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 cgroup_path(tg->css.cgroup, group_path, PATH_MAX);
120 print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
127 SEQ_printf(m, "%15s %5d %9Ld.%06ld %9Ld %5d ",
128 p->comm, task_pid_nr(p),
129 SPLIT_NS(p->se.vruntime),
130 (long long)(p->nvcsw + p->nivcsw),
132 #ifdef CONFIG_SCHEDSTATS
133 SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
134 SPLIT_NS(p->se.vruntime),
135 SPLIT_NS(p->se.sum_exec_runtime),
136 SPLIT_NS(p->se.statistics.sum_sleep_runtime));
138 SEQ_printf(m, "%15Ld %15Ld %15Ld.%06ld %15Ld.%06ld %15Ld.%06ld",
139 0LL, 0LL, 0LL, 0L, 0LL, 0L, 0LL, 0L);
141 #ifdef CONFIG_NUMA_BALANCING
142 SEQ_printf(m, " %d", cpu_to_node(task_cpu(p)));
144 #ifdef CONFIG_CGROUP_SCHED
145 SEQ_printf(m, " %s", task_group_path(task_group(p)));
151 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
153 struct task_struct *g, *p;
157 "\nrunnable tasks:\n"
158 " task PID tree-key switches prio"
159 " exec-runtime sum-exec sum-sleep\n"
160 "------------------------------------------------------"
161 "----------------------------------------------------\n");
163 read_lock_irqsave(&tasklist_lock, flags);
165 do_each_thread(g, p) {
166 if (task_cpu(p) != rq_cpu)
169 print_task(m, rq, p);
170 } while_each_thread(g, p);
172 read_unlock_irqrestore(&tasklist_lock, flags);
175 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
177 s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
178 spread, rq0_min_vruntime, spread0;
179 struct rq *rq = cpu_rq(cpu);
180 struct sched_entity *last;
183 #ifdef CONFIG_FAIR_GROUP_SCHED
184 SEQ_printf(m, "\ncfs_rq[%d]:%s\n", cpu, task_group_path(cfs_rq->tg));
186 SEQ_printf(m, "\ncfs_rq[%d]:\n", cpu);
188 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "exec_clock",
189 SPLIT_NS(cfs_rq->exec_clock));
191 raw_spin_lock_irqsave(&rq->lock, flags);
192 if (cfs_rq->rb_leftmost)
193 MIN_vruntime = (__pick_first_entity(cfs_rq))->vruntime;
194 last = __pick_last_entity(cfs_rq);
196 max_vruntime = last->vruntime;
197 min_vruntime = cfs_rq->min_vruntime;
198 rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
199 raw_spin_unlock_irqrestore(&rq->lock, flags);
200 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "MIN_vruntime",
201 SPLIT_NS(MIN_vruntime));
202 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "min_vruntime",
203 SPLIT_NS(min_vruntime));
204 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "max_vruntime",
205 SPLIT_NS(max_vruntime));
206 spread = max_vruntime - MIN_vruntime;
207 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread",
209 spread0 = min_vruntime - rq0_min_vruntime;
210 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread0",
212 SEQ_printf(m, " .%-30s: %d\n", "nr_spread_over",
213 cfs_rq->nr_spread_over);
214 SEQ_printf(m, " .%-30s: %d\n", "nr_running", cfs_rq->nr_running);
215 SEQ_printf(m, " .%-30s: %ld\n", "load", cfs_rq->load.weight);
217 SEQ_printf(m, " .%-30s: %ld\n", "runnable_load_avg",
218 cfs_rq->runnable_load_avg);
219 SEQ_printf(m, " .%-30s: %ld\n", "blocked_load_avg",
220 cfs_rq->blocked_load_avg);
221 #ifdef CONFIG_FAIR_GROUP_SCHED
222 SEQ_printf(m, " .%-30s: %ld\n", "tg_load_contrib",
223 cfs_rq->tg_load_contrib);
224 SEQ_printf(m, " .%-30s: %d\n", "tg_runnable_contrib",
225 cfs_rq->tg_runnable_contrib);
