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>
19 static DEFINE_SPINLOCK(sched_debug_lock);
22 * This allows printing both to /proc/sched_debug and
25 #define SEQ_printf(m, x...) \
34 * Ease the printing of nsec fields:
36 static long long nsec_high(unsigned long long nsec)
38 if ((long long)nsec < 0) {
40 do_div(nsec, 1000000);
43 do_div(nsec, 1000000);
48 static unsigned long nsec_low(unsigned long long nsec)
50 if ((long long)nsec < 0)
53 return do_div(nsec, 1000000);
56 #define SPLIT_NS(x) nsec_high(x), nsec_low(x)
58 #ifdef CONFIG_FAIR_GROUP_SCHED
59 static void print_cfs_group_stats(struct seq_file *m, int cpu, struct task_group *tg)
61 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))
72 PN(se->sum_exec_runtime);
73 #ifdef CONFIG_SCHEDSTATS
74 PN(se->statistics.wait_start);
75 PN(se->statistics.sleep_start);
76 PN(se->statistics.block_start);
77 PN(se->statistics.sleep_max);
78 PN(se->statistics.block_max);
79 PN(se->statistics.exec_max);
80 PN(se->statistics.slice_max);
81 PN(se->statistics.wait_max);
82 PN(se->statistics.wait_sum);
83 P(se->statistics.wait_count);
91 #ifdef CONFIG_CGROUP_SCHED
92 static char group_path[PATH_MAX];
94 static char *task_group_path(struct task_group *tg)
97 * May be NULL if the underlying cgroup isn't fully-created yet
99 if (!tg->css.cgroup) {
100 group_path[0] = '\0';
103 cgroup_path(tg->css.cgroup, group_path, PATH_MAX);
109 print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
116 SEQ_printf(m, "%15s %5d %9Ld.%06ld %9Ld %5d ",
118 SPLIT_NS(p->se.vruntime),
119 (long long)(p->nvcsw + p->nivcsw),
121 #ifdef CONFIG_SCHEDSTATS
122 SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
123 SPLIT_NS(p->se.vruntime),
124 SPLIT_NS(p->se.sum_exec_runtime),
125 SPLIT_NS(p->se.statistics.sum_sleep_runtime));
127 SEQ_printf(m, "%15Ld %15Ld %15Ld.%06ld %15Ld.%06ld %15Ld.%06ld",
128 0LL, 0LL, 0LL, 0L, 0LL, 0L, 0LL, 0L);
130 #ifdef CONFIG_CGROUP_SCHED
131 SEQ_printf(m, " %s", task_group_path(task_group(p)));
137 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
139 struct task_struct *g, *p;
143 "\nrunnable tasks:\n"
144 " task PID tree-key switches prio"
145 " exec-runtime sum-exec sum-sleep\n"
146 "------------------------------------------------------"
147 "----------------------------------------------------\n");
149 read_lock_irqsave(&tasklist_lock, flags);
151 do_each_thread(g, p) {
152 if (!p->se.on_rq || task_cpu(p) != rq_cpu)
155 print_task(m, rq, p);
156 } while_each_thread(g, p);
158 read_unlock_irqrestore(&tasklist_lock, flags);
161 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
163 s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
164 spread, rq0_min_vruntime, spread0;
165 struct rq *rq = cpu_rq(cpu);
166 struct sched_entity *last;
169 #ifdef CONFIG_FAIR_GROUP_SCHED
170 SEQ_printf(m, "\ncfs_rq[%d]:%s\n", cpu, task_group_path(cfs_rq->tg));
172 SEQ_printf(m, "\ncfs_rq[%d]:\n", cpu);
174 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "exec_clock",
175 SPLIT_NS(cfs_rq->exec_clock));
177 raw_spin_lock_irqsave(&rq->lock, flags);
178 if (cfs_rq->rb_leftmost)
179 MIN_vruntime = (__pick_next_entity(cfs_rq))->vruntime;
180 last = __pick_last_entity(cfs_rq);
182 max_vruntime = last->vruntime;
183 min_vruntime = cfs_rq->min_vruntime;
