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>
18 #include <linux/mempolicy.h>
22 static DEFINE_SPINLOCK(sched_debug_lock
);
25 * This allows printing both to /proc/sched_debug and
28 #define SEQ_printf(m, x...) \
37 * Ease the printing of nsec fields:
39 static long long nsec_high(unsigned long long nsec
)
41 if ((long long)nsec
< 0) {
43 do_div(nsec
, 1000000);
46 do_div(nsec
, 1000000);
51 static unsigned long nsec_low(unsigned long long nsec
)
53 if ((long long)nsec
< 0)
56 return do_div(nsec
, 1000000);
59 #define SPLIT_NS(x) nsec_high(x), nsec_low(x)
61 #ifdef CONFIG_FAIR_GROUP_SCHED
62 static void print_cfs_group_stats(struct seq_file
*m
, int cpu
, struct task_group
*tg
)
64 struct sched_entity
*se
= tg
->se
[cpu
];
67 SEQ_printf(m, " .%-30s: %lld\n", #F, (long long)F)
69 SEQ_printf(m, " .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F))
72 struct sched_avg
*avg
= &cpu_rq(cpu
)->avg
;
73 P(avg
->runnable_avg_sum
);
81 PN(se
->sum_exec_runtime
);
82 #ifdef CONFIG_SCHEDSTATS
83 PN(se
->statistics
.wait_start
);
84 PN(se
->statistics
.sleep_start
);
85 PN(se
->statistics
.block_start
);
86 PN(se
->statistics
.sleep_max
);
87 PN(se
->statistics
.block_max
);
88 PN(se
->statistics
.exec_max
);
89 PN(se
->statistics
.slice_max
);
90 PN(se
->statistics
.wait_max
);
91 PN(se
->statistics
.wait_sum
);
92 P(se
->statistics
.wait_count
);
96 P(se
->avg
.runnable_avg_sum
);
97 P(se
->avg
.running_avg_sum
);
98 P(se
->avg
.avg_period
);
99 P(se
->avg
.load_avg_contrib
);
100 P(se
->avg
.utilization_avg_contrib
);
101 P(se
->avg
.decay_count
);
108 #ifdef CONFIG_CGROUP_SCHED
109 static char group_path
[PATH_MAX
];
111 static char *task_group_path(struct task_group
*tg
)
113 if (autogroup_path(tg
, group_path
, PATH_MAX
))
116 return cgroup_path(tg
->css
.cgroup
, group_path
, PATH_MAX
);
121 print_task(struct seq_file
*m
, struct rq
*rq
, struct task_struct
*p
)
128 SEQ_printf(m
, "%15s %5d %9Ld.%06ld %9Ld %5d ",
129 p
->comm
, task_pid_nr(p
),
130 SPLIT_NS(p
->se
.vruntime
),
131 (long long)(p
->nvcsw
+ p
->nivcsw
),
133 #ifdef CONFIG_SCHEDSTATS
134 SEQ_printf(m
, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
135 SPLIT_NS(p
->se
.vruntime
),
136 SPLIT_NS(p
->se
.sum_exec_runtime
),
137 SPLIT_NS(p
->se
.statistics
.sum_sleep_runtime
));
139 SEQ_printf(m
, "%15Ld %15Ld %15Ld.%06ld %15Ld.%06ld %15Ld.%06ld",
140 0LL, 0LL, 0LL, 0L, 0LL, 0L, 0LL, 0L);
142 #ifdef CONFIG_NUMA_BALANCING
143 SEQ_printf(m
, " %d", task_node(p
));
145 #ifdef CONFIG_CGROUP_SCHED
146 SEQ_printf(m
, " %s", task_group_path(task_group(p
)));
152 static void print_rq(struct seq_file
*m
, struct rq
*rq
, int rq_cpu
)
154 struct task_struct
*g
, *p
;
157 "\nrunnable tasks:\n"
158 " task PID tree-key switches prio"
159 " exec-runtime sum-exec sum-sleep\n"
160 "------------------------------------------------------"
161 "----------------------------------------------------\n");
164 for_each_process_thread(g
, p
) {
165 if (task_cpu(p
) != rq_cpu
)
168 print_task(m
, rq
, p
);
173 void print_cfs_rq(struct seq_file
*m
, int cpu
, struct cfs_rq
*cfs_rq
)
175 s64 MIN_vruntime
= -1, min_vruntime
, max_vruntime
= -1,
176 spread
, rq0_min_vruntime
, spread0
;
177 struct rq
*rq
= cpu_rq(cpu
);
178 struct sched_entity
*last
;
181 #ifdef CONFIG_FAIR_GROUP_SCHED
182 SEQ_printf(m
, "\ncfs_rq[%d]:%s\n", cpu
, task_group_path(cfs_rq
->tg
));
184 SEQ_printf(m
, "\ncfs_rq[%d]:\n", cpu
);
186 SEQ_printf(m
, " .%-30s: %Ld.%06ld\n", "exec_clock",
