d452007-05-09 02:35:14 -0700314 p->vm_stat_diff[i] = 0;
Christoph Lametera7f75e22008-02-04 22:29:16 -0800315 local_irq_restore(flags);
316 atomic_long_add(v, &zone->vm_stat[i]);
317 global_diff[i] += v;
Christoph Lameter4037d452007-05-09 02:35:14 -0700318#ifdef CONFIG_NUMA
319 /* 3 seconds idle till flush */
320 p->expire = 3;
321#endif
Christoph Lameter2244b952006-06-30 01:55:33 -0700322 }
Dimitri Sivanich468fd622008-04-28 02:13:37 -0700323 cond_resched();
Christoph Lameter4037d452007-05-09 02:35:14 -0700324#ifdef CONFIG_NUMA
325 /*
326 * Deal with draining the remote pageset of this
327 * processor
328 *
329 * Check if there are pages remaining in this pageset
330 * if not then there is nothing to expire.
331 */
Christoph Lameter3dfa5722008-02-04 22:29:19 -0800332 if (!p->expire || !p->pcp.count)
Christoph Lameter4037d452007-05-09 02:35:14 -0700333 continue;
334
335 /*
336 * We never drain zones local to this processor.
337 */
338 if (zone_to_nid(zone) == numa_node_id()) {
339 p->expire = 0;
340 continue;
341 }
342
343 p->expire--;
344 if (p->expire)
345 continue;
346
Christoph Lameter3dfa5722008-02-04 22:29:19 -0800347 if (p->pcp.count)
348 drain_zone_pages(zone, &p->pcp);
Christoph Lameter4037d452007-05-09 02:35:14 -0700349#endif
Christoph Lameter2244b952006-06-30 01:55:33 -0700350 }
Christoph Lametera7f75e22008-02-04 22:29:16 -0800351
352 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
353 if (global_diff[i])
354 atomic_long_add(global_diff[i], &vm_stat[i]);
Christoph Lameter2244b952006-06-30 01:55:33 -0700355}
356
Christoph Lameter2244b952006-06-30 01:55:33 -0700357#endif
358
Christoph Lameterca889e62006-06-30 01:55:44 -0700359#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 */
Mel Gorman18ea7e72008-04-28 02:12:14 -0700366void zone_statistics(struct zone *preferred_zone, struct zone *z)
Christoph Lameterca889e62006-06-30 01:55:44 -0700367{
Mel Gorman18ea7e72008-04-28 02:12:14 -0700368 if (z->zone_pgdat == preferred_zone->zone_pgdat) {
Christoph Lameterca889e62006-06-30 01:55:44 -0700369 __inc_zone_state(z, NUMA_HIT);
370 } else {
371 __inc_zone_state(z, NUMA_MISS);
Mel Gorman18ea7e72008-04-28 02:12:14 -0700372 __inc_zone_state(preferred_zone, NUMA_FOREIGN);
Christoph Lameterca889e62006-06-30 01:55:44 -0700373 }
Christoph Lameter5d292342006-09-27 01:50:10 -0700374 if (z->node == numa_node_id())
Christoph Lameterca889e62006-06-30 01:55:44 -0700375 __inc_zone_state(z, NUMA_LOCAL);
376 else
377 __inc_zone_state(z, NUMA_OTHER);
378}
379#endif
380
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700381#ifdef CONFIG_PROC_FS
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +0400382#include <linux/proc_fs.h>
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700383#include <linux/seq_file.h>
384
Mel Gorman467c9962007-10-16 01:26:02 -0700385static char * const migratetype_names[MIGRATE_TYPES] = {
386 "Unmovable",
387 "Reclaimable",
388 "Movable",
389 "Reserve",
KOSAKI Motohiro91446b02008-04-15 14:34:42 -0700390 "Isolate",
Mel Gorman467c9962007-10-16 01:26:02 -0700391};
392
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700393static void *frag_start(struct seq_file *m, loff_t *pos)
394{
395 pg_data_t *pgdat;
396 loff_t node = *pos;
397 for (pgdat = first_online_pgdat();
398 pgdat && node;
399 pgdat = next_online_pgdat(pgdat))
400 --node;
401
402 return pgdat;
403}
404
405static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
406{
407 pg_data_t *pgdat = (pg_data_t *)arg;
408
409 (*pos)++;
410 return next_online_pgdat(pgdat);
411}
412
413static void frag_stop(struct seq_file *m, void *arg)
414{
415}
416
Mel Gorman467c9962007-10-16 01:26:02 -0700417/* Walk all the zones in a node and print using a callback */
418static void walk_zones_in_node(struct seq_file *m, pg_data_t *pgdat,
419 void (*print)(struct seq_file *m, pg_data_t *, struct zone *))
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700420{
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700421 struct zone *zone;
422 struct zone *node_zones = pgdat->node_zones;
