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))
76 PN(se->sum_exec_runtime);
77 #ifdef CONFIG_SCHEDSTATS
78 if (schedstat_enabled()) {
79 PN(se->statistics.wait_start);
80 PN(se->statistics.sleep_start);
81 PN(se->statistics.block_start);
82 PN(se->statistics.sleep_max);
83 PN(se->statistics.block_max);
84 PN(se->statistics.exec_max);
85 PN(se->statistics.slice_max);
86 PN(se->statistics.wait_max);
87 PN(se->statistics.wait_sum);
88 P(se->statistics.wait_count);
101 #ifdef CONFIG_CGROUP_SCHED
102 static char group_path[PATH_MAX];
104 static char *task_group_path(struct task_group *tg)
106 if (autogroup_path(tg, group_path, PATH_MAX))
109 return cgroup_path(tg->css.cgroup, group_path, PATH_MAX);
114 print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
121 SEQ_printf(m, "%15s %5d %9Ld.%06ld %9Ld %5d ",
122 p->comm, task_pid_nr(p),
123 SPLIT_NS(p->se.vruntime),
124 (long long)(p->nvcsw + p->nivcsw),
126 #ifdef CONFIG_SCHEDSTATS
127 if (schedstat_enabled()) {
128 SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
129 SPLIT_NS(p->se.statistics.wait_sum),
130 SPLIT_NS(p->se.sum_exec_runtime),
131 SPLIT_NS(p->se.statistics.sum_sleep_runtime));
134 SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
136 SPLIT_NS(p->se.sum_exec_runtime),
139 #ifdef CONFIG_NUMA_BALANCING
140 SEQ_printf(m, " %d %d", task_node(p), task_numa_group_id(p));
142 #ifdef CONFIG_CGROUP_SCHED
143 SEQ_printf(m, " %s", task_group_path(task_group(p)));
149 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
151 struct task_struct *g, *p;
154 "\nrunnable tasks:\n"
155 " task PID tree-key switches prio"
156 " wait-time sum-exec sum-sleep\n"
157 "------------------------------------------------------"
158 "----------------------------------------------------\n");
161 for_each_process_thread(g, p) {
162 if (task_cpu(p) != rq_cpu)
165 print_task(m, rq, p);
170 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
172 s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
173 spread, rq0_min_vruntime, spread0;
174 struct rq *rq = cpu_rq(cpu);
175 struct sched_entity *last;
178 #ifdef CONFIG_FAIR_GROUP_SCHED
179 SEQ_printf(m, "\ncfs_rq[%d]:%s\n", cpu, task_group_path(cfs_rq->tg));
181 SEQ_printf(m, "\ncfs_rq[%d]:\n", cpu);
183 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "exec_clock",
184 SPLIT_NS(cfs_rq->exec_clock));
186 raw_spin_lock_irqsave(&rq->lock, flags);
187 if (cfs_rq->rb_leftmost)
188 MIN_vruntime = (__pick_first_entity(cfs_rq))->vruntime;
189 last = __pick_last_entity(cfs_rq);
191 max_vruntime = last->vruntime;
192 min_vruntime = cfs_rq->min_vruntime;
193 rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
194 raw_spin_unlock_irqrestore(&rq->lock, flags);
195 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "MIN_vruntime",
196 SPLIT_NS(MIN_vruntime));
197 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "min_vruntime",
198 SPLIT_NS(min_vruntime));
199 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "max_vruntime",
200 SPLIT_NS(max_vruntime));
201 spread = max_vruntime - MIN_vruntime;
202 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread",
204 spread0 = min_vruntime - rq0_min_vruntime;
205 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread0",
207 SEQ_printf(m, " .%-30s: %d\n", "nr_spread_over",
208 cfs_rq->nr_spread_over);
209 SEQ_printf(m, " .%-30s: %d\n", "nr_running", cfs_rq->nr_running);
210 SEQ_printf(m, " .%-30s: %ld\n", "load", cfs_rq->load.weight);
212 SEQ_printf(m, " .%-30s: %lu\n", "load_avg",
213 cfs_rq->avg.load_avg);
214 SEQ_printf(m, " .%-30s: %lu\n", "runnable_load_avg",
215 cfs_rq->runnable_load_avg);
216 SEQ_printf(m, " .%-30s: %lu\n", "util_avg",
217 cfs_rq->avg.util_avg);
218 SEQ_printf(m, " .%-30s: %ld\n", "removed_load_avg",
219 atomic_long_read(&cfs_rq->removed_load_avg));
220 SEQ_printf(m, " .%-30s: %ld\n", "removed_util_avg",
