1
2
3
4
5
6
7
8
9
10
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
26
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
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 if (schedstat_enabled()) {
84 PN(se->statistics->wait_start);
85 PN(se->statistics->sleep_start);
86 PN(se->statistics->block_start);
87 PN(se->statistics->sleep_max);
88 PN(se->statistics->block_max);
89 PN(se->statistics->exec_max);
90 PN(se->statistics->slice_max);
91 PN(se->statistics->wait_max);
92 PN(se->statistics->wait_sum);
93 P(se->statistics->wait_count);
94 }
95#endif
96 P(se->load.weight);
97#ifdef CONFIG_SMP
98 P(se->avg.runnable_avg_sum);
99 P(se->avg.runnable_avg_period);
100 P(se->avg.load_avg_contrib);
101 P(se->avg.decay_count);
102#endif
103#undef PN
104#undef P
105}
106#endif
107
108#ifdef CONFIG_CGROUP_SCHED
109static char group_path[PATH_MAX];
110
111static char *task_group_path(struct task_group *tg)
112{
113 if (autogroup_path(tg, group_path, PATH_MAX))
114 return group_path;
115
116 cgroup_path(tg->css.cgroup, group_path, PATH_MAX);
117 return group_path;
118}
119#endif
120
121static void
122print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
123{
124 if (rq->curr == p)
125 SEQ_printf(m, "R");
126 else
127 SEQ_printf(m, " ");
128
129 SEQ_printf(m, "%15s %5d %9Ld.%06ld %9Ld %5d ",
130 p->comm, p->pid,
131 SPLIT_NS(p->se.vruntime),
132 (long long)(p->nvcsw + p->nivcsw),
133 p->prio);
134
135 SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
136 SPLIT_NS(schedstat_val(p, se.statistics->wait_sum)),
137 SPLIT_NS(p->se.sum_exec_runtime),
138 SPLIT_NS(schedstat_val(p, se.statistics->sum_sleep_runtime)));
139
140#ifdef CONFIG_NUMA_BALANCING
141 SEQ_printf(m, " %d", cpu_to_node(task_cpu(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 unsigned long flags;
154
155 SEQ_printf(m,
156 "\nrunnable tasks:\n"
157 " task PID tree-key switches prio"
158 " wait-time sum-exec sum-sleep\n"
159 "------------------------------------------------------"
160 "----------------------------------------------------\n");
161
162 qread_lock_irqsave(&tasklist_lock, flags);
163
164 do_each_thread(g, p) {
165 if (task_cpu(p) != rq_cpu)
166 continue;
167
168 print_task(m, rq, p);
169 } while_each_thread(g, p);
170
171 qread_unlock_irqrestore(&tasklist_lock, flags);
172}
173
174void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
175{
176 s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
177 spread, rq0_min_vruntime, spread0;
178 struct rq *rq = cpu_rq(cpu);
179 struct sched_entity *last;
180 unsigned long flags;
181
182#ifdef CONFIG_FAIR_GROUP_SCHED
183 SEQ_printf(m, "\ncfs_rq[%d]:%s\n", cpu, task_group_path(cfs_rq->tg));
184#else
185 SEQ_printf(m, "\ncfs_rq[%d]:\n", cpu);
186#endif
187 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "exec_clock",
188 SPLIT_NS(cfs_rq->exec_clock));
189
190 raw_spin_lock_irqsave(&rq->lock, flags);
191 if (cfs_rq->rb_leftmost)
192 MIN_vruntime = (__pick_first_entity(cfs_rq))->vruntime;
193 last = __pick_last_entity(cfs_rq);
194 if (last)
195 max_vruntime = last->vruntime;
196 min_vruntime = cfs_rq->min_vruntime;
197 rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
198 raw_spin_unlock_irqrestore(&rq->lock, flags);
199 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "MIN_vruntime",
200 SPLIT_NS(MIN_vruntime));
201 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "min_vruntime",
202 SPLIT_NS(min_vruntime));
203 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "max_vruntime",
204 SPLIT_NS(max_vruntime));
205 spread = max_vruntime - MIN_vruntime;
206 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread",
207 SPLIT_NS(spread));
208 spread0 = min_vruntime - rq0_min_vruntime;
209 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread0",
210 SPLIT_NS(spread0));
