2 * kernel/time/sched_debug.c
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>
20 * This allows printing both to /proc/sched_debug and
23 #define SEQ_printf(m, x...) \
32 * Ease the printing of nsec fields:
34 static long long nsec_high(unsigned long long nsec)
36 if ((long long)nsec < 0) {
38 do_div(nsec, 1000000);
41 do_div(nsec, 1000000);
46 static unsigned long nsec_low(unsigned long long nsec)
48 if ((long long)nsec < 0)
51 return do_div(nsec, 1000000);
54 #define SPLIT_NS(x) nsec_high(x), nsec_low(x)
56 #ifdef CONFIG_FAIR_GROUP_SCHED
57 static void print_cfs_group_stats(struct seq_file *m, int cpu,
58 struct task_group *tg)
60 struct sched_entity *se = tg->se[cpu];
65 SEQ_printf(m, " .%-30s: %lld\n", #F, (long long)F)
67 SEQ_printf(m, " .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F))
71 PN(se->sum_exec_runtime);
72 #ifdef CONFIG_SCHEDSTATS
73 PN(se->statistics.wait_start);
74 PN(se->statistics.sleep_start);
75 PN(se->statistics.block_start);
76 PN(se->statistics.sleep_max);
77 PN(se->statistics.block_max);
78 PN(se->statistics.exec_max);
79 PN(se->statistics.slice_max);
80 PN(se->statistics.wait_max);
81 PN(se->statistics.wait_sum);
82 P(se->statistics.wait_count);
91 print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
98 SEQ_printf(m, "%15s %5d %9Ld.%06ld %9Ld %5d ",
100 SPLIT_NS(p->se.vruntime),
101 (long long)(p->nvcsw + p->nivcsw),
103 #ifdef CONFIG_SCHEDSTATS
104 SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
105 SPLIT_NS(p->se.vruntime),
106 SPLIT_NS(p->se.sum_exec_runtime),
107 SPLIT_NS(p->se.statistics.sum_sleep_runtime));
109 SEQ_printf(m, "%15Ld %15Ld %15Ld.%06ld %15Ld.%06ld %15Ld.%06ld",
110 0LL, 0LL, 0LL, 0L, 0LL, 0L, 0LL, 0L);
113 #ifdef CONFIG_CGROUP_SCHED
118 cgroup_path(task_group(p)->css.cgroup, path, sizeof(path));
120 SEQ_printf(m, " %s", path);
126 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
128 struct task_struct *g, *p;
132 "\nrunnable tasks:\n"
133 " task PID tree-key switches prio"
134 " exec-runtime sum-exec sum-sleep\n"
135 "------------------------------------------------------"
136 "----------------------------------------------------\n");
138 read_lock_irqsave(&tasklist_lock, flags);
140 do_each_thread(g, p) {
141 if (!p->se.on_rq || task_cpu(p) != rq_cpu)
144 print_task(m, rq, p);
145 } while_each_thread(g, p);
147 read_unlock_irqrestore(&tasklist_lock, flags);
150 #if defined(CONFIG_CGROUP_SCHED) && \
151 (defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED))
152 static void task_group_path(struct task_group *tg, char *buf, int buflen)
154 /* may be NULL if the underlying cgroup isn't fully-created yet */
155 if (!tg->css.cgroup) {
159 cgroup_path(tg->css.cgroup, buf, buflen);
163 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
165 s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
166 spread, rq0_min_vruntime, spread0;
167 struct rq *rq = cpu_rq(cpu);
168 struct sched_entity *last;
171 #if defined(CONFIG_CGROUP_SCHED) && defined(CONFIG_FAIR_GROUP_SCHED)
173 struct task_group *tg = cfs_rq->tg;
175 task_group_path(tg, path, sizeof(path));
177 SEQ_printf(m, "\ncfs_rq[%d]:%s\n", cpu, path);
179 SEQ_printf(m, "\ncfs_rq[%d]:\n", cpu);
181 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "exec_clock",
182 SPLIT_NS(cfs_rq->exec_clock));
184 raw_spin_lock_irqsave(&rq->lock, flags);
185 if (cfs_rq->rb_leftmost)