187 SPLIT_NS(cfs_rq
->exec_clock
));
189 raw_spin_lock_irqsave(&rq
->lock
, flags
);
190 if (cfs_rq
->rb_leftmost
)
191 MIN_vruntime
= (__pick_first_entity(cfs_rq
))->vruntime
;
192 last
= __pick_last_entity(cfs_rq
);
194 max_vruntime
= last
->vruntime
;
195 min_vruntime
= cfs_rq
->min_vruntime
;
196 rq0_min_vruntime
= cpu_rq(0)->cfs
.min_vruntime
;
197 raw_spin_unlock_irqrestore(&rq
->lock
, flags
);
198 SEQ_printf(m
, " .%-30s: %Ld.%06ld\n", "MIN_vruntime",
199 SPLIT_NS(MIN_vruntime
));
200 SEQ_printf(m
, " .%-30s: %Ld.%06ld\n", "min_vruntime",
201 SPLIT_NS(min_vruntime
));
202 SEQ_printf(m
, " .%-30s: %Ld.%06ld\n", "max_vruntime",
203 SPLIT_NS(max_vruntime
));
204 spread
= max_vruntime
- MIN_vruntime
;
205 SEQ_printf(m
, " .%-30s: %Ld.%06ld\n", "spread",
207 spread0
= min_vruntime
- rq0_min_vruntime
;
208 SEQ_printf(m
, " .%-30s: %Ld.%06ld\n", "spread0",
210 SEQ_printf(m
, " .%-30s: %d\n", "nr_spread_over",
211 cfs_rq
->nr_spread_over
);
212 SEQ_printf(m
, " .%-30s: %d\n", "nr_running", cfs_rq
->nr_running
);
213 SEQ_printf(m
, " .%-30s: %ld\n", "load", cfs_rq
->load
.weight
);
215 SEQ_printf(m
, " .%-30s: %ld\n", "runnable_load_avg",
216 cfs_rq
->runnable_load_avg
);
217 SEQ_printf(m
, " .%-30s: %ld\n", "blocked_load_avg",
218 cfs_rq
->blocked_load_avg
);
219 SEQ_printf(m
, " .%-30s: %ld\n", "utilization_load_avg",
220 cfs_rq
->utilization_load_avg
);
221 #ifdef CONFIG_FAIR_GROUP_SCHED
222 SEQ_printf(m
, " .%-30s: %ld\n", "tg_load_contrib",
223 cfs_rq
->tg_load_contrib
);
224 SEQ_printf(m
, " .%-30s: %d\n", "tg_runnable_contrib",
225 cfs_rq
->tg_runnable_contrib
);
226 SEQ_printf(m
, " .%-30s: %ld\n", "tg_load_avg",
227 atomic_long_read(&cfs_rq
->tg
->load_avg
));
228 SEQ_printf(m
, " .%-30s: %d\n", "tg->runnable_avg",
229 atomic_read(&cfs_rq
->tg
->runnable_avg
));
232 #ifdef CONFIG_CFS_BANDWIDTH
233 SEQ_printf(m
, " .%-30s: %d\n", "tg->cfs_bandwidth.timer_active",
234 cfs_rq
->tg
->cfs_bandwidth
.timer_active
);
235 SEQ_printf(m
, " .%-30s: %d\n", "throttled",
237 SEQ_printf(m
, " .%-30s: %d\n", "throttle_count",
238 cfs_rq
->throttle_count
);
241 #ifdef CONFIG_FAIR_GROUP_SCHED
242 print_cfs_group_stats(m
, cpu
, cfs_rq
->tg
);
246 void print_rt_rq(struct seq_file
*m
, int cpu
, struct rt_rq
*rt_rq
)
248 #ifdef CONFIG_RT_GROUP_SCHED
249 SEQ_printf(m
, "\nrt_rq[%d]:%s\n", cpu
, task_group_path(rt_rq
->tg
));
251 SEQ_printf(m
, "\nrt_rq[%d]:\n", cpu
);
255 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
257 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
268 void print_dl_rq(struct seq_file
*m
, int cpu
, struct dl_rq
*dl_rq
)
270 SEQ_printf(m
, "\ndl_rq[%d]:\n", cpu
);
271 SEQ_printf(m
, " .%-30s: %ld\n", "dl_nr_running", dl_rq
->dl_nr_running
);
274 extern __read_mostly
int sched_clock_running
;
276 static void print_cpu(struct seq_file
*m
, int cpu
)
278 struct rq
*rq
= cpu_rq(cpu
);
283 unsigned int freq
= cpu_khz
? : 1;
285 SEQ_printf(m
, "cpu#%d, %u.%03u MHz\n",
286 cpu
, freq
/ 1000, (freq
% 1000));
289 SEQ_printf(m
, "cpu#%d\n", cpu
);
294 if (sizeof(rq->x) == 4) \
295 SEQ_printf(m, " .%-30s: %ld\n", #x, (long)(rq->x)); \
297 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rq->x));\
301 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
304 SEQ_printf(m
, " .%-30s: %lu\n", "load",
308 P(nr_uninterruptible
);
310 SEQ_printf(m
, " .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq
->curr
)));
321 #ifdef CONFIG_SCHEDSTATS
322 #define P(n) SEQ_printf(m, " .%-30s: %d\n", #n, rq->n);
