projects/linux·blob v4.0 · refs

668 lines · 15.4 KiB

1/*
2 * kernel/sched/debug.c
3 *
4 * Print the CFS rbtree
5 *
6 * Copyright(C) 2007, Red Hat, Inc., Ingo Molnar
7 *
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.
11 */
12
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>
19
20#include "sched.h"
21
22static DEFINE_SPINLOCK(sched_debug_lock);
23
24/*
25 * This allows printing both to /proc/sched_debug and
26 * to the console
27 */
28#define SEQ_printf(m, x...) \
29 do { \
30 if (m) \
31 seq_printf(m, x); \
32 else \
33 printk(x); \
34 } while (0)
35
36/*
37 * Ease the printing of nsec fields:
38 */
39static long long nsec_high(unsigned long long nsec)
40{
41 if ((long long)nsec < 0) {
42 nsec = -nsec;
43 do_div(nsec, 1000000);
44 return -nsec;
45 }
46 do_div(nsec, 1000000);
47
48 return nsec;
49}
50
51static unsigned long nsec_low(unsigned long long nsec)
52{
53 if ((long long)nsec < 0)
54 nsec = -nsec;
55
56 return do_div(nsec, 1000000);
57}
58
59#define SPLIT_NS(x) nsec_high(x), nsec_low(x)
60
61#ifdef CONFIG_FAIR_GROUP_SCHED
62static void print_cfs_group_stats(struct seq_file *m, int cpu, struct task_group *tg)
63{
64 struct sched_entity *se = tg->se[cpu];
65
66#define P(F) \
67 SEQ_printf(m, " .%-30s: %lld\n", #F, (long long)F)
68#define PN(F) \
69 SEQ_printf(m, " .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F))
70
71 if (!se) {
72 struct sched_avg *avg = &cpu_rq(cpu)->avg;
73 P(avg->runnable_avg_sum);
74 P(avg->runnable_avg_period);
75 return;
76 }
77
78
79 PN(se->exec_start);
80 PN(se->vruntime);
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);
93#endif
94 P(se->load.weight);
95#ifdef CONFIG_SMP
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);
100#endif
101#undef PN
102#undef P
103}
104#endif
105
106#ifdef CONFIG_CGROUP_SCHED
107static char group_path[PATH_MAX];
108
109static char *task_group_path(struct task_group *tg)
110{
111 if (autogroup_path(tg, group_path, PATH_MAX))
112 return group_path;
113
114 return cgroup_path(tg->css.cgroup, group_path, PATH_MAX);
115}
116#endif
117
118static void
119print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
120{
121 if (rq->curr == p)
122 SEQ_printf(m, "R");
123 else
124 SEQ_printf(m, " ");
125
126 SEQ_printf(m, "%15s %5d %9Ld.%06ld %9Ld %5d ",
127 p->comm, task_pid_nr(p),
128 SPLIT_NS(p->se.vruntime),
129 (long long)(p->nvcsw + p->nivcsw),
130 p->prio);
131#ifdef CONFIG_SCHEDSTATS
132 SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
133 SPLIT_NS(p->se.vruntime),
134 SPLIT_NS(p->se.sum_exec_runtime),
135 SPLIT_NS(p->se.statistics.sum_sleep_runtime));
136#else
137 SEQ_printf(m, "%15Ld %15Ld %15Ld.%06ld %15Ld.%06ld %15Ld.%06ld",
138 0LL, 0LL, 0LL, 0L, 0LL, 0L, 0LL, 0L);
139#endif
140#ifdef CONFIG_NUMA_BALANCING
141 SEQ_printf(m, " %d", task_node(p));
142#endif
143#ifdef CONFIG_CGROUP_SCHED
144 SEQ_printf(m, " %s", task_group_path(task_group(p)));
145#endif
146
147 SEQ_printf(m, "\n");
148}
149
150static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
151{
152 struct task_struct *g, *p;
153
154 SEQ_printf(m,
155 "\nrunnable tasks:\n"
156 " task PID tree-key switches prio"
157 " exec-runtime sum-exec sum-sleep\n"
158 "------------------------------------------------------"
159 "----------------------------------------------------\n");
160
161 rcu_read_lock();
162 for_each_process_thread(g, p) {
163 if (task_cpu(p) != rq_cpu)
164 continue;
165
166 print_task(m, rq, p);
167 }
168 rcu_read_unlock();
169}
170
171void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
172{
