blob: ceda39b63d7ea37c6e0e3fa3bf7fef3457627684 [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)
142 zone_pcp(zone, cpu)->stat_threshold = threshold;
143 }
Christoph Lameter2244b952006-06-30 01:55:33 -0700144}
145
146/*
147 * For use when we know that interrupts are disabled.
148 */
149void __mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
150 int delta)
151{
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700152 struct per_cpu_pageset *pcp = zone_pcp(zone, smp_processor_id());
153 s8 *p = pcp->vm_stat_diff + item;
Christoph Lameter2244b952006-06-30 01:55:33 -0700154 long x;
155
Christoph Lameter2244b952006-06-30 01:55:33 -0700156 x = delta + *p;
157
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700158 if (unlikely(x > pcp->stat_threshold || x < -pcp->stat_threshold)) {
Christoph Lameter2244b952006-06-30 01:55:33 -0700159 zone_page_state_add(x, zone, item);
160 x = 0;
161 }
Christoph Lameter2244b952006-06-30 01:55:33 -0700162 *p = x;
163}
164EXPORT_SYMBOL(__mod_zone_page_state);
165
166/*
167 * For an unknown interrupt state
168 */
169void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
170 int delta)
171{
172 unsigned long flags;
173
174 local_irq_save(flags);
175 __mod_zone_page_state(zone, item, delta);
176 local_irq_restore(flags);
177}
178EXPORT_SYMBOL(mod_zone_page_state);
179
180/*
181 * Optimized increment and decrement functions.
182 *
183 * These are only for a single page and therefore can take a struct page *
184 * argument instead of struct zone *. This allows the inclusion of the code
185 * generated for page_zone(page) into the optimized functions.
186 *
187 * No overflow check is necessary and therefore the differential can be
188 * incremented or decremented in place which may allow the compilers to
189 * generate better code.
Christoph Lameter2244b952006-06-30 01:55:33 -0700190 * The increment or decrement is known and therefore one boundary check can
191 * be omitted.
192 *
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700193 * NOTE: These functions are very performance sensitive. Change only
194 * with care.
195 *
Christoph Lameter2244b952006-06-30 01:55:33 -0700196 * Some processors have inc/dec instructions that are atomic vs an interrupt.
197 * However, the code must first determine the differential location in a zone
198 * based on the processor number and then inc/dec the counter. There is no
199 * guarantee without disabling preemption that the processor will not change
200 * in between and therefore the atomicity vs. interrupt cannot be exploited
201 * in a useful way here.
202 */
Christoph Lameterc8785382007-02-10 01:43:01 -0800203void __inc_zone_state(struct zone *zone, enum zone_stat_item item)
Christoph Lameter2244b952006-06-30 01:55:33 -0700204{
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700205 struct per_cpu_pageset *pcp = zone_pcp(zone, smp_processor_id());
206 s8 *p = pcp->vm_stat_diff + item;
Christoph Lameter2244b952006-06-30 01:55:33 -0700207
208 (*p)++;
209
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700210 if (unlikely(*p > pcp->stat_threshold)) {
211 int overstep = pcp->stat_threshold / 2;
212
213 zone_page_state_add(*p + overstep, zone, item);
214 *p = -overstep;
Christoph Lameter2244b952006-06-30 01:55:33 -0700215 }
216}
Christoph Lameterca889e62006-06-30 01:55:44 -0700217
218void __inc_zone_page_state(struct page *page, enum zone_stat_item item)
219{
220 __inc_zone_state(page_zone(page), item);
221}
Christoph Lameter2244b952006-06-30 01:55:33 -0700222EXPORT_SYMBOL(__inc_zone_page_state);
223
Christoph Lameterc8785382007-02-10 01:43:01 -0800224void __dec_zone_state(struct zone *zone, enum zone_stat_item item)
Christoph Lameter2244b952006-06-30 01:55:33 -0700225{
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700226 struct per_cpu_pageset *pcp = zone_pcp(zone, smp_processor_id());
227 s8 *p = pcp->vm_stat_diff + item;
Christoph Lameter2244b952006-06-30 01:55:33 -0700228
229 (*p)--;
230
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700231 if (unlikely(*p < - pcp->stat_threshold)) {
232 int overstep = pcp->stat_threshold / 2;
233
234 zone_page_state_add(*p - overstep, zone, item);
235 *p = overstep;
Christoph Lameter2244b952006-06-30 01:55:33 -0700236 }
