blob: 5caaf3f7b8abca176ce0cfe8c068eb4781530eda [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",
Mel Gorman599d0c92016-07-28 15:45:31 -0700924 "nr_zone_anon_lru",
925 "nr_zone_file_lru",
Mel Gorman11fb9982016-07-28 15:46:20 -0700926 "nr_zone_write_pending",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700927 "nr_mlock",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700928 "nr_slab_reclaimable",
929 "nr_slab_unreclaimable",
930 "nr_page_table_pages",
931 "nr_kernel_stack",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700932 "nr_bounce",
Minchan Kim91537fe2016-07-26 15:24:45 -0700933#if IS_ENABLED(CONFIG_ZSMALLOC)
934 "nr_zspages",
935#endif
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700936#ifdef CONFIG_NUMA
937 "numa_hit",
938 "numa_miss",
939 "numa_foreign",
940 "numa_interleave",
941 "numa_local",
942 "numa_other",
943#endif
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -0700944 "nr_free_cma",
Konstantin Khlebnikov09316c02014-10-09 15:29:32 -0700945
Mel Gorman599d0c92016-07-28 15:45:31 -0700946 /* Node-based counters */
947 "nr_inactive_anon",
948 "nr_active_anon",
949 "nr_inactive_file",
950 "nr_active_file",
951 "nr_unevictable",
952 "nr_isolated_anon",
953 "nr_isolated_file",
954 "nr_pages_scanned",
Mel Gorman1e6b10852016-07-28 15:46:08 -0700955 "workingset_refault",
956 "workingset_activate",
957 "workingset_nodereclaim",
Mel Gorman50658e22016-07-28 15:46:14 -0700958 "nr_anon_pages",
959 "nr_mapped",
Mel Gorman11fb9982016-07-28 15:46:20 -0700960 "nr_file_pages",
961 "nr_dirty",
962 "nr_writeback",
963 "nr_writeback_temp",
964 "nr_shmem",
965 "nr_shmem_hugepages",
966 "nr_shmem_pmdmapped",
967 "nr_anon_transparent_hugepages",
968 "nr_unstable",
Mel Gormanc4a25632016-07-28 15:46:23 -0700969 "nr_vmscan_write",
970 "nr_vmscan_immediate_reclaim",
971 "nr_dirtied",
972 "nr_written",
Mel Gorman599d0c92016-07-28 15:45:31 -0700973
Konstantin Khlebnikov09316c02014-10-09 15:29:32 -0700974 /* enum writeback_stat_item counters */
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700975 "nr_dirty_threshold",
976 "nr_dirty_background_threshold",
977
978#ifdef CONFIG_VM_EVENT_COUNTERS
Konstantin Khlebnikov09316c02014-10-09 15:29:32 -0700979 /* enum vm_event_item counters */
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700980 "pgpgin",
981 "pgpgout",
982 "pswpin",
983 "pswpout",
984
985 TEXTS_FOR_ZONES("pgalloc")
Mel Gorman7cc30fc2016-07-28 15:46:59 -0700986 TEXTS_FOR_ZONES("allocstall")
987 TEXTS_FOR_ZONES("pgskip")
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700988
989 "pgfree",
990 "pgactivate",
991 "pgdeactivate",
992
993 "pgfault",
994 "pgmajfault",
Minchan Kim854e9ed2016-01-15 16:54:53 -0800995 "pglazyfreed",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -0700996
Mel Gorman599d0c92016-07-28 15:45:31 -0700997 "pgrefill",
998 "pgsteal_kswapd",
999 "pgsteal_direct",
1000 "pgscan_kswapd",
1001 "pgscan_direct",
Mel Gorman68243e72012-07-31 16:44:39 -07001002 "pgscan_direct_throttle",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001003
1004#ifdef CONFIG_NUMA
1005 "zone_reclaim_failed",
1006#endif
1007 "pginodesteal",
1008 "slabs_scanned",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001009 "kswapd_inodesteal",
1010 "kswapd_low_wmark_hit_quickly",
1011 "kswapd_high_wmark_hit_quickly",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001012 "pageoutrun",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001013
1014 "pgrotated",
1015
Dave Hansen5509a5d2014-04-03 14:48:19 -07001016 "drop_pagecache",
1017 "drop_slab",
1018
Mel Gorman03c5a6e2012-11-02 14:52:48 +00001019#ifdef CONFIG_NUMA_BALANCING
1020 "numa_pte_updates",
Mel Gorman72403b42013-11-12 15:08:32 -08001021 "numa_huge_pte_updates",
Mel Gorman03c5a6e2012-11-02 14:52:48 +00001022 "numa_hint_faults",
1023 "numa_hint_faults_local",
1024 "numa_pages_migrated",
1025#endif
Mel Gorman5647bc22012-10-19 10:46:20 +01001026#ifdef CONFIG_MIGRATION
1027 "pgmigrate_success",
1028 "pgmigrate_fail",
1029#endif
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001030#ifdef CONFIG_COMPACTION
Mel Gorman397487d2012-10-19 12:00:10 +01001031 "compact_migrate_scanned",
1032 "compact_free_scanned",
1033 "compact_isolated",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001034 "compact_stall",
1035 "compact_fail",
1036 "compact_success",