173 s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
174 spread, rq0_min_vruntime, spread0;
175 struct rq *rq = cpu_rq(cpu);
176 struct sched_entity *last;
177 unsigned long flags;
178
179#ifdef CONFIG_FAIR_GROUP_SCHED
180 SEQ_printf(m, "\ncfs_rq[%d]:%s\n", cpu, task_group_path(cfs_rq->tg));
181#else
182 SEQ_printf(m, "\ncfs_rq[%d]:\n", cpu);
183#endif
184 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "exec_clock",
185 SPLIT_NS(cfs_rq->exec_clock));
186
187 raw_spin_lock_irqsave(&rq->lock, flags);
188 if (cfs_rq->rb_leftmost)
189 MIN_vruntime = (__pick_first_entity(cfs_rq))->vruntime;
190 last = __pick_last_entity(cfs_rq);
191 if (last)
192 max_vruntime = last->vruntime;
193 min_vruntime = cfs_rq->min_vruntime;
194 rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
195 raw_spin_unlock_irqrestore(&rq->lock, flags);
196 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "MIN_vruntime",
197 SPLIT_NS(MIN_vruntime));
198 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "min_vruntime",
199 SPLIT_NS(min_vruntime));
200 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "max_vruntime",
201 SPLIT_NS(max_vruntime));
202 spread = max_vruntime - MIN_vruntime;
203 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread",
204 SPLIT_NS(spread));
205 spread0 = min_vruntime - rq0_min_vruntime;
206 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread0",
207 SPLIT_NS(spread0));
208 SEQ_printf(m, " .%-30s: %d\n", "nr_spread_over",
209 cfs_rq->nr_spread_over);
210 SEQ_printf(m, " .%-30s: %d\n", "nr_running", cfs_rq->nr_running);
211 SEQ_printf(m, " .%-30s: %ld\n", "load", cfs_rq->load.weight);
212#ifdef CONFIG_SMP
213 SEQ_printf(m, " .%-30s: %ld\n", "runnable_load_avg",
214 cfs_rq->runnable_load_avg);
215 SEQ_printf(m, " .%-30s: %ld\n", "blocked_load_avg",
216 cfs_rq->blocked_load_avg);
217#ifdef CONFIG_FAIR_GROUP_SCHED
218 SEQ_printf(m, " .%-30s: %ld\n", "tg_load_contrib",
219 cfs_rq->tg_load_contrib);
220 SEQ_printf(m, " .%-30s: %d\n", "tg_runnable_contrib",
221 cfs_rq->tg_runnable_contrib);
222 SEQ_printf(m, " .%-30s: %ld\n", "tg_load_avg",
223 atomic_long_read(&cfs_rq->tg->load_avg));
224 SEQ_printf(m, " .%-30s: %d\n", "tg->runnable_avg",
225 atomic_read(&cfs_rq->tg->runnable_avg));
226#endif
227#endif
228#ifdef CONFIG_CFS_BANDWIDTH
229 SEQ_printf(m, " .%-30s: %d\n", "tg->cfs_bandwidth.timer_active",
230 cfs_rq->tg->cfs_bandwidth.timer_active);
231 SEQ_printf(m, " .%-30s: %d\n", "throttled",
232 cfs_rq->throttled);
233 SEQ_printf(m, " .%-30s: %d\n", "throttle_count",
234 cfs_rq->throttle_count);
235#endif
236
237#ifdef CONFIG_FAIR_GROUP_SCHED
238 print_cfs_group_stats(m, cpu, cfs_rq->tg);
239#endif
240}
241
242void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
243{
244#ifdef CONFIG_RT_GROUP_SCHED
245 SEQ_printf(m, "\nrt_rq[%d]:%s\n", cpu, task_group_path(rt_rq->tg));
246#else
247 SEQ_printf(m, "\nrt_rq[%d]:\n", cpu);
248#endif
249
250#define P(x) \
251 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
252#define PN(x) \
253 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
254
255 P(rt_nr_running);
256 P(rt_throttled);
257 PN(rt_time);
258 PN(rt_runtime);
259
260#undef PN
261#undef P
262}
263
264void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq)
265{
266 SEQ_printf(m, "\ndl_rq[%d]:\n", cpu);
267 SEQ_printf(m, " .%-30s: %ld\n", "dl_nr_running", dl_rq->dl_nr_running);
268}
269
270extern __read_mostly int sched_clock_running;
271
272static void print_cpu(struct seq_file *m, int cpu)
273{
274 struct rq *rq = cpu_rq(cpu);
275 unsigned long flags;
276
