blob: 9a4441bbeef26d7bf39fe8bb9148a7e8a4ef5632 [file] [log] [blame]
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001/*
2 * linux/mm/vmstat.c
3 *
4 * Manages VM statistics
5 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
Christoph Lameter2244b952006-06-30 01:55:33 -07006 *
7 * zoned VM statistics
8 * Copyright (C) 2006 Silicon Graphics, Inc.,
9 * Christoph Lameter <christoph@lameter.com>
Christoph Lameter7cc36bb2014-10-09 15:29:43 -070010 * Copyright (C) 2008-2014 Christoph Lameter
Christoph Lameterf6ac2352006-06-30 01:55:32 -070011 */
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +040012#include <linux/fs.h>
Christoph Lameterf6ac2352006-06-30 01:55:32 -070013#include <linux/mm.h>
Alexey Dobriyan4e950f62007-07-30 02:36:13 +040014#include <linux/err.h>
Christoph Lameter2244b952006-06-30 01:55:33 -070015#include <linux/module.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090016#include <linux/slab.h>
Christoph Lameterdf9ecab2006-08-31 21:27:35 -070017#include <linux/cpu.h>
Christoph Lameter7cc36bb2014-10-09 15:29:43 -070018#include <linux/cpumask.h>
Adrian Bunkc748e132008-07-23 21:27:03 -070019#include <linux/vmstat.h>
Andrew Morton3c486872015-02-10 14:09:43 -080020#include <linux/proc_fs.h>
21#include <linux/seq_file.h>
22#include <linux/debugfs.h>
Alexey Dobriyane8edc6e2007-05-21 01:22:52 +040023#include <linux/sched.h>
Mel Gormanf1a5ab12010-05-24 14:32:26 -070024#include <linux/math64.h>
Michael Rubin79da8262010-10-26 14:21:36 -070025#include <linux/writeback.h>
Namhyung Kim36deb0b2010-10-26 14:22:04 -070026#include <linux/compaction.h>
Lisa Du6e543d52013-09-11 14:22:36 -070027#include <linux/mm_inline.h>
Joonsoo Kim48c96a32014-12-12 16:56:01 -080028#include <linux/page_ext.h>
29#include <linux/page_owner.h>
Lisa Du6e543d52013-09-11 14:22:36 -070030
31#include "internal.h"
Christoph Lameterf6ac2352006-06-30 01:55:32 -070032
Christoph Lameterf8891e52006-06-30 01:55:45 -070033#ifdef CONFIG_VM_EVENT_COUNTERS
34DEFINE_PER_CPU(struct vm_event_state, vm_event_states) = {{0}};
35EXPORT_PER_CPU_SYMBOL(vm_event_states);
36
Minchan Kim31f961a2010-08-09 17:18:59 -070037static void sum_vm_events(unsigned long *ret)
Christoph Lameterf8891e52006-06-30 01:55:45 -070038{
Christoph Lameter9eccf2a2008-02-04 22:29:22 -080039 int cpu;
Christoph Lameterf8891e52006-06-30 01:55:45 -070040 int i;
41
42 memset(ret, 0, NR_VM_EVENT_ITEMS * sizeof(unsigned long));
43
Minchan Kim31f961a2010-08-09 17:18:59 -070044 for_each_online_cpu(cpu) {
Christoph Lameterf8891e52006-06-30 01:55:45 -070045 struct vm_event_state *this = &per_cpu(vm_event_states, cpu);
46
Christoph Lameterf8891e52006-06-30 01:55:45 -070047 for (i = 0; i < NR_VM_EVENT_ITEMS; i++)
48 ret[i] += this->event[i];
49 }
50}
51
52/*
53 * Accumulate the vm event counters across all CPUs.
54 * The result is unavoidably approximate - it can change
55 * during and after execution of this function.
56*/
57void all_vm_events(unsigned long *ret)
58{
KOSAKI Motohirob5be1132008-05-12 14:02:06 -070059 get_online_cpus();
Minchan Kim31f961a2010-08-09 17:18:59 -070060 sum_vm_events(ret);
KOSAKI Motohirob5be1132008-05-12 14:02:06 -070061 put_online_cpus();
Christoph Lameterf8891e52006-06-30 01:55:45 -070062}
Heiko Carstens32dd66f2006-07-10 04:44:31 -070063EXPORT_SYMBOL_GPL(all_vm_events);
Christoph Lameterf8891e52006-06-30 01:55:45 -070064
Christoph Lameterf8891e52006-06-30 01:55:45 -070065/*
66 * Fold the foreign cpu events into our own.
67 *
68 * This is adding to the events on one processor
69 * but keeps the global counts constant.
70 */
71void vm_events_fold_cpu(int cpu)
72{
73 struct vm_event_state *fold_state = &per_cpu(vm_event_states, cpu);
74 int i;
75
76 for (i = 0; i < NR_VM_EVENT_ITEMS; i++) {
77 count_vm_events(i, fold_state->event[i]);
78 fold_state->event[i] = 0;
79 }
80}
Christoph Lameterf8891e52006-06-30 01:55:45 -070081
82#endif /* CONFIG_VM_EVENT_COUNTERS */
83
Christoph Lameter2244b952006-06-30 01:55:33 -070084/*
85 * Manage combined zone based / global counters
86 *
87 * vm_stat contains the global counters
88 */
Mel Gorman75ef7182016-07-28 15:45:24 -070089atomic_long_t vm_zone_stat[NR_VM_ZONE_STAT_ITEMS] __cacheline_aligned_in_smp;
90atomic_long_t vm_node_stat[NR_VM_NODE_STAT_ITEMS] __cacheline_aligned_in_smp;
91EXPORT_SYMBOL(vm_zone_stat);
92EXPORT_SYMBOL(vm_node_stat);
Christoph Lameter2244b952006-06-30 01:55:33 -070093
94#ifdef CONFIG_SMP
95
Mel Gormanb44129b2011-01-13 15:45:43 -080096int calculate_pressure_threshold(struct zone *zone)
Mel Gorman88f5acf2011-01-13 15:45:41 -080097{
98 int threshold;
99 int watermark_distance;
100
101 /*
102 * As vmstats are not up to date, there is drift between the estimated
103 * and real values. For high thresholds and a high number of CPUs, it
104 * is possible for the min watermark to be breached while the estimated
105 * value looks fine. The pressure threshold is a reduced value such
106 * that even the maximum amount of drift will not accidentally breach
107 * the min watermark
108 */
109 watermark_distance = low_wmark_pages(zone) - min_wmark_pages(zone);
110 threshold = max(1, (int)(watermark_distance / num_online_cpus()));
111
112 /*
113 * Maximum threshold is 125
114 */
115 threshold = min(125, threshold);
116
117 return threshold;
118}
119
Mel Gormanb44129b2011-01-13 15:45:43 -0800120int calculate_normal_threshold(struct zone *zone)
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700121{
122 int threshold;
123 int mem; /* memory in 128 MB units */
124
125 /*
126 * The threshold scales with the number of processors and the amount
127 * of memory per zone. More memory means that we can defer updates for
128 * longer, more processors could lead to more contention.
129 * fls() is used to have a cheap way of logarithmic scaling.
130 *
131 * Some sample thresholds:
132 *
133 * Threshold Processors (fls) Zonesize fls(mem+1)
134 * ------------------------------------------------------------------
135 * 8 1 1 0.9-1 GB 4
136 * 16 2 2 0.9-1 GB 4
137 * 20 2 2 1-2 GB 5
138 * 24 2 2 2-4 GB 6
139 * 28 2 2 4-8 GB 7
140 * 32 2 2 8-16 GB 8
141 * 4 2 2 <128M 1
142 * 30 4 3 2-4 GB 5
143 * 48 4 3 8-16 GB 8
144 * 32 8 4 1-2 GB 4
145 * 32 8 4 0.9-1GB 4
146 * 10 16 5 <128M 1
147 * 40 16 5 900M 4
148 * 70 64 7 2-4 GB 5
149 * 84 64 7 4-8 GB 6
150 * 108 512 9 4-8 GB 6
151 * 125 1024 10 8-16 GB 8
152 * 125 1024 10 16-32 GB 9
153 */
154
Jiang Liub40da042013-02-22 16:33:52 -0800155 mem = zone->managed_pages >> (27 - PAGE_SHIFT);
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700156
157 threshold = 2 * fls(num_online_cpus()) * (1 + fls(mem));
158
159 /*
160 * Maximum threshold is 125
161 */
162 threshold = min(125, threshold);
163
164 return threshold;
165}
Christoph Lameter2244b952006-06-30 01:55:33 -0700166
167/*
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700168 * Refresh the thresholds for each zone.
Christoph Lameter2244b952006-06-30 01:55:33 -0700169 */
KOSAKI Motohiroa6cccdc2011-05-24 17:11:33 -0700170void refresh_zone_stat_thresholds(void)
Christoph Lameter2244b952006-06-30 01:55:33 -0700171{
Mel Gorman75ef7182016-07-28 15:45:24 -0700172 struct pglist_data *pgdat;
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700173 struct zone *zone;
174 int cpu;
175 int threshold;
176
Mel Gorman75ef7182016-07-28 15:45:24 -0700177 /* Zero current pgdat thresholds */
178 for_each_online_pgdat(pgdat) {
179 for_each_online_cpu(cpu) {
180 per_cpu_ptr(pgdat->per_cpu_nodestats, cpu)->stat_threshold = 0;
181 }
182 }
183
KOSAKI Motohiroee99c712009-03-31 15:19:31 -0700184 for_each_populated_zone(zone) {
Mel Gorman75ef7182016-07-28 15:45:24 -0700185 struct pglist_data *pgdat = zone->zone_pgdat;
Christoph Lameteraa454842010-09-09 16:38:17 -0700186 unsigned long max_drift, tolerate_drift;
187
Mel Gormanb44129b2011-01-13 15:45:43 -0800188 threshold = calculate_normal_threshold(zone);
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700189
Mel Gorman75ef7182016-07-28 15:45:24 -0700190 for_each_online_cpu(cpu) {
191 int pgdat_threshold;
192
Christoph Lameter99dcc3e2010-01-05 15:34:51 +0900193 per_cpu_ptr(zone->pageset, cpu)->stat_threshold
194 = threshold;
Christoph Lameteraa454842010-09-09 16:38:17 -0700195
Mel Gorman75ef7182016-07-28 15:45:24 -0700196 /* Base nodestat threshold on the largest populated zone. */
197 pgdat_threshold = per_cpu_ptr(pgdat->per_cpu_nodestats, cpu)->stat_threshold;
198 per_cpu_ptr(pgdat->per_cpu_nodestats, cpu)->stat_threshold
199 = max(threshold, pgdat_threshold);
200 }
201
Christoph Lameteraa454842010-09-09 16:38:17 -0700202 /*
203 * Only set percpu_drift_mark if there is a danger that
204 * NR_FREE_PAGES reports the low watermark is ok when in fact
205 * the min watermark could be breached by an allocation
206 */
207 tolerate_drift = low_wmark_pages(zone) - min_wmark_pages(zone);
208 max_drift = num_online_cpus() * threshold;
209 if (max_drift > tolerate_drift)
210 zone->percpu_drift_mark = high_wmark_pages(zone) +
211 max_drift;
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700212 }
Christoph Lameter2244b952006-06-30 01:55:33 -0700213}
214
Mel Gormanb44129b2011-01-13 15:45:43 -0800215void set_pgdat_percpu_threshold(pg_data_t *pgdat,
216 int (*calculate_pressure)(struct zone *))
Mel Gorman88f5acf2011-01-13 15:45:41 -0800217{
218 struct zone *zone;
219 int cpu;
220 int threshold;
221 int i;
222
Mel Gorman88f5acf2011-01-13 15:45:41 -0800223 for (i = 0; i < pgdat->nr_zones; i++) {
224 zone = &pgdat->node_zones[i];
225 if (!zone->percpu_drift_mark)
226 continue;
227
Mel Gormanb44129b2011-01-13 15:45:43 -0800228 threshold = (*calculate_pressure)(zone);
Mel Gormanbb0b6df2014-08-06 16:07:18 -0700229 for_each_online_cpu(cpu)
Mel Gorman88f5acf2011-01-13 15:45:41 -0800230 per_cpu_ptr(zone->pageset, cpu)->stat_threshold
231 = threshold;
232 }
Mel Gorman88f5acf2011-01-13 15:45:41 -0800233}
234
Christoph Lameter2244b952006-06-30 01:55:33 -0700235/*
Jianyu Zhanbea04b02014-06-04 16:09:51 -0700236 * For use when we know that interrupts are disabled,
237 * or when we know that preemption is disabled and that
238 * particular counter cannot be updated from interrupt context.
