blob: 75eda9c2b2602fe24b4c431f797c5e0fc563ebda [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
Kemi Wang1d90ca82017-09-08 16:12:52 -070033#define NUMA_STATS_THRESHOLD (U16_MAX - 2)
34
Kemi Wang45180852017-11-15 17:38:22 -080035#ifdef CONFIG_NUMA
36int sysctl_vm_numa_stat = ENABLE_NUMA_STAT;
37
38/* zero numa counters within a zone */
39static void zero_zone_numa_counters(struct zone *zone)
40{
41 int item, cpu;
42
43 for (item = 0; item < NR_VM_NUMA_STAT_ITEMS; item++) {
44 atomic_long_set(&zone->vm_numa_stat[item], 0);
45 for_each_online_cpu(cpu)
46 per_cpu_ptr(zone->pageset, cpu)->vm_numa_stat_diff[item]
47 = 0;
48 }
49}
50
51/* zero numa counters of all the populated zones */
52static void zero_zones_numa_counters(void)
53{
54 struct zone *zone;
55
56 for_each_populated_zone(zone)
57 zero_zone_numa_counters(zone);
58}
59
60/* zero global numa counters */
61static void zero_global_numa_counters(void)
62{
63 int item;
64
65 for (item = 0; item < NR_VM_NUMA_STAT_ITEMS; item++)
66 atomic_long_set(&vm_numa_stat[item], 0);
67}
68
69static void invalid_numa_statistics(void)
70{
71 zero_zones_numa_counters();
72 zero_global_numa_counters();
73}
74
75static DEFINE_MUTEX(vm_numa_stat_lock);
76
77int sysctl_vm_numa_stat_handler(struct ctl_table *table, int write,
78 void __user *buffer, size_t *length, loff_t *ppos)
79{
80 int ret, oldval;
81
82 mutex_lock(&vm_numa_stat_lock);
83 if (write)
84 oldval = sysctl_vm_numa_stat;
85 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
86 if (ret || !write)
87 goto out;
88
89 if (oldval == sysctl_vm_numa_stat)
90 goto out;
91 else if (sysctl_vm_numa_stat == ENABLE_NUMA_STAT) {
92 static_branch_enable(&vm_numa_stat_key);
93 pr_info("enable numa statistics\n");
94 } else {
95 static_branch_disable(&vm_numa_stat_key);
96 invalid_numa_statistics();
97 pr_info("disable numa statistics, and clear numa counters\n");
98 }
99
100out:
101 mutex_unlock(&vm_numa_stat_lock);
102 return ret;
103}
104#endif
105
Christoph Lameterf8891e52006-06-30 01:55:45 -0700106#ifdef CONFIG_VM_EVENT_COUNTERS
107DEFINE_PER_CPU(struct vm_event_state, vm_event_states) = {{0}};
108EXPORT_PER_CPU_SYMBOL(vm_event_states);
109
Minchan Kim31f961a2010-08-09 17:18:59 -0700110static void sum_vm_events(unsigned long *ret)
Christoph Lameterf8891e52006-06-30 01:55:45 -0700111{
Christoph Lameter9eccf2a2008-02-04 22:29:22 -0800112 int cpu;
Christoph Lameterf8891e52006-06-30 01:55:45 -0700113 int i;
114
115 memset(ret, 0, NR_VM_EVENT_ITEMS * sizeof(unsigned long));
116
Minchan Kim31f961a2010-08-09 17:18:59 -0700117 for_each_online_cpu(cpu) {
Christoph Lameterf8891e52006-06-30 01:55:45 -0700118 struct vm_event_state *this = &per_cpu(vm_event_states, cpu);
119
Christoph Lameterf8891e52006-06-30 01:55:45 -0700120 for (i = 0; i < NR_VM_EVENT_ITEMS; i++)
121 ret[i] += this->event[i];
122 }
123}
124
125/*
126 * Accumulate the vm event counters across all CPUs.
127 * The result is unavoidably approximate - it can change
128 * during and after execution of this function.
129*/
130void all_vm_events(unsigned long *ret)
131{
KOSAKI Motohirob5be1132008-05-12 14:02:06 -0700132 get_online_cpus();
Minchan Kim31f961a2010-08-09 17:18:59 -0700133 sum_vm_events(ret);
KOSAKI Motohirob5be1132008-05-12 14:02:06 -0700134 put_online_cpus();
Christoph Lameterf8891e52006-06-30 01:55:45 -0700135}
Heiko Carstens32dd66f2006-07-10 04:44:31 -0700136EXPORT_SYMBOL_GPL(all_vm_events);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700137
Christoph Lameterf8891e52006-06-30 01:55:45 -0700138/*
139 * Fold the foreign cpu events into our own.
140 *
141 * This is adding to the events on one processor
142 * but keeps the global counts constant.
143 */
144void vm_events_fold_cpu(int cpu)
145{
146 struct vm_event_state *fold_state = &per_cpu(vm_event_states, cpu);
147 int i;
148
149 for (i = 0; i < NR_VM_EVENT_ITEMS; i++) {
150 count_vm_events(i, fold_state->event[i]);
151 fold_state->event[i] = 0;
152 }
153}
Christoph Lameterf8891e52006-06-30 01:55:45 -0700154
155#endif /* CONFIG_VM_EVENT_COUNTERS */
156
Christoph Lameter2244b952006-06-30 01:55:33 -0700157/*
158 * Manage combined zone based / global counters
159 *
160 * vm_stat contains the global counters
161 */
Mel Gorman75ef7182016-07-28 15:45:24 -0700162atomic_long_t vm_zone_stat[NR_VM_ZONE_STAT_ITEMS] __cacheline_aligned_in_smp;
Kemi Wang3a321d22017-09-08 16:12:48 -0700163atomic_long_t vm_numa_stat[NR_VM_NUMA_STAT_ITEMS] __cacheline_aligned_in_smp;
Mel Gorman75ef7182016-07-28 15:45:24 -0700164atomic_long_t vm_node_stat[NR_VM_NODE_STAT_ITEMS] __cacheline_aligned_in_smp;
165EXPORT_SYMBOL(vm_zone_stat);
Kemi Wang3a321d22017-09-08 16:12:48 -0700166EXPORT_SYMBOL(vm_numa_stat);
Mel Gorman75ef7182016-07-28 15:45:24 -0700167EXPORT_SYMBOL(vm_node_stat);
Christoph Lameter2244b952006-06-30 01:55:33 -0700168
169#ifdef CONFIG_SMP
170
Mel Gormanb44129b2011-01-13 15:45:43 -0800171int calculate_pressure_threshold(struct zone *zone)
Mel Gorman88f5acf2011-01-13 15:45:41 -0800172{
173 int threshold;
174 int watermark_distance;
175
176 /*
177 * As vmstats are not up to date, there is drift between the estimated
178 * and real values. For high thresholds and a high number of CPUs, it
179 * is possible for the min watermark to be breached while the estimated
180 * value looks fine. The pressure threshold is a reduced value such
181 * that even the maximum amount of drift will not accidentally breach
182 * the min watermark
183 */
184 watermark_distance = low_wmark_pages(zone) - min_wmark_pages(zone);
185 threshold = max(1, (int)(watermark_distance / num_online_cpus()));
186
187 /*
188 * Maximum threshold is 125
189 */
190 threshold = min(125, threshold);
191
192 return threshold;
193}
194
Mel Gormanb44129b2011-01-13 15:45:43 -0800195int calculate_normal_threshold(struct zone *zone)
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700196{
197 int threshold;
198 int mem; /* memory in 128 MB units */
199
200 /*
201 * The threshold scales with the number of processors and the amount
202 * of memory per zone. More memory means that we can defer updates for
203 * longer, more processors could lead to more contention.
204 * fls() is used to have a cheap way of logarithmic scaling.
205 *
206 * Some sample thresholds:
207 *
208 * Threshold Processors (fls) Zonesize fls(mem+1)
209 * ------------------------------------------------------------------
210 * 8 1 1 0.9-1 GB 4
211 * 16 2 2 0.9-1 GB 4
212 * 20 2 2 1-2 GB 5
213 * 24 2 2 2-4 GB 6
214 * 28 2 2 4-8 GB 7
215 * 32 2 2 8-16 GB 8
216 * 4 2 2 <128M 1
217 * 30 4 3 2-4 GB 5
218 * 48 4 3 8-16 GB 8
219 * 32 8 4 1-2 GB 4
220 * 32 8 4 0.9-1GB 4
221 * 10 16 5 <128M 1
222 * 40 16 5 900M 4
223 * 70 64 7 2-4 GB 5
224 * 84 64 7 4-8 GB 6
225 * 108 512 9 4-8 GB 6
226 * 125 1024 10 8-16 GB 8
227 * 125 1024 10 16-32 GB 9
228 */
229
Jiang Liub40da042013-02-22 16:33:52 -0800230 mem = zone->managed_pages >> (27 - PAGE_SHIFT);
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700231
232 threshold = 2 * fls(num_online_cpus()) * (1 + fls(mem));
233
234 /*
235 * Maximum threshold is 125
236 */
237 threshold = min(125, threshold);
238
239 return threshold;
240}
Christoph Lameter2244b952006-06-30 01:55:33 -0700241
242/*
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700243 * Refresh the thresholds for each zone.
Christoph Lameter2244b952006-06-30 01:55:33 -0700244 */
KOSAKI Motohiroa6cccdc2011-05-24 17:11:33 -0700245void refresh_zone_stat_thresholds(void)
Christoph Lameter2244b952006-06-30 01:55:33 -0700246{
Mel Gorman75ef7182016-07-28 15:45:24 -0700247 struct pglist_data *pgdat;
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700248 struct zone *zone;
249 int cpu;
250 int threshold;
251
Mel Gorman75ef7182016-07-28 15:45:24 -0700252 /* Zero current pgdat thresholds */
253 for_each_online_pgdat(pgdat) {
254 for_each_online_cpu(cpu) {
255 per_cpu_ptr(pgdat->per_cpu_nodestats, cpu)->stat_threshold = 0;
256 }
257 }
258
KOSAKI Motohiroee99c712009-03-31 15:19:31 -0700259 for_each_populated_zone(zone) {
Mel Gorman75ef7182016-07-28 15:45:24 -0700260 struct pglist_data *pgdat = zone->zone_pgdat;
Christoph Lameteraa454842010-09-09 16:38:17 -0700261 unsigned long max_drift, tolerate_drift;
262
Mel Gormanb44129b2011-01-13 15:45:43 -0800263 threshold = calculate_normal_threshold(zone);
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700264
Mel Gorman75ef7182016-07-28 15:45:24 -0700265 for_each_online_cpu(cpu) {
266 int pgdat_threshold;
267
Christoph Lameter99dcc3e2010-01-05 15:34:51 +0900268 per_cpu_ptr(zone->pageset, cpu)->stat_threshold
269 = threshold;
Kemi Wang1d90ca82017-09-08 16:12:52 -0700270
Mel Gorman75ef7182016-07-28 15:45:24 -0700271 /* Base nodestat threshold on the largest populated zone. */
272 pgdat_threshold = per_cpu_ptr(pgdat->per_cpu_nodestats, cpu)->stat_threshold;
273 per_cpu_ptr(pgdat->per_cpu_nodestats, cpu)->stat_threshold
274 = max(threshold, pgdat_threshold);
275 }
276
Christoph Lameteraa454842010-09-09 16:38:17 -0700277 /*
278 * Only set percpu_drift_mark if there is a danger that
279 * NR_FREE_PAGES reports the low watermark is ok when in fact
280 * the min watermark could be breached by an allocation
281 */
282 tolerate_drift = low_wmark_pages(zone) - min_wmark_pages(zone);
283 max_drift = num_online_cpus() * threshold;
284 if (max_drift > tolerate_drift)
285 zone->percpu_drift_mark = high_wmark_pages(zone) +
286 max_drift;
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700287 }
Christoph Lameter2244b952006-06-30 01:55:33 -0700288}
289
Mel Gormanb44129b2011-01-13 15:45:43 -0800290void set_pgdat_percpu_threshold(pg_data_t *pgdat,
291 int (*calculate_pressure)(struct zone *))
Mel Gorman88f5acf2011-01-13 15:45:41 -0800292{
293 struct zone *zone;
294 int cpu;
295 int threshold;
296 int i;
297
Mel Gorman88f5acf2011-01-13 15:45:41 -0800298 for (i = 0; i < pgdat->nr_zones; i++) {
299 zone = &pgdat->node_zones[i];
300 if (!zone->percpu_drift_mark)
301 continue;
302
Mel Gormanb44129b2011-01-13 15:45:43 -0800303 threshold = (*calculate_pressure)(zone);
Kemi Wang1d90ca82017-09-08 16:12:52 -0700304 for_each_online_cpu(cpu)
Mel Gorman88f5acf2011-01-13 15:45:41 -0800305 per_cpu_ptr(zone->pageset, cpu)->stat_threshold
306 = threshold;
307 }
Mel Gorman88f5acf2011-01-13 15:45:41 -0800308}
309
Christoph Lameter2244b952006-06-30 01:55:33 -0700310/*
Jianyu Zhanbea04b02014-06-04 16:09:51 -0700311 * For use when we know that interrupts are disabled,
312 * or when we know that preemption is disabled and that
313 * particular counter cannot be updated from interrupt context.
