blob: 772b39b87d955078b4f0d7a1b7114751e0ef9a74 [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
Mel Gormanb44129b2011-01-13 15:45:43 -080086int calculate_pressure_threshold(struct zone *zone)
Mel Gorman88f5acf2011-01-13 15:45:41 -080087{
88 int threshold;
89 int watermark_distance;
90
91 /*
92 * As vmstats are not up to date, there is drift between the estimated
93 * and real values. For high thresholds and a high number of CPUs, it
94 * is possible for the min watermark to be breached while the estimated
95 * value looks fine. The pressure threshold is a reduced value such
96 * that even the maximum amount of drift will not accidentally breach
97 * the min watermark
98 */
99 watermark_distance = low_wmark_pages(zone) - min_wmark_pages(zone);
100 threshold = max(1, (int)(watermark_distance / num_online_cpus()));
101
102 /*
103 * Maximum threshold is 125
104 */
105 threshold = min(125, threshold);
106
107 return threshold;
108}
109
Mel Gormanb44129b2011-01-13 15:45:43 -0800110int calculate_normal_threshold(struct zone *zone)
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700111{
112 int threshold;
113 int mem; /* memory in 128 MB units */
114
115 /*
116 * The threshold scales with the number of processors and the amount
117 * of memory per zone. More memory means that we can defer updates for
118 * longer, more processors could lead to more contention.
119 * fls() is used to have a cheap way of logarithmic scaling.
120 *
121 * Some sample thresholds:
122 *
123 * Threshold Processors (fls) Zonesize fls(mem+1)
124 * ------------------------------------------------------------------
125 * 8 1 1 0.9-1 GB 4
126 * 16 2 2 0.9-1 GB 4
127 * 20 2 2 1-2 GB 5
128 * 24 2 2 2-4 GB 6
129 * 28 2 2 4-8 GB 7
130 * 32 2 2 8-16 GB 8
131 * 4 2 2 <128M 1
132 * 30 4 3 2-4 GB 5
133 * 48 4 3 8-16 GB 8
134 * 32 8 4 1-2 GB 4
135 * 32 8 4 0.9-1GB 4
136 * 10 16 5 <128M 1
137 * 40 16 5 900M 4
138 * 70 64 7 2-4 GB 5
139 * 84 64 7 4-8 GB 6
140 * 108 512 9 4-8 GB 6
141 * 125 1024 10 8-16 GB 8
142 * 125 1024 10 16-32 GB 9
143 */
144
145 mem = zone->present_pages >> (27 - PAGE_SHIFT);
146
147 threshold = 2 * fls(num_online_cpus()) * (1 + fls(mem));
148
149 /*
150 * Maximum threshold is 125
151 */
152 threshold = min(125, threshold);
153
154 return threshold;
155}
Christoph Lameter2244b952006-06-30 01:55:33 -0700156
157/*
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700158 * Refresh the thresholds for each zone.
Christoph Lameter2244b952006-06-30 01:55:33 -0700159 */
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700160static void refresh_zone_stat_thresholds(void)
Christoph Lameter2244b952006-06-30 01:55:33 -0700161{
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700162 struct zone *zone;
163 int cpu;
164 int threshold;
165
KOSAKI Motohiroee99c712009-03-31 15:19:31 -0700166 for_each_populated_zone(zone) {
Christoph Lameteraa454842010-09-09 16:38:17 -0700167 unsigned long max_drift, tolerate_drift;
168
Mel Gormanb44129b2011-01-13 15:45:43 -0800169 threshold = calculate_normal_threshold(zone);
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700170
171 for_each_online_cpu(cpu)
Christoph Lameter99dcc3e2010-01-05 15:34:51 +0900172 per_cpu_ptr(zone->pageset, cpu)->stat_threshold
173 = threshold;
Christoph Lameteraa454842010-09-09 16:38:17 -0700174
175 /*
176 * Only set percpu_drift_mark if there is a danger that
177 * NR_FREE_PAGES reports the low watermark is ok when in fact
178 * the min watermark could be breached by an allocation
179 */
180 tolerate_drift = low_wmark_pages(zone) - min_wmark_pages(zone);
181 max_drift = num_online_cpus() * threshold;
182 if (max_drift > tolerate_drift)
183 zone->percpu_drift_mark = high_wmark_pages(zone) +
184 max_drift;
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700185 }
Christoph Lameter2244b952006-06-30 01:55:33 -0700186}
187
Mel Gormanb44129b2011-01-13 15:45:43 -0800188void set_pgdat_percpu_threshold(pg_data_t *pgdat,
189 int (*calculate_pressure)(struct zone *))
Mel Gorman88f5acf2011-01-13 15:45:41 -0800190{
191 struct zone *zone;
192 int cpu;
193 int threshold;
194 int i;
195
Mel Gorman88f5acf2011-01-13 15:45:41 -0800196 for (i = 0; i < pgdat->nr_zones; i++) {
197 zone = &pgdat->node_zones[i];
198 if (!zone->percpu_drift_mark)
199 continue;
200
Mel Gormanb44129b2011-01-13 15:45:43 -0800201 threshold = (*calculate_pressure)(zone);
202 for_each_possible_cpu(cpu)
Mel Gorman88f5acf2011-01-13 15:45:41 -0800203 per_cpu_ptr(zone->pageset, cpu)->stat_threshold
204 = threshold;
205 }
Mel Gorman88f5acf2011-01-13 15:45:41 -0800206}
207
Christoph Lameter2244b952006-06-30 01:55:33 -0700208/*
209 * For use when we know that interrupts are disabled.
210 */
211void __mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
212 int delta)
213{
Christoph Lameter12938a92010-12-06 11:16:20 -0600214 struct per_cpu_pageset __percpu *pcp = zone->pageset;
215 s8 __percpu *p = pcp->vm_stat_diff + item;
Christoph Lameter2244b952006-06-30 01:55:33 -0700216 long x;
Christoph Lameter12938a92010-12-06 11:16:20 -0600217 long t;
Christoph Lameter2244b952006-06-30 01:55:33 -0700218
Christoph Lameter12938a92010-12-06 11:16:20 -0600219 x = delta + __this_cpu_read(*p);
Christoph Lameter2244b952006-06-30 01:55:33 -0700220
Christoph Lameter12938a92010-12-06 11:16:20 -0600221 t = __this_cpu_read(pcp->stat_threshold);
222
223 if (unlikely(x > t || x < -t)) {
Christoph Lameter2244b952006-06-30 01:55:33 -0700224 zone_page_state_add(x, zone, item);
225 x = 0;
226 }
Christoph Lameter12938a92010-12-06 11:16:20 -0600227 __this_cpu_write(*p, x);
Christoph Lameter2244b952006-06-30 01:55:33 -0700228}
229EXPORT_SYMBOL(__mod_zone_page_state);
230
231/*
Christoph Lameter2244b952006-06-30 01:55:33 -0700232 * Optimized increment and decrement functions.
233 *
234 * These are only for a single page and therefore can take a struct page *
235 * argument instead of struct zone *. This allows the inclusion of the code
236 * generated for page_zone(page) into the optimized functions.
237 *
238 * No overflow check is necessary and therefore the differential can be
239 * incremented or decremented in place which may allow the compilers to
240 * generate better code.