277#ifdef CONFIG_X86
278 {
279 unsigned int freq = cpu_khz ? : 1;
280
281 SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
282 cpu, freq / 1000, (freq % 1000));
283 }
284#else
285 SEQ_printf(m, "cpu#%d\n", cpu);
286#endif
287
288#define P(x) \
289do { \
290 if (sizeof(rq->x) == 4) \
291 SEQ_printf(m, " .%-30s: %ld\n", #x, (long)(rq->x)); \
292 else \
293 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rq->x));\
294} while (0)
295
296#define PN(x) \
297 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
298
299 P(nr_running);
300 SEQ_printf(m, " .%-30s: %lu\n", "load",
301 rq->load.weight);
302 P(nr_switches);
303 P(nr_load_updates);
304 P(nr_uninterruptible);
305 PN(next_balance);
306 SEQ_printf(m, " .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr)));
307 PN(clock);
308 PN(clock_task);
309 P(cpu_load[0]);
310 P(cpu_load[1]);
311 P(cpu_load[2]);
312 P(cpu_load[3]);
313 P(cpu_load[4]);
314#undef P
315#undef PN
316
317#ifdef CONFIG_SCHEDSTATS
318#define P(n) SEQ_printf(m, " .%-30s: %d\n", #n, rq->n);
319#define P64(n) SEQ_printf(m, " .%-30s: %Ld\n", #n, rq->n);
320
321 P(yld_count);
322
323 P(sched_count);
324 P(sched_goidle);
325#ifdef CONFIG_SMP
326 P64(avg_idle);
327 P64(max_idle_balance_cost);
328#endif
329
330 P(ttwu_count);
331 P(ttwu_local);
332
333#undef P
334#undef P64
335#endif
336 spin_lock_irqsave(&sched_debug_lock, flags);
337 print_cfs_stats(m, cpu);
338 print_rt_stats(m, cpu);
339 print_dl_stats(m, cpu);
340
341 print_rq(m, rq, cpu);
342 spin_unlock_irqrestore(&sched_debug_lock, flags);
343 SEQ_printf(m, "\n");
344}
345
346static const char *sched_tunable_scaling_names[] = {
347 "none",
348 "logaritmic",
349 "linear"
350};
351
352static void sched_debug_header(struct seq_file *m)
353{
354 u64 ktime, sched_clk, cpu_clk;
355 unsigned long flags;
356
357 local_irq_save(flags);
358 ktime = ktime_to_ns(ktime_get());
359 sched_clk = sched_clock();
360 cpu_clk = local_clock();
361 local_irq_restore(flags);
362
363 SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n",
364 init_utsname()->release,
365 (int)strcspn(init_utsname()->version, " "),
366 init_utsname()->version);
367
368#define P(x) \
369 SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
370#define PN(x) \
371 SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
372 PN(ktime);
373 PN(sched_clk);
374 PN(cpu_clk);
375 P(jiffies);
376#ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
377 P(sched_clock_stable());
378#endif
379#undef PN
380#undef P
381
382 SEQ_printf(m, "\n");
383 SEQ_printf(m, "sysctl_sched\n");
384
385#define P(x) \
386 SEQ_printf(m, " .%-40s: %Ld\n", #x, (long long)(x))
387#define PN(x) \
388 SEQ_printf(m, " .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
389 PN(sysctl_sched_latency);
390 PN(sysctl_sched_min_granularity);
391 PN(sysctl_sched_wakeup_granularity);
392 P(sysctl_sched_child_runs_first);
393 P(sysctl_sched_features);
394#undef PN
395#undef P
396
397 SEQ_printf(m, " .%-40s: %d (%s)\n",
398 "sysctl_sched_tunable_scaling",
399 sysctl_sched_tunable_scaling,
400 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
401 SEQ_printf(m, "\n");
402}
403
404static int sched_debug_show(struct seq_file *m, void *v)
405{
406 int cpu = (unsigned long)(v - 2);
407
408 if (cpu != -1)
409 print_cpu(m, cpu);
410 else
411 sched_debug_header(m);
412
413 return 0;
414}
415
416void sysrq_sched_debug_show(void)
417{
418 int cpu;
419
420 sched_debug_header(NULL);
421 for_each_online_cpu(cpu)
422 print_cpu(NULL, cpu);
423
424}
425
426/*
427 * This itererator needs some explanation.
