blob: c7370579111b872943c56bd694dc0bd02121092d [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 */
Dimitri Sivanicha1cb2c62011-10-31 17:09:46 -070081atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS] __cacheline_aligned_in_smp;
Christoph Lameter2244b952006-06-30 01:55:33 -070082EXPORT_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 */
KOSAKI Motohiroa6cccdc2011-05-24 17:11:33 -0700160void 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
Heiko Carstens41561532012-01-12 17:17:30 -0800298#ifdef CONFIG_HAVE_CMPXCHG_LOCAL
Christoph Lameter7c839122010-12-14 10:28:46 -0600299/*
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
Christoph Lameterd3bc2362011-04-14 15:21:58 -0700324 * rescheduled while executing here. However, the next
325 * counter update will apply the threshold again and
326 * therefore bring the counter under the threshold again.
327 *
328 * Most of the time the thresholds are the same anyways
329 * for all cpus in a zone.
Christoph Lameter7c839122010-12-14 10:28:46 -0600330 */
331 t = this_cpu_read(pcp->stat_threshold);
332
333 o = this_cpu_read(*p);
334 n = delta + o;
335
336 if (n > t || n < -t) {
337 int os = overstep_mode * (t >> 1) ;
338
339 /* Overflow must be added to zone counters */
340 z = n + os;
341 n = -os;
342 }
343 } while (this_cpu_cmpxchg(*p, o, n) != o);
344
345 if (z)
346 zone_page_state_add(z, zone, item);
347}
348
349void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
350 int delta)
351{
352 mod_state(zone, item, delta, 0);
353}
354EXPORT_SYMBOL(mod_zone_page_state);
355
356void inc_zone_state(struct zone *zone, enum zone_stat_item item)
357{
358 mod_state(zone, item, 1, 1);
359}
360
361void inc_zone_page_state(struct page *page, enum zone_stat_item item)
362{
363 mod_state(page_zone(page), item, 1, 1);
364}
365EXPORT_SYMBOL(inc_zone_page_state);
366
367void dec_zone_page_state(struct page *page, enum zone_stat_item item)
368{
369 mod_state(page_zone(page), item, -1, -1);
370}
371EXPORT_SYMBOL(dec_zone_page_state);
372#else
373/*
374 * Use interrupt disable to serialize counter updates
375 */
376void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
377 int delta)
378{
379 unsigned long flags;
380
381 local_irq_save(flags);
382 __mod_zone_page_state(zone, item, delta);
383 local_irq_restore(flags);
384}
385EXPORT_SYMBOL(mod_zone_page_state);
386
Christoph Lameterca889e62006-06-30 01:55:44 -0700387void inc_zone_state(struct zone *zone, enum zone_stat_item item)
388{
389 unsigned long flags;
390
391 local_irq_save(flags);
392 __inc_zone_state(zone, item);
393 local_irq_restore(flags);
394}
395
Christoph Lameter2244b952006-06-30 01:55:33 -0700396void inc_zone_page_state(struct page *page, enum zone_stat_item item)
397{
398 unsigned long flags;
399 struct zone *zone;
Christoph Lameter2244b952006-06-30 01:55:33 -0700400
401 zone = page_zone(page);
402 local_irq_save(flags);
Christoph Lameterca889e62006-06-30 01:55:44 -0700403 __inc_zone_state(zone, item);
Christoph Lameter2244b952006-06-30 01:55:33 -0700404 local_irq_restore(flags);
405}
406EXPORT_SYMBOL(inc_zone_page_state);
407
408void dec_zone_page_state(struct page *page, enum zone_stat_item item)
409{
410 unsigned long flags;
Christoph Lameter2244b952006-06-30 01:55:33 -0700411
Christoph Lameter2244b952006-06-30 01:55:33 -0700412 local_irq_save(flags);
Christoph Lametera302eb42006-08-31 21:27:34 -0700413 __dec_zone_page_state(page, item);
Christoph Lameter2244b952006-06-30 01:55:33 -0700414 local_irq_restore(flags);
415}
416EXPORT_SYMBOL(dec_zone_page_state);
Christoph Lameter7c839122010-12-14 10:28:46 -0600417#endif
Christoph Lameter2244b952006-06-30 01:55:33 -0700418
419/*
420 * Update the zone counters for one cpu.
Christoph Lameter4037d452007-05-09 02:35:14 -0700421 *
Christoph Lametera7f75e22008-02-04 22:29:16 -0800422 * The cpu specified must be either the current cpu or a processor that
423 * is not online. If it is the current cpu then the execution thread must
424 * be pinned to the current cpu.
425 *
Christoph Lameter4037d452007-05-09 02:35:14 -0700426 * Note that refresh_cpu_vm_stats strives to only access
427 * node local memory. The per cpu pagesets on remote zones are placed
428 * in the memory local to the processor using that pageset. So the
429 * loop over all zones will access a series of cachelines local to
430 * the processor.
431 *
432 * The call to zone_page_state_add updates the cachelines with the
433 * statistics in the remote zone struct as well as the global cachelines
434 * with the global counters. These could cause remote node cache line
435 * bouncing and will have to be only done when necessary.