Christoph Lameter2244b952006-06-30 01:55:33 -0700241 * The increment or decrement is known and therefore one boundary check can
242 * be omitted.
243 *
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700244 * NOTE: These functions are very performance sensitive. Change only
245 * with care.
246 *
Christoph Lameter2244b952006-06-30 01:55:33 -0700247 * Some processors have inc/dec instructions that are atomic vs an interrupt.
248 * However, the code must first determine the differential location in a zone
249 * based on the processor number and then inc/dec the counter. There is no
250 * guarantee without disabling preemption that the processor will not change
251 * in between and therefore the atomicity vs. interrupt cannot be exploited
252 * in a useful way here.
253 */
Christoph Lameterc8785382007-02-10 01:43:01 -0800254void __inc_zone_state(struct zone *zone, enum zone_stat_item item)
Christoph Lameter2244b952006-06-30 01:55:33 -0700255{
Christoph Lameter12938a92010-12-06 11:16:20 -0600256 struct per_cpu_pageset __percpu *pcp = zone->pageset;
257 s8 __percpu *p = pcp->vm_stat_diff + item;
258 s8 v, t;
Christoph Lameter2244b952006-06-30 01:55:33 -0700259
Christoph Lameter908ee0f2010-12-06 11:40:02 -0600260 v = __this_cpu_inc_return(*p);
Christoph Lameter12938a92010-12-06 11:16:20 -0600261 t = __this_cpu_read(pcp->stat_threshold);
262 if (unlikely(v > t)) {
263 s8 overstep = t >> 1;
Christoph Lameter2244b952006-06-30 01:55:33 -0700264
Christoph Lameter12938a92010-12-06 11:16:20 -0600265 zone_page_state_add(v + overstep, zone, item);
266 __this_cpu_write(*p, -overstep);
Christoph Lameter2244b952006-06-30 01:55:33 -0700267 }
268}
Christoph Lameterca889e62006-06-30 01:55:44 -0700269
270void __inc_zone_page_state(struct page *page, enum zone_stat_item item)
271{
272 __inc_zone_state(page_zone(page), item);
273}
Christoph Lameter2244b952006-06-30 01:55:33 -0700274EXPORT_SYMBOL(__inc_zone_page_state);
275
Christoph Lameterc8785382007-02-10 01:43:01 -0800276void __dec_zone_state(struct zone *zone, enum zone_stat_item item)
Christoph Lameter2244b952006-06-30 01:55:33 -0700277{
Christoph Lameter12938a92010-12-06 11:16:20 -0600278 struct per_cpu_pageset __percpu *pcp = zone->pageset;
279 s8 __percpu *p = pcp->vm_stat_diff + item;
280 s8 v, t;
Christoph Lameter2244b952006-06-30 01:55:33 -0700281
Christoph Lameter908ee0f2010-12-06 11:40:02 -0600282 v = __this_cpu_dec_return(*p);
Christoph Lameter12938a92010-12-06 11:16:20 -0600283 t = __this_cpu_read(pcp->stat_threshold);
284 if (unlikely(v < - t)) {
285 s8 overstep = t >> 1;
Christoph Lameter2244b952006-06-30 01:55:33 -0700286
Christoph Lameter12938a92010-12-06 11:16:20 -0600287 zone_page_state_add(v - overstep, zone, item);
288 __this_cpu_write(*p, overstep);
Christoph Lameter2244b952006-06-30 01:55:33 -0700289 }
290}
Christoph Lameterc8785382007-02-10 01:43:01 -0800291
292void __dec_zone_page_state(struct page *page, enum zone_stat_item item)
293{
294 __dec_zone_state(page_zone(page), item);
295}
Christoph Lameter2244b952006-06-30 01:55:33 -0700296EXPORT_SYMBOL(__dec_zone_page_state);
297
Christoph Lameter7c839122010-12-14 10:28:46 -0600298#ifdef CONFIG_CMPXCHG_LOCAL
299/*
300 * If we have cmpxchg_local support then we do not need to incur the overhead
301 * that comes with local_irq_save/restore if we use this_cpu_cmpxchg.
302 *
303 * mod_state() modifies the zone counter state through atomic per cpu
304 * operations.
305 *
306 * Overstep mode specifies how overstep should handled:
307 * 0 No overstepping
308 * 1 Overstepping half of threshold
309 * -1 Overstepping minus half of threshold
310*/
311static inline void mod_state(struct zone *zone,
312 enum zone_stat_item item, int delta, int overstep_mode)
313{
314 struct per_cpu_pageset __percpu *pcp = zone->pageset;
315 s8 __percpu *p = pcp->vm_stat_diff + item;
316 long o, n, t, z;
317
318 do {
319 z = 0; /* overflow to zone counters */
320
321 /*
322 * The fetching of the stat_threshold is racy. We may apply
323 * a counter threshold to the wrong the cpu if we get
324 * rescheduled while executing here. However, the following
325 * will apply the threshold again and therefore bring the
326 * counter under the threshold.
327 */
328 t = this_cpu_read(pcp->stat_threshold);
329
330 o = this_cpu_read(*p);
331 n = delta + o;
332
333 if (n > t || n < -t) {
334 int os = overstep_mode * (t >> 1) ;
335
336 /* Overflow must be added to zone counters */
337 z = n + os;
338 n = -os;
339 }
340 } while (this_cpu_cmpxchg(*p, o, n) != o);
341
342 if (z)
343 zone_page_state_add(z, zone, item);
344}
345
346void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
347 int delta)
348{
349 mod_state(zone, item, delta, 0);
350}
351EXPORT_SYMBOL(mod_zone_page_state);
352
353void inc_zone_state(struct zone *zone, enum zone_stat_item item)
354{
355 mod_state(zone, item, 1, 1);
356}
357
358void inc_zone_page_state(struct page *page, enum zone_stat_item item)
359{
360 mod_state(page_zone(page), item, 1, 1);
361}
362EXPORT_SYMBOL(inc_zone_page_state);
363
364void dec_zone_page_state(struct page *page, enum zone_stat_item item)
365{
366 mod_state(page_zone(page), item, -1, -1);
367}
368EXPORT_SYMBOL(dec_zone_page_state);
369#else
370/*
371 * Use interrupt disable to serialize counter updates
372 */
373void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
374 int delta)
375{
376 unsigned long flags;
377
378 local_irq_save(flags);
379 __mod_zone_page_state(zone, item, delta);
380 local_irq_restore(flags);
381}
382EXPORT_SYMBOL(mod_zone_page_state);
383
Christoph Lameterca889e62006-06-30 01:55:44 -0700384void inc_zone_state(struct zone *zone, enum zone_stat_item item)
385{
386 unsigned long flags;
387
388 local_irq_save(flags);
389 __inc_zone_state(zone, item);
390 local_irq_restore(flags);
391}
392
Christoph Lameter2244b952006-06-30 01:55:33 -0700393void inc_zone_page_state(struct page *page, enum zone_stat_item item)
394{
395 unsigned long flags;
396 struct zone *zone;
Christoph Lameter2244b952006-06-30 01:55:33 -0700397
398 zone = page_zone(page);
399 local_irq_save(flags);
Christoph Lameterca889e62006-06-30 01:55:44 -0700400 __inc_zone_state(zone, item);
Christoph Lameter2244b952006-06-30 01:55:33 -0700401 local_irq_restore(flags);
402}
403EXPORT_SYMBOL(inc_zone_page_state);
404
405void dec_zone_page_state(struct page *page, enum zone_stat_item item)
406{
407 unsigned long flags;
Christoph Lameter2244b952006-06-30 01:55:33 -0700408
Christoph Lameter2244b952006-06-30 01:55:33 -0700409 local_irq_save(flags);
Christoph Lametera302eb42006-08-31 21:27:34 -0700410 __dec_zone_page_state(page, item);
Christoph Lameter2244b952006-06-30 01:55:33 -0700411 local_irq_restore(flags);
412}
413EXPORT_SYMBOL(dec_zone_page_state);
Christoph Lameter7c839122010-12-14 10:28:46 -0600414#endif
Christoph Lameter2244b952006-06-30 01:55:33 -0700415
416/*
417 * Update the zone counters for one cpu.
