blob: 00e81e99c6ee24e419446b8c476faea0b8ef75bd [file] [log] [blame]
Thomas Gleixner457c8992019-05-19 13:08:55 +01001// SPDX-License-Identifier: GPL-2.0-only
Christoph Lameterf6ac2352006-06-30 01:55:32 -07002/*
3 * linux/mm/vmstat.c
4 *
5 * Manages VM statistics
6 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
Christoph Lameter2244b952006-06-30 01:55:33 -07007 *
8 * zoned VM statistics
9 * Copyright (C) 2006 Silicon Graphics, Inc.,
10 * Christoph Lameter <christoph@lameter.com>
Christoph Lameter7cc36bb2014-10-09 15:29:43 -070011 * Copyright (C) 2008-2014 Christoph Lameter
Christoph Lameterf6ac2352006-06-30 01:55:32 -070012 */
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +040013#include <linux/fs.h>
Christoph Lameterf6ac2352006-06-30 01:55:32 -070014#include <linux/mm.h>
Alexey Dobriyan4e950f62007-07-30 02:36:13 +040015#include <linux/err.h>
Christoph Lameter2244b952006-06-30 01:55:33 -070016#include <linux/module.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090017#include <linux/slab.h>
Christoph Lameterdf9ecab2006-08-31 21:27:35 -070018#include <linux/cpu.h>
Christoph Lameter7cc36bb2014-10-09 15:29:43 -070019#include <linux/cpumask.h>
Adrian Bunkc748e132008-07-23 21:27:03 -070020#include <linux/vmstat.h>
Andrew Morton3c486872015-02-10 14:09:43 -080021#include <linux/proc_fs.h>
22#include <linux/seq_file.h>
23#include <linux/debugfs.h>
Alexey Dobriyane8edc6e2007-05-21 01:22:52 +040024#include <linux/sched.h>
Mel Gormanf1a5ab12010-05-24 14:32:26 -070025#include <linux/math64.h>
Michael Rubin79da8262010-10-26 14:21:36 -070026#include <linux/writeback.h>
Namhyung Kim36deb0b2010-10-26 14:22:04 -070027#include <linux/compaction.h>
Lisa Du6e543d52013-09-11 14:22:36 -070028#include <linux/mm_inline.h>
Joonsoo Kim48c96a32014-12-12 16:56:01 -080029#include <linux/page_owner.h>
Marcelo Tosattibe5e0152023-06-07 17:28:07 -030030#include <linux/sched/isolation.h>
Lisa Du6e543d52013-09-11 14:22:36 -070031
32#include "internal.h"
Christoph Lameterf6ac2352006-06-30 01:55:32 -070033
Kemi Wang45180852017-11-15 17:38:22 -080034#ifdef CONFIG_NUMA
35int sysctl_vm_numa_stat = ENABLE_NUMA_STAT;
36
37/* zero numa counters within a zone */
38static void zero_zone_numa_counters(struct zone *zone)
39{
40 int item, cpu;
41
Mel Gormanf19298b2021-06-28 19:41:44 -070042 for (item = 0; item < NR_VM_NUMA_EVENT_ITEMS; item++) {
43 atomic_long_set(&zone->vm_numa_event[item], 0);
44 for_each_online_cpu(cpu) {
45 per_cpu_ptr(zone->per_cpu_zonestats, cpu)->vm_numa_event[item]
Kemi Wang45180852017-11-15 17:38:22 -080046 = 0;
Mel Gormanf19298b2021-06-28 19:41:44 -070047 }
Kemi Wang45180852017-11-15 17:38:22 -080048 }
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
Mel Gormanf19298b2021-06-28 19:41:44 -070065 for (item = 0; item < NR_VM_NUMA_EVENT_ITEMS; item++)
66 atomic_long_set(&vm_numa_event[item], 0);
Kemi Wang45180852017-11-15 17:38:22 -080067}
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,
Christoph Hellwig32927392020-04-24 08:43:38 +020078 void *buffer, size_t *length, loff_t *ppos)
Kemi Wang45180852017-11-15 17:38:22 -080079{
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{
Sebastian Andrzej Siewior7625ecc2021-08-03 16:16:03 +0200132 cpus_read_lock();
Minchan Kim31f961a2010-08-09 17:18:59 -0700133 sum_vm_events(ret);
Sebastian Andrzej Siewior7625ecc2021-08-03 16:16:03 +0200134 cpus_read_unlock();
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;
163atomic_long_t vm_node_stat[NR_VM_NODE_STAT_ITEMS] __cacheline_aligned_in_smp;
Mel Gormanf19298b2021-06-28 19:41:44 -0700164atomic_long_t vm_numa_event[NR_VM_NUMA_EVENT_ITEMS] __cacheline_aligned_in_smp;
Mel Gorman75ef7182016-07-28 15:45:24 -0700165EXPORT_SYMBOL(vm_zone_stat);
166EXPORT_SYMBOL(vm_node_stat);
Christoph Lameter2244b952006-06-30 01:55:33 -0700167
Geert Uytterhoevenebeac3e2021-11-05 13:40:28 -0700168#ifdef CONFIG_NUMA
169static void fold_vm_zone_numa_events(struct zone *zone)
170{
171 unsigned long zone_numa_events[NR_VM_NUMA_EVENT_ITEMS] = { 0, };
172 int cpu;
173 enum numa_stat_item item;
174
175 for_each_online_cpu(cpu) {
176 struct per_cpu_zonestat *pzstats;
177
178 pzstats = per_cpu_ptr(zone->per_cpu_zonestats, cpu);
179 for (item = 0; item < NR_VM_NUMA_EVENT_ITEMS; item++)
180 zone_numa_events[item] += xchg(&pzstats->vm_numa_event[item], 0);
181 }
182
183 for (item = 0; item < NR_VM_NUMA_EVENT_ITEMS; item++)
184 zone_numa_event_add(zone_numa_events[item], zone, item);
185}
186
187void fold_vm_numa_events(void)
188{
189 struct zone *zone;
190
191 for_each_populated_zone(zone)
192 fold_vm_zone_numa_events(zone);
193}
194#endif
195
Christoph Lameter2244b952006-06-30 01:55:33 -0700196#ifdef CONFIG_SMP
197
Mel Gormanb44129b2011-01-13 15:45:43 -0800198int calculate_pressure_threshold(struct zone *zone)
Mel Gorman88f5acf2011-01-13 15:45:41 -0800199{
200 int threshold;
201 int watermark_distance;
202
203 /*
204 * As vmstats are not up to date, there is drift between the estimated
205 * and real values. For high thresholds and a high number of CPUs, it
206 * is possible for the min watermark to be breached while the estimated
207 * value looks fine. The pressure threshold is a reduced value such
208 * that even the maximum amount of drift will not accidentally breach
209 * the min watermark
210 */
211 watermark_distance = low_wmark_pages(zone) - min_wmark_pages(zone);
212 threshold = max(1, (int)(watermark_distance / num_online_cpus()));
213
214 /*
215 * Maximum threshold is 125
216 */
217 threshold = min(125, threshold);
218
219 return threshold;
220}
221
Mel Gormanb44129b2011-01-13 15:45:43 -0800222int calculate_normal_threshold(struct zone *zone)
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700223{
224 int threshold;
225 int mem; /* memory in 128 MB units */
226
227 /*
228 * The threshold scales with the number of processors and the amount
229 * of memory per zone. More memory means that we can defer updates for
230 * longer, more processors could lead to more contention.
231 * fls() is used to have a cheap way of logarithmic scaling.
232 *
233 * Some sample thresholds:
234 *
Miaohe Linea15ba12021-09-02 15:01:03 -0700235 * Threshold Processors (fls) Zonesize fls(mem)+1
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700236 * ------------------------------------------------------------------
237 * 8 1 1 0.9-1 GB 4
238 * 16 2 2 0.9-1 GB 4
239 * 20 2 2 1-2 GB 5
240 * 24 2 2 2-4 GB 6
241 * 28 2 2 4-8 GB 7
242 * 32 2 2 8-16 GB 8
243 * 4 2 2 <128M 1
244 * 30 4 3 2-4 GB 5
245 * 48 4 3 8-16 GB 8
246 * 32 8 4 1-2 GB 4
247 * 32 8 4 0.9-1GB 4
248 * 10 16 5 <128M 1
249 * 40 16 5 900M 4
250 * 70 64 7 2-4 GB 5
251 * 84 64 7 4-8 GB 6
252 * 108 512 9 4-8 GB 6
253 * 125 1024 10 8-16 GB 8
254 * 125 1024 10 16-32 GB 9
255 */
256
Arun KS9705bea2018-12-28 00:34:24 -0800257 mem = zone_managed_pages(zone) >> (27 - PAGE_SHIFT);
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700258
259 threshold = 2 * fls(num_online_cpus()) * (1 + fls(mem));
260
261 /*
262 * Maximum threshold is 125
263 */
264 threshold = min(125, threshold);
265
266 return threshold;
267}
Christoph Lameter2244b952006-06-30 01:55:33 -0700268
269/*
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700270 * Refresh the thresholds for each zone.
Christoph Lameter2244b952006-06-30 01:55:33 -0700271 */
KOSAKI Motohiroa6cccdc2011-05-24 17:11:33 -0700272void refresh_zone_stat_thresholds(void)
Christoph Lameter2244b952006-06-30 01:55:33 -0700273{
Mel Gorman75ef7182016-07-28 15:45:24 -0700274 struct pglist_data *pgdat;
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700275 struct zone *zone;
276 int cpu;
277 int threshold;
278
Mel Gorman75ef7182016-07-28 15:45:24 -0700279 /* Zero current pgdat thresholds */
280 for_each_online_pgdat(pgdat) {
281 for_each_online_cpu(cpu) {
282 per_cpu_ptr(pgdat->per_cpu_nodestats, cpu)->stat_threshold = 0;
283 }
284 }
285
KOSAKI Motohiroee99c712009-03-31 15:19:31 -0700286 for_each_populated_zone(zone) {
Mel Gorman75ef7182016-07-28 15:45:24 -0700287 struct pglist_data *pgdat = zone->zone_pgdat;
Christoph Lameteraa454842010-09-09 16:38:17 -0700288 unsigned long max_drift, tolerate_drift;
289
Mel Gormanb44129b2011-01-13 15:45:43 -0800290 threshold = calculate_normal_threshold(zone);
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700291
Mel Gorman75ef7182016-07-28 15:45:24 -0700292 for_each_online_cpu(cpu) {
293 int pgdat_threshold;
294
Mel Gorman28f836b2021-06-28 19:41:38 -0700295 per_cpu_ptr(zone->per_cpu_zonestats, cpu)->stat_threshold
Christoph Lameter99dcc3e2010-01-05 15:34:51 +0900296 = threshold;
Kemi Wang1d90ca82017-09-08 16:12:52 -0700297
Mel Gorman75ef7182016-07-28 15:45:24 -0700298 /* Base nodestat threshold on the largest populated zone. */
299 pgdat_threshold = per_cpu_ptr(pgdat->per_cpu_nodestats, cpu)->stat_threshold;
300 per_cpu_ptr(pgdat->per_cpu_nodestats, cpu)->stat_threshold
301 = max(threshold, pgdat_threshold);
302 }
303
Christoph Lameteraa454842010-09-09 16:38:17 -0700304 /*
305 * Only set percpu_drift_mark if there is a danger that
306 * NR_FREE_PAGES reports the low watermark is ok when in fact
307 * the min watermark could be breached by an allocation
308 */
309 tolerate_drift = low_wmark_pages(zone) - min_wmark_pages(zone);
310 max_drift = num_online_cpus() * threshold;
311 if (max_drift > tolerate_drift)
312 zone->percpu_drift_mark = high_wmark_pages(zone) +
313 max_drift;
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700314 }
Christoph Lameter2244b952006-06-30 01:55:33 -0700315}
316
Mel Gormanb44129b2011-01-13 15:45:43 -0800317void set_pgdat_percpu_threshold(pg_data_t *pgdat,
318 int (*calculate_pressure)(struct zone *))
Mel Gorman88f5acf2011-01-13 15:45:41 -0800319{
320 struct zone *zone;
321 int cpu;
322 int threshold;
323 int i;
324
Mel Gorman88f5acf2011-01-13 15:45:41 -0800325 for (i = 0; i < pgdat->nr_zones; i++) {
326 zone = &pgdat->node_zones[i];
327 if (!zone->percpu_drift_mark)
328 continue;
329
Mel Gormanb44129b2011-01-13 15:45:43 -0800330 threshold = (*calculate_pressure)(zone);
Kemi Wang1d90ca82017-09-08 16:12:52 -0700331 for_each_online_cpu(cpu)
Mel Gorman28f836b2021-06-28 19:41:38 -0700332 per_cpu_ptr(zone->per_cpu_zonestats, cpu)->stat_threshold
Mel Gorman88f5acf2011-01-13 15:45:41 -0800333 = threshold;
334 }
Mel Gorman88f5acf2011-01-13 15:45:41 -0800335}
336
Christoph Lameter2244b952006-06-30 01:55:33 -0700337/*
Jianyu Zhanbea04b02014-06-04 16:09:51 -0700338 * For use when we know that interrupts are disabled,
339 * or when we know that preemption is disabled and that
340 * particular counter cannot be updated from interrupt context.
Christoph Lameter2244b952006-06-30 01:55:33 -0700341 */
342void __mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
Heiko Carstens6cdb18a2015-12-29 14:54:32 -0800343 long delta)
Christoph Lameter2244b952006-06-30 01:55:33 -0700344{
Mel Gorman28f836b2021-06-28 19:41:38 -0700345 struct per_cpu_zonestat __percpu *pcp = zone->per_cpu_zonestats;
Christoph Lameter12938a92010-12-06 11:16:20 -0600346 s8 __percpu *p = pcp->vm_stat_diff + item;
Christoph Lameter2244b952006-06-30 01:55:33 -0700347 long x;
Christoph Lameter12938a92010-12-06 11:16:20 -0600348 long t;
Christoph Lameter2244b952006-06-30 01:55:33 -0700349
Ingo Molnarc68ed792021-09-08 15:17:57 -0700350 /*
351 * Accurate vmstat updates require a RMW. On !PREEMPT_RT kernels,
352 * atomicity is provided by IRQs being disabled -- either explicitly
353 * or via local_lock_irq. On PREEMPT_RT, local_lock_irq only disables
354 * CPU migrations and preemption potentially corrupts a counter so
355 * disable preemption.