Christoph Lameter2244b952006-06-30 01:55:33 -0700239 */
240void __mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
Heiko Carstens6cdb18a2015-12-29 14:54:32 -0800241 long delta)
Christoph Lameter2244b952006-06-30 01:55:33 -0700242{
Christoph Lameter12938a92010-12-06 11:16:20 -0600243 struct per_cpu_pageset __percpu *pcp = zone->pageset;
244 s8 __percpu *p = pcp->vm_stat_diff + item;
Christoph Lameter2244b952006-06-30 01:55:33 -0700245 long x;
Christoph Lameter12938a92010-12-06 11:16:20 -0600246 long t;
Christoph Lameter2244b952006-06-30 01:55:33 -0700247
Christoph Lameter12938a92010-12-06 11:16:20 -0600248 x = delta + __this_cpu_read(*p);
Christoph Lameter2244b952006-06-30 01:55:33 -0700249
Christoph Lameter12938a92010-12-06 11:16:20 -0600250 t = __this_cpu_read(pcp->stat_threshold);
251
252 if (unlikely(x > t || x < -t)) {
Christoph Lameter2244b952006-06-30 01:55:33 -0700253 zone_page_state_add(x, zone, item);
254 x = 0;
255 }
Christoph Lameter12938a92010-12-06 11:16:20 -0600256 __this_cpu_write(*p, x);
Christoph Lameter2244b952006-06-30 01:55:33 -0700257}
258EXPORT_SYMBOL(__mod_zone_page_state);
259
Mel Gorman75ef7182016-07-28 15:45:24 -0700260void __mod_node_page_state(struct pglist_data *pgdat, enum node_stat_item item,
261 long delta)
262{
263 struct per_cpu_nodestat __percpu *pcp = pgdat->per_cpu_nodestats;
264 s8 __percpu *p = pcp->vm_node_stat_diff + item;
265 long x;
266 long t;
267
268 x = delta + __this_cpu_read(*p);
269
270 t = __this_cpu_read(pcp->stat_threshold);
271
272 if (unlikely(x > t || x < -t)) {
273 node_page_state_add(x, pgdat, item);
274 x = 0;
275 }
276 __this_cpu_write(*p, x);
277}
278EXPORT_SYMBOL(__mod_node_page_state);
279
Christoph Lameter2244b952006-06-30 01:55:33 -0700280/*
Christoph Lameter2244b952006-06-30 01:55:33 -0700281 * Optimized increment and decrement functions.
282 *
283 * These are only for a single page and therefore can take a struct page *
284 * argument instead of struct zone *. This allows the inclusion of the code
285 * generated for page_zone(page) into the optimized functions.
286 *
287 * No overflow check is necessary and therefore the differential can be
288 * incremented or decremented in place which may allow the compilers to
289 * generate better code.
Christoph Lameter2244b952006-06-30 01:55:33 -0700290 * The increment or decrement is known and therefore one boundary check can
291 * be omitted.
292 *
Christoph Lameterdf9ecab2006-08-31 21:27:35 -0700293 * NOTE: These functions are very performance sensitive. Change only
294 * with care.
295 *
Christoph Lameter2244b952006-06-30 01:55:33 -0700296 * Some processors have inc/dec instructions that are atomic vs an interrupt.
297 * However, the code must first determine the differential location in a zone
298 * based on the processor number and then inc/dec the counter. There is no
299 * guarantee without disabling preemption that the processor will not change
300 * in between and therefore the atomicity vs. interrupt cannot be exploited
301 * in a useful way here.
302 */
Christoph Lameterc8785382007-02-10 01:43:01 -0800303void __inc_zone_state(struct zone *zone, enum zone_stat_item item)
Christoph Lameter2244b952006-06-30 01:55:33 -0700304{
Christoph Lameter12938a92010-12-06 11:16:20 -0600305 struct per_cpu_pageset __percpu *pcp = zone->pageset;
306 s8 __percpu *p = pcp->vm_stat_diff + item;
307 s8 v, t;
Christoph Lameter2244b952006-06-30 01:55:33 -0700308
Christoph Lameter908ee0f2010-12-06 11:40:02 -0600309 v = __this_cpu_inc_return(*p);
Christoph Lameter12938a92010-12-06 11:16:20 -0600310 t = __this_cpu_read(pcp->stat_threshold);
311 if (unlikely(v > t)) {
312 s8 overstep = t >> 1;
Christoph Lameter2244b952006-06-30 01:55:33 -0700313
Christoph Lameter12938a92010-12-06 11:16:20 -0600314 zone_page_state_add(v + overstep, zone, item);
315 __this_cpu_write(*p, -overstep);
Christoph Lameter2244b952006-06-30 01:55:33 -0700316 }
317}
Christoph Lameterca889e62006-06-30 01:55:44 -0700318
Mel Gorman75ef7182016-07-28 15:45:24 -0700319void __inc_node_state(struct pglist_data *pgdat, enum node_stat_item item)
320{
321 struct per_cpu_nodestat __percpu *pcp = pgdat->per_cpu_nodestats;
322 s8 __percpu *p = pcp->vm_node_stat_diff + item;
323 s8 v, t;
324
325 v = __this_cpu_inc_return(*p);
326 t = __this_cpu_read(pcp->stat_threshold);
327 if (unlikely(v > t)) {
328 s8 overstep = t >> 1;
329
330 node_page_state_add(v + overstep, pgdat, item);
331 __this_cpu_write(*p, -overstep);
332 }
333}
334
Christoph Lameterca889e62006-06-30 01:55:44 -0700335void __inc_zone_page_state(struct page *page, enum zone_stat_item item)
336{
337 __inc_zone_state(page_zone(page), item);
338}
Christoph Lameter2244b952006-06-30 01:55:33 -0700339EXPORT_SYMBOL(__inc_zone_page_state);
340
Mel Gorman75ef7182016-07-28 15:45:24 -0700341void __inc_node_page_state(struct page *page, enum node_stat_item item)
342{
343 __inc_node_state(page_pgdat(page), item);
344}
345EXPORT_SYMBOL(__inc_node_page_state);
346
Christoph Lameterc8785382007-02-10 01:43:01 -0800347void __dec_zone_state(struct zone *zone, enum zone_stat_item item)
Christoph Lameter2244b952006-06-30 01:55:33 -0700348{
Christoph Lameter12938a92010-12-06 11:16:20 -0600349 struct per_cpu_pageset __percpu *pcp = zone->pageset;
350 s8 __percpu *p = pcp->vm_stat_diff + item;
351 s8 v, t;
Christoph Lameter2244b952006-06-30 01:55:33 -0700352
Christoph Lameter908ee0f2010-12-06 11:40:02 -0600353 v = __this_cpu_dec_return(*p);
Christoph Lameter12938a92010-12-06 11:16:20 -0600354 t = __this_cpu_read(pcp->stat_threshold);
355 if (unlikely(v < - t)) {
356 s8 overstep = t >> 1;
Christoph Lameter2244b952006-06-30 01:55:33 -0700357
Christoph Lameter12938a92010-12-06 11:16:20 -0600358 zone_page_state_add(v - overstep, zone, item);
359 __this_cpu_write(*p, overstep);
Christoph Lameter2244b952006-06-30 01:55:33 -0700360 }
361}
Christoph Lameterc8785382007-02-10 01:43:01 -0800362
Mel Gorman75ef7182016-07-28 15:45:24 -0700363void __dec_node_state(struct pglist_data *pgdat, enum node_stat_item item)
364{
365 struct per_cpu_nodestat __percpu *pcp = pgdat->per_cpu_nodestats;
366 s8 __percpu *p = pcp->vm_node_stat_diff + item;
367 s8 v, t;
368
369 v = __this_cpu_dec_return(*p);
370 t = __this_cpu_read(pcp->stat_threshold);
371 if (unlikely(v < - t)) {
372 s8 overstep = t >> 1;
373
374 node_page_state_add(v - overstep, pgdat, item);
375 __this_cpu_write(*p, overstep);
376 }
377}
378
Christoph Lameterc8785382007-02-10 01:43:01 -0800379void __dec_zone_page_state(struct page *page, enum zone_stat_item item)
380{
381 __dec_zone_state(page_zone(page), item);
382}
Christoph Lameter2244b952006-06-30 01:55:33 -0700383EXPORT_SYMBOL(__dec_zone_page_state);
384
Mel Gorman75ef7182016-07-28 15:45:24 -0700385void __dec_node_page_state(struct page *page, enum node_stat_item item)
386{
387 __dec_node_state(page_pgdat(page), item);
388}
389EXPORT_SYMBOL(__dec_node_page_state);
390
Heiko Carstens41561532012-01-12 17:17:30 -0800391#ifdef CONFIG_HAVE_CMPXCHG_LOCAL
Christoph Lameter7c839122010-12-14 10:28:46 -0600392/*
393 * If we have cmpxchg_local support then we do not need to incur the overhead
394 * that comes with local_irq_save/restore if we use this_cpu_cmpxchg.
395 *
396 * mod_state() modifies the zone counter state through atomic per cpu
397 * operations.
398 *
399 * Overstep mode specifies how overstep should handled:
400 * 0 No overstepping
401 * 1 Overstepping half of threshold
402 * -1 Overstepping minus half of threshold
403*/
Mel Gorman75ef7182016-07-28 15:45:24 -0700404static inline void mod_zone_state(struct zone *zone,
405 enum zone_stat_item item, long delta, int overstep_mode)
Christoph Lameter7c839122010-12-14 10:28:46 -0600406{
407 struct per_cpu_pageset __percpu *pcp = zone->pageset;
408 s8 __percpu *p = pcp->vm_stat_diff + item;
409 long o, n, t, z;
410
411 do {
412 z = 0; /* overflow to zone counters */
413
414 /*
415 * The fetching of the stat_threshold is racy. We may apply
416 * a counter threshold to the wrong the cpu if we get
Christoph Lameterd3bc2362011-04-14 15:21:58 -0700417 * rescheduled while executing here. However, the next
418 * counter update will apply the threshold again and
419 * therefore bring the counter under the threshold again.
420 *
421 * Most of the time the thresholds are the same anyways
422 * for all cpus in a zone.