Vlastimil Babka698b1b32016-03-17 14:18:08 -07001037 "compact_daemon_wake",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001038#endif
1039
1040#ifdef CONFIG_HUGETLB_PAGE
1041 "htlb_buddy_alloc_success",
1042 "htlb_buddy_alloc_fail",
1043#endif
1044 "unevictable_pgs_culled",
1045 "unevictable_pgs_scanned",
1046 "unevictable_pgs_rescued",
1047 "unevictable_pgs_mlocked",
1048 "unevictable_pgs_munlocked",
1049 "unevictable_pgs_cleared",
1050 "unevictable_pgs_stranded",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001051
1052#ifdef CONFIG_TRANSPARENT_HUGEPAGE
1053 "thp_fault_alloc",
1054 "thp_fault_fallback",
1055 "thp_collapse_alloc",
1056 "thp_collapse_alloc_failed",
Kirill A. Shutemov95ecedc2016-07-26 15:25:31 -07001057 "thp_file_alloc",
1058 "thp_file_mapped",
Kirill A. Shutemov122afea2016-01-15 16:52:46 -08001059 "thp_split_page",
1060 "thp_split_page_failed",
Kirill A. Shutemovf9719a02016-03-17 14:18:45 -07001061 "thp_deferred_split_page",
Kirill A. Shutemov122afea2016-01-15 16:52:46 -08001062 "thp_split_pmd",
Kirill A. Shutemovd8a8e1f2012-12-12 13:51:09 -08001063 "thp_zero_page_alloc",
1064 "thp_zero_page_alloc_failed",
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001065#endif
Konstantin Khlebnikov09316c02014-10-09 15:29:32 -07001066#ifdef CONFIG_MEMORY_BALLOON
1067 "balloon_inflate",
1068 "balloon_deflate",
1069#ifdef CONFIG_BALLOON_COMPACTION
1070 "balloon_migrate",
1071#endif
1072#endif /* CONFIG_MEMORY_BALLOON */
Mel Gormanec659932014-01-21 14:33:16 -08001073#ifdef CONFIG_DEBUG_TLBFLUSH
Dave Hansen6df46862013-09-11 14:20:24 -07001074#ifdef CONFIG_SMP
Dave Hansen9824cf92013-09-11 14:20:23 -07001075 "nr_tlb_remote_flush",
1076 "nr_tlb_remote_flush_received",
Mel Gormanec659932014-01-21 14:33:16 -08001077#endif /* CONFIG_SMP */
Dave Hansen9824cf92013-09-11 14:20:23 -07001078 "nr_tlb_local_flush_all",
1079 "nr_tlb_local_flush_one",
Mel Gormanec659932014-01-21 14:33:16 -08001080#endif /* CONFIG_DEBUG_TLBFLUSH */
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001081
Davidlohr Bueso4f115142014-06-04 16:06:46 -07001082#ifdef CONFIG_DEBUG_VM_VMACACHE
1083 "vmacache_find_calls",
1084 "vmacache_find_hits",
Davidlohr Buesof5f302e2014-12-12 16:56:10 -08001085 "vmacache_full_flushes",
Davidlohr Bueso4f115142014-06-04 16:06:46 -07001086#endif
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001087#endif /* CONFIG_VM_EVENTS_COUNTERS */
1088};
David Rientjes0d6617c2011-09-14 16:21:05 -07001089#endif /* CONFIG_PROC_FS || CONFIG_SYSFS || CONFIG_NUMA */
KOSAKI Motohirofa25c502011-05-24 17:11:28 -07001090
1091
Andrew Morton3c486872015-02-10 14:09:43 -08001092#if (defined(CONFIG_DEBUG_FS) && defined(CONFIG_COMPACTION)) || \
1093 defined(CONFIG_PROC_FS)
1094static void *frag_start(struct seq_file *m, loff_t *pos)
1095{
1096 pg_data_t *pgdat;
1097 loff_t node = *pos;
1098
1099 for (pgdat = first_online_pgdat();
1100 pgdat && node;
1101 pgdat = next_online_pgdat(pgdat))
1102 --node;
1103
1104 return pgdat;
1105}
1106
1107static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
1108{
1109 pg_data_t *pgdat = (pg_data_t *)arg;
1110
1111 (*pos)++;
1112 return next_online_pgdat(pgdat);
1113}
1114
1115static void frag_stop(struct seq_file *m, void *arg)
1116{
1117}
1118
1119/* Walk all the zones in a node and print using a callback */
1120static void walk_zones_in_node(struct seq_file *m, pg_data_t *pgdat,
1121 void (*print)(struct seq_file *m, pg_data_t *, struct zone *))
1122{
1123 struct zone *zone;
1124 struct zone *node_zones = pgdat->node_zones;
1125 unsigned long flags;
1126
1127 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
1128 if (!populated_zone(zone))
1129 continue;
1130
1131 spin_lock_irqsave(&zone->lock, flags);
1132 print(m, pgdat, zone);
1133 spin_unlock_irqrestore(&zone->lock, flags);
1134 }
1135}
1136#endif
1137
Mel Gormand7a57522010-05-24 14:32:25 -07001138#ifdef CONFIG_PROC_FS
Mel Gorman467c9962007-10-16 01:26:02 -07001139static void frag_show_print(struct seq_file *m, pg_data_t *pgdat,
1140 struct zone *zone)
1141{
1142 int order;
1143
1144 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
1145 for (order = 0; order < MAX_ORDER; ++order)
1146 seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
1147 seq_putc(m, '\n');
1148}
1149
1150/*
1151 * This walks the free areas for each zone.