Christoph Lameter2244b952006-06-30 01:55:33 -0700436 */
437void refresh_cpu_vm_stats(int cpu)
438{
439 struct zone *zone;
440 int i;
Christoph Lametera7f75e22008-02-04 22:29:16 -0800441 int global_diff[NR_VM_ZONE_STAT_ITEMS] = { 0, };
Christoph Lameter2244b952006-06-30 01:55:33 -0700442
KOSAKI Motohiroee99c712009-03-31 15:19:31 -0700443 for_each_populated_zone(zone) {
Christoph Lameter4037d452007-05-09 02:35:14 -0700444 struct per_cpu_pageset *p;
Christoph Lameter2244b952006-06-30 01:55:33 -0700445
Christoph Lameter99dcc3e2010-01-05 15:34:51 +0900446 p = per_cpu_ptr(zone->pageset, cpu);
Christoph Lameter2244b952006-06-30 01:55:33 -0700447
448 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
Christoph Lameter4037d452007-05-09 02:35:14 -0700449 if (p->vm_stat_diff[i]) {
Christoph Lametera7f75e22008-02-04 22:29:16 -0800450 unsigned long flags;
451 int v;
452
Christoph Lameter2244b952006-06-30 01:55:33 -0700453 local_irq_save(flags);
Christoph Lametera7f75e22008-02-04 22:29:16 -0800454 v = p->vm_stat_diff[i];
Christoph Lameter4037d452007-05-09 02:35:14 -0700455 p->vm_stat_diff[i] = 0;
Christoph Lametera7f75e22008-02-04 22:29:16 -0800456 local_irq_restore(flags);
457 atomic_long_add(v, &zone->vm_stat[i]);
458 global_diff[i] += v;
Christoph Lameter4037d452007-05-09 02:35:14 -0700459#ifdef CONFIG_NUMA
460 /* 3 seconds idle till flush */
461 p->expire = 3;
462#endif
Christoph Lameter2244b952006-06-30 01:55:33 -0700463 }
Dimitri Sivanich468fd622008-04-28 02:13:37 -0700464 cond_resched();
Christoph Lameter4037d452007-05-09 02:35:14 -0700465#ifdef CONFIG_NUMA
466 /*
467 * Deal with draining the remote pageset of this
468 * processor
469 *
470 * Check if there are pages remaining in this pageset
471 * if not then there is nothing to expire.
472 */
Christoph Lameter3dfa5722008-02-04 22:29:19 -0800473 if (!p->expire || !p->pcp.count)
Christoph Lameter4037d452007-05-09 02:35:14 -0700474 continue;
475
476 /*
477 * We never drain zones local to this processor.
478 */
479 if (zone_to_nid(zone) == numa_node_id()) {
480 p->expire = 0;
481 continue;
482 }
483
484 p->expire--;
485 if (p->expire)
486 continue;
487
Christoph Lameter3dfa5722008-02-04 22:29:19 -0800488 if (p->pcp.count)
489 drain_zone_pages(zone, &p->pcp);
Christoph Lameter4037d452007-05-09 02:35:14 -0700490#endif
Christoph Lameter2244b952006-06-30 01:55:33 -0700491 }
Christoph Lametera7f75e22008-02-04 22:29:16 -0800492
493 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
494 if (global_diff[i])
495 atomic_long_add(global_diff[i], &vm_stat[i]);
Christoph Lameter2244b952006-06-30 01:55:33 -0700496}
497
Minchan Kim5a883812012-10-08 16:33:39 -0700498void drain_zonestat(struct zone *zone, struct per_cpu_pageset *pset)
499{
500 int i;
501
502 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
503 if (pset->vm_stat_diff[i]) {
504 int v = pset->vm_stat_diff[i];
505 pset->vm_stat_diff[i] = 0;
506 atomic_long_add(v, &zone->vm_stat[i]);
507 atomic_long_add(v, &vm_stat[i]);
508 }
509}
Christoph Lameter2244b952006-06-30 01:55:33 -0700510#endif
511
Christoph Lameterca889e62006-06-30 01:55:44 -0700512#ifdef CONFIG_NUMA
513/*
514 * zonelist = the list of zones passed to the allocator
515 * z = the zone from which the allocation occurred.
516 *
517 * Must be called with interrupts disabled.
Andi Kleen78afd562011-03-22 16:33:12 -0700518 *
519 * When __GFP_OTHER_NODE is set assume the node of the preferred
520 * zone is the local node. This is useful for daemons who allocate
521 * memory on behalf of other processes.
Christoph Lameterca889e62006-06-30 01:55:44 -0700522 */
Andi Kleen78afd562011-03-22 16:33:12 -0700523void zone_statistics(struct zone *preferred_zone, struct zone *z, gfp_t flags)
Christoph Lameterca889e62006-06-30 01:55:44 -0700524{
Mel Gorman18ea7e72008-04-28 02:12:14 -0700525 if (z->zone_pgdat == preferred_zone->zone_pgdat) {
Christoph Lameterca889e62006-06-30 01:55:44 -0700526 __inc_zone_state(z, NUMA_HIT);
527 } else {
528 __inc_zone_state(z, NUMA_MISS);
Mel Gorman18ea7e72008-04-28 02:12:14 -0700529 __inc_zone_state(preferred_zone, NUMA_FOREIGN);
Christoph Lameterca889e62006-06-30 01:55:44 -0700530 }
Andi Kleen78afd562011-03-22 16:33:12 -0700531 if (z->node == ((flags & __GFP_OTHER_NODE) ?
