blob: 42eac4d33216b81c307a87016e821051bc86146e [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>
Michael Rubin79da8262010-10-26 14:21:36 -070020#include <linux/writeback.h>
Namhyung Kim36deb0b2010-10-26 14:22:04 -070021#include <linux/compaction.h>
Christoph Lameterf6ac2352006-06-30 01:55:32 -070022
Christoph Lameterf8891e52006-06-30 01:55:45 -070023#ifdef CONFIG_VM_EVENT_COUNTERS
24DEFINE_PER_CPU(struct vm_event_state, vm_event_states) = {{0}};
25EXPORT_PER_CPU_SYMBOL(vm_event_states);
26
Minchan Kim31f961a2010-08-09 17:18:59 -070027static void sum_vm_events(unsigned long *ret)
Christoph Lameterf8891e52006-06-30 01:55:45 -070028{
Christoph Lameter9eccf2a2008-02-04 22:29:22 -080029 int cpu;
Christoph Lameterf8891e52006-06-30 01:55:45 -070030 int i;
31
32 memset(ret, 0, NR_VM_EVENT_ITEMS * sizeof(unsigned long));
33
Minchan Kim31f961a2010-08-09 17:18:59 -070034 for_each_online_cpu(cpu) {
Christoph Lameterf8891e52006-06-30 01:55:45 -070035 struct vm_event_state *this = &per_cpu(vm_event_states, cpu);
36
Christoph Lameterf8891e52006-06-30 01:55:45 -070037 for (i = 0; i < NR_VM_EVENT_ITEMS; i++)
38 ret[i] += this->event[i];
39 }
40}
41
42/*
43 * Accumulate the vm event counters across all CPUs.
44 * The result is unavoidably approximate - it can change
45 * during and after execution of this function.
46*/
47void all_vm_events(unsigned long *ret)
48{
KOSAKI Motohirob5be1132008-05-12 14:02:06 -070049 get_online_cpus();
Minchan Kim31f961a2010-08-09 17:18:59 -070050 sum_vm_events(ret);
KOSAKI Motohirob5be1132008-05-12 14:02:06 -070051 put_online_cpus();
Christoph Lameterf8891e52006-06-30 01:55:45 -070052}
Heiko Carstens32dd66f2006-07-10 04:44:31 -070053EXPORT_SYMBOL_GPL(all_vm_events);
Christoph Lameterf8891e52006-06-30 01:55:45 -070054
55#ifdef CONFIG_HOTPLUG
56/*
57 * Fold the foreign cpu events into our own.
58 *
59 * This is adding to the events on one processor
60 * but keeps the global counts constant.
61 */
62void vm_events_fold_cpu(int cpu)
63{
64 struct vm_event_state *fold_state = &per_cpu(vm_event_states, cpu);
65 int i;
66
67 for (i = 0; i < NR_VM_EVENT_ITEMS; i++) {
68 count_vm_events(i, fold_state->event[i]);
69 fold_state->event[i] = 0;
70 }
71}
72#endif /* CONFIG_HOTPLUG */
73
74#endif /* CONFIG_VM_EVENT_COUNTERS */
75
Christoph Lameter2244b952006-06-30 01:55:33 -070076/*
77 * Manage combined zone based / global counters
78 *
79 * vm_stat contains the global counters
80 */
81atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
82EXPORT_SYMBOL(vm_stat);
83
84#ifdef CONFIG_SMP
85
Christoph Lameterdf9ecab2006-08-31 21:27:35 -070086static int calculate_threshold(struct zone *zone)
87{
88 int threshold;
89 int mem; /* memory in 128 MB units */
90
91 /*
92 * The threshold scales with the number of processors and the amount
93 * of memory per zone. More memory means that we can defer updates for
94 * longer, more processors could lead to more contention.
95 * fls() is used to have a cheap way of logarithmic scaling.
96 *
97 * Some sample thresholds:
98 *
99 * Threshold Processors (fls) Zonesize fls(mem+1)
100 * ------------------------------------------------------------------
101 * 8 1 1 0.9-1 GB 4
102 * 16 2 2 0.9-1 GB 4
103 * 20 2 2 1-2 GB 5
104 * 24 2 2 2-4 GB 6
105 * 28 2 2 4-8 GB 7
106 * 32 2 2 8-16 GB 8
107 * 4 2 2 <128M 1
108 * 30 4 3 2-4 GB 5
109 * 48 4 3 8-16 GB 8
110 * 32 8 4 1-2 GB 4
111 * 32 8 4 0.9-1GB 4
112 * 10 16 5 <128M 1
113 * 40 16 5 900M 4
114 * 70 64 7 2-4 GB 5
115 * 84 64 7 4-8 GB 6
116 * 108 512 9 4-8 GB 6
117 * 125 1024 10 8-16 GB 8
118 * 125 1024 10 16-32 GB 9
119 */
120
121 mem = zone->present_pages >> (27 - PAGE_SHIFT);
122
123 threshold = 2 * fls(num_online_cpus()) * (1 + fls(mem));
124
125 /*
126 * Maximum threshold is 125
127 */
128 threshold = min(125, threshold);
129
130 return threshold;
131}
Christoph Lameter2244b952006-06-30 01:55:33 -0700132
133/*
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700134 * Refresh the thresholds for each zone.
Christoph Lameter2244b952006-06-30 01:55:33 -0700135 */
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700136static void refresh_zone_stat_thresholds(void)
Christoph Lameter2244b952006-06-30 01:55:33 -0700137{
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700138 struct zone *zone;
139 int cpu;
140 int threshold;
141
KOSAKI Motohiroee99c712009-03-31 15:19:31 -0700142 for_each_populated_zone(zone) {
Christoph Lameteraa454842010-09-09 16:38:17 -0700143 unsigned long max_drift, tolerate_drift;
144
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700145 threshold = calculate_threshold(zone);
146
147 for_each_online_cpu(cpu)
Christoph Lameter99dcc3e2010-01-05 15:34:51 +0900148 per_cpu_ptr(zone->pageset, cpu)->stat_threshold
149 = threshold;
Christoph Lameteraa454842010-09-09 16:38:17 -0700150
151 /*
152 * Only set percpu_drift_mark if there is a danger that
153 * NR_FREE_PAGES reports the low watermark is ok when in fact
154 * the min watermark could be breached by an allocation
155 */
156 tolerate_drift = low_wmark_pages(zone) - min_wmark_pages(zone);
157 max_drift = num_online_cpus() * threshold;
158 if (max_drift > tolerate_drift)
159 zone->percpu_drift_mark = high_wmark_pages(zone) +
160 max_drift;
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700161 }
Christoph Lameter2244b952006-06-30 01:55:33 -0700162}
163
164/*
165 * For use when we know that interrupts are disabled.
