Loading...
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 unsigned long flags;
154
155 SEQ_printf(m,
156 "\nrunnable tasks:\n"
157 " task PID tree-key switches prio"
158 " exec-runtime sum-exec sum-sleep\n"
159 "------------------------------------------------------"
160 "----------------------------------------------------\n");
161
162 read_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 read_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_SMP
216 SEQ_printf(m, " .%-30s: %ld\n", "runnable_load_avg",
217 cfs_rq->runnable_load_avg);
218 SEQ_printf(m, " .%-30s: %ld\n", "blocked_load_avg",
219 cfs_rq->blocked_load_avg);
220#ifdef CONFIG_FAIR_GROUP_SCHED
221 SEQ_printf(m, " .%-30s: %ld\n", "tg_load_contrib",
222 cfs_rq->tg_load_contrib);
223 SEQ_printf(m, " .%-30s: %d\n", "tg_runnable_contrib",
224 cfs_rq->tg_runnable_contrib);
225 SEQ_printf(m, " .%-30s: %ld\n", "tg_load_avg",
226 atomic_long_read(&cfs_rq->tg->load_avg));
227 SEQ_printf(m, " .%-30s: %d\n", "tg->runnable_avg",
228 atomic_read(&cfs_rq->tg->runnable_avg));
229#endif
230#endif
231#ifdef CONFIG_CFS_BANDWIDTH
232 SEQ_printf(m, " .%-30s: %d\n", "tg->cfs_bandwidth.timer_active",
233 cfs_rq->tg->cfs_bandwidth.timer_active);
234 SEQ_printf(m, " .%-30s: %d\n", "throttled",
235 cfs_rq->throttled);
236 SEQ_printf(m, " .%-30s: %d\n", "throttle_count",
237 cfs_rq->throttle_count);
238#endif
239
240#ifdef CONFIG_FAIR_GROUP_SCHED
241 print_cfs_group_stats(m, cpu, cfs_rq->tg);
242#endif
243}
244
245void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
246{
247#ifdef CONFIG_RT_GROUP_SCHED
248 SEQ_printf(m, "\nrt_rq[%d]:%s\n", cpu, task_group_path(rt_rq->tg));
249#else
250 SEQ_printf(m, "\nrt_rq[%d]:\n", cpu);
251#endif
252
253#define P(x) \
254 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
255#define PN(x) \
256 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
257
258 P(rt_nr_running);
259 P(rt_throttled);
260 PN(rt_time);
261 PN(rt_runtime);
262
263#undef PN
264#undef P
265}
266
267extern __read_mostly int sched_clock_running;
268
269static void print_cpu(struct seq_file *m, int cpu)
270{
271 struct rq *rq = cpu_rq(cpu);
272 unsigned long flags;
273
274#ifdef CONFIG_X86
275 {
276 unsigned int freq = cpu_khz ? : 1;
277
278 SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
279 cpu, freq / 1000, (freq % 1000));
280 }
281#else
282 SEQ_printf(m, "cpu#%d\n", cpu);
283#endif
284
285#define P(x) \
286do { \
287 if (sizeof(rq->x) == 4) \
288 SEQ_printf(m, " .%-30s: %ld\n", #x, (long)(rq->x)); \
289 else \
290 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rq->x));\
291} while (0)
292
293#define PN(x) \
294 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
295
296 P(nr_running);
297 SEQ_printf(m, " .%-30s: %lu\n", "load",
298 rq->load.weight);
299 P(nr_switches);
300 P(nr_load_updates);
301 P(nr_uninterruptible);
302 PN(next_balance);
303 SEQ_printf(m, " .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr)));
304 PN(clock);