532 preferred_zone->node : numa_node_id()))
Christoph Lameterca889e62006-06-30 01:55:44 -0700533 __inc_zone_state(z, NUMA_LOCAL);
534 else
535 __inc_zone_state(z, NUMA_OTHER);
536}
537#endif
538
Mel Gormand7a57522010-05-24 14:32:25 -0700539#ifdef CONFIG_COMPACTION
Namhyung Kim36deb0b2010-10-26 14:22:04 -0700540
Mel Gormand7a57522010-05-24 14:32:25 -0700541struct contig_page_info {
542 unsigned long free_pages;
543 unsigned long free_blocks_total;
544 unsigned long free_blocks_suitable;
545};
546
547/*
548 * Calculate the number of free pages in a zone, how many contiguous
549 * pages are free and how many are large enough to satisfy an allocation of
550 * the target size. Note that this function makes no attempt to estimate
551 * how many suitable free blocks there *might* be if MOVABLE pages were
552 * migrated. Calculating that is possible, but expensive and can be
553 * figured out from userspace
554 */
555static void fill_contig_page_info(struct zone *zone,
556 unsigned int suitable_order,
557 struct contig_page_info *info)
558{
559 unsigned int order;
560
561 info->free_pages = 0;
562 info->free_blocks_total = 0;
563 info->free_blocks_suitable = 0;
564
565 for (order = 0; order < MAX_ORDER; order++) {
566 unsigned long blocks;
567
568 /* Count number of free blocks */
569 blocks = zone->free_area[order].nr_free;
570 info->free_blocks_total += blocks;
571
572 /* Count free base pages */
573 info->free_pages += blocks << order;
574
575 /* Count the suitable free blocks */
576 if (order >= suitable_order)
577 info->free_blocks_suitable += blocks <<
578 (order - suitable_order);
579 }
580}
Mel Gormanf1a5ab12010-05-24 14:32:26 -0700581
582/*
583 * A fragmentation index only makes sense if an allocation of a requested
584 * size would fail. If that is true, the fragmentation index indicates
585 * whether external fragmentation or a lack of memory was the problem.
586 * The value can be used to determine if page reclaim or compaction
587 * should be used
588 */
Mel Gorman56de7262010-05-24 14:32:30 -0700589static int __fragmentation_index(unsigned int order, struct contig_page_info *info)
Mel Gormanf1a5ab12010-05-24 14:32:26 -0700590{
591 unsigned long requested = 1UL << order;
592
593 if (!info->free_blocks_total)
594 return 0;
595
596 /* Fragmentation index only makes sense when a request would fail */
597 if (info->free_blocks_suitable)
598 return -1000;
599
600 /*
601 * Index is between 0 and 1 so return within 3 decimal places
602 *
603 * 0 => allocation would fail due to lack of memory
604 * 1 => allocation would fail due to fragmentation
605 */
606 return 1000 - div_u64( (1000+(div_u64(info->free_pages * 1000ULL, requested))), info->free_blocks_total);
607}
Mel Gorman56de7262010-05-24 14:32:30 -0700608
609/* Same as __fragmentation index but allocs contig_page_info on stack */
610int fragmentation_index(struct zone *zone, unsigned int order)
611{
612 struct contig_page_info info;
613
614 fill_contig_page_info(zone, order, &info);
615 return __fragmentation_index(order, &info);
616}
Mel Gormand7a57522010-05-24 14:32:25 -0700617#endif
618
619#if defined(CONFIG_PROC_FS) || defined(CONFIG_COMPACTION)
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +0400620#include <linux/proc_fs.h>
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700621#include <linux/seq_file.h>
622
Mel Gorman467c9962007-10-16 01:26:02 -0700623static char * const migratetype_names[MIGRATE_TYPES] = {
624 "Unmovable",
625 "Reclaimable",
626 "Movable",
627 "Reserve",
Michal Nazarewicz47118af2011-12-29 13:09:50 +0100628#ifdef CONFIG_CMA
629 "CMA",
630#endif
KOSAKI Motohiro91446b02008-04-15 14:34:42 -0700631 "Isolate",
Mel Gorman467c9962007-10-16 01:26:02 -0700632};
633
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700634static void *frag_start(struct seq_file *m, loff_t *pos)
635{
636 pg_data_t *pgdat;
637 loff_t node = *pos;
638 for (pgdat = first_online_pgdat();
639 pgdat && node;
640 pgdat = next_online_pgdat(pgdat))
641 --node;
642
643 return pgdat;
644}
645
646static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
647{
648 pg_data_t *pgdat = (pg_data_t *)arg;
649
650 (*pos)++;
651 return next_online_pgdat(pgdat);
652}
653
654static void frag_stop(struct seq_file *m, void *arg)
655{
656}
657
Mel Gorman467c9962007-10-16 01:26:02 -0700658/* Walk all the zones in a node and print using a callback */
659static void walk_zones_in_node(struct seq_file *m, pg_data_t *pgdat,
660 void (*print)(struct seq_file *m, pg_data_t *, struct zone *))
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700661{
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700662 struct zone *zone;
663 struct zone *node_zones = pgdat->node_zones;
664 unsigned long flags;