221 atomic_long_read(&cfs_rq->removed_util_avg));
222 #ifdef CONFIG_FAIR_GROUP_SCHED
223 SEQ_printf(m, " .%-30s: %lu\n", "tg_load_avg_contrib",
224 cfs_rq->tg_load_avg_contrib);
225 SEQ_printf(m, " .%-30s: %ld\n", "tg_load_avg",
226 atomic_long_read(&cfs_rq->tg->load_avg));
229 #ifdef CONFIG_CFS_BANDWIDTH
230 SEQ_printf(m, " .%-30s: %d\n", "throttled",
232 SEQ_printf(m, " .%-30s: %d\n", "throttle_count",
233 cfs_rq->throttle_count);
236 #ifdef CONFIG_FAIR_GROUP_SCHED
237 print_cfs_group_stats(m, cpu, cfs_rq->tg);
241 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
243 #ifdef CONFIG_RT_GROUP_SCHED
244 SEQ_printf(m, "\nrt_rq[%d]:%s\n", cpu, task_group_path(rt_rq->tg));
246 SEQ_printf(m, "\nrt_rq[%d]:\n", cpu);
250 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
252 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
263 void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq)
265 SEQ_printf(m, "\ndl_rq[%d]:\n", cpu);
266 SEQ_printf(m, " .%-30s: %ld\n", "dl_nr_running", dl_rq->dl_nr_running);
269 extern __read_mostly int sched_clock_running;
271 static void print_cpu(struct seq_file *m, int cpu)
273 struct rq *rq = cpu_rq(cpu);
278 unsigned int freq = cpu_khz ? : 1;
280 SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
281 cpu, freq / 1000, (freq % 1000));
284 SEQ_printf(m, "cpu#%d\n", cpu);
289 if (sizeof(rq->x) == 4) \
290 SEQ_printf(m, " .%-30s: %ld\n", #x, (long)(rq->x)); \
292 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rq->x));\
296 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
299 SEQ_printf(m, " .%-30s: %lu\n", "load",
303 P(nr_uninterruptible);
305 SEQ_printf(m, " .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr)));
316 #ifdef CONFIG_SCHEDSTATS
317 #define P(n) SEQ_printf(m, " .%-30s: %d\n", #n, rq->n);
318 #define P64(n) SEQ_printf(m, " .%-30s: %Ld\n", #n, rq->n);
322 P64(max_idle_balance_cost);
325 if (schedstat_enabled()) {
336 spin_lock_irqsave(&sched_debug_lock, flags);
337 print_cfs_stats(m, cpu);
338 print_rt_stats(m, cpu);
339 print_dl_stats(m, cpu);
341 print_rq(m, rq, cpu);
342 spin_unlock_irqrestore(&sched_debug_lock, flags);
346 static const char *sched_tunable_scaling_names[] = {
352 static void sched_debug_header(struct seq_file *m)
354 u64 ktime, sched_clk, cpu_clk;
357 local_irq_save(flags);
358 ktime = ktime_to_ns(ktime_get());
359 sched_clk = sched_clock();
360 cpu_clk = local_clock();
361 local_irq_restore(flags);
363 SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n",
364 init_utsname()->release,
365 (int)strcspn(init_utsname()->version, " "),
366 init_utsname()->version);
369 SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
371 SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
376 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
377 P(sched_clock_stable());
383 SEQ_printf(m, "sysctl_sched\n");
386 SEQ_printf(m, " .%-40s: %Ld\n", #x, (long long)(x))
388 SEQ_printf(m, " .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
389 PN(sysctl_sched_latency);
390 PN(sysctl_sched_min_granularity);
391 PN(sysctl_sched_wakeup_granularity);
392 P(sysctl_sched_child_runs_first);
393 P(sysctl_sched_features);
397 SEQ_printf(m, " .%-40s: %d (%s)\n",
398 "sysctl_sched_tunable_scaling",
399 sysctl_sched_tunable_scaling,
400 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
404 static int sched_debug_show(struct seq_file *m, void *v)
406 int cpu = (unsigned long)(v - 2);
411 sched_debug_header(m);
416 void sysrq_sched_debug_show(void)
420 sched_debug_header(NULL);
421 for_each_online_cpu(cpu)
422 print_cpu(NULL, cpu);
427 * This itererator needs some explanation.
428 * It returns 1 for the header position.
429 * This means 2 is cpu 0.
430 * In a hotplugged system some cpus, including cpu 0, may be missing so we have
431 * to use cpumask_* to iterate over the cpus.