Christoph Lameter7c839122010-12-14 10:28:46 -0600423 */
424 t = this_cpu_read(pcp->stat_threshold);
425
426 o = this_cpu_read(*p);
427 n = delta + o;
428
429 if (n > t || n < -t) {
430 int os = overstep_mode * (t >> 1) ;
431
432 /* Overflow must be added to zone counters */
433 z = n + os;
434 n = -os;
435 }
436 } while (this_cpu_cmpxchg(*p, o, n) != o);
437
438 if (z)
439 zone_page_state_add(z, zone, item);
440}
441
442void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
Heiko Carstens6cdb18a2015-12-29 14:54:32 -0800443 long delta)
Christoph Lameter7c839122010-12-14 10:28:46 -0600444{
Mel Gorman75ef7182016-07-28 15:45:24 -0700445 mod_zone_state(zone, item, delta, 0);
Christoph Lameter7c839122010-12-14 10:28:46 -0600446}
447EXPORT_SYMBOL(mod_zone_page_state);
448
Christoph Lameter7c839122010-12-14 10:28:46 -0600449void inc_zone_page_state(struct page *page, enum zone_stat_item item)
450{
Mel Gorman75ef7182016-07-28 15:45:24 -0700451 mod_zone_state(page_zone(page), item, 1, 1);
Christoph Lameter7c839122010-12-14 10:28:46 -0600452}
453EXPORT_SYMBOL(inc_zone_page_state);
454
455void dec_zone_page_state(struct page *page, enum zone_stat_item item)
456{
Mel Gorman75ef7182016-07-28 15:45:24 -0700457 mod_zone_state(page_zone(page), item, -1, -1);
Christoph Lameter7c839122010-12-14 10:28:46 -0600458}
459EXPORT_SYMBOL(dec_zone_page_state);
Mel Gorman75ef7182016-07-28 15:45:24 -0700460
461static inline void mod_node_state(struct pglist_data *pgdat,
462 enum node_stat_item item, int delta, int overstep_mode)
463{
464 struct per_cpu_nodestat __percpu *pcp = pgdat->per_cpu_nodestats;
465 s8 __percpu *p = pcp->vm_node_stat_diff + item;
466 long o, n, t, z;
467
468 do {
469 z = 0; /* overflow to node counters */
470
471 /*
472 * The fetching of the stat_threshold is racy. We may apply
473 * a counter threshold to the wrong the cpu if we get
474 * rescheduled while executing here. However, the next
475 * counter update will apply the threshold again and
476 * therefore bring the counter under the threshold again.
477 *
478 * Most of the time the thresholds are the same anyways
479 * for all cpus in a node.
480 */
481 t = this_cpu_read(pcp->stat_threshold);
482
483 o = this_cpu_read(*p);
484 n = delta + o;
485
486 if (n > t || n < -t) {
487 int os = overstep_mode * (t >> 1) ;
488
489 /* Overflow must be added to node counters */
490 z = n + os;
491 n = -os;
492 }
493 } while (this_cpu_cmpxchg(*p, o, n) != o);
494
495 if (z)
496 node_page_state_add(z, pgdat, item);
497}
498
499void mod_node_page_state(struct pglist_data *pgdat, enum node_stat_item item,
500 long delta)
501{
502 mod_node_state(pgdat, item, delta, 0);
503}
504EXPORT_SYMBOL(mod_node_page_state);
505
506void inc_node_state(struct pglist_data *pgdat, enum node_stat_item item)
507{
508 mod_node_state(pgdat, item, 1, 1);
509}
510
511void inc_node_page_state(struct page *page, enum node_stat_item item)
512{
513 mod_node_state(page_pgdat(page), item, 1, 1);
514}
515EXPORT_SYMBOL(inc_node_page_state);
516
517void dec_node_page_state(struct page *page, enum node_stat_item item)
518{
519 mod_node_state(page_pgdat(page), item, -1, -1);
520}
521EXPORT_SYMBOL(dec_node_page_state);
Christoph Lameter7c839122010-12-14 10:28:46 -0600522#else
523/*
524 * Use interrupt disable to serialize counter updates
525 */
526void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
Heiko Carstens6cdb18a2015-12-29 14:54:32 -0800527 long delta)
Christoph Lameter7c839122010-12-14 10:28:46 -0600528{
529 unsigned long flags;
530
531 local_irq_save(flags);
532 __mod_zone_page_state(zone, item, delta);
533 local_irq_restore(flags);
534}
535EXPORT_SYMBOL(mod_zone_page_state);
536
Christoph Lameter2244b952006-06-30 01:55:33 -0700537void inc_zone_page_state(struct page *page, enum zone_stat_item item)
538{
539 unsigned long flags;
540 struct zone *zone;
Christoph Lameter2244b952006-06-30 01:55:33 -0700541
542 zone = page_zone(page);
543 local_irq_save(flags);
Christoph Lameterca889e62006-06-30 01:55:44 -0700544 __inc_zone_state(zone, item);
Christoph Lameter2244b952006-06-30 01:55:33 -0700545 local_irq_restore(flags);
546}
547EXPORT_SYMBOL(inc_zone_page_state);
548
549void dec_zone_page_state(struct page *page, enum zone_stat_item item)
550{
551 unsigned long flags;
Christoph Lameter2244b952006-06-30 01:55:33 -0700552
Christoph Lameter2244b952006-06-30 01:55:33 -0700553 local_irq_save(flags);
Christoph Lametera302eb42006-08-31 21:27:34 -0700554 __dec_zone_page_state(page, item);
Christoph Lameter2244b952006-06-30 01:55:33 -0700555 local_irq_restore(flags);
556}
557EXPORT_SYMBOL(dec_zone_page_state);
558
Mel Gorman75ef7182016-07-28 15:45:24 -0700559void inc_node_state(struct pglist_data *pgdat, enum node_stat_item item)
560{
561 unsigned long flags;
562
563 local_irq_save(flags);
564 __inc_node_state(pgdat, item);
565 local_irq_restore(flags);
566}
567EXPORT_SYMBOL(inc_node_state);
568
569void mod_node_page_state(struct pglist_data *pgdat, enum node_stat_item item,
570 long delta)
571{
572 unsigned long flags;
573
574 local_irq_save(flags);
575 __mod_node_page_state(pgdat, item, delta);
576 local_irq_restore(flags);
577}
578EXPORT_SYMBOL(mod_node_page_state);
579
580void inc_node_page_state(struct page *page, enum node_stat_item item)
581{
582 unsigned long flags;
583 struct pglist_data *pgdat;
584
585 pgdat = page_pgdat(page);
586 local_irq_save(flags);
587 __inc_node_state(pgdat, item);
588 local_irq_restore(flags);
589}
590EXPORT_SYMBOL(inc_node_page_state);
591
592void dec_node_page_state(struct page *page, enum node_stat_item item)
593{
594 unsigned long flags;
595
596 local_irq_save(flags);
597 __dec_node_page_state(page, item);
598 local_irq_restore(flags);
599}
600EXPORT_SYMBOL(dec_node_page_state);
601#endif
Christoph Lameter7cc36bb2014-10-09 15:29:43 -0700602
603/*
604 * Fold a differential into the global counters.
605 * Returns the number of counters updated.
606 */
Mel Gorman75ef7182016-07-28 15:45:24 -0700607static int fold_diff(int *zone_diff, int *node_diff)
Christoph Lameter4edb0742013-09-11 14:21:31 -0700608{
609 int i;
Christoph Lameter7cc36bb2014-10-09 15:29:43 -0700610 int changes = 0;
Christoph Lameter4edb0742013-09-11 14:21:31 -0700611
612 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
Mel Gorman75ef7182016-07-28 15:45:24 -0700613 if (zone_diff[i]) {
614 atomic_long_add(zone_diff[i], &vm_zone_stat[i]);
615 changes++;
616 }
617
618 for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++)
619 if (node_diff[i]) {
620 atomic_long_add(node_diff[i], &vm_node_stat[i]);
Christoph Lameter7cc36bb2014-10-09 15:29:43 -0700621 changes++;
622 }
623 return changes;
Christoph Lameter4edb0742013-09-11 14:21:31 -0700624}
625
Christoph Lameter2244b952006-06-30 01:55:33 -0700626/*
Christoph Lameter2bb921e2013-09-11 14:21:30 -0700627 * Update the zone counters for the current cpu.
Christoph Lametera7f75e22008-02-04 22:29:16 -0800628 *
Christoph Lameter4037d452007-05-09 02:35:14 -0700629 * Note that refresh_cpu_vm_stats strives to only access
630 * node local memory. The per cpu pagesets on remote zones are placed
631 * in the memory local to the processor using that pageset. So the
632 * loop over all zones will access a series of cachelines local to
633 * the processor.
634 *
635 * The call to zone_page_state_add updates the cachelines with the
636 * statistics in the remote zone struct as well as the global cachelines
637 * with the global counters. These could cause remote node cache line
638 * bouncing and will have to be only done when necessary.
Christoph Lameter7cc36bb2014-10-09 15:29:43 -0700639 *
640 * The function returns the number of global counters updated.
Christoph Lameter2244b952006-06-30 01:55:33 -0700641 */
Christoph Lameter0eb77e92016-01-14 15:21:40 -0800642static int refresh_cpu_vm_stats(bool do_pagesets)
Christoph Lameter2244b952006-06-30 01:55:33 -0700643{
Mel Gorman75ef7182016-07-28 15:45:24 -0700644 struct pglist_data *pgdat;
Christoph Lameter2244b952006-06-30 01:55:33 -0700645 struct zone *zone;
646 int i;
Mel Gorman75ef7182016-07-28 15:45:24 -0700647 int global_zone_diff[NR_VM_ZONE_STAT_ITEMS] = { 0, };
648 int global_node_diff[NR_VM_NODE_STAT_ITEMS] = { 0, };
Christoph Lameter7cc36bb2014-10-09 15:29:43 -0700649 int changes = 0;
Christoph Lameter2244b952006-06-30 01:55:33 -0700650
KOSAKI Motohiroee99c712009-03-31 15:19:31 -0700651 for_each_populated_zone(zone) {
Christoph Lameterfbc2edb2013-09-11 14:21:32 -0700652 struct per_cpu_pageset __percpu *p = zone->pageset;
Christoph Lameter2244b952006-06-30 01:55:33 -0700653
Christoph Lameterfbc2edb2013-09-11 14:21:32 -0700654 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) {
655 int v;
Christoph Lameter2244b952006-06-30 01:55:33 -0700656
Christoph Lameterfbc2edb2013-09-11 14:21:32 -0700657 v = this_cpu_xchg(p->vm_stat_diff[i], 0);
658 if (v) {
Christoph Lametera7f75e22008-02-04 22:29:16 -0800659
Christoph Lametera7f75e22008-02-04 22:29:16 -0800660 atomic_long_add(v, &zone->vm_stat[i]);
Mel Gorman75ef7182016-07-28 15:45:24 -0700661 global_zone_diff[i] += v;
Christoph Lameter4037d452007-05-09 02:35:14 -0700662#ifdef CONFIG_NUMA
663 /* 3 seconds idle till flush */
Christoph Lameterfbc2edb2013-09-11 14:21:32 -0700664 __this_cpu_write(p->expire, 3);
Christoph Lameter4037d452007-05-09 02:35:14 -0700665#endif
Christoph Lameter2244b952006-06-30 01:55:33 -0700666 }
Christoph Lameterfbc2edb2013-09-11 14:21:32 -0700667 }
Christoph Lameter4037d452007-05-09 02:35:14 -0700668#ifdef CONFIG_NUMA
Christoph Lameter0eb77e92016-01-14 15:21:40 -0800669 if (do_pagesets) {
670 cond_resched();
671 /*
672 * Deal with draining the remote pageset of this
673 * processor
674 *
675 * Check if there are pages remaining in this pageset
676 * if not then there is nothing to expire.
677 */
678 if (!__this_cpu_read(p->expire) ||
Christoph Lameterfbc2edb2013-09-11 14:21:32 -0700679 !__this_cpu_read(p->pcp.count))
Christoph Lameter0eb77e92016-01-14 15:21:40 -0800680 continue;
Christoph Lameter4037d452007-05-09 02:35:14 -0700681
Christoph Lameter0eb77e92016-01-14 15:21:40 -0800682 /*
683 * We never drain zones local to this processor.