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700665
666 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
667 if (!populated_zone(zone))
668 continue;
669
670 spin_lock_irqsave(&zone->lock, flags);
Mel Gorman467c9962007-10-16 01:26:02 -0700671 print(m, pgdat, zone);
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700672 spin_unlock_irqrestore(&zone->lock, flags);
Mel Gorman467c9962007-10-16 01:26:02 -0700673 }
674}
Mel Gormand7a57522010-05-24 14:32:25 -0700675#endif
Mel Gorman467c9962007-10-16 01:26:02 -0700676
David Rientjes0d6617c2011-09-14 16:21:05 -0700677#if defined(CONFIG_PROC_FS) || defined(CONFIG_SYSFS) || defined(CONFIG_NUMA)
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700678#ifdef CONFIG_ZONE_DMA
679#define TEXT_FOR_DMA(xx) xx "_dma",
680#else
681#define TEXT_FOR_DMA(xx)
682#endif
683
684#ifdef CONFIG_ZONE_DMA32
685#define TEXT_FOR_DMA32(xx) xx "_dma32",
686#else
687#define TEXT_FOR_DMA32(xx)
688#endif
689
690#ifdef CONFIG_HIGHMEM
691#define TEXT_FOR_HIGHMEM(xx) xx "_high",
692#else
693#define TEXT_FOR_HIGHMEM(xx)
694#endif
695
696#define TEXTS_FOR_ZONES(xx) TEXT_FOR_DMA(xx) TEXT_FOR_DMA32(xx) xx "_normal", \
697 TEXT_FOR_HIGHMEM(xx) xx "_movable",
698
699const char * const vmstat_text[] = {
700 /* Zoned VM counters */
701 "nr_free_pages",
702 "nr_inactive_anon",
703 "nr_active_anon",
704 "nr_inactive_file",
705 "nr_active_file",
706 "nr_unevictable",
707 "nr_mlock",
708 "nr_anon_pages",
709 "nr_mapped",
710 "nr_file_pages",
711 "nr_dirty",
712 "nr_writeback",
713 "nr_slab_reclaimable",
714 "nr_slab_unreclaimable",
715 "nr_page_table_pages",
716 "nr_kernel_stack",
717 "nr_unstable",
718 "nr_bounce",
719 "nr_vmscan_write",
Mel Gorman49ea7eb2011-10-31 17:07:59 -0700720 "nr_vmscan_immediate_reclaim",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700721 "nr_writeback_temp",
722 "nr_isolated_anon",
723 "nr_isolated_file",
724 "nr_shmem",
725 "nr_dirtied",
726 "nr_written",
727
728#ifdef CONFIG_NUMA
729 "numa_hit",
730 "numa_miss",
731 "numa_foreign",
732 "numa_interleave",
733 "numa_local",
734 "numa_other",
735#endif
736 "nr_anon_transparent_hugepages",
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -0700737 "nr_free_cma",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700738 "nr_dirty_threshold",
739 "nr_dirty_background_threshold",
740
741#ifdef CONFIG_VM_EVENT_COUNTERS
742 "pgpgin",
743 "pgpgout",
744 "pswpin",
745 "pswpout",
746
747 TEXTS_FOR_ZONES("pgalloc")
748
749 "pgfree",
750 "pgactivate",
751 "pgdeactivate",
752
753 "pgfault",
754 "pgmajfault",
755
756 TEXTS_FOR_ZONES("pgrefill")
Ying Han904249a2012-04-25 16:01:48 -0700757 TEXTS_FOR_ZONES("pgsteal_kswapd")
758 TEXTS_FOR_ZONES("pgsteal_direct")
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700759 TEXTS_FOR_ZONES("pgscan_kswapd")
760 TEXTS_FOR_ZONES("pgscan_direct")
Mel Gorman68243e72012-07-31 16:44:39 -0700761 "pgscan_direct_throttle",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700762
763#ifdef CONFIG_NUMA
764 "zone_reclaim_failed",
765#endif
766 "pginodesteal",
767 "slabs_scanned",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700768 "kswapd_inodesteal",
769 "kswapd_low_wmark_hit_quickly",
770 "kswapd_high_wmark_hit_quickly",
771 "kswapd_skip_congestion_wait",
772 "pageoutrun",
773 "allocstall",
774
775 "pgrotated",
776
777#ifdef CONFIG_COMPACTION
778 "compact_blocks_moved",
779 "compact_pages_moved",
780 "compact_pagemigrate_failed",
781 "compact_stall",
782 "compact_fail",
783 "compact_success",
784#endif
785
786#ifdef CONFIG_HUGETLB_PAGE
787 "htlb_buddy_alloc_success",
788 "htlb_buddy_alloc_fail",
789#endif
790 "unevictable_pgs_culled",
791 "unevictable_pgs_scanned",
792 "unevictable_pgs_rescued",
793 "unevictable_pgs_mlocked",
794 "unevictable_pgs_munlocked",
795 "unevictable_pgs_cleared",
796 "unevictable_pgs_stranded",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700797
798#ifdef CONFIG_TRANSPARENT_HUGEPAGE
799 "thp_fault_alloc",
800 "thp_fault_fallback",
801 "thp_collapse_alloc",
802 "thp_collapse_alloc_failed",
803 "thp_split",
804#endif
805
806#endif /* CONFIG_VM_EVENTS_COUNTERS */
807};
David Rientjes0d6617c2011-09-14 16:21:05 -0700808#endif /* CONFIG_PROC_FS || CONFIG_SYSFS || CONFIG_NUMA */
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700809
810
Mel Gormand7a57522010-05-24 14:32:25 -0700811#ifdef CONFIG_PROC_FS
Mel Gorman467c9962007-10-16 01:26:02 -0700812static void frag_show_print(struct seq_file *m, pg_data_t *pgdat,
813 struct zone *zone)
814{
815 int order;
816
817 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
818 for (order = 0; order < MAX_ORDER; ++order)
819 seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
820 seq_putc(m, '\n');
821}
822
823/*
824 * This walks the free areas for each zone.