237}
Christoph Lameterc8785382007-02-10 01:43:01 -0800238
239void __dec_zone_page_state(struct page *page, enum zone_stat_item item)
240{
241 __dec_zone_state(page_zone(page), item);
242}
Christoph Lameter2244b952006-06-30 01:55:33 -0700243EXPORT_SYMBOL(__dec_zone_page_state);
244
Christoph Lameterca889e62006-06-30 01:55:44 -0700245void inc_zone_state(struct zone *zone, enum zone_stat_item item)
246{
247 unsigned long flags;
248
249 local_irq_save(flags);
250 __inc_zone_state(zone, item);
251 local_irq_restore(flags);
252}
253
Christoph Lameter2244b952006-06-30 01:55:33 -0700254void inc_zone_page_state(struct page *page, enum zone_stat_item item)
255{
256 unsigned long flags;
257 struct zone *zone;
Christoph Lameter2244b952006-06-30 01:55:33 -0700258
259 zone = page_zone(page);
260 local_irq_save(flags);
Christoph Lameterca889e62006-06-30 01:55:44 -0700261 __inc_zone_state(zone, item);
Christoph Lameter2244b952006-06-30 01:55:33 -0700262 local_irq_restore(flags);
263}
264EXPORT_SYMBOL(inc_zone_page_state);
265
266void dec_zone_page_state(struct page *page, enum zone_stat_item item)
267{
268 unsigned long flags;
Christoph Lameter2244b952006-06-30 01:55:33 -0700269
Christoph Lameter2244b952006-06-30 01:55:33 -0700270 local_irq_save(flags);
Christoph Lametera302eb42006-08-31 21:27:34 -0700271 __dec_zone_page_state(page, item);
Christoph Lameter2244b952006-06-30 01:55:33 -0700272 local_irq_restore(flags);
273}
274EXPORT_SYMBOL(dec_zone_page_state);
275
276/*
277 * Update the zone counters for one cpu.
Christoph Lameter4037d452007-05-09 02:35:14 -0700278 *
Christoph Lametera7f75e22008-02-04 22:29:16 -0800279 * The cpu specified must be either the current cpu or a processor that
280 * is not online. If it is the current cpu then the execution thread must
281 * be pinned to the current cpu.
282 *
Christoph Lameter4037d452007-05-09 02:35:14 -0700283 * Note that refresh_cpu_vm_stats strives to only access
284 * node local memory. The per cpu pagesets on remote zones are placed
285 * in the memory local to the processor using that pageset. So the
286 * loop over all zones will access a series of cachelines local to
287 * the processor.
288 *
289 * The call to zone_page_state_add updates the cachelines with the
290 * statistics in the remote zone struct as well as the global cachelines
291 * with the global counters. These could cause remote node cache line
292 * bouncing and will have to be only done when necessary.
Christoph Lameter2244b952006-06-30 01:55:33 -0700293 */
294void refresh_cpu_vm_stats(int cpu)
295{
296 struct zone *zone;
297 int i;
Christoph Lametera7f75e22008-02-04 22:29:16 -0800298 int global_diff[NR_VM_ZONE_STAT_ITEMS] = { 0, };
Christoph Lameter2244b952006-06-30 01:55:33 -0700299
KOSAKI Motohiroee99c712009-03-31 15:19:31 -0700300 for_each_populated_zone(zone) {
Christoph Lameter4037d452007-05-09 02:35:14 -0700301 struct per_cpu_pageset *p;
Christoph Lameter2244b952006-06-30 01:55:33 -0700302
Christoph Lameter4037d452007-05-09 02:35:14 -0700303 p = zone_pcp(zone, cpu);
Christoph Lameter2244b952006-06-30 01:55:33 -0700304
305 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
Christoph Lameter4037d452007-05-09 02:35:14 -0700306 if (p->vm_stat_diff[i]) {
Christoph Lametera7f75e22008-02-04 22:29:16 -0800307 unsigned long flags;
308 int v;
309
Christoph Lameter2244b952006-06-30 01:55:33 -0700310 local_irq_save(flags);
Christoph Lametera7f75e22008-02-04 22:29:16 -0800311 v = p->vm_stat_diff[i];
Christoph Lameter4037d452007-05-09 02:35:14 -0700312 p->vm_stat_diff[i] = 0;
Christoph Lametera7f75e22008-02-04 22:29:16 -0800313 local_irq_restore(flags);
314 atomic_long_add(v, &zone->vm_stat[i]);
315 global_diff[i] += v;
Christoph Lameter4037d452007-05-09 02:35:14 -0700316#ifdef CONFIG_NUMA
317 /* 3 seconds idle till flush */
318 p->expire = 3;
319#endif
Christoph Lameter2244b952006-06-30 01:55:33 -0700320 }
Dimitri Sivanich468fd622008-04-28 02:13:37 -0700321 cond_resched();
Christoph Lameter4037d452007-05-09 02:35:14 -0700322#ifdef CONFIG_NUMA
323 /*
324 * Deal with draining the remote pageset of this
325 * processor
326 *
327 * Check if there are pages remaining in this pageset
328 * if not then there is nothing to expire.
