2
* kernel/time/sched_debug.c
6
* Copyright(C) 2007, Red Hat, Inc., Ingo Molnar
8
* This program is free software; you can redistribute it and/or modify
9
* it under the terms of the GNU General Public License version 2 as
10
* published by the Free Software Foundation.
13
#include <linux/proc_fs.h>
14
#include <linux/sched.h>
15
#include <linux/seq_file.h>
16
#include <linux/kallsyms.h>
17
#include <linux/utsname.h>
19
static DEFINE_SPINLOCK(sched_debug_lock);
22
* This allows printing both to /proc/sched_debug and
25
#define SEQ_printf(m, x...) \
34
* Ease the printing of nsec fields:
36
static long long nsec_high(unsigned long long nsec)
38
if ((long long)nsec < 0) {
40
do_div(nsec, 1000000);
43
do_div(nsec, 1000000);
48
static unsigned long nsec_low(unsigned long long nsec)
50
if ((long long)nsec < 0)
53
return do_div(nsec, 1000000);
56
#define SPLIT_NS(x) nsec_high(x), nsec_low(x)
58
#ifdef CONFIG_FAIR_GROUP_SCHED
59
static void print_cfs_group_stats(struct seq_file *m, int cpu, struct task_group *tg)
61
struct sched_entity *se = tg->se[cpu];
66
SEQ_printf(m, " .%-30s: %lld\n", #F, (long long)F)
68
SEQ_printf(m, " .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F))
72
PN(se->sum_exec_runtime);
73
#ifdef CONFIG_SCHEDSTATS
74
PN(se->statistics.wait_start);
75
PN(se->statistics.sleep_start);
76
PN(se->statistics.block_start);
77
PN(se->statistics.sleep_max);
78
PN(se->statistics.block_max);
79
PN(se->statistics.exec_max);
80
PN(se->statistics.slice_max);
81
PN(se->statistics.wait_max);
82
PN(se->statistics.wait_sum);
83
P(se->statistics.wait_count);
91
#ifdef CONFIG_CGROUP_SCHED
92
static char group_path[PATH_MAX];
94
static char *task_group_path(struct task_group *tg)
96
if (autogroup_path(tg, group_path, PATH_MAX))
100
* May be NULL if the underlying cgroup isn't fully-created yet
102
if (!tg->css.cgroup) {
103
group_path[0] = '\0';
106
cgroup_path(tg->css.cgroup, group_path, PATH_MAX);
112
print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
119
SEQ_printf(m, "%15s %5d %9Ld.%06ld %9Ld %5d ",
121
SPLIT_NS(p->se.vruntime),
122
(long long)(p->nvcsw + p->nivcsw),
124
#ifdef CONFIG_SCHEDSTATS
125
SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
126
SPLIT_NS(p->se.vruntime),
127
SPLIT_NS(p->se.sum_exec_runtime),
128
SPLIT_NS(p->se.statistics.sum_sleep_runtime));
130
SEQ_printf(m, "%15Ld %15Ld %15Ld.%06ld %15Ld.%06ld %15Ld.%06ld",
131
0LL, 0LL, 0LL, 0L, 0LL, 0L, 0LL, 0L);
133
#ifdef CONFIG_CGROUP_SCHED
134
SEQ_printf(m, " %s", task_group_path(task_group(p)));
140
static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
142
struct task_struct *g, *p;
146
"\nrunnable tasks:\n"
147
" task PID tree-key switches prio"
148
" exec-runtime sum-exec sum-sleep\n"
149
"------------------------------------------------------"
150
"----------------------------------------------------\n");
152
read_lock_irqsave(&tasklist_lock, flags);
154
do_each_thread(g, p) {
155
if (!p->on_rq || task_cpu(p) != rq_cpu)
158
print_task(m, rq, p);
159
} while_each_thread(g, p);
161
read_unlock_irqrestore(&tasklist_lock, flags);
164
void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
166
s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
167
spread, rq0_min_vruntime, spread0;
168
struct rq *rq = cpu_rq(cpu);
169
struct sched_entity *last;
172
#ifdef CONFIG_FAIR_GROUP_SCHED
173
SEQ_printf(m, "\ncfs_rq[%d]:%s\n", cpu, task_group_path(cfs_rq->tg));
175
SEQ_printf(m, "\ncfs_rq[%d]:\n", cpu);
