blob: 084c6725b3734430483e7ea4fcf74e2ab67f7bfa [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 Lameter7cc36bb2014-10-09 15:29:43 -070010 * Copyright (C) 2008-2014 Christoph Lameter
Christoph Lameterf6ac2352006-06-30 01:55:32 -070011 */
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +040012#include <linux/fs.h>
Christoph Lameterf6ac2352006-06-30 01:55:32 -070013#include <linux/mm.h>
Alexey Dobriyan4e950f62007-07-30 02:36:13 +040014#include <linux/err.h>
Christoph Lameter2244b952006-06-30 01:55:33 -070015#include <linux/module.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090016#include <linux/slab.h>
Christoph Lameterdf9ecab2006-08-31 21:27:35 -070017#include <linux/cpu.h>
Christoph Lameter7cc36bb2014-10-09 15:29:43 -070018#include <linux/cpumask.h>
Adrian Bunkc748e132008-07-23 21:27:03 -070019#include <linux/vmstat.h>
Andrew Morton3c486872015-02-10 14:09:43 -080020#include <linux/proc_fs.h>
21#include <linux/seq_file.h>
22#include <linux/debugfs.h>
Alexey Dobriyane8edc6e2007-05-21 01:22:52 +040023#include <linux/sched.h>
Mel Gormanf1a5ab12010-05-24 14:32:26 -070024#include <linux/math64.h>
Michael Rubin79da8262010-10-26 14:21:36 -070025#include <linux/writeback.h>
Namhyung Kim36deb0b2010-10-26 14:22:04 -070026#include <linux/compaction.h>
Lisa Du6e543d52013-09-11 14:22:36 -070027#include <linux/mm_inline.h>
Joonsoo Kim48c96a32014-12-12 16:56:01 -080028#include <linux/page_ext.h>
29#include <linux/page_owner.h>
Lisa Du6e543d52013-09-11 14:22:36 -070030
31#include "internal.h"
Christoph Lameterf6ac2352006-06-30 01:55:32 -070032
Christoph Lameterf8891e52006-06-30 01:55:45 -070033#ifdef CONFIG_VM_EVENT_COUNTERS
34DEFINE_PER_CPU(struct vm_event_state, vm_event_states) = {{0}};
35EXPORT_PER_CPU_SYMBOL(vm_event_states);
36
Minchan Kim31f961a2010-08-09 17:18:59 -070037static void sum_vm_events(unsigned long *ret)
Christoph Lameterf8891e52006-06-30 01:55:45 -070038{
Christoph Lameter9eccf2a2008-02-04 22:29:22 -080039 int cpu;
Christoph Lameterf8891e52006-06-30 01:55:45 -070040 int i;
41
42 memset(ret, 0, NR_VM_EVENT_ITEMS * sizeof(unsigned long));
43
Minchan Kim31f961a2010-08-09 17:18:59 -070044 for_each_online_cpu(cpu) {
Christoph Lameterf8891e52006-06-30 01:55:45 -070045 struct vm_event_state *this = &per_cpu(vm_event_states, cpu);
46
Christoph Lameterf8891e52006-06-30 01:55:45 -070047 for (i = 0; i < NR_VM_EVENT_ITEMS; i++)
48 ret[i] += this->event[i];
49 }
50}
51
52/*
53 * Accumulate the vm event counters across all CPUs.
54 * The result is unavoidably approximate - it can change
55 * during and after execution of this function.
56*/
57void all_vm_events(unsigned long *ret)
58{
KOSAKI Motohirob5be1132008-05-12 14:02:06 -070059 get_online_cpus();
Minchan Kim31f961a2010-08-09 17:18:59 -070060 sum_vm_events(ret);
KOSAKI Motohirob5be1132008-05-12 14:02:06 -070061 put_online_cpus();
Christoph Lameterf8891e52006-06-30 01:55:45 -070062}
Heiko Carstens32dd66f2006-07-10 04:44:31 -070063EXPORT_SYMBOL_GPL(all_vm_events);
Christoph Lameterf8891e52006-06-30 01:55:45 -070064
Christoph Lameterf8891e52006-06-30 01:55:45 -070065/*
66 * Fold the foreign cpu events into our own.
67 *
68 * This is adding to the events on one processor
69 * but keeps the global counts constant.
70 */
71void vm_events_fold_cpu(int cpu)
72{
73 struct vm_event_state *fold_state = &per_cpu(vm_event_states, cpu);
74 int i;
75
76 for (i = 0; i < NR_VM_EVENT_ITEMS; i++) {
77 count_vm_events(i, fold_state->event[i]);
78 fold_state->event[i] = 0;
79 }
80}
Christoph Lameterf8891e52006-06-30 01:55:45 -070081
82#endif /* CONFIG_VM_EVENT_COUNTERS */
83
Christoph Lameter2244b952006-06-30 01:55:33 -070084/*
85 * Manage combined zone based / global counters
86 *
87 * vm_stat contains the global counters
88 */
Dimitri Sivanicha1cb2c62011-10-31 17:09:46 -070089atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS] __cacheline_aligned_in_smp;
Christoph Lameter2244b952006-06-30 01:55:33 -070090EXPORT_SYMBOL(vm_stat);
91
92#ifdef CONFIG_SMP
93
Mel Gormanb44129b2011-01-13 15:45:43 -080094int calculate_pressure_threshold(struct zone *zone)
Mel Gorman88f5acf2011-01-13 15:45:41 -080095{
96 int threshold;
97 int watermark_distance;
98
99 /*
100 * As vmstats are not up to date, there is drift between the estimated
101 * and real values. For high thresholds and a high number of CPUs, it
102 * is possible for the min watermark to be breached while the estimated
103 * value looks fine. The pressure threshold is a reduced value such
104 * that even the maximum amount of drift will not accidentally breach
105 * the min watermark
106 */
107 watermark_distance = low_wmark_pages(zone) - min_wmark_pages(zone);
108 threshold = max(1, (int)(watermark_distance / num_online_cpus()));
109
110 /*
111 * Maximum threshold is 125
112 */
113 threshold = min(125, threshold);
114
115 return threshold;
116}
117
Mel Gormanb44129b2011-01-13 15:45:43 -0800118int calculate_normal_threshold(struct zone *zone)
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700119{
120 int threshold;
121 int mem; /* memory in 128 MB units */
122
123 /*
124 * The threshold scales with the number of processors and the amount
125 * of memory per zone. More memory means that we can defer updates for
126 * longer, more processors could lead to more contention.
127 * fls() is used to have a cheap way of logarithmic scaling.
128 *
129 * Some sample thresholds:
130 *
131 * Threshold Processors (fls) Zonesize fls(mem+1)
132 * ------------------------------------------------------------------
133 * 8 1 1 0.9-1 GB 4
134 * 16 2 2 0.9-1 GB 4
135 * 20 2 2 1-2 GB 5
136 * 24 2 2 2-4 GB 6
137 * 28 2 2 4-8 GB 7
138 * 32 2 2 8-16 GB 8
139 * 4 2 2 <128M 1
140 * 30 4 3 2-4 GB 5
141 * 48 4 3 8-16 GB 8
142 * 32 8 4 1-2 GB 4
143 * 32 8 4 0.9-1GB 4
144 * 10 16 5 <128M 1
145 * 40 16 5 900M 4
146 * 70 64 7 2-4 GB 5
147 * 84 64 7 4-8 GB 6
148 * 108 512 9 4-8 GB 6
149 * 125 1024 10 8-16 GB 8
150 * 125 1024 10 16-32 GB 9
151 */
152
Jiang Liub40da042013-02-22 16:33:52 -0800153 mem = zone->managed_pages >> (27 - PAGE_SHIFT);
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700154
155 threshold = 2 * fls(num_online_cpus()) * (1 + fls(mem));
156
157 /*
158 * Maximum threshold is 125
159 */
160 threshold = min(125, threshold);
161
162 return threshold;
163}
Christoph Lameter2244b952006-06-30 01:55:33 -0700164
165/*
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700166 * Refresh the thresholds for each zone.
Christoph Lameter2244b952006-06-30 01:55:33 -0700167 */
KOSAKI Motohiroa6cccdc2011-05-24 17:11:33 -0700168void refresh_zone_stat_thresholds(void)
Christoph Lameter2244b952006-06-30 01:55:33 -0700169{
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700170 struct zone *zone;
171 int cpu;
172 int threshold;
173
KOSAKI Motohiroee99c712009-03-31 15:19:31 -0700174 for_each_populated_zone(zone) {
Christoph Lameteraa454842010-09-09 16:38:17 -0700175 unsigned long max_drift, tolerate_drift;
176
Mel Gormanb44129b2011-01-13 15:45:43 -0800177 threshold = calculate_normal_threshold(zone);
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700178
179 for_each_online_cpu(cpu)
Christoph Lameter99dcc3e2010-01-05 15:34:51 +0900180 per_cpu_ptr(zone->pageset, cpu)->stat_threshold
181 = threshold;
Christoph Lameteraa454842010-09-09 16:38:17 -0700182
183 /*
184 * Only set percpu_drift_mark if there is a danger that
185 * NR_FREE_PAGES reports the low watermark is ok when in fact
186 * the min watermark could be breached by an allocation
187 */
188 tolerate_drift = low_wmark_pages(zone) - min_wmark_pages(zone);
189 max_drift = num_online_cpus() * threshold;
190 if (max_drift > tolerate_drift)
191 zone->percpu_drift_mark = high_wmark_pages(zone) +
192 max_drift;
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700193 }
Christoph Lameter2244b952006-06-30 01:55:33 -0700194}
195
Mel Gormanb44129b2011-01-13 15:45:43 -0800196void set_pgdat_percpu_threshold(pg_data_t *pgdat,
197 int (*calculate_pressure)(struct zone *))
Mel Gorman88f5acf2011-01-13 15:45:41 -0800198{
199 struct zone *zone;
200 int cpu;
201 int threshold;
202 int i;
203
Mel Gorman88f5acf2011-01-13 15:45:41 -0800204 for (i = 0; i < pgdat->nr_zones; i++) {
205 zone = &pgdat->node_zones[i];
206 if (!zone->percpu_drift_mark)
207 continue;
208
Mel Gormanb44129b2011-01-13 15:45:43 -0800209 threshold = (*calculate_pressure)(zone);
Mel Gormanbb0b6df2014-08-06 16:07:18 -0700210 for_each_online_cpu(cpu)
Mel Gorman88f5acf2011-01-13 15:45:41 -0800211 per_cpu_ptr(zone->pageset, cpu)->stat_threshold
212 = threshold;
213 }
Mel Gorman88f5acf2011-01-13 15:45:41 -0800214}
215
Christoph Lameter2244b952006-06-30 01:55:33 -0700216/*
Jianyu Zhanbea04b02014-06-04 16:09:51 -0700217 * For use when we know that interrupts are disabled,
218 * or when we know that preemption is disabled and that
219 * particular counter cannot be updated from interrupt context.