423 unsigned long flags;
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700424
425 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
426 if (!populated_zone(zone))
427 continue;
428
429 spin_lock_irqsave(&zone->lock, flags);
Mel Gorman467c9962007-10-16 01:26:02 -0700430 print(m, pgdat, zone);
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700431 spin_unlock_irqrestore(&zone->lock, flags);
Mel Gorman467c9962007-10-16 01:26:02 -0700432 }
433}
434
435static void frag_show_print(struct seq_file *m, pg_data_t *pgdat,
436 struct zone *zone)
437{
438 int order;
439
440 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
441 for (order = 0; order < MAX_ORDER; ++order)
442 seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
443 seq_putc(m, '\n');
444}
445
446/*
447 * This walks the free areas for each zone.
448 */
449static int frag_show(struct seq_file *m, void *arg)
450{
451 pg_data_t *pgdat = (pg_data_t *)arg;
452 walk_zones_in_node(m, pgdat, frag_show_print);
453 return 0;
454}
455
456static void pagetypeinfo_showfree_print(struct seq_file *m,
457 pg_data_t *pgdat, struct zone *zone)
458{
459 int order, mtype;
460
461 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++) {
462 seq_printf(m, "Node %4d, zone %8s, type %12s ",
463 pgdat->node_id,
464 zone->name,
465 migratetype_names[mtype]);
466 for (order = 0; order < MAX_ORDER; ++order) {
467 unsigned long freecount = 0;
468 struct free_area *area;
469 struct list_head *curr;
470
471 area = &(zone->free_area[order]);
472
473 list_for_each(curr, &area->free_list[mtype])
474 freecount++;
475 seq_printf(m, "%6lu ", freecount);
476 }
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700477 seq_putc(m, '\n');
478 }
Mel Gorman467c9962007-10-16 01:26:02 -0700479}
480
481/* Print out the free pages at each order for each migatetype */
482static int pagetypeinfo_showfree(struct seq_file *m, void *arg)
483{
484 int order;
485 pg_data_t *pgdat = (pg_data_t *)arg;
486
487 /* Print header */
488 seq_printf(m, "%-43s ", "Free pages count per migrate type at order");
489 for (order = 0; order < MAX_ORDER; ++order)
490 seq_printf(m, "%6d ", order);
491 seq_putc(m, '\n');
492
493 walk_zones_in_node(m, pgdat, pagetypeinfo_showfree_print);
494
495 return 0;
496}
497
498static void pagetypeinfo_showblockcount_print(struct seq_file *m,
499 pg_data_t *pgdat, struct zone *zone)
500{
501 int mtype;
502 unsigned long pfn;
503 unsigned long start_pfn = zone->zone_start_pfn;
504 unsigned long end_pfn = start_pfn + zone->spanned_pages;
505 unsigned long count[MIGRATE_TYPES] = { 0, };
506
507 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
508 struct page *page;
509
510 if (!pfn_valid(pfn))
511 continue;
512
513 page = pfn_to_page(pfn);
Mel Gormaneb335752009-05-13 17:34:48 +0100514
515 /* Watch for unexpected holes punched in the memmap */
516 if (!memmap_valid_within(pfn, page, zone))
Mel Gormane80d6a22008-08-14 11:10:14 +0100517 continue;
Mel Gormaneb335752009-05-13 17:34:48 +0100518
Mel Gorman467c9962007-10-16 01:26:02 -0700519 mtype = get_pageblock_migratetype(page);
520
Mel Gormane80d6a22008-08-14 11:10:14 +0100521 if (mtype < MIGRATE_TYPES)
522 count[mtype]++;
Mel Gorman467c9962007-10-16 01:26:02 -0700523 }
524
525 /* Print counts */
526 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
527 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
528 seq_printf(m, "%12lu ", count[mtype]);
529 seq_putc(m, '\n');
530}
531
532/* Print out the free pages at each order for each migratetype */
533static int pagetypeinfo_showblockcount(struct seq_file *m, void *arg)
534{
535 int mtype;
536 pg_data_t *pgdat = (pg_data_t *)arg;
537
538 seq_printf(m, "\n%-23s", "Number of blocks type ");
539 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
540 seq_printf(m, "%12s ", migratetype_names[mtype]);
541 seq_putc(m, '\n');
542 walk_zones_in_node(m, pgdat, pagetypeinfo_showblockcount_print);
543
544 return 0;
545}
546
547/*
548 * This prints out statistics in relation to grouping pages by mobility.