323 #define P64(n) SEQ_printf(m, " .%-30s: %Ld\n", #n, rq->n);
331 P64(max_idle_balance_cost
);
340 spin_lock_irqsave(&sched_debug_lock
, flags
);
341 print_cfs_stats(m
, cpu
);
342 print_rt_stats(m
, cpu
);
343 print_dl_stats(m
, cpu
);
345 print_rq(m
, rq
, cpu
);
346 spin_unlock_irqrestore(&sched_debug_lock
, flags
);
350 static const char *sched_tunable_scaling_names
[] = {
356 static void sched_debug_header(struct seq_file
*m
)
358 u64 ktime
, sched_clk
, cpu_clk
;
361 local_irq_save(flags
);
362 ktime
= ktime_to_ns(ktime_get());
363 sched_clk
= sched_clock();
364 cpu_clk
= local_clock();
365 local_irq_restore(flags
);
367 SEQ_printf(m
, "Sched Debug Version: v0.11, %s %.*s\n",
368 init_utsname()->release
,
369 (int)strcspn(init_utsname()->version
, " "),
370 init_utsname()->version
);
373 SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
375 SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
380 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
381 P(sched_clock_stable());
387 SEQ_printf(m
, "sysctl_sched\n");
390 SEQ_printf(m, " .%-40s: %Ld\n", #x, (long long)(x))
392 SEQ_printf(m, " .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
393 PN(sysctl_sched_latency
);
394 PN(sysctl_sched_min_granularity
);
395 PN(sysctl_sched_wakeup_granularity
);
396 P(sysctl_sched_child_runs_first
);
397 P(sysctl_sched_features
);
401 SEQ_printf(m
, " .%-40s: %d (%s)\n",
402 "sysctl_sched_tunable_scaling",
403 sysctl_sched_tunable_scaling
,
404 sched_tunable_scaling_names
[sysctl_sched_tunable_scaling
]);
408 static int sched_debug_show(struct seq_file
*m
, void *v
)
410 int cpu
= (unsigned long)(v
- 2);
415 sched_debug_header(m
);
420 void sysrq_sched_debug_show(void)
424 sched_debug_header(NULL
);
425 for_each_online_cpu(cpu
)
426 print_cpu(NULL
, cpu
);
431 * This itererator needs some explanation.
432 * It returns 1 for the header position.
433 * This means 2 is cpu 0.
434 * In a hotplugged system some cpus, including cpu 0, may be missing so we have
435 * to use cpumask_* to iterate over the cpus.
437 static void *sched_debug_start(struct seq_file
*file
, loff_t
*offset
)
439 unsigned long n
= *offset
;
447 n
= cpumask_next(n
- 1, cpu_online_mask
);
449 n
= cpumask_first(cpu_online_mask
);
454 return (void *)(unsigned long)(n
+ 2);
458 static void *sched_debug_next(struct seq_file
*file
, void *data
, loff_t
*offset
)
461 return sched_debug_start(file
, offset
);
464 static void sched_debug_stop(struct seq_file
*file
, void *data
)
468 static const struct seq_operations sched_debug_sops
= {
469 .start
= sched_debug_start
,
470 .next
= sched_debug_next
,
471 .stop
= sched_debug_stop
,
472 .show
= sched_debug_show
,
475 static int sched_debug_release(struct inode
*inode
, struct file
*file
)
477 seq_release(inode
, file
);
482 static int sched_debug_open(struct inode
*inode
, struct file
*filp
)
486 ret
= seq_open(filp
, &sched_debug_sops
);
491 static const struct file_operations sched_debug_fops
= {
492 .open
= sched_debug_open
,
495 .release
= sched_debug_release
,
498 static int __init
init_sched_debug_procfs(void)
500 struct proc_dir_entry
*pe
;
502 pe
= proc_create("sched_debug", 0444, NULL
, &sched_debug_fops
);
508 __initcall(init_sched_debug_procfs
);
511 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)F)
513 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)p->F)
515 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