Christoph Lameter2244b952006-06-30 01:55:33 -0700220 */
221void __mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
Heiko Carstens6cdb18a2015-12-29 14:54:32 -0800222 long delta)
Christoph Lameter2244b952006-06-30 01:55:33 -0700223{
Christoph Lameter12938a92010-12-06 11:16:20 -0600224 struct per_cpu_pageset __percpu *pcp = zone->pageset;
225 s8 __percpu *p = pcp->vm_stat_diff + item;
Christoph Lameter2244b952006-06-30 01:55:33 -0700226 long x;
Christoph Lameter12938a92010-12-06 11:16:20 -0600227 long t;
Christoph Lameter2244b952006-06-30 01:55:33 -0700228
Christoph Lameter12938a92010-12-06 11:16:20 -0600229 x = delta + __this_cpu_read(*p);
Christoph Lameter2244b952006-06-30 01:55:33 -0700230
Christoph Lameter12938a92010-12-06 11:16:20 -0600231 t = __this_cpu_read(pcp->stat_threshold);
232
233 if (unlikely(x > t || x < -t)) {
Christoph Lameter2244b952006-06-30 01:55:33 -0700234 zone_page_state_add(x, zone, item);
235 x = 0;
236 }
Christoph Lameter12938a92010-12-06 11:16:20 -0600237 __this_cpu_write(*p, x);
Christoph Lameter2244b952006-06-30 01:55:33 -0700238}
239EXPORT_SYMBOL(__mod_zone_page_state);
240
241/*
Christoph Lameter2244b952006-06-30 01:55:33 -0700242 * Optimized increment and decrement functions.
243 *
244 * These are only for a single page and therefore can take a struct page *
245 * argument instead of struct zone *. This allows the inclusion of the code
246 * generated for page_zone(page) into the optimized functions.
247 *
248 * No overflow check is necessary and therefore the differential can be
249 * incremented or decremented in place which may allow the compilers to
250 * generate better code.
Christoph Lameter2244b952006-06-30 01:55:33 -0700251 * The increment or decrement is known and therefore one boundary check can
252 * be omitted.
253 *
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700254 * NOTE: These functions are very performance sensitive. Change only
255 * with care.
256 *
Christoph Lameter2244b952006-06-30 01:55:33 -0700257 * Some processors have inc/dec instructions that are atomic vs an interrupt.
258 * However, the code must first determine the differential location in a zone
259 * based on the processor number and then inc/dec the counter. There is no
260 * guarantee without disabling preemption that the processor will not change
261 * in between and therefore the atomicity vs. interrupt cannot be exploited
262 * in a useful way here.
263 */
Christoph Lameterc8785382007-02-10 01:43:01 -0800264void __inc_zone_state(struct zone *zone, enum zone_stat_item item)
Christoph Lameter2244b952006-06-30 01:55:33 -0700265{
Christoph Lameter12938a92010-12-06 11:16:20 -0600266 struct per_cpu_pageset __percpu *pcp = zone->pageset;
267 s8 __percpu *p = pcp->vm_stat_diff + item;
268 s8 v, t;
Christoph Lameter2244b952006-06-30 01:55:33 -0700269
Christoph Lameter908ee0f2010-12-06 11:40:02 -0600270 v = __this_cpu_inc_return(*p);
Christoph Lameter12938a92010-12-06 11:16:20 -0600271 t = __this_cpu_read(pcp->stat_threshold);
272 if (unlikely(v > t)) {
273 s8 overstep = t >> 1;
Christoph Lameter2244b952006-06-30 01:55:33 -0700274
Christoph Lameter12938a92010-12-06 11:16:20 -0600275 zone_page_state_add(v + overstep, zone, item);
276 __this_cpu_write(*p, -overstep);
Christoph Lameter2244b952006-06-30 01:55:33 -0700277 }
278}
Christoph Lameterca889e62006-06-30 01:55:44 -0700279
280void __inc_zone_page_state(struct page *page, enum zone_stat_item item)
281{
282 __inc_zone_state(page_zone(page), item);
283}
Christoph Lameter2244b952006-06-30 01:55:33 -0700284EXPORT_SYMBOL(__inc_zone_page_state);
285
Christoph Lameterc8785382007-02-10 01:43:01 -0800286void __dec_zone_state(struct zone *zone, enum zone_stat_item item)
Christoph Lameter2244b952006-06-30 01:55:33 -0700287{
Christoph Lameter12938a92010-12-06 11:16:20 -0600288 struct per_cpu_pageset __percpu *pcp = zone->pageset;
289 s8 __percpu *p = pcp->vm_stat_diff + item;
290 s8 v, t;
Christoph Lameter2244b952006-06-30 01:55:33 -0700291
Christoph Lameter908ee0f2010-12-06 11:40:02 -0600292 v = __this_cpu_dec_return(*p);
Christoph Lameter12938a92010-12-06 11:16:20 -0600293 t = __this_cpu_read(pcp->stat_threshold);
294 if (unlikely(v < - t)) {
295 s8 overstep = t >> 1;
Christoph Lameter2244b952006-06-30 01:55:33 -0700296
Christoph Lameter12938a92010-12-06 11:16:20 -0600297 zone_page_state_add(v - overstep, zone, item);
298 __this_cpu_write(*p, overstep);
Christoph Lameter2244b952006-06-30 01:55:33 -0700299 }
300}
Christoph Lameterc8785382007-02-10 01:43:01 -0800301
302void __dec_zone_page_state(struct page *page, enum zone_stat_item item)
303{
304 __dec_zone_state(page_zone(page), item);
305}
Christoph Lameter2244b952006-06-30 01:55:33 -0700306EXPORT_SYMBOL(__dec_zone_page_state);
307
Heiko Carstens41561532012-01-12 17:17:30 -0800308#ifdef CONFIG_HAVE_CMPXCHG_LOCAL
Christoph Lameter7c839122010-12-14 10:28:46 -0600309/*
310 * If we have cmpxchg_local support then we do not need to incur the overhead
311 * that comes with local_irq_save/restore if we use this_cpu_cmpxchg.
312 *
313 * mod_state() modifies the zone counter state through atomic per cpu
314 * operations.
315 *
316 * Overstep mode specifies how overstep should handled:
317 * 0 No overstepping
318 * 1 Overstepping half of threshold
319 * -1 Overstepping minus half of threshold
320*/
Heiko Carstens6cdb18a2015-12-29 14:54:32 -0800321static inline void mod_state(struct zone *zone, enum zone_stat_item item,
322 long delta, int overstep_mode)
Christoph Lameter7c839122010-12-14 10:28:46 -0600323{
324 struct per_cpu_pageset __percpu *pcp = zone->pageset;
325 s8 __percpu *p = pcp->vm_stat_diff + item;
326 long o, n, t, z;
327
328 do {
329 z = 0; /* overflow to zone counters */
330
331 /*
332 * The fetching of the stat_threshold is racy. We may apply
333 * a counter threshold to the wrong the cpu if we get
Christoph Lameterd3bc2362011-04-14 15:21:58 -0700334 * rescheduled while executing here. However, the next
335 * counter update will apply the threshold again and
336 * therefore bring the counter under the threshold again.
337 *
338 * Most of the time the thresholds are the same anyways
339 * for all cpus in a zone.
Christoph Lameter7c839122010-12-14 10:28:46 -0600340 */
341 t = this_cpu_read(pcp->stat_threshold);
342
343 o = this_cpu_read(*p);
344 n = delta + o;
345
346 if (n > t || n < -t) {
347 int os = overstep_mode * (t >> 1) ;
348
349 /* Overflow must be added to zone counters */
350 z = n + os;
351 n = -os;
352 }
353 } while (this_cpu_cmpxchg(*p, o, n) != o);
354
355 if (z)
356 zone_page_state_add(z, zone, item);
357}
358
359void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
Heiko Carstens6cdb18a2015-12-29 14:54:32 -0800360 long delta)
Christoph Lameter7c839122010-12-14 10:28:46 -0600361{
362 mod_state(zone, item, delta, 0);
363}
364EXPORT_SYMBOL(mod_zone_page_state);
365
366void inc_zone_state(struct zone *zone, enum zone_stat_item item)
367{
368 mod_state(zone, item, 1, 1);
369}
370
371void inc_zone_page_state(struct page *page, enum zone_stat_item item)
372{
373 mod_state(page_zone(page), item, 1, 1);
374}
375EXPORT_SYMBOL(inc_zone_page_state);
376
377void dec_zone_page_state(struct page *page, enum zone_stat_item item)
378{
379 mod_state(page_zone(page), item, -1, -1);
380}
381EXPORT_SYMBOL(dec_zone_page_state);
382#else
383/*
384 * Use interrupt disable to serialize counter updates
385 */
386void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
Heiko Carstens6cdb18a2015-12-29 14:54:32 -0800387 long delta)
Christoph Lameter7c839122010-12-14 10:28:46 -0600388{
389 unsigned long flags;
390
391 local_irq_save(flags);
392 __mod_zone_page_state(zone, item, delta);
393 local_irq_restore(flags);
394}
395EXPORT_SYMBOL(mod_zone_page_state);
396
Christoph Lameterca889e62006-06-30 01:55:44 -0700397void inc_zone_state(struct zone *zone, enum zone_stat_item item)
398{
399 unsigned long flags;
400
401 local_irq_save(flags);
402 __inc_zone_state(zone, item);
403 local_irq_restore(flags);
404}
405
Christoph Lameter2244b952006-06-30 01:55:33 -0700406void inc_zone_page_state(struct page *page, enum zone_stat_item item)
407{
408 unsigned long flags;
409 struct zone *zone;
Christoph Lameter2244b952006-06-30 01:55:33 -0700410
411 zone = page_zone(page);
412 local_irq_save(flags);
Christoph Lameterca889e62006-06-30 01:55:44 -0700413 __inc_zone_state(zone, item);
Christoph Lameter2244b952006-06-30 01:55:33 -0700414 local_irq_restore(flags);
415}
416EXPORT_SYMBOL(inc_zone_page_state);
417
418void dec_zone_page_state(struct page *page, enum zone_stat_item item)
419{
420 unsigned long flags;
Christoph Lameter2244b952006-06-30 01:55:33 -0700421
Christoph Lameter2244b952006-06-30 01:55:33 -0700422 local_irq_save(flags);
Christoph Lametera302eb42006-08-31 21:27:34 -0700423 __dec_zone_page_state(page, item);
Christoph Lameter2244b952006-06-30 01:55:33 -0700424 local_irq_restore(flags);
425}
426EXPORT_SYMBOL(dec_zone_page_state);
Christoph Lameter7c839122010-12-14 10:28:46 -0600427#endif
Christoph Lameter2244b952006-06-30 01:55:33 -0700428
Christoph Lameter7cc36bb2014-10-09 15:29:43 -0700429
430/*
431 * Fold a differential into the global counters.