549 * It is expensive to collect so do not constantly read the file.
550 */
551static int pagetypeinfo_show(struct seq_file *m, void *arg)
552{
553 pg_data_t *pgdat = (pg_data_t *)arg;
554
KOSAKI Motohiro41b25a32008-04-30 00:52:13 -0700555 /* check memoryless node */
556 if (!node_state(pgdat->node_id, N_HIGH_MEMORY))
557 return 0;
558
Mel Gorman467c9962007-10-16 01:26:02 -0700559 seq_printf(m, "Page block order: %d\n", pageblock_order);
560 seq_printf(m, "Pages per block: %lu\n", pageblock_nr_pages);
561 seq_putc(m, '\n');
562 pagetypeinfo_showfree(m, pgdat);
563 pagetypeinfo_showblockcount(m, pgdat);
564
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700565 return 0;
566}
567
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +0400568static const struct seq_operations fragmentation_op = {
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700569 .start = frag_start,
570 .next = frag_next,
571 .stop = frag_stop,
572 .show = frag_show,
573};
574
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +0400575static int fragmentation_open(struct inode *inode, struct file *file)
576{
577 return seq_open(file, &fragmentation_op);
578}
579
580static const struct file_operations fragmentation_file_operations = {
581 .open = fragmentation_open,
582 .read = seq_read,
583 .llseek = seq_lseek,
584 .release = seq_release,
585};
586
Alexey Dobriyan74e2e8e2008-10-06 04:15:36 +0400587static const struct seq_operations pagetypeinfo_op = {
Mel Gorman467c9962007-10-16 01:26:02 -0700588 .start = frag_start,
589 .next = frag_next,
590 .stop = frag_stop,
591 .show = pagetypeinfo_show,
592};
593
Alexey Dobriyan74e2e8e2008-10-06 04:15:36 +0400594static int pagetypeinfo_open(struct inode *inode, struct file *file)
595{
596 return seq_open(file, &pagetypeinfo_op);
597}
598
599static const struct file_operations pagetypeinfo_file_ops = {
600 .open = pagetypeinfo_open,
601 .read = seq_read,
602 .llseek = seq_lseek,
603 .release = seq_release,
604};
605
Christoph Lameter4b51d662007-02-10 01:43:10 -0800606#ifdef CONFIG_ZONE_DMA
607#define TEXT_FOR_DMA(xx) xx "_dma",
608#else
609#define TEXT_FOR_DMA(xx)
610#endif
611
Christoph Lameter27bf71c2006-09-25 23:31:15 -0700612#ifdef CONFIG_ZONE_DMA32
613#define TEXT_FOR_DMA32(xx) xx "_dma32",
614#else
615#define TEXT_FOR_DMA32(xx)
616#endif
617
618#ifdef CONFIG_HIGHMEM
619#define TEXT_FOR_HIGHMEM(xx) xx "_high",
620#else
621#define TEXT_FOR_HIGHMEM(xx)
622#endif
623
Christoph Lameter4b51d662007-02-10 01:43:10 -0800624#define TEXTS_FOR_ZONES(xx) TEXT_FOR_DMA(xx) TEXT_FOR_DMA32(xx) xx "_normal", \
Mel Gorman2a1e2742007-07-17 04:03:12 -0700625 TEXT_FOR_HIGHMEM(xx) xx "_movable",
Christoph Lameter27bf71c2006-09-25 23:31:15 -0700626
Helge Deller15ad7cd2006-12-06 20:40:36 -0800627static const char * const vmstat_text[] = {
Christoph Lameter2244b952006-06-30 01:55:33 -0700628 /* Zoned VM counters */