Christoph Lameter4037d452007-05-09 02:35:14 -0700418 *
Christoph Lametera7f75e22008-02-04 22:29:16 -0800419 * The cpu specified must be either the current cpu or a processor that
420 * is not online. If it is the current cpu then the execution thread must
421 * be pinned to the current cpu.
422 *
Christoph Lameter4037d452007-05-09 02:35:14 -0700423 * Note that refresh_cpu_vm_stats strives to only access
424 * node local memory. The per cpu pagesets on remote zones are placed
425 * in the memory local to the processor using that pageset. So the
426 * loop over all zones will access a series of cachelines local to
427 * the processor.
428 *
429 * The call to zone_page_state_add updates the cachelines with the
430 * statistics in the remote zone struct as well as the global cachelines
431 * with the global counters. These could cause remote node cache line
432 * bouncing and will have to be only done when necessary.
Christoph Lameter2244b952006-06-30 01:55:33 -0700433 */
434void refresh_cpu_vm_stats(int cpu)
435{
436 struct zone *zone;
437 int i;
Christoph Lametera7f75e22008-02-04 22:29:16 -0800438 int global_diff[NR_VM_ZONE_STAT_ITEMS] = { 0, };
Christoph Lameter2244b952006-06-30 01:55:33 -0700439
KOSAKI Motohiroee99c712009-03-31 15:19:31 -0700440 for_each_populated_zone(zone) {
Christoph Lameter4037d452007-05-09 02:35:14 -0700441 struct per_cpu_pageset *p;
Christoph Lameter2244b952006-06-30 01:55:33 -0700442
Christoph Lameter99dcc3e2010-01-05 15:34:51 +0900443 p = per_cpu_ptr(zone->pageset, cpu);
Christoph Lameter2244b952006-06-30 01:55:33 -0700444
445 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
Christoph Lameter4037d452007-05-09 02:35:14 -0700446 if (p->vm_stat_diff[i]) {
Christoph Lametera7f75e22008-02-04 22:29:16 -0800447 unsigned long flags;
448 int v;
449
Christoph Lameter2244b952006-06-30 01:55:33 -0700450 local_irq_save(flags);
Christoph Lametera7f75e22008-02-04 22:29:16 -0800451 v = p->vm_stat_diff[i];
Christoph Lameter4037d452007-05-09 02:35:14 -0700452 p->vm_stat_diff[i] = 0;
Christoph Lametera7f75e22008-02-04 22:29:16 -0800453 local_irq_restore(flags);
454 atomic_long_add(v, &zone->vm_stat[i]);
455 global_diff[i] += v;
Christoph Lameter4037d452007-05-09 02:35:14 -0700456#ifdef CONFIG_NUMA
457 /* 3 seconds idle till flush */
458 p->expire = 3;
459#endif
Christoph Lameter2244b952006-06-30 01:55:33 -0700460 }
Dimitri Sivanich468fd622008-04-28 02:13:37 -0700461 cond_resched();
Christoph Lameter4037d452007-05-09 02:35:14 -0700462#ifdef CONFIG_NUMA
463 /*
464 * Deal with draining the remote pageset of this
465 * processor
466 *
467 * Check if there are pages remaining in this pageset
468 * if not then there is nothing to expire.
469 */
Christoph Lameter3dfa5722008-02-04 22:29:19 -0800470 if (!p->expire || !p->pcp.count)
Christoph Lameter4037d452007-05-09 02:35:14 -0700471 continue;
472
473 /*
474 * We never drain zones local to this processor.
475 */
476 if (zone_to_nid(zone) == numa_node_id()) {
477 p->expire = 0;
478 continue;
479 }
480
481 p->expire--;
482 if (p->expire)
483 continue;
484
Christoph Lameter3dfa5722008-02-04 22:29:19 -0800485 if (p->pcp.count)
486 drain_zone_pages(zone, &p->pcp);
Christoph Lameter4037d452007-05-09 02:35:14 -0700487#endif
Christoph Lameter2244b952006-06-30 01:55:33 -0700488 }
Christoph Lametera7f75e22008-02-04 22:29:16 -0800489
490 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
491 if (global_diff[i])
492 atomic_long_add(global_diff[i], &vm_stat[i]);
Christoph Lameter2244b952006-06-30 01:55:33 -0700493}
494
Christoph Lameter2244b952006-06-30 01:55:33 -0700495#endif
496
Christoph Lameterca889e62006-06-30 01:55:44 -0700497#ifdef CONFIG_NUMA
498/*
499 * zonelist = the list of zones passed to the allocator
500 * z = the zone from which the allocation occurred.
501 *
502 * Must be called with interrupts disabled.
Andi Kleen78afd562011-03-22 16:33:12 -0700503 *
504 * When __GFP_OTHER_NODE is set assume the node of the preferred
505 * zone is the local node. This is useful for daemons who allocate
506 * memory on behalf of other processes.
Christoph Lameterca889e62006-06-30 01:55:44 -0700507 */
Andi Kleen78afd562011-03-22 16:33:12 -0700508void zone_statistics(struct zone *preferred_zone, struct zone *z, gfp_t flags)
Christoph Lameterca889e62006-06-30 01:55:44 -0700509{
Mel Gorman18ea7e72008-04-28 02:12:14 -0700510 if (z->zone_pgdat == preferred_zone->zone_pgdat) {
Christoph Lameterca889e62006-06-30 01:55:44 -0700511 __inc_zone_state(z, NUMA_HIT);
512 } else {
513 __inc_zone_state(z, NUMA_MISS);
Mel Gorman18ea7e72008-04-28 02:12:14 -0700514 __inc_zone_state(preferred_zone, NUMA_FOREIGN);
Christoph Lameterca889e62006-06-30 01:55:44 -0700515 }
Andi Kleen78afd562011-03-22 16:33:12 -0700516 if (z->node == ((flags & __GFP_OTHER_NODE) ?
