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 PN(se->statistics.wait_start);
79 PN(se->statistics.sleep_start);
80 PN(se->statistics.block_start);
81 PN(se->statistics.sleep_max);
82 PN(se->statistics.block_max);
83 PN(se->statistics.exec_max);
84 PN(se->statistics.slice_max);
85 PN(se->statistics.wait_max);
86 PN(se->statistics.wait_sum);
87 P(se->statistics.wait_count);
99 #ifdef CONFIG_CGROUP_SCHED
100 static char group_path[PATH_MAX];
102 static char *task_group_path(struct task_group *tg)
104 if (autogroup_path(tg, group_path, PATH_MAX))
107 return 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 ",
120 p->comm, task_pid_nr(p),
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.statistics.wait_sum),
127 SPLIT_NS(p->se.sum_exec_runtime),
128 SPLIT_NS(p->se.statistics.sum_sleep_runtime));
130 SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
132 SPLIT_NS(p->se.sum_exec_runtime),
135 #ifdef CONFIG_NUMA_BALANCING
136 SEQ_printf(m, " %d %d", task_node(p), task_numa_group_id(p));
138 #ifdef CONFIG_CGROUP_SCHED
139 SEQ_printf(m, " %s", task_group_path(task_group(p)));
145 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
147 struct task_struct *g, *p;
150 "\nrunnable tasks:\n"
151 " task PID tree-key switches prio"
152 " wait-time sum-exec sum-sleep\n"
153 "------------------------------------------------------"
154 "----------------------------------------------------\n");
157 for_each_process_thread(g, p) {
158 if (task_cpu(p) != rq_cpu)
161 print_task(m, rq, p);
166 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
168 s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
169 spread, rq0_min_vruntime, spread0;
170 struct rq *rq = cpu_rq(cpu);
171 struct sched_entity *last;
174 #ifdef CONFIG_FAIR_GROUP_SCHED
175 SEQ_printf(m, "\ncfs_rq[%d]:%s\n", cpu, task_group_path(cfs_rq->tg));
177 SEQ_printf(m, "\ncfs_rq[%d]:\n", cpu);
179 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "exec_clock",
180 SPLIT_NS(cfs_rq->exec_clock));
182 raw_spin_lock_irqsave(&rq->lock, flags);
183 if (cfs_rq->rb_leftmost)
184 MIN_vruntime = (__pick_first_entity(cfs_rq))->vruntime;
185 last = __pick_last_entity(cfs_rq);
187 max_vruntime = last->vruntime;
188 min_vruntime = cfs_rq->min_vruntime;
189 rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
190 raw_spin_unlock_irqrestore(&rq->lock, flags);
191 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "MIN_vruntime",
192 SPLIT_NS(MIN_vruntime));
193 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "min_vruntime",
194 SPLIT_NS(min_vruntime));
195 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "max_vruntime",
196 SPLIT_NS(max_vruntime));
197 spread = max_vruntime - MIN_vruntime;
198 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread",
200 spread0 = min_vruntime - rq0_min_vruntime;
201 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread0",
203 SEQ_printf(m, " .%-30s: %d\n", "nr_spread_over",
204 cfs_rq->nr_spread_over);
205 SEQ_printf(m, " .%-30s: %d\n", "nr_running", cfs_rq->nr_running);
206 SEQ_printf(m, " .%-30s: %ld\n", "load", cfs_rq->load.weight);
208 SEQ_printf(m, " .%-30s: %lu\n", "load_avg",
209 cfs_rq->avg.load_avg);
210 SEQ_printf(m, " .%-30s: %lu\n", "runnable_load_avg",
211 cfs_rq->runnable_load_avg);
212 SEQ_printf(m, " .%-30s: %lu\n", "util_avg",
213 cfs_rq->avg.util_avg);
214 SEQ_printf(m, " .%-30s: %ld\n", "removed_load_avg",
215 atomic_long_read(&cfs_rq->removed_load_avg));
216 SEQ_printf(m, " .%-30s: %ld\n", "removed_util_avg",
217 atomic_long_read(&cfs_rq->removed_util_avg));
218 #ifdef CONFIG_FAIR_GROUP_SCHED
219 SEQ_printf(m, " .%-30s: %lu\n", "tg_load_avg_contrib",
