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))
114 * May be NULL if the underlying cgroup isn't fully-created yet
116 if (!tg->css.cgroup) {
117 group_path[0] = '\0';
120 cgroup_path(tg->css.cgroup, group_path, PATH_MAX);
126 print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
133 SEQ_printf(m, "%15s %5d %9Ld.%06ld %9Ld %5d ",
135 SPLIT_NS(p->se.vruntime),
136 (long long)(p->nvcsw + p->nivcsw),
138 #ifdef CONFIG_SCHEDSTATS
139 SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
140 SPLIT_NS(p->se.vruntime),
141 SPLIT_NS(p->se.sum_exec_runtime),
142 SPLIT_NS(p->se.statistics.sum_sleep_runtime));
144 SEQ_printf(m, "%15Ld %15Ld %15Ld.%06ld %15Ld.%06ld %15Ld.%06ld",
145 0LL, 0LL, 0LL, 0L, 0LL, 0L, 0LL, 0L);
147 #ifdef CONFIG_CGROUP_SCHED
148 SEQ_printf(m, " %s", task_group_path(task_group(p)));
154 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
156 struct task_struct *g, *p;
160 "\nrunnable tasks:\n"
161 " task PID tree-key switches prio"
162 " exec-runtime sum-exec sum-sleep\n"
163 "------------------------------------------------------"
164 "----------------------------------------------------\n");
166 read_lock_irqsave(&tasklist_lock, flags);
168 do_each_thread(g, p) {
169 if (!p->on_rq || task_cpu(p) != rq_cpu)
172 print_task(m, rq, p);
173 } while_each_thread(g, p);
175 read_unlock_irqrestore(&tasklist_lock, flags);
178 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
180 s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
181 spread, rq0_min_vruntime, spread0;
182 struct rq *rq = cpu_rq(cpu);
183 struct sched_entity *last;
186 #ifdef CONFIG_FAIR_GROUP_SCHED
187 SEQ_printf(m, "\ncfs_rq[%d]:%s\n", cpu, task_group_path(cfs_rq->tg));
189 SEQ_printf(m, "\ncfs_rq[%d]:\n", cpu);
191 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "exec_clock",
192 SPLIT_NS(cfs_rq->exec_clock));
194 raw_spin_lock_irqsave(&rq->lock, flags);
195 if (cfs_rq->rb_leftmost)
196 MIN_vruntime = (__pick_first_entity(cfs_rq))->vruntime;
197 last = __pick_last_entity(cfs_rq);
199 max_vruntime = last->vruntime;
200 min_vruntime = cfs_rq->min_vruntime;
201 rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
202 raw_spin_unlock_irqrestore(&rq->lock, flags);
203 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "MIN_vruntime",
204 SPLIT_NS(MIN_vruntime));
205 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "min_vruntime",
206 SPLIT_NS(min_vruntime));
207 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "max_vruntime",
208 SPLIT_NS(max_vruntime));
209 spread = max_vruntime - MIN_vruntime;
210 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread",
212 spread0 = min_vruntime - rq0_min_vruntime;
213 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread0",
215 SEQ_printf(m, " .%-30s: %d\n", "nr_spread_over",
216 cfs_rq->nr_spread_over);
217 SEQ_printf(m, " .%-30s: %d\n", "nr_running", cfs_rq->nr_running);
218 SEQ_printf(m, " .%-30s: %ld\n", "load", cfs_rq->load.weight);
219 #ifdef CONFIG_FAIR_GROUP_SCHED
221 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "load_avg",
222 SPLIT_NS(cfs_rq->load_avg));
223 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "load_period",
224 SPLIT_NS(cfs_rq->load_period));
225 SEQ_printf(m, " .%-30s: %ld\n", "load_contrib",
226 cfs_rq->load_contribution);
227 SEQ_printf(m, " .%-30s: %d\n", "load_tg",
228 atomic_read(&cfs_rq->tg->load_weight));
229 SEQ_printf(m, " .%-30s: %lld\n", "runnable_load_avg",
230 cfs_rq->runnable_load_avg);
231 SEQ_printf(m, " .%-30s: %lld\n", "blocked_load_avg",
232 cfs_rq->blocked_load_avg);
233 SEQ_printf(m, " .%-30s: %ld\n", "tg_load_avg",
234 atomic64_read(&cfs_rq->tg->load_avg));
235 SEQ_printf(m, " .%-30s: %lld\n", "tg_load_contrib",
236 cfs_rq->tg_load_contrib);
237 SEQ_printf(m, " .%-30s: %d\n", "tg_runnable_contrib",
238 cfs_rq->tg_runnable_contrib);
239 SEQ_printf(m, " .%-30s: %d\n", "tg->runnable_avg",
240 atomic_read(&cfs_rq->tg->runnable_avg));
243 print_cfs_group_stats(m, cpu, cfs_rq->tg);
247 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
249 #ifdef CONFIG_RT_GROUP_SCHED
250 SEQ_printf(m, "\nrt_rq[%d]:%s\n", cpu, task_group_path(rt_rq->tg));
252 SEQ_printf(m, "\nrt_rq[%d]:\n", cpu);