356 */
Thomas Gleixner7a025e92022-08-25 18:41:26 +0200357 preempt_disable_nested();
Ingo Molnarc68ed792021-09-08 15:17:57 -0700358
Christoph Lameter12938a92010-12-06 11:16:20 -0600359 x = delta + __this_cpu_read(*p);
Christoph Lameter2244b952006-06-30 01:55:33 -0700360
Christoph Lameter12938a92010-12-06 11:16:20 -0600361 t = __this_cpu_read(pcp->stat_threshold);
362
Miaohe Lin40610072020-10-15 20:07:36 -0700363 if (unlikely(abs(x) > t)) {
Christoph Lameter2244b952006-06-30 01:55:33 -0700364 zone_page_state_add(x, zone, item);
365 x = 0;
366 }
Christoph Lameter12938a92010-12-06 11:16:20 -0600367 __this_cpu_write(*p, x);
Ingo Molnarc68ed792021-09-08 15:17:57 -0700368
Thomas Gleixner7a025e92022-08-25 18:41:26 +0200369 preempt_enable_nested();
Christoph Lameter2244b952006-06-30 01:55:33 -0700370}
371EXPORT_SYMBOL(__mod_zone_page_state);
372
Mel Gorman75ef7182016-07-28 15:45:24 -0700373void __mod_node_page_state(struct pglist_data *pgdat, enum node_stat_item item,
374 long delta)
375{
376 struct per_cpu_nodestat __percpu *pcp = pgdat->per_cpu_nodestats;
377 s8 __percpu *p = pcp->vm_node_stat_diff + item;
378 long x;
379 long t;
380
Roman Gushchinea426c22020-08-06 23:20:35 -0700381 if (vmstat_item_in_bytes(item)) {
Johannes Weiner629484a2021-02-25 17:16:51 -0800382 /*
383 * Only cgroups use subpage accounting right now; at
384 * the global level, these items still change in
385 * multiples of whole pages. Store them as pages
386 * internally to keep the per-cpu counters compact.
387 */
Roman Gushchinea426c22020-08-06 23:20:35 -0700388 VM_WARN_ON_ONCE(delta & (PAGE_SIZE - 1));
389 delta >>= PAGE_SHIFT;
390 }
391
Ingo Molnarc68ed792021-09-08 15:17:57 -0700392 /* See __mod_node_page_state */
Thomas Gleixner7a025e92022-08-25 18:41:26 +0200393 preempt_disable_nested();
Ingo Molnarc68ed792021-09-08 15:17:57 -0700394
Mel Gorman75ef7182016-07-28 15:45:24 -0700395 x = delta + __this_cpu_read(*p);
396
397 t = __this_cpu_read(pcp->stat_threshold);
398
Miaohe Lin40610072020-10-15 20:07:36 -0700399 if (unlikely(abs(x) > t)) {
Mel Gorman75ef7182016-07-28 15:45:24 -0700400 node_page_state_add(x, pgdat, item);
401 x = 0;
402 }
403 __this_cpu_write(*p, x);
Ingo Molnarc68ed792021-09-08 15:17:57 -0700404
Thomas Gleixner7a025e92022-08-25 18:41:26 +0200405 preempt_enable_nested();
Mel Gorman75ef7182016-07-28 15:45:24 -0700406}
407EXPORT_SYMBOL(__mod_node_page_state);
408
Christoph Lameter2244b952006-06-30 01:55:33 -0700409/*
Christoph Lameter2244b952006-06-30 01:55:33 -0700410 * Optimized increment and decrement functions.
411 *
412 * These are only for a single page and therefore can take a struct page *
413 * argument instead of struct zone *. This allows the inclusion of the code
414 * generated for page_zone(page) into the optimized functions.
415 *
416 * No overflow check is necessary and therefore the differential can be
417 * incremented or decremented in place which may allow the compilers to
418 * generate better code.
Christoph Lameter2244b952006-06-30 01:55:33 -0700419 * The increment or decrement is known and therefore one boundary check can
420 * be omitted.
421 *
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700422 * NOTE: These functions are very performance sensitive. Change only
423 * with care.
424 *
Christoph Lameter2244b952006-06-30 01:55:33 -0700425 * Some processors have inc/dec instructions that are atomic vs an interrupt.
426 * However, the code must first determine the differential location in a zone
427 * based on the processor number and then inc/dec the counter. There is no
428 * guarantee without disabling preemption that the processor will not change
429 * in between and therefore the atomicity vs. interrupt cannot be exploited
430 * in a useful way here.
431 */
Christoph Lameterc8785382007-02-10 01:43:01 -0800432void __inc_zone_state(struct zone *zone, enum zone_stat_item item)
Christoph Lameter2244b952006-06-30 01:55:33 -0700433{
Mel Gorman28f836b2021-06-28 19:41:38 -0700434 struct per_cpu_zonestat __percpu *pcp = zone->per_cpu_zonestats;
Christoph Lameter12938a92010-12-06 11:16:20 -0600435 s8 __percpu *p = pcp->vm_stat_diff + item;
436 s8 v, t;
Christoph Lameter2244b952006-06-30 01:55:33 -0700437
Ingo Molnarc68ed792021-09-08 15:17:57 -0700438 /* See __mod_node_page_state */
Thomas Gleixner7a025e92022-08-25 18:41:26 +0200439 preempt_disable_nested();
Ingo Molnarc68ed792021-09-08 15:17:57 -0700440
Christoph Lameter908ee0f2010-12-06 11:40:02 -0600441 v = __this_cpu_inc_return(*p);
Christoph Lameter12938a92010-12-06 11:16:20 -0600442 t = __this_cpu_read(pcp->stat_threshold);
443 if (unlikely(v > t)) {
444 s8 overstep = t >> 1;
Christoph Lameter2244b952006-06-30 01:55:33 -0700445
Christoph Lameter12938a92010-12-06 11:16:20 -0600446 zone_page_state_add(v + overstep, zone, item);
447 __this_cpu_write(*p, -overstep);
Christoph Lameter2244b952006-06-30 01:55:33 -0700448 }
Ingo Molnarc68ed792021-09-08 15:17:57 -0700449
Thomas Gleixner7a025e92022-08-25 18:41:26 +0200450 preempt_enable_nested();
Christoph Lameter2244b952006-06-30 01:55:33 -0700451}
Christoph Lameterca889e62006-06-30 01:55:44 -0700452
Mel Gorman75ef7182016-07-28 15:45:24 -0700453void __inc_node_state(struct pglist_data *pgdat, enum node_stat_item item)
454{
455 struct per_cpu_nodestat __percpu *pcp = pgdat->per_cpu_nodestats;
456 s8 __percpu *p = pcp->vm_node_stat_diff + item;
457 s8 v, t;
458
Roman Gushchinea426c22020-08-06 23:20:35 -0700459 VM_WARN_ON_ONCE(vmstat_item_in_bytes(item));
460
Ingo Molnarc68ed792021-09-08 15:17:57 -0700461 /* See __mod_node_page_state */
Thomas Gleixner7a025e92022-08-25 18:41:26 +0200462 preempt_disable_nested();
Ingo Molnarc68ed792021-09-08 15:17:57 -0700463
Mel Gorman75ef7182016-07-28 15:45:24 -0700464 v = __this_cpu_inc_return(*p);
465 t = __this_cpu_read(pcp->stat_threshold);
466 if (unlikely(v > t)) {
467 s8 overstep = t >> 1;
468
469 node_page_state_add(v + overstep, pgdat, item);
470 __this_cpu_write(*p, -overstep);
471 }
Ingo Molnarc68ed792021-09-08 15:17:57 -0700472
Thomas Gleixner7a025e92022-08-25 18:41:26 +0200473 preempt_enable_nested();
Mel Gorman75ef7182016-07-28 15:45:24 -0700474}
475
Christoph Lameterca889e62006-06-30 01:55:44 -0700476void __inc_zone_page_state(struct page *page, enum zone_stat_item item)
477{
478 __inc_zone_state(page_zone(page), item);
479}
Christoph Lameter2244b952006-06-30 01:55:33 -0700480EXPORT_SYMBOL(__inc_zone_page_state);
481
Mel Gorman75ef7182016-07-28 15:45:24 -0700482void __inc_node_page_state(struct page *page, enum node_stat_item item)
483{
484 __inc_node_state(page_pgdat(page), item);
485}
486EXPORT_SYMBOL(__inc_node_page_state);
487
Christoph Lameterc8785382007-02-10 01:43:01 -0800488void __dec_zone_state(struct zone *zone, enum zone_stat_item item)
Christoph Lameter2244b952006-06-30 01:55:33 -0700489{
Mel Gorman28f836b2021-06-28 19:41:38 -0700490 struct per_cpu_zonestat __percpu *pcp = zone->per_cpu_zonestats;
Christoph Lameter12938a92010-12-06 11:16:20 -0600491 s8 __percpu *p = pcp->vm_stat_diff + item;
492 s8 v, t;
Christoph Lameter2244b952006-06-30 01:55:33 -0700493
Ingo Molnarc68ed792021-09-08 15:17:57 -0700494 /* See __mod_node_page_state */
Thomas Gleixner7a025e92022-08-25 18:41:26 +0200495 preempt_disable_nested();
Ingo Molnarc68ed792021-09-08 15:17:57 -0700496
Christoph Lameter908ee0f2010-12-06 11:40:02 -0600497 v = __this_cpu_dec_return(*p);
Christoph Lameter12938a92010-12-06 11:16:20 -0600498 t = __this_cpu_read(pcp->stat_threshold);
499 if (unlikely(v < - t)) {
500 s8 overstep = t >> 1;
Christoph Lameter2244b952006-06-30 01:55:33 -0700501
Christoph Lameter12938a92010-12-06 11:16:20 -0600502 zone_page_state_add(v - overstep, zone, item);
503 __this_cpu_write(*p, overstep);
Christoph Lameter2244b952006-06-30 01:55:33 -0700504 }
Ingo Molnarc68ed792021-09-08 15:17:57 -0700505
Thomas Gleixner7a025e92022-08-25 18:41:26 +0200506 preempt_enable_nested();
Christoph Lameter2244b952006-06-30 01:55:33 -0700507}
Christoph Lameterc8785382007-02-10 01:43:01 -0800508
Mel Gorman75ef7182016-07-28 15:45:24 -0700509void __dec_node_state(struct pglist_data *pgdat, enum node_stat_item item)
510{
511 struct per_cpu_nodestat __percpu *pcp = pgdat->per_cpu_nodestats;
512 s8 __percpu *p = pcp->vm_node_stat_diff + item;
513 s8 v, t;
514
Roman Gushchinea426c22020-08-06 23:20:35 -0700515 VM_WARN_ON_ONCE(vmstat_item_in_bytes(item));
516
Ingo Molnarc68ed792021-09-08 15:17:57 -0700517 /* See __mod_node_page_state */
Thomas Gleixner7a025e92022-08-25 18:41:26 +0200518 preempt_disable_nested();
Ingo Molnarc68ed792021-09-08 15:17:57 -0700519
Mel Gorman75ef7182016-07-28 15:45:24 -0700520 v = __this_cpu_dec_return(*p);
521 t = __this_cpu_read(pcp->stat_threshold);
522 if (unlikely(v < - t)) {
523 s8 overstep = t >> 1;
524
525 node_page_state_add(v - overstep, pgdat, item);
526 __this_cpu_write(*p, overstep);
527 }
Ingo Molnarc68ed792021-09-08 15:17:57 -0700528
Thomas Gleixner7a025e92022-08-25 18:41:26 +0200529 preempt_enable_nested();
Mel Gorman75ef7182016-07-28 15:45:24 -0700530}
531
Christoph Lameterc8785382007-02-10 01:43:01 -0800532void __dec_zone_page_state(struct page *page, enum zone_stat_item item)
533{
534 __dec_zone_state(page_zone(page), item);
535}
Christoph Lameter2244b952006-06-30 01:55:33 -0700536EXPORT_SYMBOL(__dec_zone_page_state);
537
Mel Gorman75ef7182016-07-28 15:45:24 -0700538void __dec_node_page_state(struct page *page, enum node_stat_item item)
539{
540 __dec_node_state(page_pgdat(page), item);
541}
542EXPORT_SYMBOL(__dec_node_page_state);
543
Heiko Carstens41561532012-01-12 17:17:30 -0800544#ifdef CONFIG_HAVE_CMPXCHG_LOCAL
Christoph Lameter7c839122010-12-14 10:28:46 -0600545/*
546 * If we have cmpxchg_local support then we do not need to incur the overhead
547 * that comes with local_irq_save/restore if we use this_cpu_cmpxchg.
548 *
549 * mod_state() modifies the zone counter state through atomic per cpu
550 * operations.
551 *
552 * Overstep mode specifies how overstep should handled:
553 * 0 No overstepping
554 * 1 Overstepping half of threshold
555 * -1 Overstepping minus half of threshold
556*/
Mel Gorman75ef7182016-07-28 15:45:24 -0700557static inline void mod_zone_state(struct zone *zone,
558 enum zone_stat_item item, long delta, int overstep_mode)
Christoph Lameter7c839122010-12-14 10:28:46 -0600559{
Mel Gorman28f836b2021-06-28 19:41:38 -0700560 struct per_cpu_zonestat __percpu *pcp = zone->per_cpu_zonestats;
Christoph Lameter7c839122010-12-14 10:28:46 -0600561 s8 __percpu *p = pcp->vm_stat_diff + item;
562 long o, n, t, z;
563
564 do {
565 z = 0; /* overflow to zone counters */
566
567 /*
568 * The fetching of the stat_threshold is racy. We may apply
569 * a counter threshold to the wrong the cpu if we get
Christoph Lameterd3bc2362011-04-14 15:21:58 -0700570 * rescheduled while executing here. However, the next
571 * counter update will apply the threshold again and
572 * therefore bring the counter under the threshold again.
573 *
574 * Most of the time the thresholds are the same anyways
575 * for all cpus in a zone.
Christoph Lameter7c839122010-12-14 10:28:46 -0600576 */
577 t = this_cpu_read(pcp->stat_threshold);
578
579 o = this_cpu_read(*p);
580 n = delta + o;
581
Miaohe Lin40610072020-10-15 20:07:36 -0700582 if (abs(n) > t) {
Christoph Lameter7c839122010-12-14 10:28:46 -0600583 int os = overstep_mode * (t >> 1) ;
584
585 /* Overflow must be added to zone counters */
586 z = n + os;
587 n = -os;
588 }
589 } while (this_cpu_cmpxchg(*p, o, n) != o);
590
591 if (z)
592 zone_page_state_add(z, zone, item);
593}
594
595void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
Heiko Carstens6cdb18a2015-12-29 14:54:32 -0800596 long delta)
Christoph Lameter7c839122010-12-14 10:28:46 -0600597{
Mel Gorman75ef7182016-07-28 15:45:24 -0700598 mod_zone_state(zone, item, delta, 0);
Christoph Lameter7c839122010-12-14 10:28:46 -0600599}
600EXPORT_SYMBOL(mod_zone_page_state);
601
Christoph Lameter7c839122010-12-14 10:28:46 -0600602void inc_zone_page_state(struct page *page, enum zone_stat_item item)
603{
Mel Gorman75ef7182016-07-28 15:45:24 -0700604 mod_zone_state(page_zone(page), item, 1, 1);
Christoph Lameter7c839122010-12-14 10:28:46 -0600605}
606EXPORT_SYMBOL(inc_zone_page_state);
607
608void dec_zone_page_state(struct page *page, enum zone_stat_item item)
609{
Mel Gorman75ef7182016-07-28 15:45:24 -0700610 mod_zone_state(page_zone(page), item, -1, -1);
Christoph Lameter7c839122010-12-14 10:28:46 -0600611}
612EXPORT_SYMBOL(dec_zone_page_state);
Mel Gorman75ef7182016-07-28 15:45:24 -0700613
614static inline void mod_node_state(struct pglist_data *pgdat,
615 enum node_stat_item item, int delta, int overstep_mode)
616{
617 struct per_cpu_nodestat __percpu *pcp = pgdat->per_cpu_nodestats;
618 s8 __percpu *p = pcp->vm_node_stat_diff + item;
619 long o, n, t, z;
620
Roman Gushchinea426c22020-08-06 23:20:35 -0700621 if (vmstat_item_in_bytes(item)) {
Johannes Weiner629484a2021-02-25 17:16:51 -0800622 /*
623 * Only cgroups use subpage accounting right now; at
624 * the global level, these items still change in
625 * multiples of whole pages. Store them as pages
626 * internally to keep the per-cpu counters compact.
