blob: f389168f9a837b9c6be4e1f9bb3d0892396315de [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>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090015#include <linux/slab.h>
Christoph Lameterdf9ecab2006-08-31 21:27:35 -070016#include <linux/cpu.h>
Adrian Bunkc748e132008-07-23 21:27:03 -070017#include <linux/vmstat.h>
Alexey Dobriyane8edc6e2007-05-21 01:22:52 +040018#include <linux/sched.h>
Mel Gormanf1a5ab12010-05-24 14:32:26 -070019#include <linux/math64.h>
Christoph Lameterf6ac2352006-06-30 01:55:32 -070020
Christoph Lameterf8891e52006-06-30 01:55:45 -070021#ifdef CONFIG_VM_EVENT_COUNTERS
22DEFINE_PER_CPU(struct vm_event_state, vm_event_states) = {{0}};
23EXPORT_PER_CPU_SYMBOL(vm_event_states);
24
Minchan Kim31f961a2010-08-09 17:18:59 -070025static void sum_vm_events(unsigned long *ret)
Christoph Lameterf8891e52006-06-30 01:55:45 -070026{
Christoph Lameter9eccf2a2008-02-04 22:29:22 -080027 int cpu;
Christoph Lameterf8891e52006-06-30 01:55:45 -070028 int i;
29
30 memset(ret, 0, NR_VM_EVENT_ITEMS * sizeof(unsigned long));
31
Minchan Kim31f961a2010-08-09 17:18:59 -070032 for_each_online_cpu(cpu) {
Christoph Lameterf8891e52006-06-30 01:55:45 -070033 struct vm_event_state *this = &per_cpu(vm_event_states, cpu);
34
Christoph Lameterf8891e52006-06-30 01:55:45 -070035 for (i = 0; i < NR_VM_EVENT_ITEMS; i++)
36 ret[i] += this->event[i];
37 }
38}
39
40/*
41 * Accumulate the vm event counters across all CPUs.
42 * The result is unavoidably approximate - it can change
43 * during and after execution of this function.
44*/
45void all_vm_events(unsigned long *ret)
46{
KOSAKI Motohirob5be1132008-05-12 14:02:06 -070047 get_online_cpus();
Minchan Kim31f961a2010-08-09 17:18:59 -070048 sum_vm_events(ret);
KOSAKI Motohirob5be1132008-05-12 14:02:06 -070049 put_online_cpus();
Christoph Lameterf8891e52006-06-30 01:55:45 -070050}
Heiko Carstens32dd66f2006-07-10 04:44:31 -070051EXPORT_SYMBOL_GPL(all_vm_events);
Christoph Lameterf8891e52006-06-30 01:55:45 -070052
53#ifdef CONFIG_HOTPLUG
54/*
55 * Fold the foreign cpu events into our own.
56 *
57 * This is adding to the events on one processor
58 * but keeps the global counts constant.
59 */
60void vm_events_fold_cpu(int cpu)
61{
62 struct vm_event_state *fold_state = &per_cpu(vm_event_states, cpu);
63 int i;
64
65 for (i = 0; i < NR_VM_EVENT_ITEMS; i++) {
66 count_vm_events(i, fold_state->event[i]);
67 fold_state->event[i] = 0;
68 }
69}
70#endif /* CONFIG_HOTPLUG */
71
72#endif /* CONFIG_VM_EVENT_COUNTERS */
73
Christoph Lameter2244b952006-06-30 01:55:33 -070074/*
75 * Manage combined zone based / global counters
76 *
77 * vm_stat contains the global counters
78 */
79atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
80EXPORT_SYMBOL(vm_stat);
81
82#ifdef CONFIG_SMP
83
Christoph Lameterdf9ecab2006-08-31 21:27:35 -070084static int calculate_threshold(struct zone *zone)
85{
86 int threshold;
87 int mem; /* memory in 128 MB units */
88
89 /*
90 * The threshold scales with the number of processors and the amount
91 * of memory per zone. More memory means that we can defer updates for
92 * longer, more processors could lead to more contention.
93 * fls() is used to have a cheap way of logarithmic scaling.
94 *
95 * Some sample thresholds:
96 *
97 * Threshold Processors (fls) Zonesize fls(mem+1)
98 * ------------------------------------------------------------------
99 * 8 1 1 0.9-1 GB 4
100 * 16 2 2 0.9-1 GB 4
101 * 20 2 2 1-2 GB 5
102 * 24 2 2 2-4 GB 6
103 * 28 2 2 4-8 GB 7
104 * 32 2 2 8-16 GB 8
105 * 4 2 2 <128M 1
106 * 30 4 3 2-4 GB 5
107 * 48 4 3 8-16 GB 8
108 * 32 8 4 1-2 GB 4
109 * 32 8 4 0.9-1GB 4
110 * 10 16 5 <128M 1
111 * 40 16 5 900M 4
112 * 70 64 7 2-4 GB 5
113 * 84 64 7 4-8 GB 6
114 * 108 512 9 4-8 GB 6
115 * 125 1024 10 8-16 GB 8
116 * 125 1024 10 16-32 GB 9
117 */
118
119 mem = zone->present_pages >> (27 - PAGE_SHIFT);
120
121 threshold = 2 * fls(num_online_cpus()) * (1 + fls(mem));
122
123 /*
124 * Maximum threshold is 125
125 */
126 threshold = min(125, threshold);
127
128 return threshold;
129}
Christoph Lameter2244b952006-06-30 01:55:33 -0700130
131/*
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700132 * Refresh the thresholds for each zone.
Christoph Lameter2244b952006-06-30 01:55:33 -0700133 */
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700134static void refresh_zone_stat_thresholds(void)
Christoph Lameter2244b952006-06-30 01:55:33 -0700135{
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700136 struct zone *zone;
137 int cpu;
138 int threshold;
139
KOSAKI Motohiroee99c712009-03-31 15:19:31 -0700140 for_each_populated_zone(zone) {
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700141 threshold = calculate_threshold(zone);
142
143 for_each_online_cpu(cpu)
Christoph Lameter99dcc3e2010-01-05 15:34:51 +0900144 per_cpu_ptr(zone->pageset, cpu)->stat_threshold
145 = threshold;
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700146 }
Christoph Lameter2244b952006-06-30 01:55:33 -0700147}
148
149/*
150 * For use when we know that interrupts are disabled.