432 * Returns the number of counters updated.
433 */
434static int fold_diff(int *diff)
Christoph Lameter4edb0742013-09-11 14:21:31 -0700435{
436 int i;
Christoph Lameter7cc36bb2014-10-09 15:29:43 -0700437 int changes = 0;
Christoph Lameter4edb0742013-09-11 14:21:31 -0700438
439 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
Christoph Lameter7cc36bb2014-10-09 15:29:43 -0700440 if (diff[i]) {
Christoph Lameter4edb0742013-09-11 14:21:31 -0700441 atomic_long_add(diff[i], &vm_stat[i]);
Christoph Lameter7cc36bb2014-10-09 15:29:43 -0700442 changes++;
443 }
444 return changes;
Christoph Lameter4edb0742013-09-11 14:21:31 -0700445}
446
Christoph Lameter2244b952006-06-30 01:55:33 -0700447/*
Christoph Lameter2bb921e2013-09-11 14:21:30 -0700448 * Update the zone counters for the current cpu.
Christoph Lametera7f75e22008-02-04 22:29:16 -0800449 *
Christoph Lameter4037d452007-05-09 02:35:14 -0700450 * Note that refresh_cpu_vm_stats strives to only access
451 * node local memory. The per cpu pagesets on remote zones are placed
452 * in the memory local to the processor using that pageset. So the
453 * loop over all zones will access a series of cachelines local to
454 * the processor.
455 *
456 * The call to zone_page_state_add updates the cachelines with the
457 * statistics in the remote zone struct as well as the global cachelines
458 * with the global counters. These could cause remote node cache line
459 * bouncing and will have to be only done when necessary.
Christoph Lameter7cc36bb2014-10-09 15:29:43 -0700460 *
461 * The function returns the number of global counters updated.
Christoph Lameter2244b952006-06-30 01:55:33 -0700462 */
Christoph Lameter0eb77e92016-01-14 15:21:40 -0800463static int refresh_cpu_vm_stats(bool do_pagesets)
Christoph Lameter2244b952006-06-30 01:55:33 -0700464{
465 struct zone *zone;
466 int i;
Christoph Lametera7f75e22008-02-04 22:29:16 -0800467 int global_diff[NR_VM_ZONE_STAT_ITEMS] = { 0, };
Christoph Lameter7cc36bb2014-10-09 15:29:43 -0700468 int changes = 0;
Christoph Lameter2244b952006-06-30 01:55:33 -0700469
KOSAKI Motohiroee99c712009-03-31 15:19:31 -0700470 for_each_populated_zone(zone) {
Christoph Lameterfbc2edb2013-09-11 14:21:32 -0700471 struct per_cpu_pageset __percpu *p = zone->pageset;
Christoph Lameter2244b952006-06-30 01:55:33 -0700472
Christoph Lameterfbc2edb2013-09-11 14:21:32 -0700473 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) {
474 int v;
Christoph Lameter2244b952006-06-30 01:55:33 -0700475
Christoph Lameterfbc2edb2013-09-11 14:21:32 -0700476 v = this_cpu_xchg(p->vm_stat_diff[i], 0);
477 if (v) {
Christoph Lametera7f75e22008-02-04 22:29:16 -0800478
Christoph Lametera7f75e22008-02-04 22:29:16 -0800479 atomic_long_add(v, &zone->vm_stat[i]);
480 global_diff[i] += v;
Christoph Lameter4037d452007-05-09 02:35:14 -0700481#ifdef CONFIG_NUMA
482 /* 3 seconds idle till flush */
Christoph Lameterfbc2edb2013-09-11 14:21:32 -0700483 __this_cpu_write(p->expire, 3);
Christoph Lameter4037d452007-05-09 02:35:14 -0700484#endif
Christoph Lameter2244b952006-06-30 01:55:33 -0700485 }
Christoph Lameterfbc2edb2013-09-11 14:21:32 -0700486 }
Christoph Lameter4037d452007-05-09 02:35:14 -0700487#ifdef CONFIG_NUMA
Christoph Lameter0eb77e92016-01-14 15:21:40 -0800488 if (do_pagesets) {
489 cond_resched();
490 /*
491 * Deal with draining the remote pageset of this
492 * processor
493 *
494 * Check if there are pages remaining in this pageset
495 * if not then there is nothing to expire.
496 */
497 if (!__this_cpu_read(p->expire) ||
Christoph Lameterfbc2edb2013-09-11 14:21:32 -0700498 !__this_cpu_read(p->pcp.count))
Christoph Lameter0eb77e92016-01-14 15:21:40 -0800499 continue;
Christoph Lameter4037d452007-05-09 02:35:14 -0700500
Christoph Lameter0eb77e92016-01-14 15:21:40 -0800501 /*
502 * We never drain zones local to this processor.
503 */
504 if (zone_to_nid(zone) == numa_node_id()) {
505 __this_cpu_write(p->expire, 0);
506 continue;
507 }
Christoph Lameter4037d452007-05-09 02:35:14 -0700508
Christoph Lameter0eb77e92016-01-14 15:21:40 -0800509 if (__this_cpu_dec_return(p->expire))
510 continue;
Christoph Lameter4037d452007-05-09 02:35:14 -0700511
Christoph Lameter0eb77e92016-01-14 15:21:40 -0800512 if (__this_cpu_read(p->pcp.count)) {
513 drain_zone_pages(zone, this_cpu_ptr(&p->pcp));
514 changes++;
515 }
Christoph Lameter7cc36bb2014-10-09 15:29:43 -0700516 }
Christoph Lameter4037d452007-05-09 02:35:14 -0700517#endif
Christoph Lameter2244b952006-06-30 01:55:33 -0700518 }
Christoph Lameter7cc36bb2014-10-09 15:29:43 -0700519 changes += fold_diff(global_diff);
520 return changes;
Christoph Lameter2244b952006-06-30 01:55:33 -0700521}
522
Cody P Schafer40f4b1e2013-04-29 15:08:38 -0700523/*
Christoph Lameter2bb921e2013-09-11 14:21:30 -0700524 * Fold the data for an offline cpu into the global array.
525 * There cannot be any access by the offline cpu and therefore
526 * synchronization is simplified.
527 */
528void cpu_vm_stats_fold(int cpu)
529{
530 struct zone *zone;
531 int i;
532 int global_diff[NR_VM_ZONE_STAT_ITEMS] = { 0, };
533
534 for_each_populated_zone(zone) {
535 struct per_cpu_pageset *p;
536
537 p = per_cpu_ptr(zone->pageset, cpu);
538
539 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
540 if (p->vm_stat_diff[i]) {
541 int v;
542
543 v = p->vm_stat_diff[i];
544 p->vm_stat_diff[i] = 0;
545 atomic_long_add(v, &zone->vm_stat[i]);
546 global_diff[i] += v;
547 }
548 }
549
Christoph Lameter4edb0742013-09-11 14:21:31 -0700550 fold_diff(global_diff);
Christoph Lameter2bb921e2013-09-11 14:21:30 -0700551}
552
553/*
Cody P Schafer40f4b1e2013-04-29 15:08:38 -0700554 * this is only called if !populated_zone(zone), which implies no other users of
555 * pset->vm_stat_diff[] exsist.
556 */
Minchan Kim5a883812012-10-08 16:33:39 -0700557void drain_zonestat(struct zone *zone, struct per_cpu_pageset *pset)
558{
559 int i;
560
561 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
562 if (pset->vm_stat_diff[i]) {
563 int v = pset->vm_stat_diff[i];
564 pset->vm_stat_diff[i] = 0;
565 atomic_long_add(v, &zone->vm_stat[i]);
566 atomic_long_add(v, &vm_stat[i]);
567 }
568}
Christoph Lameter2244b952006-06-30 01:55:33 -0700569#endif
570
Christoph Lameterca889e62006-06-30 01:55:44 -0700571#ifdef CONFIG_NUMA
572/*
573 * zonelist = the list of zones passed to the allocator
574 * z = the zone from which the allocation occurred.
575 *
576 * Must be called with interrupts disabled.
Andi Kleen78afd562011-03-22 16:33:12 -0700577 *
578 * When __GFP_OTHER_NODE is set assume the node of the preferred
579 * zone is the local node. This is useful for daemons who allocate
580 * memory on behalf of other processes.
Christoph Lameterca889e62006-06-30 01:55:44 -0700581 */
Andi Kleen78afd562011-03-22 16:33:12 -0700582void zone_statistics(struct zone *preferred_zone, struct zone *z, gfp_t flags)
Christoph Lameterca889e62006-06-30 01:55:44 -0700583{
Mel Gorman18ea7e72008-04-28 02:12:14 -0700584 if (z->zone_pgdat == preferred_zone->zone_pgdat) {
Christoph Lameterca889e62006-06-30 01:55:44 -0700585 __inc_zone_state(z, NUMA_HIT);
586 } else {
587 __inc_zone_state(z, NUMA_MISS);
Mel Gorman18ea7e72008-04-28 02:12:14 -0700588 __inc_zone_state(preferred_zone, NUMA_FOREIGN);
Christoph Lameterca889e62006-06-30 01:55:44 -0700589 }
Andi Kleen78afd562011-03-22 16:33:12 -0700590 if (z->node == ((flags & __GFP_OTHER_NODE) ?