627 */
Roman Gushchinea426c22020-08-06 23:20:35 -0700628 VM_WARN_ON_ONCE(delta & (PAGE_SIZE - 1));
629 delta >>= PAGE_SHIFT;
630 }
631
Mel Gorman75ef7182016-07-28 15:45:24 -0700632 do {
633 z = 0; /* overflow to node counters */
634
635 /*
636 * The fetching of the stat_threshold is racy. We may apply
637 * a counter threshold to the wrong the cpu if we get
638 * rescheduled while executing here. However, the next
639 * counter update will apply the threshold again and
640 * therefore bring the counter under the threshold again.
641 *
642 * Most of the time the thresholds are the same anyways
643 * for all cpus in a node.
644 */
645 t = this_cpu_read(pcp->stat_threshold);
646
647 o = this_cpu_read(*p);
648 n = delta + o;
649
Miaohe Lin40610072020-10-15 20:07:36 -0700650 if (abs(n) > t) {
Mel Gorman75ef7182016-07-28 15:45:24 -0700651 int os = overstep_mode * (t >> 1) ;
652
653 /* Overflow must be added to node counters */
654 z = n + os;
655 n = -os;
656 }
657 } while (this_cpu_cmpxchg(*p, o, n) != o);
658
659 if (z)
660 node_page_state_add(z, pgdat, item);
661}
662
663void mod_node_page_state(struct pglist_data *pgdat, enum node_stat_item item,
664 long delta)
665{
666 mod_node_state(pgdat, item, delta, 0);
667}
668EXPORT_SYMBOL(mod_node_page_state);
669
670void inc_node_state(struct pglist_data *pgdat, enum node_stat_item item)
671{
672 mod_node_state(pgdat, item, 1, 1);
673}
674
675void inc_node_page_state(struct page *page, enum node_stat_item item)
676{
677 mod_node_state(page_pgdat(page), item, 1, 1);
678}
679EXPORT_SYMBOL(inc_node_page_state);
680
681void dec_node_page_state(struct page *page, enum node_stat_item item)
682{
683 mod_node_state(page_pgdat(page), item, -1, -1);
684}
685EXPORT_SYMBOL(dec_node_page_state);
Christoph Lameter7c839122010-12-14 10:28:46 -0600686#else
687/*
688 * Use interrupt disable to serialize counter updates
689 */
690void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
Heiko Carstens6cdb18a2015-12-29 14:54:32 -0800691 long delta)
Christoph Lameter7c839122010-12-14 10:28:46 -0600692{
693 unsigned long flags;
694
695 local_irq_save(flags);
696 __mod_zone_page_state(zone, item, delta);
697 local_irq_restore(flags);
698}
699EXPORT_SYMBOL(mod_zone_page_state);
700
Christoph Lameter2244b952006-06-30 01:55:33 -0700701void inc_zone_page_state(struct page *page, enum zone_stat_item item)
702{
703 unsigned long flags;
704 struct zone *zone;
Christoph Lameter2244b952006-06-30 01:55:33 -0700705
706 zone = page_zone(page);
707 local_irq_save(flags);
Christoph Lameterca889e62006-06-30 01:55:44 -0700708 __inc_zone_state(zone, item);
Christoph Lameter2244b952006-06-30 01:55:33 -0700709 local_irq_restore(flags);
710}
711EXPORT_SYMBOL(inc_zone_page_state);
712
713void dec_zone_page_state(struct page *page, enum zone_stat_item item)
714{
715 unsigned long flags;
Christoph Lameter2244b952006-06-30 01:55:33 -0700716
Christoph Lameter2244b952006-06-30 01:55:33 -0700717 local_irq_save(flags);
Christoph Lametera302eb42006-08-31 21:27:34 -0700718 __dec_zone_page_state(page, item);
Christoph Lameter2244b952006-06-30 01:55:33 -0700719 local_irq_restore(flags);
720}
721EXPORT_SYMBOL(dec_zone_page_state);
722
Mel Gorman75ef7182016-07-28 15:45:24 -0700723void inc_node_state(struct pglist_data *pgdat, enum node_stat_item item)
724{
725 unsigned long flags;
726
727 local_irq_save(flags);
728 __inc_node_state(pgdat, item);
729 local_irq_restore(flags);
730}
731EXPORT_SYMBOL(inc_node_state);
732
733void mod_node_page_state(struct pglist_data *pgdat, enum node_stat_item item,
734 long delta)
735{
736 unsigned long flags;
737
738 local_irq_save(flags);
739 __mod_node_page_state(pgdat, item, delta);
740 local_irq_restore(flags);
741}
742EXPORT_SYMBOL(mod_node_page_state);
743
744void inc_node_page_state(struct page *page, enum node_stat_item item)
745{
746 unsigned long flags;
747 struct pglist_data *pgdat;
748
749 pgdat = page_pgdat(page);
750 local_irq_save(flags);
751 __inc_node_state(pgdat, item);
752 local_irq_restore(flags);
753}
754EXPORT_SYMBOL(inc_node_page_state);
755
756void dec_node_page_state(struct page *page, enum node_stat_item item)
757{
758 unsigned long flags;
759
760 local_irq_save(flags);
761 __dec_node_page_state(page, item);
762 local_irq_restore(flags);
763}
764EXPORT_SYMBOL(dec_node_page_state);
765#endif
Christoph Lameter7cc36bb2014-10-09 15:29:43 -0700766
767/*
768 * Fold a differential into the global counters.
769 * Returns the number of counters updated.
770 */
Mel Gorman75ef7182016-07-28 15:45:24 -0700771static int fold_diff(int *zone_diff, int *node_diff)
Christoph Lameter4edb0742013-09-11 14:21:31 -0700772{
773 int i;
Christoph Lameter7cc36bb2014-10-09 15:29:43 -0700774 int changes = 0;
Christoph Lameter4edb0742013-09-11 14:21:31 -0700775
776 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
Mel Gorman75ef7182016-07-28 15:45:24 -0700777 if (zone_diff[i]) {
778 atomic_long_add(zone_diff[i], &vm_zone_stat[i]);
779 changes++;
780 }
781
782 for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++)
783 if (node_diff[i]) {
784 atomic_long_add(node_diff[i], &vm_node_stat[i]);
Christoph Lameter7cc36bb2014-10-09 15:29:43 -0700785 changes++;
786 }
787 return changes;
Christoph Lameter4edb0742013-09-11 14:21:31 -0700788}
Mel Gormanf19298b2021-06-28 19:41:44 -0700789
Christoph Lameter2244b952006-06-30 01:55:33 -0700790/*
Christoph Lameter2bb921e2013-09-11 14:21:30 -0700791 * Update the zone counters for the current cpu.
Christoph Lametera7f75e22008-02-04 22:29:16 -0800792 *
Christoph Lameter4037d452007-05-09 02:35:14 -0700793 * Note that refresh_cpu_vm_stats strives to only access
794 * node local memory. The per cpu pagesets on remote zones are placed
795 * in the memory local to the processor using that pageset. So the
796 * loop over all zones will access a series of cachelines local to
797 * the processor.
798 *
799 * The call to zone_page_state_add updates the cachelines with the
800 * statistics in the remote zone struct as well as the global cachelines
801 * with the global counters. These could cause remote node cache line
802 * bouncing and will have to be only done when necessary.
Christoph Lameter7cc36bb2014-10-09 15:29:43 -0700803 *
804 * The function returns the number of global counters updated.
Christoph Lameter2244b952006-06-30 01:55:33 -0700805 */
Christoph Lameter0eb77e92016-01-14 15:21:40 -0800806static int refresh_cpu_vm_stats(bool do_pagesets)
Christoph Lameter2244b952006-06-30 01:55:33 -0700807{
Mel Gorman75ef7182016-07-28 15:45:24 -0700808 struct pglist_data *pgdat;
Christoph Lameter2244b952006-06-30 01:55:33 -0700809 struct zone *zone;
810 int i;
Mel Gorman75ef7182016-07-28 15:45:24 -0700811 int global_zone_diff[NR_VM_ZONE_STAT_ITEMS] = { 0, };
812 int global_node_diff[NR_VM_NODE_STAT_ITEMS] = { 0, };
Christoph Lameter7cc36bb2014-10-09 15:29:43 -0700813 int changes = 0;
Christoph Lameter2244b952006-06-30 01:55:33 -0700814
KOSAKI Motohiroee99c712009-03-31 15:19:31 -0700815 for_each_populated_zone(zone) {
Mel Gorman28f836b2021-06-28 19:41:38 -0700816 struct per_cpu_zonestat __percpu *pzstats = zone->per_cpu_zonestats;
817#ifdef CONFIG_NUMA
818 struct per_cpu_pages __percpu *pcp = zone->per_cpu_pageset;
819#endif
Christoph Lameter2244b952006-06-30 01:55:33 -0700820
Christoph Lameterfbc2edb2013-09-11 14:21:32 -0700821 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) {
822 int v;
Christoph Lameter2244b952006-06-30 01:55:33 -0700823
Mel Gorman28f836b2021-06-28 19:41:38 -0700824 v = this_cpu_xchg(pzstats->vm_stat_diff[i], 0);
Christoph Lameterfbc2edb2013-09-11 14:21:32 -0700825 if (v) {
Christoph Lametera7f75e22008-02-04 22:29:16 -0800826
Christoph Lametera7f75e22008-02-04 22:29:16 -0800827 atomic_long_add(v, &zone->vm_stat[i]);
Mel Gorman75ef7182016-07-28 15:45:24 -0700828 global_zone_diff[i] += v;
Christoph Lameter4037d452007-05-09 02:35:14 -0700829#ifdef CONFIG_NUMA
830 /* 3 seconds idle till flush */
Mel Gorman28f836b2021-06-28 19:41:38 -0700831 __this_cpu_write(pcp->expire, 3);
Christoph Lameter4037d452007-05-09 02:35:14 -0700832#endif
Christoph Lameter2244b952006-06-30 01:55:33 -0700833 }
Christoph Lameterfbc2edb2013-09-11 14:21:32 -0700834 }
Christoph Lameter4037d452007-05-09 02:35:14 -0700835#ifdef CONFIG_NUMA
Kemi Wang3a321d22017-09-08 16:12:48 -0700836
Christoph Lameter0eb77e92016-01-14 15:21:40 -0800837 if (do_pagesets) {
838 cond_resched();
839 /*
840 * Deal with draining the remote pageset of this
841 * processor
842 *
843 * Check if there are pages remaining in this pageset
844 * if not then there is nothing to expire.
845 */
Mel Gorman28f836b2021-06-28 19:41:38 -0700846 if (!__this_cpu_read(pcp->expire) ||
847 !__this_cpu_read(pcp->count))
Christoph Lameter0eb77e92016-01-14 15:21:40 -0800848 continue;
Christoph Lameter4037d452007-05-09 02:35:14 -0700849
Christoph Lameter0eb77e92016-01-14 15:21:40 -0800850 /*
851 * We never drain zones local to this processor.
852 */
853 if (zone_to_nid(zone) == numa_node_id()) {
Mel Gorman28f836b2021-06-28 19:41:38 -0700854 __this_cpu_write(pcp->expire, 0);
Christoph Lameter0eb77e92016-01-14 15:21:40 -0800855 continue;
856 }
Christoph Lameter4037d452007-05-09 02:35:14 -0700857
Mel Gorman28f836b2021-06-28 19:41:38 -0700858 if (__this_cpu_dec_return(pcp->expire))
Christoph Lameter0eb77e92016-01-14 15:21:40 -0800859 continue;
Christoph Lameter4037d452007-05-09 02:35:14 -0700860
Mel Gorman28f836b2021-06-28 19:41:38 -0700861 if (__this_cpu_read(pcp->count)) {
862 drain_zone_pages(zone, this_cpu_ptr(pcp));
Christoph Lameter0eb77e92016-01-14 15:21:40 -0800863 changes++;
864 }
Christoph Lameter7cc36bb2014-10-09 15:29:43 -0700865 }
Christoph Lameter4037d452007-05-09 02:35:14 -0700866#endif
Christoph Lameter2244b952006-06-30 01:55:33 -0700867 }
Mel Gorman75ef7182016-07-28 15:45:24 -0700868
869 for_each_online_pgdat(pgdat) {
870 struct per_cpu_nodestat __percpu *p = pgdat->per_cpu_nodestats;
871
872 for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) {
873 int v;
874
875 v = this_cpu_xchg(p->vm_node_stat_diff[i], 0);
876 if (v) {
877 atomic_long_add(v, &pgdat->vm_stat[i]);
878 global_node_diff[i] += v;
879 }
880 }
881 }
882
883 changes += fold_diff(global_zone_diff, global_node_diff);
Christoph Lameter7cc36bb2014-10-09 15:29:43 -0700884 return changes;
Christoph Lameter2244b952006-06-30 01:55:33 -0700885}
886
Cody P Schafer40f4b1e2013-04-29 15:08:38 -0700887/*
Christoph Lameter2bb921e2013-09-11 14:21:30 -0700888 * Fold the data for an offline cpu into the global array.
889 * There cannot be any access by the offline cpu and therefore
890 * synchronization is simplified.