1152 */
1153static int frag_show(struct seq_file *m, void *arg)
1154{
1155 pg_data_t *pgdat = (pg_data_t *)arg;
1156 walk_zones_in_node(m, pgdat, frag_show_print);
1157 return 0;
1158}
1159
1160static void pagetypeinfo_showfree_print(struct seq_file *m,
1161 pg_data_t *pgdat, struct zone *zone)
1162{
1163 int order, mtype;
1164
1165 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++) {
1166 seq_printf(m, "Node %4d, zone %8s, type %12s ",
1167 pgdat->node_id,
1168 zone->name,
1169 migratetype_names[mtype]);
1170 for (order = 0; order < MAX_ORDER; ++order) {
1171 unsigned long freecount = 0;
1172 struct free_area *area;
1173 struct list_head *curr;
1174
1175 area = &(zone->free_area[order]);
1176
1177 list_for_each(curr, &area->free_list[mtype])
1178 freecount++;
1179 seq_printf(m, "%6lu ", freecount);
1180 }
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001181 seq_putc(m, '\n');
1182 }
Mel Gorman467c9962007-10-16 01:26:02 -07001183}
1184
1185/* Print out the free pages at each order for each migatetype */
1186static int pagetypeinfo_showfree(struct seq_file *m, void *arg)
1187{
1188 int order;
1189 pg_data_t *pgdat = (pg_data_t *)arg;
1190
1191 /* Print header */
1192 seq_printf(m, "%-43s ", "Free pages count per migrate type at order");
1193 for (order = 0; order < MAX_ORDER; ++order)
1194 seq_printf(m, "%6d ", order);
1195 seq_putc(m, '\n');
1196
1197 walk_zones_in_node(m, pgdat, pagetypeinfo_showfree_print);
1198
1199 return 0;
1200}
1201
1202static void pagetypeinfo_showblockcount_print(struct seq_file *m,
1203 pg_data_t *pgdat, struct zone *zone)
1204{
1205 int mtype;
1206 unsigned long pfn;
1207 unsigned long start_pfn = zone->zone_start_pfn;
Cody P Schafer108bcc92013-02-22 16:35:23 -08001208 unsigned long end_pfn = zone_end_pfn(zone);
Mel Gorman467c9962007-10-16 01:26:02 -07001209 unsigned long count[MIGRATE_TYPES] = { 0, };
1210
1211 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
1212 struct page *page;
1213
1214 if (!pfn_valid(pfn))
1215 continue;
1216
1217 page = pfn_to_page(pfn);
Mel Gormaneb335752009-05-13 17:34:48 +01001218
1219 /* Watch for unexpected holes punched in the memmap */
1220 if (!memmap_valid_within(pfn, page, zone))
Mel Gormane80d6a22008-08-14 11:10:14 +01001221 continue;
Mel Gormaneb335752009-05-13 17:34:48 +01001222
Joonsoo Kima91c43c2016-05-19 17:12:10 -07001223 if (page_zone(page) != zone)
1224 continue;
1225
Mel Gorman467c9962007-10-16 01:26:02 -07001226 mtype = get_pageblock_migratetype(page);
1227
Mel Gormane80d6a22008-08-14 11:10:14 +01001228 if (mtype < MIGRATE_TYPES)
1229 count[mtype]++;
Mel Gorman467c9962007-10-16 01:26:02 -07001230 }
1231
1232 /* Print counts */
1233 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
1234 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
1235 seq_printf(m, "%12lu ", count[mtype]);
1236 seq_putc(m, '\n');
1237}
1238
1239/* Print out the free pages at each order for each migratetype */
1240static int pagetypeinfo_showblockcount(struct seq_file *m, void *arg)
1241{
1242 int mtype;
1243 pg_data_t *pgdat = (pg_data_t *)arg;
1244
1245 seq_printf(m, "\n%-23s", "Number of blocks type ");
1246 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
1247 seq_printf(m, "%12s ", migratetype_names[mtype]);
1248 seq_putc(m, '\n');
1249 walk_zones_in_node(m, pgdat, pagetypeinfo_showblockcount_print);
1250
1251 return 0;
1252}
1253
Joonsoo Kim48c96a32014-12-12 16:56:01 -08001254#ifdef CONFIG_PAGE_OWNER
1255static void pagetypeinfo_showmixedcount_print(struct seq_file *m,
1256 pg_data_t *pgdat,
1257 struct zone *zone)
1258{
1259 struct page *page;
1260 struct page_ext *page_ext;
1261 unsigned long pfn = zone->zone_start_pfn, block_end_pfn;
1262 unsigned long end_pfn = pfn + zone->spanned_pages;
1263 unsigned long count[MIGRATE_TYPES] = { 0, };
1264 int pageblock_mt, page_mt;
1265 int i;
1266
1267 /* Scan block by block. First and last block may be incomplete */
1268 pfn = zone->zone_start_pfn;
1269
1270 /*
1271 * Walk the zone in pageblock_nr_pages steps. If a page block spans
1272 * a zone boundary, it will be double counted between zones. This does
1273 * not matter as the mixed block count will still be correct
1274 */
1275 for (; pfn < end_pfn; ) {
1276 if (!pfn_valid(pfn)) {
1277 pfn = ALIGN(pfn + 1, MAX_ORDER_NR_PAGES);
1278 continue;
1279 }
1280
1281 block_end_pfn = ALIGN(pfn + 1, pageblock_nr_pages);
1282 block_end_pfn = min(block_end_pfn, end_pfn);
1283
1284 page = pfn_to_page(pfn);
Mel Gorman0b423ca2016-05-19 17:14:27 -07001285 pageblock_mt = get_pageblock_migratetype(page);
Joonsoo Kim48c96a32014-12-12 16:56:01 -08001286
1287 for (; pfn < block_end_pfn; pfn++) {
1288 if (!pfn_valid_within(pfn))
1289 continue;
1290
1291 page = pfn_to_page(pfn);
Joonsoo Kima91c43c2016-05-19 17:12:10 -07001292
1293 if (page_zone(page) != zone)
1294 continue;
1295
Joonsoo Kim48c96a32014-12-12 16:56:01 -08001296 if (PageBuddy(page)) {
1297 pfn += (1UL << page_order(page)) - 1;
1298 continue;
1299 }
1300
1301 if (PageReserved(page))
1302 continue;
1303
1304 page_ext = lookup_page_ext(page);
Yang Shif86e4272016-06-03 14:55:38 -07001305 if (unlikely(!page_ext))
1306 continue;
Joonsoo Kim48c96a32014-12-12 16:56:01 -08001307
1308 if (!test_bit(PAGE_EXT_OWNER, &page_ext->flags))
1309 continue;
1310
1311 page_mt = gfpflags_to_migratetype(page_ext->gfp_mask);
1312 if (pageblock_mt != page_mt) {
1313 if (is_migrate_cma(pageblock_mt))
1314 count[MIGRATE_MOVABLE]++;
1315 else
1316 count[pageblock_mt]++;
1317
1318 pfn = block_end_pfn;
1319 break;
1320 }
1321 pfn += (1UL << page_ext->order) - 1;
1322 }
1323 }
1324
1325 /* Print counts */
1326 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
1327 for (i = 0; i < MIGRATE_TYPES; i++)
1328 seq_printf(m, "%12lu ", count[i]);
1329 seq_putc(m, '\n');
1330}
1331#endif /* CONFIG_PAGE_OWNER */
1332
1333/*
1334 * Print out the number of pageblocks for each migratetype that contain pages
1335 * of other types. This gives an indication of how well fallbacks are being
1336 * contained by rmqueue_fallback(). It requires information from PAGE_OWNER
1337 * to determine what is going on
1338 */
1339static void pagetypeinfo_showmixedcount(struct seq_file *m, pg_data_t *pgdat)
1340{
1341#ifdef CONFIG_PAGE_OWNER
1342 int mtype;
1343
Vlastimil Babka7dd80b82016-03-15 14:56:12 -07001344 if (!static_branch_unlikely(&page_owner_inited))
Joonsoo Kim48c96a32014-12-12 16:56:01 -08001345 return;
1346
1347 drain_all_pages(NULL);
1348
1349 seq_printf(m, "\n%-23s", "Number of mixed blocks ");
1350 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
1351 seq_printf(m, "%12s ", migratetype_names[mtype]);
1352 seq_putc(m, '\n');
1353
1354 walk_zones_in_node(m, pgdat, pagetypeinfo_showmixedcount_print);
1355#endif /* CONFIG_PAGE_OWNER */
1356}
1357
Mel Gorman467c9962007-10-16 01:26:02 -07001358/*
1359 * This prints out statistics in relation to grouping pages by mobility.