591 preferred_zone->node : numa_node_id()))
Christoph Lameterca889e62006-06-30 01:55:44 -0700592 __inc_zone_state(z, NUMA_LOCAL);
593 else
594 __inc_zone_state(z, NUMA_OTHER);
595}
Andrew Mortonc2d42c12015-11-05 18:48:43 -0800596
597/*
598 * Determine the per node value of a stat item.
599 */
600unsigned long node_page_state(int node, enum zone_stat_item item)
601{
602 struct zone *zones = NODE_DATA(node)->node_zones;
603
604 return
605#ifdef CONFIG_ZONE_DMA
606 zone_page_state(&zones[ZONE_DMA], item) +
607#endif
608#ifdef CONFIG_ZONE_DMA32
609 zone_page_state(&zones[ZONE_DMA32], item) +
610#endif
611#ifdef CONFIG_HIGHMEM
612 zone_page_state(&zones[ZONE_HIGHMEM], item) +
613#endif
614 zone_page_state(&zones[ZONE_NORMAL], item) +
615 zone_page_state(&zones[ZONE_MOVABLE], item);
616}
617
Christoph Lameterca889e62006-06-30 01:55:44 -0700618#endif
619
Mel Gormand7a57522010-05-24 14:32:25 -0700620#ifdef CONFIG_COMPACTION
Namhyung Kim36deb0b2010-10-26 14:22:04 -0700621
Mel Gormand7a57522010-05-24 14:32:25 -0700622struct contig_page_info {
623 unsigned long free_pages;
624 unsigned long free_blocks_total;
625 unsigned long free_blocks_suitable;
626};
627
628/*
629 * Calculate the number of free pages in a zone, how many contiguous
630 * pages are free and how many are large enough to satisfy an allocation of
631 * the target size. Note that this function makes no attempt to estimate
632 * how many suitable free blocks there *might* be if MOVABLE pages were
633 * migrated. Calculating that is possible, but expensive and can be
634 * figured out from userspace
635 */
636static void fill_contig_page_info(struct zone *zone,
637 unsigned int suitable_order,
638 struct contig_page_info *info)
639{
640 unsigned int order;
641
642 info->free_pages = 0;
643 info->free_blocks_total = 0;
644 info->free_blocks_suitable = 0;
645
646 for (order = 0; order < MAX_ORDER; order++) {
647 unsigned long blocks;
648
649 /* Count number of free blocks */
650 blocks = zone->free_area[order].nr_free;
651 info->free_blocks_total += blocks;
652
653 /* Count free base pages */
654 info->free_pages += blocks << order;
655
656 /* Count the suitable free blocks */
657 if (order >= suitable_order)
658 info->free_blocks_suitable += blocks <<
659 (order - suitable_order);
660 }
661}
Mel Gormanf1a5ab12010-05-24 14:32:26 -0700662
663/*
664 * A fragmentation index only makes sense if an allocation of a requested
665 * size would fail. If that is true, the fragmentation index indicates
666 * whether external fragmentation or a lack of memory was the problem.
667 * The value can be used to determine if page reclaim or compaction
668 * should be used
669 */
Mel Gorman56de7262010-05-24 14:32:30 -0700670static int __fragmentation_index(unsigned int order, struct contig_page_info *info)
Mel Gormanf1a5ab12010-05-24 14:32:26 -0700671{
672 unsigned long requested = 1UL << order;
673
674 if (!info->free_blocks_total)
675 return 0;
676
677 /* Fragmentation index only makes sense when a request would fail */
678 if (info->free_blocks_suitable)
679 return -1000;
680
681 /*
682 * Index is between 0 and 1 so return within 3 decimal places
683 *
684 * 0 => allocation would fail due to lack of memory
685 * 1 => allocation would fail due to fragmentation
686 */
687 return 1000 - div_u64( (1000+(div_u64(info->free_pages * 1000ULL, requested))), info->free_blocks_total);
688}
Mel Gorman56de7262010-05-24 14:32:30 -0700689
690/* Same as __fragmentation index but allocs contig_page_info on stack */
691int fragmentation_index(struct zone *zone, unsigned int order)
692{
693 struct contig_page_info info;
694
695 fill_contig_page_info(zone, order, &info);
696 return __fragmentation_index(order, &info);
697}
Mel Gormand7a57522010-05-24 14:32:25 -0700698#endif
699
David Rientjes0d6617c2011-09-14 16:21:05 -0700700#if defined(CONFIG_PROC_FS) || defined(CONFIG_SYSFS) || defined(CONFIG_NUMA)
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700701#ifdef CONFIG_ZONE_DMA
702#define TEXT_FOR_DMA(xx) xx "_dma",
703#else
704#define TEXT_FOR_DMA(xx)
705#endif
706
707#ifdef CONFIG_ZONE_DMA32
708#define TEXT_FOR_DMA32(xx) xx "_dma32",
709#else
710#define TEXT_FOR_DMA32(xx)
711#endif
712
713#ifdef CONFIG_HIGHMEM
714#define TEXT_FOR_HIGHMEM(xx) xx "_high",
715#else
716#define TEXT_FOR_HIGHMEM(xx)
717#endif
718
719#define TEXTS_FOR_ZONES(xx) TEXT_FOR_DMA(xx) TEXT_FOR_DMA32(xx) xx "_normal", \
720 TEXT_FOR_HIGHMEM(xx) xx "_movable",
721
722const char * const vmstat_text[] = {
Konstantin Khlebnikov09316c02014-10-09 15:29:32 -0700723 /* enum zone_stat_item countes */
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700724 "nr_free_pages",
Johannes Weiner81c0a2b2013-09-11 14:20:47 -0700725 "nr_alloc_batch",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700726 "nr_inactive_anon",
727 "nr_active_anon",
728 "nr_inactive_file",
729 "nr_active_file",
730 "nr_unevictable",
731 "nr_mlock",
732 "nr_anon_pages",
733 "nr_mapped",
734 "nr_file_pages",
735 "nr_dirty",
736 "nr_writeback",
737 "nr_slab_reclaimable",
738 "nr_slab_unreclaimable",
739 "nr_page_table_pages",
740 "nr_kernel_stack",
741 "nr_unstable",
742 "nr_bounce",
743 "nr_vmscan_write",
Mel Gorman49ea7eb2011-10-31 17:07:59 -0700744 "nr_vmscan_immediate_reclaim",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700745 "nr_writeback_temp",
746 "nr_isolated_anon",
747 "nr_isolated_file",
748 "nr_shmem",
749 "nr_dirtied",
750 "nr_written",
Mel Gorman0d5d8232014-08-06 16:07:16 -0700751 "nr_pages_scanned",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700752
753#ifdef CONFIG_NUMA
754 "numa_hit",
755 "numa_miss",
756 "numa_foreign",
757 "numa_interleave",
758 "numa_local",
759 "numa_other",
760#endif
Johannes Weinera5289102014-04-03 14:47:51 -0700761 "workingset_refault",
762 "workingset_activate",
Johannes Weiner449dd692014-04-03 14:47:56 -0700763 "workingset_nodereclaim",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700764 "nr_anon_transparent_hugepages",
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -0700765 "nr_free_cma",
Konstantin Khlebnikov09316c02014-10-09 15:29:32 -0700766
767 /* enum writeback_stat_item counters */
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700768 "nr_dirty_threshold",
769 "nr_dirty_background_threshold",
770
771#ifdef CONFIG_VM_EVENT_COUNTERS
Konstantin Khlebnikov09316c02014-10-09 15:29:32 -0700772 /* enum vm_event_item counters */
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700773 "pgpgin",
774 "pgpgout",
775 "pswpin",
776 "pswpout",
777
778 TEXTS_FOR_ZONES("pgalloc")
779
780 "pgfree",
781 "pgactivate",
782 "pgdeactivate",
783
784 "pgfault",
785 "pgmajfault",
Minchan Kim854e9ed2016-01-15 16:54:53 -0800786 "pglazyfreed",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700787
788 TEXTS_FOR_ZONES("pgrefill")
Ying Han904249a2012-04-25 16:01:48 -0700789 TEXTS_FOR_ZONES("pgsteal_kswapd")
790 TEXTS_FOR_ZONES("pgsteal_direct")
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700791 TEXTS_FOR_ZONES("pgscan_kswapd")
792 TEXTS_FOR_ZONES("pgscan_direct")
Mel Gorman68243e72012-07-31 16:44:39 -0700793 "pgscan_direct_throttle",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700794
795#ifdef CONFIG_NUMA
796 "zone_reclaim_failed",
797#endif
798 "pginodesteal",
799 "slabs_scanned",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700800 "kswapd_inodesteal",
801 "kswapd_low_wmark_hit_quickly",
802 "kswapd_high_wmark_hit_quickly",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700803 "pageoutrun",
804 "allocstall",
805
806 "pgrotated",
807
Dave Hansen5509a5d2014-04-03 14:48:19 -0700808 "drop_pagecache",
809 "drop_slab",
810
Mel Gorman03c5a6e2012-11-02 14:52:48 +0000811#ifdef CONFIG_NUMA_BALANCING
812 "numa_pte_updates",
Mel Gorman72403b42013-11-12 15:08:32 -0800813 "numa_huge_pte_updates",
Mel Gorman03c5a6e2012-11-02 14:52:48 +0000814 "numa_hint_faults",
815 "numa_hint_faults_local",
816 "numa_pages_migrated",
817#endif
Mel Gorman5647bc22012-10-19 10:46:20 +0100818#ifdef CONFIG_MIGRATION
819 "pgmigrate_success",
820 "pgmigrate_fail",
821#endif
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700822#ifdef CONFIG_COMPACTION
Mel Gorman397487d2012-10-19 12:00:10 +0100823 "compact_migrate_scanned",
824 "compact_free_scanned",
825 "compact_isolated",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700826 "compact_stall",
827 "compact_fail",
828 "compact_success",
829#endif
830
831#ifdef CONFIG_HUGETLB_PAGE
832 "htlb_buddy_alloc_success",
833 "htlb_buddy_alloc_fail",
834#endif
835 "unevictable_pgs_culled",
836 "unevictable_pgs_scanned",
837 "unevictable_pgs_rescued",
838 "unevictable_pgs_mlocked",
839 "unevictable_pgs_munlocked",
840 "unevictable_pgs_cleared",
841 "unevictable_pgs_stranded",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700842
843#ifdef CONFIG_TRANSPARENT_HUGEPAGE
844 "thp_fault_alloc",
845 "thp_fault_fallback",
846 "thp_collapse_alloc",
847 "thp_collapse_alloc_failed",
Kirill A. Shutemov122afea2016-01-15 16:52:46 -0800848 "thp_split_page",
849 "thp_split_page_failed",
850 "thp_split_pmd",
Kirill A. Shutemovd8a8e1f2012-12-12 13:51:09 -0800851 "thp_zero_page_alloc",
852 "thp_zero_page_alloc_failed",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700853#endif
Konstantin Khlebnikov09316c02014-10-09 15:29:32 -0700854#ifdef CONFIG_MEMORY_BALLOON
855 "balloon_inflate",
856 "balloon_deflate",
857#ifdef CONFIG_BALLOON_COMPACTION
858 "balloon_migrate",
859#endif
860#endif /* CONFIG_MEMORY_BALLOON */