891 */
892void cpu_vm_stats_fold(int cpu)
893{
Mel Gorman75ef7182016-07-28 15:45:24 -0700894 struct pglist_data *pgdat;
Christoph Lameter2bb921e2013-09-11 14:21:30 -0700895 struct zone *zone;
896 int i;
Mel Gorman75ef7182016-07-28 15:45:24 -0700897 int global_zone_diff[NR_VM_ZONE_STAT_ITEMS] = { 0, };
898 int global_node_diff[NR_VM_NODE_STAT_ITEMS] = { 0, };
Christoph Lameter2bb921e2013-09-11 14:21:30 -0700899
900 for_each_populated_zone(zone) {
Mel Gorman28f836b2021-06-28 19:41:38 -0700901 struct per_cpu_zonestat *pzstats;
Christoph Lameter2bb921e2013-09-11 14:21:30 -0700902
Mel Gorman28f836b2021-06-28 19:41:38 -0700903 pzstats = per_cpu_ptr(zone->per_cpu_zonestats, cpu);
Christoph Lameter2bb921e2013-09-11 14:21:30 -0700904
Mel Gormanf19298b2021-06-28 19:41:44 -0700905 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) {
Mel Gorman28f836b2021-06-28 19:41:38 -0700906 if (pzstats->vm_stat_diff[i]) {
Christoph Lameter2bb921e2013-09-11 14:21:30 -0700907 int v;
908
Mel Gorman28f836b2021-06-28 19:41:38 -0700909 v = pzstats->vm_stat_diff[i];
910 pzstats->vm_stat_diff[i] = 0;
Christoph Lameter2bb921e2013-09-11 14:21:30 -0700911 atomic_long_add(v, &zone->vm_stat[i]);
Mel Gorman75ef7182016-07-28 15:45:24 -0700912 global_zone_diff[i] += v;
Christoph Lameter2bb921e2013-09-11 14:21:30 -0700913 }
Mel Gormanf19298b2021-06-28 19:41:44 -0700914 }
Kemi Wang3a321d22017-09-08 16:12:48 -0700915#ifdef CONFIG_NUMA
Mel Gormanf19298b2021-06-28 19:41:44 -0700916 for (i = 0; i < NR_VM_NUMA_EVENT_ITEMS; i++) {
917 if (pzstats->vm_numa_event[i]) {
918 unsigned long v;
Kemi Wang3a321d22017-09-08 16:12:48 -0700919
Mel Gormanf19298b2021-06-28 19:41:44 -0700920 v = pzstats->vm_numa_event[i];
921 pzstats->vm_numa_event[i] = 0;
922 zone_numa_event_add(v, zone, i);
Kemi Wang3a321d22017-09-08 16:12:48 -0700923 }
Mel Gormanf19298b2021-06-28 19:41:44 -0700924 }
Kemi Wang3a321d22017-09-08 16:12:48 -0700925#endif
Christoph Lameter2bb921e2013-09-11 14:21:30 -0700926 }
927
Mel Gorman75ef7182016-07-28 15:45:24 -0700928 for_each_online_pgdat(pgdat) {
929 struct per_cpu_nodestat *p;
930
931 p = per_cpu_ptr(pgdat->per_cpu_nodestats, cpu);
932
933 for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++)
934 if (p->vm_node_stat_diff[i]) {
935 int v;
936
937 v = p->vm_node_stat_diff[i];
938 p->vm_node_stat_diff[i] = 0;
939 atomic_long_add(v, &pgdat->vm_stat[i]);
940 global_node_diff[i] += v;
941 }
942 }
943
944 fold_diff(global_zone_diff, global_node_diff);
Christoph Lameter2bb921e2013-09-11 14:21:30 -0700945}
946
947/*
Cody P Schafer40f4b1e2013-04-29 15:08:38 -0700948 * this is only called if !populated_zone(zone), which implies no other users of
Ingo Molnarf0953a12021-05-06 18:06:47 -0700949 * pset->vm_stat_diff[] exist.
Cody P Schafer40f4b1e2013-04-29 15:08:38 -0700950 */
Mel Gorman28f836b2021-06-28 19:41:38 -0700951void drain_zonestat(struct zone *zone, struct per_cpu_zonestat *pzstats)
Minchan Kim5a883812012-10-08 16:33:39 -0700952{
Mel Gormanf19298b2021-06-28 19:41:44 -0700953 unsigned long v;
Minchan Kim5a883812012-10-08 16:33:39 -0700954 int i;
955
Mel Gormanf19298b2021-06-28 19:41:44 -0700956 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) {
Mel Gorman28f836b2021-06-28 19:41:38 -0700957 if (pzstats->vm_stat_diff[i]) {
Mel Gormanf19298b2021-06-28 19:41:44 -0700958 v = pzstats->vm_stat_diff[i];
Mel Gorman28f836b2021-06-28 19:41:38 -0700959 pzstats->vm_stat_diff[i] = 0;
Mel Gormanf19298b2021-06-28 19:41:44 -0700960 zone_page_state_add(v, zone, i);
Minchan Kim5a883812012-10-08 16:33:39 -0700961 }
Mel Gormanf19298b2021-06-28 19:41:44 -0700962 }
Kemi Wang3a321d22017-09-08 16:12:48 -0700963
964#ifdef CONFIG_NUMA
Mel Gormanf19298b2021-06-28 19:41:44 -0700965 for (i = 0; i < NR_VM_NUMA_EVENT_ITEMS; i++) {
966 if (pzstats->vm_numa_event[i]) {
967 v = pzstats->vm_numa_event[i];
968 pzstats->vm_numa_event[i] = 0;
969 zone_numa_event_add(v, zone, i);
Kemi Wang3a321d22017-09-08 16:12:48 -0700970 }
Mel Gormanf19298b2021-06-28 19:41:44 -0700971 }
Kemi Wang3a321d22017-09-08 16:12:48 -0700972#endif
Minchan Kim5a883812012-10-08 16:33:39 -0700973}
Christoph Lameter2244b952006-06-30 01:55:33 -0700974#endif
975
Christoph Lameterca889e62006-06-30 01:55:44 -0700976#ifdef CONFIG_NUMA
977/*
Mel Gorman75ef7182016-07-28 15:45:24 -0700978 * Determine the per node value of a stat item. This function
979 * is called frequently in a NUMA machine, so try to be as
980 * frugal as possible.
Andrew Mortonc2d42c12015-11-05 18:48:43 -0800981 */
Mel Gorman75ef7182016-07-28 15:45:24 -0700982unsigned long sum_zone_node_page_state(int node,
983 enum zone_stat_item item)
Andrew Mortonc2d42c12015-11-05 18:48:43 -0800984{
985 struct zone *zones = NODE_DATA(node)->node_zones;
Joonsoo Kime87d59f2016-05-19 17:12:29 -0700986 int i;
987 unsigned long count = 0;
Andrew Mortonc2d42c12015-11-05 18:48:43 -0800988
Joonsoo Kime87d59f2016-05-19 17:12:29 -0700989 for (i = 0; i < MAX_NR_ZONES; i++)
990 count += zone_page_state(zones + i, item);
991
992 return count;
Andrew Mortonc2d42c12015-11-05 18:48:43 -0800993}
994
Mel Gormanf19298b2021-06-28 19:41:44 -0700995/* Determine the per node value of a numa stat item. */
996unsigned long sum_zone_numa_event_state(int node,
Kemi Wang3a321d22017-09-08 16:12:48 -0700997 enum numa_stat_item item)
998{
999 struct zone *zones = NODE_DATA(node)->node_zones;
Kemi Wang3a321d22017-09-08 16:12:48 -07001000 unsigned long count = 0;
Mel Gormanf19298b2021-06-28 19:41:44 -07001001 int i;
Kemi Wang3a321d22017-09-08 16:12:48 -07001002
1003 for (i = 0; i < MAX_NR_ZONES; i++)
Mel Gormanf19298b2021-06-28 19:41:44 -07001004 count += zone_numa_event_state(zones + i, item);
Kemi Wang3a321d22017-09-08 16:12:48 -07001005
1006 return count;
1007}
1008
Mel Gorman75ef7182016-07-28 15:45:24 -07001009/*
1010 * Determine the per node value of a stat item.
1011 */
Roman Gushchinea426c22020-08-06 23:20:35 -07001012unsigned long node_page_state_pages(struct pglist_data *pgdat,
1013 enum node_stat_item item)
Mel Gorman75ef7182016-07-28 15:45:24 -07001014{
1015 long x = atomic_long_read(&pgdat->vm_stat[item]);
1016#ifdef CONFIG_SMP
1017 if (x < 0)
1018 x = 0;
1019#endif
1020 return x;
1021}
Roman Gushchinea426c22020-08-06 23:20:35 -07001022
1023unsigned long node_page_state(struct pglist_data *pgdat,
1024 enum node_stat_item item)
1025{
1026 VM_WARN_ON_ONCE(vmstat_item_in_bytes(item));
1027
1028 return node_page_state_pages(pgdat, item);
1029}
Christoph Lameterca889e62006-06-30 01:55:44 -07001030#endif
1031
Mel Gormand7a57522010-05-24 14:32:25 -07001032#ifdef CONFIG_COMPACTION
Namhyung Kim36deb0b2010-10-26 14:22:04 -07001033
Mel Gormand7a57522010-05-24 14:32:25 -07001034struct contig_page_info {
1035 unsigned long free_pages;
1036 unsigned long free_blocks_total;
1037 unsigned long free_blocks_suitable;
1038};
1039
1040/*
1041 * Calculate the number of free pages in a zone, how many contiguous
1042 * pages are free and how many are large enough to satisfy an allocation of
1043 * the target size. Note that this function makes no attempt to estimate
1044 * how many suitable free blocks there *might* be if MOVABLE pages were
1045 * migrated. Calculating that is possible, but expensive and can be
1046 * figured out from userspace
1047 */
1048static void fill_contig_page_info(struct zone *zone,
1049 unsigned int suitable_order,
1050 struct contig_page_info *info)
1051{
1052 unsigned int order;
1053
1054 info->free_pages = 0;
1055 info->free_blocks_total = 0;
1056 info->free_blocks_suitable = 0;
1057
Kirill A. Shutemov23baf832023-03-15 14:31:33 +03001058 for (order = 0; order <= MAX_ORDER; order++) {
Mel Gormand7a57522010-05-24 14:32:25 -07001059 unsigned long blocks;
1060
Liu Shixinaf1c31a2021-11-05 13:43:59 -07001061 /*
1062 * Count number of free blocks.
1063 *
1064 * Access to nr_free is lockless as nr_free is used only for
1065 * diagnostic purposes. Use data_race to avoid KCSAN warning.
1066 */
1067 blocks = data_race(zone->free_area[order].nr_free);
Mel Gormand7a57522010-05-24 14:32:25 -07001068 info->free_blocks_total += blocks;
1069
1070 /* Count free base pages */
1071 info->free_pages += blocks << order;
1072
1073 /* Count the suitable free blocks */
1074 if (order >= suitable_order)
1075 info->free_blocks_suitable += blocks <<
1076 (order - suitable_order);
1077 }
1078}
Mel Gormanf1a5ab12010-05-24 14:32:26 -07001079
1080/*
1081 * A fragmentation index only makes sense if an allocation of a requested
1082 * size would fail. If that is true, the fragmentation index indicates
1083 * whether external fragmentation or a lack of memory was the problem.
1084 * The value can be used to determine if page reclaim or compaction
1085 * should be used
1086 */
Mel Gorman56de7262010-05-24 14:32:30 -07001087static int __fragmentation_index(unsigned int order, struct contig_page_info *info)
Mel Gormanf1a5ab12010-05-24 14:32:26 -07001088{
1089 unsigned long requested = 1UL << order;
1090
Kirill A. Shutemov23baf832023-03-15 14:31:33 +03001091 if (WARN_ON_ONCE(order > MAX_ORDER))
Wen Yang88d6ac42017-09-06 16:24:06 -07001092 return 0;
1093
Mel Gormanf1a5ab12010-05-24 14:32:26 -07001094 if (!info->free_blocks_total)
1095 return 0;
1096
1097 /* Fragmentation index only makes sense when a request would fail */
1098 if (info->free_blocks_suitable)
1099 return -1000;
1100
1101 /*
1102 * Index is between 0 and 1 so return within 3 decimal places
1103 *
1104 * 0 => allocation would fail due to lack of memory
1105 * 1 => allocation would fail due to fragmentation
1106 */
1107 return 1000 - div_u64( (1000+(div_u64(info->free_pages * 1000ULL, requested))), info->free_blocks_total);
1108}
Mel Gorman56de7262010-05-24 14:32:30 -07001109
Nitin Guptafacdaa92020-08-11 18:31:00 -07001110/*
1111 * Calculates external fragmentation within a zone wrt the given order.
1112 * It is defined as the percentage of pages found in blocks of size
1113 * less than 1 << order. It returns values in range [0, 100].