1360 * It is expensive to collect so do not constantly read the file.
1361 */
1362static int pagetypeinfo_show(struct seq_file *m, void *arg)
1363{
1364 pg_data_t *pgdat = (pg_data_t *)arg;
1365
KOSAKI Motohiro41b25a32008-04-30 00:52:13 -07001366 /* check memoryless node */
Lai Jiangshana47b53c2012-12-12 13:51:37 -08001367 if (!node_state(pgdat->node_id, N_MEMORY))
KOSAKI Motohiro41b25a32008-04-30 00:52:13 -07001368 return 0;
1369
Mel Gorman467c9962007-10-16 01:26:02 -07001370 seq_printf(m, "Page block order: %d\n", pageblock_order);
1371 seq_printf(m, "Pages per block: %lu\n", pageblock_nr_pages);
1372 seq_putc(m, '\n');
1373 pagetypeinfo_showfree(m, pgdat);
1374 pagetypeinfo_showblockcount(m, pgdat);
Joonsoo Kim48c96a32014-12-12 16:56:01 -08001375 pagetypeinfo_showmixedcount(m, pgdat);
Mel Gorman467c9962007-10-16 01:26:02 -07001376
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001377 return 0;
1378}
1379
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +04001380static const struct seq_operations fragmentation_op = {
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001381 .start = frag_start,
1382 .next = frag_next,
1383 .stop = frag_stop,
1384 .show = frag_show,
1385};
1386
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +04001387static int fragmentation_open(struct inode *inode, struct file *file)
1388{
1389 return seq_open(file, &fragmentation_op);
1390}
1391
1392static const struct file_operations fragmentation_file_operations = {
1393 .open = fragmentation_open,
1394 .read = seq_read,
1395 .llseek = seq_lseek,
1396 .release = seq_release,
1397};
1398
Alexey Dobriyan74e2e8e2008-10-06 04:15:36 +04001399static const struct seq_operations pagetypeinfo_op = {
Mel Gorman467c9962007-10-16 01:26:02 -07001400 .start = frag_start,
1401 .next = frag_next,
1402 .stop = frag_stop,
1403 .show = pagetypeinfo_show,
1404};
1405
Alexey Dobriyan74e2e8e2008-10-06 04:15:36 +04001406static int pagetypeinfo_open(struct inode *inode, struct file *file)
1407{
1408 return seq_open(file, &pagetypeinfo_op);
1409}
1410
1411static const struct file_operations pagetypeinfo_file_ops = {
1412 .open = pagetypeinfo_open,
1413 .read = seq_read,
1414 .llseek = seq_lseek,
1415 .release = seq_release,
1416};
1417
Mel Gorman467c9962007-10-16 01:26:02 -07001418static void zoneinfo_show_print(struct seq_file *m, pg_data_t *pgdat,
1419 struct zone *zone)
1420{
1421 int i;
1422 seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
1423 seq_printf(m,
1424 "\n pages free %lu"
1425 "\n min %lu"
1426 "\n low %lu"
1427 "\n high %lu"
Mel Gorman599d0c92016-07-28 15:45:31 -07001428 "\n node_scanned %lu"
Mel Gorman467c9962007-10-16 01:26:02 -07001429 "\n spanned %lu"
Jiang Liu9feedc92012-12-12 13:52:12 -08001430 "\n present %lu"
1431 "\n managed %lu",
Mel Gorman88f5acf2011-01-13 15:45:41 -08001432 zone_page_state(zone, NR_FREE_PAGES),
Mel Gorman41858962009-06-16 15:32:12 -07001433 min_wmark_pages(zone),
1434 low_wmark_pages(zone),
1435 high_wmark_pages(zone),
Mel Gorman599d0c92016-07-28 15:45:31 -07001436 node_page_state(zone->zone_pgdat, NR_PAGES_SCANNED),
Mel Gorman467c9962007-10-16 01:26:02 -07001437 zone->spanned_pages,
Jiang Liu9feedc92012-12-12 13:52:12 -08001438 zone->present_pages,
1439 zone->managed_pages);
Mel Gorman467c9962007-10-16 01:26:02 -07001440
1441 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
Mel Gorman599d0c92016-07-28 15:45:31 -07001442 seq_printf(m, "\n %-12s %lu", vmstat_text[i],
Mel Gorman467c9962007-10-16 01:26:02 -07001443 zone_page_state(zone, i));
1444
1445 seq_printf(m,
Mel Gorman3484b2d2014-08-06 16:07:14 -07001446 "\n protection: (%ld",
Mel Gorman467c9962007-10-16 01:26:02 -07001447 zone->lowmem_reserve[0]);
1448 for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
Mel Gorman3484b2d2014-08-06 16:07:14 -07001449 seq_printf(m, ", %ld", zone->lowmem_reserve[i]);
Mel Gorman467c9962007-10-16 01:26:02 -07001450 seq_printf(m,
1451 ")"
1452 "\n pagesets");
1453 for_each_online_cpu(i) {
1454 struct per_cpu_pageset *pageset;
Mel Gorman467c9962007-10-16 01:26:02 -07001455
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001456 pageset = per_cpu_ptr(zone->pageset, i);
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001457 seq_printf(m,
1458 "\n cpu: %i"
1459 "\n count: %i"
1460 "\n high: %i"
1461 "\n batch: %i",
1462 i,
1463 pageset->pcp.count,
1464 pageset->pcp.high,
1465 pageset->pcp.batch);
Mel Gorman467c9962007-10-16 01:26:02 -07001466#ifdef CONFIG_SMP
1467 seq_printf(m, "\n vm stats threshold: %d",
1468 pageset->stat_threshold);
1469#endif
1470 }
1471 seq_printf(m,
Mel Gorman599d0c92016-07-28 15:45:31 -07001472 "\n node_unreclaimable: %u"
1473 "\n start_pfn: %lu"
1474 "\n node_inactive_ratio: %u",
1475 !pgdat_reclaimable(zone->zone_pgdat),
Rik van Riel556adec2008-10-18 20:26:34 -07001476 zone->zone_start_pfn,
Mel Gorman599d0c92016-07-28 15:45:31 -07001477 zone->zone_pgdat->inactive_ratio);
Mel Gorman467c9962007-10-16 01:26:02 -07001478 seq_putc(m, '\n');
1479}
1480
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001481/*
1482 * Output information about zones in @pgdat.