684 */
685 if (zone_to_nid(zone) == numa_node_id()) {
686 __this_cpu_write(p->expire, 0);
687 continue;
688 }
Christoph Lameter4037d452007-05-09 02:35:14 -0700689
Christoph Lameter0eb77e92016-01-14 15:21:40 -0800690 if (__this_cpu_dec_return(p->expire))
691 continue;
Christoph Lameter4037d452007-05-09 02:35:14 -0700692
Christoph Lameter0eb77e92016-01-14 15:21:40 -0800693 if (__this_cpu_read(p->pcp.count)) {
694 drain_zone_pages(zone, this_cpu_ptr(&p->pcp));
695 changes++;
696 }
Christoph Lameter7cc36bb2014-10-09 15:29:43 -0700697 }
Christoph Lameter4037d452007-05-09 02:35:14 -0700698#endif
Christoph Lameter2244b952006-06-30 01:55:33 -0700699 }
Mel Gorman75ef7182016-07-28 15:45:24 -0700700
701 for_each_online_pgdat(pgdat) {
702 struct per_cpu_nodestat __percpu *p = pgdat->per_cpu_nodestats;
703
704 for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) {
705 int v;
706
707 v = this_cpu_xchg(p->vm_node_stat_diff[i], 0);
708 if (v) {
709 atomic_long_add(v, &pgdat->vm_stat[i]);
710 global_node_diff[i] += v;
711 }
712 }
713 }
714
715 changes += fold_diff(global_zone_diff, global_node_diff);
Christoph Lameter7cc36bb2014-10-09 15:29:43 -0700716 return changes;
Christoph Lameter2244b952006-06-30 01:55:33 -0700717}
718
Cody P Schafer40f4b1e2013-04-29 15:08:38 -0700719/*
Christoph Lameter2bb921e2013-09-11 14:21:30 -0700720 * Fold the data for an offline cpu into the global array.
721 * There cannot be any access by the offline cpu and therefore
722 * synchronization is simplified.
723 */
724void cpu_vm_stats_fold(int cpu)
725{
Mel Gorman75ef7182016-07-28 15:45:24 -0700726 struct pglist_data *pgdat;
Christoph Lameter2bb921e2013-09-11 14:21:30 -0700727 struct zone *zone;
728 int i;
Mel Gorman75ef7182016-07-28 15:45:24 -0700729 int global_zone_diff[NR_VM_ZONE_STAT_ITEMS] = { 0, };
730 int global_node_diff[NR_VM_NODE_STAT_ITEMS] = { 0, };
Christoph Lameter2bb921e2013-09-11 14:21:30 -0700731
732 for_each_populated_zone(zone) {
733 struct per_cpu_pageset *p;
734
735 p = per_cpu_ptr(zone->pageset, cpu);
736
737 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
738 if (p->vm_stat_diff[i]) {
739 int v;
740
741 v = p->vm_stat_diff[i];
742 p->vm_stat_diff[i] = 0;
743 atomic_long_add(v, &zone->vm_stat[i]);
Mel Gorman75ef7182016-07-28 15:45:24 -0700744 global_zone_diff[i] += v;
Christoph Lameter2bb921e2013-09-11 14:21:30 -0700745 }
746 }
747
Mel Gorman75ef7182016-07-28 15:45:24 -0700748 for_each_online_pgdat(pgdat) {
749 struct per_cpu_nodestat *p;
750
751 p = per_cpu_ptr(pgdat->per_cpu_nodestats, cpu);
752
753 for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++)
754 if (p->vm_node_stat_diff[i]) {
755 int v;
756
757 v = p->vm_node_stat_diff[i];
758 p->vm_node_stat_diff[i] = 0;
759 atomic_long_add(v, &pgdat->vm_stat[i]);
760 global_node_diff[i] += v;
761 }
762 }
763
764 fold_diff(global_zone_diff, global_node_diff);
Christoph Lameter2bb921e2013-09-11 14:21:30 -0700765}
766
767/*
Cody P Schafer40f4b1e2013-04-29 15:08:38 -0700768 * this is only called if !populated_zone(zone), which implies no other users of
769 * pset->vm_stat_diff[] exsist.
770 */
Minchan Kim5a883812012-10-08 16:33:39 -0700771void drain_zonestat(struct zone *zone, struct per_cpu_pageset *pset)
772{
773 int i;
774
775 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
776 if (pset->vm_stat_diff[i]) {
777 int v = pset->vm_stat_diff[i];
778 pset->vm_stat_diff[i] = 0;
779 atomic_long_add(v, &zone->vm_stat[i]);
Mel Gorman75ef7182016-07-28 15:45:24 -0700780 atomic_long_add(v, &vm_zone_stat[i]);
Minchan Kim5a883812012-10-08 16:33:39 -0700781 }
782}
Christoph Lameter2244b952006-06-30 01:55:33 -0700783#endif
784
Christoph Lameterca889e62006-06-30 01:55:44 -0700785#ifdef CONFIG_NUMA
786/*
Mel Gorman75ef7182016-07-28 15:45:24 -0700787 * Determine the per node value of a stat item. This function
788 * is called frequently in a NUMA machine, so try to be as
789 * frugal as possible.
Andrew Mortonc2d42c12015-11-05 18:48:43 -0800790 */
Mel Gorman75ef7182016-07-28 15:45:24 -0700791unsigned long sum_zone_node_page_state(int node,
792 enum zone_stat_item item)
Andrew Mortonc2d42c12015-11-05 18:48:43 -0800793{
794 struct zone *zones = NODE_DATA(node)->node_zones;
Joonsoo Kime87d59f2016-05-19 17:12:29 -0700795 int i;
796 unsigned long count = 0;
Andrew Mortonc2d42c12015-11-05 18:48:43 -0800797
Joonsoo Kime87d59f2016-05-19 17:12:29 -0700798 for (i = 0; i < MAX_NR_ZONES; i++)
799 count += zone_page_state(zones + i, item);
800
801 return count;
Andrew Mortonc2d42c12015-11-05 18:48:43 -0800802}
803
Mel Gorman75ef7182016-07-28 15:45:24 -0700804/*
805 * Determine the per node value of a stat item.
806 */
807unsigned long node_page_state(struct pglist_data *pgdat,
808 enum node_stat_item item)
809{
810 long x = atomic_long_read(&pgdat->vm_stat[item]);
811#ifdef CONFIG_SMP
812 if (x < 0)
813 x = 0;
814#endif
815 return x;
816}
Christoph Lameterca889e62006-06-30 01:55:44 -0700817#endif
818
Mel Gormand7a57522010-05-24 14:32:25 -0700819#ifdef CONFIG_COMPACTION
Namhyung Kim36deb0b2010-10-26 14:22:04 -0700820
Mel Gormand7a57522010-05-24 14:32:25 -0700821struct contig_page_info {
822 unsigned long free_pages;
823 unsigned long free_blocks_total;
824 unsigned long free_blocks_suitable;
825};
826
827/*
828 * Calculate the number of free pages in a zone, how many contiguous
829 * pages are free and how many are large enough to satisfy an allocation of
830 * the target size. Note that this function makes no attempt to estimate
831 * how many suitable free blocks there *might* be if MOVABLE pages were
832 * migrated. Calculating that is possible, but expensive and can be
833 * figured out from userspace
834 */
835static void fill_contig_page_info(struct zone *zone,
836 unsigned int suitable_order,
837 struct contig_page_info *info)
838{
839 unsigned int order;
840
841 info->free_pages = 0;
842 info->free_blocks_total = 0;
843 info->free_blocks_suitable = 0;
844
845 for (order = 0; order < MAX_ORDER; order++) {
846 unsigned long blocks;
847
848 /* Count number of free blocks */
849 blocks = zone->free_area[order].nr_free;
850 info->free_blocks_total += blocks;
851
852 /* Count free base pages */
853 info->free_pages += blocks << order;
854
855 /* Count the suitable free blocks */
856 if (order >= suitable_order)
857 info->free_blocks_suitable += blocks <<
858 (order - suitable_order);
859 }
860}
Mel Gormanf1a5ab12010-05-24 14:32:26 -0700861
862/*
863 * A fragmentation index only makes sense if an allocation of a requested
864 * size would fail. If that is true, the fragmentation index indicates
865 * whether external fragmentation or a lack of memory was the problem.
866 * The value can be used to determine if page reclaim or compaction
867 * should be used
868 */
Mel Gorman56de7262010-05-24 14:32:30 -0700869static int __fragmentation_index(unsigned int order, struct contig_page_info *info)
Mel Gormanf1a5ab12010-05-24 14:32:26 -0700870{
871 unsigned long requested = 1UL << order;
872
873 if (!info->free_blocks_total)
874 return 0;
875
876 /* Fragmentation index only makes sense when a request would fail */
877 if (info->free_blocks_suitable)
878 return -1000;
879
880 /*
881 * Index is between 0 and 1 so return within 3 decimal places
882 *
883 * 0 => allocation would fail due to lack of memory
884 * 1 => allocation would fail due to fragmentation
885 */
886 return 1000 - div_u64( (1000+(div_u64(info->free_pages * 1000ULL, requested))), info->free_blocks_total);
887}
Mel Gorman56de7262010-05-24 14:32:30 -0700888
889/* Same as __fragmentation index but allocs contig_page_info on stack */
890int fragmentation_index(struct zone *zone, unsigned int order)
891{
892 struct contig_page_info info;
893
894 fill_contig_page_info(zone, order, &info);
895 return __fragmentation_index(order, &info);
896}
Mel Gormand7a57522010-05-24 14:32:25 -0700897#endif
898
David Rientjes0d6617c2011-09-14 16:21:05 -0700899#if defined(CONFIG_PROC_FS) || defined(CONFIG_SYSFS) || defined(CONFIG_NUMA)
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700900#ifdef CONFIG_ZONE_DMA
901#define TEXT_FOR_DMA(xx) xx "_dma",
902#else
903#define TEXT_FOR_DMA(xx)
904#endif
905
906#ifdef CONFIG_ZONE_DMA32
907#define TEXT_FOR_DMA32(xx) xx "_dma32",
908#else
909#define TEXT_FOR_DMA32(xx)
910#endif
911
912#ifdef CONFIG_HIGHMEM
913#define TEXT_FOR_HIGHMEM(xx) xx "_high",
914#else
915#define TEXT_FOR_HIGHMEM(xx)
916#endif
917
918#define TEXTS_FOR_ZONES(xx) TEXT_FOR_DMA(xx) TEXT_FOR_DMA32(xx) xx "_normal", \
919 TEXT_FOR_HIGHMEM(xx) xx "_movable",
920
921const char * const vmstat_text[] = {
Konstantin Khlebnikov09316c02014-10-09 15:29:32 -0700922 /* enum zone_stat_item countes */
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700923 "nr_free_pages",
Minchan Kim71c799f2016-07-28 15:47:26 -0700924 "nr_zone_inactive_anon",
925 "nr_zone_active_anon",
926 "nr_zone_inactive_file",
927 "nr_zone_active_file",
928 "nr_zone_unevictable",
Mel Gorman5a1c84b2016-07-28 15:47:31 -0700929 "nr_zone_write_pending",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700930 "nr_mlock",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700931 "nr_page_table_pages",
932 "nr_kernel_stack",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700933 "nr_bounce",
Minchan Kim91537fe2016-07-26 15:24:45 -0700934#if IS_ENABLED(CONFIG_ZSMALLOC)