220 cfs_rq->tg_load_avg_contrib);
221 SEQ_printf(m, " .%-30s: %ld\n", "tg_load_avg",
222 atomic_long_read(&cfs_rq->tg->load_avg));
225 #ifdef CONFIG_CFS_BANDWIDTH
226 SEQ_printf(m, " .%-30s: %d\n", "throttled",
228 SEQ_printf(m, " .%-30s: %d\n", "throttle_count",
229 cfs_rq->throttle_count);
232 #ifdef CONFIG_FAIR_GROUP_SCHED
233 print_cfs_group_stats(m, cpu, cfs_rq->tg);
237 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
239 #ifdef CONFIG_RT_GROUP_SCHED
240 SEQ_printf(m, "\nrt_rq[%d]:%s\n", cpu, task_group_path(rt_rq->tg));
242 SEQ_printf(m, "\nrt_rq[%d]:\n", cpu);
246 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
248 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
262 void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq)
264 SEQ_printf(m, "\ndl_rq[%d]:\n", cpu);
265 SEQ_printf(m, " .%-30s: %ld\n", "dl_nr_running", dl_rq->dl_nr_running);
268 extern __read_mostly int sched_clock_running;
270 static void print_cpu(struct seq_file *m, int cpu)
272 struct rq *rq = cpu_rq(cpu);
277 unsigned int freq = cpu_khz ? : 1;
279 SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
280 cpu, freq / 1000, (freq % 1000));
283 SEQ_printf(m, "cpu#%d\n", cpu);
288 if (sizeof(rq->x) == 4) \
289 SEQ_printf(m, " .%-30s: %ld\n", #x, (long)(rq->x)); \
291 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rq->x));\
295 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
298 SEQ_printf(m, " .%-30s: %lu\n", "load",
302 P(nr_uninterruptible);
304 SEQ_printf(m, " .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr)));
315 #ifdef CONFIG_SCHEDSTATS
316 #define P(n) SEQ_printf(m, " .%-30s: %d\n", #n, rq->n);
317 #define P64(n) SEQ_printf(m, " .%-30s: %Ld\n", #n, rq->n);
325 P64(max_idle_balance_cost);
334 spin_lock_irqsave(&sched_debug_lock, flags);
335 print_cfs_stats(m, cpu);
336 print_rt_stats(m, cpu);
337 print_dl_stats(m, cpu);
339 print_rq(m, rq, cpu);
340 spin_unlock_irqrestore(&sched_debug_lock, flags);
344 static const char *sched_tunable_scaling_names[] = {
350 static void sched_debug_header(struct seq_file *m)
352 u64 ktime, sched_clk, cpu_clk;
355 local_irq_save(flags);
356 ktime = ktime_to_ns(ktime_get());
357 sched_clk = sched_clock();
358 cpu_clk = local_clock();
359 local_irq_restore(flags);
361 SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n",
362 init_utsname()->release,
363 (int)strcspn(init_utsname()->version, " "),
364 init_utsname()->version);
367 SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
369 SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
374 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
375 P(sched_clock_stable());
381 SEQ_printf(m, "sysctl_sched\n");
384 SEQ_printf(m, " .%-40s: %Ld\n", #x, (long long)(x))
386 SEQ_printf(m, " .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
387 PN(sysctl_sched_latency);
388 PN(sysctl_sched_min_granularity);
389 PN(sysctl_sched_wakeup_granularity);
390 P(sysctl_sched_child_runs_first);
391 P(sysctl_sched_features);
395 SEQ_printf(m, " .%-40s: %d (%s)\n",
396 "sysctl_sched_tunable_scaling",
397 sysctl_sched_tunable_scaling,
398 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
402 static int sched_debug_show(struct seq_file *m, void *v)
404 int cpu = (unsigned long)(v - 2);
409 sched_debug_header(m);
414 void sysrq_sched_debug_show(void)
418 sched_debug_header(NULL);
419 for_each_online_cpu(cpu)
420 print_cpu(NULL, cpu);
425 * This itererator needs some explanation.
426 * It returns 1 for the header position.
427 * This means 2 is cpu 0.
428 * In a hotplugged system some cpus, including cpu 0, may be missing so we have
429 * to use cpumask_* to iterate over the cpus.