151 */
152void __mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
153 int delta)
154{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +0900155 struct per_cpu_pageset *pcp = this_cpu_ptr(zone->pageset);
156
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700157 s8 *p = pcp->vm_stat_diff + item;
Christoph Lameter2244b952006-06-30 01:55:33 -0700158 long x;
159
Christoph Lameter2244b952006-06-30 01:55:33 -0700160 x = delta + *p;
161
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700162 if (unlikely(x > pcp->stat_threshold || x < -pcp->stat_threshold)) {
Christoph Lameter2244b952006-06-30 01:55:33 -0700163 zone_page_state_add(x, zone, item);
164 x = 0;
165 }
Christoph Lameter2244b952006-06-30 01:55:33 -0700166 *p = x;
167}
168EXPORT_SYMBOL(__mod_zone_page_state);
169
170/*
171 * For an unknown interrupt state
172 */
173void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
174 int delta)
175{
176 unsigned long flags;
177
178 local_irq_save(flags);
179 __mod_zone_page_state(zone, item, delta);
180 local_irq_restore(flags);
181}
182EXPORT_SYMBOL(mod_zone_page_state);
183
184/*
185 * Optimized increment and decrement functions.
186 *
187 * These are only for a single page and therefore can take a struct page *
188 * argument instead of struct zone *. This allows the inclusion of the code
189 * generated for page_zone(page) into the optimized functions.
190 *
191 * No overflow check is necessary and therefore the differential can be
192 * incremented or decremented in place which may allow the compilers to
193 * generate better code.
Christoph Lameter2244b952006-06-30 01:55:33 -0700194 * The increment or decrement is known and therefore one boundary check can
195 * be omitted.
196 *
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700197 * NOTE: These functions are very performance sensitive. Change only
198 * with care.
199 *
Christoph Lameter2244b952006-06-30 01:55:33 -0700200 * Some processors have inc/dec instructions that are atomic vs an interrupt.
201 * However, the code must first determine the differential location in a zone
202 * based on the processor number and then inc/dec the counter. There is no
203 * guarantee without disabling preemption that the processor will not change
204 * in between and therefore the atomicity vs. interrupt cannot be exploited
205 * in a useful way here.
206 */
Christoph Lameterc8785382007-02-10 01:43:01 -0800207void __inc_zone_state(struct zone *zone, enum zone_stat_item item)
Christoph Lameter2244b952006-06-30 01:55:33 -0700208{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +0900209 struct per_cpu_pageset *pcp = this_cpu_ptr(zone->pageset);
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700210 s8 *p = pcp->vm_stat_diff + item;
Christoph Lameter2244b952006-06-30 01:55:33 -0700211
212 (*p)++;
213
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700214 if (unlikely(*p > pcp->stat_threshold)) {
215 int overstep = pcp->stat_threshold / 2;
216
217 zone_page_state_add(*p + overstep, zone, item);
218 *p = -overstep;
Christoph Lameter2244b952006-06-30 01:55:33 -0700219 }
220}
Christoph Lameterca889e62006-06-30 01:55:44 -0700221
222void __inc_zone_page_state(struct page *page, enum zone_stat_item item)
223{
224 __inc_zone_state(page_zone(page), item);
225}
Christoph Lameter2244b952006-06-30 01:55:33 -0700226EXPORT_SYMBOL(__inc_zone_page_state);
227
Christoph Lameterc8785382007-02-10 01:43:01 -0800228void __dec_zone_state(struct zone *zone, enum zone_stat_item item)
Christoph Lameter2244b952006-06-30 01:55:33 -0700229{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +0900230 struct per_cpu_pageset *pcp = this_cpu_ptr(zone->pageset);
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700231 s8 *p = pcp->vm_stat_diff + item;
Christoph Lameter2244b952006-06-30 01:55:33 -0700232
233 (*p)--;
234
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700235 if (unlikely(*p < - pcp->stat_threshold)) {
236 int overstep = pcp->stat_threshold / 2;
237
238 zone_page_state_add(*p - overstep, zone, item);
239 *p = overstep;
Christoph Lameter2244b952006-06-30 01:55:33 -0700240 }
241}
Christoph Lameterc8785382007-02-10 01:43:01 -0800242
243void __dec_zone_page_state(struct page *page, enum zone_stat_item item)
244{
245 __dec_zone_state(page_zone(page), item);
246}
Christoph Lameter2244b952006-06-30 01:55:33 -0700247EXPORT_SYMBOL(__dec_zone_page_state);
248
Christoph Lameterca889e62006-06-30 01:55:44 -0700249void inc_zone_state(struct zone *zone, enum zone_stat_item item)
250{
251 unsigned long flags;
252
253 local_irq_save(flags);
254 __inc_zone_state(zone, item);
255 local_irq_restore(flags);
256}
257
Christoph Lameter2244b952006-06-30 01:55:33 -0700258void inc_zone_page_state(struct page *page, enum zone_stat_item item)
259{
260 unsigned long flags;
261 struct zone *zone;
Christoph Lameter2244b952006-06-30 01:55:33 -0700262
263 zone = page_zone(page);
264 local_irq_save(flags);
Christoph Lameterca889e62006-06-30 01:55:44 -0700265 __inc_zone_state(zone, item);
Christoph Lameter2244b952006-06-30 01:55:33 -0700266 local_irq_restore(flags);
267}
268EXPORT_SYMBOL(inc_zone_page_state);
269
270void dec_zone_page_state(struct page *page, enum zone_stat_item item)
271{
272 unsigned long flags;
Christoph Lameter2244b952006-06-30 01:55:33 -0700273
Christoph Lameter2244b952006-06-30 01:55:33 -0700274 local_irq_save(flags);
Christoph Lametera302eb42006-08-31 21:27:34 -0700275 __dec_zone_page_state(page, item);
Christoph Lameter2244b952006-06-30 01:55:33 -0700276 local_irq_restore(flags);
277}
278EXPORT_SYMBOL(dec_zone_page_state);
279
280/*
281 * Update the zone counters for one cpu.
Christoph Lameter4037d452007-05-09 02:35:14 -0700282 *
Christoph Lametera7f75e22008-02-04 22:29:16 -0800283 * The cpu specified must be either the current cpu or a processor that
284 * is not online. If it is the current cpu then the execution thread must
285 * be pinned to the current cpu.
286 *
Christoph Lameter4037d452007-05-09 02:35:14 -0700287 * Note that refresh_cpu_vm_stats strives to only access
288 * node local memory. The per cpu pagesets on remote zones are placed
289 * in the memory local to the processor using that pageset. So the
290 * loop over all zones will access a series of cachelines local to
291 * the processor.
292 *
293 * The call to zone_page_state_add updates the cachelines with the
294 * statistics in the remote zone struct as well as the global cachelines
295 * with the global counters. These could cause remote node cache line
296 * bouncing and will have to be only done when necessary.