517 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
520 static void sched_show_numa(struct task_struct
*p
, struct seq_file
*m
)
522 #ifdef CONFIG_NUMA_BALANCING
523 struct mempolicy
*pol
;
527 P(mm
->numa_scan_seq
);
531 if (pol
&& !(pol
->flags
& MPOL_F_MORON
))
536 SEQ_printf(m
, "numa_migrations, %ld\n", xchg(&p
->numa_pages_migrated
, 0));
538 for_each_online_node(node
) {
539 for (i
= 0; i
< 2; i
++) {
540 unsigned long nr_faults
= -1;
541 int cpu_current
, home_node
;
544 nr_faults
= p
->numa_faults
[2*node
+ i
];
546 cpu_current
= !i
? (task_node(p
) == node
) :
547 (pol
&& node_isset(node
, pol
->v
.nodes
));
549 home_node
= (p
->numa_preferred_nid
== node
);
551 SEQ_printf(m
, "numa_faults_memory, %d, %d, %d, %d, %ld\n",
552 i
, node
, cpu_current
, home_node
, nr_faults
);
560 void proc_sched_show_task(struct task_struct
*p
, struct seq_file
*m
)
562 unsigned long nr_switches
;
564 SEQ_printf(m
, "%s (%d, #threads: %d)\n", p
->comm
, task_pid_nr(p
),
567 "---------------------------------------------------------"
570 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)F)
572 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)p->F)
574 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
576 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
580 PN(se
.sum_exec_runtime
);
582 nr_switches
= p
->nvcsw
+ p
->nivcsw
;
584 #ifdef CONFIG_SCHEDSTATS
585 PN(se
.statistics
.wait_start
);
586 PN(se
.statistics
.sleep_start
);
587 PN(se
.statistics
.block_start
);
588 PN(se
.statistics
.sleep_max
);
589 PN(se
.statistics
.block_max
);
590 PN(se
.statistics
.exec_max
);
591 PN(se
.statistics
.slice_max
);
592 PN(se
.statistics
.wait_max
);
593 PN(se
.statistics
.wait_sum
);
594 P(se
.statistics
.wait_count
);
595 PN(se
.statistics
.iowait_sum
);
596 P(se
.statistics
.iowait_count
);
598 P(se
.statistics
.nr_migrations_cold
);
599 P(se
.statistics
.nr_failed_migrations_affine
);
600 P(se
.statistics
.nr_failed_migrations_running
);
601 P(se
.statistics
.nr_failed_migrations_hot
);
602 P(se
.statistics
.nr_forced_migrations
);
603 P(se
.statistics
.nr_wakeups
);
604 P(se
.statistics
.nr_wakeups_sync
);
605 P(se
.statistics
.nr_wakeups_migrate
);
606 P(se
.statistics
.nr_wakeups_local
);
607 P(se
.statistics
.nr_wakeups_remote
);
608 P(se
.statistics
.nr_wakeups_affine
);
609 P(se
.statistics
.nr_wakeups_affine_attempts
);
610 P(se
.statistics
.nr_wakeups_passive
);
611 P(se
.statistics
.nr_wakeups_idle
);
614 u64 avg_atom
, avg_per_cpu
;
616 avg_atom
= p
->se
.sum_exec_runtime
;
618 avg_atom
= div64_ul(avg_atom
, nr_switches
);
622 avg_per_cpu
= p
->se
.sum_exec_runtime
;
623 if (p
->se
.nr_migrations
) {
624 avg_per_cpu
= div64_u64(avg_per_cpu
,
625 p
->se
.nr_migrations
);
635 SEQ_printf(m
, "%-45s:%21Ld\n",
636 "nr_voluntary_switches", (long long)p
->nvcsw
);
637 SEQ_printf(m
, "%-45s:%21Ld\n",
638 "nr_involuntary_switches", (long long)p
->nivcsw
);
642 P(se
.avg
.runnable_avg_sum
);
643 P(se
.avg
.running_avg_sum
);
644 P(se
.avg
.avg_period
);
645 P(se
.avg
.load_avg_contrib
);
646 P(se
.avg
.utilization_avg_contrib
);
647 P(se
.avg
.decay_count
);
657 unsigned int this_cpu
= raw_smp_processor_id();
660 t0
= cpu_clock(this_cpu
);
661 t1
= cpu_clock(this_cpu
);
662 SEQ_printf(m
, "%-45s:%21Ld\n",
663 "clock-delta", (long long)(t1
-t0
));
666 sched_show_numa(p
, m
);
669 void proc_sched_set_task(struct task_struct
*p
)
671 #ifdef CONFIG_SCHEDSTATS
672 memset(&p
->se
.statistics
, 0, sizeof(p
->se
.statistics
));