166 */
167void __mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
168 int delta)
169{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +0900170 struct per_cpu_pageset *pcp = this_cpu_ptr(zone->pageset);
171
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700172 s8 *p = pcp->vm_stat_diff + item;
Christoph Lameter2244b952006-06-30 01:55:33 -0700173 long x;
174
Christoph Lameter2244b952006-06-30 01:55:33 -0700175 x = delta + *p;
176
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700177 if (unlikely(x > pcp->stat_threshold || x < -pcp->stat_threshold)) {
Christoph Lameter2244b952006-06-30 01:55:33 -0700178 zone_page_state_add(x, zone, item);
179 x = 0;
180 }
Christoph Lameter2244b952006-06-30 01:55:33 -0700181 *p = x;
182}
183EXPORT_SYMBOL(__mod_zone_page_state);
184
185/*
186 * For an unknown interrupt state
187 */
188void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
189 int delta)
190{
191 unsigned long flags;
192
193 local_irq_save(flags);
194 __mod_zone_page_state(zone, item, delta);
195 local_irq_restore(flags);
196}
197EXPORT_SYMBOL(mod_zone_page_state);
198
199/*
200 * Optimized increment and decrement functions.
201 *
202 * These are only for a single page and therefore can take a struct page *
203 * argument instead of struct zone *. This allows the inclusion of the code
204 * generated for page_zone(page) into the optimized functions.
205 *
206 * No overflow check is necessary and therefore the differential can be
207 * incremented or decremented in place which may allow the compilers to
208 * generate better code.
Christoph Lameter2244b952006-06-30 01:55:33 -0700209 * The increment or decrement is known and therefore one boundary check can
210 * be omitted.
211 *
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700212 * NOTE: These functions are very performance sensitive. Change only
213 * with care.
214 *
Christoph Lameter2244b952006-06-30 01:55:33 -0700215 * Some processors have inc/dec instructions that are atomic vs an interrupt.
216 * However, the code must first determine the differential location in a zone
217 * based on the processor number and then inc/dec the counter. There is no
218 * guarantee without disabling preemption that the processor will not change
219 * in between and therefore the atomicity vs. interrupt cannot be exploited
220 * in a useful way here.
221 */
Christoph Lameterc8785382007-02-10 01:43:01 -0800222void __inc_zone_state(struct zone *zone, enum zone_stat_item item)
Christoph Lameter2244b952006-06-30 01:55:33 -0700223{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +0900224 struct per_cpu_pageset *pcp = this_cpu_ptr(zone->pageset);
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700225 s8 *p = pcp->vm_stat_diff + item;
Christoph Lameter2244b952006-06-30 01:55:33 -0700226
227 (*p)++;
228
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700229 if (unlikely(*p > pcp->stat_threshold)) {
230 int overstep = pcp->stat_threshold / 2;
231
232 zone_page_state_add(*p + overstep, zone, item);
233 *p = -overstep;
Christoph Lameter2244b952006-06-30 01:55:33 -0700234 }
235}
Christoph Lameterca889e62006-06-30 01:55:44 -0700236
237void __inc_zone_page_state(struct page *page, enum zone_stat_item item)
238{
239 __inc_zone_state(page_zone(page), item);
240}
Christoph Lameter2244b952006-06-30 01:55:33 -0700241EXPORT_SYMBOL(__inc_zone_page_state);
242
Christoph Lameterc8785382007-02-10 01:43:01 -0800243void __dec_zone_state(struct zone *zone, enum zone_stat_item item)
Christoph Lameter2244b952006-06-30 01:55:33 -0700244{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +0900245 struct per_cpu_pageset *pcp = this_cpu_ptr(zone->pageset);
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700246 s8 *p = pcp->vm_stat_diff + item;
Christoph Lameter2244b952006-06-30 01:55:33 -0700247
248 (*p)--;
249
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700250 if (unlikely(*p < - pcp->stat_threshold)) {
251 int overstep = pcp->stat_threshold / 2;
252
253 zone_page_state_add(*p - overstep, zone, item);
254 *p = overstep;
Christoph Lameter2244b952006-06-30 01:55:33 -0700255 }
256}
Christoph Lameterc8785382007-02-10 01:43:01 -0800257
258void __dec_zone_page_state(struct page *page, enum zone_stat_item item)
259{
260 __dec_zone_state(page_zone(page), item);
261}
Christoph Lameter2244b952006-06-30 01:55:33 -0700262EXPORT_SYMBOL(__dec_zone_page_state);
263
Christoph Lameterca889e62006-06-30 01:55:44 -0700264void inc_zone_state(struct zone *zone, enum zone_stat_item item)
265{
266 unsigned long flags;
267
268 local_irq_save(flags);
269 __inc_zone_state(zone, item);
270 local_irq_restore(flags);
271}
272
Christoph Lameter2244b952006-06-30 01:55:33 -0700273void inc_zone_page_state(struct page *page, enum zone_stat_item item)
274{
275 unsigned long flags;
276 struct zone *zone;
Christoph Lameter2244b952006-06-30 01:55:33 -0700277
278 zone = page_zone(page);
279 local_irq_save(flags);
Christoph Lameterca889e62006-06-30 01:55:44 -0700280 __inc_zone_state(zone, item);
Christoph Lameter2244b952006-06-30 01:55:33 -0700281 local_irq_restore(flags);
282}
283EXPORT_SYMBOL(inc_zone_page_state);
284
285void dec_zone_page_state(struct page *page, enum zone_stat_item item)
286{
287 unsigned long flags;
Christoph Lameter2244b952006-06-30 01:55:33 -0700288
Christoph Lameter2244b952006-06-30 01:55:33 -0700289 local_irq_save(flags);
Christoph Lametera302eb42006-08-31 21:27:34 -0700290 __dec_zone_page_state(page, item);
Christoph Lameter2244b952006-06-30 01:55:33 -0700291 local_irq_restore(flags);
292}
293EXPORT_SYMBOL(dec_zone_page_state);
294
295/*
296 * Update the zone counters for one cpu.
Christoph Lameter4037d452007-05-09 02:35:14 -0700297 *
Christoph Lametera7f75e22008-02-04 22:29:16 -0800298 * The cpu specified must be either the current cpu or a processor that
299 * is not online. If it is the current cpu then the execution thread must
300 * be pinned to the current cpu.
301 *
Christoph Lameter4037d452007-05-09 02:35:14 -0700302 * Note that refresh_cpu_vm_stats strives to only access
303 * node local memory. The per cpu pagesets on remote zones are placed
304 * in the memory local to the processor using that pageset. So the
305 * loop over all zones will access a series of cachelines local to
306 * the processor.
307 *
308 * The call to zone_page_state_add updates the cachelines with the
309 * statistics in the remote zone struct as well as the global cachelines
310 * with the global counters. These could cause remote node cache line
311 * bouncing and will have to be only done when necessary.