433 static void *sched_debug_start(struct seq_file *file, loff_t *offset)
435 unsigned long n = *offset;
443 n = cpumask_next(n - 1, cpu_online_mask);
445 n = cpumask_first(cpu_online_mask);
450 return (void *)(unsigned long)(n + 2);
454 static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
457 return sched_debug_start(file, offset);
460 static void sched_debug_stop(struct seq_file *file, void *data)
464 static const struct seq_operations sched_debug_sops = {
465 .start = sched_debug_start,
466 .next = sched_debug_next,
467 .stop = sched_debug_stop,
468 .show = sched_debug_show,
471 static int sched_debug_release(struct inode *inode, struct file *file)
473 seq_release(inode, file);
478 static int sched_debug_open(struct inode *inode, struct file *filp)
482 ret = seq_open(filp, &sched_debug_sops);
487 static const struct file_operations sched_debug_fops = {
488 .open = sched_debug_open,
491 .release = sched_debug_release,
494 static int __init init_sched_debug_procfs(void)
496 struct proc_dir_entry *pe;
498 pe = proc_create("sched_debug", 0444, NULL, &sched_debug_fops);
504 __initcall(init_sched_debug_procfs);
507 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)F)
509 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)p->F)
511 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
513 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
516 #ifdef CONFIG_NUMA_BALANCING
517 void print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
518 unsigned long tpf, unsigned long gsf, unsigned long gpf)
520 SEQ_printf(m, "numa_faults node=%d ", node);
521 SEQ_printf(m, "task_private=%lu task_shared=%lu ", tsf, tpf);
522 SEQ_printf(m, "group_private=%lu group_shared=%lu\n", gsf, gpf);
527 static void sched_show_numa(struct task_struct *p, struct seq_file *m)
529 #ifdef CONFIG_NUMA_BALANCING
530 struct mempolicy *pol;
533 P(mm->numa_scan_seq);
537 if (pol && !(pol->flags & MPOL_F_MORON))
542 P(numa_pages_migrated);
543 P(numa_preferred_nid);
544 P(total_numa_faults);
545 SEQ_printf(m, "current_node=%d, numa_group_id=%d\n",
546 task_node(p), task_numa_group_id(p));
547 show_numa_stats(p, m);
552 void proc_sched_show_task(struct task_struct *p, struct seq_file *m)
554 unsigned long nr_switches;
556 SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr(p),
559 "---------------------------------------------------------"
562 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)F)
564 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)p->F)
566 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
568 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
572 PN(se.sum_exec_runtime);
574 nr_switches = p->nvcsw + p->nivcsw;
576 #ifdef CONFIG_SCHEDSTATS
579 if (schedstat_enabled()) {
580 u64 avg_atom, avg_per_cpu;
582 PN(se.statistics.sum_sleep_runtime);
583 PN(se.statistics.wait_start);
584 PN(se.statistics.sleep_start);
585 PN(se.statistics.block_start);
586 PN(se.statistics.sleep_max);
587 PN(se.statistics.block_max);
588 PN(se.statistics.exec_max);
589 PN(se.statistics.slice_max);
590 PN(se.statistics.wait_max);
591 PN(se.statistics.wait_sum);
592 P(se.statistics.wait_count);
593 PN(se.statistics.iowait_sum);
594 P(se.statistics.iowait_count);
595 P(se.statistics.nr_migrations_cold);
596 P(se.statistics.nr_failed_migrations_affine);
597 P(se.statistics.nr_failed_migrations_running);
598 P(se.statistics.nr_failed_migrations_hot);
599 P(se.statistics.nr_forced_migrations);
600 P(se.statistics.nr_wakeups);
601 P(se.statistics.nr_wakeups_sync);
602 P(se.statistics.nr_wakeups_migrate);
603 P(se.statistics.nr_wakeups_local);
604 P(se.statistics.nr_wakeups_remote);
605 P(se.statistics.nr_wakeups_affine);
606 P(se.statistics.nr_wakeups_affine_attempts);
607 P(se.statistics.nr_wakeups_passive);
608 P(se.statistics.nr_wakeups_idle);
610 avg_atom = p->se.sum_exec_runtime;
612 avg_atom = div64_ul(avg_atom, nr_switches);
616 avg_per_cpu = p->se.sum_exec_runtime;
617 if (p->se.nr_migrations) {
618 avg_per_cpu = div64_u64(avg_per_cpu,
619 p->se.nr_migrations);
629 SEQ_printf(m, "%-45s:%21Ld\n",
630 "nr_voluntary_switches", (long long)p->nvcsw);
631 SEQ_printf(m, "%-45s:%21Ld\n",
632 "nr_involuntary_switches", (long long)p->nivcsw);
640 P(se.avg.last_update_time);
650 unsigned int this_cpu = raw_smp_processor_id();
653 t0 = cpu_clock(this_cpu);
654 t1 = cpu_clock(this_cpu);
655 SEQ_printf(m, "%-45s:%21Ld\n",
656 "clock-delta", (long long)(t1-t0));
659 sched_show_numa(p, m);
662 void proc_sched_set_task(struct task_struct *p)
664 #ifdef CONFIG_SCHEDSTATS
665 memset(&p->se.statistics, 0, sizeof(p->se.statistics));