blob: 843125a0e255ccb35127f804498aea71d9f26e52 [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>
Lisa Du6e543d52013-09-11 14:22:36 -070022#include <linux/mm_inline.h>
23
24#include "internal.h"
Christoph Lameterf6ac2352006-06-30 01:55:32 -070025
Christoph Lameterf8891e52006-06-30 01:55:45 -070026#ifdef CONFIG_VM_EVENT_COUNTERS
27DEFINE_PER_CPU(struct vm_event_state, vm_event_states) = {{0}};
28EXPORT_PER_CPU_SYMBOL(vm_event_states);
29
Minchan Kim31f961a2010-08-09 17:18:59 -070030static void sum_vm_events(unsigned long *ret)
Christoph Lameterf8891e52006-06-30 01:55:45 -070031{
Christoph Lameter9eccf2a2008-02-04 22:29:22 -080032 int cpu;
Christoph Lameterf8891e52006-06-30 01:55:45 -070033 int i;
34
35 memset(ret, 0, NR_VM_EVENT_ITEMS * sizeof(unsigned long));
36
Minchan Kim31f961a2010-08-09 17:18:59 -070037 for_each_online_cpu(cpu) {
Christoph Lameterf8891e52006-06-30 01:55:45 -070038 struct vm_event_state *this = &per_cpu(vm_event_states, cpu);
39
Christoph Lameterf8891e52006-06-30 01:55:45 -070040 for (i = 0; i < NR_VM_EVENT_ITEMS; i++)
41 ret[i] += this->event[i];
42 }
43}
44
45/*
46 * Accumulate the vm event counters across all CPUs.
47 * The result is unavoidably approximate - it can change
48 * during and after execution of this function.
49*/
50void all_vm_events(unsigned long *ret)
51{
KOSAKI Motohirob5be1132008-05-12 14:02:06 -070052 get_online_cpus();
Minchan Kim31f961a2010-08-09 17:18:59 -070053 sum_vm_events(ret);
KOSAKI Motohirob5be1132008-05-12 14:02:06 -070054 put_online_cpus();
Christoph Lameterf8891e52006-06-30 01:55:45 -070055}
Heiko Carstens32dd66f2006-07-10 04:44:31 -070056EXPORT_SYMBOL_GPL(all_vm_events);
Christoph Lameterf8891e52006-06-30 01:55:45 -070057
Christoph Lameterf8891e52006-06-30 01:55:45 -070058/*
59 * Fold the foreign cpu events into our own.
60 *
61 * This is adding to the events on one processor
62 * but keeps the global counts constant.
63 */
64void vm_events_fold_cpu(int cpu)
65{
66 struct vm_event_state *fold_state = &per_cpu(vm_event_states, cpu);
67 int i;
68
69 for (i = 0; i < NR_VM_EVENT_ITEMS; i++) {
70 count_vm_events(i, fold_state->event[i]);
71 fold_state->event[i] = 0;
72 }
73}
Christoph Lameterf8891e52006-06-30 01:55:45 -070074
75#endif /* CONFIG_VM_EVENT_COUNTERS */
76
Christoph Lameter2244b952006-06-30 01:55:33 -070077/*
78 * Manage combined zone based / global counters
79 *
80 * vm_stat contains the global counters
81 */
Dimitri Sivanicha1cb2c62011-10-31 17:09:46 -070082atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS] __cacheline_aligned_in_smp;
Christoph Lameter2244b952006-06-30 01:55:33 -070083EXPORT_SYMBOL(vm_stat);
84
85#ifdef CONFIG_SMP
86
Mel Gormanb44129b2011-01-13 15:45:43 -080087int calculate_pressure_threshold(struct zone *zone)
Mel Gorman88f5acf2011-01-13 15:45:41 -080088{
89 int threshold;
90 int watermark_distance;
91
92 /*
93 * As vmstats are not up to date, there is drift between the estimated
94 * and real values. For high thresholds and a high number of CPUs, it
95 * is possible for the min watermark to be breached while the estimated
96 * value looks fine. The pressure threshold is a reduced value such
97 * that even the maximum amount of drift will not accidentally breach
98 * the min watermark
99 */
100 watermark_distance = low_wmark_pages(zone) - min_wmark_pages(zone);
101 threshold = max(1, (int)(watermark_distance / num_online_cpus()));
102
103 /*
104 * Maximum threshold is 125
105 */
106 threshold = min(125, threshold);
107
108 return threshold;
109}
110
Mel Gormanb44129b2011-01-13 15:45:43 -0800111int calculate_normal_threshold(struct zone *zone)
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700112{
113 int threshold;
114 int mem; /* memory in 128 MB units */
115
116 /*
117 * The threshold scales with the number of processors and the amount
118 * of memory per zone. More memory means that we can defer updates for
119 * longer, more processors could lead to more contention.
120 * fls() is used to have a cheap way of logarithmic scaling.
121 *
122 * Some sample thresholds:
123 *
124 * Threshold Processors (fls) Zonesize fls(mem+1)
125 * ------------------------------------------------------------------
126 * 8 1 1 0.9-1 GB 4
127 * 16 2 2 0.9-1 GB 4
128 * 20 2 2 1-2 GB 5
129 * 24 2 2 2-4 GB 6
130 * 28 2 2 4-8 GB 7
131 * 32 2 2 8-16 GB 8
132 * 4 2 2 <128M 1
133 * 30 4 3 2-4 GB 5
134 * 48 4 3 8-16 GB 8
135 * 32 8 4 1-2 GB 4
136 * 32 8 4 0.9-1GB 4
137 * 10 16 5 <128M 1
138 * 40 16 5 900M 4
139 * 70 64 7 2-4 GB 5
140 * 84 64 7 4-8 GB 6
141 * 108 512 9 4-8 GB 6
142 * 125 1024 10 8-16 GB 8
143 * 125 1024 10 16-32 GB 9
144 */
145
Jiang Liub40da042013-02-22 16:33:52 -0800146 mem = zone->managed_pages >> (27 - PAGE_SHIFT);
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700147
148 threshold = 2 * fls(num_online_cpus()) * (1 + fls(mem));
149
150 /*
151 * Maximum threshold is 125
152 */
153 threshold = min(125, threshold);
154
155 return threshold;
156}
Christoph Lameter2244b952006-06-30 01:55:33 -0700157
158/*
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700159 * Refresh the thresholds for each zone.
Christoph Lameter2244b952006-06-30 01:55:33 -0700160 */
KOSAKI Motohiroa6cccdc2011-05-24 17:11:33 -0700161void refresh_zone_stat_thresholds(void)
Christoph Lameter2244b952006-06-30 01:55:33 -0700162{
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700163 struct zone *zone;
164 int cpu;
165 int threshold;
166
KOSAKI Motohiroee99c712009-03-31 15:19:31 -0700167 for_each_populated_zone(zone) {
Christoph Lameteraa454842010-09-09 16:38:17 -0700168 unsigned long max_drift, tolerate_drift;
169
Mel Gormanb44129b2011-01-13 15:45:43 -0800170 threshold = calculate_normal_threshold(zone);
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700171
172 for_each_online_cpu(cpu)
Christoph Lameter99dcc3e2010-01-05 15:34:51 +0900173 per_cpu_ptr(zone->pageset, cpu)->stat_threshold
174 = threshold;
Christoph Lameteraa454842010-09-09 16:38:17 -0700175
176 /*
177 * Only set percpu_drift_mark if there is a danger that
178 * NR_FREE_PAGES reports the low watermark is ok when in fact
179 * the min watermark could be breached by an allocation
180 */
181 tolerate_drift = low_wmark_pages(zone) - min_wmark_pages(zone);
182 max_drift = num_online_cpus() * threshold;
183 if (max_drift > tolerate_drift)
184 zone->percpu_drift_mark = high_wmark_pages(zone) +
185 max_drift;
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700186 }
Christoph Lameter2244b952006-06-30 01:55:33 -0700187}
188
Mel Gormanb44129b2011-01-13 15:45:43 -0800189void set_pgdat_percpu_threshold(pg_data_t *pgdat,
190 int (*calculate_pressure)(struct zone *))
Mel Gorman88f5acf2011-01-13 15:45:41 -0800191{
192 struct zone *zone;
193 int cpu;
194 int threshold;
195 int i;
196
Mel Gorman88f5acf2011-01-13 15:45:41 -0800197 for (i = 0; i < pgdat->nr_zones; i++) {
198 zone = &pgdat->node_zones[i];
199 if (!zone->percpu_drift_mark)
200 continue;
201
Mel Gormanb44129b2011-01-13 15:45:43 -0800202 threshold = (*calculate_pressure)(zone);
203 for_each_possible_cpu(cpu)
Mel Gorman88f5acf2011-01-13 15:45:41 -0800204 per_cpu_ptr(zone->pageset, cpu)->stat_threshold
205 = threshold;
206 }
Mel Gorman88f5acf2011-01-13 15:45:41 -0800207}
208
Christoph Lameter2244b952006-06-30 01:55:33 -0700209/*
210 * For use when we know that interrupts are disabled.