Mel Gormanec659932014-01-21 14:33:16 -0800861#ifdef CONFIG_DEBUG_TLBFLUSH
Dave Hansen6df46862013-09-11 14:20:24 -0700862#ifdef CONFIG_SMP
Dave Hansen9824cf92013-09-11 14:20:23 -0700863 "nr_tlb_remote_flush",
864 "nr_tlb_remote_flush_received",
Mel Gormanec659932014-01-21 14:33:16 -0800865#endif /* CONFIG_SMP */
Dave Hansen9824cf92013-09-11 14:20:23 -0700866 "nr_tlb_local_flush_all",
867 "nr_tlb_local_flush_one",
Mel Gormanec659932014-01-21 14:33:16 -0800868#endif /* CONFIG_DEBUG_TLBFLUSH */
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700869
Davidlohr Bueso4f115142014-06-04 16:06:46 -0700870#ifdef CONFIG_DEBUG_VM_VMACACHE
871 "vmacache_find_calls",
872 "vmacache_find_hits",
Davidlohr Buesof5f302e2014-12-12 16:56:10 -0800873 "vmacache_full_flushes",
Davidlohr Bueso4f115142014-06-04 16:06:46 -0700874#endif
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700875#endif /* CONFIG_VM_EVENTS_COUNTERS */
876};
David Rientjes0d6617c2011-09-14 16:21:05 -0700877#endif /* CONFIG_PROC_FS || CONFIG_SYSFS || CONFIG_NUMA */
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700878
879
Andrew Morton3c486872015-02-10 14:09:43 -0800880#if (defined(CONFIG_DEBUG_FS) && defined(CONFIG_COMPACTION)) || \
881 defined(CONFIG_PROC_FS)
882static void *frag_start(struct seq_file *m, loff_t *pos)
883{
884 pg_data_t *pgdat;
885 loff_t node = *pos;
886
887 for (pgdat = first_online_pgdat();
888 pgdat && node;
889 pgdat = next_online_pgdat(pgdat))
890 --node;
891
892 return pgdat;
893}
894
895static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
896{
897 pg_data_t *pgdat = (pg_data_t *)arg;
898
899 (*pos)++;
900 return next_online_pgdat(pgdat);
901}
902
903static void frag_stop(struct seq_file *m, void *arg)
904{
905}
906
907/* Walk all the zones in a node and print using a callback */
908static void walk_zones_in_node(struct seq_file *m, pg_data_t *pgdat,
909 void (*print)(struct seq_file *m, pg_data_t *, struct zone *))
910{
911 struct zone *zone;
912 struct zone *node_zones = pgdat->node_zones;
913 unsigned long flags;
914
915 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
916 if (!populated_zone(zone))
917 continue;
918
919 spin_lock_irqsave(&zone->lock, flags);
920 print(m, pgdat, zone);
921 spin_unlock_irqrestore(&zone->lock, flags);
922 }
923}
924#endif
925
Mel Gormand7a57522010-05-24 14:32:25 -0700926#ifdef CONFIG_PROC_FS
Andrew Morton3c486872015-02-10 14:09:43 -0800927static char * const migratetype_names[MIGRATE_TYPES] = {
928 "Unmovable",
Andrew Morton3c486872015-02-10 14:09:43 -0800929 "Movable",
Vlastimil Babka475a2f92015-12-11 13:40:29 -0800930 "Reclaimable",
Mel Gorman0aaa29a2015-11-06 16:28:37 -0800931 "HighAtomic",
Andrew Morton3c486872015-02-10 14:09:43 -0800932#ifdef CONFIG_CMA
933 "CMA",
934#endif
935#ifdef CONFIG_MEMORY_ISOLATION
936 "Isolate",
937#endif
938};
939
Mel Gorman467c9962007-10-16 01:26:02 -0700940static void frag_show_print(struct seq_file *m, pg_data_t *pgdat,
941 struct zone *zone)
942{
943 int order;
944
945 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
946 for (order = 0; order < MAX_ORDER; ++order)
947 seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
948 seq_putc(m, '\n');
949}
950
951/*
952 * This walks the free areas for each zone.
953 */
954static int frag_show(struct seq_file *m, void *arg)
955{
956 pg_data_t *pgdat = (pg_data_t *)arg;
957 walk_zones_in_node(m, pgdat, frag_show_print);
958 return 0;
959}
960
961static void pagetypeinfo_showfree_print(struct seq_file *m,
962 pg_data_t *pgdat, struct zone *zone)
963{
964 int order, mtype;
965
966 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++) {
967 seq_printf(m, "Node %4d, zone %8s, type %12s ",
968 pgdat->node_id,
969 zone->name,
970 migratetype_names[mtype]);
971 for (order = 0; order < MAX_ORDER; ++order) {
972 unsigned long freecount = 0;
973 struct free_area *area;
974 struct list_head *curr;
975
976 area = &(zone->free_area[order]);
977
978 list_for_each(curr, &area->free_list[mtype])
979 freecount++;
980 seq_printf(m, "%6lu ", freecount);
981 }
Christoph Lameterf6ac2352006-06-30 01:55:32 -0700982 seq_putc(m, '\n');
983 }
Mel Gorman467c9962007-10-16 01:26:02 -0700984}
985
986/* Print out the free pages at each order for each migatetype */
987static int pagetypeinfo_showfree(struct seq_file *m, void *arg)
988{
989 int order;
990 pg_data_t *pgdat = (pg_data_t *)arg;
991
992 /* Print header */
993 seq_printf(m, "%-43s ", "Free pages count per migrate type at order");
994 for (order = 0; order < MAX_ORDER; ++order)
995 seq_printf(m, "%6d ", order);
996 seq_putc(m, '\n');
997
998 walk_zones_in_node(m, pgdat, pagetypeinfo_showfree_print);
999
1000 return 0;
1001}
1002
1003static void pagetypeinfo_showblockcount_print(struct seq_file *m,
1004 pg_data_t *pgdat, struct zone *zone)
1005{
1006 int mtype;
1007 unsigned long pfn;
1008 unsigned long start_pfn = zone->zone_start_pfn;
Cody P Schafer108bcc92013-02-22 16:35:23 -08001009 unsigned long end_pfn = zone_end_pfn(zone);
Mel Gorman467c9962007-10-16 01:26:02 -07001010 unsigned long count[MIGRATE_TYPES] = { 0, };
1011
1012 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
1013 struct page *page;
1014
1015 if (!pfn_valid(pfn))
1016 continue;
1017
1018 page = pfn_to_page(pfn);
Mel Gormaneb335752009-05-13 17:34:48 +01001019
1020 /* Watch for unexpected holes punched in the memmap */
1021 if (!memmap_valid_within(pfn, page, zone))
Mel Gormane80d6a22008-08-14 11:10:14 +01001022 continue;
Mel Gormaneb335752009-05-13 17:34:48 +01001023
Mel Gorman467c9962007-10-16 01:26:02 -07001024 mtype = get_pageblock_migratetype(page);
1025
Mel Gormane80d6a22008-08-14 11:10:14 +01001026 if (mtype < MIGRATE_TYPES)
1027 count[mtype]++;
Mel Gorman467c9962007-10-16 01:26:02 -07001028 }
1029
1030 /* Print counts */
1031 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
1032 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
1033 seq_printf(m, "%12lu ", count[mtype]);
1034 seq_putc(m, '\n');
1035}
1036
1037/* Print out the free pages at each order for each migratetype */
1038static int pagetypeinfo_showblockcount(struct seq_file *m, void *arg)
1039{
1040 int mtype;
1041 pg_data_t *pgdat = (pg_data_t *)arg;
1042
1043 seq_printf(m, "\n%-23s", "Number of blocks type ");
1044 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
1045 seq_printf(m, "%12s ", migratetype_names[mtype]);
1046 seq_putc(m, '\n');
1047 walk_zones_in_node(m, pgdat, pagetypeinfo_showblockcount_print);
1048
1049 return 0;
1050}
1051
Joonsoo Kim48c96a32014-12-12 16:56:01 -08001052#ifdef CONFIG_PAGE_OWNER
1053static void pagetypeinfo_showmixedcount_print(struct seq_file *m,
1054 pg_data_t *pgdat,
1055 struct zone *zone)
1056{
1057 struct page *page;
1058 struct page_ext *page_ext;
1059 unsigned long pfn = zone->zone_start_pfn, block_end_pfn;
1060 unsigned long end_pfn = pfn + zone->spanned_pages;
1061 unsigned long count[MIGRATE_TYPES] = { 0, };
1062 int pageblock_mt, page_mt;
1063 int i;
1064
1065 /* Scan block by block. First and last block may be incomplete */
1066 pfn = zone->zone_start_pfn;
1067
1068 /*
1069 * Walk the zone in pageblock_nr_pages steps. If a page block spans
1070 * a zone boundary, it will be double counted between zones. This does
1071 * not matter as the mixed block count will still be correct
1072 */
1073 for (; pfn < end_pfn; ) {
1074 if (!pfn_valid(pfn)) {
1075 pfn = ALIGN(pfn + 1, MAX_ORDER_NR_PAGES);
1076 continue;
1077 }
1078
1079 block_end_pfn = ALIGN(pfn + 1, pageblock_nr_pages);
1080 block_end_pfn = min(block_end_pfn, end_pfn);
1081
1082 page = pfn_to_page(pfn);
1083 pageblock_mt = get_pfnblock_migratetype(page, pfn);
1084
1085 for (; pfn < block_end_pfn; pfn++) {
1086 if (!pfn_valid_within(pfn))
1087 continue;
1088
1089 page = pfn_to_page(pfn);
1090 if (PageBuddy(page)) {
1091 pfn += (1UL << page_order(page)) - 1;
1092 continue;
1093 }
1094
1095 if (PageReserved(page))
1096 continue;
1097
1098 page_ext = lookup_page_ext(page);
1099
1100 if (!test_bit(PAGE_EXT_OWNER, &page_ext->flags))
1101 continue;
1102
1103 page_mt = gfpflags_to_migratetype(page_ext->gfp_mask);
1104 if (pageblock_mt != page_mt) {
1105 if (is_migrate_cma(pageblock_mt))
1106 count[MIGRATE_MOVABLE]++;
1107 else
1108 count[pageblock_mt]++;
1109
1110 pfn = block_end_pfn;
1111 break;
1112 }
1113 pfn += (1UL << page_ext->order) - 1;
1114 }
1115 }
1116
1117 /* Print counts */
1118 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
1119 for (i = 0; i < MIGRATE_TYPES; i++)
1120 seq_printf(m, "%12lu ", count[i]);
1121 seq_putc(m, '\n');
1122}
1123#endif /* CONFIG_PAGE_OWNER */
1124
1125/*
1126 * Print out the number of pageblocks for each migratetype that contain pages
1127 * of other types. This gives an indication of how well fallbacks are being
1128 * contained by rmqueue_fallback(). It requires information from PAGE_OWNER
1129 * to determine what is going on
1130 */
1131static void pagetypeinfo_showmixedcount(struct seq_file *m, pg_data_t *pgdat)
1132{
1133#ifdef CONFIG_PAGE_OWNER
1134 int mtype;
1135
1136 if (!page_owner_inited)
1137 return;
1138
1139 drain_all_pages(NULL);
1140
1141 seq_printf(m, "\n%-23s", "Number of mixed blocks ");
1142 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
1143 seq_printf(m, "%12s ", migratetype_names[mtype]);
1144 seq_putc(m, '\n');
1145
1146 walk_zones_in_node(m, pgdat, pagetypeinfo_showmixedcount_print);
1147#endif /* CONFIG_PAGE_OWNER */
1148}
1149
Mel Gorman467c9962007-10-16 01:26:02 -07001150/*
1151 * This prints out statistics in relation to grouping pages by mobility.