428 * It returns 1 for the header position.
429 * This means 2 is cpu 0.
430 * In a hotplugged system some cpus, including cpu 0, may be missing so we have
431 * to use cpumask_* to iterate over the cpus.
432 */
433static void *sched_debug_start(struct seq_file *file, loff_t *offset)
434{
435 unsigned long n = *offset;
436
437 if (n == 0)
438 return (void *) 1;
439
440 n--;
441
442 if (n > 0)
443 n = cpumask_next(n - 1, cpu_online_mask);
444 else
445 n = cpumask_first(cpu_online_mask);
446
447 *offset = n + 1;
448
449 if (n < nr_cpu_ids)
450 return (void *)(unsigned long)(n + 2);
451 return NULL;
452}
453
454static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
455{
456 (*offset)++;
457 return sched_debug_start(file, offset);
458}
459
460static void sched_debug_stop(struct seq_file *file, void *data)
461{
462}
463
464static const struct seq_operations sched_debug_sops = {
465 .start = sched_debug_start,
466 .next = sched_debug_next,
467 .stop = sched_debug_stop,
468 .show = sched_debug_show,
469};
470
471static int sched_debug_release(struct inode *inode, struct file *file)
472{
473 seq_release(inode, file);
474
475 return 0;
476}
477
478static int sched_debug_open(struct inode *inode, struct file *filp)
479{
480 int ret = 0;
481
482 ret = seq_open(filp, &sched_debug_sops);
483
484 return ret;
485}
486
487static const struct file_operations sched_debug_fops = {
488 .open = sched_debug_open,
489 .read = seq_read,
490 .llseek = seq_lseek,
491 .release = sched_debug_release,
492};
493
494static int __init init_sched_debug_procfs(void)
495{
496 struct proc_dir_entry *pe;
497
498 pe = proc_create("sched_debug", 0444, NULL, &sched_debug_fops);
499 if (!pe)
500 return -ENOMEM;
501 return 0;
502}
503
504__initcall(init_sched_debug_procfs);
505
506#define __P(F) \
507 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)F)
508#define P(F) \
509 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)p->F)
510#define __PN(F) \
511 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
512#define PN(F) \
513 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
514
515
516static void sched_show_numa(struct task_struct *p, struct seq_file *m)
517{
518#ifdef CONFIG_NUMA_BALANCING
519 struct mempolicy *pol;
520 int node, i;
521
522 if (p->mm)
523 P(mm->numa_scan_seq);
524
525 task_lock(p);
526 pol = p->mempolicy;
527 if (pol && !(pol->flags & MPOL_F_MORON))
528 pol = NULL;
529 mpol_get(pol);
530 task_unlock(p);
531
532 SEQ_printf(m, "numa_migrations, %ld\n", xchg(&p->numa_pages_migrated, 0));
533
534 for_each_online_node(node) {
535 for (i = 0; i < 2; i++) {
536 unsigned long nr_faults = -1;
537 int cpu_current, home_node;
538
539 if (p->numa_faults)
540 nr_faults = p->numa_faults[2*node + i];
541
542 cpu_current = !i ? (task_node(p) == node) :
543 (pol && node_isset(node, pol->v.nodes));
544
545 home_node = (p->numa_preferred_nid == node);
546
547 SEQ_printf(m, "numa_faults_memory, %d, %d, %d, %d, %ld\n",
548 i, node, cpu_current, home_node, nr_faults);
549 }
550 }
551
552 mpol_put(pol);
553#endif
554}
555