Christoph Lameter2244b952006-06-30 01:55:33 -0700314 */
315void __mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
Heiko Carstens6cdb18a2015-12-29 14:54:32 -0800316 long delta)
Christoph Lameter2244b952006-06-30 01:55:33 -0700317{
Christoph Lameter12938a92010-12-06 11:16:20 -0600318 struct per_cpu_pageset __percpu *pcp = zone->pageset;
319 s8 __percpu *p = pcp->vm_stat_diff + item;
Christoph Lameter2244b952006-06-30 01:55:33 -0700320 long x;
Christoph Lameter12938a92010-12-06 11:16:20 -0600321 long t;
Christoph Lameter2244b952006-06-30 01:55:33 -0700322
Christoph Lameter12938a92010-12-06 11:16:20 -0600323 x = delta + __this_cpu_read(*p);
Christoph Lameter2244b952006-06-30 01:55:33 -0700324
Christoph Lameter12938a92010-12-06 11:16:20 -0600325 t = __this_cpu_read(pcp->stat_threshold);
326
327 if (unlikely(x > t || x < -t)) {
Christoph Lameter2244b952006-06-30 01:55:33 -0700328 zone_page_state_add(x, zone, item);
329 x = 0;
330 }
Christoph Lameter12938a92010-12-06 11:16:20 -0600331 __this_cpu_write(*p, x);
Christoph Lameter2244b952006-06-30 01:55:33 -0700332}
333EXPORT_SYMBOL(__mod_zone_page_state);
334
Mel Gorman75ef7182016-07-28 15:45:24 -0700335void __mod_node_page_state(struct pglist_data *pgdat, enum node_stat_item item,
336 long delta)
337{
338 struct per_cpu_nodestat __percpu *pcp = pgdat->per_cpu_nodestats;
339 s8 __percpu *p = pcp->vm_node_stat_diff + item;
340 long x;
341 long t;
342
343 x = delta + __this_cpu_read(*p);
344
345 t = __this_cpu_read(pcp->stat_threshold);
346
347 if (unlikely(x > t || x < -t)) {
348 node_page_state_add(x, pgdat, item);
349 x = 0;
350 }
351 __this_cpu_write(*p, x);
352}
353EXPORT_SYMBOL(__mod_node_page_state);
354
Christoph Lameter2244b952006-06-30 01:55:33 -0700355/*
Christoph Lameter2244b952006-06-30 01:55:33 -0700356 * Optimized increment and decrement functions.
357 *
358 * These are only for a single page and therefore can take a struct page *
359 * argument instead of struct zone *. This allows the inclusion of the code
360 * generated for page_zone(page) into the optimized functions.
361 *
362 * No overflow check is necessary and therefore the differential can be
363 * incremented or decremented in place which may allow the compilers to
364 * generate better code.
Christoph Lameter2244b952006-06-30 01:55:33 -0700365 * The increment or decrement is known and therefore one boundary check can
366 * be omitted.
367 *
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700368 * NOTE: These functions are very performance sensitive. Change only
369 * with care.
370 *
Christoph Lameter2244b952006-06-30 01:55:33 -0700371 * Some processors have inc/dec instructions that are atomic vs an interrupt.
372 * However, the code must first determine the differential location in a zone
373 * based on the processor number and then inc/dec the counter. There is no
374 * guarantee without disabling preemption that the processor will not change
375 * in between and therefore the atomicity vs. interrupt cannot be exploited
376 * in a useful way here.
377 */
Christoph Lameterc8785382007-02-10 01:43:01 -0800378void __inc_zone_state(struct zone *zone, enum zone_stat_item item)
Christoph Lameter2244b952006-06-30 01:55:33 -0700379{
Christoph Lameter12938a92010-12-06 11:16:20 -0600380 struct per_cpu_pageset __percpu *pcp = zone->pageset;
381 s8 __percpu *p = pcp->vm_stat_diff + item;
382 s8 v, t;
Christoph Lameter2244b952006-06-30 01:55:33 -0700383
Christoph Lameter908ee0f2010-12-06 11:40:02 -0600384 v = __this_cpu_inc_return(*p);
Christoph Lameter12938a92010-12-06 11:16:20 -0600385 t = __this_cpu_read(pcp->stat_threshold);
386 if (unlikely(v > t)) {
387 s8 overstep = t >> 1;
Christoph Lameter2244b952006-06-30 01:55:33 -0700388
Christoph Lameter12938a92010-12-06 11:16:20 -0600389 zone_page_state_add(v + overstep, zone, item);
390 __this_cpu_write(*p, -overstep);
Christoph Lameter2244b952006-06-30 01:55:33 -0700391 }
392}
Christoph Lameterca889e62006-06-30 01:55:44 -0700393
Mel Gorman75ef7182016-07-28 15:45:24 -0700394void __inc_node_state(struct pglist_data *pgdat, enum node_stat_item item)
395{
396 struct per_cpu_nodestat __percpu *pcp = pgdat->per_cpu_nodestats;
397 s8 __percpu *p = pcp->vm_node_stat_diff + item;
398 s8 v, t;
399
400 v = __this_cpu_inc_return(*p);
401 t = __this_cpu_read(pcp->stat_threshold);
402 if (unlikely(v > t)) {
403 s8 overstep = t >> 1;
404
405 node_page_state_add(v + overstep, pgdat, item);
406 __this_cpu_write(*p, -overstep);
407 }
408}
409
Christoph Lameterca889e62006-06-30 01:55:44 -0700410void __inc_zone_page_state(struct page *page, enum zone_stat_item item)
411{
412 __inc_zone_state(page_zone(page), item);
413}
Christoph Lameter2244b952006-06-30 01:55:33 -0700414EXPORT_SYMBOL(__inc_zone_page_state);
415
Mel Gorman75ef7182016-07-28 15:45:24 -0700416void __inc_node_page_state(struct page *page, enum node_stat_item item)
417{
418 __inc_node_state(page_pgdat(page), item);
419}
420EXPORT_SYMBOL(__inc_node_page_state);
421
Christoph Lameterc8785382007-02-10 01:43:01 -0800422void __dec_zone_state(struct zone *zone, enum zone_stat_item item)
Christoph Lameter2244b952006-06-30 01:55:33 -0700423{
Christoph Lameter12938a92010-12-06 11:16:20 -0600424 struct per_cpu_pageset __percpu *pcp = zone->pageset;
425 s8 __percpu *p = pcp->vm_stat_diff + item;
426 s8 v, t;
Christoph Lameter2244b952006-06-30 01:55:33 -0700427
Christoph Lameter908ee0f2010-12-06 11:40:02 -0600428 v = __this_cpu_dec_return(*p);
Christoph Lameter12938a92010-12-06 11:16:20 -0600429 t = __this_cpu_read(pcp->stat_threshold);
430 if (unlikely(v < - t)) {
431 s8 overstep = t >> 1;
Christoph Lameter2244b952006-06-30 01:55:33 -0700432
Christoph Lameter12938a92010-12-06 11:16:20 -0600433 zone_page_state_add(v - overstep, zone, item);
434 __this_cpu_write(*p, overstep);
Christoph Lameter2244b952006-06-30 01:55:33 -0700435 }
436}
Christoph Lameterc8785382007-02-10 01:43:01 -0800437
Mel Gorman75ef7182016-07-28 15:45:24 -0700438void __dec_node_state(struct pglist_data *pgdat, enum node_stat_item item)
439{
440 struct per_cpu_nodestat __percpu *pcp = pgdat->per_cpu_nodestats;
441 s8 __percpu *p = pcp->vm_node_stat_diff + item;
442 s8 v, t;
443
444 v = __this_cpu_dec_return(*p);
445 t = __this_cpu_read(pcp->stat_threshold);
446 if (unlikely(v < - t)) {
447 s8 overstep = t >> 1;
448
449 node_page_state_add(v - overstep, pgdat, item);
450 __this_cpu_write(*p, overstep);
451 }
452}
453
Christoph Lameterc8785382007-02-10 01:43:01 -0800454void __dec_zone_page_state(struct page *page, enum zone_stat_item item)
455{
456 __dec_zone_state(page_zone(page), item);
457}
Christoph Lameter2244b952006-06-30 01:55:33 -0700458EXPORT_SYMBOL(__dec_zone_page_state);
459
Mel Gorman75ef7182016-07-28 15:45:24 -0700460void __dec_node_page_state(struct page *page, enum node_stat_item item)
461{
462 __dec_node_state(page_pgdat(page), item);
463}
464EXPORT_SYMBOL(__dec_node_page_state);
465
Heiko Carstens41561532012-01-12 17:17:30 -0800466#ifdef CONFIG_HAVE_CMPXCHG_LOCAL
Christoph Lameter7c839122010-12-14 10:28:46 -0600467/*
468 * If we have cmpxchg_local support then we do not need to incur the overhead
469 * that comes with local_irq_save/restore if we use this_cpu_cmpxchg.
470 *
471 * mod_state() modifies the zone counter state through atomic per cpu
472 * operations.
473 *
474 * Overstep mode specifies how overstep should handled:
475 * 0 No overstepping
476 * 1 Overstepping half of threshold
477 * -1 Overstepping minus half of threshold
478*/
Mel Gorman75ef7182016-07-28 15:45:24 -0700479static inline void mod_zone_state(struct zone *zone,
480 enum zone_stat_item item, long delta, int overstep_mode)
Christoph Lameter7c839122010-12-14 10:28:46 -0600481{
482 struct per_cpu_pageset __percpu *pcp = zone->pageset;
483 s8 __percpu *p = pcp->vm_stat_diff + item;
484 long o, n, t, z;
485
486 do {
487 z = 0; /* overflow to zone counters */
488
489 /*
490 * The fetching of the stat_threshold is racy. We may apply
491 * a counter threshold to the wrong the cpu if we get
Christoph Lameterd3bc2362011-04-14 15:21:58 -0700492 * rescheduled while executing here. However, the next
493 * counter update will apply the threshold again and
494 * therefore bring the counter under the threshold again.
495 *
496 * Most of the time the thresholds are the same anyways
497 * for all cpus in a zone.
Christoph Lameter7c839122010-12-14 10:28:46 -0600498 */
499 t = this_cpu_read(pcp->stat_threshold);
500
501 o = this_cpu_read(*p);
502 n = delta + o;
503
504 if (n > t || n < -t) {
505 int os = overstep_mode * (t >> 1) ;
506
507 /* Overflow must be added to zone counters */
508 z = n + os;
509 n = -os;
510 }
511 } while (this_cpu_cmpxchg(*p, o, n) != o);
512
513 if (z)
514 zone_page_state_add(z, zone, item);
515}
516
517void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
Heiko Carstens6cdb18a2015-12-29 14:54:32 -0800518 long delta)
Christoph Lameter7c839122010-12-14 10:28:46 -0600519{
Mel Gorman75ef7182016-07-28 15:45:24 -0700520 mod_zone_state(zone, item, delta, 0);
Christoph Lameter7c839122010-12-14 10:28:46 -0600521}
522EXPORT_SYMBOL(mod_zone_page_state);
523
Christoph Lameter7c839122010-12-14 10:28:46 -0600524void inc_zone_page_state(struct page *page, enum zone_stat_item item)
525{
Mel Gorman75ef7182016-07-28 15:45:24 -0700526 mod_zone_state(page_zone(page), item, 1, 1);
Christoph Lameter7c839122010-12-14 10:28:46 -0600527}
528EXPORT_SYMBOL(inc_zone_page_state);
529
530void dec_zone_page_state(struct page *page, enum zone_stat_item item)
531{
Mel Gorman75ef7182016-07-28 15:45:24 -0700532 mod_zone_state(page_zone(page), item, -1, -1);
Christoph Lameter7c839122010-12-14 10:28:46 -0600533}
534EXPORT_SYMBOL(dec_zone_page_state);
Mel Gorman75ef7182016-07-28 15:45:24 -0700535
536static inline void mod_node_state(struct pglist_data *pgdat,
537 enum node_stat_item item, int delta, int overstep_mode)
538{
539 struct per_cpu_nodestat __percpu *pcp = pgdat->per_cpu_nodestats;
540 s8 __percpu *p = pcp->vm_node_stat_diff + item;
541 long o, n, t, z;
542
543 do {
544 z = 0; /* overflow to node counters */
545
546 /*
547 * The fetching of the stat_threshold is racy. We may apply
548 * a counter threshold to the wrong the cpu if we get
549 * rescheduled while executing here. However, the next
550 * counter update will apply the threshold again and
551 * therefore bring the counter under the threshold again.
552 *
553 * Most of the time the thresholds are the same anyways
554 * for all cpus in a node.