305 P(cpu_load[0]);
306 P(cpu_load[1]);
307 P(cpu_load[2]);
308 P(cpu_load[3]);
309 P(cpu_load[4]);
310#undef P
311#undef PN
312
313#ifdef CONFIG_SCHEDSTATS
314#define P(n) SEQ_printf(m, " .%-30s: %d\n", #n, rq->n);
315#define P64(n) SEQ_printf(m, " .%-30s: %Ld\n", #n, rq->n);
316
317 P(yld_count);
318
319 P(sched_count);
320 P(sched_goidle);
321#ifdef CONFIG_SMP
322 P64(avg_idle);
323 P64(max_idle_balance_cost);
324#endif
325
326 P(ttwu_count);
327 P(ttwu_local);
328
329#undef P
330#undef P64
331#endif
332 spin_lock_irqsave(&sched_debug_lock, flags);
333 print_cfs_stats(m, cpu);
334 print_rt_stats(m, cpu);
335
336 rcu_read_lock();
337 print_rq(m, rq, cpu);
338 rcu_read_unlock();
339 spin_unlock_irqrestore(&sched_debug_lock, flags);
340 SEQ_printf(m, "\n");
341}
342
343static const char *sched_tunable_scaling_names[] = {
344 "none",
345 "logaritmic",
346 "linear"
347};
348
349static void sched_debug_header(struct seq_file *m)
350{
351 u64 ktime, sched_clk, cpu_clk;
352 unsigned long flags;
353
354 local_irq_save(flags);
355 ktime = ktime_to_ns(ktime_get());
356 sched_clk = sched_clock();
357 cpu_clk = local_clock();
358 local_irq_restore(flags);
359
360 SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n",
361 init_utsname()->release,
362 (int)strcspn(init_utsname()->version, " "),
363 init_utsname()->version);
364
365#define P(x) \
366 SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
367#define PN(x) \
368 SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
369 PN(ktime);
370 PN(sched_clk);
371 PN(cpu_clk);
372 P(jiffies);
373#ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
374 P(sched_clock_stable());
375#endif
376#undef PN
377#undef P
378
379 SEQ_printf(m, "\n");
380 SEQ_printf(m, "sysctl_sched\n");
381
382#define P(x) \
383 SEQ_printf(m, " .%-40s: %Ld\n", #x, (long long)(x))
384#define PN(x) \
385 SEQ_printf(m, " .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
386 PN(sysctl_sched_latency);
387 PN(sysctl_sched_min_granularity);
388 PN(sysctl_sched_wakeup_granularity);
389 P(sysctl_sched_child_runs_first);
390 P(sysctl_sched_features);
391#undef PN
392#undef P
393
394 SEQ_printf(m, " .%-40s: %d (%s)\n",
395 "sysctl_sched_tunable_scaling",
396 sysctl_sched_tunable_scaling,
397 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
398 SEQ_printf(m, "\n");
399}
400
401static int sched_debug_show(struct seq_file *m, void *v)
402{
403 int cpu = (unsigned long)(v - 2);
404
405 if (cpu != -1)
406 print_cpu(m, cpu);
407 else
408 sched_debug_header(m);
409
410 return 0;
411}
412
413void sysrq_sched_debug_show(void)
414{
415 int cpu;
416
417 sched_debug_header(NULL);
418 for_each_online_cpu(cpu)
419 print_cpu(NULL, cpu);
420
421}
422
423/*
424 * This itererator needs some explanation.
425 * It returns 1 for the header position.
426 * This means 2 is cpu 0.
427 * In a hotplugged system some cpus, including cpu 0, may be missing so we have
428 * to use cpumask_* to iterate over the cpus.