211 */
212void __mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
213 int delta)
214{
Christoph Lameter12938a92010-12-06 11:16:20 -0600215 struct per_cpu_pageset __percpu *pcp = zone->pageset;
216 s8 __percpu *p = pcp->vm_stat_diff + item;
Christoph Lameter2244b952006-06-30 01:55:33 -0700217 long x;
Christoph Lameter12938a92010-12-06 11:16:20 -0600218 long t;
Christoph Lameter2244b952006-06-30 01:55:33 -0700219
Christoph Lameter12938a92010-12-06 11:16:20 -0600220 x = delta + __this_cpu_read(*p);
Christoph Lameter2244b952006-06-30 01:55:33 -0700221
Christoph Lameter12938a92010-12-06 11:16:20 -0600222 t = __this_cpu_read(pcp->stat_threshold);
223
224 if (unlikely(x > t || x < -t)) {
Christoph Lameter2244b952006-06-30 01:55:33 -0700225 zone_page_state_add(x, zone, item);
226 x = 0;
227 }
Christoph Lameter12938a92010-12-06 11:16:20 -0600228 __this_cpu_write(*p, x);
Christoph Lameter2244b952006-06-30 01:55:33 -0700229}
230EXPORT_SYMBOL(__mod_zone_page_state);
231
232/*
Christoph Lameter2244b952006-06-30 01:55:33 -0700233 * Optimized increment and decrement functions.
234 *
235 * These are only for a single page and therefore can take a struct page *
236 * argument instead of struct zone *. This allows the inclusion of the code
237 * generated for page_zone(page) into the optimized functions.
238 *
239 * No overflow check is necessary and therefore the differential can be
240 * incremented or decremented in place which may allow the compilers to
241 * generate better code.
Christoph Lameter2244b952006-06-30 01:55:33 -0700242 * The increment or decrement is known and therefore one boundary check can
243 * be omitted.
244 *
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700245 * NOTE: These functions are very performance sensitive. Change only
246 * with care.
247 *
Christoph Lameter2244b952006-06-30 01:55:33 -0700248 * Some processors have inc/dec instructions that are atomic vs an interrupt.
249 * However, the code must first determine the differential location in a zone
250 * based on the processor number and then inc/dec the counter. There is no
251 * guarantee without disabling preemption that the processor will not change
252 * in between and therefore the atomicity vs. interrupt cannot be exploited
253 * in a useful way here.
254 */
Christoph Lameterc8785382007-02-10 01:43:01 -0800255void __inc_zone_state(struct zone *zone, enum zone_stat_item item)
Christoph Lameter2244b952006-06-30 01:55:33 -0700256{
Christoph Lameter12938a92010-12-06 11:16:20 -0600257 struct per_cpu_pageset __percpu *pcp = zone->pageset;
258 s8 __percpu *p = pcp->vm_stat_diff + item;
259 s8 v, t;
Christoph Lameter2244b952006-06-30 01:55:33 -0700260
Christoph Lameter908ee0f2010-12-06 11:40:02 -0600261 v = __this_cpu_inc_return(*p);
Christoph Lameter12938a92010-12-06 11:16:20 -0600262 t = __this_cpu_read(pcp->stat_threshold);
263 if (unlikely(v > t)) {
264 s8 overstep = t >> 1;
Christoph Lameter2244b952006-06-30 01:55:33 -0700265
Christoph Lameter12938a92010-12-06 11:16:20 -0600266 zone_page_state_add(v + overstep, zone, item);
267 __this_cpu_write(*p, -overstep);
Christoph Lameter2244b952006-06-30 01:55:33 -0700268 }
269}
Christoph Lameterca889e62006-06-30 01:55:44 -0700270
271void __inc_zone_page_state(struct page *page, enum zone_stat_item item)
272{
273 __inc_zone_state(page_zone(page), item);
274}
Christoph Lameter2244b952006-06-30 01:55:33 -0700275EXPORT_SYMBOL(__inc_zone_page_state);
276
Christoph Lameterc8785382007-02-10 01:43:01 -0800277void __dec_zone_state(struct zone *zone, enum zone_stat_item item)
Christoph Lameter2244b952006-06-30 01:55:33 -0700278{
Christoph Lameter12938a92010-12-06 11:16:20 -0600279 struct per_cpu_pageset __percpu *pcp = zone->pageset;
280 s8 __percpu *p = pcp->vm_stat_diff + item;
281 s8 v, t;
Christoph Lameter2244b952006-06-30 01:55:33 -0700282
Christoph Lameter908ee0f2010-12-06 11:40:02 -0600283 v = __this_cpu_dec_return(*p);
Christoph Lameter12938a92010-12-06 11:16:20 -0600284 t = __this_cpu_read(pcp->stat_threshold);
285 if (unlikely(v < - t)) {
286 s8 overstep = t >> 1;
Christoph Lameter2244b952006-06-30 01:55:33 -0700287
Christoph Lameter12938a92010-12-06 11:16:20 -0600288 zone_page_state_add(v - overstep, zone, item);
289 __this_cpu_write(*p, overstep);
Christoph Lameter2244b952006-06-30 01:55:33 -0700290 }
291}
Christoph Lameterc8785382007-02-10 01:43:01 -0800292
293void __dec_zone_page_state(struct page *page, enum zone_stat_item item)
294{
295 __dec_zone_state(page_zone(page), item);
296}
Christoph Lameter2244b952006-06-30 01:55:33 -0700297EXPORT_SYMBOL(__dec_zone_page_state);
298
Heiko Carstens41561532012-01-12 17:17:30 -0800299#ifdef CONFIG_HAVE_CMPXCHG_LOCAL
Christoph Lameter7c839122010-12-14 10:28:46 -0600300/*
301 * If we have cmpxchg_local support then we do not need to incur the overhead
302 * that comes with local_irq_save/restore if we use this_cpu_cmpxchg.
303 *
304 * mod_state() modifies the zone counter state through atomic per cpu
305 * operations.
306 *
307 * Overstep mode specifies how overstep should handled:
308 * 0 No overstepping
309 * 1 Overstepping half of threshold
310 * -1 Overstepping minus half of threshold
311*/
312static inline void mod_state(struct zone *zone,
313 enum zone_stat_item item, int delta, int overstep_mode)
314{
315 struct per_cpu_pageset __percpu *pcp = zone->pageset;
316 s8 __percpu *p = pcp->vm_stat_diff + item;
317 long o, n, t, z;
318
319 do {
320 z = 0; /* overflow to zone counters */
321
322 /*
323 * The fetching of the stat_threshold is racy. We may apply
324 * a counter threshold to the wrong the cpu if we get
Christoph Lameterd3bc2362011-04-14 15:21:58 -0700325 * rescheduled while executing here. However, the next
326 * counter update will apply the threshold again and
327 * therefore bring the counter under the threshold again.
328 *
329 * Most of the time the thresholds are the same anyways
330 * for all cpus in a zone.