825 */
826static int frag_show(struct seq_file *m, void *arg)
827{
828 pg_data_t *pgdat = (pg_data_t *)arg;
829 walk_zones_in_node(m, pgdat, frag_show_print);
830 return 0;
831}
832
833static void pagetypeinfo_showfree_print(struct seq_file *m,
834 pg_data_t *pgdat, struct zone *zone)
835{
836 int order, mtype;
837
838 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++) {
839 seq_printf(m, "Node %4d, zone %8s, type %12s ",
840 pgdat->node_id,
841 zone->name,
842 migratetype_names[mtype]);
843 for (order = 0; order < MAX_ORDER; ++order) {
844 unsigned long freecount = 0;
845 struct free_area *area;
846 struct list_head *curr;
847
848 area = &(zone->free_area[order]);
849
850 list_for_each(curr, &area->free_list[mtype])
851 freecount++;
852 seq_printf(m, "%6lu ", freecount);
853 }
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700854 seq_putc(m, '\n');
855 }
Mel Gorman467c9962007-10-16 01:26:02 -0700856}
857
858/* Print out the free pages at each order for each migatetype */
859static int pagetypeinfo_showfree(struct seq_file *m, void *arg)
860{
861 int order;
862 pg_data_t *pgdat = (pg_data_t *)arg;
863
864 /* Print header */
865 seq_printf(m, "%-43s ", "Free pages count per migrate type at order");
866 for (order = 0; order < MAX_ORDER; ++order)
867 seq_printf(m, "%6d ", order);
868 seq_putc(m, '\n');
869
870 walk_zones_in_node(m, pgdat, pagetypeinfo_showfree_print);
871
872 return 0;
873}
874
875static void pagetypeinfo_showblockcount_print(struct seq_file *m,
876 pg_data_t *pgdat, struct zone *zone)
877{
878 int mtype;
879 unsigned long pfn;
880 unsigned long start_pfn = zone->zone_start_pfn;
881 unsigned long end_pfn = start_pfn + zone->spanned_pages;
882 unsigned long count[MIGRATE_TYPES] = { 0, };
883
884 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
885 struct page *page;
886
887 if (!pfn_valid(pfn))
888 continue;
889
890 page = pfn_to_page(pfn);
Mel Gormaneb335752009-05-13 17:34:48 +0100891
892 /* Watch for unexpected holes punched in the memmap */
893 if (!memmap_valid_within(pfn, page, zone))
Mel Gormane80d6a22008-08-14 11:10:14 +0100894 continue;
Mel Gormaneb335752009-05-13 17:34:48 +0100895
Mel Gorman467c9962007-10-16 01:26:02 -0700896 mtype = get_pageblock_migratetype(page);
897
Mel Gormane80d6a22008-08-14 11:10:14 +0100898 if (mtype < MIGRATE_TYPES)
899 count[mtype]++;
Mel Gorman467c9962007-10-16 01:26:02 -0700900 }
901
902 /* Print counts */
903 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
904 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
905 seq_printf(m, "%12lu ", count[mtype]);
906 seq_putc(m, '\n');
907}
908
909/* Print out the free pages at each order for each migratetype */
910static int pagetypeinfo_showblockcount(struct seq_file *m, void *arg)
911{
912 int mtype;
913 pg_data_t *pgdat = (pg_data_t *)arg;
914
915 seq_printf(m, "\n%-23s", "Number of blocks type ");
916 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
917 seq_printf(m, "%12s ", migratetype_names[mtype]);
918 seq_putc(m, '\n');
919 walk_zones_in_node(m, pgdat, pagetypeinfo_showblockcount_print);
920
921 return 0;
922}
923
924/*
925 * This prints out statistics in relation to grouping pages by mobility.
926 * It is expensive to collect so do not constantly read the file.
927 */
928static int pagetypeinfo_show(struct seq_file *m, void *arg)
929{
930 pg_data_t *pgdat = (pg_data_t *)arg;
931
KOSAKI Motohiro41b25a32008-04-30 00:52:13 -0700932 /* check memoryless node */
933 if (!node_state(pgdat->node_id, N_HIGH_MEMORY))
934 return 0;
935
Mel Gorman467c9962007-10-16 01:26:02 -0700936 seq_printf(m, "Page block order: %d\n", pageblock_order);
937 seq_printf(m, "Pages per block: %lu\n", pageblock_nr_pages);
938 seq_putc(m, '\n');
939 pagetypeinfo_showfree(m, pgdat);
940 pagetypeinfo_showblockcount(m, pgdat);
941
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700942 return 0;
943}
944
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +0400945static const struct seq_operations fragmentation_op = {
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700946 .start = frag_start,
947 .next = frag_next,
948 .stop = frag_stop,
949 .show = frag_show,
950};
951
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +0400952static int fragmentation_open(struct inode *inode, struct file *file)
953{
954 return seq_open(file, &fragmentation_op);
955}
956
957static const struct file_operations fragmentation_file_operations = {
958 .open = fragmentation_open,
959 .read = seq_read,
960 .llseek = seq_lseek,
961 .release = seq_release,
962};
963
Alexey Dobriyan74e2e8e2008-10-06 04:15:36 +0400964static const struct seq_operations pagetypeinfo_op = {
Mel Gorman467c9962007-10-16 01:26:02 -0700965 .start = frag_start,
966 .next = frag_next,
967 .stop = frag_stop,
968 .show = pagetypeinfo_show,
969};
970
Alexey Dobriyan74e2e8e2008-10-06 04:15:36 +0400971static int pagetypeinfo_open(struct inode *inode, struct file *file)
972{
973 return seq_open(file, &pagetypeinfo_op);
974}
975
976static const struct file_operations pagetypeinfo_file_ops = {
977 .open = pagetypeinfo_open,
978 .read = seq_read,
979 .llseek = seq_lseek,
980 .release = seq_release,
981};
982
Mel Gorman467c9962007-10-16 01:26:02 -0700983static void zoneinfo_show_print(struct seq_file *m, pg_data_t *pgdat,
984 struct zone *zone)
985{
986 int i;
987 seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
988 seq_printf(m,
989 "\n pages free %lu"
990 "\n min %lu"
991 "\n low %lu"
992 "\n high %lu"
Wu Fengguang08d9ae72009-06-16 15:32:30 -0700993 "\n scanned %lu"
Mel Gorman467c9962007-10-16 01:26:02 -0700994 "\n spanned %lu"
995 "\n present %lu",
Mel Gorman88f5acf2011-01-13 15:45:41 -0800996 zone_page_state(zone, NR_FREE_PAGES),
Mel Gorman41858962009-06-16 15:32:12 -0700997 min_wmark_pages(zone),
998 low_wmark_pages(zone),
999 high_wmark_pages(zone),
Mel Gorman467c9962007-10-16 01:26:02 -07001000 zone->pages_scanned,
Mel Gorman467c9962007-10-16 01:26:02 -07001001 zone->spanned_pages,
1002 zone->present_pages);
1003
1004 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
1005 seq_printf(m, "\n %-12s %lu", vmstat_text[i],
1006 zone_page_state(zone, i));
1007
1008 seq_printf(m,
1009 "\n protection: (%lu",
1010 zone->lowmem_reserve[0]);
1011 for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
1012 seq_printf(m, ", %lu", zone->lowmem_reserve[i]);
1013 seq_printf(m,
1014 ")"
1015 "\n pagesets");
1016 for_each_online_cpu(i) {
1017 struct per_cpu_pageset *pageset;
Mel Gorman467c9962007-10-16 01:26:02 -07001018
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001019 pageset = per_cpu_ptr(zone->pageset, i);
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001020 seq_printf(m,
1021 "\n cpu: %i"
1022 "\n count: %i"
1023 "\n high: %i"
1024 "\n batch: %i",
1025 i,
1026 pageset->pcp.count,
1027 pageset->pcp.high,
1028 pageset->pcp.batch);
Mel Gorman467c9962007-10-16 01:26:02 -07001029#ifdef CONFIG_SMP
1030 seq_printf(m, "\n vm stats threshold: %d",
1031 pageset->stat_threshold);
1032#endif
1033 }
1034 seq_printf(m,
1035 "\n all_unreclaimable: %u"
Rik van Riel556adec2008-10-18 20:26:34 -07001036 "\n start_pfn: %lu"
1037 "\n inactive_ratio: %u",
KOSAKI Motohiro93e4a892010-03-05 13:41:55 -08001038 zone->all_unreclaimable,
Rik van Riel556adec2008-10-18 20:26:34 -07001039 zone->zone_start_pfn,
1040 zone->inactive_ratio);
Mel Gorman467c9962007-10-16 01:26:02 -07001041 seq_putc(m, '\n');
1042}
1043
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001044/*
1045 * Output information about zones in @pgdat.