211 SEQ_printf(m, " .%-30s: %d\n", "nr_spread_over",
212 cfs_rq->nr_spread_over);
213 SEQ_printf(m, " .%-30s: %d\n", "nr_running", cfs_rq->nr_running);
214 SEQ_printf(m, " .%-30s: %ld\n", "load", cfs_rq->load.weight);
215#ifdef CONFIG_FAIR_GROUP_SCHED
216#ifdef CONFIG_SMP
217 SEQ_printf(m, " .%-30s: %lld\n", "runnable_load_avg",
218 cfs_rq->runnable_load_avg);
219 SEQ_printf(m, " .%-30s: %lld\n", "blocked_load_avg",
220 cfs_rq->blocked_load_avg);
221 SEQ_printf(m, " .%-30s: %lld\n", "tg_load_avg",
222 (unsigned long long)atomic64_read(&cfs_rq->tg->load_avg));
223 SEQ_printf(m, " .%-30s: %lld\n", "tg_load_contrib",
224 cfs_rq->tg_load_contrib);
225 SEQ_printf(m, " .%-30s: %d\n", "tg_runnable_contrib",
226 cfs_rq->tg_runnable_contrib);
227 SEQ_printf(m, " .%-30s: %d\n", "tg->runnable_avg",
228 atomic_read(&cfs_rq->tg->runnable_avg));
229#endif
230#ifdef CONFIG_CFS_BANDWIDTH
231 SEQ_printf(m, " .%-30s: %d\n", "tg->cfs_bandwidth.timer_active",
232 cfs_rq->tg->cfs_bandwidth.timer_active);
233 SEQ_printf(m, " .%-30s: %d\n", "throttled",
234 cfs_rq->throttled);
235 SEQ_printf(m, " .%-30s: %d\n", "throttle_count",
236 cfs_rq->throttle_count);
237#endif
238
239 print_cfs_group_stats(m, cpu, cfs_rq->tg);
240#endif
241}
242
243void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
244{
245#ifdef CONFIG_RT_GROUP_SCHED
246 SEQ_printf(m, "\nrt_rq[%d]:%s\n", cpu, task_group_path(rt_rq->tg));
247#else
248 SEQ_printf(m, "\nrt_rq[%d]:\n", cpu);
249#endif
250
251#define P(x) \
252 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
253#define PN(x) \
254 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
255
256 P(rt_nr_running);
257 P(rt_throttled);
258 PN(rt_time);
259 PN(rt_runtime);
260
261#undef PN
262#undef P
263}
264
265extern __read_mostly int sched_clock_running;
266
267static void print_cpu(struct seq_file *m, int cpu)
268{
269 struct rq *rq = cpu_rq(cpu);
270 unsigned long flags;
271
272#ifdef CONFIG_X86
273 {
274 unsigned int freq = cpu_khz ? : 1;
275
276 SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
277 cpu, freq / 1000, (freq % 1000));
278 }
279#else
280 SEQ_printf(m, "cpu#%d\n", cpu);
281#endif
282
283#define P(x) \
284do { \
285 if (sizeof(rq->x) == 4) \
286 SEQ_printf(m, " .%-30s: %ld\n", #x, (long)(rq->x)); \
287 else \
288 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rq->x));\
289} while (0)
290
291#define PN(x) \
292 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
293
294 P(nr_running);
295 SEQ_printf(m, " .%-30s: %lu\n", "load",
296 rq->load.weight);
297 P(nr_switches);
298 P(nr_load_updates);
299 P(nr_uninterruptible);
300 PN(next_balance);
301 P(curr->pid);
302 PN(clock);
303 P(cpu_load[0]);
304 P(cpu_load[1]);
305 P(cpu_load[2]);
306 P(cpu_load[3]);
307 P(cpu_load[4]);
308#undef P
309#undef PN
310
311#ifdef CONFIG_SCHEDSTATS
312#define P(n) SEQ_printf(m, " .%-30s: %d\n", #n, rq->n);
313#define P64(n) SEQ_printf(m, " .%-30s: %Ld\n", #n, rq->n);
314
315#ifdef CONFIG_SMP
316 P64(avg_idle);
317 P64(max_idle_balance_cost);
318#endif
319
320 if (schedstat_enabled()) {
321 P(yld_count);
322 P(sched_count);
323 P(sched_goidle);
324 P(ttwu_count);
325 P(ttwu_local);
326 }
327
328#undef P
329#undef P64
330#endif
331 spin_lock_irqsave(&sched_debug_lock, flags);
332 print_cfs_stats(m, cpu);
333 print_rt_stats(m, cpu);
334
335 rcu_read_lock();
336 print_rq(m, rq, cpu);
337 rcu_read_unlock();
338 spin_unlock_irqrestore(&sched_debug_lock, flags);
339 SEQ_printf(m, "\n");
340}
341
342static const char *sched_tunable_scaling_names[] = {
343 "none",
344 "logaritmic",
345 "linear"
346};
347
348static void sched_debug_header(struct seq_file *m)
349{
350 u64 ktime, sched_clk, cpu_clk;
351 unsigned long flags;
352
353 local_irq_save(flags);
354 ktime = ktime_to_ns(ktime_get());