Christoph Lameter7c839122010-12-14 10:28:46 -0600331 */
332 t = this_cpu_read(pcp->stat_threshold);
333
334 o = this_cpu_read(*p);
335 n = delta + o;
336
337 if (n > t || n < -t) {
338 int os = overstep_mode * (t >> 1) ;
339
340 /* Overflow must be added to zone counters */
341 z = n + os;
342 n = -os;
343 }
344 } while (this_cpu_cmpxchg(*p, o, n) != o);
345
346 if (z)
347 zone_page_state_add(z, zone, item);
348}
349
350void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
351 int delta)
352{
353 mod_state(zone, item, delta, 0);
354}
355EXPORT_SYMBOL(mod_zone_page_state);
356
357void inc_zone_state(struct zone *zone, enum zone_stat_item item)
358{
359 mod_state(zone, item, 1, 1);
360}
361
362void inc_zone_page_state(struct page *page, enum zone_stat_item item)
363{
364 mod_state(page_zone(page), item, 1, 1);
365}
366EXPORT_SYMBOL(inc_zone_page_state);
367
368void dec_zone_page_state(struct page *page, enum zone_stat_item item)
369{
370 mod_state(page_zone(page), item, -1, -1);
371}
372EXPORT_SYMBOL(dec_zone_page_state);
373#else
374/*
375 * Use interrupt disable to serialize counter updates
376 */
377void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
378 int delta)
379{
380 unsigned long flags;
381
382 local_irq_save(flags);
383 __mod_zone_page_state(zone, item, delta);
384 local_irq_restore(flags);
385}
386EXPORT_SYMBOL(mod_zone_page_state);
387
Christoph Lameterca889e62006-06-30 01:55:44 -0700388void inc_zone_state(struct zone *zone, enum zone_stat_item item)
389{
390 unsigned long flags;
391
392 local_irq_save(flags);
393 __inc_zone_state(zone, item);
394 local_irq_restore(flags);
395}
396
Christoph Lameter2244b952006-06-30 01:55:33 -0700397void inc_zone_page_state(struct page *page, enum zone_stat_item item)
398{
399 unsigned long flags;
400 struct zone *zone;
Christoph Lameter2244b952006-06-30 01:55:33 -0700401
402 zone = page_zone(page);
403 local_irq_save(flags);
Christoph Lameterca889e62006-06-30 01:55:44 -0700404 __inc_zone_state(zone, item);
Christoph Lameter2244b952006-06-30 01:55:33 -0700405 local_irq_restore(flags);
406}
407EXPORT_SYMBOL(inc_zone_page_state);
408
409void dec_zone_page_state(struct page *page, enum zone_stat_item item)
410{
411 unsigned long flags;
Christoph Lameter2244b952006-06-30 01:55:33 -0700412
Christoph Lameter2244b952006-06-30 01:55:33 -0700413 local_irq_save(flags);
Christoph Lametera302eb42006-08-31 21:27:34 -0700414 __dec_zone_page_state(page, item);
Christoph Lameter2244b952006-06-30 01:55:33 -0700415 local_irq_restore(flags);
416}
417EXPORT_SYMBOL(dec_zone_page_state);
Christoph Lameter7c839122010-12-14 10:28:46 -0600418#endif
Christoph Lameter2244b952006-06-30 01:55:33 -0700419
Christoph Lameter4edb0742013-09-11 14:21:31 -0700420static inline void fold_diff(int *diff)
421{
422 int i;
423
424 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
425 if (diff[i])
426 atomic_long_add(diff[i], &vm_stat[i]);
427}
428
Christoph Lameter2244b952006-06-30 01:55:33 -0700429/*
Christoph Lameter2bb921e2013-09-11 14:21:30 -0700430 * Update the zone counters for the current cpu.
Christoph Lametera7f75e22008-02-04 22:29:16 -0800431 *
Christoph Lameter4037d452007-05-09 02:35:14 -0700432 * Note that refresh_cpu_vm_stats strives to only access
433 * node local memory. The per cpu pagesets on remote zones are placed
434 * in the memory local to the processor using that pageset. So the
435 * loop over all zones will access a series of cachelines local to
436 * the processor.
437 *
438 * The call to zone_page_state_add updates the cachelines with the
439 * statistics in the remote zone struct as well as the global cachelines
440 * with the global counters. These could cause remote node cache line
441 * bouncing and will have to be only done when necessary.
Christoph Lameter2244b952006-06-30 01:55:33 -0700442 */
Christoph Lameterfbc2edb2013-09-11 14:21:32 -0700443static void refresh_cpu_vm_stats(void)
Christoph Lameter2244b952006-06-30 01:55:33 -0700444{
445 struct zone *zone;
446 int i;
Christoph Lametera7f75e22008-02-04 22:29:16 -0800447 int global_diff[NR_VM_ZONE_STAT_ITEMS] = { 0, };
Christoph Lameter2244b952006-06-30 01:55:33 -0700448
KOSAKI Motohiroee99c712009-03-31 15:19:31 -0700449 for_each_populated_zone(zone) {
Christoph Lameterfbc2edb2013-09-11 14:21:32 -0700450 struct per_cpu_pageset __percpu *p = zone->pageset;
Christoph Lameter2244b952006-06-30 01:55:33 -0700451
Christoph Lameterfbc2edb2013-09-11 14:21:32 -0700452 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) {
453 int v;
Christoph Lameter2244b952006-06-30 01:55:33 -0700454
Christoph Lameterfbc2edb2013-09-11 14:21:32 -0700455 v = this_cpu_xchg(p->vm_stat_diff[i], 0);
456 if (v) {
Christoph Lametera7f75e22008-02-04 22:29:16 -0800457
Christoph Lametera7f75e22008-02-04 22:29:16 -0800458 atomic_long_add(v, &zone->vm_stat[i]);
459 global_diff[i] += v;
Christoph Lameter4037d452007-05-09 02:35:14 -0700460#ifdef CONFIG_NUMA
461 /* 3 seconds idle till flush */
Christoph Lameterfbc2edb2013-09-11 14:21:32 -0700462 __this_cpu_write(p->expire, 3);
Christoph Lameter4037d452007-05-09 02:35:14 -0700463#endif
Christoph Lameter2244b952006-06-30 01:55:33 -0700464 }
Christoph Lameterfbc2edb2013-09-11 14:21:32 -0700465 }
Dimitri Sivanich468fd622008-04-28 02:13:37 -0700466 cond_resched();
Christoph Lameter4037d452007-05-09 02:35:14 -0700467#ifdef CONFIG_NUMA
468 /*
469 * Deal with draining the remote pageset of this
470 * processor
471 *
472 * Check if there are pages remaining in this pageset
473 * if not then there is nothing to expire.
474 */
Christoph Lameterfbc2edb2013-09-11 14:21:32 -0700475 if (!__this_cpu_read(p->expire) ||
476 !__this_cpu_read(p->pcp.count))
Christoph Lameter4037d452007-05-09 02:35:14 -0700477 continue;
478
479 /*
480 * We never drain zones local to this processor.
481 */
482 if (zone_to_nid(zone) == numa_node_id()) {
Christoph Lameterfbc2edb2013-09-11 14:21:32 -0700483 __this_cpu_write(p->expire, 0);
Christoph Lameter4037d452007-05-09 02:35:14 -0700484 continue;
485 }
486
Christoph Lameterfbc2edb2013-09-11 14:21:32 -0700487
488 if (__this_cpu_dec_return(p->expire))
Christoph Lameter4037d452007-05-09 02:35:14 -0700489 continue;
490
Christoph Lameterfbc2edb2013-09-11 14:21:32 -0700491 if (__this_cpu_read(p->pcp.count))
492 drain_zone_pages(zone, __this_cpu_ptr(&p->pcp));
Christoph Lameter4037d452007-05-09 02:35:14 -0700493#endif
Christoph Lameter2244b952006-06-30 01:55:33 -0700494 }
Christoph Lameter4edb0742013-09-11 14:21:31 -0700495 fold_diff(global_diff);
Christoph Lameter2244b952006-06-30 01:55:33 -0700496}
497
Cody P Schafer40f4b1e2013-04-29 15:08:38 -0700498/*
Christoph Lameter2bb921e2013-09-11 14:21:30 -0700499 * Fold the data for an offline cpu into the global array.
500 * There cannot be any access by the offline cpu and therefore
501 * synchronization is simplified.
502 */
503void cpu_vm_stats_fold(int cpu)
504{
505 struct zone *zone;
506 int i;
507 int global_diff[NR_VM_ZONE_STAT_ITEMS] = { 0, };
508
509 for_each_populated_zone(zone) {
510 struct per_cpu_pageset *p;
511
512 p = per_cpu_ptr(zone->pageset, cpu);
513
514 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
515 if (p->vm_stat_diff[i]) {
516 int v;
517
518 v = p->vm_stat_diff[i];
519 p->vm_stat_diff[i] = 0;
520 atomic_long_add(v, &zone->vm_stat[i]);
521 global_diff[i] += v;
522 }
523 }
524
Christoph Lameter4edb0742013-09-11 14:21:31 -0700525 fold_diff(global_diff);
Christoph Lameter2bb921e2013-09-11 14:21:30 -0700526}
527
528/*
Cody P Schafer40f4b1e2013-04-29 15:08:38 -0700529 * this is only called if !populated_zone(zone), which implies no other users of
530 * pset->vm_stat_diff[] exsist.