256 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
258 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
269 extern __read_mostly int sched_clock_running;
271 static void print_cpu(struct seq_file *m, int cpu)
273 struct rq *rq = cpu_rq(cpu);
278 unsigned int freq = cpu_khz ? : 1;
280 SEQ_printf(m, "\ncpu#%d, %u.%03u MHz\n",
281 cpu, freq / 1000, (freq % 1000));
284 SEQ_printf(m, "\ncpu#%d\n", cpu);
289 if (sizeof(rq->x) == 4) \
290 SEQ_printf(m, " .%-30s: %ld\n", #x, (long)(rq->x)); \
292 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rq->x));\
296 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
299 SEQ_printf(m, " .%-30s: %lu\n", "load",
303 P(nr_uninterruptible);
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);
333 spin_lock_irqsave(&sched_debug_lock, flags);
334 print_cfs_stats(m, cpu);
335 print_rt_stats(m, cpu);
338 print_rq(m, rq, cpu);
340 spin_unlock_irqrestore(&sched_debug_lock, flags);
343 static const char *sched_tunable_scaling_names[] = {
349 static int sched_debug_show(struct seq_file *m, void *v)
351 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: v0.10, %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", "sysctl_sched_tunable_scaling",
396 sysctl_sched_tunable_scaling,
397 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
399 for_each_online_cpu(cpu)
407 void sysrq_sched_debug_show(void)
409 sched_debug_show(NULL, NULL);
412 static int sched_debug_open(struct inode *inode, struct file *filp)
414 return single_open(filp, sched_debug_show, NULL);
417 static const struct file_operations sched_debug_fops = {
418 .open = sched_debug_open,
421 .release = single_release,
424 static int __init init_sched_debug_procfs(void)
426 struct proc_dir_entry *pe;
428 pe = proc_create("sched_debug", 0444, NULL, &sched_debug_fops);
434 __initcall(init_sched_debug_procfs);
436 void proc_sched_show_task(struct task_struct *p, struct seq_file *m)
438 unsigned long nr_switches;
440 SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, p->pid,
443 "---------------------------------------------------------\n");
445 SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)F)
447 SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)p->F)
449 SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
451 SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
455 PN(se.sum_exec_runtime);
457 nr_switches = p->nvcsw + p->nivcsw;
459 #ifdef CONFIG_SCHEDSTATS
460 PN(se.statistics.wait_start);
461 PN(se.statistics.sleep_start);
462 PN(se.statistics.block_start);
463 PN(se.statistics.sleep_max);
464 PN(se.statistics.block_max);
465 PN(se.statistics.exec_max);
466 PN(se.statistics.slice_max);
467 PN(se.statistics.wait_max);
468 PN(se.statistics.wait_sum);
469 P(se.statistics.wait_count);
470 PN(se.statistics.iowait_sum);
471 P(se.statistics.iowait_count);
473 P(se.statistics.nr_migrations_cold);
474 P(se.statistics.nr_failed_migrations_affine);
475 P(se.statistics.nr_failed_migrations_running);
476 P(se.statistics.nr_failed_migrations_hot);
477 P(se.statistics.nr_forced_migrations);
478 P(se.statistics.nr_wakeups);
479 P(se.statistics.nr_wakeups_sync);
480 P(se.statistics.nr_wakeups_migrate);
481 P(se.statistics.nr_wakeups_local);
482 P(se.statistics.nr_wakeups_remote);
483 P(se.statistics.nr_wakeups_affine);
484 P(se.statistics.nr_wakeups_affine_attempts);
485 P(se.statistics.nr_wakeups_passive);
486 P(se.statistics.nr_wakeups_idle);
489 u64 avg_atom, avg_per_cpu;
491 avg_atom = p->se.sum_exec_runtime;
493 do_div(avg_atom, nr_switches);
497 avg_per_cpu = p->se.sum_exec_runtime;
498 if (p->se.nr_migrations) {
499 avg_per_cpu = div64_u64(avg_per_cpu,
500 p->se.nr_migrations);
510 SEQ_printf(m, "%-35s:%21Ld\n",
511 "nr_voluntary_switches", (long long)p->nvcsw);
512 SEQ_printf(m, "%-35s:%21Ld\n",
513 "nr_involuntary_switches", (long long)p->nivcsw);
524 unsigned int this_cpu = raw_smp_processor_id();
527 t0 = cpu_clock(this_cpu);
528 t1 = cpu_clock(this_cpu);
529 SEQ_printf(m, "%-35s:%21Ld\n",
530 "clock-delta", (long long)(t1-t0));
534 void proc_sched_set_task(struct task_struct *p)
536 #ifdef CONFIG_SCHEDSTATS
537 memset(&p->se.statistics, 0, sizeof(p->se.statistics));