Christoph Lameter2244b952006-06-30 01:55:33 -0700312 */
313void refresh_cpu_vm_stats(int cpu)
314{
315 struct zone *zone;
316 int i;
Christoph Lametera7f75e22008-02-04 22:29:16 -0800317 int global_diff[NR_VM_ZONE_STAT_ITEMS] = { 0, };
Christoph Lameter2244b952006-06-30 01:55:33 -0700318
KOSAKI Motohiroee99c712009-03-31 15:19:31 -0700319 for_each_populated_zone(zone) {
Christoph Lameter4037d452007-05-09 02:35:14 -0700320 struct per_cpu_pageset *p;
Christoph Lameter2244b952006-06-30 01:55:33 -0700321
Christoph Lameter99dcc3e2010-01-05 15:34:51 +0900322 p = per_cpu_ptr(zone->pageset, cpu);
Christoph Lameter2244b952006-06-30 01:55:33 -0700323
324 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
Christoph Lameter4037d452007-05-09 02:35:14 -0700325 if (p->vm_stat_diff[i]) {
Christoph Lametera7f75e22008-02-04 22:29:16 -0800326 unsigned long flags;
327 int v;
328
Christoph Lameter2244b952006-06-30 01:55:33 -0700329 local_irq_save(flags);
Christoph Lametera7f75e22008-02-04 22:29:16 -0800330 v = p->vm_stat_diff[i];
Christoph Lameter4037d452007-05-09 02:35:14 -0700331 p->vm_stat_diff[i] = 0;
Christoph Lametera7f75e22008-02-04 22:29:16 -0800332 local_irq_restore(flags);
333 atomic_long_add(v, &zone->vm_stat[i]);
334 global_diff[i] += v;
Christoph Lameter4037d452007-05-09 02:35:14 -0700335#ifdef CONFIG_NUMA
336 /* 3 seconds idle till flush */
337 p->expire = 3;
338#endif
Christoph Lameter2244b952006-06-30 01:55:33 -0700339 }
Dimitri Sivanich468fd622008-04-28 02:13:37 -0700340 cond_resched();
Christoph Lameter4037d452007-05-09 02:35:14 -0700341#ifdef CONFIG_NUMA
342 /*
343 * Deal with draining the remote pageset of this
344 * processor
345 *
346 * Check if there are pages remaining in this pageset
347 * if not then there is nothing to expire.
348 */
Christoph Lameter3dfa5722008-02-04 22:29:19 -0800349 if (!p->expire || !p->pcp.count)
Christoph Lameter4037d452007-05-09 02:35:14 -0700350 continue;
351
352 /*
353 * We never drain zones local to this processor.
354 */
355 if (zone_to_nid(zone) == numa_node_id()) {
356 p->expire = 0;
357 continue;
358 }
359
360 p->expire--;
361 if (p->expire)
362 continue;
363
Christoph Lameter3dfa5722008-02-04 22:29:19 -0800364 if (p->pcp.count)
365 drain_zone_pages(zone, &p->pcp);
Christoph Lameter4037d452007-05-09 02:35:14 -0700366#endif
Christoph Lameter2244b952006-06-30 01:55:33 -0700367 }
Christoph Lametera7f75e22008-02-04 22:29:16 -0800368
369 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
370 if (global_diff[i])
371 atomic_long_add(global_diff[i], &vm_stat[i]);
Christoph Lameter2244b952006-06-30 01:55:33 -0700372}
373
Christoph Lameter2244b952006-06-30 01:55:33 -0700374#endif
375
Christoph Lameterca889e62006-06-30 01:55:44 -0700376#ifdef CONFIG_NUMA
377/*
378 * zonelist = the list of zones passed to the allocator
379 * z = the zone from which the allocation occurred.
380 *
381 * Must be called with interrupts disabled.
382 */
Mel Gorman18ea7e72008-04-28 02:12:14 -0700383void zone_statistics(struct zone *preferred_zone, struct zone *z)
Christoph Lameterca889e62006-06-30 01:55:44 -0700384{
Mel Gorman18ea7e72008-04-28 02:12:14 -0700385 if (z->zone_pgdat == preferred_zone->zone_pgdat) {
Christoph Lameterca889e62006-06-30 01:55:44 -0700386 __inc_zone_state(z, NUMA_HIT);
387 } else {
388 __inc_zone_state(z, NUMA_MISS);
Mel Gorman18ea7e72008-04-28 02:12:14 -0700389 __inc_zone_state(preferred_zone, NUMA_FOREIGN);
Christoph Lameterca889e62006-06-30 01:55:44 -0700390 }
Christoph Lameter5d292342006-09-27 01:50:10 -0700391 if (z->node == numa_node_id())
Christoph Lameterca889e62006-06-30 01:55:44 -0700392 __inc_zone_state(z, NUMA_LOCAL);
393 else
394 __inc_zone_state(z, NUMA_OTHER);
395}
396#endif
397
Mel Gormand7a57522010-05-24 14:32:25 -0700398#ifdef CONFIG_COMPACTION
Namhyung Kim36deb0b2010-10-26 14:22:04 -0700399
Mel Gormand7a57522010-05-24 14:32:25 -0700400struct contig_page_info {
401 unsigned long free_pages;
402 unsigned long free_blocks_total;
403 unsigned long free_blocks_suitable;
404};
405
406/*
407 * Calculate the number of free pages in a zone, how many contiguous
408 * pages are free and how many are large enough to satisfy an allocation of
409 * the target size. Note that this function makes no attempt to estimate
410 * how many suitable free blocks there *might* be if MOVABLE pages were
411 * migrated. Calculating that is possible, but expensive and can be
412 * figured out from userspace
413 */
414static void fill_contig_page_info(struct zone *zone,
415 unsigned int suitable_order,
416 struct contig_page_info *info)
417{
418 unsigned int order;
419
420 info->free_pages = 0;
421 info->free_blocks_total = 0;
422 info->free_blocks_suitable = 0;
423
424 for (order = 0; order < MAX_ORDER; order++) {
425 unsigned long blocks;
426
427 /* Count number of free blocks */
428 blocks = zone->free_area[order].nr_free;
429 info->free_blocks_total += blocks;
430
431 /* Count free base pages */
432 info->free_pages += blocks << order;
433
434 /* Count the suitable free blocks */
435 if (order >= suitable_order)
436 info->free_blocks_suitable += blocks <<
437 (order - suitable_order);
438 }
439}
Mel Gormanf1a5ab12010-05-24 14:32:26 -0700440
441/*
442 * A fragmentation index only makes sense if an allocation of a requested
443 * size would fail. If that is true, the fragmentation index indicates
444 * whether external fragmentation or a lack of memory was the problem.