1483 */
1484static int zoneinfo_show(struct seq_file *m, void *arg)
1485{
Mel Gorman467c9962007-10-16 01:26:02 -07001486 pg_data_t *pgdat = (pg_data_t *)arg;
1487 walk_zones_in_node(m, pgdat, zoneinfo_show_print);
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001488 return 0;
1489}
1490
Alexey Dobriyan5c9fe622008-10-06 04:19:42 +04001491static const struct seq_operations zoneinfo_op = {
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001492 .start = frag_start, /* iterate over all zones. The same as in
1493 * fragmentation. */
1494 .next = frag_next,
1495 .stop = frag_stop,
1496 .show = zoneinfo_show,
1497};
1498
Alexey Dobriyan5c9fe622008-10-06 04:19:42 +04001499static int zoneinfo_open(struct inode *inode, struct file *file)
1500{
1501 return seq_open(file, &zoneinfo_op);
1502}
1503
1504static const struct file_operations proc_zoneinfo_file_operations = {
1505 .open = zoneinfo_open,
1506 .read = seq_read,
1507 .llseek = seq_lseek,
1508 .release = seq_release,
1509};
1510
Michael Rubin79da8262010-10-26 14:21:36 -07001511enum writeback_stat_item {
1512 NR_DIRTY_THRESHOLD,
1513 NR_DIRTY_BG_THRESHOLD,
1514 NR_VM_WRITEBACK_STAT_ITEMS,
1515};
1516
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001517static void *vmstat_start(struct seq_file *m, loff_t *pos)
1518{
Christoph Lameter2244b952006-06-30 01:55:33 -07001519 unsigned long *v;
Michael Rubin79da8262010-10-26 14:21:36 -07001520 int i, stat_items_size;
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001521
1522 if (*pos >= ARRAY_SIZE(vmstat_text))
1523 return NULL;
Michael Rubin79da8262010-10-26 14:21:36 -07001524 stat_items_size = NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long) +
Mel Gorman75ef7182016-07-28 15:45:24 -07001525 NR_VM_NODE_STAT_ITEMS * sizeof(unsigned long) +
Michael Rubin79da8262010-10-26 14:21:36 -07001526 NR_VM_WRITEBACK_STAT_ITEMS * sizeof(unsigned long);
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001527
Christoph Lameterf8891e52006-06-30 01:55:45 -07001528#ifdef CONFIG_VM_EVENT_COUNTERS
Michael Rubin79da8262010-10-26 14:21:36 -07001529 stat_items_size += sizeof(struct vm_event_state);
Christoph Lameterf8891e52006-06-30 01:55:45 -07001530#endif
Michael Rubin79da8262010-10-26 14:21:36 -07001531
1532 v = kmalloc(stat_items_size, GFP_KERNEL);
Christoph Lameter2244b952006-06-30 01:55:33 -07001533 m->private = v;
1534 if (!v)
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001535 return ERR_PTR(-ENOMEM);
Christoph Lameter2244b952006-06-30 01:55:33 -07001536 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
1537 v[i] = global_page_state(i);
Michael Rubin79da8262010-10-26 14:21:36 -07001538 v += NR_VM_ZONE_STAT_ITEMS;
1539
Mel Gorman75ef7182016-07-28 15:45:24 -07001540 for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++)
1541 v[i] = global_node_page_state(i);
1542 v += NR_VM_NODE_STAT_ITEMS;
1543
Michael Rubin79da8262010-10-26 14:21:36 -07001544 global_dirty_limits(v + NR_DIRTY_BG_THRESHOLD,
1545 v + NR_DIRTY_THRESHOLD);
1546 v += NR_VM_WRITEBACK_STAT_ITEMS;
1547
Christoph Lameterf8891e52006-06-30 01:55:45 -07001548#ifdef CONFIG_VM_EVENT_COUNTERS
Michael Rubin79da8262010-10-26 14:21:36 -07001549 all_vm_events(v);
1550 v[PGPGIN] /= 2; /* sectors -> kbytes */
1551 v[PGPGOUT] /= 2;
Christoph Lameterf8891e52006-06-30 01:55:45 -07001552#endif
Wu Fengguangff8b16d2010-11-04 01:56:49 +08001553 return (unsigned long *)m->private + *pos;
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001554}
1555
1556static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
1557{
1558 (*pos)++;
1559 if (*pos >= ARRAY_SIZE(vmstat_text))
1560 return NULL;
1561 return (unsigned long *)m->private + *pos;
1562}
1563
1564static int vmstat_show(struct seq_file *m, void *arg)
1565{
1566 unsigned long *l = arg;
1567 unsigned long off = l - (unsigned long *)m->private;
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001568 seq_printf(m, "%s %lu\n", vmstat_text[off], *l);
1569 return 0;
1570}
1571
1572static void vmstat_stop(struct seq_file *m, void *arg)
1573{
1574 kfree(m->private);
1575 m->private = NULL;
1576}
1577
Alexey Dobriyanb6aa44a2008-10-06 04:17:48 +04001578static const struct seq_operations vmstat_op = {
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001579 .start = vmstat_start,
1580 .next = vmstat_next,
1581 .stop = vmstat_stop,
1582 .show = vmstat_show,
1583};
1584
Alexey Dobriyanb6aa44a2008-10-06 04:17:48 +04001585static int vmstat_open(struct inode *inode, struct file *file)
1586{
1587 return seq_open(file, &vmstat_op);
1588}
1589
1590static const struct file_operations proc_vmstat_file_operations = {
1591 .open = vmstat_open,
1592 .read = seq_read,
1593 .llseek = seq_lseek,
1594 .release = seq_release,
1595};
Christoph Lameterf6ac2352006-06-30 01:55:32 -07001596#endif /* CONFIG_PROC_FS */
1597
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001598#ifdef CONFIG_SMP
Michal Hocko373ccbe2015-12-11 13:40:32 -08001599static struct workqueue_struct *vmstat_wq;
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001600static DEFINE_PER_CPU(struct delayed_work, vmstat_work);
Christoph Lameter77461ab2007-05-09 02:35:13 -07001601int sysctl_stat_interval __read_mostly = HZ;
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001602
Hugh Dickins52b6f462016-05-19 17:12:50 -07001603#ifdef CONFIG_PROC_FS
1604static void refresh_vm_stats(struct work_struct *work)
1605{
1606 refresh_cpu_vm_stats(true);
1607}
1608
1609int vmstat_refresh(struct ctl_table *table, int write,
1610 void __user *buffer, size_t *lenp, loff_t *ppos)
1611{
1612 long val;
1613 int err;
1614 int i;
1615
1616 /*
1617 * The regular update, every sysctl_stat_interval, may come later
1618 * than expected: leaving a significant amount in per_cpu buckets.