355 sched_clk = sched_clock();
356 cpu_clk = local_clock();
357 local_irq_restore(flags);
358
359 SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n",
360 init_utsname()->release,
361 (int)strcspn(init_utsname()->version, " "),
362 init_utsname()->version);
363
364#define P(x) \
365 SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
366#define PN(x) \
367 SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
368 PN(ktime);
369 PN(sched_clk);
370 PN(cpu_clk);
371 P(jiffies);
372#ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
373 P(sched_clock_stable());
374#endif
375#undef PN
376#undef P
377
378 SEQ_printf(m, "\n");
379 SEQ_printf(m, "sysctl_sched\n");
380
381#define P(x) \
382 SEQ_printf(m, " .%-40s: %Ld\n", #x, (long long)(x))
383#define PN(x) \
384 SEQ_printf(m, " .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
385 PN(sysctl_sched_latency);
386 PN(sysctl_sched_min_granularity);
387 PN(sysctl_sched_wakeup_granularity);
388 P(sysctl_sched_child_runs_first);
389 P(sysctl_sched_features);
390#undef PN
391#undef P
392
393 SEQ_printf(m, " .%-40s: %d (%s)\n",
394 "sysctl_sched_tunable_scaling",
395 sysctl_sched_tunable_scaling,
396 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
397 SEQ_printf(m, "\n");
398}
399
400static int sched_debug_show(struct seq_file *m, void *v)
401{
402 int cpu = (unsigned long)(v - 2);
403
404 if (cpu != -1)
405 print_cpu(m, cpu);
406 else
407 sched_debug_header(m);
408
409 return 0;
410}
411
412void sysrq_sched_debug_show(void)
413{
414 int cpu;
415
416 sched_debug_header(NULL);
417 for_each_online_cpu(cpu)
418 print_cpu(NULL, cpu);
419
420}
421
422
423
424
425
426
427
428
429static void *sched_debug_start(struct seq_file *file, loff_t *offset)
430{
431 unsigned long n = *offset;
432
433 if (n == 0)
434 return (void *) 1;
435
436 n--;
437
438 if (n > 0)
439 n = cpumask_next(n - 1, cpu_online_mask);
440 else
441 n = cpumask_first(cpu_online_mask);
442
443 *offset = n + 1;
444
445 if (n < nr_cpu_ids)
446 return (void *)(unsigned long)(n + 2);
447 return NULL;
448}
449
450static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
451{
452 (*offset)++;
453 return sched_debug_start(file, offset);
454}
455
456static void sched_debug_stop(struct seq_file *file, void *data)
457{
458}
459
460static const struct seq_operations sched_debug_sops = {
461 .start = sched_debug_start,
462 .next = sched_debug_next,
463 .stop = sched_debug_stop,
464 .show = sched_debug_show,
465};
466
467static int sched_debug_release(struct inode *inode, struct file *file)
468{
469 seq_release(inode, file);
470
471 return 0;
472}
473
474static int sched_debug_open(struct inode *inode, struct file *filp)
475{
476 int ret = 0;
477
478 ret = seq_open(filp, &sched_debug_sops);
479
480 return ret;
481}
482
483static const struct file_operations sched_debug_fops = {
484 .open = sched_debug_open,
485 .read = seq_read,
486 .llseek = seq_lseek,
487 .release = sched_debug_release,
488};
489
490static int __init init_sched_debug_procfs(void)
491{
492 struct proc_dir_entry *pe;
493
494 pe = proc_create("sched_debug", 0444, NULL, &sched_debug_fops);
495 if (!pe)
496 return -ENOMEM;
497 return 0;
498}
499
500__initcall(init_sched_debug_procfs);
501
502#define __P(F) \
503 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)F)
504#define P(F) \
505 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)p->F)
506#define __PN(F) \
507 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
508#define PN(F) \
509 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
510
511
512static void sched_show_numa(struct task_struct *p, struct seq_file *m)
513{
514#ifdef CONFIG_NUMA_BALANCING
515 struct mempolicy *pol;
516 int node, i;
517
518 if (p->mm)
519 P(mm->numa_scan_seq);
520
521 task_lock(p);
522 pol = p->mempolicy;