445 * The value can be used to determine if page reclaim or compaction
446 * should be used
447 */
Mel Gorman56de7262010-05-24 14:32:30 -0700448static int __fragmentation_index(unsigned int order, struct contig_page_info *info)
Mel Gormanf1a5ab12010-05-24 14:32:26 -0700449{
450 unsigned long requested = 1UL << order;
451
452 if (!info->free_blocks_total)
453 return 0;
454
455 /* Fragmentation index only makes sense when a request would fail */
456 if (info->free_blocks_suitable)
457 return -1000;
458
459 /*
460 * Index is between 0 and 1 so return within 3 decimal places
461 *
462 * 0 => allocation would fail due to lack of memory
463 * 1 => allocation would fail due to fragmentation
464 */
465 return 1000 - div_u64( (1000+(div_u64(info->free_pages * 1000ULL, requested))), info->free_blocks_total);
466}
Mel Gorman56de7262010-05-24 14:32:30 -0700467
468/* Same as __fragmentation index but allocs contig_page_info on stack */
469int fragmentation_index(struct zone *zone, unsigned int order)
470{
471 struct contig_page_info info;
472
473 fill_contig_page_info(zone, order, &info);
474 return __fragmentation_index(order, &info);
475}
Mel Gormand7a57522010-05-24 14:32:25 -0700476#endif
477
478#if defined(CONFIG_PROC_FS) || defined(CONFIG_COMPACTION)
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +0400479#include <linux/proc_fs.h>
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700480#include <linux/seq_file.h>
481
Mel Gorman467c9962007-10-16 01:26:02 -0700482static char * const migratetype_names[MIGRATE_TYPES] = {
483 "Unmovable",
484 "Reclaimable",
485 "Movable",
486 "Reserve",
KOSAKI Motohiro91446b02008-04-15 14:34:42 -0700487 "Isolate",
Mel Gorman467c9962007-10-16 01:26:02 -0700488};
489
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700490static void *frag_start(struct seq_file *m, loff_t *pos)
491{
492 pg_data_t *pgdat;
493 loff_t node = *pos;
494 for (pgdat = first_online_pgdat();
495 pgdat && node;
496 pgdat = next_online_pgdat(pgdat))
497 --node;
498
499 return pgdat;
500}
501
502static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
503{
504 pg_data_t *pgdat = (pg_data_t *)arg;
505
506 (*pos)++;
507 return next_online_pgdat(pgdat);
508}
509
510static void frag_stop(struct seq_file *m, void *arg)
511{
512}
513
Mel Gorman467c9962007-10-16 01:26:02 -0700514/* Walk all the zones in a node and print using a callback */
515static void walk_zones_in_node(struct seq_file *m, pg_data_t *pgdat,
516 void (*print)(struct seq_file *m, pg_data_t *, struct zone *))
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700517{
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700518 struct zone *zone;
519 struct zone *node_zones = pgdat->node_zones;
520 unsigned long flags;
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700521
522 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
523 if (!populated_zone(zone))
524 continue;
525
526 spin_lock_irqsave(&zone->lock, flags);
Mel Gorman467c9962007-10-16 01:26:02 -0700527 print(m, pgdat, zone);
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700528 spin_unlock_irqrestore(&zone->lock, flags);
Mel Gorman467c9962007-10-16 01:26:02 -0700529 }
530}
Mel Gormand7a57522010-05-24 14:32:25 -0700531#endif
Mel Gorman467c9962007-10-16 01:26:02 -0700532
Mel Gormand7a57522010-05-24 14:32:25 -0700533#ifdef CONFIG_PROC_FS
Mel Gorman467c9962007-10-16 01:26:02 -0700534static void frag_show_print(struct seq_file *m, pg_data_t *pgdat,
535 struct zone *zone)
536{
537 int order;
538
539 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
540 for (order = 0; order < MAX_ORDER; ++order)
541 seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
542 seq_putc(m, '\n');
543}
544
545/*
546 * This walks the free areas for each zone.
547 */
548static int frag_show(struct seq_file *m, void *arg)
549{
550 pg_data_t *pgdat = (pg_data_t *)arg;
551 walk_zones_in_node(m, pgdat, frag_show_print);
552 return 0;
553}
554
555static void pagetypeinfo_showfree_print(struct seq_file *m,
556 pg_data_t *pgdat, struct zone *zone)
557{
558 int order, mtype;
559
560 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++) {
561 seq_printf(m, "Node %4d, zone %8s, type %12s ",
562 pgdat->node_id,
563 zone->name,
564 migratetype_names[mtype]);
565 for (order = 0; order < MAX_ORDER; ++order) {
566 unsigned long freecount = 0;
567 struct free_area *area;
568 struct list_head *curr;
569
570 area = &(zone->free_area[order]);
571
572 list_for_each(curr, &area->free_list[mtype])
573 freecount++;
574 seq_printf(m, "%6lu ", freecount);
575 }
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700576 seq_putc(m, '\n');
577 }
Mel Gorman467c9962007-10-16 01:26:02 -0700578}
579
580/* Print out the free pages at each order for each migatetype */
581static int pagetypeinfo_showfree(struct seq_file *m, void *arg)
582{
583 int order;
584 pg_data_t *pgdat = (pg_data_t *)arg;
585
586 /* Print header */
587 seq_printf(m, "%-43s ", "Free pages count per migrate type at order");
588 for (order = 0; order < MAX_ORDER; ++order)
589 seq_printf(m, "%6d ", order);
590 seq_putc(m, '\n');
591
592 walk_zones_in_node(m, pgdat, pagetypeinfo_showfree_print);
593
594 return 0;
595}
596
597static void pagetypeinfo_showblockcount_print(struct seq_file *m,
598 pg_data_t *pgdat, struct zone *zone)
599{
600 int mtype;
601 unsigned long pfn;
602 unsigned long start_pfn = zone->zone_start_pfn;
603 unsigned long end_pfn = start_pfn + zone->spanned_pages;
604 unsigned long count[MIGRATE_TYPES] = { 0, };
605
606 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
607 struct page *page;
608
609 if (!pfn_valid(pfn))
610 continue;
611
612 page = pfn_to_page(pfn);
Mel Gormaneb335752009-05-13 17:34:48 +0100613
614 /* Watch for unexpected holes punched in the memmap */
615 if (!memmap_valid_within(pfn, page, zone))
Mel Gormane80d6a22008-08-14 11:10:14 +0100616 continue;
Mel Gormaneb335752009-05-13 17:34:48 +0100617
Mel Gorman467c9962007-10-16 01:26:02 -0700618 mtype = get_pageblock_migratetype(page);
619
Mel Gormane80d6a22008-08-14 11:10:14 +0100620 if (mtype < MIGRATE_TYPES)
621 count[mtype]++;
Mel Gorman467c9962007-10-16 01:26:02 -0700622 }
623
624 /* Print counts */
625 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
626 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
627 seq_printf(m, "%12lu ", count[mtype]);
628 seq_putc(m, '\n');
629}
630
631/* Print out the free pages at each order for each migratetype */
632static int pagetypeinfo_showblockcount(struct seq_file *m, void *arg)
633{
634 int mtype;
635 pg_data_t *pgdat = (pg_data_t *)arg;
636
637 seq_printf(m, "\n%-23s", "Number of blocks type ");
638 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
639 seq_printf(m, "%12s ", migratetype_names[mtype]);
640 seq_putc(m, '\n');
641 walk_zones_in_node(m, pgdat, pagetypeinfo_showblockcount_print);
642
643 return 0;
644}
645
646/*
647 * This prints out statistics in relation to grouping pages by mobility.