431 static void *sched_debug_start(struct seq_file *file, loff_t *offset)
433 unsigned long n = *offset;
441 n = cpumask_next(n - 1, cpu_online_mask);
443 n = cpumask_first(cpu_online_mask);
448 return (void *)(unsigned long)(n + 2);
452 static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
455 return sched_debug_start(file, offset);
458 static void sched_debug_stop(struct seq_file *file, void *data)
462 static const struct seq_operations sched_debug_sops = {
463 .start = sched_debug_start,
464 .next = sched_debug_next,
465 .stop = sched_debug_stop,
466 .show = sched_debug_show,
469 static int sched_debug_release(struct inode *inode, struct file *file)
471 seq_release(inode, file);
476 static int sched_debug_open(struct inode *inode, struct file *filp)
480 ret = seq_open(filp, &sched_debug_sops);
485 static const struct file_operations sched_debug_fops = {
486 .open = sched_debug_open,
489 .release = sched_debug_release,
492 static int __init init_sched_debug_procfs(void)
494 struct proc_dir_entry *pe;
496 pe = proc_create("sched_debug", 0444, NULL, &sched_debug_fops);
502 __initcall(init_sched_debug_procfs);
505 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)F)
507 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)p->F)
509 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
511 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
514 #ifdef CONFIG_NUMA_BALANCING
515 void print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
516 unsigned long tpf, unsigned long gsf, unsigned long gpf)
518 SEQ_printf(m, "numa_faults node=%d ", node);
519 SEQ_printf(m, "task_private=%lu task_shared=%lu ", tsf, tpf);
520 SEQ_printf(m, "group_private=%lu group_shared=%lu\n", gsf, gpf);
525 static void sched_show_numa(struct task_struct *p, struct seq_file *m)
527 #ifdef CONFIG_NUMA_BALANCING
528 struct mempolicy *pol;
531 P(mm->numa_scan_seq);
535 if (pol && !(pol->flags & MPOL_F_MORON))
540 P(numa_pages_migrated);
541 P(numa_preferred_nid);
542 P(total_numa_faults);
543 SEQ_printf(m, "current_node=%d, numa_group_id=%d\n",
544 task_node(p), task_numa_group_id(p));
545 show_numa_stats(p, m);
550 void proc_sched_show_task(struct task_struct *p, struct seq_file *m)
552 unsigned long nr_switches;
554 SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr(p),
557 "---------------------------------------------------------"
560 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)F)
562 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)p->F)
564 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
566 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
570 PN(se.sum_exec_runtime);
572 nr_switches = p->nvcsw + p->nivcsw;
574 #ifdef CONFIG_SCHEDSTATS
575 PN(se.statistics.sum_sleep_runtime);
576 PN(se.statistics.wait_start);
577 PN(se.statistics.sleep_start);
578 PN(se.statistics.block_start);
579 PN(se.statistics.sleep_max);
580 PN(se.statistics.block_max);
581 PN(se.statistics.exec_max);
582 PN(se.statistics.slice_max);
583 PN(se.statistics.wait_max);
584 PN(se.statistics.wait_sum);
585 P(se.statistics.wait_count);
586 PN(se.statistics.iowait_sum);
587 P(se.statistics.iowait_count);
589 P(se.statistics.nr_migrations_cold);
590 P(se.statistics.nr_failed_migrations_affine);
591 P(se.statistics.nr_failed_migrations_running);
592 P(se.statistics.nr_failed_migrations_hot);
593 P(se.statistics.nr_forced_migrations);
594 P(se.statistics.nr_wakeups);
595 P(se.statistics.nr_wakeups_sync);
596 P(se.statistics.nr_wakeups_migrate);
597 P(se.statistics.nr_wakeups_local);
598 P(se.statistics.nr_wakeups_remote);
599 P(se.statistics.nr_wakeups_affine);
600 P(se.statistics.nr_wakeups_affine_attempts);
601 P(se.statistics.nr_wakeups_passive);
602 P(se.statistics.nr_wakeups_idle);
605 u64 avg_atom, avg_per_cpu;
607 avg_atom = p->se.sum_exec_runtime;
609 avg_atom = div64_ul(avg_atom, nr_switches);
613 avg_per_cpu = p->se.sum_exec_runtime;
614 if (p->se.nr_migrations) {
615 avg_per_cpu = div64_u64(avg_per_cpu,
616 p->se.nr_migrations);
626 SEQ_printf(m, "%-45s:%21Ld\n",
627 "nr_voluntary_switches", (long long)p->nvcsw);
628 SEQ_printf(m, "%-45s:%21Ld\n",
629 "nr_involuntary_switches", (long long)p->nivcsw);
637 P(se.avg.last_update_time);
641 #ifdef CONFIG_PREEMPT_RT_FULL
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));