Christoph Lameter2244b952006-06-30 01:55:33 -0700297 */
298void refresh_cpu_vm_stats(int cpu)
299{
300 struct zone *zone;
301 int i;
Christoph Lametera7f75e22008-02-04 22:29:16 -0800302 int global_diff[NR_VM_ZONE_STAT_ITEMS] = { 0, };
Christoph Lameter2244b952006-06-30 01:55:33 -0700303
KOSAKI Motohiroee99c712009-03-31 15:19:31 -0700304 for_each_populated_zone(zone) {
Christoph Lameter4037d452007-05-09 02:35:14 -0700305 struct per_cpu_pageset *p;
Christoph Lameter2244b952006-06-30 01:55:33 -0700306
Christoph Lameter99dcc3e2010-01-05 15:34:51 +0900307 p = per_cpu_ptr(zone->pageset, cpu);
Christoph Lameter2244b952006-06-30 01:55:33 -0700308
309 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
Christoph Lameter4037d452007-05-09 02:35:14 -0700310 if (p->vm_stat_diff[i]) {
Christoph Lametera7f75e22008-02-04 22:29:16 -0800311 unsigned long flags;
312 int v;
313
Christoph Lameter2244b952006-06-30 01:55:33 -0700314 local_irq_save(flags);
Christoph Lametera7f75e22008-02-04 22:29:16 -0800315 v = p->vm_stat_diff[i];
Christoph Lameter4037d452007-05-09 02:35:14 -0700316 p->vm_stat_diff[i] = 0;
Christoph Lametera7f75e22008-02-04 22:29:16 -0800317 local_irq_restore(flags);
318 atomic_long_add(v, &zone->vm_stat[i]);
319 global_diff[i] += v;
Christoph Lameter4037d452007-05-09 02:35:14 -0700320#ifdef CONFIG_NUMA
321 /* 3 seconds idle till flush */
322 p->expire = 3;
323#endif
Christoph Lameter2244b952006-06-30 01:55:33 -0700324 }
Dimitri Sivanich468fd622008-04-28 02:13:37 -0700325 cond_resched();
Christoph Lameter4037d452007-05-09 02:35:14 -0700326#ifdef CONFIG_NUMA
327 /*
328 * Deal with draining the remote pageset of this
329 * processor
330 *
331 * Check if there are pages remaining in this pageset
332 * if not then there is nothing to expire.
333 */
Christoph Lameter3dfa5722008-02-04 22:29:19 -0800334 if (!p->expire || !p->pcp.count)
Christoph Lameter4037d452007-05-09 02:35:14 -0700335 continue;
336
337 /*
338 * We never drain zones local to this processor.
339 */
340 if (zone_to_nid(zone) == numa_node_id()) {
341 p->expire = 0;
342 continue;
343 }
344
345 p->expire--;
346 if (p->expire)
347 continue;
348
Christoph Lameter3dfa5722008-02-04 22:29:19 -0800349 if (p->pcp.count)
350 drain_zone_pages(zone, &p->pcp);
Christoph Lameter4037d452007-05-09 02:35:14 -0700351#endif
Christoph Lameter2244b952006-06-30 01:55:33 -0700352 }
Christoph Lametera7f75e22008-02-04 22:29:16 -0800353
354 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
355 if (global_diff[i])
356 atomic_long_add(global_diff[i], &vm_stat[i]);
Christoph Lameter2244b952006-06-30 01:55:33 -0700357}
358
Christoph Lameter2244b952006-06-30 01:55:33 -0700359#endif
360
Christoph Lameterca889e62006-06-30 01:55:44 -0700361#ifdef CONFIG_NUMA
362/*
363 * zonelist = the list of zones passed to the allocator
364 * z = the zone from which the allocation occurred.
365 *
366 * Must be called with interrupts disabled.
367 */
Mel Gorman18ea7e72008-04-28 02:12:14 -0700368void zone_statistics(struct zone *preferred_zone, struct zone *z)
Christoph Lameterca889e62006-06-30 01:55:44 -0700369{
Mel Gorman18ea7e72008-04-28 02:12:14 -0700370 if (z->zone_pgdat == preferred_zone->zone_pgdat) {
Christoph Lameterca889e62006-06-30 01:55:44 -0700371 __inc_zone_state(z, NUMA_HIT);
372 } else {
373 __inc_zone_state(z, NUMA_MISS);
Mel Gorman18ea7e72008-04-28 02:12:14 -0700374 __inc_zone_state(preferred_zone, NUMA_FOREIGN);
Christoph Lameterca889e62006-06-30 01:55:44 -0700375 }
Christoph Lameter5d292342006-09-27 01:50:10 -0700376 if (z->node == numa_node_id())
Christoph Lameterca889e62006-06-30 01:55:44 -0700377 __inc_zone_state(z, NUMA_LOCAL);
378 else
379 __inc_zone_state(z, NUMA_OTHER);
380}
381#endif
382
Mel Gormand7a57522010-05-24 14:32:25 -0700383#ifdef CONFIG_COMPACTION
384struct contig_page_info {
385 unsigned long free_pages;
386 unsigned long free_blocks_total;
387 unsigned long free_blocks_suitable;
388};
389
390/*
391 * Calculate the number of free pages in a zone, how many contiguous
392 * pages are free and how many are large enough to satisfy an allocation of
393 * the target size. Note that this function makes no attempt to estimate
394 * how many suitable free blocks there *might* be if MOVABLE pages were
395 * migrated. Calculating that is possible, but expensive and can be
396 * figured out from userspace
397 */
398static void fill_contig_page_info(struct zone *zone,
399 unsigned int suitable_order,
400 struct contig_page_info *info)
401{
402 unsigned int order;
403
404 info->free_pages = 0;
405 info->free_blocks_total = 0;
406 info->free_blocks_suitable = 0;
407
408 for (order = 0; order < MAX_ORDER; order++) {
409 unsigned long blocks;
410
411 /* Count number of free blocks */
412 blocks = zone->free_area[order].nr_free;
413 info->free_blocks_total += blocks;
414
415 /* Count free base pages */
416 info->free_pages += blocks << order;
417
418 /* Count the suitable free blocks */
419 if (order >= suitable_order)
420 info->free_blocks_suitable += blocks <<
421 (order - suitable_order);
422 }
423}
Mel Gormanf1a5ab12010-05-24 14:32:26 -0700424
425/*
426 * A fragmentation index only makes sense if an allocation of a requested
427 * size would fail. If that is true, the fragmentation index indicates
428 * whether external fragmentation or a lack of memory was the problem.