531 */
Minchan Kim5a883812012-10-08 16:33:39 -0700532void drain_zonestat(struct zone *zone, struct per_cpu_pageset *pset)
533{
534 int i;
535
536 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
537 if (pset->vm_stat_diff[i]) {
538 int v = pset->vm_stat_diff[i];
539 pset->vm_stat_diff[i] = 0;
540 atomic_long_add(v, &zone->vm_stat[i]);
541 atomic_long_add(v, &vm_stat[i]);
542 }
543}
Christoph Lameter2244b952006-06-30 01:55:33 -0700544#endif
545
Christoph Lameterca889e62006-06-30 01:55:44 -0700546#ifdef CONFIG_NUMA
547/*
548 * zonelist = the list of zones passed to the allocator
549 * z = the zone from which the allocation occurred.
550 *
551 * Must be called with interrupts disabled.
Andi Kleen78afd562011-03-22 16:33:12 -0700552 *
553 * When __GFP_OTHER_NODE is set assume the node of the preferred
554 * zone is the local node. This is useful for daemons who allocate
555 * memory on behalf of other processes.
Christoph Lameterca889e62006-06-30 01:55:44 -0700556 */
Andi Kleen78afd562011-03-22 16:33:12 -0700557void zone_statistics(struct zone *preferred_zone, struct zone *z, gfp_t flags)
Christoph Lameterca889e62006-06-30 01:55:44 -0700558{
Mel Gorman18ea7e72008-04-28 02:12:14 -0700559 if (z->zone_pgdat == preferred_zone->zone_pgdat) {
Christoph Lameterca889e62006-06-30 01:55:44 -0700560 __inc_zone_state(z, NUMA_HIT);
561 } else {
562 __inc_zone_state(z, NUMA_MISS);
Mel Gorman18ea7e72008-04-28 02:12:14 -0700563 __inc_zone_state(preferred_zone, NUMA_FOREIGN);
Christoph Lameterca889e62006-06-30 01:55:44 -0700564 }
Andi Kleen78afd562011-03-22 16:33:12 -0700565 if (z->node == ((flags & __GFP_OTHER_NODE) ?
566 preferred_zone->node : numa_node_id()))
Christoph Lameterca889e62006-06-30 01:55:44 -0700567 __inc_zone_state(z, NUMA_LOCAL);
568 else
569 __inc_zone_state(z, NUMA_OTHER);
570}
571#endif
572
Mel Gormand7a57522010-05-24 14:32:25 -0700573#ifdef CONFIG_COMPACTION
Namhyung Kim36deb0b2010-10-26 14:22:04 -0700574
Mel Gormand7a57522010-05-24 14:32:25 -0700575struct contig_page_info {
576 unsigned long free_pages;
577 unsigned long free_blocks_total;
578 unsigned long free_blocks_suitable;
579};
580
581/*
582 * Calculate the number of free pages in a zone, how many contiguous
583 * pages are free and how many are large enough to satisfy an allocation of
584 * the target size. Note that this function makes no attempt to estimate
585 * how many suitable free blocks there *might* be if MOVABLE pages were
586 * migrated. Calculating that is possible, but expensive and can be
587 * figured out from userspace
588 */
589static void fill_contig_page_info(struct zone *zone,
590 unsigned int suitable_order,
591 struct contig_page_info *info)
592{
593 unsigned int order;
594
595 info->free_pages = 0;
596 info->free_blocks_total = 0;
597 info->free_blocks_suitable = 0;
598
599 for (order = 0; order < MAX_ORDER; order++) {
600 unsigned long blocks;
601
602 /* Count number of free blocks */
603 blocks = zone->free_area[order].nr_free;
604 info->free_blocks_total += blocks;
605
606 /* Count free base pages */
607 info->free_pages += blocks << order;
608
609 /* Count the suitable free blocks */
610 if (order >= suitable_order)
611 info->free_blocks_suitable += blocks <<
612 (order - suitable_order);
613 }
614}
Mel Gormanf1a5ab12010-05-24 14:32:26 -0700615
616/*
617 * A fragmentation index only makes sense if an allocation of a requested
618 * size would fail. If that is true, the fragmentation index indicates
619 * whether external fragmentation or a lack of memory was the problem.
620 * The value can be used to determine if page reclaim or compaction
621 * should be used
622 */
Mel Gorman56de7262010-05-24 14:32:30 -0700623static int __fragmentation_index(unsigned int order, struct contig_page_info *info)
Mel Gormanf1a5ab12010-05-24 14:32:26 -0700624{
625 unsigned long requested = 1UL << order;
626
627 if (!info->free_blocks_total)
628 return 0;
629
630 /* Fragmentation index only makes sense when a request would fail */
631 if (info->free_blocks_suitable)
632 return -1000;
633
634 /*
635 * Index is between 0 and 1 so return within 3 decimal places
636 *
637 * 0 => allocation would fail due to lack of memory
638 * 1 => allocation would fail due to fragmentation
639 */
640 return 1000 - div_u64( (1000+(div_u64(info->free_pages * 1000ULL, requested))), info->free_blocks_total);
641}
Mel Gorman56de7262010-05-24 14:32:30 -0700642
643/* Same as __fragmentation index but allocs contig_page_info on stack */
644int fragmentation_index(struct zone *zone, unsigned int order)
645{
646 struct contig_page_info info;
647
648 fill_contig_page_info(zone, order, &info);
649 return __fragmentation_index(order, &info);
650}
Mel Gormand7a57522010-05-24 14:32:25 -0700651#endif
652
653#if defined(CONFIG_PROC_FS) || defined(CONFIG_COMPACTION)
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +0400654#include <linux/proc_fs.h>
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700655#include <linux/seq_file.h>
656
Mel Gorman467c9962007-10-16 01:26:02 -0700657static char * const migratetype_names[MIGRATE_TYPES] = {
658 "Unmovable",
659 "Reclaimable",
660 "Movable",
661 "Reserve",
Michal Nazarewicz47118af2011-12-29 13:09:50 +0100662#ifdef CONFIG_CMA
663 "CMA",
664#endif
Minchan Kim194159f2013-02-22 16:33:58 -0800665#ifdef CONFIG_MEMORY_ISOLATION
KOSAKI Motohiro91446b02008-04-15 14:34:42 -0700666 "Isolate",
Minchan Kim194159f2013-02-22 16:33:58 -0800667#endif
Mel Gorman467c9962007-10-16 01:26:02 -0700668};
669
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700670static void *frag_start(struct seq_file *m, loff_t *pos)
671{
672 pg_data_t *pgdat;
673 loff_t node = *pos;
674 for (pgdat = first_online_pgdat();
675 pgdat && node;
676 pgdat = next_online_pgdat(pgdat))
677 --node;
678
679 return pgdat;
680}
681
682static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
683{
684 pg_data_t *pgdat = (pg_data_t *)arg;
685
686 (*pos)++;
687 return next_online_pgdat(pgdat);
688}
689
690static void frag_stop(struct seq_file *m, void *arg)
691{
692}
693
Mel Gorman467c9962007-10-16 01:26:02 -0700694/* Walk all the zones in a node and print using a callback */
695static void walk_zones_in_node(struct seq_file *m, pg_data_t *pgdat,
696 void (*print)(struct seq_file *m, pg_data_t *, struct zone *))
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700697{
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700698 struct zone *zone;