517 preferred_zone->node : numa_node_id()))
Christoph Lameterca889e62006-06-30 01:55:44 -0700518 __inc_zone_state(z, NUMA_LOCAL);
519 else
520 __inc_zone_state(z, NUMA_OTHER);
521}
522#endif
523
Mel Gormand7a57522010-05-24 14:32:25 -0700524#ifdef CONFIG_COMPACTION
Namhyung Kim36deb0b2010-10-26 14:22:04 -0700525
Mel Gormand7a57522010-05-24 14:32:25 -0700526struct contig_page_info {
527 unsigned long free_pages;
528 unsigned long free_blocks_total;
529 unsigned long free_blocks_suitable;
530};
531
532/*
533 * Calculate the number of free pages in a zone, how many contiguous
534 * pages are free and how many are large enough to satisfy an allocation of
535 * the target size. Note that this function makes no attempt to estimate
536 * how many suitable free blocks there *might* be if MOVABLE pages were
537 * migrated. Calculating that is possible, but expensive and can be
538 * figured out from userspace
539 */
540static void fill_contig_page_info(struct zone *zone,
541 unsigned int suitable_order,
542 struct contig_page_info *info)
543{
544 unsigned int order;
545
546 info->free_pages = 0;
547 info->free_blocks_total = 0;
548 info->free_blocks_suitable = 0;
549
550 for (order = 0; order < MAX_ORDER; order++) {
551 unsigned long blocks;
552
553 /* Count number of free blocks */
554 blocks = zone->free_area[order].nr_free;
555 info->free_blocks_total += blocks;
556
557 /* Count free base pages */
558 info->free_pages += blocks << order;
559
560 /* Count the suitable free blocks */
561 if (order >= suitable_order)
562 info->free_blocks_suitable += blocks <<
563 (order - suitable_order);
564 }
565}
Mel Gormanf1a5ab12010-05-24 14:32:26 -0700566
567/*
568 * A fragmentation index only makes sense if an allocation of a requested
569 * size would fail. If that is true, the fragmentation index indicates
570 * whether external fragmentation or a lack of memory was the problem.
571 * The value can be used to determine if page reclaim or compaction
572 * should be used
573 */
Mel Gorman56de7262010-05-24 14:32:30 -0700574static int __fragmentation_index(unsigned int order, struct contig_page_info *info)
Mel Gormanf1a5ab12010-05-24 14:32:26 -0700575{
576 unsigned long requested = 1UL << order;
577
578 if (!info->free_blocks_total)
579 return 0;
580
581 /* Fragmentation index only makes sense when a request would fail */
582 if (info->free_blocks_suitable)
583 return -1000;
584
585 /*
586 * Index is between 0 and 1 so return within 3 decimal places
587 *
588 * 0 => allocation would fail due to lack of memory
589 * 1 => allocation would fail due to fragmentation
590 */
591 return 1000 - div_u64( (1000+(div_u64(info->free_pages * 1000ULL, requested))), info->free_blocks_total);
592}
Mel Gorman56de7262010-05-24 14:32:30 -0700593
594/* Same as __fragmentation index but allocs contig_page_info on stack */
595int fragmentation_index(struct zone *zone, unsigned int order)
596{
597 struct contig_page_info info;
598
599 fill_contig_page_info(zone, order, &info);
600 return __fragmentation_index(order, &info);
601}
Mel Gormand7a57522010-05-24 14:32:25 -0700602#endif
603
604#if defined(CONFIG_PROC_FS) || defined(CONFIG_COMPACTION)
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +0400605#include <linux/proc_fs.h>
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700606#include <linux/seq_file.h>
607
Mel Gorman467c9962007-10-16 01:26:02 -0700608static char * const migratetype_names[MIGRATE_TYPES] = {
609 "Unmovable",
610 "Reclaimable",
611 "Movable",
612 "Reserve",
KOSAKI Motohiro91446b02008-04-15 14:34:42 -0700613 "Isolate",
Mel Gorman467c9962007-10-16 01:26:02 -0700614};
615
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700616static void *frag_start(struct seq_file *m, loff_t *pos)
617{
618 pg_data_t *pgdat;
619 loff_t node = *pos;
620 for (pgdat = first_online_pgdat();
621 pgdat && node;
622 pgdat = next_online_pgdat(pgdat))
623 --node;
624
625 return pgdat;
626}
627
628static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
629{
630 pg_data_t *pgdat = (pg_data_t *)arg;
631
632 (*pos)++;
633 return next_online_pgdat(pgdat);
634}
635
636static void frag_stop(struct seq_file *m, void *arg)
637{
638}
639
Mel Gorman467c9962007-10-16 01:26:02 -0700640/* Walk all the zones in a node and print using a callback */
641static void walk_zones_in_node(struct seq_file *m, pg_data_t *pgdat,
642 void (*print)(struct seq_file *m, pg_data_t *, struct zone *))
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700643{
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700644 struct zone *zone;
645 struct zone *node_zones = pgdat->node_zones;
646 unsigned long flags;
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700647
648 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
649 if (!populated_zone(zone))
650 continue;
651
652 spin_lock_irqsave(&zone->lock, flags);
Mel Gorman467c9962007-10-16 01:26:02 -0700653 print(m, pgdat, zone);
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700654 spin_unlock_irqrestore(&zone->lock, flags);
Mel Gorman467c9962007-10-16 01:26:02 -0700655 }
656}
Mel Gormand7a57522010-05-24 14:32:25 -0700657#endif
Mel Gorman467c9962007-10-16 01:26:02 -0700658
Mel Gormand7a57522010-05-24 14:32:25 -0700659#ifdef CONFIG_PROC_FS
Mel Gorman467c9962007-10-16 01:26:02 -0700660static void frag_show_print(struct seq_file *m, pg_data_t *pgdat,
661 struct zone *zone)
662{
663 int order;
664
665 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
666 for (order = 0; order < MAX_ORDER; ++order)
667 seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
668 seq_putc(m, '\n');
669}
670
671/*
672 * This walks the free areas for each zone.