1046 */
1047static int zoneinfo_show(struct seq_file *m, void *arg)
1048{
Mel Gorman467c9962007-10-16 01:26:02 -07001049 pg_data_t *pgdat = (pg_data_t *)arg;
1050 walk_zones_in_node(m, pgdat, zoneinfo_show_print);
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001051 return 0;
1052}
1053
Alexey Dobriyan5c9fe622008-10-06 04:19:42 +04001054static const struct seq_operations zoneinfo_op = {
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001055 .start = frag_start, /* iterate over all zones. The same as in
1056 * fragmentation. */
1057 .next = frag_next,
1058 .stop = frag_stop,
1059 .show = zoneinfo_show,
1060};
1061
Alexey Dobriyan5c9fe622008-10-06 04:19:42 +04001062static int zoneinfo_open(struct inode *inode, struct file *file)
1063{
1064 return seq_open(file, &zoneinfo_op);
1065}
1066
1067static const struct file_operations proc_zoneinfo_file_operations = {
1068 .open = zoneinfo_open,
1069 .read = seq_read,
1070 .llseek = seq_lseek,
1071 .release = seq_release,
1072};
1073
Michael Rubin79da8262010-10-26 14:21:36 -07001074enum writeback_stat_item {
1075 NR_DIRTY_THRESHOLD,
1076 NR_DIRTY_BG_THRESHOLD,
1077 NR_VM_WRITEBACK_STAT_ITEMS,
1078};
1079
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001080static void *vmstat_start(struct seq_file *m, loff_t *pos)
1081{
Christoph Lameter2244b952006-06-30 01:55:33 -07001082 unsigned long *v;
Michael Rubin79da8262010-10-26 14:21:36 -07001083 int i, stat_items_size;
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001084
1085 if (*pos >= ARRAY_SIZE(vmstat_text))
1086 return NULL;
Michael Rubin79da8262010-10-26 14:21:36 -07001087 stat_items_size = NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long) +
1088 NR_VM_WRITEBACK_STAT_ITEMS * sizeof(unsigned long);
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001089
Christoph Lameterf8891e52006-06-30 01:55:45 -07001090#ifdef CONFIG_VM_EVENT_COUNTERS
Michael Rubin79da8262010-10-26 14:21:36 -07001091 stat_items_size += sizeof(struct vm_event_state);
Christoph Lameterf8891e52006-06-30 01:55:45 -07001092#endif
Michael Rubin79da8262010-10-26 14:21:36 -07001093
1094 v = kmalloc(stat_items_size, GFP_KERNEL);
Christoph Lameter2244b952006-06-30 01:55:33 -07001095 m->private = v;
1096 if (!v)
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001097 return ERR_PTR(-ENOMEM);
Christoph Lameter2244b952006-06-30 01:55:33 -07001098 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
1099 v[i] = global_page_state(i);
Michael Rubin79da8262010-10-26 14:21:36 -07001100 v += NR_VM_ZONE_STAT_ITEMS;
1101
1102 global_dirty_limits(v + NR_DIRTY_BG_THRESHOLD,
1103 v + NR_DIRTY_THRESHOLD);
1104 v += NR_VM_WRITEBACK_STAT_ITEMS;
1105
Christoph Lameterf8891e52006-06-30 01:55:45 -07001106#ifdef CONFIG_VM_EVENT_COUNTERS
Michael Rubin79da8262010-10-26 14:21:36 -07001107 all_vm_events(v);
1108 v[PGPGIN] /= 2; /* sectors -> kbytes */
1109 v[PGPGOUT] /= 2;
Christoph Lameterf8891e52006-06-30 01:55:45 -07001110#endif
Wu Fengguangff8b16d2010-11-04 01:56:49 +08001111 return (unsigned long *)m->private + *pos;
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001112}
1113
1114static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
1115{
1116 (*pos)++;
1117 if (*pos >= ARRAY_SIZE(vmstat_text))
1118 return NULL;
1119 return (unsigned long *)m->private + *pos;
1120}
1121
1122static int vmstat_show(struct seq_file *m, void *arg)
1123{
1124 unsigned long *l = arg;
1125 unsigned long off = l - (unsigned long *)m->private;
1126
1127 seq_printf(m, "%s %lu\n", vmstat_text[off], *l);
1128 return 0;
1129}
1130
1131static void vmstat_stop(struct seq_file *m, void *arg)
1132{
1133 kfree(m->private);
1134 m->private = NULL;
1135}
1136
Alexey Dobriyanb6aa44a2008-10-06 04:17:48 +04001137static const struct seq_operations vmstat_op = {
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001138 .start = vmstat_start,
1139 .next = vmstat_next,
1140 .stop = vmstat_stop,
1141 .show = vmstat_show,
1142};
1143
Alexey Dobriyanb6aa44a2008-10-06 04:17:48 +04001144static int vmstat_open(struct inode *inode, struct file *file)
1145{
1146 return seq_open(file, &vmstat_op);
1147}
1148
1149static const struct file_operations proc_vmstat_file_operations = {
1150 .open = vmstat_open,
1151 .read = seq_read,
1152 .llseek = seq_lseek,
1153 .release = seq_release,
1154};
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001155#endif /* CONFIG_PROC_FS */
1156
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001157#ifdef CONFIG_SMP
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001158static DEFINE_PER_CPU(struct delayed_work, vmstat_work);