935 "nr_zspages",
936#endif
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700937#ifdef CONFIG_NUMA
938 "numa_hit",
939 "numa_miss",
940 "numa_foreign",
941 "numa_interleave",
942 "numa_local",
943 "numa_other",
944#endif
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -0700945 "nr_free_cma",
Konstantin Khlebnikov09316c02014-10-09 15:29:32 -0700946
Mel Gorman599d0c92016-07-28 15:45:31 -0700947 /* Node-based counters */
948 "nr_inactive_anon",
949 "nr_active_anon",
950 "nr_inactive_file",
951 "nr_active_file",
952 "nr_unevictable",
Johannes Weiner385386c2017-07-06 15:40:43 -0700953 "nr_slab_reclaimable",
954 "nr_slab_unreclaimable",
Mel Gorman599d0c92016-07-28 15:45:31 -0700955 "nr_isolated_anon",
956 "nr_isolated_file",
Mel Gorman1e6b10852016-07-28 15:46:08 -0700957 "workingset_refault",
958 "workingset_activate",
959 "workingset_nodereclaim",
Mel Gorman50658e22016-07-28 15:46:14 -0700960 "nr_anon_pages",
961 "nr_mapped",
Mel Gorman11fb9982016-07-28 15:46:20 -0700962 "nr_file_pages",
963 "nr_dirty",
964 "nr_writeback",
965 "nr_writeback_temp",
966 "nr_shmem",
967 "nr_shmem_hugepages",
968 "nr_shmem_pmdmapped",
969 "nr_anon_transparent_hugepages",
970 "nr_unstable",
Mel Gormanc4a25632016-07-28 15:46:23 -0700971 "nr_vmscan_write",
972 "nr_vmscan_immediate_reclaim",
973 "nr_dirtied",
974 "nr_written",
Mel Gorman599d0c92016-07-28 15:45:31 -0700975
Konstantin Khlebnikov09316c02014-10-09 15:29:32 -0700976 /* enum writeback_stat_item counters */
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700977 "nr_dirty_threshold",
978 "nr_dirty_background_threshold",
979
980#ifdef CONFIG_VM_EVENT_COUNTERS
Konstantin Khlebnikov09316c02014-10-09 15:29:32 -0700981 /* enum vm_event_item counters */
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700982 "pgpgin",
983 "pgpgout",
984 "pswpin",
985 "pswpout",
986
987 TEXTS_FOR_ZONES("pgalloc")
Mel Gorman7cc30fc2016-07-28 15:46:59 -0700988 TEXTS_FOR_ZONES("allocstall")
989 TEXTS_FOR_ZONES("pgskip")
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700990
991 "pgfree",
992 "pgactivate",
993 "pgdeactivate",
Shaohua Lif7ad2a62017-05-03 14:52:29 -0700994 "pglazyfree",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700995
996 "pgfault",
997 "pgmajfault",
Minchan Kim854e9ed2016-01-15 16:54:53 -0800998 "pglazyfreed",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700999
Mel Gorman599d0c92016-07-28 15:45:31 -07001000 "pgrefill",
1001 "pgsteal_kswapd",
1002 "pgsteal_direct",
1003 "pgscan_kswapd",
1004 "pgscan_direct",
Mel Gorman68243e72012-07-31 16:44:39 -07001005 "pgscan_direct_throttle",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001006
1007#ifdef CONFIG_NUMA
1008 "zone_reclaim_failed",
1009#endif
1010 "pginodesteal",
1011 "slabs_scanned",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001012 "kswapd_inodesteal",
1013 "kswapd_low_wmark_hit_quickly",
1014 "kswapd_high_wmark_hit_quickly",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001015 "pageoutrun",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001016
1017 "pgrotated",
1018
Dave Hansen5509a5d2014-04-03 14:48:19 -07001019 "drop_pagecache",
1020 "drop_slab",
Konstantin Khlebnikov8e675f72017-07-06 15:40:28 -07001021 "oom_kill",
Dave Hansen5509a5d2014-04-03 14:48:19 -07001022
Mel Gorman03c5a6e2012-11-02 14:52:48 +00001023#ifdef CONFIG_NUMA_BALANCING
1024 "numa_pte_updates",
Mel Gorman72403b42013-11-12 15:08:32 -08001025 "numa_huge_pte_updates",
Mel Gorman03c5a6e2012-11-02 14:52:48 +00001026 "numa_hint_faults",
1027 "numa_hint_faults_local",
1028 "numa_pages_migrated",
1029#endif
Mel Gorman5647bc22012-10-19 10:46:20 +01001030#ifdef CONFIG_MIGRATION
1031 "pgmigrate_success",
1032 "pgmigrate_fail",
1033#endif
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001034#ifdef CONFIG_COMPACTION
Mel Gorman397487d2012-10-19 12:00:10 +01001035 "compact_migrate_scanned",
1036 "compact_free_scanned",
1037 "compact_isolated",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001038 "compact_stall",
1039 "compact_fail",
1040 "compact_success",
Vlastimil Babka698b1b32016-03-17 14:18:08 -07001041 "compact_daemon_wake",
David Rientjes7f354a52017-02-22 15:44:50 -08001042 "compact_daemon_migrate_scanned",
1043 "compact_daemon_free_scanned",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001044#endif
1045
1046#ifdef CONFIG_HUGETLB_PAGE
1047 "htlb_buddy_alloc_success",
1048 "htlb_buddy_alloc_fail",
1049#endif
1050 "unevictable_pgs_culled",
1051 "unevictable_pgs_scanned",
1052 "unevictable_pgs_rescued",
1053 "unevictable_pgs_mlocked",
1054 "unevictable_pgs_munlocked",
1055 "unevictable_pgs_cleared",
1056 "unevictable_pgs_stranded",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001057
1058#ifdef CONFIG_TRANSPARENT_HUGEPAGE
1059 "thp_fault_alloc",
1060 "thp_fault_fallback",
1061 "thp_collapse_alloc",
1062 "thp_collapse_alloc_failed",
Kirill A. Shutemov95ecedc2016-07-26 15:25:31 -07001063 "thp_file_alloc",
1064 "thp_file_mapped",
Kirill A. Shutemov122afea2016-01-15 16:52:46 -08001065 "thp_split_page",
1066 "thp_split_page_failed",
Kirill A. Shutemovf9719a02016-03-17 14:18:45 -07001067 "thp_deferred_split_page",
Kirill A. Shutemov122afea2016-01-15 16:52:46 -08001068 "thp_split_pmd",
Yisheng Xiece9311c2017-03-09 16:17:00 -08001069#ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
1070 "thp_split_pud",
1071#endif
Kirill A. Shutemovd8a8e1f2012-12-12 13:51:09 -08001072 "thp_zero_page_alloc",
1073 "thp_zero_page_alloc_failed",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001074#endif
Konstantin Khlebnikov09316c02014-10-09 15:29:32 -07001075#ifdef CONFIG_MEMORY_BALLOON
1076 "balloon_inflate",
1077 "balloon_deflate",
1078#ifdef CONFIG_BALLOON_COMPACTION
1079 "balloon_migrate",
1080#endif
1081#endif /* CONFIG_MEMORY_BALLOON */
Mel Gormanec659932014-01-21 14:33:16 -08001082#ifdef CONFIG_DEBUG_TLBFLUSH
Dave Hansen6df46862013-09-11 14:20:24 -07001083#ifdef CONFIG_SMP
Dave Hansen9824cf92013-09-11 14:20:23 -07001084 "nr_tlb_remote_flush",
1085 "nr_tlb_remote_flush_received",
Mel Gormanec659932014-01-21 14:33:16 -08001086#endif /* CONFIG_SMP */
Dave Hansen9824cf92013-09-11 14:20:23 -07001087 "nr_tlb_local_flush_all",
1088 "nr_tlb_local_flush_one",
Mel Gormanec659932014-01-21 14:33:16 -08001089#endif /* CONFIG_DEBUG_TLBFLUSH */
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001090
Davidlohr Bueso4f115142014-06-04 16:06:46 -07001091#ifdef CONFIG_DEBUG_VM_VMACACHE
1092 "vmacache_find_calls",
1093 "vmacache_find_hits",
Davidlohr Buesof5f302e2014-12-12 16:56:10 -08001094 "vmacache_full_flushes",
Davidlohr Bueso4f115142014-06-04 16:06:46 -07001095#endif
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001096#endif /* CONFIG_VM_EVENTS_COUNTERS */
1097};
David Rientjes0d6617c2011-09-14 16:21:05 -07001098#endif /* CONFIG_PROC_FS || CONFIG_SYSFS || CONFIG_NUMA */
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001099
1100
Andrew Morton3c486872015-02-10 14:09:43 -08001101#if (defined(CONFIG_DEBUG_FS) && defined(CONFIG_COMPACTION)) || \
1102 defined(CONFIG_PROC_FS)
1103static void *frag_start(struct seq_file *m, loff_t *pos)
1104{
1105 pg_data_t *pgdat;
1106 loff_t node = *pos;
1107
1108 for (pgdat = first_online_pgdat();
1109 pgdat && node;
1110 pgdat = next_online_pgdat(pgdat))
1111 --node;
1112
1113 return pgdat;
1114}
1115
1116static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
1117{
1118 pg_data_t *pgdat = (pg_data_t *)arg;
1119
1120 (*pos)++;
1121 return next_online_pgdat(pgdat);
1122}
1123
1124static void frag_stop(struct seq_file *m, void *arg)
1125{
1126}
1127
David Rientjesb2bd8592017-05-03 14:52:59 -07001128/*
1129 * Walk zones in a node and print using a callback.
1130 * If @assert_populated is true, only use callback for zones that are populated.
1131 */
Andrew Morton3c486872015-02-10 14:09:43 -08001132static void walk_zones_in_node(struct seq_file *m, pg_data_t *pgdat,
Vinayak Menon727c0802017-07-10 15:49:17 -07001133 bool assert_populated, bool nolock,
Andrew Morton3c486872015-02-10 14:09:43 -08001134 void (*print)(struct seq_file *m, pg_data_t *, struct zone *))
1135{
1136 struct zone *zone;
1137 struct zone *node_zones = pgdat->node_zones;
1138 unsigned long flags;
1139
1140 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
David Rientjesb2bd8592017-05-03 14:52:59 -07001141 if (assert_populated && !populated_zone(zone))
Andrew Morton3c486872015-02-10 14:09:43 -08001142 continue;
1143
Vinayak Menon727c0802017-07-10 15:49:17 -07001144 if (!nolock)
1145 spin_lock_irqsave(&zone->lock, flags);
Andrew Morton3c486872015-02-10 14:09:43 -08001146 print(m, pgdat, zone);
Vinayak Menon727c0802017-07-10 15:49:17 -07001147 if (!nolock)
1148 spin_unlock_irqrestore(&zone->lock, flags);
Andrew Morton3c486872015-02-10 14:09:43 -08001149 }
1150}
1151#endif
1152
Mel Gormand7a57522010-05-24 14:32:25 -07001153#ifdef CONFIG_PROC_FS
Mel Gorman467c9962007-10-16 01:26:02 -07001154static void frag_show_print(struct seq_file *m, pg_data_t *pgdat,
1155 struct zone *zone)
1156{
1157 int order;
1158
1159 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
1160 for (order = 0; order < MAX_ORDER; ++order)
1161 seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
1162 seq_putc(m, '\n');
1163}
1164
1165/*
1166 * This walks the free areas for each zone.