429 */
430static void *sched_debug_start(struct seq_file *file, loff_t *offset)
431{
432 unsigned long n = *offset;
433
434 if (n == 0)
435 return (void *) 1;
436
437 n--;
438
439 if (n > 0)
440 n = cpumask_next(n - 1, cpu_online_mask);
441 else
442 n = cpumask_first(cpu_online_mask);
443
444 *offset = n + 1;
445
446 if (n < nr_cpu_ids)
447 return (void *)(unsigned long)(n + 2);
448 return NULL;
449}
450
451static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
452{
453 (*offset)++;
454 return sched_debug_start(file, offset);
455}
456
457static void sched_debug_stop(struct seq_file *file, void *data)
458{
459}
460
461static const struct seq_operations sched_debug_sops = {
462 .start = sched_debug_start,
463 .next = sched_debug_next,
464 .stop = sched_debug_stop,
465 .show = sched_debug_show,
466};
467
468static int sched_debug_release(struct inode *inode, struct file *file)
469{
470 seq_release(inode, file);
471
472 return 0;
473}
474
475static int sched_debug_open(struct inode *inode, struct file *filp)
476{
477 int ret = 0;
478
479 ret = seq_open(filp, &sched_debug_sops);
480
481 return ret;
482}
483
484static const struct file_operations sched_debug_fops = {
485 .open = sched_debug_open,
486 .read = seq_read,
487 .llseek = seq_lseek,
488 .release = sched_debug_release,
489};
490
491static int __init init_sched_debug_procfs(void)
492{
493 struct proc_dir_entry *pe;
494
495 pe = proc_create("sched_debug", 0444, NULL, &sched_debug_fops);
496 if (!pe)
497 return -ENOMEM;
498 return 0;
499}
500
501__initcall(init_sched_debug_procfs);
502
503#define __P(F) \
504 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)F)
505#define P(F) \
506 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)p->F)
507#define __PN(F) \
508 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
509#define PN(F) \
510 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
511
512
513static void sched_show_numa(struct task_struct *p, struct seq_file *m)
514{
515#ifdef CONFIG_NUMA_BALANCING
516 struct mempolicy *pol;
517 int node, i;
518
519 if (p->mm)
520 P(mm->numa_scan_seq);
521
522 task_lock(p);
523 pol = p->mempolicy;
524 if (pol && !(pol->flags & MPOL_F_MORON))
525 pol = NULL;
526 mpol_get(pol);
527 task_unlock(p);
528
529 SEQ_printf(m, "numa_migrations, %ld\n", xchg(&p->numa_pages_migrated, 0));
530
531 for_each_online_node(node) {
532 for (i = 0; i < 2; i++) {
533 unsigned long nr_faults = -1;
534 int cpu_current, home_node;
535
536 if (p->numa_faults_memory)
537 nr_faults = p->numa_faults_memory[2*node + i];
538
539 cpu_current = !i ? (task_node(p) == node) :
540 (pol && node_isset(node, pol->v.nodes));
541
542 home_node = (p->numa_preferred_nid == node);
543
544 SEQ_printf(m, "numa_faults_memory, %d, %d, %d, %d, %ld\n",
545 i, node, cpu_current, home_node, nr_faults);
546 }
547 }
548
549 mpol_put(pol);
550#endif
551}
552
553void proc_sched_show_task(struct task_struct *p, struct seq_file *m)
554{
555 unsigned long nr_switches;
556
557 SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr(p),
558 get_nr_threads(p));
559 SEQ_printf(m,
560 "---------------------------------------------------------"
561 "----------\n");
562#define __P(F) \
563 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)F)
564#define P(F) \
565 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)p->F)
566#define __PN(F) \
567 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
568#define PN(F) \
569 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
570