556void proc_sched_show_task(struct task_struct *p, struct seq_file *m)
557{
558 unsigned long nr_switches;
559
560 SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr(p),
561 get_nr_threads(p));
562 SEQ_printf(m,
563 "---------------------------------------------------------"
564 "----------\n");
565#define __P(F) \
566 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)F)
567#define P(F) \
568 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)p->F)
569#define __PN(F) \
570 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
571#define PN(F) \
572 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
573
574 PN(se.exec_start);
575 PN(se.vruntime);
576 PN(se.sum_exec_runtime);
577
578 nr_switches = p->nvcsw + p->nivcsw;
579
580#ifdef CONFIG_SCHEDSTATS
581 PN(se.statistics.wait_start);
582 PN(se.statistics.sleep_start);
583 PN(se.statistics.block_start);
584 PN(se.statistics.sleep_max);
585 PN(se.statistics.block_max);
586 PN(se.statistics.exec_max);
587 PN(se.statistics.slice_max);
588 PN(se.statistics.wait_max);
589 PN(se.statistics.wait_sum);
590 P(se.statistics.wait_count);
591 PN(se.statistics.iowait_sum);
592 P(se.statistics.iowait_count);
593 P(se.nr_migrations);
594 P(se.statistics.nr_migrations_cold);
595 P(se.statistics.nr_failed_migrations_affine);
596 P(se.statistics.nr_failed_migrations_running);
597 P(se.statistics.nr_failed_migrations_hot);
598 P(se.statistics.nr_forced_migrations);
599 P(se.statistics.nr_wakeups);
600 P(se.statistics.nr_wakeups_sync);
601 P(se.statistics.nr_wakeups_migrate);
602 P(se.statistics.nr_wakeups_local);
603 P(se.statistics.nr_wakeups_remote);
604 P(se.statistics.nr_wakeups_affine);
605 P(se.statistics.nr_wakeups_affine_attempts);
606 P(se.statistics.nr_wakeups_passive);
607 P(se.statistics.nr_wakeups_idle);
608
609 {
610 u64 avg_atom, avg_per_cpu;
611
612 avg_atom = p->se.sum_exec_runtime;
613 if (nr_switches)
614 avg_atom = div64_ul(avg_atom, nr_switches);
615 else
616 avg_atom = -1LL;
617
618 avg_per_cpu = p->se.sum_exec_runtime;
619 if (p->se.nr_migrations) {
620 avg_per_cpu = div64_u64(avg_per_cpu,
621 p->se.nr_migrations);
622 } else {
623 avg_per_cpu = -1LL;
624 }
625
626 __PN(avg_atom);
627 __PN(avg_per_cpu);
628 }
629#endif
630 __P(nr_switches);
631 SEQ_printf(m, "%-45s:%21Ld\n",
632 "nr_voluntary_switches", (long long)p->nvcsw);
633 SEQ_printf(m, "%-45s:%21Ld\n",
634 "nr_involuntary_switches", (long long)p->nivcsw);
635
636 P(se.load.weight);
637#ifdef CONFIG_SMP
638 P(se.avg.runnable_avg_sum);
639 P(se.avg.runnable_avg_period);
640 P(se.avg.load_avg_contrib);
641 P(se.avg.decay_count);
642#endif
643 P(policy);
644 P(prio);
645#undef PN
646#undef __PN
647#undef P
648#undef __P
649
650 {
651 unsigned int this_cpu = raw_smp_processor_id();
652 u64 t0, t1;
653
654 t0 = cpu_clock(this_cpu);
655 t1 = cpu_clock(this_cpu);
656 SEQ_printf(m, "%-45s:%21Ld\n",
657 "clock-delta", (long long)(t1-t0));
658 }
659
660 sched_show_numa(p, m);
661}
662
663void proc_sched_set_task(struct task_struct *p)
664{
665#ifdef CONFIG_SCHEDSTATS
666 memset(&p->se.statistics, 0, sizeof(p->se.statistics));
667#endif
668}