Christoph Lameterd23ad422007-02-10 01:43:02 -0800629 "nr_free_pages",
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700630 "nr_inactive_anon",
631 "nr_active_anon",
632 "nr_inactive_file",
633 "nr_active_file",
Lee Schermerhorn7b854122008-10-18 20:26:40 -0700634 "nr_unevictable",
Nick Piggin5344b7e2008-10-18 20:26:51 -0700635 "nr_mlock",
Christoph Lameterf3dbd342006-06-30 01:55:36 -0700636 "nr_anon_pages",
Christoph Lameter65ba55f2006-06-30 01:55:34 -0700637 "nr_mapped",
Christoph Lameter347ce432006-06-30 01:55:35 -0700638 "nr_file_pages",
Christoph Lameter51ed4492007-02-10 01:43:02 -0800639 "nr_dirty",
640 "nr_writeback",
Christoph Lameter972d1a72006-09-25 23:31:51 -0700641 "nr_slab_reclaimable",
642 "nr_slab_unreclaimable",
Christoph Lameterdf849a12006-06-30 01:55:38 -0700643 "nr_page_table_pages",
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -0700644 "nr_kernel_stack",
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700645 "nr_unstable",
Christoph Lameterd2c5e302006-06-30 01:55:41 -0700646 "nr_bounce",
Andrew Mortone129b5c2006-09-27 01:50:00 -0700647 "nr_vmscan_write",
Miklos Szeredifc3ba692008-04-30 00:54:38 -0700648 "nr_writeback_temp",
KOSAKI Motohiroa7312862009-09-21 17:01:37 -0700649 "nr_isolated_anon",
650 "nr_isolated_file",
KOSAKI Motohiro4b021082009-09-21 17:01:33 -0700651 "nr_shmem",
Christoph Lameterca889e62006-06-30 01:55:44 -0700652#ifdef CONFIG_NUMA
653 "numa_hit",
654 "numa_miss",
655 "numa_foreign",
656 "numa_interleave",
657 "numa_local",
658 "numa_other",
659#endif
660
Christoph Lameterf8891e52006-06-30 01:55:45 -0700661#ifdef CONFIG_VM_EVENT_COUNTERS
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700662 "pgpgin",
663 "pgpgout",
664 "pswpin",
665 "pswpout",
666
Christoph Lameter27bf71c2006-09-25 23:31:15 -0700667 TEXTS_FOR_ZONES("pgalloc")
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700668
669 "pgfree",
670 "pgactivate",
671 "pgdeactivate",
672
673 "pgfault",
674 "pgmajfault",
675
Christoph Lameter27bf71c2006-09-25 23:31:15 -0700676 TEXTS_FOR_ZONES("pgrefill")
677 TEXTS_FOR_ZONES("pgsteal")
678 TEXTS_FOR_ZONES("pgscan_kswapd")
679 TEXTS_FOR_ZONES("pgscan_direct")
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700680
Mel Gorman24cf725182009-06-16 15:33:23 -0700681#ifdef CONFIG_NUMA
682 "zone_reclaim_failed",
683#endif
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700684 "pginodesteal",
685 "slabs_scanned",
686 "kswapd_steal",
687 "kswapd_inodesteal",
KOSAKI Motohirobb3ab592009-12-14 17:58:55 -0800688 "kswapd_low_wmark_hit_quickly",
689 "kswapd_high_wmark_hit_quickly",
690 "kswapd_skip_congestion_wait",
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700691 "pageoutrun",
692 "allocstall",
693
694 "pgrotated",
Adam Litke3b116302008-04-28 02:13:06 -0700695#ifdef CONFIG_HUGETLB_PAGE
696 "htlb_buddy_alloc_success",
697 "htlb_buddy_alloc_fail",