571 PN(se.exec_start);
572 PN(se.vruntime);
573 PN(se.sum_exec_runtime);
574
575 nr_switches = p->nvcsw + p->nivcsw;
576
577#ifdef CONFIG_SCHEDSTATS
578 PN(se.statistics.wait_start);
579 PN(se.statistics.sleep_start);
580 PN(se.statistics.block_start);
581 PN(se.statistics.sleep_max);
582 PN(se.statistics.block_max);
583 PN(se.statistics.exec_max);
584 PN(se.statistics.slice_max);
585 PN(se.statistics.wait_max);
586 PN(se.statistics.wait_sum);
587 P(se.statistics.wait_count);
588 PN(se.statistics.iowait_sum);
589 P(se.statistics.iowait_count);
590 P(se.nr_migrations);
591 P(se.statistics.nr_migrations_cold);
592 P(se.statistics.nr_failed_migrations_affine);
593 P(se.statistics.nr_failed_migrations_running);
594 P(se.statistics.nr_failed_migrations_hot);
595 P(se.statistics.nr_forced_migrations);
596 P(se.statistics.nr_wakeups);
597 P(se.statistics.nr_wakeups_sync);
598 P(se.statistics.nr_wakeups_migrate);
599 P(se.statistics.nr_wakeups_local);
600 P(se.statistics.nr_wakeups_remote);
601 P(se.statistics.nr_wakeups_affine);
602 P(se.statistics.nr_wakeups_affine_attempts);
603 P(se.statistics.nr_wakeups_passive);
604 P(se.statistics.nr_wakeups_idle);
605
606 {
607 u64 avg_atom, avg_per_cpu;
608
609 avg_atom = p->se.sum_exec_runtime;
610 if (nr_switches)
611 do_div(avg_atom, nr_switches);
612 else
613 avg_atom = -1LL;
614
615 avg_per_cpu = p->se.sum_exec_runtime;
616 if (p->se.nr_migrations) {
617 avg_per_cpu = div64_u64(avg_per_cpu,
618 p->se.nr_migrations);
619 } else {
620 avg_per_cpu = -1LL;
621 }
622
623 __PN(avg_atom);
624 __PN(avg_per_cpu);
625 }
626#endif
627 __P(nr_switches);
628 SEQ_printf(m, "%-45s:%21Ld\n",
629 "nr_voluntary_switches", (long long)p->nvcsw);
630 SEQ_printf(m, "%-45s:%21Ld\n",
631 "nr_involuntary_switches", (long long)p->nivcsw);
632
633 P(se.load.weight);
634#ifdef CONFIG_SMP
635 P(se.avg.runnable_avg_sum);
636 P(se.avg.runnable_avg_period);
637 P(se.avg.load_avg_contrib);
638 P(se.avg.decay_count);
639#endif
640 P(policy);
641 P(prio);
642#undef PN
643#undef __PN
644#undef P
645#undef __P
646
647 {
648 unsigned int this_cpu = raw_smp_processor_id();
649 u64 t0, t1;
650
651 t0 = cpu_clock(this_cpu);
652 t1 = cpu_clock(this_cpu);
653 SEQ_printf(m, "%-45s:%21Ld\n",
654 "clock-delta", (long long)(t1-t0));
655 }
656
657 sched_show_numa(p, m);
658}
659
660void proc_sched_set_task(struct task_struct *p)
661{
662#ifdef CONFIG_SCHEDSTATS
663 memset(&p->se.statistics, 0, sizeof(p->se.statistics));
664#endif
665}
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * kernel/sched/debug.c
4 *
5 * Print the CFS rbtree and other debugging details
6 *
7 * Copyright(C) 2007, Red Hat, Inc., Ingo Molnar
8 */
9#include "sched.h"
10
11/*
12 * This allows printing both to /proc/sched_debug and
13 * to the console
14 */
15#define SEQ_printf(m, x...) \
16 do { \
17 if (m) \
18 seq_printf(m, x); \
19 else \
20 pr_cont(x); \
21 } while (0)
22
23/*
24 * Ease the printing of nsec fields:
25 */
26static long long nsec_high(unsigned long long nsec)
27{
28 if ((long long)nsec < 0) {
29 nsec = -nsec;
30 do_div(nsec, 1000000);
31 return -nsec;
32 }
33 do_div(nsec, 1000000);
34
35 return nsec;
36}
37
38static unsigned long nsec_low(unsigned long long nsec)
39{
40 if ((long long)nsec < 0)
41 nsec = -nsec;
42
43 return do_div(nsec, 1000000);
44}
45
46#define SPLIT_NS(x) nsec_high(x), nsec_low(x)
47
48#define SCHED_FEAT(name, enabled) \
49 #name ,
50
51static const char * const sched_feat_names[] = {
52#include "features.h"
53};
54
55#undef SCHED_FEAT
56