Christoph Lameter77461ab2007-05-09 02:35:13 -07001159int sysctl_stat_interval __read_mostly = HZ;
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001160
1161static void vmstat_update(struct work_struct *w)
1162{
1163 refresh_cpu_vm_stats(smp_processor_id());
Christoph Lameter77461ab2007-05-09 02:35:13 -07001164 schedule_delayed_work(&__get_cpu_var(vmstat_work),
Anton Blanchard98f4ebb2009-04-02 16:56:39 -07001165 round_jiffies_relative(sysctl_stat_interval));
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001166}
1167
Randy Dunlap42614fc2007-11-14 17:00:12 -08001168static void __cpuinit start_cpu_timer(int cpu)
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001169{
Tejun Heo1871e522009-10-29 22:34:13 +09001170 struct delayed_work *work = &per_cpu(vmstat_work, cpu);
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001171
Tejun Heo203b42f2012-08-21 13:18:23 -07001172 INIT_DEFERRABLE_WORK(work, vmstat_update);
Tejun Heo1871e522009-10-29 22:34:13 +09001173 schedule_delayed_work_on(cpu, work, __round_jiffies_relative(HZ, cpu));
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001174}
1175
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001176/*
1177 * Use the cpu notifier to insure that the thresholds are recalculated
1178 * when necessary.
1179 */
1180static int __cpuinit vmstat_cpuup_callback(struct notifier_block *nfb,
1181 unsigned long action,
1182 void *hcpu)
1183{
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001184 long cpu = (long)hcpu;
1185
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001186 switch (action) {
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001187 case CPU_ONLINE:
1188 case CPU_ONLINE_FROZEN:
KAMEZAWA Hiroyuki5ee28a42010-09-09 16:38:14 -07001189 refresh_zone_stat_thresholds();
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001190 start_cpu_timer(cpu);
Christoph Lameterad596922010-01-05 15:34:51 +09001191 node_set_state(cpu_to_node(cpu), N_CPU);
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001192 break;
1193 case CPU_DOWN_PREPARE:
1194 case CPU_DOWN_PREPARE_FROZEN:
Tejun Heoafe2c512010-12-14 16:21:17 +01001195 cancel_delayed_work_sync(&per_cpu(vmstat_work, cpu));
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001196 per_cpu(vmstat_work, cpu).work.func = NULL;
1197 break;
1198 case CPU_DOWN_FAILED:
1199 case CPU_DOWN_FAILED_FROZEN:
1200 start_cpu_timer(cpu);
1201 break;
Andy Whitcroftce421c72006-12-06 20:33:08 -08001202 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07001203 case CPU_DEAD_FROZEN:
Andy Whitcroftce421c72006-12-06 20:33:08 -08001204 refresh_zone_stat_thresholds();
1205 break;
1206 default:
1207 break;
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001208 }
1209 return NOTIFY_OK;
1210}
1211
1212static struct notifier_block __cpuinitdata vmstat_notifier =
1213 { &vmstat_cpuup_callback, NULL, 0 };
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +04001214#endif
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001215
Adrian Bunke2fc88d2007-10-16 01:26:27 -07001216static int __init setup_vmstat(void)
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001217{
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +04001218#ifdef CONFIG_SMP
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001219 int cpu;
1220
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001221 register_cpu_notifier(&vmstat_notifier);
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001222
1223 for_each_online_cpu(cpu)
1224 start_cpu_timer(cpu);
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +04001225#endif
1226#ifdef CONFIG_PROC_FS
1227 proc_create("buddyinfo", S_IRUGO, NULL, &fragmentation_file_operations);
Alexey Dobriyan74e2e8e2008-10-06 04:15:36 +04001228 proc_create("pagetypeinfo", S_IRUGO, NULL, &pagetypeinfo_file_ops);
Alexey Dobriyanb6aa44a2008-10-06 04:17:48 +04001229 proc_create("vmstat", S_IRUGO, NULL, &proc_vmstat_file_operations);
Alexey Dobriyan5c9fe622008-10-06 04:19:42 +04001230 proc_create("zoneinfo", S_IRUGO, NULL, &proc_zoneinfo_file_operations);
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +04001231#endif
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001232 return 0;
1233}
1234module_init(setup_vmstat)
Mel Gormand7a57522010-05-24 14:32:25 -07001235
1236#if defined(CONFIG_DEBUG_FS) && defined(CONFIG_COMPACTION)
1237#include <linux/debugfs.h>
1238
Mel Gormand7a57522010-05-24 14:32:25 -07001239
1240/*
1241 * Return an index indicating how much of the available free memory is
1242 * unusable for an allocation of the requested size.
