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>
21 static DEFINE_SPINLOCK(sched_debug_lock);
24 * This allows printing both to /proc/sched_debug and
27 #define SEQ_printf(m, x...) \
36 * Ease the printing of nsec fields:
38 static long long nsec_high(unsigned long long nsec)
40 if ((long long)nsec < 0) {
42 do_div(nsec, 1000000);
45 do_div(nsec, 1000000);
50 static unsigned long nsec_low(unsigned long long nsec)
52 if ((long long)nsec < 0)
55 return do_div(nsec, 1000000);
58 #define SPLIT_NS(x) nsec_high(x), nsec_low(x)
60 #ifdef CONFIG_FAIR_GROUP_SCHED
61 static void print_cfs_group_stats(struct seq_file *m, int cpu, struct task_group *tg)
63 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))
71 struct sched_avg *avg = &cpu_rq(cpu)->avg;
72 P(avg->runnable_avg_sum);
73 P(avg->runnable_avg_period);
80 PN(se->sum_exec_runtime);
81 #ifdef CONFIG_SCHEDSTATS
82 PN(se->statistics.wait_start);
83 PN(se->statistics.sleep_start);
84 PN(se->statistics.block_start);
85 PN(se->statistics.sleep_max);
86 PN(se->statistics.block_max);
87 PN(se->statistics.exec_max);
88 PN(se->statistics.slice_max);
89 PN(se->statistics.wait_max);
90 PN(se->statistics.wait_sum);
91 P(se->statistics.wait_count);
95 P(se->avg.runnable_avg_sum);
96 P(se->avg.runnable_avg_period);
97 P(se->avg.load_avg_contrib);
98 P(se->avg.decay_count);
105 #ifdef CONFIG_CGROUP_SCHED
106 static char group_path[PATH_MAX];
108 static char *task_group_path(struct task_group *tg)
110 if (autogroup_path(tg, group_path, PATH_MAX))
113 cgroup_path(tg->css.cgroup, group_path, PATH_MAX);
119 print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
126 SEQ_printf(m, "%15s %5d %9Ld.%06ld %9Ld %5d ",
128 SPLIT_NS(p->se.vruntime),
129 (long long)(p->nvcsw + p->nivcsw),
131 #ifdef CONFIG_SCHEDSTATS
132 SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
133 SPLIT_NS(p->se.vruntime),
134 SPLIT_NS(p->se.sum_exec_runtime),
135 SPLIT_NS(p->se.statistics.sum_sleep_runtime));
137 SEQ_printf(m, "%15Ld %15Ld %15Ld.%06ld %15Ld.%06ld %15Ld.%06ld",
138 0LL, 0LL, 0LL, 0L, 0LL, 0L, 0LL, 0L);
140 #ifdef CONFIG_CGROUP_SCHED
141 SEQ_printf(m, " %s", task_group_path(task_group(p)));
147 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
149 struct task_struct *g, *p;
153 "\nrunnable tasks:\n"
154 " task PID tree-key switches prio"
155 " exec-runtime sum-exec sum-sleep\n"
156 "------------------------------------------------------"
157 "----------------------------------------------------\n");
159 read_lock_irqsave(&tasklist_lock, flags);
161 do_each_thread(g, p) {
162 if (!p->on_rq || task_cpu(p) != rq_cpu)
165 print_task(m, rq, p);
166 } while_each_thread(g, p);
168 read_unlock_irqrestore(&tasklist_lock, flags);
171 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
173 s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
174 spread, rq0_min_vruntime, spread0;
175 struct rq *rq = cpu_rq(cpu);
176 struct sched_entity *last;
179 #ifdef CONFIG_FAIR_GROUP_SCHED
180 SEQ_printf(m, "\ncfs_rq[%d]:%s\n", cpu, task_group_path(cfs_rq->tg));
182 SEQ_printf(m, "\ncfs_rq[%d]:\n", cpu);
184 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "exec_clock",
185 SPLIT_NS(cfs_rq->exec_clock));
187 raw_spin_lock_irqsave(&rq->lock, flags);
188 if (cfs_rq->rb_leftmost)
189 MIN_vruntime = (__pick_first_entity(cfs_rq))->vruntime;
190 last = __pick_last_entity(cfs_rq);
192 max_vruntime = last->vruntime;
193 min_vruntime = cfs_rq->min_vruntime;
194 rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
195 raw_spin_unlock_irqrestore(&rq->lock, flags);
196 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "MIN_vruntime",
197 SPLIT_NS(MIN_vruntime));
198 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "min_vruntime",
199 SPLIT_NS(min_vruntime));
200 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "max_vruntime",
201 SPLIT_NS(max_vruntime));