429 * The value can be used to determine if page reclaim or compaction
430 * should be used
431 */
Mel Gorman56de7262010-05-24 14:32:30 -0700432static int __fragmentation_index(unsigned int order, struct contig_page_info *info)
Mel Gormanf1a5ab12010-05-24 14:32:26 -0700433{
434 unsigned long requested = 1UL << order;
435
436 if (!info->free_blocks_total)
437 return 0;
438
439 /* Fragmentation index only makes sense when a request would fail */
440 if (info->free_blocks_suitable)
441 return -1000;
442
443 /*
444 * Index is between 0 and 1 so return within 3 decimal places
445 *
446 * 0 => allocation would fail due to lack of memory
447 * 1 => allocation would fail due to fragmentation
448 */
449 return 1000 - div_u64( (1000+(div_u64(info->free_pages * 1000ULL, requested))), info->free_blocks_total);
450}
Mel Gorman56de7262010-05-24 14:32:30 -0700451
452/* Same as __fragmentation index but allocs contig_page_info on stack */
453int fragmentation_index(struct zone *zone, unsigned int order)
454{
455 struct contig_page_info info;
456
457 fill_contig_page_info(zone, order, &info);
458 return __fragmentation_index(order, &info);
459}
Mel Gormand7a57522010-05-24 14:32:25 -0700460#endif
461
462#if defined(CONFIG_PROC_FS) || defined(CONFIG_COMPACTION)
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +0400463#include <linux/proc_fs.h>
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700464#include <linux/seq_file.h>
465
Mel Gorman467c9962007-10-16 01:26:02 -0700466static char * const migratetype_names[MIGRATE_TYPES] = {
467 "Unmovable",
468 "Reclaimable",
469 "Movable",
470 "Reserve",
KOSAKI Motohiro91446b02008-04-15 14:34:42 -0700471 "Isolate",
Mel Gorman467c9962007-10-16 01:26:02 -0700472};
473
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700474static void *frag_start(struct seq_file *m, loff_t *pos)
475{
476 pg_data_t *pgdat;
477 loff_t node = *pos;
478 for (pgdat = first_online_pgdat();
479 pgdat && node;
480 pgdat = next_online_pgdat(pgdat))
481 --node;
482
483 return pgdat;
484}
485
486static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
487{
488 pg_data_t *pgdat = (pg_data_t *)arg;
489
490 (*pos)++;
491 return next_online_pgdat(pgdat);
492}
493
494static void frag_stop(struct seq_file *m, void *arg)
495{
496}
497
Mel Gorman467c9962007-10-16 01:26:02 -0700498/* Walk all the zones in a node and print using a callback */
499static void walk_zones_in_node(struct seq_file *m, pg_data_t *pgdat,
500 void (*print)(struct seq_file *m, pg_data_t *, struct zone *))
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700501{
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700502 struct zone *zone;
503 struct zone *node_zones = pgdat->node_zones;
504 unsigned long flags;
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700505
506 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
507 if (!populated_zone(zone))
508 continue;
509
510 spin_lock_irqsave(&zone->lock, flags);
Mel Gorman467c9962007-10-16 01:26:02 -0700511 print(m, pgdat, zone);
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700512 spin_unlock_irqrestore(&zone->lock, flags);
Mel Gorman467c9962007-10-16 01:26:02 -0700513 }
514}
Mel Gormand7a57522010-05-24 14:32:25 -0700515#endif
Mel Gorman467c9962007-10-16 01:26:02 -0700516
Mel Gormand7a57522010-05-24 14:32:25 -0700517#ifdef CONFIG_PROC_FS
Mel Gorman467c9962007-10-16 01:26:02 -0700518static void frag_show_print(struct seq_file *m, pg_data_t *pgdat,
519 struct zone *zone)
520{
521 int order;
522
523 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
524 for (order = 0; order < MAX_ORDER; ++order)
525 seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
526 seq_putc(m, '\n');
527}
528
529/*
530 * This walks the free areas for each zone.
531 */
532static int frag_show(struct seq_file *m, void *arg)
533{
534 pg_data_t *pgdat = (pg_data_t *)arg;
535 walk_zones_in_node(m, pgdat, frag_show_print);
536 return 0;
537}
538
539static void pagetypeinfo_showfree_print(struct seq_file *m,
540 pg_data_t *pgdat, struct zone *zone)
541{
542 int order, mtype;
543
544 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++) {
545 seq_printf(m, "Node %4d, zone %8s, type %12s ",
546 pgdat->node_id,
547 zone->name,
548 migratetype_names[mtype]);
549 for (order = 0; order < MAX_ORDER; ++order) {
550 unsigned long freecount = 0;
551 struct free_area *area;
552 struct list_head *curr;
553
554 area = &(zone->free_area[order]);
555
556 list_for_each(curr, &area->free_list[mtype])
557 freecount++;
558 seq_printf(m, "%6lu ", freecount);
559 }
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700560 seq_putc(m, '\n');
561 }
Mel Gorman467c9962007-10-16 01:26:02 -0700562}
563
564/* Print out the free pages at each order for each migatetype */
565static int pagetypeinfo_showfree(struct seq_file *m, void *arg)
566{
567 int order;
568 pg_data_t *pgdat = (pg_data_t *)arg;
569
570 /* Print header */
571 seq_printf(m, "%-43s ", "Free pages count per migrate type at order");
572 for (order = 0; order < MAX_ORDER; ++order)
573 seq_printf(m, "%6d ", order);
574 seq_putc(m, '\n');
575
576 walk_zones_in_node(m, pgdat, pagetypeinfo_showfree_print);
577
578 return 0;
579}
580
581static void pagetypeinfo_showblockcount_print(struct seq_file *m,
582 pg_data_t *pgdat, struct zone *zone)
583{
584 int mtype;
585 unsigned long pfn;
586 unsigned long start_pfn = zone->zone_start_pfn;
587 unsigned long end_pfn = start_pfn + zone->spanned_pages;
588 unsigned long count[MIGRATE_TYPES] = { 0, };
589
590 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
591 struct page *page;
592
593 if (!pfn_valid(pfn))
594 continue;
595
596 page = pfn_to_page(pfn);
Mel Gormaneb335752009-05-13 17:34:48 +0100597
598 /* Watch for unexpected holes punched in the memmap */
599 if (!memmap_valid_within(pfn, page, zone))
Mel Gormane80d6a22008-08-14 11:10:14 +0100600 continue;
Mel Gormaneb335752009-05-13 17:34:48 +0100601
Mel Gorman467c9962007-10-16 01:26:02 -0700602 mtype = get_pageblock_migratetype(page);
603
Mel Gormane80d6a22008-08-14 11:10:14 +0100604 if (mtype < MIGRATE_TYPES)
605 count[mtype]++;
Mel Gorman467c9962007-10-16 01:26:02 -0700606 }
607
608 /* Print counts */
609 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
610 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
611 seq_printf(m, "%12lu ", count[mtype]);
612 seq_putc(m, '\n');
613}
614
615/* Print out the free pages at each order for each migratetype */
616static int pagetypeinfo_showblockcount(struct seq_file *m, void *arg)
617{
618 int mtype;
619 pg_data_t *pgdat = (pg_data_t *)arg;
620
621 seq_printf(m, "\n%-23s", "Number of blocks type ");
622 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
623 seq_printf(m, "%12s ", migratetype_names[mtype]);
624 seq_putc(m, '\n');
625 walk_zones_in_node(m, pgdat, pagetypeinfo_showblockcount_print);
626
627 return 0;
628}
629
630/*
631 * This prints out statistics in relation to grouping pages by mobility.