1619 * This is particularly misleading when checking a quantity of HUGE
1620 * pages, immediately after running a test. /proc/sys/vm/stat_refresh,
1621 * which can equally be echo'ed to or cat'ted from (by root),
1622 * can be used to update the stats just before reading them.
1623 *
1624 * Oh, and since global_page_state() etc. are so careful to hide
1625 * transiently negative values, report an error here if any of
1626 * the stats is negative, so we know to go looking for imbalance.
1627 */
1628 err = schedule_on_each_cpu(refresh_vm_stats);
1629 if (err)
1630 return err;
1631 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) {
Mel Gorman75ef7182016-07-28 15:45:24 -07001632 val = atomic_long_read(&vm_zone_stat[i]);
Hugh Dickins52b6f462016-05-19 17:12:50 -07001633 if (val < 0) {
1634 switch (i) {
Hugh Dickins52b6f462016-05-19 17:12:50 -07001635 case NR_PAGES_SCANNED:
1636 /*
Mel Gormane6cbd7f2016-07-28 15:46:50 -07001637 * This is often seen to go negative in
Hugh Dickins52b6f462016-05-19 17:12:50 -07001638 * recent kernels, but not to go permanently
1639 * negative. Whilst it would be nicer not to
1640 * have exceptions, rooting them out would be
1641 * another task, of rather low priority.
1642 */
1643 break;
1644 default:
1645 pr_warn("%s: %s %ld\n",
1646 __func__, vmstat_text[i], val);
1647 err = -EINVAL;
1648 break;
1649 }
1650 }
1651 }
1652 if (err)
1653 return err;
1654 if (write)
1655 *ppos += *lenp;
1656 else
1657 *lenp = 0;
1658 return 0;
1659}
1660#endif /* CONFIG_PROC_FS */
1661
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001662static void vmstat_update(struct work_struct *w)
1663{
Christoph Lameter0eb77e92016-01-14 15:21:40 -08001664 if (refresh_cpu_vm_stats(true)) {
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001665 /*
1666 * Counters were updated so we expect more updates
1667 * to occur in the future. Keep on running the
1668 * update worker thread.
1669 */
Christoph Lameter7b8da4c2016-05-20 16:58:21 -07001670 queue_delayed_work_on(smp_processor_id(), vmstat_wq,
Michal Hockof01f17d2016-02-05 15:36:24 -08001671 this_cpu_ptr(&vmstat_work),
1672 round_jiffies_relative(sysctl_stat_interval));
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001673 }
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001674}
1675
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001676/*
Christoph Lameter0eb77e92016-01-14 15:21:40 -08001677 * Switch off vmstat processing and then fold all the remaining differentials
1678 * until the diffs stay at zero. The function is used by NOHZ and can only be
1679 * invoked when tick processing is not active.
1680 */
Christoph Lameter0eb77e92016-01-14 15:21:40 -08001681/*
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001682 * Check if the diffs for a certain cpu indicate that
1683 * an update is needed.
1684 */
1685static bool need_update(int cpu)
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001686{
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001687 struct zone *zone;
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001688
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001689 for_each_populated_zone(zone) {
1690 struct per_cpu_pageset *p = per_cpu_ptr(zone->pageset, cpu);
1691
1692 BUILD_BUG_ON(sizeof(p->vm_stat_diff[0]) != 1);
1693 /*
1694 * The fast way of checking if there are any vmstat diffs.
1695 * This works because the diffs are byte sized items.
1696 */
1697 if (memchr_inv(p->vm_stat_diff, 0, NR_VM_ZONE_STAT_ITEMS))
1698 return true;
1699
1700 }
1701 return false;
1702}
1703
Christoph Lameter7b8da4c2016-05-20 16:58:21 -07001704/*
1705 * Switch off vmstat processing and then fold all the remaining differentials
1706 * until the diffs stay at zero. The function is used by NOHZ and can only be
1707 * invoked when tick processing is not active.
1708 */
Michal Hockof01f17d2016-02-05 15:36:24 -08001709void quiet_vmstat(void)
1710{
1711 if (system_state != SYSTEM_RUNNING)
1712 return;
1713
Christoph Lameter7b8da4c2016-05-20 16:58:21 -07001714 if (!delayed_work_pending(this_cpu_ptr(&vmstat_work)))
Michal Hockof01f17d2016-02-05 15:36:24 -08001715 return;
1716
1717 if (!need_update(smp_processor_id()))
1718 return;
1719
1720 /*
1721 * Just refresh counters and do not care about the pending delayed
1722 * vmstat_update. It doesn't fire that often to matter and canceling
1723 * it would be too expensive from this path.
1724 * vmstat_shepherd will take care about that for us.
1725 */
1726 refresh_cpu_vm_stats(false);
1727}
1728
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001729/*
1730 * Shepherd worker thread that checks the
1731 * differentials of processors that have their worker
1732 * threads for vm statistics updates disabled because of
1733 * inactivity.