555 */
556 t = this_cpu_read(pcp->stat_threshold);
557
558 o = this_cpu_read(*p);
559 n = delta + o;
560
561 if (n > t || n < -t) {
562 int os = overstep_mode * (t >> 1) ;
563
564 /* Overflow must be added to node counters */
565 z = n + os;
566 n = -os;
567 }
568 } while (this_cpu_cmpxchg(*p, o, n) != o);
569
570 if (z)
571 node_page_state_add(z, pgdat, item);
572}
573
574void mod_node_page_state(struct pglist_data *pgdat, enum node_stat_item item,
575 long delta)
576{
577 mod_node_state(pgdat, item, delta, 0);
578}
579EXPORT_SYMBOL(mod_node_page_state);
580
581void inc_node_state(struct pglist_data *pgdat, enum node_stat_item item)
582{
583 mod_node_state(pgdat, item, 1, 1);
584}
585
586void inc_node_page_state(struct page *page, enum node_stat_item item)
587{
588 mod_node_state(page_pgdat(page), item, 1, 1);
589}
590EXPORT_SYMBOL(inc_node_page_state);
591
592void dec_node_page_state(struct page *page, enum node_stat_item item)
593{
594 mod_node_state(page_pgdat(page), item, -1, -1);
595}
596EXPORT_SYMBOL(dec_node_page_state);
Christoph Lameter7c839122010-12-14 10:28:46 -0600597#else
598/*
599 * Use interrupt disable to serialize counter updates
600 */
601void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
Heiko Carstens6cdb18a2015-12-29 14:54:32 -0800602 long delta)
Christoph Lameter7c839122010-12-14 10:28:46 -0600603{
604 unsigned long flags;
605
606 local_irq_save(flags);
607 __mod_zone_page_state(zone, item, delta);
608 local_irq_restore(flags);
609}
610EXPORT_SYMBOL(mod_zone_page_state);
611
Christoph Lameter2244b952006-06-30 01:55:33 -0700612void inc_zone_page_state(struct page *page, enum zone_stat_item item)
613{
614 unsigned long flags;
615 struct zone *zone;
Christoph Lameter2244b952006-06-30 01:55:33 -0700616
617 zone = page_zone(page);
618 local_irq_save(flags);
Christoph Lameterca889e62006-06-30 01:55:44 -0700619 __inc_zone_state(zone, item);
Christoph Lameter2244b952006-06-30 01:55:33 -0700620 local_irq_restore(flags);
621}
622EXPORT_SYMBOL(inc_zone_page_state);
623
624void dec_zone_page_state(struct page *page, enum zone_stat_item item)
625{
626 unsigned long flags;
Christoph Lameter2244b952006-06-30 01:55:33 -0700627
Christoph Lameter2244b952006-06-30 01:55:33 -0700628 local_irq_save(flags);
Christoph Lametera302eb42006-08-31 21:27:34 -0700629 __dec_zone_page_state(page, item);
Christoph Lameter2244b952006-06-30 01:55:33 -0700630 local_irq_restore(flags);
631}
632EXPORT_SYMBOL(dec_zone_page_state);
633
Mel Gorman75ef7182016-07-28 15:45:24 -0700634void inc_node_state(struct pglist_data *pgdat, enum node_stat_item item)
635{
636 unsigned long flags;
637
638 local_irq_save(flags);
639 __inc_node_state(pgdat, item);
640 local_irq_restore(flags);
641}
642EXPORT_SYMBOL(inc_node_state);
643
644void mod_node_page_state(struct pglist_data *pgdat, enum node_stat_item item,
645 long delta)
646{
647 unsigned long flags;
648
649 local_irq_save(flags);
650 __mod_node_page_state(pgdat, item, delta);
651 local_irq_restore(flags);
652}
653EXPORT_SYMBOL(mod_node_page_state);
654
655void inc_node_page_state(struct page *page, enum node_stat_item item)
656{
657 unsigned long flags;
658 struct pglist_data *pgdat;
659
660 pgdat = page_pgdat(page);
661 local_irq_save(flags);
662 __inc_node_state(pgdat, item);
663 local_irq_restore(flags);
664}
665EXPORT_SYMBOL(inc_node_page_state);
666
667void dec_node_page_state(struct page *page, enum node_stat_item item)
668{
669 unsigned long flags;
670
671 local_irq_save(flags);
672 __dec_node_page_state(page, item);
673 local_irq_restore(flags);
674}
675EXPORT_SYMBOL(dec_node_page_state);
676#endif
Christoph Lameter7cc36bb2014-10-09 15:29:43 -0700677
678/*
679 * Fold a differential into the global counters.
680 * Returns the number of counters updated.
681 */
Kemi Wang3a321d22017-09-08 16:12:48 -0700682#ifdef CONFIG_NUMA
683static int fold_diff(int *zone_diff, int *numa_diff, int *node_diff)
684{
685 int i;
686 int changes = 0;
687
688 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
689 if (zone_diff[i]) {
690 atomic_long_add(zone_diff[i], &vm_zone_stat[i]);
691 changes++;
692 }
693
694 for (i = 0; i < NR_VM_NUMA_STAT_ITEMS; i++)
695 if (numa_diff[i]) {
696 atomic_long_add(numa_diff[i], &vm_numa_stat[i]);
697 changes++;
698 }
699
700 for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++)
701 if (node_diff[i]) {
702 atomic_long_add(node_diff[i], &vm_node_stat[i]);
703 changes++;
704 }
705 return changes;
706}
707#else
Mel Gorman75ef7182016-07-28 15:45:24 -0700708static int fold_diff(int *zone_diff, int *node_diff)
Christoph Lameter4edb0742013-09-11 14:21:31 -0700709{
710 int i;
Christoph Lameter7cc36bb2014-10-09 15:29:43 -0700711 int changes = 0;
Christoph Lameter4edb0742013-09-11 14:21:31 -0700712
713 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
Mel Gorman75ef7182016-07-28 15:45:24 -0700714 if (zone_diff[i]) {
715 atomic_long_add(zone_diff[i], &vm_zone_stat[i]);
716 changes++;
717 }
718
719 for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++)
720 if (node_diff[i]) {
721 atomic_long_add(node_diff[i], &vm_node_stat[i]);
Christoph Lameter7cc36bb2014-10-09 15:29:43 -0700722 changes++;
723 }
724 return changes;
Christoph Lameter4edb0742013-09-11 14:21:31 -0700725}
Kemi Wang3a321d22017-09-08 16:12:48 -0700726#endif /* CONFIG_NUMA */
Christoph Lameter4edb0742013-09-11 14:21:31 -0700727
Christoph Lameter2244b952006-06-30 01:55:33 -0700728/*
Christoph Lameter2bb921e2013-09-11 14:21:30 -0700729 * Update the zone counters for the current cpu.
Christoph Lametera7f75e22008-02-04 22:29:16 -0800730 *
Christoph Lameter4037d452007-05-09 02:35:14 -0700731 * Note that refresh_cpu_vm_stats strives to only access
732 * node local memory. The per cpu pagesets on remote zones are placed
733 * in the memory local to the processor using that pageset. So the
734 * loop over all zones will access a series of cachelines local to
735 * the processor.
736 *
737 * The call to zone_page_state_add updates the cachelines with the
738 * statistics in the remote zone struct as well as the global cachelines
739 * with the global counters. These could cause remote node cache line
740 * bouncing and will have to be only done when necessary.
Christoph Lameter7cc36bb2014-10-09 15:29:43 -0700741 *
742 * The function returns the number of global counters updated.
Christoph Lameter2244b952006-06-30 01:55:33 -0700743 */
Christoph Lameter0eb77e92016-01-14 15:21:40 -0800744static int refresh_cpu_vm_stats(bool do_pagesets)
Christoph Lameter2244b952006-06-30 01:55:33 -0700745{
Mel Gorman75ef7182016-07-28 15:45:24 -0700746 struct pglist_data *pgdat;
Christoph Lameter2244b952006-06-30 01:55:33 -0700747 struct zone *zone;
748 int i;
Mel Gorman75ef7182016-07-28 15:45:24 -0700749 int global_zone_diff[NR_VM_ZONE_STAT_ITEMS] = { 0, };
Kemi Wang3a321d22017-09-08 16:12:48 -0700750#ifdef CONFIG_NUMA
751 int global_numa_diff[NR_VM_NUMA_STAT_ITEMS] = { 0, };
752#endif
Mel Gorman75ef7182016-07-28 15:45:24 -0700753 int global_node_diff[NR_VM_NODE_STAT_ITEMS] = { 0, };
Christoph Lameter7cc36bb2014-10-09 15:29:43 -0700754 int changes = 0;
Christoph Lameter2244b952006-06-30 01:55:33 -0700755
KOSAKI Motohiroee99c712009-03-31 15:19:31 -0700756 for_each_populated_zone(zone) {
Christoph Lameterfbc2edb2013-09-11 14:21:32 -0700757 struct per_cpu_pageset __percpu *p = zone->pageset;
Christoph Lameter2244b952006-06-30 01:55:33 -0700758
Christoph Lameterfbc2edb2013-09-11 14:21:32 -0700759 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) {
760 int v;
Christoph Lameter2244b952006-06-30 01:55:33 -0700761
Christoph Lameterfbc2edb2013-09-11 14:21:32 -0700762 v = this_cpu_xchg(p->vm_stat_diff[i], 0);
763 if (v) {
Christoph Lametera7f75e22008-02-04 22:29:16 -0800764
Christoph Lametera7f75e22008-02-04 22:29:16 -0800765 atomic_long_add(v, &zone->vm_stat[i]);
Mel Gorman75ef7182016-07-28 15:45:24 -0700766 global_zone_diff[i] += v;
Christoph Lameter4037d452007-05-09 02:35:14 -0700767#ifdef CONFIG_NUMA
768 /* 3 seconds idle till flush */
Christoph Lameterfbc2edb2013-09-11 14:21:32 -0700769 __this_cpu_write(p->expire, 3);
Christoph Lameter4037d452007-05-09 02:35:14 -0700770#endif
Christoph Lameter2244b952006-06-30 01:55:33 -0700771 }
Christoph Lameterfbc2edb2013-09-11 14:21:32 -0700772 }
Christoph Lameter4037d452007-05-09 02:35:14 -0700773#ifdef CONFIG_NUMA
Kemi Wang3a321d22017-09-08 16:12:48 -0700774 for (i = 0; i < NR_VM_NUMA_STAT_ITEMS; i++) {
775 int v;
776
777 v = this_cpu_xchg(p->vm_numa_stat_diff[i], 0);
778 if (v) {
779
780 atomic_long_add(v, &zone->vm_numa_stat[i]);
781 global_numa_diff[i] += v;
782 __this_cpu_write(p->expire, 3);
783 }
784 }
785
Christoph Lameter0eb77e92016-01-14 15:21:40 -0800786 if (do_pagesets) {
787 cond_resched();
788 /*
789 * Deal with draining the remote pageset of this
790 * processor
791 *
792 * Check if there are pages remaining in this pageset
793 * if not then there is nothing to expire.
794 */
795 if (!__this_cpu_read(p->expire) ||
Christoph Lameterfbc2edb2013-09-11 14:21:32 -0700796 !__this_cpu_read(p->pcp.count))
Christoph Lameter0eb77e92016-01-14 15:21:40 -0800797 continue;
Christoph Lameter4037d452007-05-09 02:35:14 -0700798
Christoph Lameter0eb77e92016-01-14 15:21:40 -0800799 /*
800 * We never drain zones local to this processor.
801 */
802 if (zone_to_nid(zone) == numa_node_id()) {
803 __this_cpu_write(p->expire, 0);
804 continue;
805 }
Christoph Lameter4037d452007-05-09 02:35:14 -0700806
Christoph Lameter0eb77e92016-01-14 15:21:40 -0800807 if (__this_cpu_dec_return(p->expire))
808 continue;
Christoph Lameter4037d452007-05-09 02:35:14 -0700809
Christoph Lameter0eb77e92016-01-14 15:21:40 -0800810 if (__this_cpu_read(p->pcp.count)) {
811 drain_zone_pages(zone, this_cpu_ptr(&p->pcp));
812 changes++;
813 }
Christoph Lameter7cc36bb2014-10-09 15:29:43 -0700814 }
Christoph Lameter4037d452007-05-09 02:35:14 -0700815#endif
Christoph Lameter2244b952006-06-30 01:55:33 -0700816 }
Mel Gorman75ef7182016-07-28 15:45:24 -0700817
818 for_each_online_pgdat(pgdat) {
819 struct per_cpu_nodestat __percpu *p = pgdat->per_cpu_nodestats;
820
821 for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) {
822 int v;
823
824 v = this_cpu_xchg(p->vm_node_stat_diff[i], 0);
825 if (v) {
826 atomic_long_add(v, &pgdat->vm_stat[i]);
827 global_node_diff[i] += v;
828 }
829 }
830 }
831
Kemi Wang3a321d22017-09-08 16:12:48 -0700832#ifdef CONFIG_NUMA
833 changes += fold_diff(global_zone_diff, global_numa_diff,
834 global_node_diff);
835#else
Mel Gorman75ef7182016-07-28 15:45:24 -0700836 changes += fold_diff(global_zone_diff, global_node_diff);
Kemi Wang3a321d22017-09-08 16:12:48 -0700837#endif
Christoph Lameter7cc36bb2014-10-09 15:29:43 -0700838 return changes;
Christoph Lameter2244b952006-06-30 01:55:33 -0700839}
840
Cody P Schafer40f4b1e2013-04-29 15:08:38 -0700841/*
Christoph Lameter2bb921e2013-09-11 14:21:30 -0700842 * Fold the data for an offline cpu into the global array.