184 rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
185 raw_spin_unlock_irqrestore(&rq->lock, flags);
186 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "MIN_vruntime",
187 SPLIT_NS(MIN_vruntime));
188 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "min_vruntime",
189 SPLIT_NS(min_vruntime));
190 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "max_vruntime",
191 SPLIT_NS(max_vruntime));
192 spread = max_vruntime - MIN_vruntime;
193 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread",
195 spread0 = min_vruntime - rq0_min_vruntime;
196 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread0",
198 SEQ_printf(m, " .%-30s: %d\n", "nr_spread_over",
199 cfs_rq->nr_spread_over);
200 SEQ_printf(m, " .%-30s: %ld\n", "nr_running", cfs_rq->nr_running);
201 SEQ_printf(m, " .%-30s: %ld\n", "load", cfs_rq->load.weight);
202 #ifdef CONFIG_FAIR_GROUP_SCHED
204 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "load_avg",
205 SPLIT_NS(cfs_rq->load_avg));
206 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "load_period",
207 SPLIT_NS(cfs_rq->load_period));
208 SEQ_printf(m, " .%-30s: %ld\n", "load_contrib",
209 cfs_rq->load_contribution);
210 SEQ_printf(m, " .%-30s: %d\n", "load_tg",
211 atomic_read(&cfs_rq->tg->load_weight));
214 print_cfs_group_stats(m, cpu, cfs_rq->tg);
218 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
220 #ifdef CONFIG_RT_GROUP_SCHED
221 SEQ_printf(m, "\nrt_rq[%d]:%s\n", cpu, task_group_path(rt_rq->tg));
223 SEQ_printf(m, "\nrt_rq[%d]:\n", cpu);
227 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
229 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
240 extern __read_mostly int sched_clock_running;
242 static void print_cpu(struct seq_file *m, int cpu)
244 struct rq *rq = cpu_rq(cpu);
249 unsigned int freq = cpu_khz ? : 1;
251 SEQ_printf(m, "\ncpu#%d, %u.%03u MHz\n",
252 cpu, freq / 1000, (freq % 1000));
255 SEQ_printf(m, "\ncpu#%d\n", cpu);
259 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rq->x))
261 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
264 SEQ_printf(m, " .%-30s: %lu\n", "load",
268 P(nr_uninterruptible);
280 #ifdef CONFIG_SCHEDSTATS
281 #define P(n) SEQ_printf(m, " .%-30s: %d\n", #n, rq->n);
282 #define P64(n) SEQ_printf(m, " .%-30s: %Ld\n", #n, rq->n);
300 spin_lock_irqsave(&sched_debug_lock, flags);
301 print_cfs_stats(m, cpu);
302 print_rt_stats(m, cpu);
305 print_rq(m, rq, cpu);
307 spin_unlock_irqrestore(&sched_debug_lock, flags);
310 static const char *sched_tunable_scaling_names[] = {
316 static int sched_debug_show(struct seq_file *m, void *v)
318 u64 ktime, sched_clk, cpu_clk;
322 local_irq_save(flags);
323 ktime = ktime_to_ns(ktime_get());
324 sched_clk = sched_clock();
325 cpu_clk = local_clock();
326 local_irq_restore(flags);
328 SEQ_printf(m, "Sched Debug Version: v0.10, %s %.*s\n",
329 init_utsname()->release,
330 (int)strcspn(init_utsname()->version, " "),
331 init_utsname()->version);
334 SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
336 SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
341 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
342 P(sched_clock_stable);
348 SEQ_printf(m, "sysctl_sched\n");
351 SEQ_printf(m, " .%-40s: %Ld\n", #x, (long long)(x))
353 SEQ_printf(m, " .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