1734 */
1735static void vmstat_shepherd(struct work_struct *w);
1736
Christoph Lameter0eb77e92016-01-14 15:21:40 -08001737static DECLARE_DEFERRABLE_WORK(shepherd, vmstat_shepherd);
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001738
1739static void vmstat_shepherd(struct work_struct *w)
1740{
1741 int cpu;
1742
1743 get_online_cpus();
1744 /* Check processors whose vmstat worker threads have been disabled */
Christoph Lameter7b8da4c2016-05-20 16:58:21 -07001745 for_each_online_cpu(cpu) {
Michal Hockof01f17d2016-02-05 15:36:24 -08001746 struct delayed_work *dw = &per_cpu(vmstat_work, cpu);
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001747
Christoph Lameter7b8da4c2016-05-20 16:58:21 -07001748 if (!delayed_work_pending(dw) && need_update(cpu))
1749 queue_delayed_work_on(cpu, vmstat_wq, dw, 0);
Michal Hockof01f17d2016-02-05 15:36:24 -08001750 }
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001751 put_online_cpus();
1752
1753 schedule_delayed_work(&shepherd,
1754 round_jiffies_relative(sysctl_stat_interval));
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001755}
1756
1757static void __init start_shepherd_timer(void)
1758{
1759 int cpu;
1760
1761 for_each_possible_cpu(cpu)
Michal Hockoccde8bd2016-02-05 15:36:27 -08001762 INIT_DEFERRABLE_WORK(per_cpu_ptr(&vmstat_work, cpu),
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001763 vmstat_update);
1764
Michal Hocko751e5f52016-01-08 11:18:29 +01001765 vmstat_wq = alloc_workqueue("vmstat", WQ_FREEZABLE|WQ_MEM_RECLAIM, 0);
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001766 schedule_delayed_work(&shepherd,
1767 round_jiffies_relative(sysctl_stat_interval));
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001768}
1769
Toshi Kani807a1bd2013-11-12 15:08:13 -08001770static void vmstat_cpu_dead(int node)
1771{
1772 int cpu;
1773
1774 get_online_cpus();
1775 for_each_online_cpu(cpu)
1776 if (cpu_to_node(cpu) == node)
1777 goto end;
1778
1779 node_clear_state(node, N_CPU);
1780end:
1781 put_online_cpus();
1782}
1783
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001784/*
1785 * Use the cpu notifier to insure that the thresholds are recalculated
1786 * when necessary.
1787 */
Paul Gortmaker0db06282013-06-19 14:53:51 -04001788static int vmstat_cpuup_callback(struct notifier_block *nfb,
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001789 unsigned long action,
1790 void *hcpu)
1791{
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001792 long cpu = (long)hcpu;
1793
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001794 switch (action) {
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001795 case CPU_ONLINE:
1796 case CPU_ONLINE_FROZEN:
KAMEZAWA Hiroyuki5ee28a42010-09-09 16:38:14 -07001797 refresh_zone_stat_thresholds();
Christoph Lameterad596922010-01-05 15:34:51 +09001798 node_set_state(cpu_to_node(cpu), N_CPU);
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001799 break;
1800 case CPU_DOWN_PREPARE:
1801 case CPU_DOWN_PREPARE_FROZEN:
Tejun Heoafe2c512010-12-14 16:21:17 +01001802 cancel_delayed_work_sync(&per_cpu(vmstat_work, cpu));
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001803 break;
1804 case CPU_DOWN_FAILED:
1805 case CPU_DOWN_FAILED_FROZEN:
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001806 break;
Andy Whitcroftce421c72006-12-06 20:33:08 -08001807 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07001808 case CPU_DEAD_FROZEN:
Andy Whitcroftce421c72006-12-06 20:33:08 -08001809 refresh_zone_stat_thresholds();
Toshi Kani807a1bd2013-11-12 15:08:13 -08001810 vmstat_cpu_dead(cpu_to_node(cpu));
Andy Whitcroftce421c72006-12-06 20:33:08 -08001811 break;
1812 default:
1813 break;
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001814 }
1815 return NOTIFY_OK;
1816}
1817
Paul Gortmaker0db06282013-06-19 14:53:51 -04001818static struct notifier_block vmstat_notifier =
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001819 { &vmstat_cpuup_callback, NULL, 0 };
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +04001820#endif
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001821
Adrian Bunke2fc88d2007-10-16 01:26:27 -07001822static int __init setup_vmstat(void)
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001823{
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +04001824#ifdef CONFIG_SMP
Srivatsa S. Bhat0be94ba2014-03-11 02:12:27 +05301825 cpu_notifier_register_begin();
1826 __register_cpu_notifier(&vmstat_notifier);
Christoph Lameterd1187ed2007-05-09 02:35:12 -07001827
Christoph Lameter7cc36bb2014-10-09 15:29:43 -07001828 start_shepherd_timer();
Srivatsa S. Bhat0be94ba2014-03-11 02:12:27 +05301829 cpu_notifier_register_done();
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +04001830#endif
1831#ifdef CONFIG_PROC_FS
1832 proc_create("buddyinfo", S_IRUGO, NULL, &fragmentation_file_operations);
Alexey Dobriyan74e2e8e2008-10-06 04:15:36 +04001833 proc_create("pagetypeinfo", S_IRUGO, NULL, &pagetypeinfo_file_ops);
Alexey Dobriyanb6aa44a2008-10-06 04:17:48 +04001834 proc_create("vmstat", S_IRUGO, NULL, &proc_vmstat_file_operations);
Alexey Dobriyan5c9fe622008-10-06 04:19:42 +04001835 proc_create("zoneinfo", S_IRUGO, NULL, &proc_zoneinfo_file_operations);
Alexey Dobriyan8f32f7e2008-10-06 04:13:52 +04001836#endif
Christoph Lameterdf9ecab2006-08-31 21:27:35 -07001837 return 0;
1838}
1839module_init(setup_vmstat)
Mel Gormand7a57522010-05-24 14:32:25 -07001840
1841#if defined(CONFIG_DEBUG_FS) && defined(CONFIG_COMPACTION)
Mel Gormand7a57522010-05-24 14:32:25 -07001842
1843/*
1844 * Return an index indicating how much of the available free memory is
1845 * unusable for an allocation of the requested size.