673 */
674static int frag_show(struct seq_file *m, void *arg)
675{
676 pg_data_t *pgdat = (pg_data_t *)arg;
677 walk_zones_in_node(m, pgdat, frag_show_print);
678 return 0;
679}
680
681static void pagetypeinfo_showfree_print(struct seq_file *m,
682 pg_data_t *pgdat, struct zone *zone)
683{
684 int order, mtype;
685
686 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++) {
687 seq_printf(m, "Node %4d, zone %8s, type %12s ",
688 pgdat->node_id,
689 zone->name,
690 migratetype_names[mtype]);
691 for (order = 0; order < MAX_ORDER; ++order) {
692 unsigned long freecount = 0;
693 struct free_area *area;
694 struct list_head *curr;
695
696 area = &(zone->free_area[order]);
697
698 list_for_each(curr, &area->free_list[mtype])
699 freecount++;
700 seq_printf(m, "%6lu ", freecount);
701 }
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700702 seq_putc(m, '\n');
703 }
Mel Gorman467c9962007-10-16 01:26:02 -0700704}
705
706/* Print out the free pages at each order for each migatetype */
707static int pagetypeinfo_showfree(struct seq_file *m, void *arg)
708{
709 int order;
710 pg_data_t *pgdat = (pg_data_t *)arg;
711
712 /* Print header */
713 seq_printf(m, "%-43s ", "Free pages count per migrate type at order");
714 for (order = 0; order < MAX_ORDER; ++order)
715 seq_printf(m, "%6d ", order);
716 seq_putc(m, '\n');
717
718 walk_zones_in_node(m, pgdat, pagetypeinfo_showfree_print);
719
720 return 0;
721}
722
723static void pagetypeinfo_showblockcount_print(struct seq_file *m,
724 pg_data_t *pgdat, struct zone *zone)
725{
726 int mtype;
727 unsigned long pfn;
728 unsigned long start_pfn = zone->zone_start_pfn;
729 unsigned long end_pfn = start_pfn + zone->spanned_pages;
730 unsigned long count[MIGRATE_TYPES] = { 0, };
731
732 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
733 struct page *page;
734
735 if (!pfn_valid(pfn))
736 continue;
737
738 page = pfn_to_page(pfn);
Mel Gormaneb335752009-05-13 17:34:48 +0100739
740 /* Watch for unexpected holes punched in the memmap */
741 if (!memmap_valid_within(pfn, page, zone))
Mel Gormane80d6a22008-08-14 11:10:14 +0100742 continue;
Mel Gormaneb335752009-05-13 17:34:48 +0100743
Mel Gorman467c9962007-10-16 01:26:02 -0700744 mtype = get_pageblock_migratetype(page);
745
Mel Gormane80d6a22008-08-14 11:10:14 +0100746 if (mtype < MIGRATE_TYPES)
747 count[mtype]++;
Mel Gorman467c9962007-10-16 01:26:02 -0700748 }
749
750 /* Print counts */
751 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
752 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
753 seq_printf(m, "%12lu ", count[mtype]);
754 seq_putc(m, '\n');
755}
756
757/* Print out the free pages at each order for each migratetype */
758static int pagetypeinfo_showblockcount(struct seq_file *m, void *arg)
759{
760 int mtype;
761 pg_data_t *pgdat = (pg_data_t *)arg;
762
763 seq_printf(m, "\n%-23s", "Number of blocks type ");
764 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
765 seq_printf(m, "%12s ", migratetype_names[mtype]);
766 seq_putc(m, '\n');
767 walk_zones_in_node(m, pgdat, pagetypeinfo_showblockcount_print);
768
769 return 0;
770}
771
772/*
773 * This prints out statistics in relation to grouping pages by mobility.
774 * It is expensive to collect so do not constantly read the file.
775 */
776static int pagetypeinfo_show(struct seq_file *m, void *arg)
777{
778 pg_data_t *pgdat = (pg_data_t *)arg;
779
KOSAKI Motohiro41b25a32008-04-30 00:52:13 -0700780 /* check memoryless node */
781 if (!node_state(pgdat->node_id, N_HIGH_MEMORY))
782 return 0;
783
Mel Gorman467c9962007-10-16 01:26:02 -0700784 seq_printf(m, "Page block order: %d\n", pageblock_order);
785 seq_printf(m, "Pages per block: %lu\n", pageblock_nr_pages);
786 seq_putc(m, '\n');
787 pagetypeinfo_showfree(m, pgdat);
788 pagetypeinfo_showblockcount(m, pgdat);
789
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700790 return 0;
791}
792
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +0400793static const struct seq_operations fragmentation_op = {
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700794 .start = frag_start,
795 .next = frag_next,
796 .stop = frag_stop,
797 .show = frag_show,
798};
799
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +0400800static int fragmentation_open(struct inode *inode, struct file *file)
801{
802 return seq_open(file, &fragmentation_op);
803}
804
805static const struct file_operations fragmentation_file_operations = {
806 .open = fragmentation_open,
807 .read = seq_read,
808 .llseek = seq_lseek,
809 .release = seq_release,
810};
811
Alexey Dobriyan74e2e8e2008-10-06 04:15:36 +0400812static const struct seq_operations pagetypeinfo_op = {
Mel Gorman467c9962007-10-16 01:26:02 -0700813 .start = frag_start,
814 .next = frag_next,
815 .stop = frag_stop,
816 .show = pagetypeinfo_show,
817};
818
Alexey Dobriyan74e2e8e2008-10-06 04:15:36 +0400819static int pagetypeinfo_open(struct inode *inode, struct file *file)
820{
821 return seq_open(file, &pagetypeinfo_op);
822}
823
824static const struct file_operations pagetypeinfo_file_ops = {
825 .open = pagetypeinfo_open,
826 .read = seq_read,
827 .llseek = seq_lseek,
828 .release = seq_release,
829};
830
Christoph Lameter4b51d662007-02-10 01:43:10 -0800831#ifdef CONFIG_ZONE_DMA
832#define TEXT_FOR_DMA(xx) xx "_dma",
833#else
834#define TEXT_FOR_DMA(xx)
835#endif
836
Christoph Lameter27bf71c2006-09-25 23:31:15 -0700837#ifdef CONFIG_ZONE_DMA32
838#define TEXT_FOR_DMA32(xx) xx "_dma32",
839#else
840#define TEXT_FOR_DMA32(xx)
841#endif
842
843#ifdef CONFIG_HIGHMEM
844#define TEXT_FOR_HIGHMEM(xx) xx "_high",
845#else
846#define TEXT_FOR_HIGHMEM(xx)
847#endif
848
Christoph Lameter4b51d662007-02-10 01:43:10 -0800849#define TEXTS_FOR_ZONES(xx) TEXT_FOR_DMA(xx) TEXT_FOR_DMA32(xx) xx "_normal", \
Mel Gorman2a1e2742007-07-17 04:03:12 -0700850 TEXT_FOR_HIGHMEM(xx) xx "_movable",
Christoph Lameter27bf71c2006-09-25 23:31:15 -0700851
Helge Deller15ad7cd2006-12-06 20:40:36 -0800852static const char * const vmstat_text[] = {
Christoph Lameter2244b952006-06-30 01:55:33 -0700853 /* Zoned VM counters */
Christoph Lameterd23ad422007-02-10 01:43:02 -0800854 "nr_free_pages",
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700855 "nr_inactive_anon",