354 PN(sysctl_sched_latency);
355 PN(sysctl_sched_min_granularity);
356 PN(sysctl_sched_wakeup_granularity);
357 P(sysctl_sched_child_runs_first);
358 P(sysctl_sched_features);
362 SEQ_printf(m, " .%-40s: %d (%s)\n", "sysctl_sched_tunable_scaling",
363 sysctl_sched_tunable_scaling,
364 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
366 for_each_online_cpu(cpu)
374 static void sysrq_sched_debug_show(void)
376 sched_debug_show(NULL, NULL);
379 static int sched_debug_open(struct inode *inode, struct file *filp)
381 return single_open(filp, sched_debug_show, NULL);
384 static const struct file_operations sched_debug_fops = {
385 .open = sched_debug_open,
388 .release = single_release,
391 static int __init init_sched_debug_procfs(void)
393 struct proc_dir_entry *pe;
395 pe = proc_create("sched_debug", 0444, NULL, &sched_debug_fops);
401 __initcall(init_sched_debug_procfs);
403 void proc_sched_show_task(struct task_struct *p, struct seq_file *m)
405 unsigned long nr_switches;
407 SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, p->pid,
410 "---------------------------------------------------------\n");
412 SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)F)
414 SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)p->F)
416 SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
418 SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
422 PN(se.sum_exec_runtime);
424 nr_switches = p->nvcsw + p->nivcsw;
426 #ifdef CONFIG_SCHEDSTATS
427 PN(se.statistics.wait_start);
428 PN(se.statistics.sleep_start);
429 PN(se.statistics.block_start);
430 PN(se.statistics.sleep_max);
431 PN(se.statistics.block_max);
432 PN(se.statistics.exec_max);
433 PN(se.statistics.slice_max);
434 PN(se.statistics.wait_max);
435 PN(se.statistics.wait_sum);
436 P(se.statistics.wait_count);
437 PN(se.statistics.iowait_sum);
438 P(se.statistics.iowait_count);
439 P(sched_info.bkl_count);
441 P(se.statistics.nr_migrations_cold);
442 P(se.statistics.nr_failed_migrations_affine);
443 P(se.statistics.nr_failed_migrations_running);
444 P(se.statistics.nr_failed_migrations_hot);
445 P(se.statistics.nr_forced_migrations);
446 P(se.statistics.nr_wakeups);
447 P(se.statistics.nr_wakeups_sync);
448 P(se.statistics.nr_wakeups_migrate);
449 P(se.statistics.nr_wakeups_local);
450 P(se.statistics.nr_wakeups_remote);
451 P(se.statistics.nr_wakeups_affine);
452 P(se.statistics.nr_wakeups_affine_attempts);
453 P(se.statistics.nr_wakeups_passive);
454 P(se.statistics.nr_wakeups_idle);
457 u64 avg_atom, avg_per_cpu;
459 avg_atom = p->se.sum_exec_runtime;
461 do_div(avg_atom, nr_switches);
465 avg_per_cpu = p->se.sum_exec_runtime;
466 if (p->se.nr_migrations) {
467 avg_per_cpu = div64_u64(avg_per_cpu,
468 p->se.nr_migrations);
478 SEQ_printf(m, "%-35s:%21Ld\n",
479 "nr_voluntary_switches", (long long)p->nvcsw);
480 SEQ_printf(m, "%-35s:%21Ld\n",
481 "nr_involuntary_switches", (long long)p->nivcsw);
492 unsigned int this_cpu = raw_smp_processor_id();
495 t0 = cpu_clock(this_cpu);
496 t1 = cpu_clock(this_cpu);
497 SEQ_printf(m, "%-35s:%21Ld\n",
498 "clock-delta", (long long)(t1-t0));
502 void proc_sched_set_task(struct task_struct *p)
504 #ifdef CONFIG_SCHEDSTATS
505 memset(&p->se.statistics, 0, sizeof(p->se.statistics));