329 */
Christoph Lameter3dfa5722008-02-04 22:29:19 -0800330 if (!p->expire || !p->pcp.count)
Christoph Lameter4037d452007-05-09 02:35:14 -0700331 continue;
332
333 /*
334 * We never drain zones local to this processor.
335 */
336 if (zone_to_nid(zone) == numa_node_id()) {
337 p->expire = 0;
338 continue;
339 }
340
341 p->expire--;
342 if (p->expire)
343 continue;
344
Christoph Lameter3dfa5722008-02-04 22:29:19 -0800345 if (p->pcp.count)
346 drain_zone_pages(zone, &p->pcp);
Christoph Lameter4037d452007-05-09 02:35:14 -0700347#endif
Christoph Lameter2244b952006-06-30 01:55:33 -0700348 }
Christoph Lametera7f75e22008-02-04 22:29:16 -0800349
350 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
351 if (global_diff[i])
352 atomic_long_add(global_diff[i], &vm_stat[i]);
Christoph Lameter2244b952006-06-30 01:55:33 -0700353}
354
Christoph Lameter2244b952006-06-30 01:55:33 -0700355#endif
356
Christoph Lameterca889e62006-06-30 01:55:44 -0700357#ifdef CONFIG_NUMA
358/*
359 * zonelist = the list of zones passed to the allocator
360 * z = the zone from which the allocation occurred.
361 *
362 * Must be called with interrupts disabled.
363 */
Mel Gorman18ea7e72008-04-28 02:12:14 -0700364void zone_statistics(struct zone *preferred_zone, struct zone *z)
Christoph Lameterca889e62006-06-30 01:55:44 -0700365{
Mel Gorman18ea7e72008-04-28 02:12:14 -0700366 if (z->zone_pgdat == preferred_zone->zone_pgdat) {
Christoph Lameterca889e62006-06-30 01:55:44 -0700367 __inc_zone_state(z, NUMA_HIT);
368 } else {
369 __inc_zone_state(z, NUMA_MISS);
Mel Gorman18ea7e72008-04-28 02:12:14 -0700370 __inc_zone_state(preferred_zone, NUMA_FOREIGN);
Christoph Lameterca889e62006-06-30 01:55:44 -0700371 }
Christoph Lameter5d292342006-09-27 01:50:10 -0700372 if (z->node == numa_node_id())
Christoph Lameterca889e62006-06-30 01:55:44 -0700373 __inc_zone_state(z, NUMA_LOCAL);
374 else
375 __inc_zone_state(z, NUMA_OTHER);
376}
377#endif
378
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700379#ifdef CONFIG_PROC_FS
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +0400380#include <linux/proc_fs.h>
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700381#include <linux/seq_file.h>
382
Mel Gorman467c9962007-10-16 01:26:02 -0700383static char * const migratetype_names[MIGRATE_TYPES] = {
384 "Unmovable",
385 "Reclaimable",
386 "Movable",
387 "Reserve",
KOSAKI Motohiro91446b02008-04-15 14:34:42 -0700388 "Isolate",
Mel Gorman467c9962007-10-16 01:26:02 -0700389};
390
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700391static void *frag_start(struct seq_file *m, loff_t *pos)
392{
393 pg_data_t *pgdat;
394 loff_t node = *pos;
395 for (pgdat = first_online_pgdat();
396 pgdat && node;
397 pgdat = next_online_pgdat(pgdat))
398 --node;
399
400 return pgdat;
401}
402
403static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
404{
405 pg_data_t *pgdat = (pg_data_t *)arg;
406
407 (*pos)++;
408 return next_online_pgdat(pgdat);
409}
410
411static void frag_stop(struct seq_file *m, void *arg)
412{
413}
414
Mel Gorman467c9962007-10-16 01:26:02 -0700415/* Walk all the zones in a node and print using a callback */
416static void walk_zones_in_node(struct seq_file *m, pg_data_t *pgdat,
417 void (*print)(struct seq_file *m, pg_data_t *, struct zone *))
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700418{
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700419 struct zone *zone;
420 struct zone *node_zones = pgdat->node_zones;
421 unsigned long flags;
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700422
423 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
424 if (!populated_zone(zone))
425 continue;
426
427 spin_lock_irqsave(&zone->lock, flags);
Mel Gorman467c9962007-10-16 01:26:02 -0700428 print(m, pgdat, zone);
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700429 spin_unlock_irqrestore(&zone->lock, flags);
Mel Gorman467c9962007-10-16 01:26:02 -0700430 }
431}
432
433static void frag_show_print(struct seq_file *m, pg_data_t *pgdat,
434 struct zone *zone)
435{
436 int order;
437
438 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
439 for (order = 0; order < MAX_ORDER; ++order)
440 seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
441 seq_putc(m, '\n');
442}
443
444/*
445 * This walks the free areas for each zone.