856 "nr_active_anon",
857 "nr_inactive_file",
858 "nr_active_file",
Lee Schermerhorn7b854122008-10-18 20:26:40 -0700859 "nr_unevictable",
Nick Piggin5344b7e2008-10-18 20:26:51 -0700860 "nr_mlock",
Christoph Lameterf3dbd342006-06-30 01:55:36 -0700861 "nr_anon_pages",
Christoph Lameter65ba55f2006-06-30 01:55:34 -0700862 "nr_mapped",
Christoph Lameter347ce432006-06-30 01:55:35 -0700863 "nr_file_pages",
Christoph Lameter51ed4492007-02-10 01:43:02 -0800864 "nr_dirty",
865 "nr_writeback",
Christoph Lameter972d1a72006-09-25 23:31:51 -0700866 "nr_slab_reclaimable",
867 "nr_slab_unreclaimable",
Christoph Lameterdf849a12006-06-30 01:55:38 -0700868 "nr_page_table_pages",
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -0700869 "nr_kernel_stack",
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700870 "nr_unstable",
Christoph Lameterd2c5e302006-06-30 01:55:41 -0700871 "nr_bounce",
Andrew Mortone129b5c2006-09-27 01:50:00 -0700872 "nr_vmscan_write",
Miklos Szeredifc3ba692008-04-30 00:54:38 -0700873 "nr_writeback_temp",
KOSAKI Motohiroa7312862009-09-21 17:01:37 -0700874 "nr_isolated_anon",
875 "nr_isolated_file",
KOSAKI Motohiro4b021082009-09-21 17:01:33 -0700876 "nr_shmem",
Michael Rubinea941f02010-10-26 14:21:35 -0700877 "nr_dirtied",
878 "nr_written",
879
Christoph Lameterca889e62006-06-30 01:55:44 -0700880#ifdef CONFIG_NUMA
881 "numa_hit",
882 "numa_miss",
883 "numa_foreign",
884 "numa_interleave",
885 "numa_local",
886 "numa_other",
887#endif
Andrea Arcangeli79134172011-01-13 15:46:58 -0800888 "nr_anon_transparent_hugepages",
Wu Fengguange1726622010-12-02 14:31:13 -0800889 "nr_dirty_threshold",
890 "nr_dirty_background_threshold",
Christoph Lameterca889e62006-06-30 01:55:44 -0700891
Christoph Lameterf8891e52006-06-30 01:55:45 -0700892#ifdef CONFIG_VM_EVENT_COUNTERS
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700893 "pgpgin",
894 "pgpgout",
895 "pswpin",
896 "pswpout",
897
Christoph Lameter27bf71c2006-09-25 23:31:15 -0700898 TEXTS_FOR_ZONES("pgalloc")
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700899
900 "pgfree",
901 "pgactivate",
902 "pgdeactivate",
903
904 "pgfault",
905 "pgmajfault",
906
Christoph Lameter27bf71c2006-09-25 23:31:15 -0700907 TEXTS_FOR_ZONES("pgrefill")
908 TEXTS_FOR_ZONES("pgsteal")
909 TEXTS_FOR_ZONES("pgscan_kswapd")
910 TEXTS_FOR_ZONES("pgscan_direct")
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700911
Mel Gorman24cf725182009-06-16 15:33:23 -0700912#ifdef CONFIG_NUMA
913 "zone_reclaim_failed",
914#endif
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700915 "pginodesteal",
916 "slabs_scanned",
917 "kswapd_steal",
918 "kswapd_inodesteal",
KOSAKI Motohirobb3ab592009-12-14 17:58:55 -0800919 "kswapd_low_wmark_hit_quickly",
920 "kswapd_high_wmark_hit_quickly",
921 "kswapd_skip_congestion_wait",
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700922 "pageoutrun",
923 "allocstall",
924
925 "pgrotated",
Mel Gorman748446b2010-05-24 14:32:27 -0700926
927#ifdef CONFIG_COMPACTION
928 "compact_blocks_moved",
929 "compact_pages_moved",
930 "compact_pagemigrate_failed",
Mel Gorman56de7262010-05-24 14:32:30 -0700931 "compact_stall",
932 "compact_fail",
933 "compact_success",
Mel Gorman748446b2010-05-24 14:32:27 -0700934#endif
935
Adam Litke3b116302008-04-28 02:13:06 -0700936#ifdef CONFIG_HUGETLB_PAGE
937 "htlb_buddy_alloc_success",
938 "htlb_buddy_alloc_fail",
939#endif
Lee Schermerhornbbfd28e2008-10-18 20:26:40 -0700940 "unevictable_pgs_culled",
941 "unevictable_pgs_scanned",
942 "unevictable_pgs_rescued",
Nick Piggin5344b7e2008-10-18 20:26:51 -0700943 "unevictable_pgs_mlocked",
944 "unevictable_pgs_munlocked",
945 "unevictable_pgs_cleared",
946 "unevictable_pgs_stranded",
Lee Schermerhorn985737c2008-10-18 20:26:53 -0700947 "unevictable_pgs_mlockfreed",
Lee Schermerhornbbfd28e2008-10-18 20:26:40 -0700948#endif
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700949};
950
Mel Gorman467c9962007-10-16 01:26:02 -0700951static void zoneinfo_show_print(struct seq_file *m, pg_data_t *pgdat,
952 struct zone *zone)
953{
954 int i;
955 seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
956 seq_printf(m,
957 "\n pages free %lu"
958 "\n min %lu"
959 "\n low %lu"
960 "\n high %lu"
Wu Fengguang08d9ae72009-06-16 15:32:30 -0700961 "\n scanned %lu"
Mel Gorman467c9962007-10-16 01:26:02 -0700962 "\n spanned %lu"
963 "\n present %lu",
Mel Gorman88f5acf2011-01-13 15:45:41 -0800964 zone_page_state(zone, NR_FREE_PAGES),
Mel Gorman41858962009-06-16 15:32:12 -0700965 min_wmark_pages(zone),
966 low_wmark_pages(zone),
967 high_wmark_pages(zone),
Mel Gorman467c9962007-10-16 01:26:02 -0700968 zone->pages_scanned,
Mel Gorman467c9962007-10-16 01:26:02 -0700969 zone->spanned_pages,
970 zone->present_pages);
971
972 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
973 seq_printf(m, "\n %-12s %lu", vmstat_text[i],
974 zone_page_state(zone, i));
975
976 seq_printf(m,
977 "\n protection: (%lu",
978 zone->lowmem_reserve[0]);
979 for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
980 seq_printf(m, ", %lu", zone->lowmem_reserve[i]);
981 seq_printf(m,
982 ")"
983 "\n pagesets");
984 for_each_online_cpu(i) {
985 struct per_cpu_pageset *pageset;
Mel Gorman467c9962007-10-16 01:26:02 -0700986
Christoph Lameter99dcc3e2010-01-05 15:34:51 +0900987 pageset = per_cpu_ptr(zone->pageset, i);
Christoph Lameter3dfa5722008-02-04 22:29:19 -0800988 seq_printf(m,
989 "\n cpu: %i"
990 "\n count: %i"
991 "\n high: %i"
992 "\n batch: %i",
993 i,
994 pageset->pcp.count,
995 pageset->pcp.high,
996 pageset->pcp.batch);
Mel Gorman467c9962007-10-16 01:26:02 -0700997#ifdef CONFIG_SMP
998 seq_printf(m, "\n vm stats threshold: %d",
999 pageset->stat_threshold);
1000#endif
1001 }
1002 seq_printf(m,
1003 "\n all_unreclaimable: %u"
Rik van Riel556adec2008-10-18 20:26:34 -07001004 "\n start_pfn: %lu"
1005 "\n inactive_ratio: %u",
KOSAKI Motohiro93e4a892010-03-05 13:41:55 -08001006 zone->all_unreclaimable,
Rik van Riel556adec2008-10-18 20:26:34 -07001007 zone->zone_start_pfn,
1008 zone->inactive_ratio);
Mel Gorman467c9962007-10-16 01:26:02 -07001009 seq_putc(m, '\n');
1010}
1011
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001012/*
1013 * Output information about zones in @pgdat.