648 * It is expensive to collect so do not constantly read the file.
649 */
650static int pagetypeinfo_show(struct seq_file *m, void *arg)
651{
652 pg_data_t *pgdat = (pg_data_t *)arg;
653
KOSAKI Motohiro41b25a32008-04-30 00:52:13 -0700654 /* check memoryless node */
655 if (!node_state(pgdat->node_id, N_HIGH_MEMORY))
656 return 0;
657
Mel Gorman467c9962007-10-16 01:26:02 -0700658 seq_printf(m, "Page block order: %d\n", pageblock_order);
659 seq_printf(m, "Pages per block: %lu\n", pageblock_nr_pages);
660 seq_putc(m, '\n');
661 pagetypeinfo_showfree(m, pgdat);
662 pagetypeinfo_showblockcount(m, pgdat);
663
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700664 return 0;
665}
666
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +0400667static const struct seq_operations fragmentation_op = {
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700668 .start = frag_start,
669 .next = frag_next,
670 .stop = frag_stop,
671 .show = frag_show,
672};
673
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +0400674static int fragmentation_open(struct inode *inode, struct file *file)
675{
676 return seq_open(file, &fragmentation_op);
677}
678
679static const struct file_operations fragmentation_file_operations = {
680 .open = fragmentation_open,
681 .read = seq_read,
682 .llseek = seq_lseek,
683 .release = seq_release,
684};
685
Alexey Dobriyan74e2e8e2008-10-06 04:15:36 +0400686static const struct seq_operations pagetypeinfo_op = {
Mel Gorman467c9962007-10-16 01:26:02 -0700687 .start = frag_start,
688 .next = frag_next,
689 .stop = frag_stop,
690 .show = pagetypeinfo_show,
691};
692
Alexey Dobriyan74e2e8e2008-10-06 04:15:36 +0400693static int pagetypeinfo_open(struct inode *inode, struct file *file)
694{
695 return seq_open(file, &pagetypeinfo_op);
696}
697
698static const struct file_operations pagetypeinfo_file_ops = {
699 .open = pagetypeinfo_open,
700 .read = seq_read,
701 .llseek = seq_lseek,
702 .release = seq_release,
703};
704
Christoph Lameter4b51d662007-02-10 01:43:10 -0800705#ifdef CONFIG_ZONE_DMA
706#define TEXT_FOR_DMA(xx) xx "_dma",
707#else
708#define TEXT_FOR_DMA(xx)
709#endif
710
Christoph Lameter27bf71c2006-09-25 23:31:15 -0700711#ifdef CONFIG_ZONE_DMA32
712#define TEXT_FOR_DMA32(xx) xx "_dma32",
713#else
714#define TEXT_FOR_DMA32(xx)
715#endif
716
717#ifdef CONFIG_HIGHMEM
718#define TEXT_FOR_HIGHMEM(xx) xx "_high",
719#else
720#define TEXT_FOR_HIGHMEM(xx)
721#endif
722
Christoph Lameter4b51d662007-02-10 01:43:10 -0800723#define TEXTS_FOR_ZONES(xx) TEXT_FOR_DMA(xx) TEXT_FOR_DMA32(xx) xx "_normal", \
Mel Gorman2a1e2742007-07-17 04:03:12 -0700724 TEXT_FOR_HIGHMEM(xx) xx "_movable",
Christoph Lameter27bf71c2006-09-25 23:31:15 -0700725
Helge Deller15ad7cd2006-12-06 20:40:36 -0800726static const char * const vmstat_text[] = {
Christoph Lameter2244b952006-06-30 01:55:33 -0700727 /* Zoned VM counters */
Christoph Lameterd23ad422007-02-10 01:43:02 -0800728 "nr_free_pages",
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700729 "nr_inactive_anon",
730 "nr_active_anon",
731 "nr_inactive_file",
732 "nr_active_file",
Lee Schermerhorn7b854122008-10-18 20:26:40 -0700733 "nr_unevictable",
Nick Piggin5344b7e2008-10-18 20:26:51 -0700734 "nr_mlock",
Christoph Lameterf3dbd342006-06-30 01:55:36 -0700735 "nr_anon_pages",
Christoph Lameter65ba55f2006-06-30 01:55:34 -0700736 "nr_mapped",
Christoph Lameter347ce432006-06-30 01:55:35 -0700737 "nr_file_pages",
Christoph Lameter51ed4492007-02-10 01:43:02 -0800738 "nr_dirty",
739 "nr_writeback",
Christoph Lameter972d1a72006-09-25 23:31:51 -0700740 "nr_slab_reclaimable",
741 "nr_slab_unreclaimable",
Christoph Lameterdf849a12006-06-30 01:55:38 -0700742 "nr_page_table_pages",
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -0700743 "nr_kernel_stack",
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700744 "nr_unstable",
Christoph Lameterd2c5e302006-06-30 01:55:41 -0700745 "nr_bounce",
Andrew Mortone129b5c2006-09-27 01:50:00 -0700746 "nr_vmscan_write",
Miklos Szeredifc3ba692008-04-30 00:54:38 -0700747 "nr_writeback_temp",
KOSAKI Motohiroa7312862009-09-21 17:01:37 -0700748 "nr_isolated_anon",
749 "nr_isolated_file",
KOSAKI Motohiro4b021082009-09-21 17:01:33 -0700750 "nr_shmem",
Michael Rubinea941f02010-10-26 14:21:35 -0700751 "nr_dirtied",
752 "nr_written",
Michael Rubin79da8262010-10-26 14:21:36 -0700753 "nr_dirty_threshold",
754 "nr_dirty_background_threshold",
Michael Rubinea941f02010-10-26 14:21:35 -0700755
Christoph Lameterca889e62006-06-30 01:55:44 -0700756#ifdef CONFIG_NUMA
757 "numa_hit",
758 "numa_miss",