202 spread = max_vruntime - MIN_vruntime;
203 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread",
205 spread0 = min_vruntime - rq0_min_vruntime;
206 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread0",
208 SEQ_printf(m, " .%-30s: %d\n", "nr_spread_over",
209 cfs_rq->nr_spread_over);
210 SEQ_printf(m, " .%-30s: %d\n", "nr_running", cfs_rq->nr_running);
211 SEQ_printf(m, " .%-30s: %ld\n", "load", cfs_rq->load.weight);
213 SEQ_printf(m, " .%-30s: %ld\n", "runnable_load_avg",
214 cfs_rq->runnable_load_avg);
215 SEQ_printf(m, " .%-30s: %ld\n", "blocked_load_avg",
216 cfs_rq->blocked_load_avg);
217 #ifdef CONFIG_FAIR_GROUP_SCHED
218 SEQ_printf(m, " .%-30s: %ld\n", "tg_load_contrib",
219 cfs_rq->tg_load_contrib);
220 SEQ_printf(m, " .%-30s: %d\n", "tg_runnable_contrib",
221 cfs_rq->tg_runnable_contrib);
222 SEQ_printf(m, " .%-30s: %ld\n", "tg_load_avg",
223 atomic_long_read(&cfs_rq->tg->load_avg));
224 SEQ_printf(m, " .%-30s: %d\n", "tg->runnable_avg",
225 atomic_read(&cfs_rq->tg->runnable_avg));
229 #ifdef CONFIG_FAIR_GROUP_SCHED
230 print_cfs_group_stats(m, cpu, cfs_rq->tg);
234 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
236 #ifdef CONFIG_RT_GROUP_SCHED
237 SEQ_printf(m, "\nrt_rq[%d]:%s\n", cpu, task_group_path(rt_rq->tg));
239 SEQ_printf(m, "\nrt_rq[%d]:\n", cpu);
243 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
245 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
256 extern __read_mostly int sched_clock_running;
258 static void print_cpu(struct seq_file *m, int cpu)
260 struct rq *rq = cpu_rq(cpu);
265 unsigned int freq = cpu_khz ? : 1;
267 SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
268 cpu, freq / 1000, (freq % 1000));
271 SEQ_printf(m, "cpu#%d\n", cpu);
276 if (sizeof(rq->x) == 4) \
277 SEQ_printf(m, " .%-30s: %ld\n", #x, (long)(rq->x)); \
279 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rq->x));\
283 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
286 SEQ_printf(m, " .%-30s: %lu\n", "load",
290 P(nr_uninterruptible);
302 #ifdef CONFIG_SCHEDSTATS
303 #define P(n) SEQ_printf(m, " .%-30s: %d\n", #n, rq->n);
304 #define P64(n) SEQ_printf(m, " .%-30s: %Ld\n", #n, rq->n);
320 spin_lock_irqsave(&sched_debug_lock, flags);
321 print_cfs_stats(m, cpu);
322 print_rt_stats(m, cpu);
325 print_rq(m, rq, cpu);
327 spin_unlock_irqrestore(&sched_debug_lock, flags);
331 static const char *sched_tunable_scaling_names[] = {
337 static void sched_debug_header(struct seq_file *m)
339 u64 ktime, sched_clk, cpu_clk;
342 local_irq_save(flags);
343 ktime = ktime_to_ns(ktime_get());
344 sched_clk = sched_clock();
345 cpu_clk = local_clock();
346 local_irq_restore(flags);
348 SEQ_printf(m, "Sched Debug Version: v0.10, %s %.*s\n",
349 init_utsname()->release,
350 (int)strcspn(init_utsname()->version, " "),
351 init_utsname()->version);
354 SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
356 SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
361 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
362 P(sched_clock_stable);
368 SEQ_printf(m, "sysctl_sched\n");
371 SEQ_printf(m, " .%-40s: %Ld\n", #x, (long long)(x))
373 SEQ_printf(m, " .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
374 PN(sysctl_sched_latency);
375 PN(sysctl_sched_min_granularity);
376 PN(sysctl_sched_wakeup_granularity);
377 P(sysctl_sched_child_runs_first);
378 P(sysctl_sched_features);
382 SEQ_printf(m, " .%-40s: %d (%s)\n",
383 "sysctl_sched_tunable_scaling",
384 sysctl_sched_tunable_scaling,
385 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
389 static int sched_debug_show(struct seq_file *m, void *v)
391 int cpu = (unsigned long)(v - 2);
396 sched_debug_header(m);
401 void sysrq_sched_debug_show(void)
405 sched_debug_header(NULL);
406 for_each_online_cpu(cpu)
407 print_cpu(NULL, cpu);
412 * This itererator needs some explanation.