1167 */
1168static int frag_show(struct seq_file *m, void *arg)
1169{
1170 pg_data_t *pgdat = (pg_data_t *)arg;
Vinayak Menon727c0802017-07-10 15:49:17 -07001171 walk_zones_in_node(m, pgdat, true, false, frag_show_print);
Mel Gorman467c9962007-10-16 01:26:02 -07001172 return 0;
1173}
1174
1175static void pagetypeinfo_showfree_print(struct seq_file *m,
1176 pg_data_t *pgdat, struct zone *zone)
1177{
1178 int order, mtype;
1179
1180 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++) {
1181 seq_printf(m, "Node %4d, zone %8s, type %12s ",
1182 pgdat->node_id,
1183 zone->name,
1184 migratetype_names[mtype]);
1185 for (order = 0; order < MAX_ORDER; ++order) {
1186 unsigned long freecount = 0;
1187 struct free_area *area;
1188 struct list_head *curr;
1189
1190 area = &(zone->free_area[order]);
1191
1192 list_for_each(curr, &area->free_list[mtype])
1193 freecount++;
1194 seq_printf(m, "%6lu ", freecount);
1195 }
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001196 seq_putc(m, '\n');
1197 }
Mel Gorman467c9962007-10-16 01:26:02 -07001198}
1199
1200/* Print out the free pages at each order for each migatetype */
1201static int pagetypeinfo_showfree(struct seq_file *m, void *arg)
1202{
1203 int order;
1204 pg_data_t *pgdat = (pg_data_t *)arg;
1205
1206 /* Print header */
1207 seq_printf(m, "%-43s ", "Free pages count per migrate type at order");
1208 for (order = 0; order < MAX_ORDER; ++order)
1209 seq_printf(m, "%6d ", order);
1210 seq_putc(m, '\n');
1211
Vinayak Menon727c0802017-07-10 15:49:17 -07001212 walk_zones_in_node(m, pgdat, true, false, pagetypeinfo_showfree_print);
Mel Gorman467c9962007-10-16 01:26:02 -07001213
1214 return 0;
1215}
1216
1217static void pagetypeinfo_showblockcount_print(struct seq_file *m,
1218 pg_data_t *pgdat, struct zone *zone)
1219{
1220 int mtype;
1221 unsigned long pfn;
1222 unsigned long start_pfn = zone->zone_start_pfn;
Cody P Schafer108bcc92013-02-22 16:35:23 -08001223 unsigned long end_pfn = zone_end_pfn(zone);
Mel Gorman467c9962007-10-16 01:26:02 -07001224 unsigned long count[MIGRATE_TYPES] = { 0, };
1225
1226 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
1227 struct page *page;
1228
Michal Hockod336e942017-07-06 15:38:07 -07001229 page = pfn_to_online_page(pfn);
1230 if (!page)
Mel Gorman467c9962007-10-16 01:26:02 -07001231 continue;
1232
Mel Gormaneb335752009-05-13 17:34:48 +01001233 /* Watch for unexpected holes punched in the memmap */
1234 if (!memmap_valid_within(pfn, page, zone))
Mel Gormane80d6a22008-08-14 11:10:14 +01001235 continue;
Mel Gormaneb335752009-05-13 17:34:48 +01001236
Joonsoo Kima91c43c2016-05-19 17:12:10 -07001237 if (page_zone(page) != zone)
1238 continue;
1239
Mel Gorman467c9962007-10-16 01:26:02 -07001240 mtype = get_pageblock_migratetype(page);
1241
Mel Gormane80d6a22008-08-14 11:10:14 +01001242 if (mtype < MIGRATE_TYPES)
1243 count[mtype]++;
Mel Gorman467c9962007-10-16 01:26:02 -07001244 }
1245
1246 /* Print counts */
1247 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
1248 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
1249 seq_printf(m, "%12lu ", count[mtype]);
1250 seq_putc(m, '\n');
1251}
1252
1253/* Print out the free pages at each order for each migratetype */
1254static int pagetypeinfo_showblockcount(struct seq_file *m, void *arg)
1255{
1256 int mtype;
1257 pg_data_t *pgdat = (pg_data_t *)arg;
1258
1259 seq_printf(m, "\n%-23s", "Number of blocks type ");
1260 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
1261 seq_printf(m, "%12s ", migratetype_names[mtype]);
1262 seq_putc(m, '\n');
Vinayak Menon727c0802017-07-10 15:49:17 -07001263 walk_zones_in_node(m, pgdat, true, false,
1264 pagetypeinfo_showblockcount_print);
Mel Gorman467c9962007-10-16 01:26:02 -07001265
1266 return 0;
1267}
1268
Joonsoo Kim48c96a32014-12-12 16:56:01 -08001269/*
1270 * Print out the number of pageblocks for each migratetype that contain pages
1271 * of other types. This gives an indication of how well fallbacks are being
1272 * contained by rmqueue_fallback(). It requires information from PAGE_OWNER
1273 * to determine what is going on
1274 */
1275static void pagetypeinfo_showmixedcount(struct seq_file *m, pg_data_t *pgdat)
1276{
1277#ifdef CONFIG_PAGE_OWNER
1278 int mtype;
1279
Vlastimil Babka7dd80b82016-03-15 14:56:12 -07001280 if (!static_branch_unlikely(&page_owner_inited))
Joonsoo Kim48c96a32014-12-12 16:56:01 -08001281 return;
1282
1283 drain_all_pages(NULL);
1284
1285 seq_printf(m, "\n%-23s", "Number of mixed blocks ");
1286 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
1287 seq_printf(m, "%12s ", migratetype_names[mtype]);
1288 seq_putc(m, '\n');
1289
Vinayak Menon727c0802017-07-10 15:49:17 -07001290 walk_zones_in_node(m, pgdat, true, true,
1291 pagetypeinfo_showmixedcount_print);
Joonsoo Kim48c96a32014-12-12 16:56:01 -08001292#endif /* CONFIG_PAGE_OWNER */
1293}
1294
Mel Gorman467c9962007-10-16 01:26:02 -07001295/*
1296 * This prints out statistics in relation to grouping pages by mobility.
1297 * It is expensive to collect so do not constantly read the file.
1298 */
1299static int pagetypeinfo_show(struct seq_file *m, void *arg)
1300{
1301 pg_data_t *pgdat = (pg_data_t *)arg;
1302
KOSAKI Motohiro41b25a32008-04-30 00:52:13 -07001303 /* check memoryless node */
Lai Jiangshana47b53c2012-12-12 13:51:37 -08001304 if (!node_state(pgdat->node_id, N_MEMORY))
KOSAKI Motohiro41b25a32008-04-30 00:52:13 -07001305 return 0;
1306
Mel Gorman467c9962007-10-16 01:26:02 -07001307 seq_printf(m, "Page block order: %d\n", pageblock_order);
1308 seq_printf(m, "Pages per block: %lu\n", pageblock_nr_pages);
1309 seq_putc(m, '\n');
1310 pagetypeinfo_showfree(m, pgdat);
1311 pagetypeinfo_showblockcount(m, pgdat);
Joonsoo Kim48c96a32014-12-12 16:56:01 -08001312 pagetypeinfo_showmixedcount(m, pgdat);
Mel Gorman467c9962007-10-16 01:26:02 -07001313
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001314 return 0;
1315}
1316
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +04001317static const struct seq_operations fragmentation_op = {
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001318 .start = frag_start,
1319 .next = frag_next,
1320 .stop = frag_stop,
1321 .show = frag_show,
1322};
1323
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +04001324static int fragmentation_open(struct inode *inode, struct file *file)
1325{
1326 return seq_open(file, &fragmentation_op);
1327}
1328
Anshuman Khandual9d85e152017-07-06 15:37:15 -07001329static const struct file_operations buddyinfo_file_operations = {
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +04001330 .open = fragmentation_open,
1331 .read = seq_read,
1332 .llseek = seq_lseek,
1333 .release = seq_release,
1334};
1335
Alexey Dobriyan74e2e8e2008-10-06 04:15:36 +04001336static const struct seq_operations pagetypeinfo_op = {
Mel Gorman467c9962007-10-16 01:26:02 -07001337 .start = frag_start,
1338 .next = frag_next,
1339 .stop = frag_stop,
1340 .show = pagetypeinfo_show,
1341};
1342
Alexey Dobriyan74e2e8e2008-10-06 04:15:36 +04001343static int pagetypeinfo_open(struct inode *inode, struct file *file)
1344{
1345 return seq_open(file, &pagetypeinfo_op);
1346}
1347
Anshuman Khandual9d85e152017-07-06 15:37:15 -07001348static const struct file_operations pagetypeinfo_file_operations = {
Alexey Dobriyan74e2e8e2008-10-06 04:15:36 +04001349 .open = pagetypeinfo_open,
1350 .read = seq_read,
1351 .llseek = seq_lseek,
1352 .release = seq_release,
1353};
1354
Mel Gormane2ecc8a2016-07-28 15:47:02 -07001355static bool is_zone_first_populated(pg_data_t *pgdat, struct zone *zone)
1356{
1357 int zid;
1358
1359 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
1360 struct zone *compare = &pgdat->node_zones[zid];
1361
1362 if (populated_zone(compare))
1363 return zone == compare;
1364 }
1365
Mel Gormane2ecc8a2016-07-28 15:47:02 -07001366 return false;
1367}
1368
Mel Gorman467c9962007-10-16 01:26:02 -07001369static void zoneinfo_show_print(struct seq_file *m, pg_data_t *pgdat,
1370 struct zone *zone)
1371{
1372 int i;
1373 seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
Mel Gormane2ecc8a2016-07-28 15:47:02 -07001374 if (is_zone_first_populated(pgdat, zone)) {
1375 seq_printf(m, "\n per-node stats");
1376 for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) {
1377 seq_printf(m, "\n %-12s %lu",
1378 vmstat_text[i + NR_VM_ZONE_STAT_ITEMS],
1379 node_page_state(pgdat, i));
1380 }
1381 }
Mel Gorman467c9962007-10-16 01:26:02 -07001382 seq_printf(m,
1383 "\n pages free %lu"
1384 "\n min %lu"
1385 "\n low %lu"
1386 "\n high %lu"
Mel Gorman467c9962007-10-16 01:26:02 -07001387 "\n spanned %lu"
Jiang Liu9feedc92012-12-12 13:52:12 -08001388 "\n present %lu"
1389 "\n managed %lu",
Mel Gorman88f5acf2011-01-13 15:45:41 -08001390 zone_page_state(zone, NR_FREE_PAGES),
Mel Gorman41858962009-06-16 15:32:12 -07001391 min_wmark_pages(zone),
1392 low_wmark_pages(zone),
1393 high_wmark_pages(zone),
Mel Gorman467c9962007-10-16 01:26:02 -07001394 zone->spanned_pages,
Jiang Liu9feedc92012-12-12 13:52:12 -08001395 zone->present_pages,
1396 zone->managed_pages);
Mel Gorman467c9962007-10-16 01:26:02 -07001397
Mel Gorman467c9962007-10-16 01:26:02 -07001398 seq_printf(m,
Mel Gorman3484b2d2014-08-06 16:07:14 -07001399 "\n protection: (%ld",
Mel Gorman467c9962007-10-16 01:26:02 -07001400 zone->lowmem_reserve[0]);
1401 for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
Mel Gorman3484b2d2014-08-06 16:07:14 -07001402 seq_printf(m, ", %ld", zone->lowmem_reserve[i]);
David Rientjes7dfb8bf2017-05-03 14:53:02 -07001403 seq_putc(m, ')');
1404
1405 /* If unpopulated, no other information is useful */
1406 if (!populated_zone(zone)) {
1407 seq_putc(m, '\n');
1408 return;
1409 }
1410
1411 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
1412 seq_printf(m, "\n %-12s %lu", vmstat_text[i],
1413 zone_page_state(zone, i));
1414
1415 seq_printf(m, "\n pagesets");
Mel Gorman467c9962007-10-16 01:26:02 -07001416 for_each_online_cpu(i) {
1417 struct per_cpu_pageset *pageset;
Mel Gorman467c9962007-10-16 01:26:02 -07001418
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001419 pageset = per_cpu_ptr(zone->pageset, i);
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001420 seq_printf(m,
1421 "\n cpu: %i"
1422 "\n count: %i"
1423 "\n high: %i"
1424 "\n batch: %i",
1425 i,
1426 pageset->pcp.count,
1427 pageset->pcp.high,
1428 pageset->pcp.batch);
Mel Gorman467c9962007-10-16 01:26:02 -07001429#ifdef CONFIG_SMP
1430 seq_printf(m, "\n vm stats threshold: %d",
1431 pageset->stat_threshold);
1432#endif
1433 }
1434 seq_printf(m,
Mel Gorman599d0c92016-07-28 15:45:31 -07001435 "\n node_unreclaimable: %u"
1436 "\n start_pfn: %lu"
1437 "\n node_inactive_ratio: %u",
Johannes Weinerc73322d2017-05-03 14:51:51 -07001438 pgdat->kswapd_failures >= MAX_RECLAIM_RETRIES,
Rik van Riel556adec2008-10-18 20:26:34 -07001439 zone->zone_start_pfn,
Mel Gorman599d0c92016-07-28 15:45:31 -07001440 zone->zone_pgdat->inactive_ratio);
Mel Gorman467c9962007-10-16 01:26:02 -07001441 seq_putc(m, '\n');
1442}
1443
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001444/*
David Rientjesb2bd8592017-05-03 14:52:59 -07001445 * Output information about zones in @pgdat. All zones are printed regardless
1446 * of whether they are populated or not: lowmem_reserve_ratio operates on the
1447 * set of all zones and userspace would not be aware of such zones if they are
1448 * suppressed here (zoneinfo displays the effect of lowmem_reserve_ratio).