1014 */
1015static int zoneinfo_show(struct seq_file *m, void *arg)
1016{
Mel Gorman467c9962007-10-16 01:26:02 -07001017 pg_data_t *pgdat = (pg_data_t *)arg;
1018 walk_zones_in_node(m, pgdat, zoneinfo_show_print);
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001019 return 0;
1020}
1021
Alexey Dobriyan5c9fe622008-10-06 04:19:42 +04001022static const struct seq_operations zoneinfo_op = {
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001023 .start = frag_start, /* iterate over all zones. The same as in
1024 * fragmentation. */
1025 .next = frag_next,
1026 .stop = frag_stop,
1027 .show = zoneinfo_show,
1028};
1029
Alexey Dobriyan5c9fe622008-10-06 04:19:42 +04001030static int zoneinfo_open(struct inode *inode, struct file *file)
1031{
1032 return seq_open(file, &zoneinfo_op);
1033}
1034
1035static const struct file_operations proc_zoneinfo_file_operations = {
1036 .open = zoneinfo_open,
1037 .read = seq_read,
1038 .llseek = seq_lseek,
1039 .release = seq_release,
1040};
1041
Michael Rubin79da8262010-10-26 14:21:36 -07001042enum writeback_stat_item {
1043 NR_DIRTY_THRESHOLD,
1044 NR_DIRTY_BG_THRESHOLD,
1045 NR_VM_WRITEBACK_STAT_ITEMS,
1046};
1047
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001048static void *vmstat_start(struct seq_file *m, loff_t *pos)
1049{
Christoph Lameter2244b952006-06-30 01:55:33 -07001050 unsigned long *v;
Michael Rubin79da8262010-10-26 14:21:36 -07001051 int i, stat_items_size;
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001052
1053 if (*pos >= ARRAY_SIZE(vmstat_text))
1054 return NULL;
Michael Rubin79da8262010-10-26 14:21:36 -07001055 stat_items_size = NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long) +
1056 NR_VM_WRITEBACK_STAT_ITEMS * sizeof(unsigned long);
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001057
Christoph Lameterf8891e52006-06-30 01:55:45 -07001058#ifdef CONFIG_VM_EVENT_COUNTERS
Michael Rubin79da8262010-10-26 14:21:36 -07001059 stat_items_size += sizeof(struct vm_event_state);
Christoph Lameterf8891e52006-06-30 01:55:45 -07001060#endif
Michael Rubin79da8262010-10-26 14:21:36 -07001061
1062 v = kmalloc(stat_items_size, GFP_KERNEL);
Christoph Lameter2244b952006-06-30 01:55:33 -07001063 m->private = v;
1064 if (!v)
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001065 return ERR_PTR(-ENOMEM);
Christoph Lameter2244b952006-06-30 01:55:33 -07001066 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
1067 v[i] = global_page_state(i);
Michael Rubin79da8262010-10-26 14:21:36 -07001068 v += NR_VM_ZONE_STAT_ITEMS;
1069
1070 global_dirty_limits(v + NR_DIRTY_BG_THRESHOLD,
1071 v + NR_DIRTY_THRESHOLD);
1072 v += NR_VM_WRITEBACK_STAT_ITEMS;
1073
Christoph Lameterf8891e52006-06-30 01:55:45 -07001074#ifdef CONFIG_VM_EVENT_COUNTERS
Michael Rubin79da8262010-10-26 14:21:36 -07001075 all_vm_events(v);
1076 v[PGPGIN] /= 2; /* sectors -> kbytes */
1077 v[PGPGOUT] /= 2;
Christoph Lameterf8891e52006-06-30 01:55:45 -07001078#endif
Wu Fengguangff8b16d2010-11-04 01:56:49 +08001079 return (unsigned long *)m->private + *pos;
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001080}
1081
1082static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
1083{
1084 (*pos)++;
1085 if (*pos >= ARRAY_SIZE(vmstat_text))
1086 return NULL;
1087 return (unsigned long *)m->private + *pos;
1088}
1089
1090static int vmstat_show(struct seq_file *m, void *arg)
1091{
1092 unsigned long *l = arg;
1093 unsigned long off = l - (unsigned long *)m->private;
1094
1095 seq_printf(m, "%s %lu\n", vmstat_text[off], *l);
1096 return 0;
1097}
1098
1099static void vmstat_stop(struct seq_file *m, void *arg)
1100{
1101 kfree(m->private);
1102 m->private = NULL;
1103}
1104
Alexey Dobriyanb6aa44a2008-10-06 04:17:48 +04001105static const struct seq_operations vmstat_op = {
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001106 .start = vmstat_start,
1107 .next = vmstat_next,
1108 .stop = vmstat_stop,
1109 .show = vmstat_show,
1110};
1111
Alexey Dobriyanb6aa44a2008-10-06 04:17:48 +04001112static int vmstat_open(struct inode *inode, struct file *file)
1113{
1114 return seq_open(file, &vmstat_op);
1115}
1116
1117static const struct file_operations proc_vmstat_file_operations = {
1118 .open = vmstat_open,
1119 .read = seq_read,
1120 .llseek = seq_lseek,
1121 .release = seq_release,
1122};
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001123#endif /* CONFIG_PROC_FS */
1124
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001125#ifdef CONFIG_SMP
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001126static DEFINE_PER_CPU(struct delayed_work, vmstat_work);
Christoph Lameter77461ab2007-05-09 02:35:13 -07001127int sysctl_stat_interval __read_mostly = HZ;
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001128
1129static void vmstat_update(struct work_struct *w)
1130{
1131 refresh_cpu_vm_stats(smp_processor_id());
Christoph Lameter77461ab2007-05-09 02:35:13 -07001132 schedule_delayed_work(&__get_cpu_var(vmstat_work),
Anton Blanchard98f4ebb2009-04-02 16:56:39 -07001133 round_jiffies_relative(sysctl_stat_interval));
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001134}
1135
Randy Dunlap42614fc2007-11-14 17:00:12 -08001136static void __cpuinit start_cpu_timer(int cpu)
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001137{
Tejun Heo1871e522009-10-29 22:34:13 +09001138 struct delayed_work *work = &per_cpu(vmstat_work, cpu);
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001139
Tejun Heo1871e522009-10-29 22:34:13 +09001140 INIT_DELAYED_WORK_DEFERRABLE(work, vmstat_update);
1141 schedule_delayed_work_on(cpu, work, __round_jiffies_relative(HZ, cpu));
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001142}
1143
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001144/*
1145 * Use the cpu notifier to insure that the thresholds are recalculated
1146 * when necessary.