446 */
447static int frag_show(struct seq_file *m, void *arg)
448{
449 pg_data_t *pgdat = (pg_data_t *)arg;
450 walk_zones_in_node(m, pgdat, frag_show_print);
451 return 0;
452}
453
454static void pagetypeinfo_showfree_print(struct seq_file *m,
455 pg_data_t *pgdat, struct zone *zone)
456{
457 int order, mtype;
458
459 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++) {
460 seq_printf(m, "Node %4d, zone %8s, type %12s ",
461 pgdat->node_id,
462 zone->name,
463 migratetype_names[mtype]);
464 for (order = 0; order < MAX_ORDER; ++order) {
465 unsigned long freecount = 0;
466 struct free_area *area;
467 struct list_head *curr;
468
469 area = &(zone->free_area[order]);
470
471 list_for_each(curr, &area->free_list[mtype])
472 freecount++;
473 seq_printf(m, "%6lu ", freecount);
474 }
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700475 seq_putc(m, '\n');
476 }
Mel Gorman467c9962007-10-16 01:26:02 -0700477}
478
479/* Print out the free pages at each order for each migatetype */
480static int pagetypeinfo_showfree(struct seq_file *m, void *arg)
481{
482 int order;
483 pg_data_t *pgdat = (pg_data_t *)arg;
484
485 /* Print header */
486 seq_printf(m, "%-43s ", "Free pages count per migrate type at order");
487 for (order = 0; order < MAX_ORDER; ++order)
488 seq_printf(m, "%6d ", order);
489 seq_putc(m, '\n');
490
491 walk_zones_in_node(m, pgdat, pagetypeinfo_showfree_print);
492
493 return 0;
494}
495
496static void pagetypeinfo_showblockcount_print(struct seq_file *m,
497 pg_data_t *pgdat, struct zone *zone)
498{
499 int mtype;
500 unsigned long pfn;
501 unsigned long start_pfn = zone->zone_start_pfn;
502 unsigned long end_pfn = start_pfn + zone->spanned_pages;
503 unsigned long count[MIGRATE_TYPES] = { 0, };
504
505 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
506 struct page *page;
507
508 if (!pfn_valid(pfn))
509 continue;
510
511 page = pfn_to_page(pfn);
Mel Gormaneb335752009-05-13 17:34:48 +0100512
513 /* Watch for unexpected holes punched in the memmap */
514 if (!memmap_valid_within(pfn, page, zone))
Mel Gormane80d6a22008-08-14 11:10:14 +0100515 continue;
Mel Gormaneb335752009-05-13 17:34:48 +0100516
Mel Gorman467c9962007-10-16 01:26:02 -0700517 mtype = get_pageblock_migratetype(page);
518
Mel Gormane80d6a22008-08-14 11:10:14 +0100519 if (mtype < MIGRATE_TYPES)
520 count[mtype]++;
Mel Gorman467c9962007-10-16 01:26:02 -0700521 }
522
523 /* Print counts */
524 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
525 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
526 seq_printf(m, "%12lu ", count[mtype]);
527 seq_putc(m, '\n');
528}
529
530/* Print out the free pages at each order for each migratetype */
531static int pagetypeinfo_showblockcount(struct seq_file *m, void *arg)
532{
533 int mtype;
534 pg_data_t *pgdat = (pg_data_t *)arg;
535
536 seq_printf(m, "\n%-23s", "Number of blocks type ");
537 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
538 seq_printf(m, "%12s ", migratetype_names[mtype]);
539 seq_putc(m, '\n');
540 walk_zones_in_node(m, pgdat, pagetypeinfo_showblockcount_print);
541
542 return 0;
543}
544
545/*
546 * This prints out statistics in relation to grouping pages by mobility.
547 * It is expensive to collect so do not constantly read the file.