843 * There cannot be any access by the offline cpu and therefore
844 * synchronization is simplified.
845 */
846void cpu_vm_stats_fold(int cpu)
847{
Mel Gorman75ef7182016-07-28 15:45:24 -0700848 struct pglist_data *pgdat;
Christoph Lameter2bb921e2013-09-11 14:21:30 -0700849 struct zone *zone;
850 int i;
Mel Gorman75ef7182016-07-28 15:45:24 -0700851 int global_zone_diff[NR_VM_ZONE_STAT_ITEMS] = { 0, };
Kemi Wang3a321d22017-09-08 16:12:48 -0700852#ifdef CONFIG_NUMA
853 int global_numa_diff[NR_VM_NUMA_STAT_ITEMS] = { 0, };
854#endif
Mel Gorman75ef7182016-07-28 15:45:24 -0700855 int global_node_diff[NR_VM_NODE_STAT_ITEMS] = { 0, };
Christoph Lameter2bb921e2013-09-11 14:21:30 -0700856
857 for_each_populated_zone(zone) {
858 struct per_cpu_pageset *p;
859
860 p = per_cpu_ptr(zone->pageset, cpu);
861
862 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
863 if (p->vm_stat_diff[i]) {
864 int v;
865
866 v = p->vm_stat_diff[i];
867 p->vm_stat_diff[i] = 0;
868 atomic_long_add(v, &zone->vm_stat[i]);
Mel Gorman75ef7182016-07-28 15:45:24 -0700869 global_zone_diff[i] += v;
Christoph Lameter2bb921e2013-09-11 14:21:30 -0700870 }
Kemi Wang3a321d22017-09-08 16:12:48 -0700871
872#ifdef CONFIG_NUMA
873 for (i = 0; i < NR_VM_NUMA_STAT_ITEMS; i++)
874 if (p->vm_numa_stat_diff[i]) {
875 int v;
876
877 v = p->vm_numa_stat_diff[i];
878 p->vm_numa_stat_diff[i] = 0;
879 atomic_long_add(v, &zone->vm_numa_stat[i]);
880 global_numa_diff[i] += v;
881 }
882#endif
Christoph Lameter2bb921e2013-09-11 14:21:30 -0700883 }
884
Mel Gorman75ef7182016-07-28 15:45:24 -0700885 for_each_online_pgdat(pgdat) {
886 struct per_cpu_nodestat *p;
887
888 p = per_cpu_ptr(pgdat->per_cpu_nodestats, cpu);
889
890 for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++)
891 if (p->vm_node_stat_diff[i]) {
892 int v;
893
894 v = p->vm_node_stat_diff[i];
895 p->vm_node_stat_diff[i] = 0;
896 atomic_long_add(v, &pgdat->vm_stat[i]);
897 global_node_diff[i] += v;
898 }
899 }
900
Kemi Wang3a321d22017-09-08 16:12:48 -0700901#ifdef CONFIG_NUMA
902 fold_diff(global_zone_diff, global_numa_diff, global_node_diff);
903#else
Mel Gorman75ef7182016-07-28 15:45:24 -0700904 fold_diff(global_zone_diff, global_node_diff);
Kemi Wang3a321d22017-09-08 16:12:48 -0700905#endif
Christoph Lameter2bb921e2013-09-11 14:21:30 -0700906}
907
908/*
Cody P Schafer40f4b1e2013-04-29 15:08:38 -0700909 * this is only called if !populated_zone(zone), which implies no other users of
910 * pset->vm_stat_diff[] exsist.
911 */
Minchan Kim5a883812012-10-08 16:33:39 -0700912void drain_zonestat(struct zone *zone, struct per_cpu_pageset *pset)
913{
914 int i;
915
916 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
917 if (pset->vm_stat_diff[i]) {
918 int v = pset->vm_stat_diff[i];
919 pset->vm_stat_diff[i] = 0;
920 atomic_long_add(v, &zone->vm_stat[i]);
Mel Gorman75ef7182016-07-28 15:45:24 -0700921 atomic_long_add(v, &vm_zone_stat[i]);
Minchan Kim5a883812012-10-08 16:33:39 -0700922 }
Kemi Wang3a321d22017-09-08 16:12:48 -0700923
924#ifdef CONFIG_NUMA
925 for (i = 0; i < NR_VM_NUMA_STAT_ITEMS; i++)
926 if (pset->vm_numa_stat_diff[i]) {
927 int v = pset->vm_numa_stat_diff[i];
928
929 pset->vm_numa_stat_diff[i] = 0;
930 atomic_long_add(v, &zone->vm_numa_stat[i]);
931 atomic_long_add(v, &vm_numa_stat[i]);
932 }
933#endif
Minchan Kim5a883812012-10-08 16:33:39 -0700934}
Christoph Lameter2244b952006-06-30 01:55:33 -0700935#endif
936
Christoph Lameterca889e62006-06-30 01:55:44 -0700937#ifdef CONFIG_NUMA
Kemi Wang3a321d22017-09-08 16:12:48 -0700938void __inc_numa_state(struct zone *zone,
939 enum numa_stat_item item)
940{
941 struct per_cpu_pageset __percpu *pcp = zone->pageset;
Kemi Wang1d90ca82017-09-08 16:12:52 -0700942 u16 __percpu *p = pcp->vm_numa_stat_diff + item;
943 u16 v;
Kemi Wang3a321d22017-09-08 16:12:48 -0700944
945 v = __this_cpu_inc_return(*p);
Kemi Wang3a321d22017-09-08 16:12:48 -0700946
Kemi Wang1d90ca82017-09-08 16:12:52 -0700947 if (unlikely(v > NUMA_STATS_THRESHOLD)) {
948 zone_numa_state_add(v, zone, item);
949 __this_cpu_write(*p, 0);
Kemi Wang3a321d22017-09-08 16:12:48 -0700950 }
951}
952
Christoph Lameterca889e62006-06-30 01:55:44 -0700953/*
Mel Gorman75ef7182016-07-28 15:45:24 -0700954 * Determine the per node value of a stat item. This function
955 * is called frequently in a NUMA machine, so try to be as
956 * frugal as possible.
Andrew Mortonc2d42c12015-11-05 18:48:43 -0800957 */
Mel Gorman75ef7182016-07-28 15:45:24 -0700958unsigned long sum_zone_node_page_state(int node,
959 enum zone_stat_item item)
Andrew Mortonc2d42c12015-11-05 18:48:43 -0800960{
961 struct zone *zones = NODE_DATA(node)->node_zones;
Joonsoo Kime87d59f2016-05-19 17:12:29 -0700962 int i;
963 unsigned long count = 0;
Andrew Mortonc2d42c12015-11-05 18:48:43 -0800964
Joonsoo Kime87d59f2016-05-19 17:12:29 -0700965 for (i = 0; i < MAX_NR_ZONES; i++)
966 count += zone_page_state(zones + i, item);
967
968 return count;
Andrew Mortonc2d42c12015-11-05 18:48:43 -0800969}
970
Kemi Wang63803222017-09-08 16:12:55 -0700971/*
972 * Determine the per node value of a numa stat item. To avoid deviation,
973 * the per cpu stat number in vm_numa_stat_diff[] is also included.
974 */
Kemi Wang3a321d22017-09-08 16:12:48 -0700975unsigned long sum_zone_numa_state(int node,
976 enum numa_stat_item item)
977{
978 struct zone *zones = NODE_DATA(node)->node_zones;
979 int i;
980 unsigned long count = 0;
981
982 for (i = 0; i < MAX_NR_ZONES; i++)
Kemi Wang63803222017-09-08 16:12:55 -0700983 count += zone_numa_state_snapshot(zones + i, item);
Kemi Wang3a321d22017-09-08 16:12:48 -0700984
985 return count;
986}
987
Mel Gorman75ef7182016-07-28 15:45:24 -0700988/*
989 * Determine the per node value of a stat item.
990 */
991unsigned long node_page_state(struct pglist_data *pgdat,
992 enum node_stat_item item)
993{
994 long x = atomic_long_read(&pgdat->vm_stat[item]);
995#ifdef CONFIG_SMP
996 if (x < 0)
997 x = 0;
998#endif
999 return x;
1000}
Christoph Lameterca889e62006-06-30 01:55:44 -07001001#endif
1002
Mel Gormand7a57522010-05-24 14:32:25 -07001003#ifdef CONFIG_COMPACTION
Namhyung Kim36deb0b2010-10-26 14:22:04 -07001004
Mel Gormand7a57522010-05-24 14:32:25 -07001005struct contig_page_info {
1006 unsigned long free_pages;
1007 unsigned long free_blocks_total;
1008 unsigned long free_blocks_suitable;
1009};
1010
1011/*
1012 * Calculate the number of free pages in a zone, how many contiguous
1013 * pages are free and how many are large enough to satisfy an allocation of
1014 * the target size. Note that this function makes no attempt to estimate
1015 * how many suitable free blocks there *might* be if MOVABLE pages were
1016 * migrated. Calculating that is possible, but expensive and can be
1017 * figured out from userspace
1018 */
1019static void fill_contig_page_info(struct zone *zone,
1020 unsigned int suitable_order,
1021 struct contig_page_info *info)
1022{
1023 unsigned int order;
1024
1025 info->free_pages = 0;
1026 info->free_blocks_total = 0;
1027 info->free_blocks_suitable = 0;
1028
1029 for (order = 0; order < MAX_ORDER; order++) {
1030 unsigned long blocks;
1031
1032 /* Count number of free blocks */
1033 blocks = zone->free_area[order].nr_free;
1034 info->free_blocks_total += blocks;
1035
1036 /* Count free base pages */
1037 info->free_pages += blocks << order;
1038
1039 /* Count the suitable free blocks */
1040 if (order >= suitable_order)
1041 info->free_blocks_suitable += blocks <<
1042 (order - suitable_order);
1043 }
1044}
Mel Gormanf1a5ab12010-05-24 14:32:26 -07001045
1046/*
1047 * A fragmentation index only makes sense if an allocation of a requested
1048 * size would fail. If that is true, the fragmentation index indicates
1049 * whether external fragmentation or a lack of memory was the problem.
