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);
74 P(avg->runnable_avg_period);
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.runnable_avg_period);
98 P(se->avg.load_avg_contrib);
99 P(se->avg.decay_count);
106 #ifdef CONFIG_CGROUP_SCHED
107 static char group_path[PATH_MAX];
109 static char *task_group_path(struct task_group *tg)
111 if (autogroup_path(tg, group_path, PATH_MAX))
114 return 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 ",
127 p->comm, task_pid_nr(p),
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_NUMA_BALANCING
141 SEQ_printf(m, " %d", task_node(p));
143 #ifdef CONFIG_CGROUP_SCHED
144 SEQ_printf(m, " %s", task_group_path(task_group(p)));
150 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
152 struct task_struct *g, *p;
155 "\nrunnable tasks:\n"
156 " task PID tree-key switches prio"
157 " exec-runtime sum-exec sum-sleep\n"
158 "------------------------------------------------------"
159 "----------------------------------------------------\n");
162 for_each_process_thread(g, p) {
163 if (task_cpu(p) != rq_cpu)
166 print_task(m, rq, p);
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));
228 #ifdef CONFIG_CFS_BANDWIDTH
229 SEQ_printf(m, " .%-30s: %d\n", "tg->cfs_bandwidth.timer_active",
230 cfs_rq->tg->cfs_bandwidth.timer_active);
231 SEQ_printf(m, " .%-30s: %d\n", "throttled",
233 SEQ_printf(m, " .%-30s: %d\n", "throttle_count",
234 cfs_rq->throttle_count);
237 #ifdef CONFIG_FAIR_GROUP_SCHED
238 print_cfs_group_stats(m, cpu, cfs_rq->tg);
242 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
244 #ifdef CONFIG_RT_GROUP_SCHED
245 SEQ_printf(m, "\nrt_rq[%d]:%s\n", cpu, task_group_path(rt_rq->tg));
247 SEQ_printf(m, "\nrt_rq[%d]:\n", cpu);
251 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
253 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
264 void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq)
266 SEQ_printf(m, "\ndl_rq[%d]:\n", cpu);
267 SEQ_printf(m, " .%-30s: %ld\n", "dl_nr_running", dl_rq->dl_nr_running);
270 extern __read_mostly int sched_clock_running;
272 static void print_cpu(struct seq_file *m, int cpu)
274 struct rq *rq = cpu_rq(cpu);
279 unsigned int freq = cpu_khz ? : 1;
281 SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
282 cpu, freq / 1000, (freq % 1000));
285 SEQ_printf(m, "cpu#%d\n", cpu);
290 if (sizeof(rq->x) == 4) \
291 SEQ_printf(m, " .%-30s: %ld\n", #x, (long)(rq->x)); \
293 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rq->x));\
297 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
300 SEQ_printf(m, " .%-30s: %lu\n", "load",
304 P(nr_uninterruptible);
306 SEQ_printf(m, " .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr)));
317 #ifdef CONFIG_SCHEDSTATS
318 #define P(n) SEQ_printf(m, " .%-30s: %d\n", #n, rq->n);
319 #define P64(n) SEQ_printf(m, " .%-30s: %Ld\n", #n, rq->n);
327 P64(max_idle_balance_cost);
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 static void sched_show_numa(struct task_struct *p, struct seq_file *m)
518 #ifdef CONFIG_NUMA_BALANCING
519 struct mempolicy *pol;
523 P(mm->numa_scan_seq);
527 if (pol && !(pol->flags & MPOL_F_MORON))
532 SEQ_printf(m, "numa_migrations, %ld\n", xchg(&p->numa_pages_migrated, 0));
534 for_each_online_node(node) {
535 for (i = 0; i < 2; i++) {
536 unsigned long nr_faults = -1;
537 int cpu_current, home_node;
540 nr_faults = p->numa_faults[2*node + i];
542 cpu_current = !i ? (task_node(p) == node) :
543 (pol && node_isset(node, pol->v.nodes));
545 home_node = (p->numa_preferred_nid == node);
547 SEQ_printf(m, "numa_faults_memory, %d, %d, %d, %d, %ld\n",
548 i, node, cpu_current, home_node, nr_faults);
556 void proc_sched_show_task(struct task_struct *p, struct seq_file *m)
558 unsigned long nr_switches;
560 SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr(p),
563 "---------------------------------------------------------"
566 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)F)
568 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)p->F)
570 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
572 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
576 PN(se.sum_exec_runtime);
578 nr_switches = p->nvcsw + p->nivcsw;
580 #ifdef CONFIG_SCHEDSTATS
581 PN(se.statistics.wait_start);
582 PN(se.statistics.sleep_start);
583 PN(se.statistics.block_start);
584 PN(se.statistics.sleep_max);
585 PN(se.statistics.block_max);
586 PN(se.statistics.exec_max);
587 PN(se.statistics.slice_max);
588 PN(se.statistics.wait_max);
589 PN(se.statistics.wait_sum);
590 P(se.statistics.wait_count);
591 PN(se.statistics.iowait_sum);
592 P(se.statistics.iowait_count);
594 P(se.statistics.nr_migrations_cold);
595 P(se.statistics.nr_failed_migrations_affine);
596 P(se.statistics.nr_failed_migrations_running);
597 P(se.statistics.nr_failed_migrations_hot);
598 P(se.statistics.nr_forced_migrations);
599 P(se.statistics.nr_wakeups);
600 P(se.statistics.nr_wakeups_sync);
601 P(se.statistics.nr_wakeups_migrate);
602 P(se.statistics.nr_wakeups_local);
603 P(se.statistics.nr_wakeups_remote);
604 P(se.statistics.nr_wakeups_affine);
605 P(se.statistics.nr_wakeups_affine_attempts);
606 P(se.statistics.nr_wakeups_passive);
607 P(se.statistics.nr_wakeups_idle);
610 u64 avg_atom, avg_per_cpu;
612 avg_atom = p->se.sum_exec_runtime;
614 avg_atom = div64_ul(avg_atom, nr_switches);
618 avg_per_cpu = p->se.sum_exec_runtime;
619 if (p->se.nr_migrations) {
620 avg_per_cpu = div64_u64(avg_per_cpu,
621 p->se.nr_migrations);
631 SEQ_printf(m, "%-45s:%21Ld\n",
632 "nr_voluntary_switches", (long long)p->nvcsw);
633 SEQ_printf(m, "%-45s:%21Ld\n",
634 "nr_involuntary_switches", (long long)p->nivcsw);
638 P(se.avg.runnable_avg_sum);
639 P(se.avg.runnable_avg_period);
640 P(se.avg.load_avg_contrib);
641 P(se.avg.decay_count);
651 unsigned int this_cpu = raw_smp_processor_id();
654 t0 = cpu_clock(this_cpu);
655 t1 = cpu_clock(this_cpu);
656 SEQ_printf(m, "%-45s:%21Ld\n",
657 "clock-delta", (long long)(t1-t0));
660 sched_show_numa(p, m);
663 void proc_sched_set_task(struct task_struct *p)
665 #ifdef CONFIG_SCHEDSTATS
666 memset(&p->se.statistics, 0, sizeof(p->se.statistics));