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001449 */
1450static int zoneinfo_show(struct seq_file *m, void *arg)
1451{
Mel Gorman467c9962007-10-16 01:26:02 -07001452 pg_data_t *pgdat = (pg_data_t *)arg;
Vinayak Menon727c0802017-07-10 15:49:17 -07001453 walk_zones_in_node(m, pgdat, false, false, zoneinfo_show_print);
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001454 return 0;
1455}
1456
Alexey Dobriyan5c9fe622008-10-06 04:19:42 +04001457static const struct seq_operations zoneinfo_op = {
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001458 .start = frag_start, /* iterate over all zones. The same as in
1459 * fragmentation. */
1460 .next = frag_next,
1461 .stop = frag_stop,
1462 .show = zoneinfo_show,
1463};
1464
Alexey Dobriyan5c9fe622008-10-06 04:19:42 +04001465static int zoneinfo_open(struct inode *inode, struct file *file)
1466{
1467 return seq_open(file, &zoneinfo_op);
1468}
1469
Anshuman Khandual9d85e152017-07-06 15:37:15 -07001470static const struct file_operations zoneinfo_file_operations = {
Alexey Dobriyan5c9fe622008-10-06 04:19:42 +04001471 .open = zoneinfo_open,
1472 .read = seq_read,
1473 .llseek = seq_lseek,
1474 .release = seq_release,
1475};
1476
Michael Rubin79da8262010-10-26 14:21:36 -07001477enum writeback_stat_item {
1478 NR_DIRTY_THRESHOLD,
1479 NR_DIRTY_BG_THRESHOLD,
1480 NR_VM_WRITEBACK_STAT_ITEMS,
1481};
1482
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001483static void *vmstat_start(struct seq_file *m, loff_t *pos)
1484{
Christoph Lameter2244b952006-06-30 01:55:33 -07001485 unsigned long *v;
Michael Rubin79da8262010-10-26 14:21:36 -07001486 int i, stat_items_size;
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001487
1488 if (*pos >= ARRAY_SIZE(vmstat_text))
1489 return NULL;
Michael Rubin79da8262010-10-26 14:21:36 -07001490 stat_items_size = NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long) +
Mel Gorman75ef7182016-07-28 15:45:24 -07001491 NR_VM_NODE_STAT_ITEMS * sizeof(unsigned long) +
Michael Rubin79da8262010-10-26 14:21:36 -07001492 NR_VM_WRITEBACK_STAT_ITEMS * sizeof(unsigned long);
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001493
Christoph Lameterf8891e52006-06-30 01:55:45 -07001494#ifdef CONFIG_VM_EVENT_COUNTERS
Michael Rubin79da8262010-10-26 14:21:36 -07001495 stat_items_size += sizeof(struct vm_event_state);
Christoph Lameterf8891e52006-06-30 01:55:45 -07001496#endif
Michael Rubin79da8262010-10-26 14:21:36 -07001497
1498 v = kmalloc(stat_items_size, GFP_KERNEL);
Christoph Lameter2244b952006-06-30 01:55:33 -07001499 m->private = v;
1500 if (!v)
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001501 return ERR_PTR(-ENOMEM);
Christoph Lameter2244b952006-06-30 01:55:33 -07001502 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
1503 v[i] = global_page_state(i);
Michael Rubin79da8262010-10-26 14:21:36 -07001504 v += NR_VM_ZONE_STAT_ITEMS;
1505
Mel Gorman75ef7182016-07-28 15:45:24 -07001506 for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++)
1507 v[i] = global_node_page_state(i);
1508 v += NR_VM_NODE_STAT_ITEMS;
1509
Michael Rubin79da8262010-10-26 14:21:36 -07001510 global_dirty_limits(v + NR_DIRTY_BG_THRESHOLD,
1511 v + NR_DIRTY_THRESHOLD);
1512 v += NR_VM_WRITEBACK_STAT_ITEMS;
1513
Christoph Lameterf8891e52006-06-30 01:55:45 -07001514#ifdef CONFIG_VM_EVENT_COUNTERS
Michael Rubin79da8262010-10-26 14:21:36 -07001515 all_vm_events(v);
1516 v[PGPGIN] /= 2; /* sectors -> kbytes */
1517 v[PGPGOUT] /= 2;
Christoph Lameterf8891e52006-06-30 01:55:45 -07001518#endif
Wu Fengguangff8b16d2010-11-04 01:56:49 +08001519 return (unsigned long *)m->private + *pos;
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001520}
1521
1522static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
1523{
1524 (*pos)++;
1525 if (*pos >= ARRAY_SIZE(vmstat_text))
1526 return NULL;
1527 return (unsigned long *)m->private + *pos;
1528}
1529
1530static int vmstat_show(struct seq_file *m, void *arg)
1531{
1532 unsigned long *l = arg;
1533 unsigned long off = l - (unsigned long *)m->private;
Alexey Dobriyan68ba0322016-10-07 17:02:14 -07001534
1535 seq_puts(m, vmstat_text[off]);
Joe Perches75ba1d02016-10-07 17:02:20 -07001536 seq_put_decimal_ull(m, " ", *l);
Alexey Dobriyan68ba0322016-10-07 17:02:14 -07001537 seq_putc(m, '\n');
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001538 return 0;
1539}
1540
1541static void vmstat_stop(struct seq_file *m, void *arg)
1542{
1543 kfree(m->private);
1544 m->private = NULL;
1545}
1546
Alexey Dobriyanb6aa44a2008-10-06 04:17:48 +04001547static const struct seq_operations vmstat_op = {
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001548 .start = vmstat_start,
1549 .next = vmstat_next,
1550 .stop = vmstat_stop,
1551 .show = vmstat_show,
1552};
1553
Alexey Dobriyanb6aa44a2008-10-06 04:17:48 +04001554static int vmstat_open(struct inode *inode, struct file *file)
1555{
1556 return seq_open(file, &vmstat_op);
1557}
1558
Anshuman Khandual9d85e152017-07-06 15:37:15 -07001559static const struct file_operations vmstat_file_operations = {
Alexey Dobriyanb6aa44a2008-10-06 04:17:48 +04001560 .open = vmstat_open,
1561 .read = seq_read,
1562 .llseek = seq_lseek,
1563 .release = seq_release,
1564};
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001565#endif /* CONFIG_PROC_FS */
1566
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001567#ifdef CONFIG_SMP
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001568static DEFINE_PER_CPU(struct delayed_work, vmstat_work);
Christoph Lameter77461ab2007-05-09 02:35:13 -07001569int sysctl_stat_interval __read_mostly = HZ;
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001570
Hugh Dickins52b6f462016-05-19 17:12:50 -07001571#ifdef CONFIG_PROC_FS
1572static void refresh_vm_stats(struct work_struct *work)
1573{
1574 refresh_cpu_vm_stats(true);
1575}
1576
1577int vmstat_refresh(struct ctl_table *table, int write,
1578 void __user *buffer, size_t *lenp, loff_t *ppos)
1579{
1580 long val;
1581 int err;
1582 int i;
1583
1584 /*
1585 * The regular update, every sysctl_stat_interval, may come later
1586 * than expected: leaving a significant amount in per_cpu buckets.
1587 * This is particularly misleading when checking a quantity of HUGE
1588 * pages, immediately after running a test. /proc/sys/vm/stat_refresh,
1589 * which can equally be echo'ed to or cat'ted from (by root),
1590 * can be used to update the stats just before reading them.
1591 *
1592 * Oh, and since global_page_state() etc. are so careful to hide
1593 * transiently negative values, report an error here if any of
1594 * the stats is negative, so we know to go looking for imbalance.
1595 */
1596 err = schedule_on_each_cpu(refresh_vm_stats);
1597 if (err)
1598 return err;
1599 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) {
Mel Gorman75ef7182016-07-28 15:45:24 -07001600 val = atomic_long_read(&vm_zone_stat[i]);
Hugh Dickins52b6f462016-05-19 17:12:50 -07001601 if (val < 0) {
Johannes Weinerc822f622017-05-03 14:52:10 -07001602 pr_warn("%s: %s %ld\n",
1603 __func__, vmstat_text[i], val);
1604 err = -EINVAL;
Hugh Dickins52b6f462016-05-19 17:12:50 -07001605 }
1606 }
1607 if (err)
1608 return err;
1609 if (write)
1610 *ppos += *lenp;
1611 else
1612 *lenp = 0;
1613 return 0;
1614}
1615#endif /* CONFIG_PROC_FS */
1616
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001617static void vmstat_update(struct work_struct *w)
1618{
Christoph Lameter0eb77e92016-01-14 15:21:40 -08001619 if (refresh_cpu_vm_stats(true)) {
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001620 /*
1621 * Counters were updated so we expect more updates
1622 * to occur in the future. Keep on running the
1623 * update worker thread.
1624 */
Michal Hockoce612872017-04-07 16:05:05 -07001625 queue_delayed_work_on(smp_processor_id(), mm_percpu_wq,
Michal Hockof01f17d2016-02-05 15:36:24 -08001626 this_cpu_ptr(&vmstat_work),
1627 round_jiffies_relative(sysctl_stat_interval));
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001628 }
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001629}
1630
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001631/*
Christoph Lameter0eb77e92016-01-14 15:21:40 -08001632 * Switch off vmstat processing and then fold all the remaining differentials
1633 * until the diffs stay at zero. The function is used by NOHZ and can only be
1634 * invoked when tick processing is not active.
1635 */
Christoph Lameter0eb77e92016-01-14 15:21:40 -08001636/*
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001637 * Check if the diffs for a certain cpu indicate that
1638 * an update is needed.
1639 */
1640static bool need_update(int cpu)
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001641{
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001642 struct zone *zone;
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001643
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001644 for_each_populated_zone(zone) {
1645 struct per_cpu_pageset *p = per_cpu_ptr(zone->pageset, cpu);
1646
1647 BUILD_BUG_ON(sizeof(p->vm_stat_diff[0]) != 1);
1648 /*
1649 * The fast way of checking if there are any vmstat diffs.
1650 * This works because the diffs are byte sized items.
1651 */
1652 if (memchr_inv(p->vm_stat_diff, 0, NR_VM_ZONE_STAT_ITEMS))
1653 return true;
1654
1655 }
1656 return false;
1657}
1658
Christoph Lameter7b8da4c2016-05-20 16:58:21 -07001659/*
1660 * Switch off vmstat processing and then fold all the remaining differentials
1661 * until the diffs stay at zero. The function is used by NOHZ and can only be
1662 * invoked when tick processing is not active.
1663 */
Michal Hockof01f17d2016-02-05 15:36:24 -08001664void quiet_vmstat(void)
1665{
1666 if (system_state != SYSTEM_RUNNING)
1667 return;
1668
Christoph Lameter7b8da4c2016-05-20 16:58:21 -07001669 if (!delayed_work_pending(this_cpu_ptr(&vmstat_work)))
Michal Hockof01f17d2016-02-05 15:36:24 -08001670 return;
1671
1672 if (!need_update(smp_processor_id()))
1673 return;
1674
1675 /*
1676 * Just refresh counters and do not care about the pending delayed
1677 * vmstat_update. It doesn't fire that often to matter and canceling
1678 * it would be too expensive from this path.
1679 * vmstat_shepherd will take care about that for us.
1680 */
1681 refresh_cpu_vm_stats(false);
1682}
1683
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001684/*
1685 * Shepherd worker thread that checks the
1686 * differentials of processors that have their worker
1687 * threads for vm statistics updates disabled because of
1688 * inactivity.