1114 */
Nitin Guptad34c0a72020-08-11 18:31:07 -07001115unsigned int extfrag_for_order(struct zone *zone, unsigned int order)
Nitin Guptafacdaa92020-08-11 18:31:00 -07001116{
1117 struct contig_page_info info;
1118
1119 fill_contig_page_info(zone, order, &info);
1120 if (info.free_pages == 0)
1121 return 0;
1122
1123 return div_u64((info.free_pages -
1124 (info.free_blocks_suitable << order)) * 100,
1125 info.free_pages);
1126}
1127
Mel Gorman56de7262010-05-24 14:32:30 -07001128/* Same as __fragmentation index but allocs contig_page_info on stack */
1129int fragmentation_index(struct zone *zone, unsigned int order)
1130{
1131 struct contig_page_info info;
1132
1133 fill_contig_page_info(zone, order, &info);
1134 return __fragmentation_index(order, &info);
1135}
Mel Gormand7a57522010-05-24 14:32:25 -07001136#endif
1137
Konstantin Khlebnikovebc5d83d2019-12-04 16:49:53 -08001138#if defined(CONFIG_PROC_FS) || defined(CONFIG_SYSFS) || \
1139 defined(CONFIG_NUMA) || defined(CONFIG_MEMCG)
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001140#ifdef CONFIG_ZONE_DMA
1141#define TEXT_FOR_DMA(xx) xx "_dma",
1142#else
1143#define TEXT_FOR_DMA(xx)
1144#endif
1145
1146#ifdef CONFIG_ZONE_DMA32
1147#define TEXT_FOR_DMA32(xx) xx "_dma32",
1148#else
1149#define TEXT_FOR_DMA32(xx)
1150#endif
1151
1152#ifdef CONFIG_HIGHMEM
1153#define TEXT_FOR_HIGHMEM(xx) xx "_high",
1154#else
1155#define TEXT_FOR_HIGHMEM(xx)
1156#endif
1157
Hao Leea39c5d32022-08-07 15:44:42 +00001158#ifdef CONFIG_ZONE_DEVICE
1159#define TEXT_FOR_DEVICE(xx) xx "_device",
1160#else
1161#define TEXT_FOR_DEVICE(xx)
1162#endif
1163
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001164#define TEXTS_FOR_ZONES(xx) TEXT_FOR_DMA(xx) TEXT_FOR_DMA32(xx) xx "_normal", \
Hao Leea39c5d32022-08-07 15:44:42 +00001165 TEXT_FOR_HIGHMEM(xx) xx "_movable", \
1166 TEXT_FOR_DEVICE(xx)
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001167
1168const char * const vmstat_text[] = {
NeilBrown8d928902020-06-01 21:48:21 -07001169 /* enum zone_stat_item counters */
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001170 "nr_free_pages",
Minchan Kim71c799f2016-07-28 15:47:26 -07001171 "nr_zone_inactive_anon",
1172 "nr_zone_active_anon",
1173 "nr_zone_inactive_file",
1174 "nr_zone_active_file",
1175 "nr_zone_unevictable",
Mel Gorman5a1c84b2016-07-28 15:47:31 -07001176 "nr_zone_write_pending",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001177 "nr_mlock",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001178 "nr_bounce",
Minchan Kim91537fe2016-07-26 15:24:45 -07001179#if IS_ENABLED(CONFIG_ZSMALLOC)
1180 "nr_zspages",
1181#endif
Kemi Wang3a321d22017-09-08 16:12:48 -07001182 "nr_free_cma",
Kirill A. Shutemovdcdfdd42023-06-06 17:26:29 +03001183#ifdef CONFIG_UNACCEPTED_MEMORY
1184 "nr_unaccepted",
1185#endif
Kemi Wang3a321d22017-09-08 16:12:48 -07001186
1187 /* enum numa_stat_item counters */
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001188#ifdef CONFIG_NUMA
1189 "numa_hit",
1190 "numa_miss",
1191 "numa_foreign",
1192 "numa_interleave",
1193 "numa_local",
1194 "numa_other",
1195#endif
Konstantin Khlebnikov09316c02014-10-09 15:29:32 -07001196
Konstantin Khlebnikov9d7ea9a2019-12-04 16:49:50 -08001197 /* enum node_stat_item counters */
Mel Gorman599d0c92016-07-28 15:45:31 -07001198 "nr_inactive_anon",
1199 "nr_active_anon",
1200 "nr_inactive_file",
1201 "nr_active_file",
1202 "nr_unevictable",
Johannes Weiner385386c2017-07-06 15:40:43 -07001203 "nr_slab_reclaimable",
1204 "nr_slab_unreclaimable",
Mel Gorman599d0c92016-07-28 15:45:31 -07001205 "nr_isolated_anon",
1206 "nr_isolated_file",
Johannes Weiner68d48e62018-10-26 15:06:39 -07001207 "workingset_nodes",
Joonsoo Kim170b04b72020-08-11 18:30:43 -07001208 "workingset_refault_anon",
1209 "workingset_refault_file",
1210 "workingset_activate_anon",
1211 "workingset_activate_file",
1212 "workingset_restore_anon",
1213 "workingset_restore_file",
Mel Gorman1e6b10852016-07-28 15:46:08 -07001214 "workingset_nodereclaim",
Mel Gorman50658e22016-07-28 15:46:14 -07001215 "nr_anon_pages",
1216 "nr_mapped",
Mel Gorman11fb9982016-07-28 15:46:20 -07001217 "nr_file_pages",
1218 "nr_dirty",
1219 "nr_writeback",
1220 "nr_writeback_temp",
1221 "nr_shmem",
1222 "nr_shmem_hugepages",
1223 "nr_shmem_pmdmapped",
Song Liu60fbf0a2019-09-23 15:37:54 -07001224 "nr_file_hugepages",
1225 "nr_file_pmdmapped",
Mel Gorman11fb9982016-07-28 15:46:20 -07001226 "nr_anon_transparent_hugepages",
Mel Gormanc4a25632016-07-28 15:46:23 -07001227 "nr_vmscan_write",
1228 "nr_vmscan_immediate_reclaim",
1229 "nr_dirtied",
1230 "nr_written",
Mel Gorman8cd7c582021-11-05 13:42:25 -07001231 "nr_throttled_written",
Vlastimil Babkab29940c2018-10-26 15:05:46 -07001232 "nr_kernel_misc_reclaimable",
John Hubbard1970dc62020-04-01 21:05:37 -07001233 "nr_foll_pin_acquired",
1234 "nr_foll_pin_released",
Shakeel Butt991e7672020-08-06 23:21:37 -07001235 "nr_kernel_stack",
1236#if IS_ENABLED(CONFIG_SHADOW_CALL_STACK)
1237 "nr_shadow_call_stack",
1238#endif
Shakeel Buttf0c0c112020-12-14 19:07:17 -08001239 "nr_page_table_pages",
Yosry Ahmedebc97a52022-08-23 00:46:36 +00001240 "nr_sec_page_table_pages",
Shakeel Buttb6038942021-02-24 12:03:55 -08001241#ifdef CONFIG_SWAP
1242 "nr_swapcached",
1243#endif
Huang Yinge39bb6b2022-03-22 14:46:20 -07001244#ifdef CONFIG_NUMA_BALANCING
1245 "pgpromote_success",
Huang Yingc6833e12022-07-13 16:39:52 +08001246 "pgpromote_candidate",
Huang Yinge39bb6b2022-03-22 14:46:20 -07001247#endif
Mel Gorman599d0c92016-07-28 15:45:31 -07001248
Konstantin Khlebnikov09316c02014-10-09 15:29:32 -07001249 /* enum writeback_stat_item counters */
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001250 "nr_dirty_threshold",
1251 "nr_dirty_background_threshold",
1252
Konstantin Khlebnikovebc5d83d2019-12-04 16:49:53 -08001253#if defined(CONFIG_VM_EVENT_COUNTERS) || defined(CONFIG_MEMCG)
Konstantin Khlebnikov09316c02014-10-09 15:29:32 -07001254 /* enum vm_event_item counters */
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001255 "pgpgin",
1256 "pgpgout",
1257 "pswpin",
1258 "pswpout",
1259
1260 TEXTS_FOR_ZONES("pgalloc")
Mel Gorman7cc30fc2016-07-28 15:46:59 -07001261 TEXTS_FOR_ZONES("allocstall")
1262 TEXTS_FOR_ZONES("pgskip")
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001263
1264 "pgfree",
1265 "pgactivate",
1266 "pgdeactivate",
Shaohua Lif7ad2a62017-05-03 14:52:29 -07001267 "pglazyfree",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001268
1269 "pgfault",
1270 "pgmajfault",
Minchan Kim854e9ed2016-01-15 16:54:53 -08001271 "pglazyfreed",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001272
Mel Gorman599d0c92016-07-28 15:45:31 -07001273 "pgrefill",
Peter Xu798a6b82020-08-21 19:49:58 -04001274 "pgreuse",
Mel Gorman599d0c92016-07-28 15:45:31 -07001275 "pgsteal_kswapd",
1276 "pgsteal_direct",
Johannes Weiner57e9cc52022-10-26 14:01:33 -04001277 "pgsteal_khugepaged",
Yang Shi668e4142021-09-02 14:59:19 -07001278 "pgdemote_kswapd",
1279 "pgdemote_direct",
Johannes Weiner57e9cc52022-10-26 14:01:33 -04001280 "pgdemote_khugepaged",
Mel Gorman599d0c92016-07-28 15:45:31 -07001281 "pgscan_kswapd",
1282 "pgscan_direct",
Johannes Weiner57e9cc52022-10-26 14:01:33 -04001283 "pgscan_khugepaged",
Mel Gorman68243e72012-07-31 16:44:39 -07001284 "pgscan_direct_throttle",
Johannes Weiner497a6c12020-06-03 16:02:34 -07001285 "pgscan_anon",
1286 "pgscan_file",
1287 "pgsteal_anon",
1288 "pgsteal_file",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001289
1290#ifdef CONFIG_NUMA
1291 "zone_reclaim_failed",
1292#endif
1293 "pginodesteal",
1294 "slabs_scanned",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001295 "kswapd_inodesteal",
1296 "kswapd_low_wmark_hit_quickly",
1297 "kswapd_high_wmark_hit_quickly",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001298 "pageoutrun",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001299
1300 "pgrotated",
1301
Dave Hansen5509a5d2014-04-03 14:48:19 -07001302 "drop_pagecache",
1303 "drop_slab",
Konstantin Khlebnikov8e675f72017-07-06 15:40:28 -07001304 "oom_kill",
Dave Hansen5509a5d2014-04-03 14:48:19 -07001305
Mel Gorman03c5a6e2012-11-02 14:52:48 +00001306#ifdef CONFIG_NUMA_BALANCING
1307 "numa_pte_updates",
Mel Gorman72403b42013-11-12 15:08:32 -08001308 "numa_huge_pte_updates",
Mel Gorman03c5a6e2012-11-02 14:52:48 +00001309 "numa_hint_faults",
1310 "numa_hint_faults_local",
1311 "numa_pages_migrated",
1312#endif
Mel Gorman5647bc22012-10-19 10:46:20 +01001313#ifdef CONFIG_MIGRATION
1314 "pgmigrate_success",
1315 "pgmigrate_fail",
Anshuman Khandual1a5bae22020-08-11 18:31:51 -07001316 "thp_migration_success",
1317 "thp_migration_fail",
1318 "thp_migration_split",
Mel Gorman5647bc22012-10-19 10:46:20 +01001319#endif
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001320#ifdef CONFIG_COMPACTION
Mel Gorman397487d2012-10-19 12:00:10 +01001321 "compact_migrate_scanned",
1322 "compact_free_scanned",
1323 "compact_isolated",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001324 "compact_stall",
1325 "compact_fail",
1326 "compact_success",
Vlastimil Babka698b1b32016-03-17 14:18:08 -07001327 "compact_daemon_wake",
David Rientjes7f354a52017-02-22 15:44:50 -08001328 "compact_daemon_migrate_scanned",
1329 "compact_daemon_free_scanned",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001330#endif
1331
1332#ifdef CONFIG_HUGETLB_PAGE
1333 "htlb_buddy_alloc_success",
1334 "htlb_buddy_alloc_fail",
1335#endif
Minchan Kimbbb26922021-05-04 18:37:19 -07001336#ifdef CONFIG_CMA
1337 "cma_alloc_success",
1338 "cma_alloc_fail",
1339#endif
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001340 "unevictable_pgs_culled",
1341 "unevictable_pgs_scanned",
1342 "unevictable_pgs_rescued",
1343 "unevictable_pgs_mlocked",
1344 "unevictable_pgs_munlocked",
1345 "unevictable_pgs_cleared",
1346 "unevictable_pgs_stranded",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001347
1348#ifdef CONFIG_TRANSPARENT_HUGEPAGE
1349 "thp_fault_alloc",
1350 "thp_fault_fallback",
David Rientjes85b9f462020-04-06 20:04:28 -07001351 "thp_fault_fallback_charge",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001352 "thp_collapse_alloc",
1353 "thp_collapse_alloc_failed",
Kirill A. Shutemov95ecedc2016-07-26 15:25:31 -07001354 "thp_file_alloc",
David Rientjesdcdf11e2020-04-06 20:04:25 -07001355 "thp_file_fallback",
David Rientjes85b9f462020-04-06 20:04:28 -07001356 "thp_file_fallback_charge",
Kirill A. Shutemov95ecedc2016-07-26 15:25:31 -07001357 "thp_file_mapped",
Kirill A. Shutemov122afea2016-01-15 16:52:46 -08001358 "thp_split_page",
1359 "thp_split_page_failed",
Kirill A. Shutemovf9719a02016-03-17 14:18:45 -07001360 "thp_deferred_split_page",
Kirill A. Shutemov122afea2016-01-15 16:52:46 -08001361 "thp_split_pmd",
Yang Yange9ea8742022-01-14 14:07:55 -08001362 "thp_scan_exceed_none_pte",
1363 "thp_scan_exceed_swap_pte",
1364 "thp_scan_exceed_share_pte",
Yisheng Xiece9311c2017-03-09 16:17:00 -08001365#ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
1366 "thp_split_pud",
1367#endif
Kirill A. Shutemovd8a8e1f2012-12-12 13:51:09 -08001368 "thp_zero_page_alloc",
1369 "thp_zero_page_alloc_failed",
Huang Ying225311a2017-09-06 16:22:30 -07001370 "thp_swpout",
Huang Yingfe490cc2017-09-06 16:22:52 -07001371 "thp_swpout_fallback",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001372#endif
Konstantin Khlebnikov09316c02014-10-09 15:29:32 -07001373#ifdef CONFIG_MEMORY_BALLOON
1374 "balloon_inflate",
1375 "balloon_deflate",
1376#ifdef CONFIG_BALLOON_COMPACTION
1377 "balloon_migrate",
1378#endif
1379#endif /* CONFIG_MEMORY_BALLOON */
Mel Gormanec659932014-01-21 14:33:16 -08001380#ifdef CONFIG_DEBUG_TLBFLUSH
Dave Hansen9824cf92013-09-11 14:20:23 -07001381 "nr_tlb_remote_flush",
1382 "nr_tlb_remote_flush_received",
1383 "nr_tlb_local_flush_all",
1384 "nr_tlb_local_flush_one",
Mel Gormanec659932014-01-21 14:33:16 -08001385#endif /* CONFIG_DEBUG_TLBFLUSH */
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001386
Huang Yingcbc65df2017-09-06 16:24:29 -07001387#ifdef CONFIG_SWAP
1388 "swap_ra",
1389 "swap_ra_hit",
Yang Yang4d45c3a2022-03-22 14:46:33 -07001390#ifdef CONFIG_KSM
1391 "ksm_swpin_copy",
1392#endif
Huang Yingcbc65df2017-09-06 16:24:29 -07001393#endif
Yang Yang94bfe852022-04-28 23:16:16 -07001394#ifdef CONFIG_KSM
1395 "cow_ksm",
1396#endif
Johannes Weinerf6498b72022-05-19 14:08:53 -07001397#ifdef CONFIG_ZSWAP
1398 "zswpin",
1399 "zswpout",
1400#endif
Saravanan D575299e2021-05-04 18:38:03 -07001401#ifdef CONFIG_X86
1402 "direct_map_level2_splits",
1403 "direct_map_level3_splits",
1404#endif
Suren Baghdasaryan52f23862023-02-27 09:36:27 -08001405#ifdef CONFIG_PER_VMA_LOCK_STATS
1406 "vma_lock_success",
1407 "vma_lock_abort",
1408 "vma_lock_retry",
1409 "vma_lock_miss",
1410#endif
Konstantin Khlebnikovebc5d83d2019-12-04 16:49:53 -08001411#endif /* CONFIG_VM_EVENT_COUNTERS || CONFIG_MEMCG */
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001412};
Konstantin Khlebnikovebc5d83d2019-12-04 16:49:53 -08001413#endif /* CONFIG_PROC_FS || CONFIG_SYSFS || CONFIG_NUMA || CONFIG_MEMCG */
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001414
Andrew Morton3c486872015-02-10 14:09:43 -08001415#if (defined(CONFIG_DEBUG_FS) && defined(CONFIG_COMPACTION)) || \
1416 defined(CONFIG_PROC_FS)
1417static void *frag_start(struct seq_file *m, loff_t *pos)
1418{
1419 pg_data_t *pgdat;
1420 loff_t node = *pos;
1421
1422 for (pgdat = first_online_pgdat();
1423 pgdat && node;
1424 pgdat = next_online_pgdat(pgdat))
1425 --node;
1426
1427 return pgdat;
1428}
1429
1430static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
1431{
1432 pg_data_t *pgdat = (pg_data_t *)arg;
1433
1434 (*pos)++;
1435 return next_online_pgdat(pgdat);
1436}
1437
1438static void frag_stop(struct seq_file *m, void *arg)
1439{
1440}
1441
David Rientjesb2bd8592017-05-03 14:52:59 -07001442/*
1443 * Walk zones in a node and print using a callback.
1444 * If @assert_populated is true, only use callback for zones that are populated.
1445 */
Andrew Morton3c486872015-02-10 14:09:43 -08001446static void walk_zones_in_node(struct seq_file *m, pg_data_t *pgdat,
Vinayak Menon727c0802017-07-10 15:49:17 -07001447 bool assert_populated, bool nolock,
Andrew Morton3c486872015-02-10 14:09:43 -08001448 void (*print)(struct seq_file *m, pg_data_t *, struct zone *))
1449{
1450 struct zone *zone;
1451 struct zone *node_zones = pgdat->node_zones;
1452 unsigned long flags;
1453
1454 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
David Rientjesb2bd8592017-05-03 14:52:59 -07001455 if (assert_populated && !populated_zone(zone))
Andrew Morton3c486872015-02-10 14:09:43 -08001456 continue;
1457
Vinayak Menon727c0802017-07-10 15:49:17 -07001458 if (!nolock)
1459 spin_lock_irqsave(&zone->lock, flags);
Andrew Morton3c486872015-02-10 14:09:43 -08001460 print(m, pgdat, zone);
Vinayak Menon727c0802017-07-10 15:49:17 -07001461 if (!nolock)
1462 spin_unlock_irqrestore(&zone->lock, flags);
Andrew Morton3c486872015-02-10 14:09:43 -08001463 }
1464}
1465#endif
1466
Mel Gormand7a57522010-05-24 14:32:25 -07001467#ifdef CONFIG_PROC_FS
Mel Gorman467c9962007-10-16 01:26:02 -07001468static void frag_show_print(struct seq_file *m, pg_data_t *pgdat,
1469 struct zone *zone)
1470{
1471 int order;
1472
1473 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
Kirill A. Shutemov23baf832023-03-15 14:31:33 +03001474 for (order = 0; order <= MAX_ORDER; ++order)
Liu Shixinaf1c31a2021-11-05 13:43:59 -07001475 /*
1476 * Access to nr_free is lockless as nr_free is used only for
1477 * printing purposes. Use data_race to avoid KCSAN warning.