698#endif
Lee Schermerhornbbfd28e2008-10-18 20:26:40 -0700699 "unevictable_pgs_culled",
700 "unevictable_pgs_scanned",
701 "unevictable_pgs_rescued",
Nick Piggin5344b7e2008-10-18 20:26:51 -0700702 "unevictable_pgs_mlocked",
703 "unevictable_pgs_munlocked",
704 "unevictable_pgs_cleared",
705 "unevictable_pgs_stranded",
Lee Schermerhorn985737c2008-10-18 20:26:53 -0700706 "unevictable_pgs_mlockfreed",
Lee Schermerhornbbfd28e2008-10-18 20:26:40 -0700707#endif
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700708};
709
Mel Gorman467c9962007-10-16 01:26:02 -0700710static void zoneinfo_show_print(struct seq_file *m, pg_data_t *pgdat,
711 struct zone *zone)
712{
713 int i;
714 seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
715 seq_printf(m,
716 "\n pages free %lu"
717 "\n min %lu"
718 "\n low %lu"
719 "\n high %lu"
Wu Fengguang08d9ae72009-06-16 15:32:30 -0700720 "\n scanned %lu"
Mel Gorman467c9962007-10-16 01:26:02 -0700721 "\n spanned %lu"
722 "\n present %lu",
723 zone_page_state(zone, NR_FREE_PAGES),
Mel Gorman41858962009-06-16 15:32:12 -0700724 min_wmark_pages(zone),
725 low_wmark_pages(zone),
726 high_wmark_pages(zone),
Mel Gorman467c9962007-10-16 01:26:02 -0700727 zone->pages_scanned,
Mel Gorman467c9962007-10-16 01:26:02 -0700728 zone->spanned_pages,
729 zone->present_pages);
730
731 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
732 seq_printf(m, "\n %-12s %lu", vmstat_text[i],
733 zone_page_state(zone, i));
734
735 seq_printf(m,
736 "\n protection: (%lu",
737 zone->lowmem_reserve[0]);
738 for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
739 seq_printf(m, ", %lu", zone->lowmem_reserve[i]);
740 seq_printf(m,
741 ")"
742 "\n pagesets");
743 for_each_online_cpu(i) {
744 struct per_cpu_pageset *pageset;
Mel Gorman467c9962007-10-16 01:26:02 -0700745
Christoph Lameter99dcc3e2010-01-05 15:34:51 +0900746 pageset = per_cpu_ptr(zone->pageset, i);
Christoph Lameter3dfa5722008-02-04 22:29:19 -0800747 seq_printf(m,
748 "\n cpu: %i"
749 "\n count: %i"
750 "\n high: %i"
751 "\n batch: %i",
752 i,
753 pageset->pcp.count,
754 pageset->pcp.high,
755 pageset->pcp.batch);
Mel Gorman467c9962007-10-16 01:26:02 -0700756#ifdef CONFIG_SMP
757 seq_printf(m, "\n vm stats threshold: %d",
758 pageset->stat_threshold);
759#endif
760 }
761 seq_printf(m,
762 "\n all_unreclaimable: %u"
763 "\n prev_priority: %i"
Rik van Riel556adec2008-10-18 20:26:34 -0700764 "\n start_pfn: %lu"
765 "\n inactive_ratio: %u",
KOSAKI Motohiro93e4a892010-03-05 13:41:55 -0800766 zone->all_unreclaimable,
Mel Gorman467c9962007-10-16 01:26:02 -0700767 zone->prev_priority,
Rik van Riel556adec2008-10-18 20:26:34 -0700768 zone->zone_start_pfn,
769 zone->inactive_ratio);
Mel Gorman467c9962007-10-16 01:26:02 -0700770 seq_putc(m, '\n');
771}
772
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700773/*
774 * Output information about zones in @pgdat.