226 SEQ_printf(m, " .%-30s: %ld\n", "tg_load_avg",
227 atomic_long_read(&cfs_rq->tg->load_avg));
228 SEQ_printf(m, " .%-30s: %d\n", "tg->runnable_avg",
229 atomic_read(&cfs_rq->tg->runnable_avg));
233 #ifdef CONFIG_FAIR_GROUP_SCHED
234 print_cfs_group_stats(m, cpu, cfs_rq->tg);
238 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
240 #ifdef CONFIG_RT_GROUP_SCHED
241 SEQ_printf(m, "\nrt_rq[%d]:%s\n", cpu, task_group_path(rt_rq->tg));
243 SEQ_printf(m, "\nrt_rq[%d]:\n", cpu);
247 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
249 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
260 extern __read_mostly int sched_clock_running;
262 static void print_cpu(struct seq_file *m, int cpu)
264 struct rq *rq = cpu_rq(cpu);
269 unsigned int freq = cpu_khz ? : 1;
271 SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
272 cpu, freq / 1000, (freq % 1000));
275 SEQ_printf(m, "cpu#%d\n", cpu);
280 if (sizeof(rq->x) == 4) \
281 SEQ_printf(m, " .%-30s: %ld\n", #x, (long)(rq->x)); \
283 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rq->x));\
287 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
290 SEQ_printf(m, " .%-30s: %lu\n", "load",
294 P(nr_uninterruptible);
296 SEQ_printf(m, " .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr)));
306 #ifdef CONFIG_SCHEDSTATS
307 #define P(n) SEQ_printf(m, " .%-30s: %d\n", #n, rq->n);
308 #define P64(n) SEQ_printf(m, " .%-30s: %Ld\n", #n, rq->n);
324 spin_lock_irqsave(&sched_debug_lock, flags);
325 print_cfs_stats(m, cpu);
326 print_rt_stats(m, cpu);
329 print_rq(m, rq, cpu);
331 spin_unlock_irqrestore(&sched_debug_lock, flags);
335 static const char *sched_tunable_scaling_names[] = {
341 static void sched_debug_header(struct seq_file *m)
343 u64 ktime, sched_clk, cpu_clk;
346 local_irq_save(flags);
347 ktime = ktime_to_ns(ktime_get());
348 sched_clk = sched_clock();
349 cpu_clk = local_clock();
350 local_irq_restore(flags);
352 SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n",
353 init_utsname()->release,
354 (int)strcspn(init_utsname()->version, " "),
355 init_utsname()->version);
358 SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
360 SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
365 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
366 P(sched_clock_stable);
372 SEQ_printf(m, "sysctl_sched\n");
375 SEQ_printf(m, " .%-40s: %Ld\n", #x, (long long)(x))
377 SEQ_printf(m, " .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
378 PN(sysctl_sched_latency);
379 PN(sysctl_sched_min_granularity);
380 PN(sysctl_sched_wakeup_granularity);
381 P(sysctl_sched_child_runs_first);
382 P(sysctl_sched_features);
386 SEQ_printf(m, " .%-40s: %d (%s)\n",
387 "sysctl_sched_tunable_scaling",
388 sysctl_sched_tunable_scaling,
389 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
393 static int sched_debug_show(struct seq_file *m, void *v)
395 int cpu = (unsigned long)(v - 2);
400 sched_debug_header(m);
405 void sysrq_sched_debug_show(void)
409 sched_debug_header(NULL);
410 for_each_online_cpu(cpu)
411 print_cpu(NULL, cpu);
416 * This itererator needs some explanation.
417 * It returns 1 for the header position.
418 * This means 2 is cpu 0.
419 * In a hotplugged system some cpus, including cpu 0, may be missing so we have
420 * to use cpumask_* to iterate over the cpus.