548 */
549static int pagetypeinfo_show(struct seq_file *m, void *arg)
550{
551 pg_data_t *pgdat = (pg_data_t *)arg;
552
KOSAKI Motohiro41b25a32008-04-30 00:52:13 -0700553 /* check memoryless node */
554 if (!node_state(pgdat->node_id, N_HIGH_MEMORY))
555 return 0;
556
Mel Gorman467c9962007-10-16 01:26:02 -0700557 seq_printf(m, "Page block order: %d\n", pageblock_order);
558 seq_printf(m, "Pages per block: %lu\n", pageblock_nr_pages);
559 seq_putc(m, '\n');
560 pagetypeinfo_showfree(m, pgdat);
561 pagetypeinfo_showblockcount(m, pgdat);
562
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700563 return 0;
564}
565
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +0400566static const struct seq_operations fragmentation_op = {
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700567 .start = frag_start,
568 .next = frag_next,
569 .stop = frag_stop,
570 .show = frag_show,
571};
572
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +0400573static int fragmentation_open(struct inode *inode, struct file *file)
574{
575 return seq_open(file, &fragmentation_op);
576}
577
578static const struct file_operations fragmentation_file_operations = {
579 .open = fragmentation_open,
580 .read = seq_read,
581 .llseek = seq_lseek,
582 .release = seq_release,
583};
584
Alexey Dobriyan74e2e8e2008-10-06 04:15:36 +0400585static const struct seq_operations pagetypeinfo_op = {
Mel Gorman467c9962007-10-16 01:26:02 -0700586 .start = frag_start,
587 .next = frag_next,
588 .stop = frag_stop,
589 .show = pagetypeinfo_show,
590};
591
Alexey Dobriyan74e2e8e2008-10-06 04:15:36 +0400592static int pagetypeinfo_open(struct inode *inode, struct file *file)
593{
594 return seq_open(file, &pagetypeinfo_op);
595}
596
597static const struct file_operations pagetypeinfo_file_ops = {
598 .open = pagetypeinfo_open,
599 .read = seq_read,
600 .llseek = seq_lseek,
601 .release = seq_release,
602};
603
Christoph Lameter4b51d662007-02-10 01:43:10 -0800604#ifdef CONFIG_ZONE_DMA
605#define TEXT_FOR_DMA(xx) xx "_dma",
606#else
607#define TEXT_FOR_DMA(xx)
608#endif
609
Christoph Lameter27bf71c2006-09-25 23:31:15 -0700610#ifdef CONFIG_ZONE_DMA32
611#define TEXT_FOR_DMA32(xx) xx "_dma32",
612#else
613#define TEXT_FOR_DMA32(xx)
614#endif
615
616#ifdef CONFIG_HIGHMEM
617#define TEXT_FOR_HIGHMEM(xx) xx "_high",
618#else
619#define TEXT_FOR_HIGHMEM(xx)
620#endif
621
Christoph Lameter4b51d662007-02-10 01:43:10 -0800622#define TEXTS_FOR_ZONES(xx) TEXT_FOR_DMA(xx) TEXT_FOR_DMA32(xx) xx "_normal", \
Mel Gorman2a1e2742007-07-17 04:03:12 -0700623 TEXT_FOR_HIGHMEM(xx) xx "_movable",
Christoph Lameter27bf71c2006-09-25 23:31:15 -0700624
Helge Deller15ad7cd2006-12-06 20:40:36 -0800625static const char * const vmstat_text[] = {
Christoph Lameter2244b952006-06-30 01:55:33 -0700626 /* Zoned VM counters */
Christoph Lameterd23ad422007-02-10 01:43:02 -0800627 "nr_free_pages",
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700628 "nr_inactive_anon",
629 "nr_active_anon",
630 "nr_inactive_file",
631 "nr_active_file",
Lee Schermerhorn7b854122008-10-18 20:26:40 -0700632 "nr_unevictable",
Nick Piggin5344b7e2008-10-18 20:26:51 -0700633 "nr_mlock",
Christoph Lameterf3dbd342006-06-30 01:55:36 -0700634 "nr_anon_pages",
Christoph Lameter65ba55f2006-06-30 01:55:34 -0700635 "nr_mapped",
Christoph Lameter347ce432006-06-30 01:55:35 -0700636 "nr_file_pages",
Christoph Lameter51ed4492007-02-10 01:43:02 -0800637 "nr_dirty",
638 "nr_writeback",
Christoph Lameter972d1a72006-09-25 23:31:51 -0700639 "nr_slab_reclaimable",
640 "nr_slab_unreclaimable",
Christoph Lameterdf849a12006-06-30 01:55:38 -0700641 "nr_page_table_pages",
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -0700642 "nr_kernel_stack",
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700643 "nr_unstable",
Christoph Lameterd2c5e302006-06-30 01:55:41 -0700644 "nr_bounce",
Andrew Mortone129b5c2006-09-27 01:50:00 -0700645 "nr_vmscan_write",
Miklos Szeredifc3ba692008-04-30 00:54:38 -0700646 "nr_writeback_temp",
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700647
Christoph Lameterca889e62006-06-30 01:55:44 -0700648#ifdef CONFIG_NUMA