1689 */
1690static void vmstat_shepherd(struct work_struct *w);
1691
Christoph Lameter0eb77e92016-01-14 15:21:40 -08001692static DECLARE_DEFERRABLE_WORK(shepherd, vmstat_shepherd);
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001693
1694static void vmstat_shepherd(struct work_struct *w)
1695{
1696 int cpu;
1697
1698 get_online_cpus();
1699 /* Check processors whose vmstat worker threads have been disabled */
Christoph Lameter7b8da4c2016-05-20 16:58:21 -07001700 for_each_online_cpu(cpu) {
Michal Hockof01f17d2016-02-05 15:36:24 -08001701 struct delayed_work *dw = &per_cpu(vmstat_work, cpu);
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001702
Christoph Lameter7b8da4c2016-05-20 16:58:21 -07001703 if (!delayed_work_pending(dw) && need_update(cpu))
Michal Hockoce612872017-04-07 16:05:05 -07001704 queue_delayed_work_on(cpu, mm_percpu_wq, dw, 0);
Michal Hockof01f17d2016-02-05 15:36:24 -08001705 }
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001706 put_online_cpus();
1707
1708 schedule_delayed_work(&shepherd,
1709 round_jiffies_relative(sysctl_stat_interval));
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001710}
1711
1712static void __init start_shepherd_timer(void)
1713{
1714 int cpu;
1715
1716 for_each_possible_cpu(cpu)
Michal Hockoccde8bd2016-02-05 15:36:27 -08001717 INIT_DEFERRABLE_WORK(per_cpu_ptr(&vmstat_work, cpu),
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001718 vmstat_update);
1719
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001720 schedule_delayed_work(&shepherd,
1721 round_jiffies_relative(sysctl_stat_interval));
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001722}
1723
Tim Chen03e86db2016-10-07 17:00:02 -07001724static void __init init_cpu_node_state(void)
1725{
Sebastian Andrzej Siewior4c501322016-11-29 15:51:14 +01001726 int node;
Tim Chen03e86db2016-10-07 17:00:02 -07001727
Sebastian Andrzej Siewior4c501322016-11-29 15:51:14 +01001728 for_each_online_node(node) {
1729 if (cpumask_weight(cpumask_of_node(node)) > 0)
1730 node_set_state(node, N_CPU);
1731 }
Tim Chen03e86db2016-10-07 17:00:02 -07001732}
1733
Sebastian Andrzej Siewior5438da92016-11-29 15:52:21 +01001734static int vmstat_cpu_online(unsigned int cpu)
1735{
1736 refresh_zone_stat_thresholds();
1737 node_set_state(cpu_to_node(cpu), N_CPU);
1738 return 0;
1739}
1740
1741static int vmstat_cpu_down_prep(unsigned int cpu)
1742{
1743 cancel_delayed_work_sync(&per_cpu(vmstat_work, cpu));
1744 return 0;
1745}
1746
1747static int vmstat_cpu_dead(unsigned int cpu)
Toshi Kani807a1bd2013-11-12 15:08:13 -08001748{
Sebastian Andrzej Siewior4c501322016-11-29 15:51:14 +01001749 const struct cpumask *node_cpus;
Sebastian Andrzej Siewior5438da92016-11-29 15:52:21 +01001750 int node;
Toshi Kani807a1bd2013-11-12 15:08:13 -08001751
Sebastian Andrzej Siewior5438da92016-11-29 15:52:21 +01001752 node = cpu_to_node(cpu);
1753
1754 refresh_zone_stat_thresholds();
Sebastian Andrzej Siewior4c501322016-11-29 15:51:14 +01001755 node_cpus = cpumask_of_node(node);
1756 if (cpumask_weight(node_cpus) > 0)
Sebastian Andrzej Siewior5438da92016-11-29 15:52:21 +01001757 return 0;
Toshi Kani807a1bd2013-11-12 15:08:13 -08001758
1759 node_clear_state(node, N_CPU);
Sebastian Andrzej Siewior5438da92016-11-29 15:52:21 +01001760 return 0;
Toshi Kani807a1bd2013-11-12 15:08:13 -08001761}
1762
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +04001763#endif
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001764
Michal Hockoce612872017-04-07 16:05:05 -07001765struct workqueue_struct *mm_percpu_wq;
1766
Michal Hocko597b7302017-03-31 15:11:47 -07001767void __init init_mm_internals(void)
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001768{
Michal Hockoce612872017-04-07 16:05:05 -07001769 int ret __maybe_unused;
Sebastian Andrzej Siewior5438da92016-11-29 15:52:21 +01001770
Michal Hocko80d136e2017-04-19 09:52:46 +02001771 mm_percpu_wq = alloc_workqueue("mm_percpu_wq", WQ_MEM_RECLAIM, 0);
Michal Hockoce612872017-04-07 16:05:05 -07001772
1773#ifdef CONFIG_SMP
Sebastian Andrzej Siewior5438da92016-11-29 15:52:21 +01001774 ret = cpuhp_setup_state_nocalls(CPUHP_MM_VMSTAT_DEAD, "mm/vmstat:dead",
1775 NULL, vmstat_cpu_dead);
1776 if (ret < 0)
1777 pr_err("vmstat: failed to register 'dead' hotplug state\n");
1778
1779 ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "mm/vmstat:online",
1780 vmstat_cpu_online,
1781 vmstat_cpu_down_prep);
1782 if (ret < 0)
1783 pr_err("vmstat: failed to register 'online' hotplug state\n");
1784
1785 get_online_cpus();
Tim Chen03e86db2016-10-07 17:00:02 -07001786 init_cpu_node_state();
Sebastian Andrzej Siewior5438da92016-11-29 15:52:21 +01001787 put_online_cpus();
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001788
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001789 start_shepherd_timer();
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +04001790#endif
1791#ifdef CONFIG_PROC_FS
Anshuman Khandual9d85e152017-07-06 15:37:15 -07001792 proc_create("buddyinfo", 0444, NULL, &buddyinfo_file_operations);
1793 proc_create("pagetypeinfo", 0444, NULL, &pagetypeinfo_file_operations);
1794 proc_create("vmstat", 0444, NULL, &vmstat_file_operations);
1795 proc_create("zoneinfo", 0444, NULL, &zoneinfo_file_operations);
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +04001796#endif
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001797}
Mel Gormand7a57522010-05-24 14:32:25 -07001798
1799#if defined(CONFIG_DEBUG_FS) && defined(CONFIG_COMPACTION)
Mel Gormand7a57522010-05-24 14:32:25 -07001800
1801/*
1802 * Return an index indicating how much of the available free memory is
1803 * unusable for an allocation of the requested size.
1804 */
1805static int unusable_free_index(unsigned int order,
1806 struct contig_page_info *info)
1807{
1808 /* No free memory is interpreted as all free memory is unusable */
1809 if (info->free_pages == 0)
1810 return 1000;
1811
1812 /*
1813 * Index should be a value between 0 and 1. Return a value to 3
1814 * decimal places.
1815 *
1816 * 0 => no fragmentation
1817 * 1 => high fragmentation
1818 */
1819 return div_u64((info->free_pages - (info->free_blocks_suitable << order)) * 1000ULL, info->free_pages);
1820
1821}
1822
1823static void unusable_show_print(struct seq_file *m,
1824 pg_data_t *pgdat, struct zone *zone)
1825{
1826 unsigned int order;
1827 int index;
1828 struct contig_page_info info;
1829
1830 seq_printf(m, "Node %d, zone %8s ",
1831 pgdat->node_id,
1832 zone->name);
1833 for (order = 0; order < MAX_ORDER; ++order) {
1834 fill_contig_page_info(zone, order, &info);
1835 index = unusable_free_index(order, &info);
1836 seq_printf(m, "%d.%03d ", index / 1000, index % 1000);
1837 }
1838
1839 seq_putc(m, '\n');
1840}
1841
1842/*
1843 * Display unusable free space index
1844 *
1845 * The unusable free space index measures how much of the available free
1846 * memory cannot be used to satisfy an allocation of a given size and is a
1847 * value between 0 and 1. The higher the value, the more of free memory is
1848 * unusable and by implication, the worse the external fragmentation is. This
1849 * can be expressed as a percentage by multiplying by 100.
1850 */
1851static int unusable_show(struct seq_file *m, void *arg)
1852{
1853 pg_data_t *pgdat = (pg_data_t *)arg;
1854
1855 /* check memoryless node */
Lai Jiangshana47b53c2012-12-12 13:51:37 -08001856 if (!node_state(pgdat->node_id, N_MEMORY))
Mel Gormand7a57522010-05-24 14:32:25 -07001857 return 0;
1858
Vinayak Menon727c0802017-07-10 15:49:17 -07001859 walk_zones_in_node(m, pgdat, true, false, unusable_show_print);
Mel Gormand7a57522010-05-24 14:32:25 -07001860
1861 return 0;
1862}
1863
1864static const struct seq_operations unusable_op = {
1865 .start = frag_start,
1866 .next = frag_next,
1867 .stop = frag_stop,
1868 .show = unusable_show,
1869};
1870
1871static int unusable_open(struct inode *inode, struct file *file)
1872{
1873 return seq_open(file, &unusable_op);
1874}
1875
1876static const struct file_operations unusable_file_ops = {
1877 .open = unusable_open,
1878 .read = seq_read,
1879 .llseek = seq_lseek,
1880 .release = seq_release,
1881};
1882
Mel Gormanf1a5ab12010-05-24 14:32:26 -07001883static void extfrag_show_print(struct seq_file *m,
1884 pg_data_t *pgdat, struct zone *zone)
1885{
1886 unsigned int order;
1887 int index;
1888
1889 /* Alloc on stack as interrupts are disabled for zone walk */
1890 struct contig_page_info info;
1891
1892 seq_printf(m, "Node %d, zone %8s ",
1893 pgdat->node_id,
1894 zone->name);
1895 for (order = 0; order < MAX_ORDER; ++order) {
1896 fill_contig_page_info(zone, order, &info);
Mel Gorman56de7262010-05-24 14:32:30 -07001897 index = __fragmentation_index(order, &info);
Mel Gormanf1a5ab12010-05-24 14:32:26 -07001898 seq_printf(m, "%d.%03d ", index / 1000, index % 1000);
1899 }
1900
1901 seq_putc(m, '\n');
1902}
1903
1904/*
1905 * Display fragmentation index for orders that allocations would fail for
1906 */
1907static int extfrag_show(struct seq_file *m, void *arg)
1908{
1909 pg_data_t *pgdat = (pg_data_t *)arg;
1910
Vinayak Menon727c0802017-07-10 15:49:17 -07001911 walk_zones_in_node(m, pgdat, true, false, extfrag_show_print);
Mel Gormanf1a5ab12010-05-24 14:32:26 -07001912
1913 return 0;
1914}
1915
1916static const struct seq_operations extfrag_op = {
1917 .start = frag_start,
1918 .next = frag_next,
1919 .stop = frag_stop,
1920 .show = extfrag_show,
1921};
1922
1923static int extfrag_open(struct inode *inode, struct file *file)
1924{
1925 return seq_open(file, &extfrag_op);
1926}
1927
1928static const struct file_operations extfrag_file_ops = {
1929 .open = extfrag_open,
1930 .read = seq_read,
1931 .llseek = seq_lseek,
1932 .release = seq_release,
1933};
1934
Mel Gormand7a57522010-05-24 14:32:25 -07001935static int __init extfrag_debug_init(void)
1936{
Sasikantha babubde8bd82012-05-29 15:06:22 -07001937 struct dentry *extfrag_debug_root;
1938
Mel Gormand7a57522010-05-24 14:32:25 -07001939 extfrag_debug_root = debugfs_create_dir("extfrag", NULL);
1940 if (!extfrag_debug_root)
1941 return -ENOMEM;
1942
1943 if (!debugfs_create_file("unusable_index", 0444,
1944 extfrag_debug_root, NULL, &unusable_file_ops))
Sasikantha babubde8bd82012-05-29 15:06:22 -07001945 goto fail;
Mel Gormand7a57522010-05-24 14:32:25 -07001946
Mel Gormanf1a5ab12010-05-24 14:32:26 -07001947 if (!debugfs_create_file("extfrag_index", 0444,
1948 extfrag_debug_root, NULL, &extfrag_file_ops))
Sasikantha babubde8bd82012-05-29 15:06:22 -07001949 goto fail;
Mel Gormanf1a5ab12010-05-24 14:32:26 -07001950
Mel Gormand7a57522010-05-24 14:32:25 -07001951 return 0;
Sasikantha babubde8bd82012-05-29 15:06:22 -07001952fail:
1953 debugfs_remove_recursive(extfrag_debug_root);
1954 return -ENOMEM;
Mel Gormand7a57522010-05-24 14:32:25 -07001955}
1956
1957module_init(extfrag_debug_init);
1958#endif