1152 * It is expensive to collect so do not constantly read the file.
1153 */
1154static int pagetypeinfo_show(struct seq_file *m, void *arg)
1155{
1156 pg_data_t *pgdat = (pg_data_t *)arg;
1157
KOSAKI Motohiro41b25a32008-04-30 00:52:13 -07001158 /* check memoryless node */
Lai Jiangshana47b53c2012-12-12 13:51:37 -08001159 if (!node_state(pgdat->node_id, N_MEMORY))
KOSAKI Motohiro41b25a32008-04-30 00:52:13 -07001160 return 0;
1161
Mel Gorman467c9962007-10-16 01:26:02 -07001162 seq_printf(m, "Page block order: %d\n", pageblock_order);
1163 seq_printf(m, "Pages per block: %lu\n", pageblock_nr_pages);
1164 seq_putc(m, '\n');
1165 pagetypeinfo_showfree(m, pgdat);
1166 pagetypeinfo_showblockcount(m, pgdat);
Joonsoo Kim48c96a32014-12-12 16:56:01 -08001167 pagetypeinfo_showmixedcount(m, pgdat);
Mel Gorman467c9962007-10-16 01:26:02 -07001168
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001169 return 0;
1170}
1171
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +04001172static const struct seq_operations fragmentation_op = {
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001173 .start = frag_start,
1174 .next = frag_next,
1175 .stop = frag_stop,
1176 .show = frag_show,
1177};
1178
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +04001179static int fragmentation_open(struct inode *inode, struct file *file)
1180{
1181 return seq_open(file, &fragmentation_op);
1182}
1183
1184static const struct file_operations fragmentation_file_operations = {
1185 .open = fragmentation_open,
1186 .read = seq_read,
1187 .llseek = seq_lseek,
1188 .release = seq_release,
1189};
1190
Alexey Dobriyan74e2e8e2008-10-06 04:15:36 +04001191static const struct seq_operations pagetypeinfo_op = {
Mel Gorman467c9962007-10-16 01:26:02 -07001192 .start = frag_start,
1193 .next = frag_next,
1194 .stop = frag_stop,
1195 .show = pagetypeinfo_show,
1196};
1197
Alexey Dobriyan74e2e8e2008-10-06 04:15:36 +04001198static int pagetypeinfo_open(struct inode *inode, struct file *file)
1199{
1200 return seq_open(file, &pagetypeinfo_op);
1201}
1202
1203static const struct file_operations pagetypeinfo_file_ops = {
1204 .open = pagetypeinfo_open,
1205 .read = seq_read,
1206 .llseek = seq_lseek,
1207 .release = seq_release,
1208};
1209
Mel Gorman467c9962007-10-16 01:26:02 -07001210static void zoneinfo_show_print(struct seq_file *m, pg_data_t *pgdat,
1211 struct zone *zone)
1212{
1213 int i;
1214 seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
1215 seq_printf(m,
1216 "\n pages free %lu"
1217 "\n min %lu"
1218 "\n low %lu"
1219 "\n high %lu"
Wu Fengguang08d9ae72009-06-16 15:32:30 -07001220 "\n scanned %lu"
Mel Gorman467c9962007-10-16 01:26:02 -07001221 "\n spanned %lu"
Jiang Liu9feedc92012-12-12 13:52:12 -08001222 "\n present %lu"
1223 "\n managed %lu",
Mel Gorman88f5acf2011-01-13 15:45:41 -08001224 zone_page_state(zone, NR_FREE_PAGES),
Mel Gorman41858962009-06-16 15:32:12 -07001225 min_wmark_pages(zone),
1226 low_wmark_pages(zone),
1227 high_wmark_pages(zone),
Mel Gorman0d5d8232014-08-06 16:07:16 -07001228 zone_page_state(zone, NR_PAGES_SCANNED),
Mel Gorman467c9962007-10-16 01:26:02 -07001229 zone->spanned_pages,
Jiang Liu9feedc92012-12-12 13:52:12 -08001230 zone->present_pages,
1231 zone->managed_pages);
Mel Gorman467c9962007-10-16 01:26:02 -07001232
1233 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
1234 seq_printf(m, "\n %-12s %lu", vmstat_text[i],
1235 zone_page_state(zone, i));
1236
1237 seq_printf(m,
Mel Gorman3484b2d2014-08-06 16:07:14 -07001238 "\n protection: (%ld",
Mel Gorman467c9962007-10-16 01:26:02 -07001239 zone->lowmem_reserve[0]);
1240 for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
Mel Gorman3484b2d2014-08-06 16:07:14 -07001241 seq_printf(m, ", %ld", zone->lowmem_reserve[i]);
Mel Gorman467c9962007-10-16 01:26:02 -07001242 seq_printf(m,
1243 ")"
1244 "\n pagesets");
1245 for_each_online_cpu(i) {
1246 struct per_cpu_pageset *pageset;
Mel Gorman467c9962007-10-16 01:26:02 -07001247
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001248 pageset = per_cpu_ptr(zone->pageset, i);
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001249 seq_printf(m,
1250 "\n cpu: %i"
1251 "\n count: %i"
1252 "\n high: %i"
1253 "\n batch: %i",
1254 i,
1255 pageset->pcp.count,
1256 pageset->pcp.high,
1257 pageset->pcp.batch);
Mel Gorman467c9962007-10-16 01:26:02 -07001258#ifdef CONFIG_SMP
1259 seq_printf(m, "\n vm stats threshold: %d",
1260 pageset->stat_threshold);
1261#endif
1262 }
1263 seq_printf(m,
1264 "\n all_unreclaimable: %u"
Rik van Riel556adec2008-10-18 20:26:34 -07001265 "\n start_pfn: %lu"
1266 "\n inactive_ratio: %u",
Lisa Du6e543d52013-09-11 14:22:36 -07001267 !zone_reclaimable(zone),
Rik van Riel556adec2008-10-18 20:26:34 -07001268 zone->zone_start_pfn,
1269 zone->inactive_ratio);
Mel Gorman467c9962007-10-16 01:26:02 -07001270 seq_putc(m, '\n');
1271}
1272
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001273/*
1274 * Output information about zones in @pgdat.