649 "numa_hit",
650 "numa_miss",
651 "numa_foreign",
652 "numa_interleave",
653 "numa_local",
654 "numa_other",
655#endif
656
Christoph Lameterf8891e52006-06-30 01:55:45 -0700657#ifdef CONFIG_VM_EVENT_COUNTERS
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700658 "pgpgin",
659 "pgpgout",
660 "pswpin",
661 "pswpout",
662
Christoph Lameter27bf71c2006-09-25 23:31:15 -0700663 TEXTS_FOR_ZONES("pgalloc")
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700664
665 "pgfree",
666 "pgactivate",
667 "pgdeactivate",
668
669 "pgfault",
670 "pgmajfault",
671
Christoph Lameter27bf71c2006-09-25 23:31:15 -0700672 TEXTS_FOR_ZONES("pgrefill")
673 TEXTS_FOR_ZONES("pgsteal")
674 TEXTS_FOR_ZONES("pgscan_kswapd")
675 TEXTS_FOR_ZONES("pgscan_direct")
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700676
Mel Gorman24cf725182009-06-16 15:33:23 -0700677#ifdef CONFIG_NUMA
678 "zone_reclaim_failed",
679#endif
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700680 "pginodesteal",
681 "slabs_scanned",
682 "kswapd_steal",
683 "kswapd_inodesteal",
684 "pageoutrun",
685 "allocstall",
686
687 "pgrotated",
Adam Litke3b116302008-04-28 02:13:06 -0700688#ifdef CONFIG_HUGETLB_PAGE
689 "htlb_buddy_alloc_success",
690 "htlb_buddy_alloc_fail",
691#endif
Lee Schermerhornbbfd28e2008-10-18 20:26:40 -0700692 "unevictable_pgs_culled",
693 "unevictable_pgs_scanned",
694 "unevictable_pgs_rescued",
Nick Piggin5344b7e2008-10-18 20:26:51 -0700695 "unevictable_pgs_mlocked",
696 "unevictable_pgs_munlocked",
697 "unevictable_pgs_cleared",
698 "unevictable_pgs_stranded",
Lee Schermerhorn985737c2008-10-18 20:26:53 -0700699 "unevictable_pgs_mlockfreed",
Lee Schermerhornbbfd28e2008-10-18 20:26:40 -0700700#endif
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700701};
702
Mel Gorman467c9962007-10-16 01:26:02 -0700703static void zoneinfo_show_print(struct seq_file *m, pg_data_t *pgdat,
704 struct zone *zone)
705{
706 int i;
707 seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
708 seq_printf(m,
709 "\n pages free %lu"
710 "\n min %lu"
711 "\n low %lu"
712 "\n high %lu"
Wu Fengguang08d9ae72009-06-16 15:32:30 -0700713 "\n scanned %lu"
Mel Gorman467c9962007-10-16 01:26:02 -0700714 "\n spanned %lu"
715 "\n present %lu",
716 zone_page_state(zone, NR_FREE_PAGES),
Mel Gorman41858962009-06-16 15:32:12 -0700717 min_wmark_pages(zone),
718 low_wmark_pages(zone),
719 high_wmark_pages(zone),
Mel Gorman467c9962007-10-16 01:26:02 -0700720 zone->pages_scanned,
Mel Gorman467c9962007-10-16 01:26:02 -0700721 zone->spanned_pages,
722 zone->present_pages);
723
724 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
725 seq_printf(m, "\n %-12s %lu", vmstat_text[i],
726 zone_page_state(zone, i));
727
728 seq_printf(m,
729 "\n protection: (%lu",
730 zone->lowmem_reserve[0]);
731 for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
732 seq_printf(m, ", %lu", zone->lowmem_reserve[i]);
733 seq_printf(m,
734 ")"
735 "\n pagesets");
736 for_each_online_cpu(i) {
737 struct per_cpu_pageset *pageset;
Mel Gorman467c9962007-10-16 01:26:02 -0700738
739 pageset = zone_pcp(zone, i);
Christoph Lameter3dfa5722008-02-04 22:29:19 -0800740 seq_printf(m,
741 "\n cpu: %i"
742 "\n count: %i"
743 "\n high: %i"
744 "\n batch: %i",
745 i,
746 pageset->pcp.count,
747 pageset->pcp.high,
748 pageset->pcp.batch);
Mel Gorman467c9962007-10-16 01:26:02 -0700749#ifdef CONFIG_SMP
750 seq_printf(m, "\n vm stats threshold: %d",
751 pageset->stat_threshold);
752#endif
753 }
754 seq_printf(m,
755 "\n all_unreclaimable: %u"
756 "\n prev_priority: %i"
Rik van Riel556adec2008-10-18 20:26:34 -0700757 "\n start_pfn: %lu"
758 "\n inactive_ratio: %u",
David Rientjese815af92007-10-16 23:25:54 -0700759 zone_is_all_unreclaimable(zone),
Mel Gorman467c9962007-10-16 01:26:02 -0700760 zone->prev_priority,
Rik van Riel556adec2008-10-18 20:26:34 -0700761 zone->zone_start_pfn,
762 zone->inactive_ratio);
Mel Gorman467c9962007-10-16 01:26:02 -0700763 seq_putc(m, '\n');
764}
765
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700766/*
767 * Output information about zones in @pgdat.