1478 */
1479 seq_printf(m, "%6lu ", data_race(zone->free_area[order].nr_free));
Mel Gorman467c9962007-10-16 01:26:02 -07001480 seq_putc(m, '\n');
1481}
1482
1483/*
1484 * This walks the free areas for each zone.
1485 */
1486static int frag_show(struct seq_file *m, void *arg)
1487{
1488 pg_data_t *pgdat = (pg_data_t *)arg;
Vinayak Menon727c0802017-07-10 15:49:17 -07001489 walk_zones_in_node(m, pgdat, true, false, frag_show_print);
Mel Gorman467c9962007-10-16 01:26:02 -07001490 return 0;
1491}
1492
1493static void pagetypeinfo_showfree_print(struct seq_file *m,
1494 pg_data_t *pgdat, struct zone *zone)
1495{
1496 int order, mtype;
1497
1498 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++) {
1499 seq_printf(m, "Node %4d, zone %8s, type %12s ",
1500 pgdat->node_id,
1501 zone->name,
1502 migratetype_names[mtype]);
Kirill A. Shutemov23baf832023-03-15 14:31:33 +03001503 for (order = 0; order <= MAX_ORDER; ++order) {
Mel Gorman467c9962007-10-16 01:26:02 -07001504 unsigned long freecount = 0;
1505 struct free_area *area;
1506 struct list_head *curr;
Michal Hocko93b3a672019-11-05 21:16:44 -08001507 bool overflow = false;
Mel Gorman467c9962007-10-16 01:26:02 -07001508
1509 area = &(zone->free_area[order]);
1510
Michal Hocko93b3a672019-11-05 21:16:44 -08001511 list_for_each(curr, &area->free_list[mtype]) {
1512 /*
1513 * Cap the free_list iteration because it might
1514 * be really large and we are under a spinlock
1515 * so a long time spent here could trigger a
1516 * hard lockup detector. Anyway this is a
1517 * debugging tool so knowing there is a handful
1518 * of pages of this order should be more than
1519 * sufficient.
1520 */
1521 if (++freecount >= 100000) {
1522 overflow = true;
1523 break;
1524 }
1525 }
1526 seq_printf(m, "%s%6lu ", overflow ? ">" : "", freecount);
1527 spin_unlock_irq(&zone->lock);
1528 cond_resched();
1529 spin_lock_irq(&zone->lock);
Mel Gorman467c9962007-10-16 01:26:02 -07001530 }
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001531 seq_putc(m, '\n');
1532 }
Mel Gorman467c9962007-10-16 01:26:02 -07001533}
1534
1535/* Print out the free pages at each order for each migatetype */
Miaohe Lin33090af2021-09-02 15:01:08 -07001536static void pagetypeinfo_showfree(struct seq_file *m, void *arg)
Mel Gorman467c9962007-10-16 01:26:02 -07001537{
1538 int order;
1539 pg_data_t *pgdat = (pg_data_t *)arg;
1540
1541 /* Print header */
1542 seq_printf(m, "%-43s ", "Free pages count per migrate type at order");
Kirill A. Shutemov23baf832023-03-15 14:31:33 +03001543 for (order = 0; order <= MAX_ORDER; ++order)
Mel Gorman467c9962007-10-16 01:26:02 -07001544 seq_printf(m, "%6d ", order);
1545 seq_putc(m, '\n');
1546
Vinayak Menon727c0802017-07-10 15:49:17 -07001547 walk_zones_in_node(m, pgdat, true, false, pagetypeinfo_showfree_print);
Mel Gorman467c9962007-10-16 01:26:02 -07001548}
1549
1550static void pagetypeinfo_showblockcount_print(struct seq_file *m,
1551 pg_data_t *pgdat, struct zone *zone)
1552{
1553 int mtype;
1554 unsigned long pfn;
1555 unsigned long start_pfn = zone->zone_start_pfn;
Cody P Schafer108bcc92013-02-22 16:35:23 -08001556 unsigned long end_pfn = zone_end_pfn(zone);
Mel Gorman467c9962007-10-16 01:26:02 -07001557 unsigned long count[MIGRATE_TYPES] = { 0, };
1558
1559 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
1560 struct page *page;
1561
Michal Hockod336e942017-07-06 15:38:07 -07001562 page = pfn_to_online_page(pfn);
1563 if (!page)
Mel Gorman467c9962007-10-16 01:26:02 -07001564 continue;
1565
Joonsoo Kima91c43c2016-05-19 17:12:10 -07001566 if (page_zone(page) != zone)
1567 continue;
1568
Mel Gorman467c9962007-10-16 01:26:02 -07001569 mtype = get_pageblock_migratetype(page);
1570
Mel Gormane80d6a22008-08-14 11:10:14 +01001571 if (mtype < MIGRATE_TYPES)
1572 count[mtype]++;
Mel Gorman467c9962007-10-16 01:26:02 -07001573 }
1574
1575 /* Print counts */
1576 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
1577 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
1578 seq_printf(m, "%12lu ", count[mtype]);
1579 seq_putc(m, '\n');
1580}
1581
SeongJae Parkf113e642017-09-06 16:24:23 -07001582/* Print out the number of pageblocks for each migratetype */
Miaohe Lin33090af2021-09-02 15:01:08 -07001583static void pagetypeinfo_showblockcount(struct seq_file *m, void *arg)
Mel Gorman467c9962007-10-16 01:26:02 -07001584{
1585 int mtype;
1586 pg_data_t *pgdat = (pg_data_t *)arg;
1587
1588 seq_printf(m, "\n%-23s", "Number of blocks type ");
1589 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
1590 seq_printf(m, "%12s ", migratetype_names[mtype]);
1591 seq_putc(m, '\n');
Vinayak Menon727c0802017-07-10 15:49:17 -07001592 walk_zones_in_node(m, pgdat, true, false,
1593 pagetypeinfo_showblockcount_print);
Mel Gorman467c9962007-10-16 01:26:02 -07001594}
1595
Joonsoo Kim48c96a32014-12-12 16:56:01 -08001596/*
1597 * Print out the number of pageblocks for each migratetype that contain pages
1598 * of other types. This gives an indication of how well fallbacks are being
1599 * contained by rmqueue_fallback(). It requires information from PAGE_OWNER
1600 * to determine what is going on
1601 */
1602static void pagetypeinfo_showmixedcount(struct seq_file *m, pg_data_t *pgdat)
1603{
1604#ifdef CONFIG_PAGE_OWNER
1605 int mtype;
1606
Vlastimil Babka7dd80b82016-03-15 14:56:12 -07001607 if (!static_branch_unlikely(&page_owner_inited))
Joonsoo Kim48c96a32014-12-12 16:56:01 -08001608 return;
1609
1610 drain_all_pages(NULL);
1611
1612 seq_printf(m, "\n%-23s", "Number of mixed blocks ");
1613 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
1614 seq_printf(m, "%12s ", migratetype_names[mtype]);
1615 seq_putc(m, '\n');
1616
Vinayak Menon727c0802017-07-10 15:49:17 -07001617 walk_zones_in_node(m, pgdat, true, true,
1618 pagetypeinfo_showmixedcount_print);
Joonsoo Kim48c96a32014-12-12 16:56:01 -08001619#endif /* CONFIG_PAGE_OWNER */
1620}
1621
Mel Gorman467c9962007-10-16 01:26:02 -07001622/*
1623 * This prints out statistics in relation to grouping pages by mobility.
1624 * It is expensive to collect so do not constantly read the file.
1625 */
1626static int pagetypeinfo_show(struct seq_file *m, void *arg)
1627{
1628 pg_data_t *pgdat = (pg_data_t *)arg;
1629
KOSAKI Motohiro41b25a32008-04-30 00:52:13 -07001630 /* check memoryless node */
Lai Jiangshana47b53c2012-12-12 13:51:37 -08001631 if (!node_state(pgdat->node_id, N_MEMORY))
KOSAKI Motohiro41b25a32008-04-30 00:52:13 -07001632 return 0;
1633
Mel Gorman467c9962007-10-16 01:26:02 -07001634 seq_printf(m, "Page block order: %d\n", pageblock_order);
1635 seq_printf(m, "Pages per block: %lu\n", pageblock_nr_pages);
1636 seq_putc(m, '\n');
1637 pagetypeinfo_showfree(m, pgdat);
1638 pagetypeinfo_showblockcount(m, pgdat);
Joonsoo Kim48c96a32014-12-12 16:56:01 -08001639 pagetypeinfo_showmixedcount(m, pgdat);
Mel Gorman467c9962007-10-16 01:26:02 -07001640
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001641 return 0;
1642}
1643
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +04001644static const struct seq_operations fragmentation_op = {
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001645 .start = frag_start,
1646 .next = frag_next,
1647 .stop = frag_stop,
1648 .show = frag_show,
1649};
1650
Alexey Dobriyan74e2e8e2008-10-06 04:15:36 +04001651static const struct seq_operations pagetypeinfo_op = {
Mel Gorman467c9962007-10-16 01:26:02 -07001652 .start = frag_start,
1653 .next = frag_next,
1654 .stop = frag_stop,
1655 .show = pagetypeinfo_show,
1656};
1657
Mel Gormane2ecc8a2016-07-28 15:47:02 -07001658static bool is_zone_first_populated(pg_data_t *pgdat, struct zone *zone)
1659{
1660 int zid;
1661
1662 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
1663 struct zone *compare = &pgdat->node_zones[zid];
1664
1665 if (populated_zone(compare))
1666 return zone == compare;
1667 }
1668
Mel Gormane2ecc8a2016-07-28 15:47:02 -07001669 return false;
1670}
1671
Mel Gorman467c9962007-10-16 01:26:02 -07001672static void zoneinfo_show_print(struct seq_file *m, pg_data_t *pgdat,
1673 struct zone *zone)
1674{
1675 int i;
1676 seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
Mel Gormane2ecc8a2016-07-28 15:47:02 -07001677 if (is_zone_first_populated(pgdat, zone)) {
1678 seq_printf(m, "\n per-node stats");
1679 for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) {
Muchun Song69473e52021-02-24 12:03:23 -08001680 unsigned long pages = node_page_state_pages(pgdat, i);
1681
1682 if (vmstat_item_print_in_thp(i))
1683 pages /= HPAGE_PMD_NR;
Konstantin Khlebnikov9d7ea9a2019-12-04 16:49:50 -08001684 seq_printf(m, "\n %-12s %lu", node_stat_name(i),
Muchun Song69473e52021-02-24 12:03:23 -08001685 pages);
Mel Gormane2ecc8a2016-07-28 15:47:02 -07001686 }
1687 }
Mel Gorman467c9962007-10-16 01:26:02 -07001688 seq_printf(m,
1689 "\n pages free %lu"
Liangcai Fana6ea8b52021-11-05 13:40:37 -07001690 "\n boost %lu"
Mel Gorman467c9962007-10-16 01:26:02 -07001691 "\n min %lu"
1692 "\n low %lu"
1693 "\n high %lu"
Mel Gorman467c9962007-10-16 01:26:02 -07001694 "\n spanned %lu"
Jiang Liu9feedc92012-12-12 13:52:12 -08001695 "\n present %lu"
David Hildenbrand3c381db2021-02-25 17:16:40 -08001696 "\n managed %lu"
1697 "\n cma %lu",
Mel Gorman88f5acf2011-01-13 15:45:41 -08001698 zone_page_state(zone, NR_FREE_PAGES),
Liangcai Fana6ea8b52021-11-05 13:40:37 -07001699 zone->watermark_boost,
Mel Gorman41858962009-06-16 15:32:12 -07001700 min_wmark_pages(zone),
1701 low_wmark_pages(zone),
1702 high_wmark_pages(zone),
Mel Gorman467c9962007-10-16 01:26:02 -07001703 zone->spanned_pages,
Jiang Liu9feedc92012-12-12 13:52:12 -08001704 zone->present_pages,
David Hildenbrand3c381db2021-02-25 17:16:40 -08001705 zone_managed_pages(zone),
1706 zone_cma_pages(zone));
Mel Gorman467c9962007-10-16 01:26:02 -07001707
Mel Gorman467c9962007-10-16 01:26:02 -07001708 seq_printf(m,
Mel Gorman3484b2d2014-08-06 16:07:14 -07001709 "\n protection: (%ld",
Mel Gorman467c9962007-10-16 01:26:02 -07001710 zone->lowmem_reserve[0]);
1711 for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
Mel Gorman3484b2d2014-08-06 16:07:14 -07001712 seq_printf(m, ", %ld", zone->lowmem_reserve[i]);
David Rientjes7dfb8bf2017-05-03 14:53:02 -07001713 seq_putc(m, ')');
1714
Baoquan Hea8a4b7a2020-08-14 17:30:07 -07001715 /* If unpopulated, no other information is useful */
1716 if (!populated_zone(zone)) {
1717 seq_putc(m, '\n');
1718 return;
1719 }
1720
David Rientjes7dfb8bf2017-05-03 14:53:02 -07001721 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
Konstantin Khlebnikov9d7ea9a2019-12-04 16:49:50 -08001722 seq_printf(m, "\n %-12s %lu", zone_stat_name(i),
1723 zone_page_state(zone, i));
David Rientjes7dfb8bf2017-05-03 14:53:02 -07001724
Kemi Wang3a321d22017-09-08 16:12:48 -07001725#ifdef CONFIG_NUMA
Mel Gormanf19298b2021-06-28 19:41:44 -07001726 for (i = 0; i < NR_VM_NUMA_EVENT_ITEMS; i++)
Konstantin Khlebnikov9d7ea9a2019-12-04 16:49:50 -08001727 seq_printf(m, "\n %-12s %lu", numa_stat_name(i),
Mel Gormanf19298b2021-06-28 19:41:44 -07001728 zone_numa_event_state(zone, i));
Kemi Wang3a321d22017-09-08 16:12:48 -07001729#endif
1730
David Rientjes7dfb8bf2017-05-03 14:53:02 -07001731 seq_printf(m, "\n pagesets");
Mel Gorman467c9962007-10-16 01:26:02 -07001732 for_each_online_cpu(i) {
Mel Gorman28f836b2021-06-28 19:41:38 -07001733 struct per_cpu_pages *pcp;
1734 struct per_cpu_zonestat __maybe_unused *pzstats;
Mel Gorman467c9962007-10-16 01:26:02 -07001735
Mel Gorman28f836b2021-06-28 19:41:38 -07001736 pcp = per_cpu_ptr(zone->per_cpu_pageset, i);
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001737 seq_printf(m,
1738 "\n cpu: %i"
1739 "\n count: %i"
1740 "\n high: %i"
1741 "\n batch: %i",
1742 i,
Mel Gorman28f836b2021-06-28 19:41:38 -07001743 pcp->count,
1744 pcp->high,
1745 pcp->batch);
Mel Gorman467c9962007-10-16 01:26:02 -07001746#ifdef CONFIG_SMP
Mel Gorman28f836b2021-06-28 19:41:38 -07001747 pzstats = per_cpu_ptr(zone->per_cpu_zonestats, i);
Mel Gorman467c9962007-10-16 01:26:02 -07001748 seq_printf(m, "\n vm stats threshold: %d",
Mel Gorman28f836b2021-06-28 19:41:38 -07001749 pzstats->stat_threshold);
Mel Gorman467c9962007-10-16 01:26:02 -07001750#endif
1751 }
1752 seq_printf(m,
Mel Gorman599d0c92016-07-28 15:45:31 -07001753 "\n node_unreclaimable: %u"
Andrey Ryabinin3a50d142017-11-15 17:34:15 -08001754 "\n start_pfn: %lu",
Johannes Weinerc73322d2017-05-03 14:51:51 -07001755 pgdat->kswapd_failures >= MAX_RECLAIM_RETRIES,
Andrey Ryabinin3a50d142017-11-15 17:34:15 -08001756 zone->zone_start_pfn);
Mel Gorman467c9962007-10-16 01:26:02 -07001757 seq_putc(m, '\n');
1758}
1759
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001760/*
David Rientjesb2bd8592017-05-03 14:52:59 -07001761 * Output information about zones in @pgdat. All zones are printed regardless
1762 * of whether they are populated or not: lowmem_reserve_ratio operates on the
1763 * set of all zones and userspace would not be aware of such zones if they are
1764 * suppressed here (zoneinfo displays the effect of lowmem_reserve_ratio).
