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", "util_avg",
211 cfs_rq->avg.util_avg);
212 SEQ_printf(m, " .%-30s: %ld\n", "removed_load_avg",
213 atomic_long_read(&cfs_rq->removed_load_avg));
214 SEQ_printf(m, " .%-30s: %ld\n", "removed_util_avg",
215 atomic_long_read(&cfs_rq->removed_util_avg));
216 #ifdef CONFIG_FAIR_GROUP_SCHED
217 SEQ_printf(m, " .%-30s: %lu\n", "tg_load_avg_contrib",
218 cfs_rq->tg_load_avg_contrib);
219 SEQ_printf(m, " .%-30s: %ld\n", "tg_load_avg",
220 atomic_long_read(&cfs_rq->tg->load_avg));
223 #ifdef CONFIG_CFS_BANDWIDTH
224 SEQ_printf(m, " .%-30s: %d\n", "throttled",
226 SEQ_printf(m, " .%-30s: %d\n", "throttle_count",
227 cfs_rq->throttle_count);
230 #ifdef CONFIG_FAIR_GROUP_SCHED
231 print_cfs_group_stats(m, cpu, cfs_rq->tg);
235 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
237 #ifdef CONFIG_RT_GROUP_SCHED
238 SEQ_printf(m, "\nrt_rq[%d]:%s\n", cpu, task_group_path(rt_rq->tg));
240 SEQ_printf(m, "\nrt_rq[%d]:\n", cpu);
244 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
246 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
257 void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq)
259 SEQ_printf(m, "\ndl_rq[%d]:\n", cpu);
260 SEQ_printf(m, " .%-30s: %ld\n", "dl_nr_running", dl_rq->dl_nr_running);
263 extern __read_mostly int sched_clock_running;
265 static void print_cpu(struct seq_file *m, int cpu)
267 struct rq *rq = cpu_rq(cpu);
272 unsigned int freq = cpu_khz ? : 1;
274 SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
275 cpu, freq / 1000, (freq % 1000));
278 SEQ_printf(m, "cpu#%d\n", cpu);
283 if (sizeof(rq->x) == 4) \
284 SEQ_printf(m, " .%-30s: %ld\n", #x, (long)(rq->x)); \
286 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rq->x));\
290 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
293 SEQ_printf(m, " .%-30s: %lu\n", "load",
297 P(nr_uninterruptible);
299 SEQ_printf(m, " .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr)));
310 #ifdef CONFIG_SCHEDSTATS
311 #define P(n) SEQ_printf(m, " .%-30s: %d\n", #n, rq->n);
312 #define P64(n) SEQ_printf(m, " .%-30s: %Ld\n", #n, rq->n);
320 P64(max_idle_balance_cost);
329 spin_lock_irqsave(&sched_debug_lock, flags);
330 print_cfs_stats(m, cpu);
331 print_rt_stats(m, cpu);
332 print_dl_stats(m, cpu);
334 print_rq(m, rq, cpu);
335 spin_unlock_irqrestore(&sched_debug_lock, flags);
339 static const char *sched_tunable_scaling_names[] = {
345 static void sched_debug_header(struct seq_file *m)
347 u64 ktime, sched_clk, cpu_clk;
350 local_irq_save(flags);
351 ktime = ktime_to_ns(ktime_get());
352 sched_clk = sched_clock();
353 cpu_clk = local_clock();
354 local_irq_restore(flags);
356 SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n",
357 init_utsname()->release,
358 (int)strcspn(init_utsname()->version, " "),
359 init_utsname()->version);
362 SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
364 SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
369 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
370 P(sched_clock_stable());
376 SEQ_printf(m, "sysctl_sched\n");
379 SEQ_printf(m, " .%-40s: %Ld\n", #x, (long long)(x))
381 SEQ_printf(m, " .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
382 PN(sysctl_sched_latency);
383 PN(sysctl_sched_min_granularity);
384 PN(sysctl_sched_wakeup_granularity);
385 P(sysctl_sched_child_runs_first);
386 P(sysctl_sched_features);
390 SEQ_printf(m, " .%-40s: %d (%s)\n",
391 "sysctl_sched_tunable_scaling",
392 sysctl_sched_tunable_scaling,
393 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
397 static int sched_debug_show(struct seq_file *m, void *v)
399 int cpu = (unsigned long)(v - 2);
404 sched_debug_header(m);
409 void sysrq_sched_debug_show(void)
413 sched_debug_header(NULL);
414 for_each_online_cpu(cpu)
415 print_cpu(NULL, cpu);
420 * This itererator needs some explanation.
421 * It returns 1 for the header position.
422 * This means 2 is cpu 0.
423 * In a hotplugged system some cpus, including cpu 0, may be missing so we have
424 * to use cpumask_* to iterate over the cpus.