768 */
769static int zoneinfo_show(struct seq_file *m, void *arg)
770{
Mel Gorman467c9962007-10-16 01:26:02 -0700771 pg_data_t *pgdat = (pg_data_t *)arg;
772 walk_zones_in_node(m, pgdat, zoneinfo_show_print);
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700773 return 0;
774}
775
Alexey Dobriyan5c9fe622008-10-06 04:19:42 +0400776static const struct seq_operations zoneinfo_op = {
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700777 .start = frag_start, /* iterate over all zones. The same as in
778 * fragmentation. */
779 .next = frag_next,
780 .stop = frag_stop,
781 .show = zoneinfo_show,
782};
783
Alexey Dobriyan5c9fe622008-10-06 04:19:42 +0400784static int zoneinfo_open(struct inode *inode, struct file *file)
785{
786 return seq_open(file, &zoneinfo_op);
787}
788
789static const struct file_operations proc_zoneinfo_file_operations = {
790 .open = zoneinfo_open,
791 .read = seq_read,
792 .llseek = seq_lseek,
793 .release = seq_release,
794};
795
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700796static void *vmstat_start(struct seq_file *m, loff_t *pos)
797{
Christoph Lameter2244b952006-06-30 01:55:33 -0700798 unsigned long *v;
Christoph Lameterf8891e52006-06-30 01:55:45 -0700799#ifdef CONFIG_VM_EVENT_COUNTERS
800 unsigned long *e;
801#endif
Christoph Lameter2244b952006-06-30 01:55:33 -0700802 int i;
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700803
804 if (*pos >= ARRAY_SIZE(vmstat_text))
805 return NULL;
806
Christoph Lameterf8891e52006-06-30 01:55:45 -0700807#ifdef CONFIG_VM_EVENT_COUNTERS
Christoph Lameter2244b952006-06-30 01:55:33 -0700808 v = kmalloc(NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long)
Christoph Lameterf8891e52006-06-30 01:55:45 -0700809 + sizeof(struct vm_event_state), GFP_KERNEL);
810#else
811 v = kmalloc(NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long),
812 GFP_KERNEL);
813#endif
Christoph Lameter2244b952006-06-30 01:55:33 -0700814 m->private = v;
815 if (!v)
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700816 return ERR_PTR(-ENOMEM);
Christoph Lameter2244b952006-06-30 01:55:33 -0700817 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
818 v[i] = global_page_state(i);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700819#ifdef CONFIG_VM_EVENT_COUNTERS
820 e = v + NR_VM_ZONE_STAT_ITEMS;
821 all_vm_events(e);
822 e[PGPGIN] /= 2; /* sectors -> kbytes */
823 e[PGPGOUT] /= 2;
824#endif
Christoph Lameter2244b952006-06-30 01:55:33 -0700825 return v + *pos;
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700826}
827
828static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
829{
830 (*pos)++;
831 if (*pos >= ARRAY_SIZE(vmstat_text))
832 return NULL;
833 return (unsigned long *)m->private + *pos;
834}
835
836static int vmstat_show(struct seq_file *m, void *arg)
837{
838 unsigned long *l = arg;
839 unsigned long off = l - (unsigned long *)m->private;
840
841 seq_printf(m, "%s %lu\n", vmstat_text[off], *l);
842 return 0;
843}
844
845static void vmstat_stop(struct seq_file *m, void *arg)
846{
847 kfree(m->private);
848 m->private = NULL;
849}
850
Alexey Dobriyanb6aa44a2008-10-06 04:17:48 +0400851static const struct seq_operations vmstat_op = {
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700852 .start = vmstat_start,
853 .next = vmstat_next,
854 .stop = vmstat_stop,
855 .show = vmstat_show,
856};
857
Alexey Dobriyanb6aa44a2008-10-06 04:17:48 +0400858static int vmstat_open(struct inode *inode, struct file *file)
859{
860 return seq_open(file, &vmstat_op);
861}
862
863static const struct file_operations proc_vmstat_file_operations = {
864 .open = vmstat_open,
865 .read = seq_read,
866 .llseek = seq_lseek,
867 .release = seq_release,
868};
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700869#endif /* CONFIG_PROC_FS */
870
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700871#ifdef CONFIG_SMP