759 "numa_foreign",
760 "numa_interleave",
761 "numa_local",
762 "numa_other",
763#endif
764
Christoph Lameterf8891e52006-06-30 01:55:45 -0700765#ifdef CONFIG_VM_EVENT_COUNTERS
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700766 "pgpgin",
767 "pgpgout",
768 "pswpin",
769 "pswpout",
770
Christoph Lameter27bf71c2006-09-25 23:31:15 -0700771 TEXTS_FOR_ZONES("pgalloc")
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700772
773 "pgfree",
774 "pgactivate",
775 "pgdeactivate",
776
777 "pgfault",
778 "pgmajfault",
779
Christoph Lameter27bf71c2006-09-25 23:31:15 -0700780 TEXTS_FOR_ZONES("pgrefill")
781 TEXTS_FOR_ZONES("pgsteal")
782 TEXTS_FOR_ZONES("pgscan_kswapd")
783 TEXTS_FOR_ZONES("pgscan_direct")
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700784
Mel Gorman24cf725182009-06-16 15:33:23 -0700785#ifdef CONFIG_NUMA
786 "zone_reclaim_failed",
787#endif
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700788 "pginodesteal",
789 "slabs_scanned",
790 "kswapd_steal",
791 "kswapd_inodesteal",
KOSAKI Motohirobb3ab592009-12-14 17:58:55 -0800792 "kswapd_low_wmark_hit_quickly",
793 "kswapd_high_wmark_hit_quickly",
794 "kswapd_skip_congestion_wait",
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700795 "pageoutrun",
796 "allocstall",
797
798 "pgrotated",
Mel Gorman748446b2010-05-24 14:32:27 -0700799
800#ifdef CONFIG_COMPACTION
801 "compact_blocks_moved",
802 "compact_pages_moved",
803 "compact_pagemigrate_failed",
Mel Gorman56de7262010-05-24 14:32:30 -0700804 "compact_stall",
805 "compact_fail",
806 "compact_success",
Mel Gorman748446b2010-05-24 14:32:27 -0700807#endif
808
Adam Litke3b116302008-04-28 02:13:06 -0700809#ifdef CONFIG_HUGETLB_PAGE
810 "htlb_buddy_alloc_success",
811 "htlb_buddy_alloc_fail",
812#endif
Lee Schermerhornbbfd28e2008-10-18 20:26:40 -0700813 "unevictable_pgs_culled",
814 "unevictable_pgs_scanned",
815 "unevictable_pgs_rescued",
Nick Piggin5344b7e2008-10-18 20:26:51 -0700816 "unevictable_pgs_mlocked",
817 "unevictable_pgs_munlocked",
818 "unevictable_pgs_cleared",
819 "unevictable_pgs_stranded",
Lee Schermerhorn985737c2008-10-18 20:26:53 -0700820 "unevictable_pgs_mlockfreed",
Lee Schermerhornbbfd28e2008-10-18 20:26:40 -0700821#endif
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700822};
823
Mel Gorman467c9962007-10-16 01:26:02 -0700824static void zoneinfo_show_print(struct seq_file *m, pg_data_t *pgdat,
825 struct zone *zone)
826{
827 int i;
828 seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
829 seq_printf(m,
830 "\n pages free %lu"
831 "\n min %lu"
832 "\n low %lu"
833 "\n high %lu"
Wu Fengguang08d9ae72009-06-16 15:32:30 -0700834 "\n scanned %lu"
Mel Gorman467c9962007-10-16 01:26:02 -0700835 "\n spanned %lu"
836 "\n present %lu",
Christoph Lameteraa454842010-09-09 16:38:17 -0700837 zone_nr_free_pages(zone),
Mel Gorman41858962009-06-16 15:32:12 -0700838 min_wmark_pages(zone),
839 low_wmark_pages(zone),
840 high_wmark_pages(zone),
Mel Gorman467c9962007-10-16 01:26:02 -0700841 zone->pages_scanned,
Mel Gorman467c9962007-10-16 01:26:02 -0700842 zone->spanned_pages,
843 zone->present_pages);
844
845 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
846 seq_printf(m, "\n %-12s %lu", vmstat_text[i],
847 zone_page_state(zone, i));
848
849 seq_printf(m,
850 "\n protection: (%lu",
851 zone->lowmem_reserve[0]);
852 for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
853 seq_printf(m, ", %lu", zone->lowmem_reserve[i]);
854 seq_printf(m,
855 ")"
856 "\n pagesets");
857 for_each_online_cpu(i) {
858 struct per_cpu_pageset *pageset;
Mel Gorman467c9962007-10-16 01:26:02 -0700859
Christoph Lameter99dcc3e2010-01-05 15:34:51 +0900860 pageset = per_cpu_ptr(zone->pageset, i);
Christoph Lameter3dfa5722008-02-04 22:29:19 -0800861 seq_printf(m,
862 "\n cpu: %i"
863 "\n count: %i"
864 "\n high: %i"
865 "\n batch: %i",
866 i,
867 pageset->pcp.count,
868 pageset->pcp.high,
869 pageset->pcp.batch);
Mel Gorman467c9962007-10-16 01:26:02 -0700870#ifdef CONFIG_SMP
871 seq_printf(m, "\n vm stats threshold: %d",
872 pageset->stat_threshold);
873#endif
874 }
875 seq_printf(m,
876 "\n all_unreclaimable: %u"
Rik van Riel556adec2008-10-18 20:26:34 -0700877 "\n start_pfn: %lu"
878 "\n inactive_ratio: %u",
KOSAKI Motohiro93e4a89a2010-03-05 13:41:55 -0800879 zone->all_unreclaimable,
Rik van Riel556adec2008-10-18 20:26:34 -0700880 zone->zone_start_pfn,
881 zone->inactive_ratio);
Mel Gorman467c9962007-10-16 01:26:02 -0700882 seq_putc(m, '\n');
883}
884
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700885/*
886 * Output information about zones in @pgdat.