1050 * The value can be used to determine if page reclaim or compaction
1051 * should be used
1052 */
Mel Gorman56de7262010-05-24 14:32:30 -07001053static int __fragmentation_index(unsigned int order, struct contig_page_info *info)
Mel Gormanf1a5ab12010-05-24 14:32:26 -07001054{
1055 unsigned long requested = 1UL << order;
1056
Wen Yang88d6ac42017-09-06 16:24:06 -07001057 if (WARN_ON_ONCE(order >= MAX_ORDER))
1058 return 0;
1059
Mel Gormanf1a5ab12010-05-24 14:32:26 -07001060 if (!info->free_blocks_total)
1061 return 0;
1062
1063 /* Fragmentation index only makes sense when a request would fail */
1064 if (info->free_blocks_suitable)
1065 return -1000;
1066
1067 /*
1068 * Index is between 0 and 1 so return within 3 decimal places
1069 *
1070 * 0 => allocation would fail due to lack of memory
1071 * 1 => allocation would fail due to fragmentation
1072 */
1073 return 1000 - div_u64( (1000+(div_u64(info->free_pages * 1000ULL, requested))), info->free_blocks_total);
1074}
Mel Gorman56de7262010-05-24 14:32:30 -07001075
1076/* Same as __fragmentation index but allocs contig_page_info on stack */
1077int fragmentation_index(struct zone *zone, unsigned int order)
1078{
1079 struct contig_page_info info;
1080
1081 fill_contig_page_info(zone, order, &info);
1082 return __fragmentation_index(order, &info);
1083}
Mel Gormand7a57522010-05-24 14:32:25 -07001084#endif
1085
David Rientjes0d6617c2011-09-14 16:21:05 -07001086#if defined(CONFIG_PROC_FS) || defined(CONFIG_SYSFS) || defined(CONFIG_NUMA)
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001087#ifdef CONFIG_ZONE_DMA
1088#define TEXT_FOR_DMA(xx) xx "_dma",
1089#else
1090#define TEXT_FOR_DMA(xx)
1091#endif
1092
1093#ifdef CONFIG_ZONE_DMA32
1094#define TEXT_FOR_DMA32(xx) xx "_dma32",
1095#else
1096#define TEXT_FOR_DMA32(xx)
1097#endif
1098
1099#ifdef CONFIG_HIGHMEM
1100#define TEXT_FOR_HIGHMEM(xx) xx "_high",
1101#else
1102#define TEXT_FOR_HIGHMEM(xx)
1103#endif
1104
1105#define TEXTS_FOR_ZONES(xx) TEXT_FOR_DMA(xx) TEXT_FOR_DMA32(xx) xx "_normal", \
1106 TEXT_FOR_HIGHMEM(xx) xx "_movable",
1107
1108const char * const vmstat_text[] = {
Konstantin Khlebnikov09316c02014-10-09 15:29:32 -07001109 /* enum zone_stat_item countes */
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001110 "nr_free_pages",
Minchan Kim71c799f2016-07-28 15:47:26 -07001111 "nr_zone_inactive_anon",
1112 "nr_zone_active_anon",
1113 "nr_zone_inactive_file",
1114 "nr_zone_active_file",
1115 "nr_zone_unevictable",
Mel Gorman5a1c84b2016-07-28 15:47:31 -07001116 "nr_zone_write_pending",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001117 "nr_mlock",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001118 "nr_page_table_pages",
1119 "nr_kernel_stack",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001120 "nr_bounce",
Minchan Kim91537fe2016-07-26 15:24:45 -07001121#if IS_ENABLED(CONFIG_ZSMALLOC)
1122 "nr_zspages",
1123#endif
Kemi Wang3a321d22017-09-08 16:12:48 -07001124 "nr_free_cma",
1125
1126 /* enum numa_stat_item counters */
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001127#ifdef CONFIG_NUMA
1128 "numa_hit",
1129 "numa_miss",
1130 "numa_foreign",
1131 "numa_interleave",
1132 "numa_local",
1133 "numa_other",
1134#endif
Konstantin Khlebnikov09316c02014-10-09 15:29:32 -07001135
Mel Gorman599d0c92016-07-28 15:45:31 -07001136 /* Node-based counters */
1137 "nr_inactive_anon",
1138 "nr_active_anon",
1139 "nr_inactive_file",
1140 "nr_active_file",
1141 "nr_unevictable",
Johannes Weiner385386c2017-07-06 15:40:43 -07001142 "nr_slab_reclaimable",
1143 "nr_slab_unreclaimable",
Mel Gorman599d0c92016-07-28 15:45:31 -07001144 "nr_isolated_anon",
1145 "nr_isolated_file",
Mel Gorman1e6b10852016-07-28 15:46:08 -07001146 "workingset_refault",
1147 "workingset_activate",
1148 "workingset_nodereclaim",
Mel Gorman50658e22016-07-28 15:46:14 -07001149 "nr_anon_pages",
1150 "nr_mapped",
Mel Gorman11fb9982016-07-28 15:46:20 -07001151 "nr_file_pages",
1152 "nr_dirty",
1153 "nr_writeback",
1154 "nr_writeback_temp",
1155 "nr_shmem",
1156 "nr_shmem_hugepages",
1157 "nr_shmem_pmdmapped",
1158 "nr_anon_transparent_hugepages",
1159 "nr_unstable",
Mel Gormanc4a25632016-07-28 15:46:23 -07001160 "nr_vmscan_write",
1161 "nr_vmscan_immediate_reclaim",
1162 "nr_dirtied",
1163 "nr_written",
Roman Gushchin7aaf7722018-05-11 16:01:53 -07001164 "", /* nr_indirectly_reclaimable */
Mel Gorman599d0c92016-07-28 15:45:31 -07001165
Konstantin Khlebnikov09316c02014-10-09 15:29:32 -07001166 /* enum writeback_stat_item counters */
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001167 "nr_dirty_threshold",
1168 "nr_dirty_background_threshold",
1169
1170#ifdef CONFIG_VM_EVENT_COUNTERS
Konstantin Khlebnikov09316c02014-10-09 15:29:32 -07001171 /* enum vm_event_item counters */
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001172 "pgpgin",
1173 "pgpgout",
1174 "pswpin",
1175 "pswpout",
1176
1177 TEXTS_FOR_ZONES("pgalloc")
Mel Gorman7cc30fc2016-07-28 15:46:59 -07001178 TEXTS_FOR_ZONES("allocstall")
1179 TEXTS_FOR_ZONES("pgskip")
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001180
1181 "pgfree",
1182 "pgactivate",
1183 "pgdeactivate",
Shaohua Lif7ad2a62017-05-03 14:52:29 -07001184 "pglazyfree",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001185
1186 "pgfault",
1187 "pgmajfault",
Minchan Kim854e9ed2016-01-15 16:54:53 -08001188 "pglazyfreed",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001189
Mel Gorman599d0c92016-07-28 15:45:31 -07001190 "pgrefill",
1191 "pgsteal_kswapd",
1192 "pgsteal_direct",
1193 "pgscan_kswapd",
1194 "pgscan_direct",
Mel Gorman68243e72012-07-31 16:44:39 -07001195 "pgscan_direct_throttle",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001196
1197#ifdef CONFIG_NUMA
1198 "zone_reclaim_failed",
1199#endif
1200 "pginodesteal",
1201 "slabs_scanned",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001202 "kswapd_inodesteal",
1203 "kswapd_low_wmark_hit_quickly",
1204 "kswapd_high_wmark_hit_quickly",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001205 "pageoutrun",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001206
1207 "pgrotated",
1208
Dave Hansen5509a5d2014-04-03 14:48:19 -07001209 "drop_pagecache",
1210 "drop_slab",
Konstantin Khlebnikov8e675f72017-07-06 15:40:28 -07001211 "oom_kill",
Dave Hansen5509a5d2014-04-03 14:48:19 -07001212
Mel Gorman03c5a6e2012-11-02 14:52:48 +00001213#ifdef CONFIG_NUMA_BALANCING
1214 "numa_pte_updates",
Mel Gorman72403b42013-11-12 15:08:32 -08001215 "numa_huge_pte_updates",
Mel Gorman03c5a6e2012-11-02 14:52:48 +00001216 "numa_hint_faults",
1217 "numa_hint_faults_local",
1218 "numa_pages_migrated",
1219#endif
Mel Gorman5647bc22012-10-19 10:46:20 +01001220#ifdef CONFIG_MIGRATION
1221 "pgmigrate_success",
1222 "pgmigrate_fail",
1223#endif
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001224#ifdef CONFIG_COMPACTION
Mel Gorman397487d2012-10-19 12:00:10 +01001225 "compact_migrate_scanned",
1226 "compact_free_scanned",
1227 "compact_isolated",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001228 "compact_stall",
1229 "compact_fail",
1230 "compact_success",
Vlastimil Babka698b1b32016-03-17 14:18:08 -07001231 "compact_daemon_wake",
David Rientjes7f354a52017-02-22 15:44:50 -08001232 "compact_daemon_migrate_scanned",
1233 "compact_daemon_free_scanned",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001234#endif
1235
1236#ifdef CONFIG_HUGETLB_PAGE
1237 "htlb_buddy_alloc_success",
1238 "htlb_buddy_alloc_fail",
1239#endif
1240 "unevictable_pgs_culled",
1241 "unevictable_pgs_scanned",
1242 "unevictable_pgs_rescued",
1243 "unevictable_pgs_mlocked",
1244 "unevictable_pgs_munlocked",
1245 "unevictable_pgs_cleared",
1246 "unevictable_pgs_stranded",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001247
1248#ifdef CONFIG_TRANSPARENT_HUGEPAGE
1249 "thp_fault_alloc",
1250 "thp_fault_fallback",
1251 "thp_collapse_alloc",
1252 "thp_collapse_alloc_failed",
Kirill A. Shutemov95ecedc2016-07-26 15:25:31 -07001253 "thp_file_alloc",
1254 "thp_file_mapped",
Kirill A. Shutemov122afea2016-01-15 16:52:46 -08001255 "thp_split_page",
1256 "thp_split_page_failed",
Kirill A. Shutemovf9719a02016-03-17 14:18:45 -07001257 "thp_deferred_split_page",
Kirill A. Shutemov122afea2016-01-15 16:52:46 -08001258 "thp_split_pmd",
Yisheng Xiece9311c2017-03-09 16:17:00 -08001259#ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
1260 "thp_split_pud",
1261#endif
Kirill A. Shutemovd8a8e1f2012-12-12 13:51:09 -08001262 "thp_zero_page_alloc",
1263 "thp_zero_page_alloc_failed",
Huang Ying225311a2017-09-06 16:22:30 -07001264 "thp_swpout",
Huang Yingfe490cc2017-09-06 16:22:52 -07001265 "thp_swpout_fallback",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001266#endif
Konstantin Khlebnikov09316c02014-10-09 15:29:32 -07001267#ifdef CONFIG_MEMORY_BALLOON
1268 "balloon_inflate",
1269 "balloon_deflate",
1270#ifdef CONFIG_BALLOON_COMPACTION
1271 "balloon_migrate",
1272#endif
1273#endif /* CONFIG_MEMORY_BALLOON */
Mel Gormanec659932014-01-21 14:33:16 -08001274#ifdef CONFIG_DEBUG_TLBFLUSH
Dave Hansen6df46862013-09-11 14:20:24 -07001275#ifdef CONFIG_SMP
Dave Hansen9824cf92013-09-11 14:20:23 -07001276 "nr_tlb_remote_flush",
1277 "nr_tlb_remote_flush_received",
Mel Gormanec659932014-01-21 14:33:16 -08001278#endif /* CONFIG_SMP */
Dave Hansen9824cf92013-09-11 14:20:23 -07001279 "nr_tlb_local_flush_all",
1280 "nr_tlb_local_flush_one",
Mel Gormanec659932014-01-21 14:33:16 -08001281#endif /* CONFIG_DEBUG_TLBFLUSH */
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001282
Davidlohr Bueso4f115142014-06-04 16:06:46 -07001283#ifdef CONFIG_DEBUG_VM_VMACACHE
1284 "vmacache_find_calls",
1285 "vmacache_find_hits",
Davidlohr Buesof5f302e2014-12-12 16:56:10 -08001286 "vmacache_full_flushes",
Davidlohr Bueso4f115142014-06-04 16:06:46 -07001287#endif
Huang Yingcbc65df2017-09-06 16:24:29 -07001288#ifdef CONFIG_SWAP
1289 "swap_ra",
1290 "swap_ra_hit",
1291#endif
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001292#endif /* CONFIG_VM_EVENTS_COUNTERS */
1293};
David Rientjes0d6617c2011-09-14 16:21:05 -07001294#endif /* CONFIG_PROC_FS || CONFIG_SYSFS || CONFIG_NUMA */
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001295
Andrew Morton3c486872015-02-10 14:09:43 -08001296#if (defined(CONFIG_DEBUG_FS) && defined(CONFIG_COMPACTION)) || \
1297 defined(CONFIG_PROC_FS)
1298static void *frag_start(struct seq_file *m, loff_t *pos)
1299{
1300 pg_data_t *pgdat;
1301 loff_t node = *pos;
1302
1303 for (pgdat = first_online_pgdat();
1304 pgdat && node;
1305 pgdat = next_online_pgdat(pgdat))
1306 --node;
1307
1308 return pgdat;
1309}
1310
1311static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
1312{
1313 pg_data_t *pgdat = (pg_data_t *)arg;
1314
1315 (*pos)++;
1316 return next_online_pgdat(pgdat);
1317}
1318
1319static void frag_stop(struct seq_file *m, void *arg)
1320{
1321}
1322
David Rientjesb2bd8592017-05-03 14:52:59 -07001323/*
1324 * Walk zones in a node and print using a callback.
1325 * If @assert_populated is true, only use callback for zones that are populated.
1326 */
Andrew Morton3c486872015-02-10 14:09:43 -08001327static void walk_zones_in_node(struct seq_file *m, pg_data_t *pgdat,
Vinayak Menon727c0802017-07-10 15:49:17 -07001328 bool assert_populated, bool nolock,
Andrew Morton3c486872015-02-10 14:09:43 -08001329 void (*print)(struct seq_file *m, pg_data_t *, struct zone *))
1330{
1331 struct zone *zone;
1332 struct zone *node_zones = pgdat->node_zones;
1333 unsigned long flags;
1334
1335 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
David Rientjesb2bd8592017-05-03 14:52:59 -07001336 if (assert_populated && !populated_zone(zone))
Andrew Morton3c486872015-02-10 14:09:43 -08001337 continue;
1338
Vinayak Menon727c0802017-07-10 15:49:17 -07001339 if (!nolock)
1340 spin_lock_irqsave(&zone->lock, flags);
Andrew Morton3c486872015-02-10 14:09:43 -08001341 print(m, pgdat, zone);
Vinayak Menon727c0802017-07-10 15:49:17 -07001342 if (!nolock)
1343 spin_unlock_irqrestore(&zone->lock, flags);
Andrew Morton3c486872015-02-10 14:09:43 -08001344 }
1345}
1346#endif
1347
Mel Gormand7a57522010-05-24 14:32:25 -07001348#ifdef CONFIG_PROC_FS
Mel Gorman467c9962007-10-16 01:26:02 -07001349static void frag_show_print(struct seq_file *m, pg_data_t *pgdat,
1350 struct zone *zone)
1351{
1352 int order;
1353
1354 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
1355 for (order = 0; order < MAX_ORDER; ++order)
1356 seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
1357 seq_putc(m, '\n');
1358}
1359
1360/*
1361 * This walks the free areas for each zone.