1275 */
1276static int zoneinfo_show(struct seq_file *m, void *arg)
1277{
Mel Gorman467c9962007-10-16 01:26:02 -07001278 pg_data_t *pgdat = (pg_data_t *)arg;
1279 walk_zones_in_node(m, pgdat, zoneinfo_show_print);
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001280 return 0;
1281}
1282
Alexey Dobriyan5c9fe622008-10-06 04:19:42 +04001283static const struct seq_operations zoneinfo_op = {
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001284 .start = frag_start, /* iterate over all zones. The same as in
1285 * fragmentation. */
1286 .next = frag_next,
1287 .stop = frag_stop,
1288 .show = zoneinfo_show,
1289};
1290
Alexey Dobriyan5c9fe622008-10-06 04:19:42 +04001291static int zoneinfo_open(struct inode *inode, struct file *file)
1292{
1293 return seq_open(file, &zoneinfo_op);
1294}
1295
1296static const struct file_operations proc_zoneinfo_file_operations = {
1297 .open = zoneinfo_open,
1298 .read = seq_read,
1299 .llseek = seq_lseek,
1300 .release = seq_release,
1301};
1302
Michael Rubin79da8262010-10-26 14:21:36 -07001303enum writeback_stat_item {
1304 NR_DIRTY_THRESHOLD,
1305 NR_DIRTY_BG_THRESHOLD,
1306 NR_VM_WRITEBACK_STAT_ITEMS,
1307};
1308
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001309static void *vmstat_start(struct seq_file *m, loff_t *pos)
1310{
Christoph Lameter2244b952006-06-30 01:55:33 -07001311 unsigned long *v;
Michael Rubin79da8262010-10-26 14:21:36 -07001312 int i, stat_items_size;
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001313
1314 if (*pos >= ARRAY_SIZE(vmstat_text))
1315 return NULL;
Michael Rubin79da8262010-10-26 14:21:36 -07001316 stat_items_size = NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long) +
1317 NR_VM_WRITEBACK_STAT_ITEMS * sizeof(unsigned long);
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001318
Christoph Lameterf8891e52006-06-30 01:55:45 -07001319#ifdef CONFIG_VM_EVENT_COUNTERS
Michael Rubin79da8262010-10-26 14:21:36 -07001320 stat_items_size += sizeof(struct vm_event_state);
Christoph Lameterf8891e52006-06-30 01:55:45 -07001321#endif
Michael Rubin79da8262010-10-26 14:21:36 -07001322
1323 v = kmalloc(stat_items_size, GFP_KERNEL);
Christoph Lameter2244b952006-06-30 01:55:33 -07001324 m->private = v;
1325 if (!v)
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001326 return ERR_PTR(-ENOMEM);
Christoph Lameter2244b952006-06-30 01:55:33 -07001327 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
1328 v[i] = global_page_state(i);
Michael Rubin79da8262010-10-26 14:21:36 -07001329 v += NR_VM_ZONE_STAT_ITEMS;
1330
1331 global_dirty_limits(v + NR_DIRTY_BG_THRESHOLD,
1332 v + NR_DIRTY_THRESHOLD);
1333 v += NR_VM_WRITEBACK_STAT_ITEMS;
1334
Christoph Lameterf8891e52006-06-30 01:55:45 -07001335#ifdef CONFIG_VM_EVENT_COUNTERS
Michael Rubin79da8262010-10-26 14:21:36 -07001336 all_vm_events(v);
1337 v[PGPGIN] /= 2; /* sectors -> kbytes */
1338 v[PGPGOUT] /= 2;
Christoph Lameterf8891e52006-06-30 01:55:45 -07001339#endif
Wu Fengguangff8b16d2010-11-04 01:56:49 +08001340 return (unsigned long *)m->private + *pos;
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001341}
1342
1343static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
1344{
1345 (*pos)++;
1346 if (*pos >= ARRAY_SIZE(vmstat_text))
1347 return NULL;
1348 return (unsigned long *)m->private + *pos;
1349}
1350
1351static int vmstat_show(struct seq_file *m, void *arg)
1352{
1353 unsigned long *l = arg;
1354 unsigned long off = l - (unsigned long *)m->private;
1355
1356 seq_printf(m, "%s %lu\n", vmstat_text[off], *l);
1357 return 0;
1358}
1359
1360static void vmstat_stop(struct seq_file *m, void *arg)
1361{
1362 kfree(m->private);
1363 m->private = NULL;
1364}
1365
Alexey Dobriyanb6aa44a2008-10-06 04:17:48 +04001366static const struct seq_operations vmstat_op = {
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001367 .start = vmstat_start,
1368 .next = vmstat_next,
1369 .stop = vmstat_stop,
1370 .show = vmstat_show,
1371};
1372
Alexey Dobriyanb6aa44a2008-10-06 04:17:48 +04001373static int vmstat_open(struct inode *inode, struct file *file)
1374{
1375 return seq_open(file, &vmstat_op);
1376}
1377
1378static const struct file_operations proc_vmstat_file_operations = {
1379 .open = vmstat_open,
1380 .read = seq_read,
1381 .llseek = seq_lseek,
1382 .release = seq_release,
1383};
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001384#endif /* CONFIG_PROC_FS */
1385
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001386#ifdef CONFIG_SMP
Michal Hocko373ccbe2015-12-11 13:40:32 -08001387static struct workqueue_struct *vmstat_wq;
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001388static DEFINE_PER_CPU(struct delayed_work, vmstat_work);
Christoph Lameter77461ab2007-05-09 02:35:13 -07001389int sysctl_stat_interval __read_mostly = HZ;
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001390static cpumask_var_t cpu_stat_off;
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001391
1392static void vmstat_update(struct work_struct *w)
1393{
Christoph Lameter0eb77e92016-01-14 15:21:40 -08001394 if (refresh_cpu_vm_stats(true)) {
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001395 /*
1396 * Counters were updated so we expect more updates
1397 * to occur in the future. Keep on running the
1398 * update worker thread.
Michal Hockof01f17d2016-02-05 15:36:24 -08001399 * If we were marked on cpu_stat_off clear the flag
1400 * so that vmstat_shepherd doesn't schedule us again.
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001401 */
Michal Hockof01f17d2016-02-05 15:36:24 -08001402 if (!cpumask_test_and_clear_cpu(smp_processor_id(),
1403 cpu_stat_off)) {
1404 queue_delayed_work_on(smp_processor_id(), vmstat_wq,
1405 this_cpu_ptr(&vmstat_work),
1406 round_jiffies_relative(sysctl_stat_interval));
1407 }
Linus Torvalds176bed12015-10-15 13:01:50 -07001408 } else {
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001409 /*
1410 * We did not update any counters so the app may be in
1411 * a mode where it does not cause counter updates.
1412 * We may be uselessly running vmstat_update.
1413 * Defer the checking for differentials to the
1414 * shepherd thread on a different processor.
1415 */
Christoph Lameter587198b2016-01-22 10:46:14 -06001416 cpumask_set_cpu(smp_processor_id(), cpu_stat_off);
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001417 }
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001418}
1419
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001420/*
Christoph Lameter0eb77e92016-01-14 15:21:40 -08001421 * Switch off vmstat processing and then fold all the remaining differentials
1422 * until the diffs stay at zero. The function is used by NOHZ and can only be
1423 * invoked when tick processing is not active.
1424 */
Christoph Lameter0eb77e92016-01-14 15:21:40 -08001425/*
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001426 * Check if the diffs for a certain cpu indicate that
1427 * an update is needed.
1428 */
1429static bool need_update(int cpu)
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001430{
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001431 struct zone *zone;
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001432
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001433 for_each_populated_zone(zone) {
1434 struct per_cpu_pageset *p = per_cpu_ptr(zone->pageset, cpu);
1435
1436 BUILD_BUG_ON(sizeof(p->vm_stat_diff[0]) != 1);
1437 /*
1438 * The fast way of checking if there are any vmstat diffs.
1439 * This works because the diffs are byte sized items.
1440 */
1441 if (memchr_inv(p->vm_stat_diff, 0, NR_VM_ZONE_STAT_ITEMS))
1442 return true;
1443
1444 }
1445 return false;
1446}
1447
Michal Hockof01f17d2016-02-05 15:36:24 -08001448void quiet_vmstat(void)
1449{
1450 if (system_state != SYSTEM_RUNNING)
1451 return;
1452
1453 /*
1454 * If we are already in hands of the shepherd then there
1455 * is nothing for us to do here.
1456 */
1457 if (cpumask_test_and_set_cpu(smp_processor_id(), cpu_stat_off))
1458 return;
1459
1460 if (!need_update(smp_processor_id()))
1461 return;
1462
1463 /*
1464 * Just refresh counters and do not care about the pending delayed
1465 * vmstat_update. It doesn't fire that often to matter and canceling
1466 * it would be too expensive from this path.
1467 * vmstat_shepherd will take care about that for us.
1468 */
1469 refresh_cpu_vm_stats(false);
1470}
1471
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001472
1473/*
1474 * Shepherd worker thread that checks the
1475 * differentials of processors that have their worker
1476 * threads for vm statistics updates disabled because of
1477 * inactivity.
1478 */
1479static void vmstat_shepherd(struct work_struct *w);
1480
Christoph Lameter0eb77e92016-01-14 15:21:40 -08001481static DECLARE_DEFERRABLE_WORK(shepherd, vmstat_shepherd);
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001482
1483static void vmstat_shepherd(struct work_struct *w)
1484{
1485 int cpu;
1486
1487 get_online_cpus();
1488 /* Check processors whose vmstat worker threads have been disabled */
Michal Hockof01f17d2016-02-05 15:36:24 -08001489 for_each_cpu(cpu, cpu_stat_off) {
1490 struct delayed_work *dw = &per_cpu(vmstat_work, cpu);
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001491
Michal Hockof01f17d2016-02-05 15:36:24 -08001492 if (need_update(cpu)) {
1493 if (cpumask_test_and_clear_cpu(cpu, cpu_stat_off))
1494 queue_delayed_work_on(cpu, vmstat_wq, dw, 0);
1495 } else {
1496 /*
1497 * Cancel the work if quiet_vmstat has put this
1498 * cpu on cpu_stat_off because the work item might
1499 * be still scheduled
1500 */
1501 cancel_delayed_work(dw);
1502 }
1503 }
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001504 put_online_cpus();
1505
1506 schedule_delayed_work(&shepherd,
1507 round_jiffies_relative(sysctl_stat_interval));
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001508}
1509
1510static void __init start_shepherd_timer(void)
1511{
1512 int cpu;
1513
1514 for_each_possible_cpu(cpu)
Michal Hockoccde8bd2016-02-05 15:36:27 -08001515 INIT_DEFERRABLE_WORK(per_cpu_ptr(&vmstat_work, cpu),
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001516 vmstat_update);
1517
1518 if (!alloc_cpumask_var(&cpu_stat_off, GFP_KERNEL))
1519 BUG();
1520 cpumask_copy(cpu_stat_off, cpu_online_mask);
1521
Michal Hocko751e5f52016-01-08 11:18:29 +01001522 vmstat_wq = alloc_workqueue("vmstat", WQ_FREEZABLE|WQ_MEM_RECLAIM, 0);
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001523 schedule_delayed_work(&shepherd,
1524 round_jiffies_relative(sysctl_stat_interval));
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001525}
1526
Toshi Kani807a1bd2013-11-12 15:08:13 -08001527static void vmstat_cpu_dead(int node)
1528{
1529 int cpu;
1530
1531 get_online_cpus();
1532 for_each_online_cpu(cpu)
1533 if (cpu_to_node(cpu) == node)
1534 goto end;
1535
1536 node_clear_state(node, N_CPU);
1537end:
1538 put_online_cpus();
1539}
1540
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001541/*
1542 * Use the cpu notifier to insure that the thresholds are recalculated
1543 * when necessary.