887 */
888static int zoneinfo_show(struct seq_file *m, void *arg)
889{
Mel Gorman467c9962007-10-16 01:26:02 -0700890 pg_data_t *pgdat = (pg_data_t *)arg;
891 walk_zones_in_node(m, pgdat, zoneinfo_show_print);
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700892 return 0;
893}
894
Alexey Dobriyan5c9fe622008-10-06 04:19:42 +0400895static const struct seq_operations zoneinfo_op = {
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700896 .start = frag_start, /* iterate over all zones. The same as in
897 * fragmentation. */
898 .next = frag_next,
899 .stop = frag_stop,
900 .show = zoneinfo_show,
901};
902
Alexey Dobriyan5c9fe622008-10-06 04:19:42 +0400903static int zoneinfo_open(struct inode *inode, struct file *file)
904{
905 return seq_open(file, &zoneinfo_op);
906}
907
908static const struct file_operations proc_zoneinfo_file_operations = {
909 .open = zoneinfo_open,
910 .read = seq_read,
911 .llseek = seq_lseek,
912 .release = seq_release,
913};
914
Michael Rubin79da8262010-10-26 14:21:36 -0700915enum writeback_stat_item {
916 NR_DIRTY_THRESHOLD,
917 NR_DIRTY_BG_THRESHOLD,
918 NR_VM_WRITEBACK_STAT_ITEMS,
919};
920
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700921static void *vmstat_start(struct seq_file *m, loff_t *pos)
922{
Christoph Lameter2244b952006-06-30 01:55:33 -0700923 unsigned long *v;
Michael Rubin79da8262010-10-26 14:21:36 -0700924 int i, stat_items_size;
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700925
926 if (*pos >= ARRAY_SIZE(vmstat_text))
927 return NULL;
Michael Rubin79da8262010-10-26 14:21:36 -0700928 stat_items_size = NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long) +
929 NR_VM_WRITEBACK_STAT_ITEMS * sizeof(unsigned long);
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700930
Christoph Lameterf8891e52006-06-30 01:55:45 -0700931#ifdef CONFIG_VM_EVENT_COUNTERS
Michael Rubin79da8262010-10-26 14:21:36 -0700932 stat_items_size += sizeof(struct vm_event_state);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700933#endif
Michael Rubin79da8262010-10-26 14:21:36 -0700934
935 v = kmalloc(stat_items_size, GFP_KERNEL);
Christoph Lameter2244b952006-06-30 01:55:33 -0700936 m->private = v;
937 if (!v)
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700938 return ERR_PTR(-ENOMEM);
Christoph Lameter2244b952006-06-30 01:55:33 -0700939 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
940 v[i] = global_page_state(i);
Michael Rubin79da8262010-10-26 14:21:36 -0700941 v += NR_VM_ZONE_STAT_ITEMS;
942
943 global_dirty_limits(v + NR_DIRTY_BG_THRESHOLD,
944 v + NR_DIRTY_THRESHOLD);
945 v += NR_VM_WRITEBACK_STAT_ITEMS;
946
Christoph Lameterf8891e52006-06-30 01:55:45 -0700947#ifdef CONFIG_VM_EVENT_COUNTERS
Michael Rubin79da8262010-10-26 14:21:36 -0700948 all_vm_events(v);
949 v[PGPGIN] /= 2; /* sectors -> kbytes */
950 v[PGPGOUT] /= 2;
Christoph Lameterf8891e52006-06-30 01:55:45 -0700951#endif
Wu Fengguangff8b16d2010-11-04 01:56:49 +0800952 return (unsigned long *)m->private + *pos;
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700953}
954
955static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
956{
957 (*pos)++;
958 if (*pos >= ARRAY_SIZE(vmstat_text))
959 return NULL;
960 return (unsigned long *)m->private + *pos;
961}
962
963static int vmstat_show(struct seq_file *m, void *arg)
964{
965 unsigned long *l = arg;
966 unsigned long off = l - (unsigned long *)m->private;
967
968 seq_printf(m, "%s %lu\n", vmstat_text[off], *l);
969 return 0;
970}
971
972static void vmstat_stop(struct seq_file *m, void *arg)
973{
974 kfree(m->private);
975 m->private = NULL;
976}
977
Alexey Dobriyanb6aa44a2008-10-06 04:17:48 +0400978static const struct seq_operations vmstat_op = {
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700979 .start = vmstat_start,
980 .next = vmstat_next,
981 .stop = vmstat_stop,
982 .show = vmstat_show,
983};
984
Alexey Dobriyanb6aa44a2008-10-06 04:17:48 +0400985static int vmstat_open(struct inode *inode, struct file *file)
986{
987 return seq_open(file, &vmstat_op);
988}
989
990static const struct file_operations proc_vmstat_file_operations = {
991 .open = vmstat_open,
992 .read = seq_read,
993 .llseek = seq_lseek,
994 .release = seq_release,
995};
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700996#endif /* CONFIG_PROC_FS */
997
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700998#ifdef CONFIG_SMP
Christoph Lameterd1187ed2007-05-09 02:35:12 -0700999static DEFINE_PER_CPU(struct delayed_work, vmstat_work);
Christoph Lameter77461ab2007-05-09 02:35:13 -07001000int sysctl_stat_interval __read_mostly = HZ;
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001001
1002static void vmstat_update(struct work_struct *w)
1003{
1004 refresh_cpu_vm_stats(smp_processor_id());
Christoph Lameter77461ab2007-05-09 02:35:13 -07001005 schedule_delayed_work(&__get_cpu_var(vmstat_work),
Anton Blanchard98f4ebb2009-04-02 16:56:39 -07001006 round_jiffies_relative(sysctl_stat_interval));
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001007}
1008
Randy Dunlap42614fc2007-11-14 17:00:12 -08001009static void __cpuinit start_cpu_timer(int cpu)
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001010{
Tejun Heo1871e522009-10-29 22:34:13 +09001011 struct delayed_work *work = &per_cpu(vmstat_work, cpu);
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001012
Tejun Heo1871e522009-10-29 22:34:13 +09001013 INIT_DELAYED_WORK_DEFERRABLE(work, vmstat_update);
1014 schedule_delayed_work_on(cpu, work, __round_jiffies_relative(HZ, cpu));
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001015}
1016
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001017/*
1018 * Use the cpu notifier to insure that the thresholds are recalculated
1019 * when necessary.
1020 */
1021static int __cpuinit vmstat_cpuup_callback(struct notifier_block *nfb,
1022 unsigned long action,
1023 void *hcpu)
1024{
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001025 long cpu = (long)hcpu;
1026
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001027 switch (action) {
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001028 case CPU_ONLINE:
1029 case CPU_ONLINE_FROZEN:
KAMEZAWA Hiroyuki5ee28a42010-09-09 16:38:14 -07001030 refresh_zone_stat_thresholds();
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001031 start_cpu_timer(cpu);
Christoph Lameterad596922010-01-05 15:34:51 +09001032 node_set_state(cpu_to_node(cpu), N_CPU);
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001033 break;
1034 case CPU_DOWN_PREPARE:
1035 case CPU_DOWN_PREPARE_FROZEN:
1036 cancel_rearming_delayed_work(&per_cpu(vmstat_work, cpu));
1037 per_cpu(vmstat_work, cpu).work.func = NULL;
1038 break;
1039 case CPU_DOWN_FAILED:
1040 case CPU_DOWN_FAILED_FROZEN:
1041 start_cpu_timer(cpu);
1042 break;
Andy Whitcroftce421c72006-12-06 20:33:08 -08001043 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07001044 case CPU_DEAD_FROZEN:
Andy Whitcroftce421c72006-12-06 20:33:08 -08001045 refresh_zone_stat_thresholds();
1046 break;
1047 default:
1048 break;
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001049 }
1050 return NOTIFY_OK;
1051}
1052
1053static struct notifier_block __cpuinitdata vmstat_notifier =
1054 { &vmstat_cpuup_callback, NULL, 0 };
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +04001055#endif
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001056
Adrian Bunke2fc88d2007-10-16 01:26:27 -07001057static int __init setup_vmstat(void)
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001058{
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +04001059#ifdef CONFIG_SMP
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001060 int cpu;
1061
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001062 refresh_zone_stat_thresholds();
1063 register_cpu_notifier(&vmstat_notifier);
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001064
1065 for_each_online_cpu(cpu)
1066 start_cpu_timer(cpu);
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +04001067#endif
1068#ifdef CONFIG_PROC_FS
1069 proc_create("buddyinfo", S_IRUGO, NULL, &fragmentation_file_operations);
Alexey Dobriyan74e2e8e2008-10-06 04:15:36 +04001070 proc_create("pagetypeinfo", S_IRUGO, NULL, &pagetypeinfo_file_ops);
Alexey Dobriyanb6aa44a2008-10-06 04:17:48 +04001071 proc_create("vmstat", S_IRUGO, NULL, &proc_vmstat_file_operations);
Alexey Dobriyan5c9fe622008-10-06 04:19:42 +04001072 proc_create("zoneinfo", S_IRUGO, NULL, &proc_zoneinfo_file_operations);
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +04001073#endif
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001074 return 0;
1075}
1076module_init(setup_vmstat)
Mel Gormand7a57522010-05-24 14:32:25 -07001077
1078#if defined(CONFIG_DEBUG_FS) && defined(CONFIG_COMPACTION)
1079#include <linux/debugfs.h>
1080
1081static struct dentry *extfrag_debug_root;
1082
1083/*
1084 * Return an index indicating how much of the available free memory is
1085 * unusable for an allocation of the requested size.