699 struct zone *node_zones = pgdat->node_zones;
700 unsigned long flags;
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700701
702 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
703 if (!populated_zone(zone))
704 continue;
705
706 spin_lock_irqsave(&zone->lock, flags);
Mel Gorman467c9962007-10-16 01:26:02 -0700707 print(m, pgdat, zone);
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700708 spin_unlock_irqrestore(&zone->lock, flags);
Mel Gorman467c9962007-10-16 01:26:02 -0700709 }
710}
Mel Gormand7a57522010-05-24 14:32:25 -0700711#endif
Mel Gorman467c9962007-10-16 01:26:02 -0700712
David Rientjes0d6617c2011-09-14 16:21:05 -0700713#if defined(CONFIG_PROC_FS) || defined(CONFIG_SYSFS) || defined(CONFIG_NUMA)
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700714#ifdef CONFIG_ZONE_DMA
715#define TEXT_FOR_DMA(xx) xx "_dma",
716#else
717#define TEXT_FOR_DMA(xx)
718#endif
719
720#ifdef CONFIG_ZONE_DMA32
721#define TEXT_FOR_DMA32(xx) xx "_dma32",
722#else
723#define TEXT_FOR_DMA32(xx)
724#endif
725
726#ifdef CONFIG_HIGHMEM
727#define TEXT_FOR_HIGHMEM(xx) xx "_high",
728#else
729#define TEXT_FOR_HIGHMEM(xx)
730#endif
731
732#define TEXTS_FOR_ZONES(xx) TEXT_FOR_DMA(xx) TEXT_FOR_DMA32(xx) xx "_normal", \
733 TEXT_FOR_HIGHMEM(xx) xx "_movable",
734
735const char * const vmstat_text[] = {
736 /* Zoned VM counters */
737 "nr_free_pages",
Johannes Weiner81c0a2b2013-09-11 14:20:47 -0700738 "nr_alloc_batch",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700739 "nr_inactive_anon",
740 "nr_active_anon",
741 "nr_inactive_file",
742 "nr_active_file",
743 "nr_unevictable",
744 "nr_mlock",
745 "nr_anon_pages",
746 "nr_mapped",
747 "nr_file_pages",
748 "nr_dirty",
749 "nr_writeback",
750 "nr_slab_reclaimable",
751 "nr_slab_unreclaimable",
752 "nr_page_table_pages",
753 "nr_kernel_stack",
754 "nr_unstable",
755 "nr_bounce",
756 "nr_vmscan_write",
Mel Gorman49ea7eb2011-10-31 17:07:59 -0700757 "nr_vmscan_immediate_reclaim",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700758 "nr_writeback_temp",
759 "nr_isolated_anon",
760 "nr_isolated_file",
761 "nr_shmem",
762 "nr_dirtied",
763 "nr_written",
764
765#ifdef CONFIG_NUMA
766 "numa_hit",
767 "numa_miss",
768 "numa_foreign",
769 "numa_interleave",
770 "numa_local",
771 "numa_other",
772#endif
Johannes Weinera5289102014-04-03 14:47:51 -0700773 "workingset_refault",
774 "workingset_activate",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700775 "nr_anon_transparent_hugepages",
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -0700776 "nr_free_cma",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700777 "nr_dirty_threshold",
778 "nr_dirty_background_threshold",
779
780#ifdef CONFIG_VM_EVENT_COUNTERS
781 "pgpgin",
782 "pgpgout",
783 "pswpin",
784 "pswpout",
785
786 TEXTS_FOR_ZONES("pgalloc")
787
788 "pgfree",
789 "pgactivate",
790 "pgdeactivate",
791
792 "pgfault",
793 "pgmajfault",
794
795 TEXTS_FOR_ZONES("pgrefill")
Ying Han904249a2012-04-25 16:01:48 -0700796 TEXTS_FOR_ZONES("pgsteal_kswapd")
797 TEXTS_FOR_ZONES("pgsteal_direct")
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700798 TEXTS_FOR_ZONES("pgscan_kswapd")
799 TEXTS_FOR_ZONES("pgscan_direct")
Mel Gorman68243e72012-07-31 16:44:39 -0700800 "pgscan_direct_throttle",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700801
802#ifdef CONFIG_NUMA
803 "zone_reclaim_failed",
804#endif
805 "pginodesteal",
806 "slabs_scanned",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700807 "kswapd_inodesteal",
808 "kswapd_low_wmark_hit_quickly",
809 "kswapd_high_wmark_hit_quickly",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700810 "pageoutrun",
811 "allocstall",
812
813 "pgrotated",
814
Mel Gorman03c5a6e2012-11-02 14:52:48 +0000815#ifdef CONFIG_NUMA_BALANCING
816 "numa_pte_updates",
Mel Gorman72403b42013-11-12 15:08:32 -0800817 "numa_huge_pte_updates",
Mel Gorman03c5a6e2012-11-02 14:52:48 +0000818 "numa_hint_faults",
819 "numa_hint_faults_local",
820 "numa_pages_migrated",
821#endif
Mel Gorman5647bc22012-10-19 10:46:20 +0100822#ifdef CONFIG_MIGRATION
823 "pgmigrate_success",
824 "pgmigrate_fail",
825#endif
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700826#ifdef CONFIG_COMPACTION
Mel Gorman397487d2012-10-19 12:00:10 +0100827 "compact_migrate_scanned",
828 "compact_free_scanned",
829 "compact_isolated",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700830 "compact_stall",
831 "compact_fail",
832 "compact_success",
833#endif
834
835#ifdef CONFIG_HUGETLB_PAGE
836 "htlb_buddy_alloc_success",
837 "htlb_buddy_alloc_fail",
838#endif
839 "unevictable_pgs_culled",
840 "unevictable_pgs_scanned",
841 "unevictable_pgs_rescued",
842 "unevictable_pgs_mlocked",
843 "unevictable_pgs_munlocked",
844 "unevictable_pgs_cleared",
845 "unevictable_pgs_stranded",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700846
847#ifdef CONFIG_TRANSPARENT_HUGEPAGE
848 "thp_fault_alloc",
849 "thp_fault_fallback",
850 "thp_collapse_alloc",
851 "thp_collapse_alloc_failed",
852 "thp_split",
Kirill A. Shutemovd8a8e1f2012-12-12 13:51:09 -0800853 "thp_zero_page_alloc",
854 "thp_zero_page_alloc_failed",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700855#endif
Mel Gormanec659932014-01-21 14:33:16 -0800856#ifdef CONFIG_DEBUG_TLBFLUSH
Dave Hansen6df46862013-09-11 14:20:24 -0700857#ifdef CONFIG_SMP
Dave Hansen9824cf92013-09-11 14:20:23 -0700858 "nr_tlb_remote_flush",
859 "nr_tlb_remote_flush_received",
Mel Gormanec659932014-01-21 14:33:16 -0800860#endif /* CONFIG_SMP */
Dave Hansen9824cf92013-09-11 14:20:23 -0700861 "nr_tlb_local_flush_all",
862 "nr_tlb_local_flush_one",
Mel Gormanec659932014-01-21 14:33:16 -0800863#endif /* CONFIG_DEBUG_TLBFLUSH */
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700864
865#endif /* CONFIG_VM_EVENTS_COUNTERS */
866};
David Rientjes0d6617c2011-09-14 16:21:05 -0700867#endif /* CONFIG_PROC_FS || CONFIG_SYSFS || CONFIG_NUMA */
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700868
869
Mel Gormand7a57522010-05-24 14:32:25 -0700870#ifdef CONFIG_PROC_FS
Mel Gorman467c9962007-10-16 01:26:02 -0700871static void frag_show_print(struct seq_file *m, pg_data_t *pgdat,
872 struct zone *zone)
873{
874 int order;
875
876 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
877 for (order = 0; order < MAX_ORDER; ++order)
878 seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
879 seq_putc(m, '\n');
880}
881
882/*
883 * This walks the free areas for each zone.