1362 */
1363static int frag_show(struct seq_file *m, void *arg)
1364{
1365 pg_data_t *pgdat = (pg_data_t *)arg;
Vinayak Menon727c0802017-07-10 15:49:17 -07001366 walk_zones_in_node(m, pgdat, true, false, frag_show_print);
Mel Gorman467c9962007-10-16 01:26:02 -07001367 return 0;
1368}
1369
1370static void pagetypeinfo_showfree_print(struct seq_file *m,
1371 pg_data_t *pgdat, struct zone *zone)
1372{
1373 int order, mtype;
1374
1375 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++) {
1376 seq_printf(m, "Node %4d, zone %8s, type %12s ",
1377 pgdat->node_id,
1378 zone->name,
1379 migratetype_names[mtype]);
1380 for (order = 0; order < MAX_ORDER; ++order) {
1381 unsigned long freecount = 0;
1382 struct free_area *area;
1383 struct list_head *curr;
1384
1385 area = &(zone->free_area[order]);
1386
1387 list_for_each(curr, &area->free_list[mtype])
1388 freecount++;
1389 seq_printf(m, "%6lu ", freecount);
1390 }
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001391 seq_putc(m, '\n');
1392 }
Mel Gorman467c9962007-10-16 01:26:02 -07001393}
1394
1395/* Print out the free pages at each order for each migatetype */
1396static int pagetypeinfo_showfree(struct seq_file *m, void *arg)
1397{
1398 int order;
1399 pg_data_t *pgdat = (pg_data_t *)arg;
1400
1401 /* Print header */
1402 seq_printf(m, "%-43s ", "Free pages count per migrate type at order");
1403 for (order = 0; order < MAX_ORDER; ++order)
1404 seq_printf(m, "%6d ", order);
1405 seq_putc(m, '\n');
1406
Vinayak Menon727c0802017-07-10 15:49:17 -07001407 walk_zones_in_node(m, pgdat, true, false, pagetypeinfo_showfree_print);
Mel Gorman467c9962007-10-16 01:26:02 -07001408
1409 return 0;
1410}
1411
1412static void pagetypeinfo_showblockcount_print(struct seq_file *m,
1413 pg_data_t *pgdat, struct zone *zone)
1414{
1415 int mtype;
1416 unsigned long pfn;
1417 unsigned long start_pfn = zone->zone_start_pfn;
Cody P Schafer108bcc92013-02-22 16:35:23 -08001418 unsigned long end_pfn = zone_end_pfn(zone);
Mel Gorman467c9962007-10-16 01:26:02 -07001419 unsigned long count[MIGRATE_TYPES] = { 0, };
1420
1421 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
1422 struct page *page;
1423
Michal Hockod336e942017-07-06 15:38:07 -07001424 page = pfn_to_online_page(pfn);
1425 if (!page)
Mel Gorman467c9962007-10-16 01:26:02 -07001426 continue;
1427
Mel Gormaneb335752009-05-13 17:34:48 +01001428 /* Watch for unexpected holes punched in the memmap */
1429 if (!memmap_valid_within(pfn, page, zone))
Mel Gormane80d6a22008-08-14 11:10:14 +01001430 continue;
Mel Gormaneb335752009-05-13 17:34:48 +01001431
Joonsoo Kima91c43c2016-05-19 17:12:10 -07001432 if (page_zone(page) != zone)
1433 continue;
1434
Mel Gorman467c9962007-10-16 01:26:02 -07001435 mtype = get_pageblock_migratetype(page);
1436
Mel Gormane80d6a22008-08-14 11:10:14 +01001437 if (mtype < MIGRATE_TYPES)
1438 count[mtype]++;
Mel Gorman467c9962007-10-16 01:26:02 -07001439 }
1440
1441 /* Print counts */
1442 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
1443 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
1444 seq_printf(m, "%12lu ", count[mtype]);
1445 seq_putc(m, '\n');
1446}
1447
SeongJae Parkf113e642017-09-06 16:24:23 -07001448/* Print out the number of pageblocks for each migratetype */
Mel Gorman467c9962007-10-16 01:26:02 -07001449static int pagetypeinfo_showblockcount(struct seq_file *m, void *arg)
1450{
1451 int mtype;
1452 pg_data_t *pgdat = (pg_data_t *)arg;
1453
1454 seq_printf(m, "\n%-23s", "Number of blocks type ");
1455 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
1456 seq_printf(m, "%12s ", migratetype_names[mtype]);
1457 seq_putc(m, '\n');
Vinayak Menon727c0802017-07-10 15:49:17 -07001458 walk_zones_in_node(m, pgdat, true, false,
1459 pagetypeinfo_showblockcount_print);
Mel Gorman467c9962007-10-16 01:26:02 -07001460
1461 return 0;
1462}
1463
Joonsoo Kim48c96a32014-12-12 16:56:01 -08001464/*
1465 * Print out the number of pageblocks for each migratetype that contain pages
1466 * of other types. This gives an indication of how well fallbacks are being
1467 * contained by rmqueue_fallback(). It requires information from PAGE_OWNER
1468 * to determine what is going on
1469 */
1470static void pagetypeinfo_showmixedcount(struct seq_file *m, pg_data_t *pgdat)
1471{
1472#ifdef CONFIG_PAGE_OWNER
1473 int mtype;
1474
Vlastimil Babka7dd80b82016-03-15 14:56:12 -07001475 if (!static_branch_unlikely(&page_owner_inited))
Joonsoo Kim48c96a32014-12-12 16:56:01 -08001476 return;
1477
1478 drain_all_pages(NULL);
1479
1480 seq_printf(m, "\n%-23s", "Number of mixed blocks ");
1481 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
1482 seq_printf(m, "%12s ", migratetype_names[mtype]);
1483 seq_putc(m, '\n');
1484
Vinayak Menon727c0802017-07-10 15:49:17 -07001485 walk_zones_in_node(m, pgdat, true, true,
1486 pagetypeinfo_showmixedcount_print);
Joonsoo Kim48c96a32014-12-12 16:56:01 -08001487#endif /* CONFIG_PAGE_OWNER */
1488}
1489
Mel Gorman467c9962007-10-16 01:26:02 -07001490/*
1491 * This prints out statistics in relation to grouping pages by mobility.
1492 * It is expensive to collect so do not constantly read the file.
1493 */
1494static int pagetypeinfo_show(struct seq_file *m, void *arg)
1495{
1496 pg_data_t *pgdat = (pg_data_t *)arg;
1497
KOSAKI Motohiro41b25a32008-04-30 00:52:13 -07001498 /* check memoryless node */
Lai Jiangshana47b53c2012-12-12 13:51:37 -08001499 if (!node_state(pgdat->node_id, N_MEMORY))
KOSAKI Motohiro41b25a32008-04-30 00:52:13 -07001500 return 0;
1501
Mel Gorman467c9962007-10-16 01:26:02 -07001502 seq_printf(m, "Page block order: %d\n", pageblock_order);
1503 seq_printf(m, "Pages per block: %lu\n", pageblock_nr_pages);
1504 seq_putc(m, '\n');
1505 pagetypeinfo_showfree(m, pgdat);
1506 pagetypeinfo_showblockcount(m, pgdat);
Joonsoo Kim48c96a32014-12-12 16:56:01 -08001507 pagetypeinfo_showmixedcount(m, pgdat);
Mel Gorman467c9962007-10-16 01:26:02 -07001508
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001509 return 0;
1510}
1511
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +04001512static const struct seq_operations fragmentation_op = {
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001513 .start = frag_start,
1514 .next = frag_next,
1515 .stop = frag_stop,
1516 .show = frag_show,
1517};
1518
Alexey Dobriyan74e2e8e2008-10-06 04:15:36 +04001519static const struct seq_operations pagetypeinfo_op = {
Mel Gorman467c9962007-10-16 01:26:02 -07001520 .start = frag_start,
1521 .next = frag_next,
1522 .stop = frag_stop,
1523 .show = pagetypeinfo_show,
1524};
1525
Mel Gormane2ecc8a2016-07-28 15:47:02 -07001526static bool is_zone_first_populated(pg_data_t *pgdat, struct zone *zone)
1527{
1528 int zid;
1529
1530 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
1531 struct zone *compare = &pgdat->node_zones[zid];
1532
1533 if (populated_zone(compare))
1534 return zone == compare;
1535 }
1536
Mel Gormane2ecc8a2016-07-28 15:47:02 -07001537 return false;
1538}
1539
Mel Gorman467c9962007-10-16 01:26:02 -07001540static void zoneinfo_show_print(struct seq_file *m, pg_data_t *pgdat,
1541 struct zone *zone)
1542{
1543 int i;
1544 seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
Mel Gormane2ecc8a2016-07-28 15:47:02 -07001545 if (is_zone_first_populated(pgdat, zone)) {
1546 seq_printf(m, "\n per-node stats");
1547 for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) {
1548 seq_printf(m, "\n %-12s %lu",
Kemi Wang3a321d22017-09-08 16:12:48 -07001549 vmstat_text[i + NR_VM_ZONE_STAT_ITEMS +
1550 NR_VM_NUMA_STAT_ITEMS],
Mel Gormane2ecc8a2016-07-28 15:47:02 -07001551 node_page_state(pgdat, i));
1552 }
1553 }
Mel Gorman467c9962007-10-16 01:26:02 -07001554 seq_printf(m,
1555 "\n pages free %lu"
1556 "\n min %lu"
1557 "\n low %lu"
1558 "\n high %lu"
Mel Gorman467c9962007-10-16 01:26:02 -07001559 "\n spanned %lu"
Jiang Liu9feedc92012-12-12 13:52:12 -08001560 "\n present %lu"
1561 "\n managed %lu",
Mel Gorman88f5acf2011-01-13 15:45:41 -08001562 zone_page_state(zone, NR_FREE_PAGES),
Mel Gorman41858962009-06-16 15:32:12 -07001563 min_wmark_pages(zone),
1564 low_wmark_pages(zone),
1565 high_wmark_pages(zone),
Mel Gorman467c9962007-10-16 01:26:02 -07001566 zone->spanned_pages,
Jiang Liu9feedc92012-12-12 13:52:12 -08001567 zone->present_pages,
1568 zone->managed_pages);
Mel Gorman467c9962007-10-16 01:26:02 -07001569
Mel Gorman467c9962007-10-16 01:26:02 -07001570 seq_printf(m,
Mel Gorman3484b2d2014-08-06 16:07:14 -07001571 "\n protection: (%ld",
Mel Gorman467c9962007-10-16 01:26:02 -07001572 zone->lowmem_reserve[0]);
1573 for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
Mel Gorman3484b2d2014-08-06 16:07:14 -07001574 seq_printf(m, ", %ld", zone->lowmem_reserve[i]);
David Rientjes7dfb8bf2017-05-03 14:53:02 -07001575 seq_putc(m, ')');
1576
1577 /* If unpopulated, no other information is useful */
1578 if (!populated_zone(zone)) {
1579 seq_putc(m, '\n');
1580 return;
1581 }
1582
1583 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
1584 seq_printf(m, "\n %-12s %lu", vmstat_text[i],
1585 zone_page_state(zone, i));
1586
Kemi Wang3a321d22017-09-08 16:12:48 -07001587#ifdef CONFIG_NUMA
1588 for (i = 0; i < NR_VM_NUMA_STAT_ITEMS; i++)
1589 seq_printf(m, "\n %-12s %lu",
1590 vmstat_text[i + NR_VM_ZONE_STAT_ITEMS],
Kemi Wang63803222017-09-08 16:12:55 -07001591 zone_numa_state_snapshot(zone, i));
Kemi Wang3a321d22017-09-08 16:12:48 -07001592#endif
1593
David Rientjes7dfb8bf2017-05-03 14:53:02 -07001594 seq_printf(m, "\n pagesets");
Mel Gorman467c9962007-10-16 01:26:02 -07001595 for_each_online_cpu(i) {
1596 struct per_cpu_pageset *pageset;
Mel Gorman467c9962007-10-16 01:26:02 -07001597
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001598 pageset = per_cpu_ptr(zone->pageset, i);
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001599 seq_printf(m,
1600 "\n cpu: %i"
1601 "\n count: %i"
1602 "\n high: %i"
1603 "\n batch: %i",
1604 i,
1605 pageset->pcp.count,
1606 pageset->pcp.high,
1607 pageset->pcp.batch);
Mel Gorman467c9962007-10-16 01:26:02 -07001608#ifdef CONFIG_SMP
1609 seq_printf(m, "\n vm stats threshold: %d",
1610 pageset->stat_threshold);
1611#endif
1612 }
1613 seq_printf(m,
Mel Gorman599d0c92016-07-28 15:45:31 -07001614 "\n node_unreclaimable: %u"
Andrey Ryabinin3a50d142017-11-15 17:34:15 -08001615 "\n start_pfn: %lu",
Johannes Weinerc73322d2017-05-03 14:51:51 -07001616 pgdat->kswapd_failures >= MAX_RECLAIM_RETRIES,
Andrey Ryabinin3a50d142017-11-15 17:34:15 -08001617 zone->zone_start_pfn);
Mel Gorman467c9962007-10-16 01:26:02 -07001618 seq_putc(m, '\n');
1619}
1620
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001621/*
David Rientjesb2bd8592017-05-03 14:52:59 -07001622 * Output information about zones in @pgdat. All zones are printed regardless
1623 * of whether they are populated or not: lowmem_reserve_ratio operates on the
1624 * set of all zones and userspace would not be aware of such zones if they are
1625 * suppressed here (zoneinfo displays the effect of lowmem_reserve_ratio).