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001765 */
1766static int zoneinfo_show(struct seq_file *m, void *arg)
1767{
Mel Gorman467c9962007-10-16 01:26:02 -07001768 pg_data_t *pgdat = (pg_data_t *)arg;
Vinayak Menon727c0802017-07-10 15:49:17 -07001769 walk_zones_in_node(m, pgdat, false, false, zoneinfo_show_print);
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001770 return 0;
1771}
1772
Alexey Dobriyan5c9fe622008-10-06 04:19:42 +04001773static const struct seq_operations zoneinfo_op = {
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001774 .start = frag_start, /* iterate over all zones. The same as in
1775 * fragmentation. */
1776 .next = frag_next,
1777 .stop = frag_stop,
1778 .show = zoneinfo_show,
1779};
1780
Konstantin Khlebnikov9d7ea9a2019-12-04 16:49:50 -08001781#define NR_VMSTAT_ITEMS (NR_VM_ZONE_STAT_ITEMS + \
Mel Gormanf19298b2021-06-28 19:41:44 -07001782 NR_VM_NUMA_EVENT_ITEMS + \
Konstantin Khlebnikov9d7ea9a2019-12-04 16:49:50 -08001783 NR_VM_NODE_STAT_ITEMS + \
1784 NR_VM_WRITEBACK_STAT_ITEMS + \
1785 (IS_ENABLED(CONFIG_VM_EVENT_COUNTERS) ? \
1786 NR_VM_EVENT_ITEMS : 0))
Michael Rubin79da8262010-10-26 14:21:36 -07001787
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001788static void *vmstat_start(struct seq_file *m, loff_t *pos)
1789{
Christoph Lameter2244b952006-06-30 01:55:33 -07001790 unsigned long *v;
Konstantin Khlebnikov9d7ea9a2019-12-04 16:49:50 -08001791 int i;
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001792
Konstantin Khlebnikov9d7ea9a2019-12-04 16:49:50 -08001793 if (*pos >= NR_VMSTAT_ITEMS)
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001794 return NULL;
1795
Konstantin Khlebnikov9d7ea9a2019-12-04 16:49:50 -08001796 BUILD_BUG_ON(ARRAY_SIZE(vmstat_text) < NR_VMSTAT_ITEMS);
Mel Gormanf19298b2021-06-28 19:41:44 -07001797 fold_vm_numa_events();
Konstantin Khlebnikov9d7ea9a2019-12-04 16:49:50 -08001798 v = kmalloc_array(NR_VMSTAT_ITEMS, sizeof(unsigned long), GFP_KERNEL);
Christoph Lameter2244b952006-06-30 01:55:33 -07001799 m->private = v;
1800 if (!v)
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001801 return ERR_PTR(-ENOMEM);
Christoph Lameter2244b952006-06-30 01:55:33 -07001802 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
Michal Hockoc41f0122017-09-06 16:23:36 -07001803 v[i] = global_zone_page_state(i);
Michael Rubin79da8262010-10-26 14:21:36 -07001804 v += NR_VM_ZONE_STAT_ITEMS;
1805
Kemi Wang3a321d22017-09-08 16:12:48 -07001806#ifdef CONFIG_NUMA
Mel Gormanf19298b2021-06-28 19:41:44 -07001807 for (i = 0; i < NR_VM_NUMA_EVENT_ITEMS; i++)
1808 v[i] = global_numa_event_state(i);
1809 v += NR_VM_NUMA_EVENT_ITEMS;
Kemi Wang3a321d22017-09-08 16:12:48 -07001810#endif
1811
Muchun Song69473e52021-02-24 12:03:23 -08001812 for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) {
Roman Gushchinea426c22020-08-06 23:20:35 -07001813 v[i] = global_node_page_state_pages(i);
Muchun Song69473e52021-02-24 12:03:23 -08001814 if (vmstat_item_print_in_thp(i))
1815 v[i] /= HPAGE_PMD_NR;
1816 }
Mel Gorman75ef7182016-07-28 15:45:24 -07001817 v += NR_VM_NODE_STAT_ITEMS;
1818
Michael Rubin79da8262010-10-26 14:21:36 -07001819 global_dirty_limits(v + NR_DIRTY_BG_THRESHOLD,
1820 v + NR_DIRTY_THRESHOLD);
1821 v += NR_VM_WRITEBACK_STAT_ITEMS;
1822
Christoph Lameterf8891e52006-06-30 01:55:45 -07001823#ifdef CONFIG_VM_EVENT_COUNTERS
Michael Rubin79da8262010-10-26 14:21:36 -07001824 all_vm_events(v);
1825 v[PGPGIN] /= 2; /* sectors -> kbytes */
1826 v[PGPGOUT] /= 2;
Christoph Lameterf8891e52006-06-30 01:55:45 -07001827#endif
Wu Fengguangff8b16d2010-11-04 01:56:49 +08001828 return (unsigned long *)m->private + *pos;
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001829}
1830
1831static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
1832{
1833 (*pos)++;
Konstantin Khlebnikov9d7ea9a2019-12-04 16:49:50 -08001834 if (*pos >= NR_VMSTAT_ITEMS)
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001835 return NULL;
1836 return (unsigned long *)m->private + *pos;
1837}
1838
1839static int vmstat_show(struct seq_file *m, void *arg)
1840{
1841 unsigned long *l = arg;
1842 unsigned long off = l - (unsigned long *)m->private;
Alexey Dobriyan68ba0322016-10-07 17:02:14 -07001843
1844 seq_puts(m, vmstat_text[off]);
Joe Perches75ba1d02016-10-07 17:02:20 -07001845 seq_put_decimal_ull(m, " ", *l);
Alexey Dobriyan68ba0322016-10-07 17:02:14 -07001846 seq_putc(m, '\n');
NeilBrown8d928902020-06-01 21:48:21 -07001847
1848 if (off == NR_VMSTAT_ITEMS - 1) {
1849 /*
1850 * We've come to the end - add any deprecated counters to avoid
1851 * breaking userspace which might depend on them being present.
1852 */
1853 seq_puts(m, "nr_unstable 0\n");
1854 }
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001855 return 0;
1856}
1857
1858static void vmstat_stop(struct seq_file *m, void *arg)
1859{
1860 kfree(m->private);
1861 m->private = NULL;
1862}
1863
Alexey Dobriyanb6aa44a2008-10-06 04:17:48 +04001864static const struct seq_operations vmstat_op = {
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001865 .start = vmstat_start,
1866 .next = vmstat_next,
1867 .stop = vmstat_stop,
1868 .show = vmstat_show,
1869};
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001870#endif /* CONFIG_PROC_FS */
1871
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001872#ifdef CONFIG_SMP
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001873static DEFINE_PER_CPU(struct delayed_work, vmstat_work);
Christoph Lameter77461ab2007-05-09 02:35:13 -07001874int sysctl_stat_interval __read_mostly = HZ;
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001875
Hugh Dickins52b6f462016-05-19 17:12:50 -07001876#ifdef CONFIG_PROC_FS
1877static void refresh_vm_stats(struct work_struct *work)
1878{
1879 refresh_cpu_vm_stats(true);
1880}
1881
1882int vmstat_refresh(struct ctl_table *table, int write,
Christoph Hellwig32927392020-04-24 08:43:38 +02001883 void *buffer, size_t *lenp, loff_t *ppos)
Hugh Dickins52b6f462016-05-19 17:12:50 -07001884{
1885 long val;
1886 int err;
1887 int i;
1888
1889 /*
1890 * The regular update, every sysctl_stat_interval, may come later
1891 * than expected: leaving a significant amount in per_cpu buckets.
1892 * This is particularly misleading when checking a quantity of HUGE
1893 * pages, immediately after running a test. /proc/sys/vm/stat_refresh,
1894 * which can equally be echo'ed to or cat'ted from (by root),
1895 * can be used to update the stats just before reading them.
1896 *
Michal Hockoc41f0122017-09-06 16:23:36 -07001897 * Oh, and since global_zone_page_state() etc. are so careful to hide
Hugh Dickins52b6f462016-05-19 17:12:50 -07001898 * transiently negative values, report an error here if any of
1899 * the stats is negative, so we know to go looking for imbalance.
1900 */
1901 err = schedule_on_each_cpu(refresh_vm_stats);
1902 if (err)
1903 return err;
1904 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) {
Hugh Dickins75083aa2021-05-04 18:37:57 -07001905 /*
1906 * Skip checking stats known to go negative occasionally.
1907 */
1908 switch (i) {
1909 case NR_ZONE_WRITE_PENDING:
1910 case NR_FREE_CMA_PAGES:
1911 continue;
1912 }
Mel Gorman75ef7182016-07-28 15:45:24 -07001913 val = atomic_long_read(&vm_zone_stat[i]);
Hugh Dickins52b6f462016-05-19 17:12:50 -07001914 if (val < 0) {
Johannes Weinerc822f622017-05-03 14:52:10 -07001915 pr_warn("%s: %s %ld\n",
Konstantin Khlebnikov9d7ea9a2019-12-04 16:49:50 -08001916 __func__, zone_stat_name(i), val);
Hugh Dickins52b6f462016-05-19 17:12:50 -07001917 }
1918 }
Hugh Dickins76d8cc32021-05-04 18:37:51 -07001919 for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) {
Hugh Dickins75083aa2021-05-04 18:37:57 -07001920 /*
1921 * Skip checking stats known to go negative occasionally.
1922 */
1923 switch (i) {
1924 case NR_WRITEBACK:
1925 continue;
1926 }
Hugh Dickins76d8cc32021-05-04 18:37:51 -07001927 val = atomic_long_read(&vm_node_stat[i]);
1928 if (val < 0) {
1929 pr_warn("%s: %s %ld\n",
1930 __func__, node_stat_name(i), val);
Hugh Dickins76d8cc32021-05-04 18:37:51 -07001931 }
1932 }
Hugh Dickins52b6f462016-05-19 17:12:50 -07001933 if (write)
1934 *ppos += *lenp;
1935 else
1936 *lenp = 0;
1937 return 0;
1938}
1939#endif /* CONFIG_PROC_FS */
1940
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001941static void vmstat_update(struct work_struct *w)
1942{
Christoph Lameter0eb77e92016-01-14 15:21:40 -08001943 if (refresh_cpu_vm_stats(true)) {
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001944 /*
1945 * Counters were updated so we expect more updates
1946 * to occur in the future. Keep on running the
1947 * update worker thread.
1948 */
Michal Hockoce612872017-04-07 16:05:05 -07001949 queue_delayed_work_on(smp_processor_id(), mm_percpu_wq,
Michal Hockof01f17d2016-02-05 15:36:24 -08001950 this_cpu_ptr(&vmstat_work),
1951 round_jiffies_relative(sysctl_stat_interval));
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001952 }
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001953}
1954
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001955/*
1956 * Check if the diffs for a certain cpu indicate that
1957 * an update is needed.
1958 */
1959static bool need_update(int cpu)
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001960{
Johannes Weiner2bbd00a2021-02-25 17:16:47 -08001961 pg_data_t *last_pgdat = NULL;
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001962 struct zone *zone;
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001963
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001964 for_each_populated_zone(zone) {
Mel Gorman28f836b2021-06-28 19:41:38 -07001965 struct per_cpu_zonestat *pzstats = per_cpu_ptr(zone->per_cpu_zonestats, cpu);
Johannes Weiner2bbd00a2021-02-25 17:16:47 -08001966 struct per_cpu_nodestat *n;
Mel Gorman28f836b2021-06-28 19:41:38 -07001967
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001968 /*
1969 * The fast way of checking if there are any vmstat diffs.
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001970 */
Miaohe Lin64632fd2021-09-02 15:01:05 -07001971 if (memchr_inv(pzstats->vm_stat_diff, 0, sizeof(pzstats->vm_stat_diff)))
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001972 return true;
Mel Gormanf19298b2021-06-28 19:41:44 -07001973
Johannes Weiner2bbd00a2021-02-25 17:16:47 -08001974 if (last_pgdat == zone->zone_pgdat)
1975 continue;
1976 last_pgdat = zone->zone_pgdat;
1977 n = per_cpu_ptr(zone->zone_pgdat->per_cpu_nodestats, cpu);
Miaohe Lin64632fd2021-09-02 15:01:05 -07001978 if (memchr_inv(n->vm_node_stat_diff, 0, sizeof(n->vm_node_stat_diff)))
1979 return true;
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001980 }
1981 return false;
1982}
1983
Christoph Lameter7b8da4c2016-05-20 16:58:21 -07001984/*
1985 * Switch off vmstat processing and then fold all the remaining differentials
1986 * until the diffs stay at zero. The function is used by NOHZ and can only be
1987 * invoked when tick processing is not active.
1988 */
Michal Hockof01f17d2016-02-05 15:36:24 -08001989void quiet_vmstat(void)
1990{
1991 if (system_state != SYSTEM_RUNNING)
1992 return;
1993
Christoph Lameter7b8da4c2016-05-20 16:58:21 -07001994 if (!delayed_work_pending(this_cpu_ptr(&vmstat_work)))
Michal Hockof01f17d2016-02-05 15:36:24 -08001995 return;
1996
1997 if (!need_update(smp_processor_id()))
1998 return;
1999
2000 /*
2001 * Just refresh counters and do not care about the pending delayed
2002 * vmstat_update. It doesn't fire that often to matter and canceling
2003 * it would be too expensive from this path.