884 */
885static int frag_show(struct seq_file *m, void *arg)
886{
887 pg_data_t *pgdat = (pg_data_t *)arg;
888 walk_zones_in_node(m, pgdat, frag_show_print);
889 return 0;
890}
891
892static void pagetypeinfo_showfree_print(struct seq_file *m,
893 pg_data_t *pgdat, struct zone *zone)
894{
895 int order, mtype;
896
897 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++) {
898 seq_printf(m, "Node %4d, zone %8s, type %12s ",
899 pgdat->node_id,
900 zone->name,
901 migratetype_names[mtype]);
902 for (order = 0; order < MAX_ORDER; ++order) {
903 unsigned long freecount = 0;
904 struct free_area *area;
905 struct list_head *curr;
906
907 area = &(zone->free_area[order]);
908
909 list_for_each(curr, &area->free_list[mtype])
910 freecount++;
911 seq_printf(m, "%6lu ", freecount);
912 }
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700913 seq_putc(m, '\n');
914 }
Mel Gorman467c9962007-10-16 01:26:02 -0700915}
916
917/* Print out the free pages at each order for each migatetype */
918static int pagetypeinfo_showfree(struct seq_file *m, void *arg)
919{
920 int order;
921 pg_data_t *pgdat = (pg_data_t *)arg;
922
923 /* Print header */
924 seq_printf(m, "%-43s ", "Free pages count per migrate type at order");
925 for (order = 0; order < MAX_ORDER; ++order)
926 seq_printf(m, "%6d ", order);
927 seq_putc(m, '\n');
928
929 walk_zones_in_node(m, pgdat, pagetypeinfo_showfree_print);
930
931 return 0;
932}
933
934static void pagetypeinfo_showblockcount_print(struct seq_file *m,
935 pg_data_t *pgdat, struct zone *zone)
936{
937 int mtype;
938 unsigned long pfn;
939 unsigned long start_pfn = zone->zone_start_pfn;
Cody P Schafer108bcc92013-02-22 16:35:23 -0800940 unsigned long end_pfn = zone_end_pfn(zone);
Mel Gorman467c9962007-10-16 01:26:02 -0700941 unsigned long count[MIGRATE_TYPES] = { 0, };
942
943 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
944 struct page *page;
945
946 if (!pfn_valid(pfn))
947 continue;
948
949 page = pfn_to_page(pfn);
Mel Gormaneb335752009-05-13 17:34:48 +0100950
951 /* Watch for unexpected holes punched in the memmap */
952 if (!memmap_valid_within(pfn, page, zone))
Mel Gormane80d6a22008-08-14 11:10:14 +0100953 continue;
Mel Gormaneb335752009-05-13 17:34:48 +0100954
Mel Gorman467c9962007-10-16 01:26:02 -0700955 mtype = get_pageblock_migratetype(page);
956
Mel Gormane80d6a22008-08-14 11:10:14 +0100957 if (mtype < MIGRATE_TYPES)
958 count[mtype]++;
Mel Gorman467c9962007-10-16 01:26:02 -0700959 }
960
961 /* Print counts */
962 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
963 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
964 seq_printf(m, "%12lu ", count[mtype]);
965 seq_putc(m, '\n');
966}
967
968/* Print out the free pages at each order for each migratetype */
969static int pagetypeinfo_showblockcount(struct seq_file *m, void *arg)
970{
971 int mtype;
972 pg_data_t *pgdat = (pg_data_t *)arg;
973
974 seq_printf(m, "\n%-23s", "Number of blocks type ");
975 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
976 seq_printf(m, "%12s ", migratetype_names[mtype]);
977 seq_putc(m, '\n');
978 walk_zones_in_node(m, pgdat, pagetypeinfo_showblockcount_print);
979
980 return 0;
981}
982
983/*
984 * This prints out statistics in relation to grouping pages by mobility.
985 * It is expensive to collect so do not constantly read the file.
986 */
987static int pagetypeinfo_show(struct seq_file *m, void *arg)
988{
989 pg_data_t *pgdat = (pg_data_t *)arg;
990
KOSAKI Motohiro41b25a32008-04-30 00:52:13 -0700991 /* check memoryless node */
Lai Jiangshana47b53c2012-12-12 13:51:37 -0800992 if (!node_state(pgdat->node_id, N_MEMORY))
KOSAKI Motohiro41b25a32008-04-30 00:52:13 -0700993 return 0;
994
Mel Gorman467c9962007-10-16 01:26:02 -0700995 seq_printf(m, "Page block order: %d\n", pageblock_order);
996 seq_printf(m, "Pages per block: %lu\n", pageblock_nr_pages);
997 seq_putc(m, '\n');
998 pagetypeinfo_showfree(m, pgdat);
999 pagetypeinfo_showblockcount(m, pgdat);
1000
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001001 return 0;
1002}
1003
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +04001004static const struct seq_operations fragmentation_op = {
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001005 .start = frag_start,
1006 .next = frag_next,
1007 .stop = frag_stop,
1008 .show = frag_show,
1009};
1010
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +04001011static int fragmentation_open(struct inode *inode, struct file *file)
1012{
1013 return seq_open(file, &fragmentation_op);
1014}
1015
1016static const struct file_operations fragmentation_file_operations = {
1017 .open = fragmentation_open,
1018 .read = seq_read,
1019 .llseek = seq_lseek,
1020 .release = seq_release,
1021};
1022
Alexey Dobriyan74e2e8e2008-10-06 04:15:36 +04001023static const struct seq_operations pagetypeinfo_op = {
Mel Gorman467c9962007-10-16 01:26:02 -07001024 .start = frag_start,
1025 .next = frag_next,
1026 .stop = frag_stop,
1027 .show = pagetypeinfo_show,
1028};
1029
Alexey Dobriyan74e2e8e2008-10-06 04:15:36 +04001030static int pagetypeinfo_open(struct inode *inode, struct file *file)
1031{
1032 return seq_open(file, &pagetypeinfo_op);
1033}
1034
1035static const struct file_operations pagetypeinfo_file_ops = {
1036 .open = pagetypeinfo_open,
1037 .read = seq_read,
1038 .llseek = seq_lseek,
1039 .release = seq_release,
1040};
1041
Mel Gorman467c9962007-10-16 01:26:02 -07001042static void zoneinfo_show_print(struct seq_file *m, pg_data_t *pgdat,
1043 struct zone *zone)
1044{
1045 int i;
1046 seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
1047 seq_printf(m,
1048 "\n pages free %lu"
1049 "\n min %lu"
1050 "\n low %lu"
1051 "\n high %lu"
Wu Fengguang08d9ae72009-06-16 15:32:30 -07001052 "\n scanned %lu"
Mel Gorman467c9962007-10-16 01:26:02 -07001053 "\n spanned %lu"
Jiang Liu9feedc92012-12-12 13:52:12 -08001054 "\n present %lu"
1055 "\n managed %lu",
Mel Gorman88f5acf2011-01-13 15:45:41 -08001056 zone_page_state(zone, NR_FREE_PAGES),
Mel Gorman41858962009-06-16 15:32:12 -07001057 min_wmark_pages(zone),
1058 low_wmark_pages(zone),
1059 high_wmark_pages(zone),
Mel Gorman467c9962007-10-16 01:26:02 -07001060 zone->pages_scanned,
Mel Gorman467c9962007-10-16 01:26:02 -07001061 zone->spanned_pages,
Jiang Liu9feedc92012-12-12 13:52:12 -08001062 zone->present_pages,
1063 zone->managed_pages);
Mel Gorman467c9962007-10-16 01:26:02 -07001064
1065 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
1066 seq_printf(m, "\n %-12s %lu", vmstat_text[i],
1067 zone_page_state(zone, i));
1068
1069 seq_printf(m,
1070 "\n protection: (%lu",
1071 zone->lowmem_reserve[0]);
1072 for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
1073 seq_printf(m, ", %lu", zone->lowmem_reserve[i]);
1074 seq_printf(m,
1075 ")"
1076 "\n pagesets");
1077 for_each_online_cpu(i) {
1078 struct per_cpu_pageset *pageset;
Mel Gorman467c9962007-10-16 01:26:02 -07001079
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001080 pageset = per_cpu_ptr(zone->pageset, i);
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001081 seq_printf(m,
1082 "\n cpu: %i"
1083 "\n count: %i"
1084 "\n high: %i"
1085 "\n batch: %i",
1086 i,
1087 pageset->pcp.count,
1088 pageset->pcp.high,
1089 pageset->pcp.batch);
Mel Gorman467c9962007-10-16 01:26:02 -07001090#ifdef CONFIG_SMP
1091 seq_printf(m, "\n vm stats threshold: %d",
1092 pageset->stat_threshold);
1093#endif
1094 }
1095 seq_printf(m,
1096 "\n all_unreclaimable: %u"
Rik van Riel556adec2008-10-18 20:26:34 -07001097 "\n start_pfn: %lu"
1098 "\n inactive_ratio: %u",
Lisa Du6e543d52013-09-11 14:22:36 -07001099 !zone_reclaimable(zone),
Rik van Riel556adec2008-10-18 20:26:34 -07001100 zone->zone_start_pfn,
1101 zone->inactive_ratio);
Mel Gorman467c9962007-10-16 01:26:02 -07001102 seq_putc(m, '\n');
1103}
1104
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001105/*
1106 * Output information about zones in @pgdat.