1086 */
1087static int unusable_free_index(unsigned int order,
1088 struct contig_page_info *info)
1089{
1090 /* No free memory is interpreted as all free memory is unusable */
1091 if (info->free_pages == 0)
1092 return 1000;
1093
1094 /*
1095 * Index should be a value between 0 and 1. Return a value to 3
1096 * decimal places.
1097 *
1098 * 0 => no fragmentation
1099 * 1 => high fragmentation
1100 */
1101 return div_u64((info->free_pages - (info->free_blocks_suitable << order)) * 1000ULL, info->free_pages);
1102
1103}
1104
1105static void unusable_show_print(struct seq_file *m,
1106 pg_data_t *pgdat, struct zone *zone)
1107{
1108 unsigned int order;
1109 int index;
1110 struct contig_page_info info;
1111
1112 seq_printf(m, "Node %d, zone %8s ",
1113 pgdat->node_id,
1114 zone->name);
1115 for (order = 0; order < MAX_ORDER; ++order) {
1116 fill_contig_page_info(zone, order, &info);
1117 index = unusable_free_index(order, &info);
1118 seq_printf(m, "%d.%03d ", index / 1000, index % 1000);
1119 }
1120
1121 seq_putc(m, '\n');
1122}
1123
1124/*
1125 * Display unusable free space index
1126 *
1127 * The unusable free space index measures how much of the available free
1128 * memory cannot be used to satisfy an allocation of a given size and is a
1129 * value between 0 and 1. The higher the value, the more of free memory is
1130 * unusable and by implication, the worse the external fragmentation is. This
1131 * can be expressed as a percentage by multiplying by 100.
1132 */
1133static int unusable_show(struct seq_file *m, void *arg)
1134{
1135 pg_data_t *pgdat = (pg_data_t *)arg;
1136
1137 /* check memoryless node */
1138 if (!node_state(pgdat->node_id, N_HIGH_MEMORY))
1139 return 0;
1140
1141 walk_zones_in_node(m, pgdat, unusable_show_print);
1142
1143 return 0;
1144}
1145
1146static const struct seq_operations unusable_op = {
1147 .start = frag_start,
1148 .next = frag_next,
1149 .stop = frag_stop,
1150 .show = unusable_show,
1151};
1152
1153static int unusable_open(struct inode *inode, struct file *file)
1154{
1155 return seq_open(file, &unusable_op);
1156}
1157
1158static const struct file_operations unusable_file_ops = {
1159 .open = unusable_open,
1160 .read = seq_read,
1161 .llseek = seq_lseek,
1162 .release = seq_release,
1163};
1164
Mel Gormanf1a5ab12010-05-24 14:32:26 -07001165static void extfrag_show_print(struct seq_file *m,
1166 pg_data_t *pgdat, struct zone *zone)
1167{
1168 unsigned int order;
1169 int index;
1170
1171 /* Alloc on stack as interrupts are disabled for zone walk */
1172 struct contig_page_info info;
1173
1174 seq_printf(m, "Node %d, zone %8s ",
1175 pgdat->node_id,
1176 zone->name);
1177 for (order = 0; order < MAX_ORDER; ++order) {
1178 fill_contig_page_info(zone, order, &info);
Mel Gorman56de7262010-05-24 14:32:30 -07001179 index = __fragmentation_index(order, &info);
Mel Gormanf1a5ab12010-05-24 14:32:26 -07001180 seq_printf(m, "%d.%03d ", index / 1000, index % 1000);
1181 }
1182
1183 seq_putc(m, '\n');
1184}
1185
1186/*
1187 * Display fragmentation index for orders that allocations would fail for
1188 */
1189static int extfrag_show(struct seq_file *m, void *arg)
1190{
1191 pg_data_t *pgdat = (pg_data_t *)arg;
1192
1193 walk_zones_in_node(m, pgdat, extfrag_show_print);
1194
1195 return 0;
1196}
1197
1198static const struct seq_operations extfrag_op = {
1199 .start = frag_start,
1200 .next = frag_next,
1201 .stop = frag_stop,
1202 .show = extfrag_show,
1203};
1204
1205static int extfrag_open(struct inode *inode, struct file *file)
1206{
1207 return seq_open(file, &extfrag_op);
1208}
1209
1210static const struct file_operations extfrag_file_ops = {
1211 .open = extfrag_open,
1212 .read = seq_read,
1213 .llseek = seq_lseek,
1214 .release = seq_release,
1215};
1216
Mel Gormand7a57522010-05-24 14:32:25 -07001217static int __init extfrag_debug_init(void)
1218{
1219 extfrag_debug_root = debugfs_create_dir("extfrag", NULL);
1220 if (!extfrag_debug_root)
1221 return -ENOMEM;
1222
1223 if (!debugfs_create_file("unusable_index", 0444,
1224 extfrag_debug_root, NULL, &unusable_file_ops))
1225 return -ENOMEM;
1226
Mel Gormanf1a5ab12010-05-24 14:32:26 -07001227 if (!debugfs_create_file("extfrag_index", 0444,
1228 extfrag_debug_root, NULL, &extfrag_file_ops))
1229 return -ENOMEM;
1230
Mel Gormand7a57522010-05-24 14:32:25 -07001231 return 0;
1232}
1233
1234module_init(extfrag_debug_init);
1235#endif