632 * It is expensive to collect so do not constantly read the file.
633 */
634static int pagetypeinfo_show(struct seq_file *m, void *arg)
635{
636 pg_data_t *pgdat = (pg_data_t *)arg;
637
KOSAKI Motohiro41b25a32008-04-30 00:52:13 -0700638 /* check memoryless node */
639 if (!node_state(pgdat->node_id, N_HIGH_MEMORY))
640 return 0;
641
Mel Gorman467c9962007-10-16 01:26:02 -0700642 seq_printf(m, "Page block order: %d\n", pageblock_order);
643 seq_printf(m, "Pages per block: %lu\n", pageblock_nr_pages);
644 seq_putc(m, '\n');
645 pagetypeinfo_showfree(m, pgdat);
646 pagetypeinfo_showblockcount(m, pgdat);
647
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700648 return 0;
649}
650
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +0400651static const struct seq_operations fragmentation_op = {
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700652 .start = frag_start,
653 .next = frag_next,
654 .stop = frag_stop,
655 .show = frag_show,
656};
657
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +0400658static int fragmentation_open(struct inode *inode, struct file *file)
659{
660 return seq_open(file, &fragmentation_op);
661}
662
663static const struct file_operations fragmentation_file_operations = {
664 .open = fragmentation_open,
665 .read = seq_read,
666 .llseek = seq_lseek,
667 .release = seq_release,
668};
669
Alexey Dobriyan74e2e8e2008-10-06 04:15:36 +0400670static const struct seq_operations pagetypeinfo_op = {
Mel Gorman467c9962007-10-16 01:26:02 -0700671 .start = frag_start,
672 .next = frag_next,
673 .stop = frag_stop,
674 .show = pagetypeinfo_show,
675};
676
Alexey Dobriyan74e2e8e2008-10-06 04:15:36 +0400677static int pagetypeinfo_open(struct inode *inode, struct file *file)
678{
679 return seq_open(file, &pagetypeinfo_op);
680}
681
682static const struct file_operations pagetypeinfo_file_ops = {
683 .open = pagetypeinfo_open,
684 .read = seq_read,
685 .llseek = seq_lseek,
686 .release = seq_release,
687};
688
Christoph Lameter4b51d662007-02-10 01:43:10 -0800689#ifdef CONFIG_ZONE_DMA
690#define TEXT_FOR_DMA(xx) xx "_dma",
691#else
692#define TEXT_FOR_DMA(xx)
693#endif
694
Christoph Lameter27bf71c2006-09-25 23:31:15 -0700695#ifdef CONFIG_ZONE_DMA32
696#define TEXT_FOR_DMA32(xx) xx "_dma32",
697#else
698#define TEXT_FOR_DMA32(xx)
699#endif
700
701#ifdef CONFIG_HIGHMEM
702#define TEXT_FOR_HIGHMEM(xx) xx "_high",
703#else
704#define TEXT_FOR_HIGHMEM(xx)
705#endif
706
Christoph Lameter4b51d662007-02-10 01:43:10 -0800707#define TEXTS_FOR_ZONES(xx) TEXT_FOR_DMA(xx) TEXT_FOR_DMA32(xx) xx "_normal", \
Mel Gorman2a1e2742007-07-17 04:03:12 -0700708 TEXT_FOR_HIGHMEM(xx) xx "_movable",
Christoph Lameter27bf71c2006-09-25 23:31:15 -0700709
Helge Deller15ad7cd2006-12-06 20:40:36 -0800710static const char * const vmstat_text[] = {
Christoph Lameter2244b952006-06-30 01:55:33 -0700711 /* Zoned VM counters */
Christoph Lameterd23ad422007-02-10 01:43:02 -0800712 "nr_free_pages",
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700713 "nr_inactive_anon",
714 "nr_active_anon",
715 "nr_inactive_file",
716 "nr_active_file",
Lee Schermerhorn7b854122008-10-18 20:26:40 -0700717 "nr_unevictable",
Nick Piggin5344b7e2008-10-18 20:26:51 -0700718 "nr_mlock",
Christoph Lameterf3dbd342006-06-30 01:55:36 -0700719 "nr_anon_pages",
Christoph Lameter65ba55f2006-06-30 01:55:34 -0700720 "nr_mapped",
Christoph Lameter347ce432006-06-30 01:55:35 -0700721 "nr_file_pages",
Christoph Lameter51ed4492007-02-10 01:43:02 -0800722 "nr_dirty",
723 "nr_writeback",
Christoph Lameter972d1a72006-09-25 23:31:51 -0700724 "nr_slab_reclaimable",
725 "nr_slab_unreclaimable",
Christoph Lameterdf849a12006-06-30 01:55:38 -0700726 "nr_page_table_pages",
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -0700727 "nr_kernel_stack",
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700728 "nr_unstable",
Christoph Lameterd2c5e302006-06-30 01:55:41 -0700729 "nr_bounce",
Andrew Mortone129b5c2006-09-27 01:50:00 -0700730 "nr_vmscan_write",
Miklos Szeredifc3ba692008-04-30 00:54:38 -0700731 "nr_writeback_temp",
KOSAKI Motohiroa7312862009-09-21 17:01:37 -0700732 "nr_isolated_anon",
733 "nr_isolated_file",
KOSAKI Motohiro4b021082009-09-21 17:01:33 -0700734 "nr_shmem",
Christoph Lameterca889e62006-06-30 01:55:44 -0700735#ifdef CONFIG_NUMA
736 "numa_hit",
737 "numa_miss",
738 "numa_foreign",
739 "numa_interleave",
740 "numa_local",
741 "numa_other",
742#endif
743