57static int sched_feat_show(struct seq_file *m, void *v)
58{
59 int i;
60
61 for (i = 0; i < __SCHED_FEAT_NR; i++) {
62 if (!(sysctl_sched_features & (1UL << i)))
63 seq_puts(m, "NO_");
64 seq_printf(m, "%s ", sched_feat_names[i]);
65 }
66 seq_puts(m, "\n");
67
68 return 0;
69}
70
71#ifdef CONFIG_JUMP_LABEL
72
73#define jump_label_key__true STATIC_KEY_INIT_TRUE
74#define jump_label_key__false STATIC_KEY_INIT_FALSE
75
76#define SCHED_FEAT(name, enabled) \
77 jump_label_key__##enabled ,
78
79struct static_key sched_feat_keys[__SCHED_FEAT_NR] = {
80#include "features.h"
81};
82
83#undef SCHED_FEAT
84
85static void sched_feat_disable(int i)
86{
87 static_key_disable_cpuslocked(&sched_feat_keys[i]);
88}
89
90static void sched_feat_enable(int i)
91{
92 static_key_enable_cpuslocked(&sched_feat_keys[i]);
93}
94#else
95static void sched_feat_disable(int i) { };
96static void sched_feat_enable(int i) { };
97#endif /* CONFIG_JUMP_LABEL */
98
99static int sched_feat_set(char *cmp)
100{
101 int i;
102 int neg = 0;
103
104 if (strncmp(cmp, "NO_", 3) == 0) {
105 neg = 1;
106 cmp += 3;
107 }
108
109 i = match_string(sched_feat_names, __SCHED_FEAT_NR, cmp);
110 if (i < 0)
111 return i;
112
113 if (neg) {
114 sysctl_sched_features &= ~(1UL << i);
115 sched_feat_disable(i);
116 } else {
117 sysctl_sched_features |= (1UL << i);
118 sched_feat_enable(i);
119 }
120
121 return 0;
122}
123
124static ssize_t
125sched_feat_write(struct file *filp, const char __user *ubuf,
126 size_t cnt, loff_t *ppos)
127{
128 char buf[64];
129 char *cmp;
130 int ret;
131 struct inode *inode;
132
133 if (cnt > 63)
134 cnt = 63;
135
136 if (copy_from_user(&buf, ubuf, cnt))
137 return -EFAULT;
138
139 buf[cnt] = 0;
140 cmp = strstrip(buf);
141
142 /* Ensure the static_key remains in a consistent state */
143 inode = file_inode(filp);
144 cpus_read_lock();
145 inode_lock(inode);
146 ret = sched_feat_set(cmp);
147 inode_unlock(inode);
148 cpus_read_unlock();
149 if (ret < 0)
150 return ret;
151
152 *ppos += cnt;
153
154 return cnt;
155}
156
157static int sched_feat_open(struct inode *inode, struct file *filp)
158{
159 return single_open(filp, sched_feat_show, NULL);
160}
161
162static const struct file_operations sched_feat_fops = {
163 .open = sched_feat_open,
164 .write = sched_feat_write,
165 .read = seq_read,
166 .llseek = seq_lseek,
167 .release = single_release,
168};
169
170#ifdef CONFIG_SMP
171
172static ssize_t sched_scaling_write(struct file *filp, const char __user *ubuf,
173 size_t cnt, loff_t *ppos)
174{
175 char buf[16];
176 unsigned int scaling;
177
178 if (cnt > 15)
179 cnt = 15;
180
181 if (copy_from_user(&buf, ubuf, cnt))
182 return -EFAULT;
183 buf[cnt] = '\0';
184
185 if (kstrtouint(buf, 10, &scaling))
186 return -EINVAL;
187
188 if (scaling >= SCHED_TUNABLESCALING_END)
189 return -EINVAL;
190
191 sysctl_sched_tunable_scaling = scaling;
192 if (sched_update_scaling())
193 return -EINVAL;
194
195 *ppos += cnt;
196 return cnt;
197}
198
199static int sched_scaling_show(struct seq_file *m, void *v)
200{
201 seq_printf(m, "%d\n", sysctl_sched_tunable_scaling);
202 return 0;
203}
204
205static int sched_scaling_open(struct inode *inode, struct file *filp)
206{
207 return single_open(filp, sched_scaling_show, NULL);
208}
209
210static const struct file_operations sched_scaling_fops = {
211 .open = sched_scaling_open,
212 .write = sched_scaling_write,
213 .read = seq_read,
214 .llseek = seq_lseek,
215 .release = single_release,
216};
217
218#endif /* SMP */
219