775 */
776static int zoneinfo_show(struct seq_file *m, void *arg)
777{
Mel Gorman467c9962007-10-16 01:26:02 -0700778 pg_data_t *pgdat = (pg_data_t *)arg;
779 walk_zones_in_node(m, pgdat, zoneinfo_show_print);
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700780 return 0;
781}
782
Alexey Dobriyan5c9fe622008-10-06 04:19:42 +0400783static const struct seq_operations zoneinfo_op = {
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700784 .start = frag_start, /* iterate over all zones. The same as in
785 * fragmentation. */
786 .next = frag_next,
787 .stop = frag_stop,
788 .show = zoneinfo_show,
789};
790
Alexey Dobriyan5c9fe622008-10-06 04:19:42 +0400791static int zoneinfo_open(struct inode *inode, struct file *file)
792{
793 return seq_open(file, &zoneinfo_op);
794}
795
796static const struct file_operations proc_zoneinfo_file_operations = {
797 .open = zoneinfo_open,
798 .read = seq_read,
799 .llseek = seq_lseek,
800 .release = seq_release,
801};
802
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700803static void *vmstat_start(struct seq_file *m, loff_t *pos)
804{
Christoph Lameter2244b952006-06-30 01:55:33 -0700805 unsigned long *v;
Christoph Lameterf8891e52006-06-30 01:55:45 -0700806#ifdef CONFIG_VM_EVENT_COUNTERS
807 unsigned long *e;
808#endif
Christoph Lameter2244b952006-06-30 01:55:33 -0700809 int i;
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700810
811 if (*pos >= ARRAY_SIZE(vmstat_text))
812 return NULL;
813
Christoph Lameterf8891e52006-06-30 01:55:45 -0700814#ifdef CONFIG_VM_EVENT_COUNTERS
Christoph Lameter2244b952006-06-30 01:55:33 -0700815 v = kmalloc(NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long)
Christoph Lameterf8891e52006-06-30 01:55:45 -0700816 + sizeof(struct vm_event_state), GFP_KERNEL);
817#else
818 v = kmalloc(NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long),
819 GFP_KERNEL);
820#endif
Christoph Lameter2244b952006-06-30 01:55:33 -0700821 m->private = v;
822 if (!v)
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700823 return ERR_PTR(-ENOMEM);
Christoph Lameter2244b952006-06-30 01:55:33 -0700824 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
825 v[i] = global_page_state(i);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700826#ifdef CONFIG_VM_EVENT_COUNTERS
827 e = v + NR_VM_ZONE_STAT_ITEMS;
828 all_vm_events(e);
829 e[PGPGIN] /= 2; /* sectors -> kbytes */
830 e[PGPGOUT] /= 2;
831#endif
Christoph Lameter2244b952006-06-30 01:55:33 -0700832 return v + *pos;
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700833}
834
835static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
836{
837 (*pos)++;
838 if (*pos >= ARRAY_SIZE(vmstat_text))
839 return NULL;
840 return (unsigned long *)m->private + *pos;
841}
842
843static int vmstat_show(struct seq_file *m, void *arg)
844{
845 unsigned long *l = arg;
846 unsigned long off = l - (unsigned long *)m->private;
847
848 seq_printf(m, "%s %lu\n", vmstat_text[off], *l);
849 return 0;
850}
851
852static void vmstat_stop(struct seq_file *m, void *arg)
853{
854 kfree(m->private);
855 m->private = NULL;
856}
857
Alexey Dobriyanb6aa44a2008-10-06 04:17:48 +0400858static const struct seq_operations vmstat_op = {
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700859 .start = vmstat_start,
860 .next = vmstat_next,
861 .stop = vmstat_stop,
862 .show = vmstat_show,
863};
864
Alexey Dobriyanb6aa44a2008-10-06 04:17:48 +0400865static int vmstat_open(struct inode *inode, struct file *file)
866{
867 return seq_open(file, &vmstat_op);
868}
869
870static const struct file_operations proc_vmstat_file_operations = {
871 .open = vmstat_open,
872 .read = seq_read,
873 .llseek = seq_lseek,
874 .release = seq_release,
875};
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700876#endif /* CONFIG_PROC_FS */
877
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700878#ifdef CONFIG_SMP
Christoph Lameterd1187ed2007-05-09 02:35:12 -0700879static DEFINE_PER_CPU(struct delayed_work, vmstat_work);
Christoph Lameter77461ab2007-05-09 02:35:13 -0700880int sysctl_stat_interval __read_mostly = HZ;
Christoph Lameterd1187ed2007-05-09 02:35:12 -0700881
882static void vmstat_update(struct work_struct *w)
883{
884 refresh_cpu_vm_stats(smp_processor_id());
Christoph Lameter77461ab2007-05-09 02:35:13 -0700885 schedule_delayed_work(&__get_cpu_var(vmstat_work),
Anton Blanchard98f4ebb2009-04-02 16:56:39 -0700886 round_jiffies_relative(sysctl_stat_interval));
Christoph Lameterd1187ed2007-05-09 02:35:12 -0700887}
888
Randy Dunlap42614fc2007-11-14 17:00:12 -0800889static void __cpuinit start_cpu_timer(int cpu)
Christoph Lameterd1187ed2007-05-09 02:35:12 -0700890{
Tejun Heo1871e522009-10-29 22:34:13 +0900891 struct delayed_work *work = &per_cpu(vmstat_work, cpu);
Christoph Lameterd1187ed2007-05-09 02:35:12 -0700892
Tejun Heo1871e522009-10-29 22:34:13 +0900893 INIT_DELAYED_WORK_DEFERRABLE(work, vmstat_update);
894 schedule_delayed_work_on(cpu, work, __round_jiffies_relative(HZ, cpu));
Christoph Lameterd1187ed2007-05-09 02:35:12 -0700895}
896
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700897/*
898 * Use the cpu notifier to insure that the thresholds are recalculated
899 * when necessary.