Christoph Lameterf8891e52006-06-30 01:55:45 -0700744#ifdef CONFIG_VM_EVENT_COUNTERS
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700745 "pgpgin",
746 "pgpgout",
747 "pswpin",
748 "pswpout",
749
Christoph Lameter27bf71c2006-09-25 23:31:15 -0700750 TEXTS_FOR_ZONES("pgalloc")
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700751
752 "pgfree",
753 "pgactivate",
754 "pgdeactivate",
755
756 "pgfault",
757 "pgmajfault",
758
Christoph Lameter27bf71c2006-09-25 23:31:15 -0700759 TEXTS_FOR_ZONES("pgrefill")
760 TEXTS_FOR_ZONES("pgsteal")
761 TEXTS_FOR_ZONES("pgscan_kswapd")
762 TEXTS_FOR_ZONES("pgscan_direct")
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700763
Mel Gorman24cf725182009-06-16 15:33:23 -0700764#ifdef CONFIG_NUMA
765 "zone_reclaim_failed",
766#endif
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700767 "pginodesteal",
768 "slabs_scanned",
769 "kswapd_steal",
770 "kswapd_inodesteal",
KOSAKI Motohirobb3ab592009-12-14 17:58:55 -0800771 "kswapd_low_wmark_hit_quickly",
772 "kswapd_high_wmark_hit_quickly",
773 "kswapd_skip_congestion_wait",
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700774 "pageoutrun",
775 "allocstall",
776
777 "pgrotated",
Mel Gorman748446b2010-05-24 14:32:27 -0700778
779#ifdef CONFIG_COMPACTION
780 "compact_blocks_moved",
781 "compact_pages_moved",
782 "compact_pagemigrate_failed",
Mel Gorman56de7262010-05-24 14:32:30 -0700783 "compact_stall",
784 "compact_fail",
785 "compact_success",
Mel Gorman748446b2010-05-24 14:32:27 -0700786#endif
787
Adam Litke3b116302008-04-28 02:13:06 -0700788#ifdef CONFIG_HUGETLB_PAGE
789 "htlb_buddy_alloc_success",
790 "htlb_buddy_alloc_fail",
791#endif
Lee Schermerhornbbfd28e2008-10-18 20:26:40 -0700792 "unevictable_pgs_culled",
793 "unevictable_pgs_scanned",
794 "unevictable_pgs_rescued",
Nick Piggin5344b7e2008-10-18 20:26:51 -0700795 "unevictable_pgs_mlocked",
796 "unevictable_pgs_munlocked",
797 "unevictable_pgs_cleared",
798 "unevictable_pgs_stranded",
Lee Schermerhorn985737c2008-10-18 20:26:53 -0700799 "unevictable_pgs_mlockfreed",
Lee Schermerhornbbfd28e2008-10-18 20:26:40 -0700800#endif
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700801};
802
Mel Gorman467c9962007-10-16 01:26:02 -0700803static void zoneinfo_show_print(struct seq_file *m, pg_data_t *pgdat,
804 struct zone *zone)
805{
806 int i;
807 seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
808 seq_printf(m,
809 "\n pages free %lu"
810 "\n min %lu"
811 "\n low %lu"
812 "\n high %lu"
Wu Fengguang08d9ae72009-06-16 15:32:30 -0700813 "\n scanned %lu"
Mel Gorman467c9962007-10-16 01:26:02 -0700814 "\n spanned %lu"
815 "\n present %lu",
816 zone_page_state(zone, NR_FREE_PAGES),
Mel Gorman41858962009-06-16 15:32:12 -0700817 min_wmark_pages(zone),
818 low_wmark_pages(zone),
819 high_wmark_pages(zone),
Mel Gorman467c9962007-10-16 01:26:02 -0700820 zone->pages_scanned,
Mel Gorman467c9962007-10-16 01:26:02 -0700821 zone->spanned_pages,
822 zone->present_pages);
823
824 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
825 seq_printf(m, "\n %-12s %lu", vmstat_text[i],
826 zone_page_state(zone, i));
827
828 seq_printf(m,
829 "\n protection: (%lu",
830 zone->lowmem_reserve[0]);
831 for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
832 seq_printf(m, ", %lu", zone->lowmem_reserve[i]);
833 seq_printf(m,
834 ")"
835 "\n pagesets");
836 for_each_online_cpu(i) {
837 struct per_cpu_pageset *pageset;
Mel Gorman467c9962007-10-16 01:26:02 -0700838
Christoph Lameter99dcc3e2010-01-05 15:34:51 +0900839 pageset = per_cpu_ptr(zone->pageset, i);
Christoph Lameter3dfa5722008-02-04 22:29:19 -0800840 seq_printf(m,
841 "\n cpu: %i"
842 "\n count: %i"
843 "\n high: %i"
844 "\n batch: %i",
845 i,
846 pageset->pcp.count,
847 pageset->pcp.high,
848 pageset->pcp.batch);
Mel Gorman467c9962007-10-16 01:26:02 -0700849#ifdef CONFIG_SMP
850 seq_printf(m, "\n vm stats threshold: %d",
851 pageset->stat_threshold);
852#endif
853 }
854 seq_printf(m,
855 "\n all_unreclaimable: %u"
Rik van Riel556adec2008-10-18 20:26:34 -0700856 "\n start_pfn: %lu"
857 "\n inactive_ratio: %u",
KOSAKI Motohiro93e4a892010-03-05 13:41:55 -0800858 zone->all_unreclaimable,
Rik van Riel556adec2008-10-18 20:26:34 -0700859 zone->zone_start_pfn,
860 zone->inactive_ratio);
Mel Gorman467c9962007-10-16 01:26:02 -0700861 seq_putc(m, '\n');
862}
863
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700864/*
865 * Output information about zones in @pgdat.