1544 */
Paul Gortmaker0db06282013-06-19 14:53:51 -04001545static int vmstat_cpuup_callback(struct notifier_block *nfb,
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001546 unsigned long action,
1547 void *hcpu)
1548{
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001549 long cpu = (long)hcpu;
1550
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001551 switch (action) {
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001552 case CPU_ONLINE:
1553 case CPU_ONLINE_FROZEN:
KAMEZAWA Hiroyuki5ee28a42010-09-09 16:38:14 -07001554 refresh_zone_stat_thresholds();
Christoph Lameterad596922010-01-05 15:34:51 +09001555 node_set_state(cpu_to_node(cpu), N_CPU);
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001556 cpumask_set_cpu(cpu, cpu_stat_off);
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001557 break;
1558 case CPU_DOWN_PREPARE:
1559 case CPU_DOWN_PREPARE_FROZEN:
Tejun Heoafe2c512010-12-14 16:21:17 +01001560 cancel_delayed_work_sync(&per_cpu(vmstat_work, cpu));
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001561 cpumask_clear_cpu(cpu, cpu_stat_off);
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001562 break;
1563 case CPU_DOWN_FAILED:
1564 case CPU_DOWN_FAILED_FROZEN:
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001565 cpumask_set_cpu(cpu, cpu_stat_off);
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001566 break;
Andy Whitcroftce421c72006-12-06 20:33:08 -08001567 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07001568 case CPU_DEAD_FROZEN:
Andy Whitcroftce421c72006-12-06 20:33:08 -08001569 refresh_zone_stat_thresholds();
Toshi Kani807a1bd2013-11-12 15:08:13 -08001570 vmstat_cpu_dead(cpu_to_node(cpu));
Andy Whitcroftce421c72006-12-06 20:33:08 -08001571 break;
1572 default:
1573 break;
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001574 }
1575 return NOTIFY_OK;
1576}
1577
Paul Gortmaker0db06282013-06-19 14:53:51 -04001578static struct notifier_block vmstat_notifier =
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001579 { &vmstat_cpuup_callback, NULL, 0 };
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +04001580#endif
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001581
Adrian Bunke2fc88d2007-10-16 01:26:27 -07001582static int __init setup_vmstat(void)
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001583{
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +04001584#ifdef CONFIG_SMP
Srivatsa S. Bhat0be94ba2014-03-11 02:12:27 +05301585 cpu_notifier_register_begin();
1586 __register_cpu_notifier(&vmstat_notifier);
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001587
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001588 start_shepherd_timer();
Srivatsa S. Bhat0be94ba2014-03-11 02:12:27 +05301589 cpu_notifier_register_done();
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +04001590#endif
1591#ifdef CONFIG_PROC_FS
1592 proc_create("buddyinfo", S_IRUGO, NULL, &fragmentation_file_operations);
Alexey Dobriyan74e2e8e2008-10-06 04:15:36 +04001593 proc_create("pagetypeinfo", S_IRUGO, NULL, &pagetypeinfo_file_ops);
Alexey Dobriyanb6aa44a2008-10-06 04:17:48 +04001594 proc_create("vmstat", S_IRUGO, NULL, &proc_vmstat_file_operations);
Alexey Dobriyan5c9fe622008-10-06 04:19:42 +04001595 proc_create("zoneinfo", S_IRUGO, NULL, &proc_zoneinfo_file_operations);
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +04001596#endif
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001597 return 0;
1598}
1599module_init(setup_vmstat)
Mel Gormand7a57522010-05-24 14:32:25 -07001600
1601#if defined(CONFIG_DEBUG_FS) && defined(CONFIG_COMPACTION)
Mel Gormand7a57522010-05-24 14:32:25 -07001602
1603/*
1604 * Return an index indicating how much of the available free memory is
1605 * unusable for an allocation of the requested size.
1606 */
1607static int unusable_free_index(unsigned int order,
1608 struct contig_page_info *info)
1609{
1610 /* No free memory is interpreted as all free memory is unusable */
1611 if (info->free_pages == 0)
1612 return 1000;
1613
1614 /*
1615 * Index should be a value between 0 and 1. Return a value to 3
1616 * decimal places.
1617 *
1618 * 0 => no fragmentation
1619 * 1 => high fragmentation
1620 */
1621 return div_u64((info->free_pages - (info->free_blocks_suitable << order)) * 1000ULL, info->free_pages);
1622
1623}
1624
1625static void unusable_show_print(struct seq_file *m,
1626 pg_data_t *pgdat, struct zone *zone)
1627{
1628 unsigned int order;
1629 int index;
1630 struct contig_page_info info;
1631
1632 seq_printf(m, "Node %d, zone %8s ",
1633 pgdat->node_id,
1634 zone->name);
1635 for (order = 0; order < MAX_ORDER; ++order) {
1636 fill_contig_page_info(zone, order, &info);
1637 index = unusable_free_index(order, &info);
1638 seq_printf(m, "%d.%03d ", index / 1000, index % 1000);
1639 }
1640
1641 seq_putc(m, '\n');
1642}
1643
1644/*
1645 * Display unusable free space index
1646 *
1647 * The unusable free space index measures how much of the available free
1648 * memory cannot be used to satisfy an allocation of a given size and is a
1649 * value between 0 and 1. The higher the value, the more of free memory is
1650 * unusable and by implication, the worse the external fragmentation is. This
1651 * can be expressed as a percentage by multiplying by 100.
1652 */
1653static int unusable_show(struct seq_file *m, void *arg)
1654{
1655 pg_data_t *pgdat = (pg_data_t *)arg;
1656
1657 /* check memoryless node */
Lai Jiangshana47b53c2012-12-12 13:51:37 -08001658 if (!node_state(pgdat->node_id, N_MEMORY))
Mel Gormand7a57522010-05-24 14:32:25 -07001659 return 0;
1660
1661 walk_zones_in_node(m, pgdat, unusable_show_print);
1662
1663 return 0;
1664}
1665
1666static const struct seq_operations unusable_op = {
1667 .start = frag_start,
1668 .next = frag_next,
1669 .stop = frag_stop,
1670 .show = unusable_show,
1671};
1672
1673static int unusable_open(struct inode *inode, struct file *file)
1674{
1675 return seq_open(file, &unusable_op);
1676}
1677
1678static const struct file_operations unusable_file_ops = {
1679 .open = unusable_open,
1680 .read = seq_read,
1681 .llseek = seq_lseek,
1682 .release = seq_release,
1683};
1684
Mel Gormanf1a5ab12010-05-24 14:32:26 -07001685static void extfrag_show_print(struct seq_file *m,
1686 pg_data_t *pgdat, struct zone *zone)
1687{
1688 unsigned int order;
1689 int index;
1690
1691 /* Alloc on stack as interrupts are disabled for zone walk */
1692 struct contig_page_info info;
1693
1694 seq_printf(m, "Node %d, zone %8s ",
1695 pgdat->node_id,
1696 zone->name);
1697 for (order = 0; order < MAX_ORDER; ++order) {
1698 fill_contig_page_info(zone, order, &info);
Mel Gorman56de7262010-05-24 14:32:30 -07001699 index = __fragmentation_index(order, &info);
Mel Gormanf1a5ab12010-05-24 14:32:26 -07001700 seq_printf(m, "%d.%03d ", index / 1000, index % 1000);
1701 }
1702
1703 seq_putc(m, '\n');
1704}
1705
1706/*
1707 * Display fragmentation index for orders that allocations would fail for
1708 */
1709static int extfrag_show(struct seq_file *m, void *arg)
1710{
1711 pg_data_t *pgdat = (pg_data_t *)arg;
1712
1713 walk_zones_in_node(m, pgdat, extfrag_show_print);
1714
1715 return 0;
1716}
1717
1718static const struct seq_operations extfrag_op = {
1719 .start = frag_start,
1720 .next = frag_next,
1721 .stop = frag_stop,
1722 .show = extfrag_show,
1723};
1724
1725static int extfrag_open(struct inode *inode, struct file *file)
1726{
1727 return seq_open(file, &extfrag_op);
1728}
1729
1730static const struct file_operations extfrag_file_ops = {
1731 .open = extfrag_open,
1732 .read = seq_read,
1733 .llseek = seq_lseek,
1734 .release = seq_release,
1735};
1736
Mel Gormand7a57522010-05-24 14:32:25 -07001737static int __init extfrag_debug_init(void)
1738{
Sasikantha babubde8bd82012-05-29 15:06:22 -07001739 struct dentry *extfrag_debug_root;
1740
Mel Gormand7a57522010-05-24 14:32:25 -07001741 extfrag_debug_root = debugfs_create_dir("extfrag", NULL);
1742 if (!extfrag_debug_root)
1743 return -ENOMEM;
1744
1745 if (!debugfs_create_file("unusable_index", 0444,
1746 extfrag_debug_root, NULL, &unusable_file_ops))
Sasikantha babubde8bd82012-05-29 15:06:22 -07001747 goto fail;
Mel Gormand7a57522010-05-24 14:32:25 -07001748
Mel Gormanf1a5ab12010-05-24 14:32:26 -07001749 if (!debugfs_create_file("extfrag_index", 0444,
1750 extfrag_debug_root, NULL, &extfrag_file_ops))
Sasikantha babubde8bd82012-05-29 15:06:22 -07001751 goto fail;
Mel Gormanf1a5ab12010-05-24 14:32:26 -07001752
Mel Gormand7a57522010-05-24 14:32:25 -07001753 return 0;
Sasikantha babubde8bd82012-05-29 15:06:22 -07001754fail:
1755 debugfs_remove_recursive(extfrag_debug_root);
1756 return -ENOMEM;
Mel Gormand7a57522010-05-24 14:32:25 -07001757}
1758
1759module_init(extfrag_debug_init);
1760#endif