900 */
901static int __cpuinit vmstat_cpuup_callback(struct notifier_block *nfb,
902 unsigned long action,
903 void *hcpu)
904{
Christoph Lameterd1187ed2007-05-09 02:35:12 -0700905 long cpu = (long)hcpu;
906
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700907 switch (action) {
Christoph Lameterd1187ed2007-05-09 02:35:12 -0700908 case CPU_ONLINE:
909 case CPU_ONLINE_FROZEN:
910 start_cpu_timer(cpu);
Christoph Lameterad596922010-01-05 15:34:51 +0900911 node_set_state(cpu_to_node(cpu), N_CPU);
Christoph Lameterd1187ed2007-05-09 02:35:12 -0700912 break;
913 case CPU_DOWN_PREPARE:
914 case CPU_DOWN_PREPARE_FROZEN:
915 cancel_rearming_delayed_work(&per_cpu(vmstat_work, cpu));
916 per_cpu(vmstat_work, cpu).work.func = NULL;
917 break;
918 case CPU_DOWN_FAILED:
919 case CPU_DOWN_FAILED_FROZEN:
920 start_cpu_timer(cpu);
921 break;
Andy Whitcroftce421c72006-12-06 20:33:08 -0800922 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -0700923 case CPU_DEAD_FROZEN:
Andy Whitcroftce421c72006-12-06 20:33:08 -0800924 refresh_zone_stat_thresholds();
925 break;
926 default:
927 break;
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700928 }
929 return NOTIFY_OK;
930}
931
932static struct notifier_block __cpuinitdata vmstat_notifier =
933 { &vmstat_cpuup_callback, NULL, 0 };
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +0400934#endif
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700935
Adrian Bunke2fc88d2007-10-16 01:26:27 -0700936static int __init setup_vmstat(void)
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700937{
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +0400938#ifdef CONFIG_SMP
Christoph Lameterd1187ed2007-05-09 02:35:12 -0700939 int cpu;
940
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700941 refresh_zone_stat_thresholds();
942 register_cpu_notifier(&vmstat_notifier);
Christoph Lameterd1187ed2007-05-09 02:35:12 -0700943
944 for_each_online_cpu(cpu)
945 start_cpu_timer(cpu);
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +0400946#endif
947#ifdef CONFIG_PROC_FS
948 proc_create("buddyinfo", S_IRUGO, NULL, &fragmentation_file_operations);
Alexey Dobriyan74e2e8e2008-10-06 04:15:36 +0400949 proc_create("pagetypeinfo", S_IRUGO, NULL, &pagetypeinfo_file_ops);
Alexey Dobriyanb6aa44a2008-10-06 04:17:48 +0400950 proc_create("vmstat", S_IRUGO, NULL, &proc_vmstat_file_operations);
Alexey Dobriyan5c9fe622008-10-06 04:19:42 +0400951 proc_create("zoneinfo", S_IRUGO, NULL, &proc_zoneinfo_file_operations);
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +0400952#endif
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700953 return 0;
954}
955module_init(setup_vmstat)