523 if (pol && !(pol->flags & MPOL_F_MORON))
524 pol = NULL;
525 mpol_get(pol);
526 task_unlock(p);
527
528 SEQ_printf(m, "numa_migrations, %ld\n", xchg(&p->numa_pages_migrated, 0));
529
530 for_each_online_node(node) {
531 for (i = 0; i < 2; i++) {
532 unsigned long nr_faults = -1;
533 int cpu_current, home_node;
534
535 if (p->numa_faults_memory)
536 nr_faults = p->numa_faults_memory[2*node + i];
537
538 cpu_current = !i ? (task_node(p) == node) :
539 (pol && node_isset(node, pol->v.nodes));
540
541 home_node = (p->numa_preferred_nid == node);
542
543 SEQ_printf(m, "numa_faults_memory, %d, %d, %d, %d, %ld\n",
544 i, node, cpu_current, home_node, nr_faults);
545 }
546 }
547
548 mpol_put(pol);
549#endif
550}
551
552void proc_sched_show_task(struct task_struct *p, struct seq_file *m)
553{
554 unsigned long nr_switches;
555
556 SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, p->pid,
557 get_nr_threads(p));
558 SEQ_printf(m,
559 "---------------------------------------------------------"
560 "----------\n");
561#define __P(F) \
562 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)F)
563#define P(F) \
564 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)p->F)
565#define __PN(F) \
566 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
567#define PN(F) \
568 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
569
570 PN(se.exec_start);
571 PN(se.vruntime);
572 PN(se.sum_exec_runtime);
573
574 nr_switches = p->nvcsw + p->nivcsw;
575
576#ifdef CONFIG_SCHEDSTATS
577 P(se.nr_migrations);
578
579 if (schedstat_enabled()) {
580 u64 avg_atom, avg_per_cpu;
581
582 PN(se.statistics->sum_sleep_runtime);
583 PN(se.statistics->wait_start);
584 PN(se.statistics->sleep_start);
585 PN(se.statistics->block_start);
586 PN(se.statistics->sleep_max);
587 PN(se.statistics->block_max);
588 PN(se.statistics->exec_max);
589 PN(se.statistics->slice_max);
590 PN(se.statistics->wait_max);
591 PN(se.statistics->wait_sum);
592 P(se.statistics->wait_count);
593 PN(se.statistics->iowait_sum);
594 P(se.statistics->iowait_count);
595 P(se.statistics->nr_migrations_cold);
596 P(se.statistics->nr_failed_migrations_affine);
597 P(se.statistics->nr_failed_migrations_running);
598 P(se.statistics->nr_failed_migrations_hot);
599 P(se.statistics->nr_forced_migrations);
600 P(se.statistics->nr_wakeups);
601 P(se.statistics->nr_wakeups_sync);
602 P(se.statistics->nr_wakeups_migrate);
603 P(se.statistics->nr_wakeups_local);
604 P(se.statistics->nr_wakeups_remote);
605 P(se.statistics->nr_wakeups_affine);
606 P(se.statistics->nr_wakeups_affine_attempts);
607 P(se.statistics->nr_wakeups_passive);
608 P(se.statistics->nr_wakeups_idle);
609
610 avg_atom = p->se.sum_exec_runtime;
611 if (nr_switches)
612 avg_atom = div64_ul(avg_atom, nr_switches);
613 else
614 avg_atom = -1LL;
615
616 avg_per_cpu = p->se.sum_exec_runtime;
617 if (p->se.nr_migrations) {
618 avg_per_cpu = div64_u64(avg_per_cpu,
619 p->se.nr_migrations);
620 } else {
621 avg_per_cpu = -1LL;
622 }
623
624 __PN(avg_atom);
625 __PN(avg_per_cpu);
626 }
627#endif
628 __P(nr_switches);
629 SEQ_printf(m, "%-45s:%21Ld\n",
630 "nr_voluntary_switches", (long long)p->nvcsw);
631 SEQ_printf(m, "%-45s:%21Ld\n",
632 "nr_involuntary_switches", (long long)p->nivcsw);
633
634 P(se.load.weight);
635 P(policy);
636 P(prio);
637#undef PN
638#undef __PN
639#undef P
640#undef __P
641
642 {
643 unsigned int this_cpu = raw_smp_processor_id();
644 u64 t0, t1;
645
646 t0 = cpu_clock(this_cpu);
647 t1 = cpu_clock(this_cpu);
648 SEQ_printf(m, "%-45s:%21Ld\n",
649 "clock-delta", (long long)(t1-t0));
650 }
651
652 sched_show_numa(p, m);
653}
654
655void proc_sched_set_task(struct task_struct *p)
656{
657#ifdef CONFIG_SCHEDSTATS
658 memset(p->se.statistics, 0, sizeof(*p->se.statistics));
659#endif
660}
661