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001626 */
1627static int zoneinfo_show(struct seq_file *m, void *arg)
1628{
Mel Gorman467c9962007-10-16 01:26:02 -07001629 pg_data_t *pgdat = (pg_data_t *)arg;
Vinayak Menon727c0802017-07-10 15:49:17 -07001630 walk_zones_in_node(m, pgdat, false, false, zoneinfo_show_print);
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001631 return 0;
1632}
1633
Alexey Dobriyan5c9fe622008-10-06 04:19:42 +04001634static const struct seq_operations zoneinfo_op = {
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001635 .start = frag_start, /* iterate over all zones. The same as in
1636 * fragmentation. */
1637 .next = frag_next,
1638 .stop = frag_stop,
1639 .show = zoneinfo_show,
1640};
1641
Michael Rubin79da8262010-10-26 14:21:36 -07001642enum writeback_stat_item {
1643 NR_DIRTY_THRESHOLD,
1644 NR_DIRTY_BG_THRESHOLD,
1645 NR_VM_WRITEBACK_STAT_ITEMS,
1646};
1647
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001648static void *vmstat_start(struct seq_file *m, loff_t *pos)
1649{
Christoph Lameter2244b952006-06-30 01:55:33 -07001650 unsigned long *v;
Michael Rubin79da8262010-10-26 14:21:36 -07001651 int i, stat_items_size;
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001652
1653 if (*pos >= ARRAY_SIZE(vmstat_text))
1654 return NULL;
Michael Rubin79da8262010-10-26 14:21:36 -07001655 stat_items_size = NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long) +
Kemi Wang3a321d22017-09-08 16:12:48 -07001656 NR_VM_NUMA_STAT_ITEMS * sizeof(unsigned long) +
Mel Gorman75ef7182016-07-28 15:45:24 -07001657 NR_VM_NODE_STAT_ITEMS * sizeof(unsigned long) +
Michael Rubin79da8262010-10-26 14:21:36 -07001658 NR_VM_WRITEBACK_STAT_ITEMS * sizeof(unsigned long);
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001659
Christoph Lameterf8891e52006-06-30 01:55:45 -07001660#ifdef CONFIG_VM_EVENT_COUNTERS
Michael Rubin79da8262010-10-26 14:21:36 -07001661 stat_items_size += sizeof(struct vm_event_state);
Christoph Lameterf8891e52006-06-30 01:55:45 -07001662#endif
Michael Rubin79da8262010-10-26 14:21:36 -07001663
1664 v = kmalloc(stat_items_size, GFP_KERNEL);
Christoph Lameter2244b952006-06-30 01:55:33 -07001665 m->private = v;
1666 if (!v)
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001667 return ERR_PTR(-ENOMEM);
Christoph Lameter2244b952006-06-30 01:55:33 -07001668 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
Michal Hockoc41f0122017-09-06 16:23:36 -07001669 v[i] = global_zone_page_state(i);
Michael Rubin79da8262010-10-26 14:21:36 -07001670 v += NR_VM_ZONE_STAT_ITEMS;
1671
Kemi Wang3a321d22017-09-08 16:12:48 -07001672#ifdef CONFIG_NUMA
1673 for (i = 0; i < NR_VM_NUMA_STAT_ITEMS; i++)
1674 v[i] = global_numa_state(i);
1675 v += NR_VM_NUMA_STAT_ITEMS;
1676#endif
1677
Mel Gorman75ef7182016-07-28 15:45:24 -07001678 for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++)
1679 v[i] = global_node_page_state(i);
1680 v += NR_VM_NODE_STAT_ITEMS;
1681
Michael Rubin79da8262010-10-26 14:21:36 -07001682 global_dirty_limits(v + NR_DIRTY_BG_THRESHOLD,
1683 v + NR_DIRTY_THRESHOLD);
1684 v += NR_VM_WRITEBACK_STAT_ITEMS;
1685
Christoph Lameterf8891e52006-06-30 01:55:45 -07001686#ifdef CONFIG_VM_EVENT_COUNTERS
Michael Rubin79da8262010-10-26 14:21:36 -07001687 all_vm_events(v);
1688 v[PGPGIN] /= 2; /* sectors -> kbytes */
1689 v[PGPGOUT] /= 2;
Christoph Lameterf8891e52006-06-30 01:55:45 -07001690#endif
Wu Fengguangff8b16d2010-11-04 01:56:49 +08001691 return (unsigned long *)m->private + *pos;
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001692}
1693
1694static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
1695{
1696 (*pos)++;
1697 if (*pos >= ARRAY_SIZE(vmstat_text))
1698 return NULL;
1699 return (unsigned long *)m->private + *pos;
1700}
1701
1702static int vmstat_show(struct seq_file *m, void *arg)
1703{
1704 unsigned long *l = arg;
1705 unsigned long off = l - (unsigned long *)m->private;
Alexey Dobriyan68ba0322016-10-07 17:02:14 -07001706
Roman Gushchin7aaf7722018-05-11 16:01:53 -07001707 /* Skip hidden vmstat items. */
1708 if (*vmstat_text[off] == '\0')
1709 return 0;
1710
Alexey Dobriyan68ba0322016-10-07 17:02:14 -07001711 seq_puts(m, vmstat_text[off]);
Joe Perches75ba1d02016-10-07 17:02:20 -07001712 seq_put_decimal_ull(m, " ", *l);
Alexey Dobriyan68ba0322016-10-07 17:02:14 -07001713 seq_putc(m, '\n');
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001714 return 0;
1715}
1716
1717static void vmstat_stop(struct seq_file *m, void *arg)
1718{
1719 kfree(m->private);
1720 m->private = NULL;
1721}
1722
Alexey Dobriyanb6aa44a2008-10-06 04:17:48 +04001723static const struct seq_operations vmstat_op = {
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001724 .start = vmstat_start,
1725 .next = vmstat_next,
1726 .stop = vmstat_stop,
1727 .show = vmstat_show,
1728};
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001729#endif /* CONFIG_PROC_FS */
1730
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001731#ifdef CONFIG_SMP
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001732static DEFINE_PER_CPU(struct delayed_work, vmstat_work);
Christoph Lameter77461ab2007-05-09 02:35:13 -07001733int sysctl_stat_interval __read_mostly = HZ;
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001734
Hugh Dickins52b6f462016-05-19 17:12:50 -07001735#ifdef CONFIG_PROC_FS
1736static void refresh_vm_stats(struct work_struct *work)
1737{
1738 refresh_cpu_vm_stats(true);
1739}
1740
1741int vmstat_refresh(struct ctl_table *table, int write,
1742 void __user *buffer, size_t *lenp, loff_t *ppos)
1743{
1744 long val;
1745 int err;
1746 int i;
1747
1748 /*
1749 * The regular update, every sysctl_stat_interval, may come later
1750 * than expected: leaving a significant amount in per_cpu buckets.
1751 * This is particularly misleading when checking a quantity of HUGE
1752 * pages, immediately after running a test. /proc/sys/vm/stat_refresh,
1753 * which can equally be echo'ed to or cat'ted from (by root),
1754 * can be used to update the stats just before reading them.
1755 *
Michal Hockoc41f0122017-09-06 16:23:36 -07001756 * Oh, and since global_zone_page_state() etc. are so careful to hide
Hugh Dickins52b6f462016-05-19 17:12:50 -07001757 * transiently negative values, report an error here if any of
1758 * the stats is negative, so we know to go looking for imbalance.
1759 */
1760 err = schedule_on_each_cpu(refresh_vm_stats);
1761 if (err)
1762 return err;
1763 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) {
Mel Gorman75ef7182016-07-28 15:45:24 -07001764 val = atomic_long_read(&vm_zone_stat[i]);
Hugh Dickins52b6f462016-05-19 17:12:50 -07001765 if (val < 0) {
Johannes Weinerc822f622017-05-03 14:52:10 -07001766 pr_warn("%s: %s %ld\n",
1767 __func__, vmstat_text[i], val);
1768 err = -EINVAL;
Hugh Dickins52b6f462016-05-19 17:12:50 -07001769 }
1770 }
Kemi Wang3a321d22017-09-08 16:12:48 -07001771#ifdef CONFIG_NUMA
1772 for (i = 0; i < NR_VM_NUMA_STAT_ITEMS; i++) {
1773 val = atomic_long_read(&vm_numa_stat[i]);
1774 if (val < 0) {
1775 pr_warn("%s: %s %ld\n",
1776 __func__, vmstat_text[i + NR_VM_ZONE_STAT_ITEMS], val);
1777 err = -EINVAL;
1778 }
1779 }
1780#endif
Hugh Dickins52b6f462016-05-19 17:12:50 -07001781 if (err)
1782 return err;
1783 if (write)
1784 *ppos += *lenp;
1785 else
1786 *lenp = 0;
1787 return 0;
1788}
1789#endif /* CONFIG_PROC_FS */
1790
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001791static void vmstat_update(struct work_struct *w)
1792{
Christoph Lameter0eb77e92016-01-14 15:21:40 -08001793 if (refresh_cpu_vm_stats(true)) {
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001794 /*
1795 * Counters were updated so we expect more updates
1796 * to occur in the future. Keep on running the
1797 * update worker thread.
1798 */
Steven J. Hillc7f26cc2018-03-28 16:01:09 -07001799 preempt_disable();
Michal Hockoce612872017-04-07 16:05:05 -07001800 queue_delayed_work_on(smp_processor_id(), mm_percpu_wq,
Michal Hockof01f17d2016-02-05 15:36:24 -08001801 this_cpu_ptr(&vmstat_work),
1802 round_jiffies_relative(sysctl_stat_interval));
Steven J. Hillc7f26cc2018-03-28 16:01:09 -07001803 preempt_enable();
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001804 }
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001805}
1806
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001807/*
Christoph Lameter0eb77e92016-01-14 15:21:40 -08001808 * Switch off vmstat processing and then fold all the remaining differentials
1809 * until the diffs stay at zero. The function is used by NOHZ and can only be
1810 * invoked when tick processing is not active.
1811 */
Christoph Lameter0eb77e92016-01-14 15:21:40 -08001812/*
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001813 * Check if the diffs for a certain cpu indicate that
1814 * an update is needed.
1815 */
1816static bool need_update(int cpu)
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001817{
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001818 struct zone *zone;
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001819
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001820 for_each_populated_zone(zone) {
1821 struct per_cpu_pageset *p = per_cpu_ptr(zone->pageset, cpu);
1822
1823 BUILD_BUG_ON(sizeof(p->vm_stat_diff[0]) != 1);
Kemi Wang3a321d22017-09-08 16:12:48 -07001824#ifdef CONFIG_NUMA
Kemi Wang1d90ca82017-09-08 16:12:52 -07001825 BUILD_BUG_ON(sizeof(p->vm_numa_stat_diff[0]) != 2);
Kemi Wang3a321d22017-09-08 16:12:48 -07001826#endif
Kemi Wang63803222017-09-08 16:12:55 -07001827
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001828 /*
1829 * The fast way of checking if there are any vmstat diffs.
1830 * This works because the diffs are byte sized items.
1831 */
1832 if (memchr_inv(p->vm_stat_diff, 0, NR_VM_ZONE_STAT_ITEMS))
1833 return true;
Kemi Wang3a321d22017-09-08 16:12:48 -07001834#ifdef CONFIG_NUMA
1835 if (memchr_inv(p->vm_numa_stat_diff, 0, NR_VM_NUMA_STAT_ITEMS))
1836 return true;
1837#endif
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001838 }
1839 return false;
1840}
1841
Christoph Lameter7b8da4c2016-05-20 16:58:21 -07001842/*
1843 * Switch off vmstat processing and then fold all the remaining differentials
1844 * until the diffs stay at zero. The function is used by NOHZ and can only be
1845 * invoked when tick processing is not active.
1846 */
Michal Hockof01f17d2016-02-05 15:36:24 -08001847void quiet_vmstat(void)
1848{
1849 if (system_state != SYSTEM_RUNNING)
1850 return;
1851
Christoph Lameter7b8da4c2016-05-20 16:58:21 -07001852 if (!delayed_work_pending(this_cpu_ptr(&vmstat_work)))
Michal Hockof01f17d2016-02-05 15:36:24 -08001853 return;
1854
1855 if (!need_update(smp_processor_id()))
1856 return;
1857
1858 /*
1859 * Just refresh counters and do not care about the pending delayed
1860 * vmstat_update. It doesn't fire that often to matter and canceling
1861 * it would be too expensive from this path.
1862 * vmstat_shepherd will take care about that for us.