413 * It returns 1 for the header position.
414 * This means 2 is cpu 0.
415 * In a hotplugged system some cpus, including cpu 0, may be missing so we have
416 * to use cpumask_* to iterate over the cpus.
418 static void *sched_debug_start(struct seq_file *file, loff_t *offset)
420 unsigned long n = *offset;
428 n = cpumask_next(n - 1, cpu_online_mask);
430 n = cpumask_first(cpu_online_mask);
435 return (void *)(unsigned long)(n + 2);
439 static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
442 return sched_debug_start(file, offset);
445 static void sched_debug_stop(struct seq_file *file, void *data)
449 static const struct seq_operations sched_debug_sops = {
450 .start = sched_debug_start,
451 .next = sched_debug_next,
452 .stop = sched_debug_stop,
453 .show = sched_debug_show,
456 static int sched_debug_release(struct inode *inode, struct file *file)
458 seq_release(inode, file);
463 static int sched_debug_open(struct inode *inode, struct file *filp)
467 ret = seq_open(filp, &sched_debug_sops);
472 static const struct file_operations sched_debug_fops = {
473 .open = sched_debug_open,
476 .release = sched_debug_release,
479 static int __init init_sched_debug_procfs(void)
481 struct proc_dir_entry *pe;
483 pe = proc_create("sched_debug", 0444, NULL, &sched_debug_fops);
489 __initcall(init_sched_debug_procfs);
491 void proc_sched_show_task(struct task_struct *p, struct seq_file *m)
493 unsigned long nr_switches;
495 SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, p->pid,
498 "---------------------------------------------------------"
501 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)F)
503 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)p->F)
505 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
507 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
511 PN(se.sum_exec_runtime);
513 nr_switches = p->nvcsw + p->nivcsw;
515 #ifdef CONFIG_SCHEDSTATS
516 PN(se.statistics.wait_start);
517 PN(se.statistics.sleep_start);
518 PN(se.statistics.block_start);
519 PN(se.statistics.sleep_max);
520 PN(se.statistics.block_max);
521 PN(se.statistics.exec_max);
522 PN(se.statistics.slice_max);
523 PN(se.statistics.wait_max);
524 PN(se.statistics.wait_sum);
525 P(se.statistics.wait_count);
526 PN(se.statistics.iowait_sum);
527 P(se.statistics.iowait_count);
529 P(se.statistics.nr_migrations_cold);
530 P(se.statistics.nr_failed_migrations_affine);
531 P(se.statistics.nr_failed_migrations_running);
532 P(se.statistics.nr_failed_migrations_hot);
533 P(se.statistics.nr_forced_migrations);
534 P(se.statistics.nr_wakeups);
535 P(se.statistics.nr_wakeups_sync);
536 P(se.statistics.nr_wakeups_migrate);
537 P(se.statistics.nr_wakeups_local);
538 P(se.statistics.nr_wakeups_remote);
539 P(se.statistics.nr_wakeups_affine);
540 P(se.statistics.nr_wakeups_affine_attempts);
541 P(se.statistics.nr_wakeups_passive);
542 P(se.statistics.nr_wakeups_idle);
545 u64 avg_atom, avg_per_cpu;
547 avg_atom = p->se.sum_exec_runtime;
549 do_div(avg_atom, nr_switches);
553 avg_per_cpu = p->se.sum_exec_runtime;
554 if (p->se.nr_migrations) {
555 avg_per_cpu = div64_u64(avg_per_cpu,
556 p->se.nr_migrations);
566 SEQ_printf(m, "%-45s:%21Ld\n",
567 "nr_voluntary_switches", (long long)p->nvcsw);
568 SEQ_printf(m, "%-45s:%21Ld\n",
569 "nr_involuntary_switches", (long long)p->nivcsw);
573 P(se.avg.runnable_avg_sum);
574 P(se.avg.runnable_avg_period);
575 P(se.avg.load_avg_contrib);
576 P(se.avg.decay_count);
586 unsigned int this_cpu = raw_smp_processor_id();
589 t0 = cpu_clock(this_cpu);
590 t1 = cpu_clock(this_cpu);
591 SEQ_printf(m, "%-45s:%21Ld\n",
592 "clock-delta", (long long)(t1-t0));
596 void proc_sched_set_task(struct task_struct *p)
598 #ifdef CONFIG_SCHEDSTATS
599 memset(&p->se.statistics, 0, sizeof(p->se.statistics));