1243 */
1244static int unusable_free_index(unsigned int order,
1245 struct contig_page_info *info)
1246{
1247 /* No free memory is interpreted as all free memory is unusable */
1248 if (info->free_pages == 0)
1249 return 1000;
1250
1251 /*
1252 * Index should be a value between 0 and 1. Return a value to 3
1253 * decimal places.
1254 *
1255 * 0 => no fragmentation
1256 * 1 => high fragmentation
1257 */
1258 return div_u64((info->free_pages - (info->free_blocks_suitable << order)) * 1000ULL, info->free_pages);
1259
1260}
1261
1262static void unusable_show_print(struct seq_file *m,
1263 pg_data_t *pgdat, struct zone *zone)
1264{
1265 unsigned int order;
1266 int index;
1267 struct contig_page_info info;
1268
1269 seq_printf(m, "Node %d, zone %8s ",
1270 pgdat->node_id,
1271 zone->name);
1272 for (order = 0; order < MAX_ORDER; ++order) {
1273 fill_contig_page_info(zone, order, &info);
1274 index = unusable_free_index(order, &info);
1275 seq_printf(m, "%d.%03d ", index / 1000, index % 1000);
1276 }
1277
1278 seq_putc(m, '\n');
1279}
1280
1281/*
1282 * Display unusable free space index
1283 *
1284 * The unusable free space index measures how much of the available free
1285 * memory cannot be used to satisfy an allocation of a given size and is a
1286 * value between 0 and 1. The higher the value, the more of free memory is
1287 * unusable and by implication, the worse the external fragmentation is. This
1288 * can be expressed as a percentage by multiplying by 100.
1289 */
1290static int unusable_show(struct seq_file *m, void *arg)
1291{
1292 pg_data_t *pgdat = (pg_data_t *)arg;
1293
1294 /* check memoryless node */
1295 if (!node_state(pgdat->node_id, N_HIGH_MEMORY))
1296 return 0;
1297
1298 walk_zones_in_node(m, pgdat, unusable_show_print);
1299
1300 return 0;
1301}
1302
1303static const struct seq_operations unusable_op = {
1304 .start = frag_start,
1305 .next = frag_next,
1306 .stop = frag_stop,
1307 .show = unusable_show,
1308};
1309
1310static int unusable_open(struct inode *inode, struct file *file)
1311{
1312 return seq_open(file, &unusable_op);
1313}
1314
1315static const struct file_operations unusable_file_ops = {
1316 .open = unusable_open,
1317 .read = seq_read,
1318 .llseek = seq_lseek,
1319 .release = seq_release,
1320};
1321
Mel Gormanf1a5ab12010-05-24 14:32:26 -07001322static void extfrag_show_print(struct seq_file *m,
1323 pg_data_t *pgdat, struct zone *zone)
1324{
1325 unsigned int order;
1326 int index;
1327
1328 /* Alloc on stack as interrupts are disabled for zone walk */
1329 struct contig_page_info info;
1330
1331 seq_printf(m, "Node %d, zone %8s ",
1332 pgdat->node_id,
1333 zone->name);
1334 for (order = 0; order < MAX_ORDER; ++order) {
1335 fill_contig_page_info(zone, order, &info);
Mel Gorman56de7262010-05-24 14:32:30 -07001336 index = __fragmentation_index(order, &info);
Mel Gormanf1a5ab12010-05-24 14:32:26 -07001337 seq_printf(m, "%d.%03d ", index / 1000, index % 1000);
1338 }
1339
1340 seq_putc(m, '\n');
1341}
1342
1343/*
1344 * Display fragmentation index for orders that allocations would fail for
1345 */
1346static int extfrag_show(struct seq_file *m, void *arg)
1347{
1348 pg_data_t *pgdat = (pg_data_t *)arg;
1349
1350 walk_zones_in_node(m, pgdat, extfrag_show_print);
1351
1352 return 0;
1353}
1354
1355static const struct seq_operations extfrag_op = {
1356 .start = frag_start,
1357 .next = frag_next,
1358 .stop = frag_stop,
1359 .show = extfrag_show,
1360};
1361
1362static int extfrag_open(struct inode *inode, struct file *file)
1363{
1364 return seq_open(file, &extfrag_op);
1365}
1366
1367static const struct file_operations extfrag_file_ops = {
1368 .open = extfrag_open,
1369 .read = seq_read,
1370 .llseek = seq_lseek,
1371 .release = seq_release,
1372};
1373
Mel Gormand7a57522010-05-24 14:32:25 -07001374static int __init extfrag_debug_init(void)
1375{
Sasikantha babubde8bd82012-05-29 15:06:22 -07001376 struct dentry *extfrag_debug_root;
1377
Mel Gormand7a57522010-05-24 14:32:25 -07001378 extfrag_debug_root = debugfs_create_dir("extfrag", NULL);
1379 if (!extfrag_debug_root)
1380 return -ENOMEM;
1381
1382 if (!debugfs_create_file("unusable_index", 0444,
1383 extfrag_debug_root, NULL, &unusable_file_ops))
Sasikantha babubde8bd82012-05-29 15:06:22 -07001384 goto fail;
Mel Gormand7a57522010-05-24 14:32:25 -07001385
Mel Gormanf1a5ab12010-05-24 14:32:26 -07001386 if (!debugfs_create_file("extfrag_index", 0444,
1387 extfrag_debug_root, NULL, &extfrag_file_ops))
Sasikantha babubde8bd82012-05-29 15:06:22 -07001388 goto fail;
Mel Gormanf1a5ab12010-05-24 14:32:26 -07001389
Mel Gormand7a57522010-05-24 14:32:25 -07001390 return 0;
Sasikantha babubde8bd82012-05-29 15:06:22 -07001391fail:
1392 debugfs_remove_recursive(extfrag_debug_root);
1393 return -ENOMEM;
Mel Gormand7a57522010-05-24 14:32:25 -07001394}
1395
1396module_init(extfrag_debug_init);
1397#endif