1107 */
1108static int zoneinfo_show(struct seq_file *m, void *arg)
1109{
Mel Gorman467c9962007-10-16 01:26:02 -07001110 pg_data_t *pgdat = (pg_data_t *)arg;
1111 walk_zones_in_node(m, pgdat, zoneinfo_show_print);
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001112 return 0;
1113}
1114
Alexey Dobriyan5c9fe622008-10-06 04:19:42 +04001115static const struct seq_operations zoneinfo_op = {
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001116 .start = frag_start, /* iterate over all zones. The same as in
1117 * fragmentation. */
1118 .next = frag_next,
1119 .stop = frag_stop,
1120 .show = zoneinfo_show,
1121};
1122
Alexey Dobriyan5c9fe622008-10-06 04:19:42 +04001123static int zoneinfo_open(struct inode *inode, struct file *file)
1124{
1125 return seq_open(file, &zoneinfo_op);
1126}
1127
1128static const struct file_operations proc_zoneinfo_file_operations = {
1129 .open = zoneinfo_open,
1130 .read = seq_read,
1131 .llseek = seq_lseek,
1132 .release = seq_release,
1133};
1134
Michael Rubin79da8262010-10-26 14:21:36 -07001135enum writeback_stat_item {
1136 NR_DIRTY_THRESHOLD,
1137 NR_DIRTY_BG_THRESHOLD,
1138 NR_VM_WRITEBACK_STAT_ITEMS,
1139};
1140
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001141static void *vmstat_start(struct seq_file *m, loff_t *pos)
1142{
Christoph Lameter2244b952006-06-30 01:55:33 -07001143 unsigned long *v;
Michael Rubin79da8262010-10-26 14:21:36 -07001144 int i, stat_items_size;
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001145
1146 if (*pos >= ARRAY_SIZE(vmstat_text))
1147 return NULL;
Michael Rubin79da8262010-10-26 14:21:36 -07001148 stat_items_size = NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long) +
1149 NR_VM_WRITEBACK_STAT_ITEMS * sizeof(unsigned long);
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001150
Christoph Lameterf8891e52006-06-30 01:55:45 -07001151#ifdef CONFIG_VM_EVENT_COUNTERS
Michael Rubin79da8262010-10-26 14:21:36 -07001152 stat_items_size += sizeof(struct vm_event_state);
Christoph Lameterf8891e52006-06-30 01:55:45 -07001153#endif
Michael Rubin79da8262010-10-26 14:21:36 -07001154
1155 v = kmalloc(stat_items_size, GFP_KERNEL);
Christoph Lameter2244b952006-06-30 01:55:33 -07001156 m->private = v;
1157 if (!v)
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001158 return ERR_PTR(-ENOMEM);
Christoph Lameter2244b952006-06-30 01:55:33 -07001159 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
1160 v[i] = global_page_state(i);
Michael Rubin79da8262010-10-26 14:21:36 -07001161 v += NR_VM_ZONE_STAT_ITEMS;
1162
1163 global_dirty_limits(v + NR_DIRTY_BG_THRESHOLD,
1164 v + NR_DIRTY_THRESHOLD);
1165 v += NR_VM_WRITEBACK_STAT_ITEMS;
1166
Christoph Lameterf8891e52006-06-30 01:55:45 -07001167#ifdef CONFIG_VM_EVENT_COUNTERS
Michael Rubin79da8262010-10-26 14:21:36 -07001168 all_vm_events(v);
1169 v[PGPGIN] /= 2; /* sectors -> kbytes */
1170 v[PGPGOUT] /= 2;
Christoph Lameterf8891e52006-06-30 01:55:45 -07001171#endif
Wu Fengguangff8b16d2010-11-04 01:56:49 +08001172 return (unsigned long *)m->private + *pos;
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001173}
1174
1175static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
1176{
1177 (*pos)++;
1178 if (*pos >= ARRAY_SIZE(vmstat_text))
1179 return NULL;
1180 return (unsigned long *)m->private + *pos;
1181}
1182
1183static int vmstat_show(struct seq_file *m, void *arg)
1184{
1185 unsigned long *l = arg;
1186 unsigned long off = l - (unsigned long *)m->private;
1187
1188 seq_printf(m, "%s %lu\n", vmstat_text[off], *l);
1189 return 0;
1190}
1191
1192static void vmstat_stop(struct seq_file *m, void *arg)
1193{
1194 kfree(m->private);
1195 m->private = NULL;
1196}
1197
Alexey Dobriyanb6aa44a2008-10-06 04:17:48 +04001198static const struct seq_operations vmstat_op = {
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001199 .start = vmstat_start,
1200 .next = vmstat_next,
1201 .stop = vmstat_stop,
1202 .show = vmstat_show,
1203};
1204
Alexey Dobriyanb6aa44a2008-10-06 04:17:48 +04001205static int vmstat_open(struct inode *inode, struct file *file)
1206{
1207 return seq_open(file, &vmstat_op);
1208}
1209
1210static const struct file_operations proc_vmstat_file_operations = {
1211 .open = vmstat_open,
1212 .read = seq_read,
1213 .llseek = seq_lseek,
1214 .release = seq_release,
1215};
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001216#endif /* CONFIG_PROC_FS */
1217
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001218#ifdef CONFIG_SMP
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001219static DEFINE_PER_CPU(struct delayed_work, vmstat_work);
Christoph Lameter77461ab2007-05-09 02:35:13 -07001220int sysctl_stat_interval __read_mostly = HZ;
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001221
1222static void vmstat_update(struct work_struct *w)
1223{
Christoph Lameterfbc2edb2013-09-11 14:21:32 -07001224 refresh_cpu_vm_stats();
Christoph Lameter77461ab2007-05-09 02:35:13 -07001225 schedule_delayed_work(&__get_cpu_var(vmstat_work),
Anton Blanchard98f4ebb2009-04-02 16:56:39 -07001226 round_jiffies_relative(sysctl_stat_interval));
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001227}
1228
Paul Gortmaker0db06282013-06-19 14:53:51 -04001229static void start_cpu_timer(int cpu)
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001230{
Tejun Heo1871e522009-10-29 22:34:13 +09001231 struct delayed_work *work = &per_cpu(vmstat_work, cpu);
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001232
Tejun Heo203b42f2012-08-21 13:18:23 -07001233 INIT_DEFERRABLE_WORK(work, vmstat_update);
Tejun Heo1871e522009-10-29 22:34:13 +09001234 schedule_delayed_work_on(cpu, work, __round_jiffies_relative(HZ, cpu));
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001235}
1236
Toshi Kani807a1bd2013-11-12 15:08:13 -08001237static void vmstat_cpu_dead(int node)
1238{
1239 int cpu;
1240
1241 get_online_cpus();
1242 for_each_online_cpu(cpu)
1243 if (cpu_to_node(cpu) == node)
1244 goto end;
1245
1246 node_clear_state(node, N_CPU);
1247end:
1248 put_online_cpus();
1249}
1250
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001251/*
1252 * Use the cpu notifier to insure that the thresholds are recalculated
1253 * when necessary.