2004 * vmstat_shepherd will take care about that for us.
2005 */
2006 refresh_cpu_vm_stats(false);
2007}
2008
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07002009/*
2010 * Shepherd worker thread that checks the
2011 * differentials of processors that have their worker
2012 * threads for vm statistics updates disabled because of
2013 * inactivity.
2014 */
2015static void vmstat_shepherd(struct work_struct *w);
2016
Christoph Lameter0eb77e92016-01-14 15:21:40 -08002017static DECLARE_DEFERRABLE_WORK(shepherd, vmstat_shepherd);
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07002018
2019static void vmstat_shepherd(struct work_struct *w)
2020{
2021 int cpu;
2022
Sebastian Andrzej Siewior7625ecc2021-08-03 16:16:03 +02002023 cpus_read_lock();
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07002024 /* Check processors whose vmstat worker threads have been disabled */
Christoph Lameter7b8da4c2016-05-20 16:58:21 -07002025 for_each_online_cpu(cpu) {
Michal Hockof01f17d2016-02-05 15:36:24 -08002026 struct delayed_work *dw = &per_cpu(vmstat_work, cpu);
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07002027
Marcelo Tosattibe5e0152023-06-07 17:28:07 -03002028 /*
2029 * In kernel users of vmstat counters either require the precise value and
2030 * they are using zone_page_state_snapshot interface or they can live with
2031 * an imprecision as the regular flushing can happen at arbitrary time and
2032 * cumulative error can grow (see calculate_normal_threshold).
2033 *
2034 * From that POV the regular flushing can be postponed for CPUs that have
2035 * been isolated from the kernel interference without critical
2036 * infrastructure ever noticing. Skip regular flushing from vmstat_shepherd
2037 * for all isolated CPUs to avoid interference with the isolated workload.
2038 */
2039 if (cpu_is_isolated(cpu))
2040 continue;
2041
Christoph Lameter7b8da4c2016-05-20 16:58:21 -07002042 if (!delayed_work_pending(dw) && need_update(cpu))
Michal Hockoce612872017-04-07 16:05:05 -07002043 queue_delayed_work_on(cpu, mm_percpu_wq, dw, 0);
Jiang Biaofbcc8182021-02-25 17:16:54 -08002044
2045 cond_resched();
Michal Hockof01f17d2016-02-05 15:36:24 -08002046 }
Sebastian Andrzej Siewior7625ecc2021-08-03 16:16:03 +02002047 cpus_read_unlock();
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07002048
2049 schedule_delayed_work(&shepherd,
2050 round_jiffies_relative(sysctl_stat_interval));
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07002051}
2052
2053static void __init start_shepherd_timer(void)
2054{
2055 int cpu;
2056
2057 for_each_possible_cpu(cpu)
Michal Hockoccde8bd2016-02-05 15:36:27 -08002058 INIT_DEFERRABLE_WORK(per_cpu_ptr(&vmstat_work, cpu),
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07002059 vmstat_update);
2060
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07002061 schedule_delayed_work(&shepherd,
2062 round_jiffies_relative(sysctl_stat_interval));
Christoph Lameterd1187ed2007-05-09 02:35:12 -07002063}
2064
Tim Chen03e86db2016-10-07 17:00:02 -07002065static void __init init_cpu_node_state(void)
2066{
Sebastian Andrzej Siewior4c501322016-11-29 15:51:14 +01002067 int node;
Tim Chen03e86db2016-10-07 17:00:02 -07002068
Sebastian Andrzej Siewior4c501322016-11-29 15:51:14 +01002069 for_each_online_node(node) {
Yury Norovb55032f2022-01-23 10:38:56 -08002070 if (!cpumask_empty(cpumask_of_node(node)))
Sebastian Andrzej Siewior4c501322016-11-29 15:51:14 +01002071 node_set_state(node, N_CPU);
2072 }
Tim Chen03e86db2016-10-07 17:00:02 -07002073}
2074
Sebastian Andrzej Siewior5438da92016-11-29 15:52:21 +01002075static int vmstat_cpu_online(unsigned int cpu)
2076{
2077 refresh_zone_stat_thresholds();
Oscar Salvador734c1572022-03-22 14:47:37 -07002078
2079 if (!node_state(cpu_to_node(cpu), N_CPU)) {
2080 node_set_state(cpu_to_node(cpu), N_CPU);
Oscar Salvador734c1572022-03-22 14:47:37 -07002081 }
2082
Sebastian Andrzej Siewior5438da92016-11-29 15:52:21 +01002083 return 0;
2084}
2085
2086static int vmstat_cpu_down_prep(unsigned int cpu)
2087{
2088 cancel_delayed_work_sync(&per_cpu(vmstat_work, cpu));
2089 return 0;
2090}
2091
2092static int vmstat_cpu_dead(unsigned int cpu)
Toshi Kani807a1bd2013-11-12 15:08:13 -08002093{
Sebastian Andrzej Siewior4c501322016-11-29 15:51:14 +01002094 const struct cpumask *node_cpus;
Sebastian Andrzej Siewior5438da92016-11-29 15:52:21 +01002095 int node;
Toshi Kani807a1bd2013-11-12 15:08:13 -08002096
Sebastian Andrzej Siewior5438da92016-11-29 15:52:21 +01002097 node = cpu_to_node(cpu);
2098
2099 refresh_zone_stat_thresholds();
Sebastian Andrzej Siewior4c501322016-11-29 15:51:14 +01002100 node_cpus = cpumask_of_node(node);
Yury Norovb55032f2022-01-23 10:38:56 -08002101 if (!cpumask_empty(node_cpus))
Sebastian Andrzej Siewior5438da92016-11-29 15:52:21 +01002102 return 0;
Toshi Kani807a1bd2013-11-12 15:08:13 -08002103
2104 node_clear_state(node, N_CPU);
Oscar Salvador734c1572022-03-22 14:47:37 -07002105
Sebastian Andrzej Siewior5438da92016-11-29 15:52:21 +01002106 return 0;
Toshi Kani807a1bd2013-11-12 15:08:13 -08002107}
2108
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +04002109#endif
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07002110
Michal Hockoce612872017-04-07 16:05:05 -07002111struct workqueue_struct *mm_percpu_wq;
2112
Michal Hocko597b7302017-03-31 15:11:47 -07002113void __init init_mm_internals(void)
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07002114{
Michal Hockoce612872017-04-07 16:05:05 -07002115 int ret __maybe_unused;
Sebastian Andrzej Siewior5438da92016-11-29 15:52:21 +01002116
Michal Hocko80d136e2017-04-19 09:52:46 +02002117 mm_percpu_wq = alloc_workqueue("mm_percpu_wq", WQ_MEM_RECLAIM, 0);
Michal Hockoce612872017-04-07 16:05:05 -07002118
2119#ifdef CONFIG_SMP
Sebastian Andrzej Siewior5438da92016-11-29 15:52:21 +01002120 ret = cpuhp_setup_state_nocalls(CPUHP_MM_VMSTAT_DEAD, "mm/vmstat:dead",
2121 NULL, vmstat_cpu_dead);
2122 if (ret < 0)
2123 pr_err("vmstat: failed to register 'dead' hotplug state\n");
2124
2125 ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "mm/vmstat:online",
2126 vmstat_cpu_online,
2127 vmstat_cpu_down_prep);
2128 if (ret < 0)
2129 pr_err("vmstat: failed to register 'online' hotplug state\n");
2130
Sebastian Andrzej Siewior7625ecc2021-08-03 16:16:03 +02002131 cpus_read_lock();
Tim Chen03e86db2016-10-07 17:00:02 -07002132 init_cpu_node_state();
Sebastian Andrzej Siewior7625ecc2021-08-03 16:16:03 +02002133 cpus_read_unlock();
Christoph Lameterd1187ed2007-05-09 02:35:12 -07002134
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07002135 start_shepherd_timer();
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +04002136#endif
2137#ifdef CONFIG_PROC_FS
Christoph Hellwigfddda2b2018-04-13 19:44:18 +02002138 proc_create_seq("buddyinfo", 0444, NULL, &fragmentation_op);
Michal Hockoabaed012019-11-05 21:16:40 -08002139 proc_create_seq("pagetypeinfo", 0400, NULL, &pagetypeinfo_op);
Christoph Hellwigfddda2b2018-04-13 19:44:18 +02002140 proc_create_seq("vmstat", 0444, NULL, &vmstat_op);
2141 proc_create_seq("zoneinfo", 0444, NULL, &zoneinfo_op);
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +04002142#endif
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07002143}
Mel Gormand7a57522010-05-24 14:32:25 -07002144
2145#if defined(CONFIG_DEBUG_FS) && defined(CONFIG_COMPACTION)
Mel Gormand7a57522010-05-24 14:32:25 -07002146
2147/*
2148 * Return an index indicating how much of the available free memory is
2149 * unusable for an allocation of the requested size.
2150 */
2151static int unusable_free_index(unsigned int order,
2152 struct contig_page_info *info)
2153{
2154 /* No free memory is interpreted as all free memory is unusable */
2155 if (info->free_pages == 0)
2156 return 1000;
2157
2158 /*
2159 * Index should be a value between 0 and 1. Return a value to 3
2160 * decimal places.
2161 *
2162 * 0 => no fragmentation
2163 * 1 => high fragmentation
2164 */
2165 return div_u64((info->free_pages - (info->free_blocks_suitable << order)) * 1000ULL, info->free_pages);
2166
2167}
2168
2169static void unusable_show_print(struct seq_file *m,
2170 pg_data_t *pgdat, struct zone *zone)
2171{
2172 unsigned int order;
2173 int index;
2174 struct contig_page_info info;
2175
2176 seq_printf(m, "Node %d, zone %8s ",
2177 pgdat->node_id,
2178 zone->name);
Kirill A. Shutemov23baf832023-03-15 14:31:33 +03002179 for (order = 0; order <= MAX_ORDER; ++order) {
Mel Gormand7a57522010-05-24 14:32:25 -07002180 fill_contig_page_info(zone, order, &info);
2181 index = unusable_free_index(order, &info);
2182 seq_printf(m, "%d.%03d ", index / 1000, index % 1000);
2183 }
2184
2185 seq_putc(m, '\n');
2186}
2187
2188/*
2189 * Display unusable free space index
2190 *
2191 * The unusable free space index measures how much of the available free
2192 * memory cannot be used to satisfy an allocation of a given size and is a
2193 * value between 0 and 1. The higher the value, the more of free memory is
2194 * unusable and by implication, the worse the external fragmentation is. This
2195 * can be expressed as a percentage by multiplying by 100.
2196 */
2197static int unusable_show(struct seq_file *m, void *arg)
2198{
2199 pg_data_t *pgdat = (pg_data_t *)arg;
2200
2201 /* check memoryless node */
Lai Jiangshana47b53c2012-12-12 13:51:37 -08002202 if (!node_state(pgdat->node_id, N_MEMORY))
Mel Gormand7a57522010-05-24 14:32:25 -07002203 return 0;
2204
Vinayak Menon727c0802017-07-10 15:49:17 -07002205 walk_zones_in_node(m, pgdat, true, false, unusable_show_print);
Mel Gormand7a57522010-05-24 14:32:25 -07002206
2207 return 0;
2208}
2209
Kefeng Wang01a99562020-06-04 16:51:08 -07002210static const struct seq_operations unusable_sops = {
Mel Gormand7a57522010-05-24 14:32:25 -07002211 .start = frag_start,
2212 .next = frag_next,
2213 .stop = frag_stop,
2214 .show = unusable_show,
2215};
2216
Kefeng Wang01a99562020-06-04 16:51:08 -07002217DEFINE_SEQ_ATTRIBUTE(unusable);
Mel Gormand7a57522010-05-24 14:32:25 -07002218
Mel Gormanf1a5ab12010-05-24 14:32:26 -07002219static void extfrag_show_print(struct seq_file *m,
2220 pg_data_t *pgdat, struct zone *zone)
2221{
2222 unsigned int order;
2223 int index;
2224
2225 /* Alloc on stack as interrupts are disabled for zone walk */
2226 struct contig_page_info info;
2227
2228 seq_printf(m, "Node %d, zone %8s ",
2229 pgdat->node_id,
2230 zone->name);
Kirill A. Shutemov23baf832023-03-15 14:31:33 +03002231 for (order = 0; order <= MAX_ORDER; ++order) {
Mel Gormanf1a5ab12010-05-24 14:32:26 -07002232 fill_contig_page_info(zone, order, &info);
Mel Gorman56de7262010-05-24 14:32:30 -07002233 index = __fragmentation_index(order, &info);
Lin Fenga9970582021-11-05 13:44:02 -07002234 seq_printf(m, "%2d.%03d ", index / 1000, index % 1000);
Mel Gormanf1a5ab12010-05-24 14:32:26 -07002235 }
2236
2237 seq_putc(m, '\n');
2238}
2239
2240/*
2241 * Display fragmentation index for orders that allocations would fail for
2242 */
2243static int extfrag_show(struct seq_file *m, void *arg)
2244{
2245 pg_data_t *pgdat = (pg_data_t *)arg;
2246
Vinayak Menon727c0802017-07-10 15:49:17 -07002247 walk_zones_in_node(m, pgdat, true, false, extfrag_show_print);
Mel Gormanf1a5ab12010-05-24 14:32:26 -07002248
2249 return 0;
2250}
2251
Kefeng Wang01a99562020-06-04 16:51:08 -07002252static const struct seq_operations extfrag_sops = {
Mel Gormanf1a5ab12010-05-24 14:32:26 -07002253 .start = frag_start,
2254 .next = frag_next,
2255 .stop = frag_stop,
2256 .show = extfrag_show,
2257};
2258
Kefeng Wang01a99562020-06-04 16:51:08 -07002259DEFINE_SEQ_ATTRIBUTE(extfrag);
Mel Gormanf1a5ab12010-05-24 14:32:26 -07002260
Mel Gormand7a57522010-05-24 14:32:25 -07002261static int __init extfrag_debug_init(void)
2262{
Sasikantha babubde8bd82012-05-29 15:06:22 -07002263 struct dentry *extfrag_debug_root;
2264
Mel Gormand7a57522010-05-24 14:32:25 -07002265 extfrag_debug_root = debugfs_create_dir("extfrag", NULL);
Mel Gormand7a57522010-05-24 14:32:25 -07002266
Greg Kroah-Hartmand9f79792019-03-05 15:46:09 -08002267 debugfs_create_file("unusable_index", 0444, extfrag_debug_root, NULL,
Kefeng Wang01a99562020-06-04 16:51:08 -07002268 &unusable_fops);
Mel Gormand7a57522010-05-24 14:32:25 -07002269
Greg Kroah-Hartmand9f79792019-03-05 15:46:09 -08002270 debugfs_create_file("extfrag_index", 0444, extfrag_debug_root, NULL,
Kefeng Wang01a99562020-06-04 16:51:08 -07002271 &extfrag_fops);
Mel Gormanf1a5ab12010-05-24 14:32:26 -07002272
Mel Gormand7a57522010-05-24 14:32:25 -07002273 return 0;
2274}
2275
2276module_init(extfrag_debug_init);
2277#endif