1147 */
1148static int __cpuinit vmstat_cpuup_callback(struct notifier_block *nfb,
1149 unsigned long action,
1150 void *hcpu)
1151{
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001152 long cpu = (long)hcpu;
1153
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001154 switch (action) {
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001155 case CPU_ONLINE:
1156 case CPU_ONLINE_FROZEN:
KAMEZAWA Hiroyuki5ee28a42010-09-09 16:38:14 -07001157 refresh_zone_stat_thresholds();
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001158 start_cpu_timer(cpu);
Christoph Lameterad596922010-01-05 15:34:51 +09001159 node_set_state(cpu_to_node(cpu), N_CPU);
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001160 break;
1161 case CPU_DOWN_PREPARE:
1162 case CPU_DOWN_PREPARE_FROZEN:
Tejun Heoafe2c512010-12-14 16:21:17 +01001163 cancel_delayed_work_sync(&per_cpu(vmstat_work, cpu));
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001164 per_cpu(vmstat_work, cpu).work.func = NULL;
1165 break;
1166 case CPU_DOWN_FAILED:
1167 case CPU_DOWN_FAILED_FROZEN:
1168 start_cpu_timer(cpu);
1169 break;
Andy Whitcroftce421c72006-12-06 20:33:08 -08001170 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07001171 case CPU_DEAD_FROZEN:
Andy Whitcroftce421c72006-12-06 20:33:08 -08001172 refresh_zone_stat_thresholds();
1173 break;
1174 default:
1175 break;
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001176 }
1177 return NOTIFY_OK;
1178}
1179
1180static struct notifier_block __cpuinitdata vmstat_notifier =
1181 { &vmstat_cpuup_callback, NULL, 0 };
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +04001182#endif
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001183
Adrian Bunke2fc88d2007-10-16 01:26:27 -07001184static int __init setup_vmstat(void)
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001185{
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +04001186#ifdef CONFIG_SMP
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001187 int cpu;
1188
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001189 refresh_zone_stat_thresholds();
1190 register_cpu_notifier(&vmstat_notifier);
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001191
1192 for_each_online_cpu(cpu)
1193 start_cpu_timer(cpu);
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +04001194#endif
1195#ifdef CONFIG_PROC_FS
1196 proc_create("buddyinfo", S_IRUGO, NULL, &fragmentation_file_operations);
Alexey Dobriyan74e2e8e2008-10-06 04:15:36 +04001197 proc_create("pagetypeinfo", S_IRUGO, NULL, &pagetypeinfo_file_ops);
Alexey Dobriyanb6aa44a2008-10-06 04:17:48 +04001198 proc_create("vmstat", S_IRUGO, NULL, &proc_vmstat_file_operations);
Alexey Dobriyan5c9fe622008-10-06 04:19:42 +04001199 proc_create("zoneinfo", S_IRUGO, NULL, &proc_zoneinfo_file_operations);
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +04001200#endif
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001201 return 0;
1202}
1203module_init(setup_vmstat)
Mel Gormand7a57522010-05-24 14:32:25 -07001204
1205#if defined(CONFIG_DEBUG_FS) && defined(CONFIG_COMPACTION)
1206#include <linux/debugfs.h>
1207
1208static struct dentry *extfrag_debug_root;
1209
1210/*
1211 * Return an index indicating how much of the available free memory is
1212 * unusable for an allocation of the requested size.
1213 */
1214static int unusable_free_index(unsigned int order,
1215 struct contig_page_info *info)
1216{
1217 /* No free memory is interpreted as all free memory is unusable */
1218 if (info->free_pages == 0)
1219 return 1000;
1220
1221 /*
1222 * Index should be a value between 0 and 1. Return a value to 3
1223 * decimal places.
1224 *
1225 * 0 => no fragmentation
1226 * 1 => high fragmentation
1227 */
1228 return div_u64((info->free_pages - (info->free_blocks_suitable << order)) * 1000ULL, info->free_pages);
1229
1230}
1231
1232static void unusable_show_print(struct seq_file *m,
1233 pg_data_t *pgdat, struct zone *zone)
1234{
1235 unsigned int order;
1236 int index;
1237 struct contig_page_info info;
1238
1239 seq_printf(m, "Node %d, zone %8s ",
1240 pgdat->node_id,
1241 zone->name);
1242 for (order = 0; order < MAX_ORDER; ++order) {
1243 fill_contig_page_info(zone, order, &info);
1244 index = unusable_free_index(order, &info);
1245 seq_printf(m, "%d.%03d ", index / 1000, index % 1000);
1246 }
1247
1248 seq_putc(m, '\n');
1249}
1250
1251/*
1252 * Display unusable free space index
1253 *
1254 * The unusable free space index measures how much of the available free
1255 * memory cannot be used to satisfy an allocation of a given size and is a
1256 * value between 0 and 1. The higher the value, the more of free memory is
1257 * unusable and by implication, the worse the external fragmentation is. This
1258 * can be expressed as a percentage by multiplying by 100.
1259 */
1260static int unusable_show(struct seq_file *m, void *arg)
1261{
1262 pg_data_t *pgdat = (pg_data_t *)arg;
1263
1264 /* check memoryless node */
1265 if (!node_state(pgdat->node_id, N_HIGH_MEMORY))
1266 return 0;
1267
1268 walk_zones_in_node(m, pgdat, unusable_show_print);
1269
1270 return 0;
1271}
1272
1273static const struct seq_operations unusable_op = {
1274 .start = frag_start,
1275 .next = frag_next,
1276 .stop = frag_stop,
1277 .show = unusable_show,
1278};
1279
1280static int unusable_open(struct inode *inode, struct file *file)
1281{
1282 return seq_open(file, &unusable_op);
1283}
1284
1285static const struct file_operations unusable_file_ops = {
1286 .open = unusable_open,
1287 .read = seq_read,
1288 .llseek = seq_lseek,
1289 .release = seq_release,
1290};
1291
Mel Gormanf1a5ab12010-05-24 14:32:26 -07001292static void extfrag_show_print(struct seq_file *m,
1293 pg_data_t *pgdat, struct zone *zone)
1294{
1295 unsigned int order;
1296 int index;
1297
1298 /* Alloc on stack as interrupts are disabled for zone walk */
1299 struct contig_page_info info;
1300
1301 seq_printf(m, "Node %d, zone %8s ",
1302 pgdat->node_id,
1303 zone->name);
1304 for (order = 0; order < MAX_ORDER; ++order) {
1305 fill_contig_page_info(zone, order, &info);
Mel Gorman56de7262010-05-24 14:32:30 -07001306 index = __fragmentation_index(order, &info);
Mel Gormanf1a5ab12010-05-24 14:32:26 -07001307 seq_printf(m, "%d.%03d ", index / 1000, index % 1000);
1308 }
1309
1310 seq_putc(m, '\n');
1311}
1312
1313/*
1314 * Display fragmentation index for orders that allocations would fail for
1315 */
1316static int extfrag_show(struct seq_file *m, void *arg)
1317{
1318 pg_data_t *pgdat = (pg_data_t *)arg;
1319
1320 walk_zones_in_node(m, pgdat, extfrag_show_print);
1321
1322 return 0;
1323}
1324
1325static const struct seq_operations extfrag_op = {
1326 .start = frag_start,
1327 .next = frag_next,
1328 .stop = frag_stop,
1329 .show = extfrag_show,
1330};
1331
1332static int extfrag_open(struct inode *inode, struct file *file)
1333{
1334 return seq_open(file, &extfrag_op);
1335}
1336
1337static const struct file_operations extfrag_file_ops = {
1338 .open = extfrag_open,
1339 .read = seq_read,
1340 .llseek = seq_lseek,
1341 .release = seq_release,
1342};
1343
Mel Gormand7a57522010-05-24 14:32:25 -07001344static int __init extfrag_debug_init(void)
1345{
1346 extfrag_debug_root = debugfs_create_dir("extfrag", NULL);
1347 if (!extfrag_debug_root)
1348 return -ENOMEM;
1349
1350 if (!debugfs_create_file("unusable_index", 0444,
1351 extfrag_debug_root, NULL, &unusable_file_ops))
1352 return -ENOMEM;
1353
Mel Gormanf1a5ab12010-05-24 14:32:26 -07001354 if (!debugfs_create_file("extfrag_index", 0444,
1355 extfrag_debug_root, NULL, &extfrag_file_ops))
1356 return -ENOMEM;
1357
Mel Gormand7a57522010-05-24 14:32:25 -07001358 return 0;
1359}
1360
1361module_init(extfrag_debug_init);
1362#endif