177
SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "exec_clock",
178
SPLIT_NS(cfs_rq->exec_clock));
180
raw_spin_lock_irqsave(&rq->lock, flags);
181
if (cfs_rq->rb_leftmost)
182
MIN_vruntime = (__pick_first_entity(cfs_rq))->vruntime;
183
last = __pick_last_entity(cfs_rq);
185
max_vruntime = last->vruntime;
186
min_vruntime = cfs_rq->min_vruntime;
187
rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
188
raw_spin_unlock_irqrestore(&rq->lock, flags);
189
SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "MIN_vruntime",
190
SPLIT_NS(MIN_vruntime));
191
SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "min_vruntime",
192
SPLIT_NS(min_vruntime));
193
SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "max_vruntime",
194
SPLIT_NS(max_vruntime));
195
spread = max_vruntime - MIN_vruntime;
196
SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread",
198
spread0 = min_vruntime - rq0_min_vruntime;
199
SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread0",
201
SEQ_printf(m, " .%-30s: %d\n", "nr_spread_over",
202
cfs_rq->nr_spread_over);
203
SEQ_printf(m, " .%-30s: %ld\n", "nr_running", cfs_rq->nr_running);
204
SEQ_printf(m, " .%-30s: %ld\n", "load", cfs_rq->load.weight);
205
#ifdef CONFIG_FAIR_GROUP_SCHED
207
SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "load_avg",
208
SPLIT_NS(cfs_rq->load_avg));
209
SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "load_period",
210
SPLIT_NS(cfs_rq->load_period));
211
SEQ_printf(m, " .%-30s: %ld\n", "load_contrib",
212
cfs_rq->load_contribution);
213
SEQ_printf(m, " .%-30s: %d\n", "load_tg",
214
atomic_read(&cfs_rq->tg->load_weight));
217
print_cfs_group_stats(m, cpu, cfs_rq->tg);
221
void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
223
#ifdef CONFIG_RT_GROUP_SCHED
224
SEQ_printf(m, "\nrt_rq[%d]:%s\n", cpu, task_group_path(rt_rq->tg));
226
SEQ_printf(m, "\nrt_rq[%d]:\n", cpu);
230
SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
232
SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
243
extern __read_mostly int sched_clock_running;
245
static void print_cpu(struct seq_file *m, int cpu)
247
struct rq *rq = cpu_rq(cpu);
252
unsigned int freq = cpu_khz ? : 1;
254
SEQ_printf(m, "\ncpu#%d, %u.%03u MHz\n",
255
cpu, freq / 1000, (freq % 1000));
258
SEQ_printf(m, "\ncpu#%d\n", cpu);
262
SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rq->x))
264
SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
267
SEQ_printf(m, " .%-30s: %lu\n", "load",
271
P(nr_uninterruptible);
283
#ifdef CONFIG_SCHEDSTATS
284
#define P(n) SEQ_printf(m, " .%-30s: %d\n", #n, rq->n);
285
#define P64(n) SEQ_printf(m, " .%-30s: %Ld\n", #n, rq->n);
302
spin_lock_irqsave(&sched_debug_lock, flags);
303
print_cfs_stats(m, cpu);
304
print_rt_stats(m, cpu);
307
print_rq(m, rq, cpu);
309
spin_unlock_irqrestore(&sched_debug_lock, flags);
312
static const char *sched_tunable_scaling_names[] = {
318
static int sched_debug_show(struct seq_file *m, void *v)
320
u64 ktime, sched_clk, cpu_clk;
324
local_irq_save(flags);
325
ktime = ktime_to_ns(ktime_get());
326
sched_clk = sched_clock();
327
cpu_clk = local_clock();
328
local_irq_restore(flags);
330
SEQ_printf(m, "Sched Debug Version: v0.10, %s %.*s\n",
331
init_utsname()->release,
332
(int)strcspn(init_utsname()->version, " "),
333
init_utsname()->version);
336
SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
338
SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
343
#ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
344
P(sched_clock_stable);
350
SEQ_printf(m, "sysctl_sched\n");
353
SEQ_printf(m, " .%-40s: %Ld\n", #x, (long long)(x))
355
SEQ_printf(m, " .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
356
PN(sysctl_sched_latency);
357
PN(sysctl_sched_min_granularity);
358
PN(sysctl_sched_wakeup_granularity);
359
P(sysctl_sched_child_runs_first);
360
P(sysctl_sched_features);
364
SEQ_printf(m, " .%-40s: %d (%s)\n", "sysctl_sched_tunable_scaling",
365
sysctl_sched_tunable_scaling,
366
sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
368
for_each_online_cpu(cpu)
376
static void sysrq_sched_debug_show(void)
378
sched_debug_show(NULL, NULL);
381
static int sched_debug_open(struct inode *inode, struct file *filp)
383
return single_open(filp, sched_debug_show, NULL);
386
static const struct file_operations sched_debug_fops = {
387
.open = sched_debug_open,
390
.release = single_release,
393
static int __init init_sched_debug_procfs(void)
395
struct proc_dir_entry *pe;
397
pe = proc_create("sched_debug", 0444, NULL, &sched_debug_fops);
403
__initcall(init_sched_debug_procfs);
405
void proc_sched_show_task(struct task_struct *p, struct seq_file *m)
407
unsigned long nr_switches;
409
SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, p->pid,
412
"---------------------------------------------------------\n");
414
SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)F)
416
SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)p->F)
418
SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
420
SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
424
PN(se.sum_exec_runtime);
426
nr_switches = p->nvcsw + p->nivcsw;
428
#ifdef CONFIG_SCHEDSTATS
429
PN(se.statistics.wait_start);
430
PN(se.statistics.sleep_start);
431
PN(se.statistics.block_start);
432
PN(se.statistics.sleep_max);
433
PN(se.statistics.block_max);
434
PN(se.statistics.exec_max);
435
PN(se.statistics.slice_max);
436
PN(se.statistics.wait_max);
437
PN(se.statistics.wait_sum);
438
P(se.statistics.wait_count);
439
PN(se.statistics.iowait_sum);
440
P(se.statistics.iowait_count);
442
P(se.statistics.nr_migrations_cold);
443
P(se.statistics.nr_failed_migrations_affine);
444
P(se.statistics.nr_failed_migrations_running);
445
P(se.statistics.nr_failed_migrations_hot);
446
P(se.statistics.nr_forced_migrations);
447
P(se.statistics.nr_wakeups);
448
P(se.statistics.nr_wakeups_sync);
449
P(se.statistics.nr_wakeups_migrate);
450
P(se.statistics.nr_wakeups_local);
451
P(se.statistics.nr_wakeups_remote);
452
P(se.statistics.nr_wakeups_affine);
453
P(se.statistics.nr_wakeups_affine_attempts);
454
P(se.statistics.nr_wakeups_passive);
455
P(se.statistics.nr_wakeups_idle);
458
u64 avg_atom, avg_per_cpu;
460
avg_atom = p->se.sum_exec_runtime;
462
do_div(avg_atom, nr_switches);
466
avg_per_cpu = p->se.sum_exec_runtime;
467
if (p->se.nr_migrations) {
468
avg_per_cpu = div64_u64(avg_per_cpu,
469
p->se.nr_migrations);
479
SEQ_printf(m, "%-35s:%21Ld\n",
480
"nr_voluntary_switches", (long long)p->nvcsw);
481
SEQ_printf(m, "%-35s:%21Ld\n",
482
"nr_involuntary_switches", (long long)p->nivcsw);
493
unsigned int this_cpu = raw_smp_processor_id();
496
t0 = cpu_clock(this_cpu);
497
t1 = cpu_clock(this_cpu);
498
SEQ_printf(m, "%-35s:%21Ld\n",
499
"clock-delta", (long long)(t1-t0));
503
void proc_sched_set_task(struct task_struct *p)
505
#ifdef CONFIG_SCHEDSTATS
506
memset(&p->se.statistics, 0, sizeof(p->se.statistics));