Christoph Lameterd1187ed2007-05-09 02:35:12 -0700872static DEFINE_PER_CPU(struct delayed_work, vmstat_work);
Christoph Lameter77461ab2007-05-09 02:35:13 -0700873int sysctl_stat_interval __read_mostly = HZ;
Christoph Lameterd1187ed2007-05-09 02:35:12 -0700874
875static void vmstat_update(struct work_struct *w)
876{
877 refresh_cpu_vm_stats(smp_processor_id());
Christoph Lameter77461ab2007-05-09 02:35:13 -0700878 schedule_delayed_work(&__get_cpu_var(vmstat_work),
Anton Blanchard98f4ebb2009-04-02 16:56:39 -0700879 round_jiffies_relative(sysctl_stat_interval));
Christoph Lameterd1187ed2007-05-09 02:35:12 -0700880}
881
Randy Dunlap42614fc2007-11-14 17:00:12 -0800882static void __cpuinit start_cpu_timer(int cpu)
Christoph Lameterd1187ed2007-05-09 02:35:12 -0700883{
884 struct delayed_work *vmstat_work = &per_cpu(vmstat_work, cpu);
885
Christoph Lameter39bf6272007-05-10 22:22:21 -0700886 INIT_DELAYED_WORK_DEFERRABLE(vmstat_work, vmstat_update);
Anton Blanchard98f4ebb2009-04-02 16:56:39 -0700887 schedule_delayed_work_on(cpu, vmstat_work,
888 __round_jiffies_relative(HZ, cpu));
Christoph Lameterd1187ed2007-05-09 02:35:12 -0700889}
890
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700891/*
892 * Use the cpu notifier to insure that the thresholds are recalculated
893 * when necessary.
894 */
895static int __cpuinit vmstat_cpuup_callback(struct notifier_block *nfb,
896 unsigned long action,
897 void *hcpu)
898{
Christoph Lameterd1187ed2007-05-09 02:35:12 -0700899 long cpu = (long)hcpu;
900
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700901 switch (action) {
Christoph Lameterd1187ed2007-05-09 02:35:12 -0700902 case CPU_ONLINE:
903 case CPU_ONLINE_FROZEN:
904 start_cpu_timer(cpu);
905 break;
906 case CPU_DOWN_PREPARE:
907 case CPU_DOWN_PREPARE_FROZEN:
908 cancel_rearming_delayed_work(&per_cpu(vmstat_work, cpu));
909 per_cpu(vmstat_work, cpu).work.func = NULL;
910 break;
911 case CPU_DOWN_FAILED:
912 case CPU_DOWN_FAILED_FROZEN:
913 start_cpu_timer(cpu);
914 break;
Andy Whitcroftce421c72006-12-06 20:33:08 -0800915 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -0700916 case CPU_DEAD_FROZEN:
Andy Whitcroftce421c72006-12-06 20:33:08 -0800917 refresh_zone_stat_thresholds();
918 break;
919 default:
920 break;
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700921 }
922 return NOTIFY_OK;
923}
924
925static struct notifier_block __cpuinitdata vmstat_notifier =
926 { &vmstat_cpuup_callback, NULL, 0 };
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +0400927#endif
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700928
Adrian Bunke2fc88d2007-10-16 01:26:27 -0700929static int __init setup_vmstat(void)
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700930{
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +0400931#ifdef CONFIG_SMP
Christoph Lameterd1187ed2007-05-09 02:35:12 -0700932 int cpu;
933
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700934 refresh_zone_stat_thresholds();
935 register_cpu_notifier(&vmstat_notifier);
Christoph Lameterd1187ed2007-05-09 02:35:12 -0700936
937 for_each_online_cpu(cpu)
938 start_cpu_timer(cpu);
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +0400939#endif
940#ifdef CONFIG_PROC_FS
941 proc_create("buddyinfo", S_IRUGO, NULL, &fragmentation_file_operations);
Alexey Dobriyan74e2e8e2008-10-06 04:15:36 +0400942 proc_create("pagetypeinfo", S_IRUGO, NULL, &pagetypeinfo_file_ops);
Alexey Dobriyanb6aa44a2008-10-06 04:17:48 +0400943 proc_create("vmstat", S_IRUGO, NULL, &proc_vmstat_file_operations);
Alexey Dobriyan5c9fe622008-10-06 04:19:42 +0400944 proc_create("zoneinfo", S_IRUGO, NULL, &proc_zoneinfo_file_operations);
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +0400945#endif
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700946 return 0;
947}
948module_init(setup_vmstat)