186 MIN_vruntime = (__pick_next_entity(cfs_rq))->vruntime;
187 last = __pick_last_entity(cfs_rq);
189 max_vruntime = last->vruntime;
190 min_vruntime = cfs_rq->min_vruntime;
191 rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
192 raw_spin_unlock_irqrestore(&rq->lock, flags);
193 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "MIN_vruntime",
194 SPLIT_NS(MIN_vruntime));
195 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "min_vruntime",
196 SPLIT_NS(min_vruntime));
197 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "max_vruntime",
198 SPLIT_NS(max_vruntime));
199 spread = max_vruntime - MIN_vruntime;
200 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread",
202 spread0 = min_vruntime - rq0_min_vruntime;
203 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread0",
205 SEQ_printf(m, " .%-30s: %d\n", "nr_spread_over",
206 cfs_rq->nr_spread_over);
207 SEQ_printf(m, " .%-30s: %ld\n", "nr_running", cfs_rq->nr_running);
208 SEQ_printf(m, " .%-30s: %ld\n", "load", cfs_rq->load.weight);
209 #ifdef CONFIG_FAIR_GROUP_SCHED
211 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "load_avg",
212 SPLIT_NS(cfs_rq->load_avg));
213 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "load_period",
214 SPLIT_NS(cfs_rq->load_period));
215 SEQ_printf(m, " .%-30s: %ld\n", "load_contrib",
216 cfs_rq->load_contribution);
217 SEQ_printf(m, " .%-30s: %d\n", "load_tg",
218 atomic_read(&tg->load_weight));
221 print_cfs_group_stats(m, cpu, cfs_rq->tg);
225 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
227 #if defined(CONFIG_CGROUP_SCHED) && defined(CONFIG_RT_GROUP_SCHED)
229 struct task_group *tg = rt_rq->tg;
231 task_group_path(tg, path, sizeof(path));
233 SEQ_printf(m, "\nrt_rq[%d]:%s\n", cpu, path);
235 SEQ_printf(m, "\nrt_rq[%d]:\n", cpu);
240 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
242 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
253 extern __read_mostly int sched_clock_running;
255 static void print_cpu(struct seq_file *m, int cpu)
257 struct rq *rq = cpu_rq(cpu);
261 unsigned int freq = cpu_khz ? : 1;
263 SEQ_printf(m, "\ncpu#%d, %u.%03u MHz\n",
264 cpu, freq / 1000, (freq % 1000));
267 SEQ_printf(m, "\ncpu#%d\n", cpu);
271 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rq->x))
273 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
276 SEQ_printf(m, " .%-30s: %lu\n", "load",
280 P(nr_uninterruptible);
292 #ifdef CONFIG_SCHEDSTATS
293 #define P(n) SEQ_printf(m, " .%-30s: %d\n", #n, rq->n);
294 #define P64(n) SEQ_printf(m, " .%-30s: %Ld\n", #n, rq->n);
312 print_cfs_stats(m, cpu);
313 print_rt_stats(m, cpu);
315 print_rq(m, rq, cpu);
318 static const char *sched_tunable_scaling_names[] = {
324 static int sched_debug_show(struct seq_file *m, void *v)
326 u64 ktime, sched_clk, cpu_clk;
330 local_irq_save(flags);
331 ktime = ktime_to_ns(ktime_get());
332 sched_clk = sched_clock();
333 cpu_clk = local_clock();
334 local_irq_restore(flags);
336 SEQ_printf(m, "Sched Debug Version: v0.10, %s %.*s\n",
337 init_utsname()->release,
338 (int)strcspn(init_utsname()->version, " "),
339 init_utsname()->version);
342 SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
344 SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
349 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
350 P(sched_clock_stable);
356 SEQ_printf(m, "sysctl_sched\n");