1846 */
1847static int unusable_free_index(unsigned int order,
1848 struct contig_page_info *info)
1849{
1850 /* No free memory is interpreted as all free memory is unusable */
1851 if (info->free_pages == 0)
1852 return 1000;
1853
1854 /*
1855 * Index should be a value between 0 and 1. Return a value to 3
1856 * decimal places.
1857 *
1858 * 0 => no fragmentation
1859 * 1 => high fragmentation
1860 */
1861 return div_u64((info->free_pages - (info->free_blocks_suitable << order)) * 1000ULL, info->free_pages);
1862
1863}
1864
1865static void unusable_show_print(struct seq_file *m,
1866 pg_data_t *pgdat, struct zone *zone)
1867{
1868 unsigned int order;
1869 int index;
1870 struct contig_page_info info;
1871
1872 seq_printf(m, "Node %d, zone %8s ",
1873 pgdat->node_id,
1874 zone->name);
1875 for (order = 0; order < MAX_ORDER; ++order) {
1876 fill_contig_page_info(zone, order, &info);
1877 index = unusable_free_index(order, &info);
1878 seq_printf(m, "%d.%03d ", index / 1000, index % 1000);
1879 }
1880
1881 seq_putc(m, '\n');
1882}
1883
1884/*
1885 * Display unusable free space index
1886 *
1887 * The unusable free space index measures how much of the available free
1888 * memory cannot be used to satisfy an allocation of a given size and is a
1889 * value between 0 and 1. The higher the value, the more of free memory is
1890 * unusable and by implication, the worse the external fragmentation is. This
1891 * can be expressed as a percentage by multiplying by 100.
1892 */
1893static int unusable_show(struct seq_file *m, void *arg)
1894{
1895 pg_data_t *pgdat = (pg_data_t *)arg;
1896
1897 /* check memoryless node */
Lai Jiangshana47b53c2012-12-12 13:51:37 -08001898 if (!node_state(pgdat->node_id, N_MEMORY))
Mel Gormand7a57522010-05-24 14:32:25 -07001899 return 0;
1900
1901 walk_zones_in_node(m, pgdat, unusable_show_print);
1902
1903 return 0;
1904}
1905
1906static const struct seq_operations unusable_op = {
1907 .start = frag_start,
1908 .next = frag_next,
1909 .stop = frag_stop,
1910 .show = unusable_show,
1911};
1912
1913static int unusable_open(struct inode *inode, struct file *file)
1914{
1915 return seq_open(file, &unusable_op);
1916}
1917
1918static const struct file_operations unusable_file_ops = {
1919 .open = unusable_open,
1920 .read = seq_read,
1921 .llseek = seq_lseek,
1922 .release = seq_release,
1923};
1924
Mel Gormanf1a5ab12010-05-24 14:32:26 -07001925static void extfrag_show_print(struct seq_file *m,
1926 pg_data_t *pgdat, struct zone *zone)
1927{
1928 unsigned int order;
1929 int index;
1930
1931 /* Alloc on stack as interrupts are disabled for zone walk */
1932 struct contig_page_info info;
1933
1934 seq_printf(m, "Node %d, zone %8s ",
1935 pgdat->node_id,
1936 zone->name);
1937 for (order = 0; order < MAX_ORDER; ++order) {
1938 fill_contig_page_info(zone, order, &info);
Mel Gorman56de7262010-05-24 14:32:30 -07001939 index = __fragmentation_index(order, &info);
Mel Gormanf1a5ab12010-05-24 14:32:26 -07001940 seq_printf(m, "%d.%03d ", index / 1000, index % 1000);
1941 }
1942
1943 seq_putc(m, '\n');
1944}
1945
1946/*
1947 * Display fragmentation index for orders that allocations would fail for
1948 */
1949static int extfrag_show(struct seq_file *m, void *arg)
1950{
1951 pg_data_t *pgdat = (pg_data_t *)arg;
1952
1953 walk_zones_in_node(m, pgdat, extfrag_show_print);
1954
1955 return 0;
1956}
1957
1958static const struct seq_operations extfrag_op = {
1959 .start = frag_start,
1960 .next = frag_next,
1961 .stop = frag_stop,
1962 .show = extfrag_show,
1963};
1964
1965static int extfrag_open(struct inode *inode, struct file *file)
1966{
1967 return seq_open(file, &extfrag_op);
1968}
1969
1970static const struct file_operations extfrag_file_ops = {
1971 .open = extfrag_open,
1972 .read = seq_read,
1973 .llseek = seq_lseek,
1974 .release = seq_release,
1975};
1976
Mel Gormand7a57522010-05-24 14:32:25 -07001977static int __init extfrag_debug_init(void)
1978{
Sasikantha babubde8bd82012-05-29 15:06:22 -07001979 struct dentry *extfrag_debug_root;
1980
Mel Gormand7a57522010-05-24 14:32:25 -07001981 extfrag_debug_root = debugfs_create_dir("extfrag", NULL);
1982 if (!extfrag_debug_root)
1983 return -ENOMEM;
1984
1985 if (!debugfs_create_file("unusable_index", 0444,
1986 extfrag_debug_root, NULL, &unusable_file_ops))
Sasikantha babubde8bd82012-05-29 15:06:22 -07001987 goto fail;
Mel Gormand7a57522010-05-24 14:32:25 -07001988
Mel Gormanf1a5ab12010-05-24 14:32:26 -07001989 if (!debugfs_create_file("extfrag_index", 0444,
1990 extfrag_debug_root, NULL, &extfrag_file_ops))
Sasikantha babubde8bd82012-05-29 15:06:22 -07001991 goto fail;
Mel Gormanf1a5ab12010-05-24 14:32:26 -07001992
Mel Gormand7a57522010-05-24 14:32:25 -07001993 return 0;
Sasikantha babubde8bd82012-05-29 15:06:22 -07001994fail:
1995 debugfs_remove_recursive(extfrag_debug_root);
1996 return -ENOMEM;
Mel Gormand7a57522010-05-24 14:32:25 -07001997}
1998
1999module_init(extfrag_debug_init);
2000#endif