422 static void *sched_debug_start(struct seq_file *file, loff_t *offset)
424 unsigned long n = *offset;
432 n = cpumask_next(n - 1, cpu_online_mask);
434 n = cpumask_first(cpu_online_mask);
439 return (void *)(unsigned long)(n + 2);
443 static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
446 return sched_debug_start(file, offset);
449 static void sched_debug_stop(struct seq_file *file, void *data)
453 static const struct seq_operations sched_debug_sops = {
454 .start = sched_debug_start,
455 .next = sched_debug_next,
456 .stop = sched_debug_stop,
457 .show = sched_debug_show,
460 static int sched_debug_release(struct inode *inode, struct file *file)
462 seq_release(inode, file);
467 static int sched_debug_open(struct inode *inode, struct file *filp)
471 ret = seq_open(filp, &sched_debug_sops);
476 static const struct file_operations sched_debug_fops = {
477 .open = sched_debug_open,
480 .release = sched_debug_release,
483 static int __init init_sched_debug_procfs(void)
485 struct proc_dir_entry *pe;
487 pe = proc_create("sched_debug", 0444, NULL, &sched_debug_fops);
493 __initcall(init_sched_debug_procfs);
496 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)F)
498 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)p->F)
500 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
502 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
505 static void sched_show_numa(struct task_struct *p, struct seq_file *m)
507 #ifdef CONFIG_NUMA_BALANCING
508 struct mempolicy *pol;
512 P(mm->numa_scan_seq);
516 if (pol && !(pol->flags & MPOL_F_MORON))
521 SEQ_printf(m, "numa_migrations, %ld\n", xchg(&p->numa_pages_migrated, 0));
523 for_each_online_node(node) {
524 for (i = 0; i < 2; i++) {
525 unsigned long nr_faults = -1;
526 int cpu_current, home_node;
529 nr_faults = p->numa_faults[2*node + i];
531 cpu_current = !i ? (task_node(p) == node) :
532 (pol && node_isset(node, pol->v.nodes));
534 home_node = (p->numa_preferred_nid == node);
536 SEQ_printf(m, "numa_faults, %d, %d, %d, %d, %ld\n",
537 i, node, cpu_current, home_node, nr_faults);
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.wait_start);
571 PN(se.statistics.sleep_start);
572 PN(se.statistics.block_start);
573 PN(se.statistics.sleep_max);
574 PN(se.statistics.block_max);
575 PN(se.statistics.exec_max);
576 PN(se.statistics.slice_max);
577 PN(se.statistics.wait_max);
578 PN(se.statistics.wait_sum);
579 P(se.statistics.wait_count);
580 PN(se.statistics.iowait_sum);
581 P(se.statistics.iowait_count);
583 P(se.statistics.nr_migrations_cold);
584 P(se.statistics.nr_failed_migrations_affine);
585 P(se.statistics.nr_failed_migrations_running);
586 P(se.statistics.nr_failed_migrations_hot);
587 P(se.statistics.nr_forced_migrations);
588 P(se.statistics.nr_wakeups);
589 P(se.statistics.nr_wakeups_sync);
590 P(se.statistics.nr_wakeups_migrate);
591 P(se.statistics.nr_wakeups_local);
592 P(se.statistics.nr_wakeups_remote);
593 P(se.statistics.nr_wakeups_affine);
594 P(se.statistics.nr_wakeups_affine_attempts);
595 P(se.statistics.nr_wakeups_passive);
596 P(se.statistics.nr_wakeups_idle);
599 u64 avg_atom, avg_per_cpu;
601 avg_atom = p->se.sum_exec_runtime;
603 do_div(avg_atom, nr_switches);
607 avg_per_cpu = p->se.sum_exec_runtime;
608 if (p->se.nr_migrations) {
609 avg_per_cpu = div64_u64(avg_per_cpu,
610 p->se.nr_migrations);
620 SEQ_printf(m, "%-45s:%21Ld\n",
621 "nr_voluntary_switches", (long long)p->nvcsw);
622 SEQ_printf(m, "%-45s:%21Ld\n",
623 "nr_involuntary_switches", (long long)p->nivcsw);
627 P(se.avg.runnable_avg_sum);
628 P(se.avg.runnable_avg_period);
629 P(se.avg.load_avg_contrib);
630 P(se.avg.decay_count);
640 unsigned int this_cpu = raw_smp_processor_id();
643 t0 = cpu_clock(this_cpu);
644 t1 = cpu_clock(this_cpu);
645 SEQ_printf(m, "%-45s:%21Ld\n",
646 "clock-delta", (long long)(t1-t0));
649 sched_show_numa(p, m);
652 void proc_sched_set_task(struct task_struct *p)
654 #ifdef CONFIG_SCHEDSTATS
655 memset(&p->se.statistics, 0, sizeof(p->se.statistics));