1254 */
Paul Gortmaker0db06282013-06-19 14:53:51 -04001255static int vmstat_cpuup_callback(struct notifier_block *nfb,
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001256 unsigned long action,
1257 void *hcpu)
1258{
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001259 long cpu = (long)hcpu;
1260
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001261 switch (action) {
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001262 case CPU_ONLINE:
1263 case CPU_ONLINE_FROZEN:
KAMEZAWA Hiroyuki5ee28a42010-09-09 16:38:14 -07001264 refresh_zone_stat_thresholds();
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001265 start_cpu_timer(cpu);
Christoph Lameterad596922010-01-05 15:34:51 +09001266 node_set_state(cpu_to_node(cpu), N_CPU);
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001267 break;
1268 case CPU_DOWN_PREPARE:
1269 case CPU_DOWN_PREPARE_FROZEN:
Tejun Heoafe2c512010-12-14 16:21:17 +01001270 cancel_delayed_work_sync(&per_cpu(vmstat_work, cpu));
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001271 per_cpu(vmstat_work, cpu).work.func = NULL;
1272 break;
1273 case CPU_DOWN_FAILED:
1274 case CPU_DOWN_FAILED_FROZEN:
1275 start_cpu_timer(cpu);
1276 break;
Andy Whitcroftce421c72006-12-06 20:33:08 -08001277 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07001278 case CPU_DEAD_FROZEN:
Andy Whitcroftce421c72006-12-06 20:33:08 -08001279 refresh_zone_stat_thresholds();
Toshi Kani807a1bd2013-11-12 15:08:13 -08001280 vmstat_cpu_dead(cpu_to_node(cpu));
Andy Whitcroftce421c72006-12-06 20:33:08 -08001281 break;
1282 default:
1283 break;
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001284 }
1285 return NOTIFY_OK;
1286}
1287
Paul Gortmaker0db06282013-06-19 14:53:51 -04001288static struct notifier_block vmstat_notifier =
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001289 { &vmstat_cpuup_callback, NULL, 0 };
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +04001290#endif
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001291
Adrian Bunke2fc88d2007-10-16 01:26:27 -07001292static int __init setup_vmstat(void)
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001293{
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +04001294#ifdef CONFIG_SMP
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001295 int cpu;
1296
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001297 register_cpu_notifier(&vmstat_notifier);
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001298
Toshi Kanid7e0b372013-11-12 15:08:12 -08001299 get_online_cpus();
1300 for_each_online_cpu(cpu) {
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001301 start_cpu_timer(cpu);
Toshi Kanid7e0b372013-11-12 15:08:12 -08001302 node_set_state(cpu_to_node(cpu), N_CPU);
1303 }
1304 put_online_cpus();
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +04001305#endif
1306#ifdef CONFIG_PROC_FS
1307 proc_create("buddyinfo", S_IRUGO, NULL, &fragmentation_file_operations);
Alexey Dobriyan74e2e8e2008-10-06 04:15:36 +04001308 proc_create("pagetypeinfo", S_IRUGO, NULL, &pagetypeinfo_file_ops);
Alexey Dobriyanb6aa44a2008-10-06 04:17:48 +04001309 proc_create("vmstat", S_IRUGO, NULL, &proc_vmstat_file_operations);
Alexey Dobriyan5c9fe622008-10-06 04:19:42 +04001310 proc_create("zoneinfo", S_IRUGO, NULL, &proc_zoneinfo_file_operations);
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +04001311#endif
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001312 return 0;
1313}
1314module_init(setup_vmstat)
Mel Gormand7a57522010-05-24 14:32:25 -07001315
1316#if defined(CONFIG_DEBUG_FS) && defined(CONFIG_COMPACTION)
1317#include <linux/debugfs.h>
1318
Mel Gormand7a57522010-05-24 14:32:25 -07001319
1320/*
1321 * Return an index indicating how much of the available free memory is
1322 * unusable for an allocation of the requested size.
1323 */
1324static int unusable_free_index(unsigned int order,
1325 struct contig_page_info *info)
1326{
1327 /* No free memory is interpreted as all free memory is unusable */
1328 if (info->free_pages == 0)
1329 return 1000;
1330
1331 /*
1332 * Index should be a value between 0 and 1. Return a value to 3
1333 * decimal places.
1334 *
1335 * 0 => no fragmentation
1336 * 1 => high fragmentation
1337 */
1338 return div_u64((info->free_pages - (info->free_blocks_suitable << order)) * 1000ULL, info->free_pages);
1339
1340}
1341
1342static void unusable_show_print(struct seq_file *m,
1343 pg_data_t *pgdat, struct zone *zone)
1344{
1345 unsigned int order;
1346 int index;
1347 struct contig_page_info info;
1348
1349 seq_printf(m, "Node %d, zone %8s ",
1350 pgdat->node_id,
1351 zone->name);
1352 for (order = 0; order < MAX_ORDER; ++order) {
1353 fill_contig_page_info(zone, order, &info);
1354 index = unusable_free_index(order, &info);
1355 seq_printf(m, "%d.%03d ", index / 1000, index % 1000);
1356 }
1357
1358 seq_putc(m, '\n');
1359}
1360
1361/*
1362 * Display unusable free space index
1363 *
1364 * The unusable free space index measures how much of the available free
1365 * memory cannot be used to satisfy an allocation of a given size and is a
1366 * value between 0 and 1. The higher the value, the more of free memory is
1367 * unusable and by implication, the worse the external fragmentation is. This
1368 * can be expressed as a percentage by multiplying by 100.
1369 */
1370static int unusable_show(struct seq_file *m, void *arg)
1371{
1372 pg_data_t *pgdat = (pg_data_t *)arg;
1373
1374 /* check memoryless node */
Lai Jiangshana47b53c2012-12-12 13:51:37 -08001375 if (!node_state(pgdat->node_id, N_MEMORY))
Mel Gormand7a57522010-05-24 14:32:25 -07001376 return 0;
1377
1378 walk_zones_in_node(m, pgdat, unusable_show_print);
1379
1380 return 0;
1381}
1382
1383static const struct seq_operations unusable_op = {
1384 .start = frag_start,
1385 .next = frag_next,
1386 .stop = frag_stop,
1387 .show = unusable_show,
1388};
1389
1390static int unusable_open(struct inode *inode, struct file *file)
1391{
1392 return seq_open(file, &unusable_op);
1393}
1394
1395static const struct file_operations unusable_file_ops = {
1396 .open = unusable_open,
1397 .read = seq_read,
1398 .llseek = seq_lseek,
1399 .release = seq_release,
1400};
1401
Mel Gormanf1a5ab12010-05-24 14:32:26 -07001402static void extfrag_show_print(struct seq_file *m,
1403 pg_data_t *pgdat, struct zone *zone)
1404{
1405 unsigned int order;
1406 int index;
1407
1408 /* Alloc on stack as interrupts are disabled for zone walk */
1409 struct contig_page_info info;
1410
1411 seq_printf(m, "Node %d, zone %8s ",
1412 pgdat->node_id,
1413 zone->name);
1414 for (order = 0; order < MAX_ORDER; ++order) {
1415 fill_contig_page_info(zone, order, &info);
Mel Gorman56de7262010-05-24 14:32:30 -07001416 index = __fragmentation_index(order, &info);
Mel Gormanf1a5ab12010-05-24 14:32:26 -07001417 seq_printf(m, "%d.%03d ", index / 1000, index % 1000);
1418 }
1419
1420 seq_putc(m, '\n');
1421}
1422
1423/*
1424 * Display fragmentation index for orders that allocations would fail for
1425 */
1426static int extfrag_show(struct seq_file *m, void *arg)
1427{
1428 pg_data_t *pgdat = (pg_data_t *)arg;
1429
1430 walk_zones_in_node(m, pgdat, extfrag_show_print);
1431
1432 return 0;
1433}
1434
1435static const struct seq_operations extfrag_op = {
1436 .start = frag_start,
1437 .next = frag_next,
1438 .stop = frag_stop,
1439 .show = extfrag_show,
1440};
1441
1442static int extfrag_open(struct inode *inode, struct file *file)
1443{
1444 return seq_open(file, &extfrag_op);
1445}
1446
1447static const struct file_operations extfrag_file_ops = {
1448 .open = extfrag_open,
1449 .read = seq_read,
1450 .llseek = seq_lseek,
1451 .release = seq_release,
1452};
1453
Mel Gormand7a57522010-05-24 14:32:25 -07001454static int __init extfrag_debug_init(void)
1455{
Sasikantha babubde8bd82012-05-29 15:06:22 -07001456 struct dentry *extfrag_debug_root;
1457
Mel Gormand7a57522010-05-24 14:32:25 -07001458 extfrag_debug_root = debugfs_create_dir("extfrag", NULL);
1459 if (!extfrag_debug_root)
1460 return -ENOMEM;
1461
1462 if (!debugfs_create_file("unusable_index", 0444,
1463 extfrag_debug_root, NULL, &unusable_file_ops))
Sasikantha babubde8bd82012-05-29 15:06:22 -07001464 goto fail;
Mel Gormand7a57522010-05-24 14:32:25 -07001465
Mel Gormanf1a5ab12010-05-24 14:32:26 -07001466 if (!debugfs_create_file("extfrag_index", 0444,
1467 extfrag_debug_root, NULL, &extfrag_file_ops))
Sasikantha babubde8bd82012-05-29 15:06:22 -07001468 goto fail;
Mel Gormanf1a5ab12010-05-24 14:32:26 -07001469
Mel Gormand7a57522010-05-24 14:32:25 -07001470 return 0;
Sasikantha babubde8bd82012-05-29 15:06:22 -07001471fail:
1472 debugfs_remove_recursive(extfrag_debug_root);
1473 return -ENOMEM;
Mel Gormand7a57522010-05-24 14:32:25 -07001474}
1475
1476module_init(extfrag_debug_init);
1477#endif