359 SEQ_printf(m, " .%-40s: %Ld\n", #x, (long long)(x))
361 SEQ_printf(m, " .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
362 PN(sysctl_sched_latency);
363 PN(sysctl_sched_min_granularity);
364 PN(sysctl_sched_wakeup_granularity);
365 P(sysctl_sched_child_runs_first);
366 P(sysctl_sched_features);
370 SEQ_printf(m, " .%-40s: %d (%s)\n", "sysctl_sched_tunable_scaling",
371 sysctl_sched_tunable_scaling,
372 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
374 for_each_online_cpu(cpu)
382 static void sysrq_sched_debug_show(void)
384 sched_debug_show(NULL, NULL);
387 static int sched_debug_open(struct inode *inode, struct file *filp)
389 return single_open(filp, sched_debug_show, NULL);
392 static const struct file_operations sched_debug_fops = {
393 .open = sched_debug_open,
396 .release = single_release,
399 static int __init init_sched_debug_procfs(void)
401 struct proc_dir_entry *pe;
403 pe = proc_create("sched_debug", 0444, NULL, &sched_debug_fops);
409 __initcall(init_sched_debug_procfs);
411 void proc_sched_show_task(struct task_struct *p, struct seq_file *m)
413 unsigned long nr_switches;
415 SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, p->pid,
418 "---------------------------------------------------------\n");
420 SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)F)
422 SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)p->F)
424 SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
426 SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
430 PN(se.sum_exec_runtime);
432 nr_switches = p->nvcsw + p->nivcsw;
434 #ifdef CONFIG_SCHEDSTATS
435 PN(se.statistics.wait_start);
436 PN(se.statistics.sleep_start);
437 PN(se.statistics.block_start);
438 PN(se.statistics.sleep_max);
439 PN(se.statistics.block_max);
440 PN(se.statistics.exec_max);
441 PN(se.statistics.slice_max);
442 PN(se.statistics.wait_max);
443 PN(se.statistics.wait_sum);
444 P(se.statistics.wait_count);
445 PN(se.statistics.iowait_sum);
446 P(se.statistics.iowait_count);
447 P(sched_info.bkl_count);
449 P(se.statistics.nr_migrations_cold);
450 P(se.statistics.nr_failed_migrations_affine);
451 P(se.statistics.nr_failed_migrations_running);
452 P(se.statistics.nr_failed_migrations_hot);
453 P(se.statistics.nr_forced_migrations);
454 P(se.statistics.nr_wakeups);
455 P(se.statistics.nr_wakeups_sync);
456 P(se.statistics.nr_wakeups_migrate);
457 P(se.statistics.nr_wakeups_local);
458 P(se.statistics.nr_wakeups_remote);
459 P(se.statistics.nr_wakeups_affine);
460 P(se.statistics.nr_wakeups_affine_attempts);
461 P(se.statistics.nr_wakeups_passive);
462 P(se.statistics.nr_wakeups_idle);
465 u64 avg_atom, avg_per_cpu;
467 avg_atom = p->se.sum_exec_runtime;
469 do_div(avg_atom, nr_switches);
473 avg_per_cpu = p->se.sum_exec_runtime;
474 if (p->se.nr_migrations) {
475 avg_per_cpu = div64_u64(avg_per_cpu,
476 p->se.nr_migrations);
486 SEQ_printf(m, "%-35s:%21Ld\n",
487 "nr_voluntary_switches", (long long)p->nvcsw);
488 SEQ_printf(m, "%-35s:%21Ld\n",
489 "nr_involuntary_switches", (long long)p->nivcsw);
500 unsigned int this_cpu = raw_smp_processor_id();
503 t0 = cpu_clock(this_cpu);
504 t1 = cpu_clock(this_cpu);
505 SEQ_printf(m, "%-35s:%21Ld\n",
506 "clock-delta", (long long)(t1-t0));
510 void proc_sched_set_task(struct task_struct *p)
512 #ifdef CONFIG_SCHEDSTATS
513 memset(&p->se.statistics, 0, sizeof(p->se.statistics));