426 static void *sched_debug_start(struct seq_file *file, loff_t *offset)
428 unsigned long n = *offset;
436 n = cpumask_next(n - 1, cpu_online_mask);
438 n = cpumask_first(cpu_online_mask);
443 return (void *)(unsigned long)(n + 2);
447 static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
450 return sched_debug_start(file, offset);
453 static void sched_debug_stop(struct seq_file *file, void *data)
457 static const struct seq_operations sched_debug_sops = {
458 .start = sched_debug_start,
459 .next = sched_debug_next,
460 .stop = sched_debug_stop,
461 .show = sched_debug_show,
464 static int sched_debug_release(struct inode *inode, struct file *file)
466 seq_release(inode, file);
471 static int sched_debug_open(struct inode *inode, struct file *filp)
475 ret = seq_open(filp, &sched_debug_sops);
480 static const struct file_operations sched_debug_fops = {
481 .open = sched_debug_open,
484 .release = sched_debug_release,
487 static int __init init_sched_debug_procfs(void)
489 struct proc_dir_entry *pe;
491 pe = proc_create("sched_debug", 0444, NULL, &sched_debug_fops);
497 __initcall(init_sched_debug_procfs);
500 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)F)
502 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)p->F)
504 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
506 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
509 #ifdef CONFIG_NUMA_BALANCING
510 void print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
511 unsigned long tpf, unsigned long gsf, unsigned long gpf)
513 SEQ_printf(m, "numa_faults node=%d ", node);
514 SEQ_printf(m, "task_private=%lu task_shared=%lu ", tsf, tpf);
515 SEQ_printf(m, "group_private=%lu group_shared=%lu\n", gsf, gpf);
520 static void sched_show_numa(struct task_struct *p, struct seq_file *m)
522 #ifdef CONFIG_NUMA_BALANCING
523 struct mempolicy *pol;
526 P(mm->numa_scan_seq);
530 if (pol && !(pol->flags & MPOL_F_MORON))
535 P(numa_pages_migrated);
536 P(numa_preferred_nid);
537 P(total_numa_faults);
538 SEQ_printf(m, "current_node=%d, numa_group_id=%d\n",
539 task_node(p), task_numa_group_id(p));
540 show_numa_stats(p, m);
545 void proc_sched_show_task(struct task_struct *p, struct seq_file *m)
547 unsigned long nr_switches;
549 SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr(p),
552 "---------------------------------------------------------"
555 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)F)
557 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)p->F)
559 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
561 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
565 PN(se.sum_exec_runtime);
567 nr_switches = p->nvcsw + p->nivcsw;
569 #ifdef CONFIG_SCHEDSTATS
570 PN(se.statistics.sum_sleep_runtime);
571 PN(se.statistics.wait_start);
572 PN(se.statistics.sleep_start);
573 PN(se.statistics.block_start);
574 PN(se.statistics.sleep_max);
575 PN(se.statistics.block_max);
576 PN(se.statistics.exec_max);
577 PN(se.statistics.slice_max);
578 PN(se.statistics.wait_max);
579 PN(se.statistics.wait_sum);
580 P(se.statistics.wait_count);
581 PN(se.statistics.iowait_sum);
582 P(se.statistics.iowait_count);
584 P(se.statistics.nr_migrations_cold);
585 P(se.statistics.nr_failed_migrations_affine);
586 P(se.statistics.nr_failed_migrations_running);
587 P(se.statistics.nr_failed_migrations_hot);
588 P(se.statistics.nr_forced_migrations);
589 P(se.statistics.nr_wakeups);
590 P(se.statistics.nr_wakeups_sync);
591 P(se.statistics.nr_wakeups_migrate);
592 P(se.statistics.nr_wakeups_local);
593 P(se.statistics.nr_wakeups_remote);
594 P(se.statistics.nr_wakeups_affine);
595 P(se.statistics.nr_wakeups_affine_attempts);
596 P(se.statistics.nr_wakeups_passive);
597 P(se.statistics.nr_wakeups_idle);
600 u64 avg_atom, avg_per_cpu;
602 avg_atom = p->se.sum_exec_runtime;
604 avg_atom = div64_ul(avg_atom, nr_switches);
608 avg_per_cpu = p->se.sum_exec_runtime;
609 if (p->se.nr_migrations) {
610 avg_per_cpu = div64_u64(avg_per_cpu,
611 p->se.nr_migrations);
621 SEQ_printf(m, "%-45s:%21Ld\n",
622 "nr_voluntary_switches", (long long)p->nvcsw);
623 SEQ_printf(m, "%-45s:%21Ld\n",
624 "nr_involuntary_switches", (long long)p->nivcsw);
632 P(se.avg.last_update_time);
642 unsigned int this_cpu = raw_smp_processor_id();
645 t0 = cpu_clock(this_cpu);
646 t1 = cpu_clock(this_cpu);
647 SEQ_printf(m, "%-45s:%21Ld\n",
648 "clock-delta", (long long)(t1-t0));
651 sched_show_numa(p, m);
654 void proc_sched_set_task(struct task_struct *p)
656 #ifdef CONFIG_SCHEDSTATS
657 memset(&p->se.statistics, 0, sizeof(p->se.statistics));