866 */
867static int zoneinfo_show(struct seq_file *m, void *arg)
868{
Mel Gorman467c9962007-10-16 01:26:02 -0700869 pg_data_t *pgdat = (pg_data_t *)arg;
870 walk_zones_in_node(m, pgdat, zoneinfo_show_print);
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700871 return 0;
872}
873
Alexey Dobriyan5c9fe622008-10-06 04:19:42 +0400874static const struct seq_operations zoneinfo_op = {
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700875 .start = frag_start, /* iterate over all zones. The same as in
876 * fragmentation. */
877 .next = frag_next,
878 .stop = frag_stop,
879 .show = zoneinfo_show,
880};
881
Alexey Dobriyan5c9fe622008-10-06 04:19:42 +0400882static int zoneinfo_open(struct inode *inode, struct file *file)
883{
884 return seq_open(file, &zoneinfo_op);
885}
886
887static const struct file_operations proc_zoneinfo_file_operations = {
888 .open = zoneinfo_open,
889 .read = seq_read,
890 .llseek = seq_lseek,
891 .release = seq_release,
892};
893
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700894static void *vmstat_start(struct seq_file *m, loff_t *pos)
895{
Christoph Lameter2244b952006-06-30 01:55:33 -0700896 unsigned long *v;
Christoph Lameterf8891e52006-06-30 01:55:45 -0700897#ifdef CONFIG_VM_EVENT_COUNTERS
898 unsigned long *e;
899#endif
Christoph Lameter2244b952006-06-30 01:55:33 -0700900 int i;
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700901
902 if (*pos >= ARRAY_SIZE(vmstat_text))
903 return NULL;
904
Christoph Lameterf8891e52006-06-30 01:55:45 -0700905#ifdef CONFIG_VM_EVENT_COUNTERS
Christoph Lameter2244b952006-06-30 01:55:33 -0700906 v = kmalloc(NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long)
Christoph Lameterf8891e52006-06-30 01:55:45 -0700907 + sizeof(struct vm_event_state), GFP_KERNEL);
908#else
909 v = kmalloc(NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long),
910 GFP_KERNEL);
911#endif
Christoph Lameter2244b952006-06-30 01:55:33 -0700912 m->private = v;
913 if (!v)
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700914 return ERR_PTR(-ENOMEM);
Christoph Lameter2244b952006-06-30 01:55:33 -0700915 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
916 v[i] = global_page_state(i);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700917#ifdef CONFIG_VM_EVENT_COUNTERS
918 e = v + NR_VM_ZONE_STAT_ITEMS;
919 all_vm_events(e);
920 e[PGPGIN] /= 2; /* sectors -> kbytes */
921 e[PGPGOUT] /= 2;
922#endif
Christoph Lameter2244b952006-06-30 01:55:33 -0700923 return v + *pos;
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700924}
925
926static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
927{
928 (*pos)++;
929 if (*pos >= ARRAY_SIZE(vmstat_text))
930 return NULL;
931 return (unsigned long *)m->private + *pos;
932}
933
934static int vmstat_show(struct seq_file *m, void *arg)
935{
936 unsigned long *l = arg;
937 unsigned long off = l - (unsigned long *)m->private;
938
939 seq_printf(m, "%s %lu\n", vmstat_text[off], *l);
940 return 0;
941}
942
943static void vmstat_stop(struct seq_file *m, void *arg)
944{
945 kfree(m->private);
946 m->private = NULL;
947}
948
Alexey Dobriyanb6aa44a2008-10-06 04:17:48 +0400949static const struct seq_operations vmstat_op = {
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700950 .start = vmstat_start,
951 .next = vmstat_next,
952 .stop = vmstat_stop,
953 .show = vmstat_show,
954};
955
Alexey Dobriyanb6aa44a2008-10-06 04:17:48 +0400956static int vmstat_open(struct inode *inode, struct file *file)
957{
958 return seq_open(file, &vmstat_op);
959}
960
961static const struct file_operations proc_vmstat_file_operations = {
962 .open = vmstat_open,
963 .read = seq_read,
964 .llseek = seq_lseek,
965 .release = seq_release,
966};
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700967#endif /* CONFIG_PROC_FS */
968
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700969#ifdef CONFIG_SMP
Christoph Lameterd1187ed2007-05-09 02:35:12 -0700970static DEFINE_PER_CPU(struct delayed_work, vmstat_work);
Christoph Lameter77461ab2007-05-09 02:35:13 -0700971int sysctl_stat_interval __read_mostly = HZ;
Christoph Lameterd1187ed2007-05-09 02:35:12 -0700972
973static void vmstat_update(struct work_struct *w)
974{
975 refresh_cpu_vm_stats(smp_processor_id());
Christoph Lameter77461ab2007-05-09 02:35:13 -0700976 schedule_delayed_work(&__get_cpu_var(vmstat_work),
Anton Blanchard98f4ebb2009-04-02 16:56:39 -0700977 round_jiffies_relative(sysctl_stat_interval));
Christoph Lameterd1187ed2007-05-09 02:35:12 -0700978}
979
Randy Dunlap42614fc2007-11-14 17:00:12 -0800980static void __cpuinit start_cpu_timer(int cpu)
Christoph Lameterd1187ed2007-05-09 02:35:12 -0700981{
Tejun Heo1871e522009-10-29 22:34:13 +0900982 struct delayed_work *work = &per_cpu(vmstat_work, cpu);
Christoph Lameterd1187ed2007-05-09 02:35:12 -0700983
Tejun Heo1871e522009-10-29 22:34:13 +0900984 INIT_DELAYED_WORK_DEFERRABLE(work, vmstat_update);
985 schedule_delayed_work_on(cpu, work, __round_jiffies_relative(HZ, cpu));
Christoph Lameterd1187ed2007-05-09 02:35:12 -0700986}
987
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700988/*
989 * Use the cpu notifier to insure that the thresholds are recalculated
990 * when necessary.
991 */
992static int __cpuinit vmstat_cpuup_callback(struct notifier_block *nfb,
993 unsigned long action,
994 void *hcpu)
995{
Christoph Lameterd1187ed2007-05-09 02:35:12 -0700996 long cpu = (long)hcpu;
997
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700998 switch (action) {
Christoph Lameterd1187ed2007-05-09 02:35:12 -0700999 case CPU_ONLINE:
1000 case CPU_ONLINE_FROZEN:
1001 start_cpu_timer(cpu);
Christoph Lameterad596922010-01-05 15:34:51 +09001002 node_set_state(cpu_to_node(cpu), N_CPU);
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001003 break;
1004 case CPU_DOWN_PREPARE:
1005 case CPU_DOWN_PREPARE_FROZEN:
1006 cancel_rearming_delayed_work(&per_cpu(vmstat_work, cpu));
1007 per_cpu(vmstat_work, cpu).work.func = NULL;
1008 break;
1009 case CPU_DOWN_FAILED:
1010 case CPU_DOWN_FAILED_FROZEN:
1011 start_cpu_timer(cpu);
1012 break;
Andy Whitcroftce421c72006-12-06 20:33:08 -08001013 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07001014 case CPU_DEAD_FROZEN:
Andy Whitcroftce421c72006-12-06 20:33:08 -08001015 refresh_zone_stat_thresholds();
1016 break;
1017 default:
1018 break;
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001019 }
1020 return NOTIFY_OK;
1021}
1022
1023static struct notifier_block __cpuinitdata vmstat_notifier =
1024 { &vmstat_cpuup_callback, NULL, 0 };
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +04001025#endif
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001026
Adrian Bunke2fc88d2007-10-16 01:26:27 -07001027static int __init setup_vmstat(void)
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001028{
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +04001029#ifdef CONFIG_SMP
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001030 int cpu;
1031
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001032 refresh_zone_stat_thresholds();
1033 register_cpu_notifier(&vmstat_notifier);
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001034
1035 for_each_online_cpu(cpu)
1036 start_cpu_timer(cpu);
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +04001037#endif
1038#ifdef CONFIG_PROC_FS
1039 proc_create("buddyinfo", S_IRUGO, NULL, &fragmentation_file_operations);
Alexey Dobriyan74e2e8e2008-10-06 04:15:36 +04001040 proc_create("pagetypeinfo", S_IRUGO, NULL, &pagetypeinfo_file_ops);
Alexey Dobriyanb6aa44a2008-10-06 04:17:48 +04001041 proc_create("vmstat", S_IRUGO, NULL, &proc_vmstat_file_operations);
Alexey Dobriyan5c9fe622008-10-06 04:19:42 +04001042 proc_create("zoneinfo", S_IRUGO, NULL, &proc_zoneinfo_file_operations);
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +04001043#endif
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001044 return 0;
1045}
1046module_init(setup_vmstat)
Mel Gormand7a57522010-05-24 14:32:25 -07001047
1048#if defined(CONFIG_DEBUG_FS) && defined(CONFIG_COMPACTION)
1049#include <linux/debugfs.h>
1050
1051static struct dentry *extfrag_debug_root;
1052
1053/*
1054 * Return an index indicating how much of the available free memory is
1055 * unusable for an allocation of the requested size.