1863 */
1864 refresh_cpu_vm_stats(false);
1865}
1866
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001867/*
1868 * Shepherd worker thread that checks the
1869 * differentials of processors that have their worker
1870 * threads for vm statistics updates disabled because of
1871 * inactivity.
1872 */
1873static void vmstat_shepherd(struct work_struct *w);
1874
Christoph Lameter0eb77e92016-01-14 15:21:40 -08001875static DECLARE_DEFERRABLE_WORK(shepherd, vmstat_shepherd);
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001876
1877static void vmstat_shepherd(struct work_struct *w)
1878{
1879 int cpu;
1880
1881 get_online_cpus();
1882 /* Check processors whose vmstat worker threads have been disabled */
Christoph Lameter7b8da4c2016-05-20 16:58:21 -07001883 for_each_online_cpu(cpu) {
Michal Hockof01f17d2016-02-05 15:36:24 -08001884 struct delayed_work *dw = &per_cpu(vmstat_work, cpu);
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001885
Christoph Lameter7b8da4c2016-05-20 16:58:21 -07001886 if (!delayed_work_pending(dw) && need_update(cpu))
Michal Hockoce612872017-04-07 16:05:05 -07001887 queue_delayed_work_on(cpu, mm_percpu_wq, dw, 0);
Michal Hockof01f17d2016-02-05 15:36:24 -08001888 }
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001889 put_online_cpus();
1890
1891 schedule_delayed_work(&shepherd,
1892 round_jiffies_relative(sysctl_stat_interval));
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001893}
1894
1895static void __init start_shepherd_timer(void)
1896{
1897 int cpu;
1898
1899 for_each_possible_cpu(cpu)
Michal Hockoccde8bd2016-02-05 15:36:27 -08001900 INIT_DEFERRABLE_WORK(per_cpu_ptr(&vmstat_work, cpu),
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001901 vmstat_update);
1902
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001903 schedule_delayed_work(&shepherd,
1904 round_jiffies_relative(sysctl_stat_interval));
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001905}
1906
Tim Chen03e86db2016-10-07 17:00:02 -07001907static void __init init_cpu_node_state(void)
1908{
Sebastian Andrzej Siewior4c501322016-11-29 15:51:14 +01001909 int node;
Tim Chen03e86db2016-10-07 17:00:02 -07001910
Sebastian Andrzej Siewior4c501322016-11-29 15:51:14 +01001911 for_each_online_node(node) {
1912 if (cpumask_weight(cpumask_of_node(node)) > 0)
1913 node_set_state(node, N_CPU);
1914 }
Tim Chen03e86db2016-10-07 17:00:02 -07001915}
1916
Sebastian Andrzej Siewior5438da92016-11-29 15:52:21 +01001917static int vmstat_cpu_online(unsigned int cpu)
1918{
1919 refresh_zone_stat_thresholds();
1920 node_set_state(cpu_to_node(cpu), N_CPU);
1921 return 0;
1922}
1923
1924static int vmstat_cpu_down_prep(unsigned int cpu)
1925{
1926 cancel_delayed_work_sync(&per_cpu(vmstat_work, cpu));
1927 return 0;
1928}
1929
1930static int vmstat_cpu_dead(unsigned int cpu)
Toshi Kani807a1bd2013-11-12 15:08:13 -08001931{
Sebastian Andrzej Siewior4c501322016-11-29 15:51:14 +01001932 const struct cpumask *node_cpus;
Sebastian Andrzej Siewior5438da92016-11-29 15:52:21 +01001933 int node;
Toshi Kani807a1bd2013-11-12 15:08:13 -08001934
Sebastian Andrzej Siewior5438da92016-11-29 15:52:21 +01001935 node = cpu_to_node(cpu);
1936
1937 refresh_zone_stat_thresholds();
Sebastian Andrzej Siewior4c501322016-11-29 15:51:14 +01001938 node_cpus = cpumask_of_node(node);
1939 if (cpumask_weight(node_cpus) > 0)
Sebastian Andrzej Siewior5438da92016-11-29 15:52:21 +01001940 return 0;
Toshi Kani807a1bd2013-11-12 15:08:13 -08001941
1942 node_clear_state(node, N_CPU);
Sebastian Andrzej Siewior5438da92016-11-29 15:52:21 +01001943 return 0;
Toshi Kani807a1bd2013-11-12 15:08:13 -08001944}
1945
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +04001946#endif
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001947
Michal Hockoce612872017-04-07 16:05:05 -07001948struct workqueue_struct *mm_percpu_wq;
1949
Michal Hocko597b7302017-03-31 15:11:47 -07001950void __init init_mm_internals(void)
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001951{
Michal Hockoce612872017-04-07 16:05:05 -07001952 int ret __maybe_unused;
Sebastian Andrzej Siewior5438da92016-11-29 15:52:21 +01001953
Michal Hocko80d136e2017-04-19 09:52:46 +02001954 mm_percpu_wq = alloc_workqueue("mm_percpu_wq", WQ_MEM_RECLAIM, 0);
Michal Hockoce612872017-04-07 16:05:05 -07001955
1956#ifdef CONFIG_SMP
Sebastian Andrzej Siewior5438da92016-11-29 15:52:21 +01001957 ret = cpuhp_setup_state_nocalls(CPUHP_MM_VMSTAT_DEAD, "mm/vmstat:dead",
1958 NULL, vmstat_cpu_dead);
1959 if (ret < 0)
1960 pr_err("vmstat: failed to register 'dead' hotplug state\n");
1961
1962 ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "mm/vmstat:online",
1963 vmstat_cpu_online,
1964 vmstat_cpu_down_prep);
1965 if (ret < 0)
1966 pr_err("vmstat: failed to register 'online' hotplug state\n");
1967
1968 get_online_cpus();
Tim Chen03e86db2016-10-07 17:00:02 -07001969 init_cpu_node_state();
Sebastian Andrzej Siewior5438da92016-11-29 15:52:21 +01001970 put_online_cpus();
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001971
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001972 start_shepherd_timer();
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +04001973#endif
1974#ifdef CONFIG_PROC_FS
Christoph Hellwigfddda2b2018-04-13 19:44:18 +02001975 proc_create_seq("buddyinfo", 0444, NULL, &fragmentation_op);
1976 proc_create_seq("pagetypeinfo", 0444, NULL, &pagetypeinfo_op);
1977 proc_create_seq("vmstat", 0444, NULL, &vmstat_op);
1978 proc_create_seq("zoneinfo", 0444, NULL, &zoneinfo_op);
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +04001979#endif
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001980}
Mel Gormand7a57522010-05-24 14:32:25 -07001981
1982#if defined(CONFIG_DEBUG_FS) && defined(CONFIG_COMPACTION)
Mel Gormand7a57522010-05-24 14:32:25 -07001983
1984/*
1985 * Return an index indicating how much of the available free memory is
1986 * unusable for an allocation of the requested size.
1987 */
1988static int unusable_free_index(unsigned int order,
1989 struct contig_page_info *info)
1990{
1991 /* No free memory is interpreted as all free memory is unusable */
1992 if (info->free_pages == 0)
1993 return 1000;
1994
1995 /*
1996 * Index should be a value between 0 and 1. Return a value to 3
1997 * decimal places.
1998 *
1999 * 0 => no fragmentation
2000 * 1 => high fragmentation
2001 */
2002 return div_u64((info->free_pages - (info->free_blocks_suitable << order)) * 1000ULL, info->free_pages);
2003
2004}
2005
2006static void unusable_show_print(struct seq_file *m,
2007 pg_data_t *pgdat, struct zone *zone)
2008{
2009 unsigned int order;
2010 int index;
2011 struct contig_page_info info;
2012
2013 seq_printf(m, "Node %d, zone %8s ",
2014 pgdat->node_id,
2015 zone->name);
2016 for (order = 0; order < MAX_ORDER; ++order) {
2017 fill_contig_page_info(zone, order, &info);
2018 index = unusable_free_index(order, &info);
2019 seq_printf(m, "%d.%03d ", index / 1000, index % 1000);
2020 }
2021
2022 seq_putc(m, '\n');
2023}
2024
2025/*
2026 * Display unusable free space index
2027 *
2028 * The unusable free space index measures how much of the available free
2029 * memory cannot be used to satisfy an allocation of a given size and is a
2030 * value between 0 and 1. The higher the value, the more of free memory is
2031 * unusable and by implication, the worse the external fragmentation is. This
2032 * can be expressed as a percentage by multiplying by 100.
2033 */
2034static int unusable_show(struct seq_file *m, void *arg)
2035{
2036 pg_data_t *pgdat = (pg_data_t *)arg;
2037
2038 /* check memoryless node */
Lai Jiangshana47b53c2012-12-12 13:51:37 -08002039 if (!node_state(pgdat->node_id, N_MEMORY))
Mel Gormand7a57522010-05-24 14:32:25 -07002040 return 0;
2041
Vinayak Menon727c0802017-07-10 15:49:17 -07002042 walk_zones_in_node(m, pgdat, true, false, unusable_show_print);
Mel Gormand7a57522010-05-24 14:32:25 -07002043
2044 return 0;
2045}
2046
2047static const struct seq_operations unusable_op = {
2048 .start = frag_start,
2049 .next = frag_next,
2050 .stop = frag_stop,
2051 .show = unusable_show,
2052};
2053
2054static int unusable_open(struct inode *inode, struct file *file)
2055{
2056 return seq_open(file, &unusable_op);
2057}
2058
2059static const struct file_operations unusable_file_ops = {
2060 .open = unusable_open,
2061 .read = seq_read,
2062 .llseek = seq_lseek,
2063 .release = seq_release,
2064};
2065
Mel Gormanf1a5ab12010-05-24 14:32:26 -07002066static void extfrag_show_print(struct seq_file *m,
2067 pg_data_t *pgdat, struct zone *zone)
2068{
2069 unsigned int order;
2070 int index;
2071
2072 /* Alloc on stack as interrupts are disabled for zone walk */
2073 struct contig_page_info info;
2074
2075 seq_printf(m, "Node %d, zone %8s ",
2076 pgdat->node_id,
2077 zone->name);
2078 for (order = 0; order < MAX_ORDER; ++order) {
2079 fill_contig_page_info(zone, order, &info);
Mel Gorman56de7262010-05-24 14:32:30 -07002080 index = __fragmentation_index(order, &info);
Mel Gormanf1a5ab12010-05-24 14:32:26 -07002081 seq_printf(m, "%d.%03d ", index / 1000, index % 1000);
2082 }
2083
2084 seq_putc(m, '\n');
2085}
2086
2087/*
2088 * Display fragmentation index for orders that allocations would fail for
2089 */
2090static int extfrag_show(struct seq_file *m, void *arg)
2091{
2092 pg_data_t *pgdat = (pg_data_t *)arg;
2093
Vinayak Menon727c0802017-07-10 15:49:17 -07002094 walk_zones_in_node(m, pgdat, true, false, extfrag_show_print);
Mel Gormanf1a5ab12010-05-24 14:32:26 -07002095
2096 return 0;
2097}
2098
2099static const struct seq_operations extfrag_op = {
2100 .start = frag_start,
2101 .next = frag_next,
2102 .stop = frag_stop,
2103 .show = extfrag_show,
2104};
2105
2106static int extfrag_open(struct inode *inode, struct file *file)
2107{
2108 return seq_open(file, &extfrag_op);
2109}
2110
2111static const struct file_operations extfrag_file_ops = {
2112 .open = extfrag_open,
2113 .read = seq_read,
2114 .llseek = seq_lseek,
2115 .release = seq_release,
2116};
2117
Mel Gormand7a57522010-05-24 14:32:25 -07002118static int __init extfrag_debug_init(void)
2119{
Sasikantha babubde8bd82012-05-29 15:06:22 -07002120 struct dentry *extfrag_debug_root;
2121
Mel Gormand7a57522010-05-24 14:32:25 -07002122 extfrag_debug_root = debugfs_create_dir("extfrag", NULL);
2123 if (!extfrag_debug_root)
2124 return -ENOMEM;
2125
2126 if (!debugfs_create_file("unusable_index", 0444,
2127 extfrag_debug_root, NULL, &unusable_file_ops))
Sasikantha babubde8bd82012-05-29 15:06:22 -07002128 goto fail;
Mel Gormand7a57522010-05-24 14:32:25 -07002129
Mel Gormanf1a5ab12010-05-24 14:32:26 -07002130 if (!debugfs_create_file("extfrag_index", 0444,
2131 extfrag_debug_root, NULL, &extfrag_file_ops))
Sasikantha babubde8bd82012-05-29 15:06:22 -07002132 goto fail;
Mel Gormanf1a5ab12010-05-24 14:32:26 -07002133
Mel Gormand7a57522010-05-24 14:32:25 -07002134 return 0;
Sasikantha babubde8bd82012-05-29 15:06:22 -07002135fail:
2136 debugfs_remove_recursive(extfrag_debug_root);
2137 return -ENOMEM;
Mel Gormand7a57522010-05-24 14:32:25 -07002138}
2139
2140module_init(extfrag_debug_init);
2141#endif