1056 */
1057static int unusable_free_index(unsigned int order,
1058 struct contig_page_info *info)
1059{
1060 /* No free memory is interpreted as all free memory is unusable */
1061 if (info->free_pages == 0)
1062 return 1000;
1063
1064 /*
1065 * Index should be a value between 0 and 1. Return a value to 3
1066 * decimal places.
1067 *
1068 * 0 => no fragmentation
1069 * 1 => high fragmentation
1070 */
1071 return div_u64((info->free_pages - (info->free_blocks_suitable << order)) * 1000ULL, info->free_pages);
1072
1073}
1074
1075static void unusable_show_print(struct seq_file *m,
1076 pg_data_t *pgdat, struct zone *zone)
1077{
1078 unsigned int order;
1079 int index;
1080 struct contig_page_info info;
1081
1082 seq_printf(m, "Node %d, zone %8s ",
1083 pgdat->node_id,
1084 zone->name);
1085 for (order = 0; order < MAX_ORDER; ++order) {
1086 fill_contig_page_info(zone, order, &info);
1087 index = unusable_free_index(order, &info);
1088 seq_printf(m, "%d.%03d ", index / 1000, index % 1000);
1089 }
1090
1091 seq_putc(m, '\n');
1092}
1093
1094/*
1095 * Display unusable free space index
1096 *
1097 * The unusable free space index measures how much of the available free
1098 * memory cannot be used to satisfy an allocation of a given size and is a
1099 * value between 0 and 1. The higher the value, the more of free memory is
1100 * unusable and by implication, the worse the external fragmentation is. This
1101 * can be expressed as a percentage by multiplying by 100.
1102 */
1103static int unusable_show(struct seq_file *m, void *arg)
1104{
1105 pg_data_t *pgdat = (pg_data_t *)arg;
1106
1107 /* check memoryless node */
1108 if (!node_state(pgdat->node_id, N_HIGH_MEMORY))
1109 return 0;
1110
1111 walk_zones_in_node(m, pgdat, unusable_show_print);
1112
1113 return 0;
1114}
1115
1116static const struct seq_operations unusable_op = {
1117 .start = frag_start,
1118 .next = frag_next,
1119 .stop = frag_stop,
1120 .show = unusable_show,
1121};
1122
1123static int unusable_open(struct inode *inode, struct file *file)
1124{
1125 return seq_open(file, &unusable_op);
1126}
1127
1128static const struct file_operations unusable_file_ops = {
1129 .open = unusable_open,
1130 .read = seq_read,
1131 .llseek = seq_lseek,
1132 .release = seq_release,
1133};
1134
Mel Gormanf1a5ab12010-05-24 14:32:26 -07001135static void extfrag_show_print(struct seq_file *m,
1136 pg_data_t *pgdat, struct zone *zone)
1137{
1138 unsigned int order;
1139 int index;
1140
1141 /* Alloc on stack as interrupts are disabled for zone walk */
1142 struct contig_page_info info;
1143
1144 seq_printf(m, "Node %d, zone %8s ",
1145 pgdat->node_id,
1146 zone->name);
1147 for (order = 0; order < MAX_ORDER; ++order) {
1148 fill_contig_page_info(zone, order, &info);
Mel Gorman56de7262010-05-24 14:32:30 -07001149 index = __fragmentation_index(order, &info);
Mel Gormanf1a5ab12010-05-24 14:32:26 -07001150 seq_printf(m, "%d.%03d ", index / 1000, index % 1000);
1151 }
1152
1153 seq_putc(m, '\n');
1154}
1155
1156/*
1157 * Display fragmentation index for orders that allocations would fail for
1158 */
1159static int extfrag_show(struct seq_file *m, void *arg)
1160{
1161 pg_data_t *pgdat = (pg_data_t *)arg;
1162
1163 walk_zones_in_node(m, pgdat, extfrag_show_print);
1164
1165 return 0;
1166}
1167
1168static const struct seq_operations extfrag_op = {
1169 .start = frag_start,
1170 .next = frag_next,
1171 .stop = frag_stop,
1172 .show = extfrag_show,
1173};
1174
1175static int extfrag_open(struct inode *inode, struct file *file)
1176{
1177 return seq_open(file, &extfrag_op);
1178}
1179
1180static const struct file_operations extfrag_file_ops = {
1181 .open = extfrag_open,
1182 .read = seq_read,
1183 .llseek = seq_lseek,
1184 .release = seq_release,
1185};
1186
Mel Gormand7a57522010-05-24 14:32:25 -07001187static int __init extfrag_debug_init(void)
1188{
1189 extfrag_debug_root = debugfs_create_dir("extfrag", NULL);
1190 if (!extfrag_debug_root)
1191 return -ENOMEM;
1192
1193 if (!debugfs_create_file("unusable_index", 0444,
1194 extfrag_debug_root, NULL, &unusable_file_ops))
1195 return -ENOMEM;
1196
Mel Gormanf1a5ab12010-05-24 14:32:26 -07001197 if (!debugfs_create_file("extfrag_index", 0444,
1198 extfrag_debug_root, NULL, &extfrag_file_ops))
1199 return -ENOMEM;
1200
Mel Gormand7a57522010-05-24 14:32:25 -07001201 return 0;
1202}
1203
1204module_init(extfrag_debug_init);
1205#endif