220#ifdef CONFIG_PREEMPT_DYNAMIC
221
222static ssize_t sched_dynamic_write(struct file *filp, const char __user *ubuf,
223 size_t cnt, loff_t *ppos)
224{
225 char buf[16];
226 int mode;
227
228 if (cnt > 15)
229 cnt = 15;
230
231 if (copy_from_user(&buf, ubuf, cnt))
232 return -EFAULT;
233
234 buf[cnt] = 0;
235 mode = sched_dynamic_mode(strstrip(buf));
236 if (mode < 0)
237 return mode;
238
239 sched_dynamic_update(mode);
240
241 *ppos += cnt;
242
243 return cnt;
244}
245
246static int sched_dynamic_show(struct seq_file *m, void *v)
247{
248 static const char * preempt_modes[] = {
249 "none", "voluntary", "full"
250 };
251 int i;
252
253 for (i = 0; i < ARRAY_SIZE(preempt_modes); i++) {
254 if (preempt_dynamic_mode == i)
255 seq_puts(m, "(");
256 seq_puts(m, preempt_modes[i]);
257 if (preempt_dynamic_mode == i)
258 seq_puts(m, ")");
259
260 seq_puts(m, " ");
261 }
262
263 seq_puts(m, "\n");
264 return 0;
265}
266
267static int sched_dynamic_open(struct inode *inode, struct file *filp)
268{
269 return single_open(filp, sched_dynamic_show, NULL);
270}
271
272static const struct file_operations sched_dynamic_fops = {
273 .open = sched_dynamic_open,
274 .write = sched_dynamic_write,
275 .read = seq_read,
276 .llseek = seq_lseek,
277 .release = single_release,
278};
279
280#endif /* CONFIG_PREEMPT_DYNAMIC */
281
282__read_mostly bool sched_debug_verbose;
283
284static const struct seq_operations sched_debug_sops;
285
286static int sched_debug_open(struct inode *inode, struct file *filp)
287{
288 return seq_open(filp, &sched_debug_sops);
289}
290
291static const struct file_operations sched_debug_fops = {
292 .open = sched_debug_open,
293 .read = seq_read,
294 .llseek = seq_lseek,
295 .release = seq_release,
296};
297
298static struct dentry *debugfs_sched;
299
300static __init int sched_init_debug(void)
301{
302 struct dentry __maybe_unused *numa;
303
304 debugfs_sched = debugfs_create_dir("sched", NULL);
305
306 debugfs_create_file("features", 0644, debugfs_sched, NULL, &sched_feat_fops);
307 debugfs_create_bool("verbose", 0644, debugfs_sched, &sched_debug_verbose);
308#ifdef CONFIG_PREEMPT_DYNAMIC
309 debugfs_create_file("preempt", 0644, debugfs_sched, NULL, &sched_dynamic_fops);
310#endif
311
312 debugfs_create_u32("latency_ns", 0644, debugfs_sched, &sysctl_sched_latency);
313 debugfs_create_u32("min_granularity_ns", 0644, debugfs_sched, &sysctl_sched_min_granularity);
314 debugfs_create_u32("wakeup_granularity_ns", 0644, debugfs_sched, &sysctl_sched_wakeup_granularity);
315
316 debugfs_create_u32("latency_warn_ms", 0644, debugfs_sched, &sysctl_resched_latency_warn_ms);
317 debugfs_create_u32("latency_warn_once", 0644, debugfs_sched, &sysctl_resched_latency_warn_once);
318
319#ifdef CONFIG_SMP
320 debugfs_create_file("tunable_scaling", 0644, debugfs_sched, NULL, &sched_scaling_fops);
321 debugfs_create_u32("migration_cost_ns", 0644, debugfs_sched, &sysctl_sched_migration_cost);
322 debugfs_create_u32("nr_migrate", 0644, debugfs_sched, &sysctl_sched_nr_migrate);
323
324 mutex_lock(&sched_domains_mutex);
325 update_sched_domain_debugfs();
326 mutex_unlock(&sched_domains_mutex);
327#endif
328
329#ifdef CONFIG_NUMA_BALANCING
330 numa = debugfs_create_dir("numa_balancing", debugfs_sched);
331
332 debugfs_create_u32("scan_delay_ms", 0644, numa, &sysctl_numa_balancing_scan_delay);
333 debugfs_create_u32("scan_period_min_ms", 0644, numa, &sysctl_numa_balancing_scan_period_min);
334 debugfs_create_u32("scan_period_max_ms", 0644, numa, &sysctl_numa_balancing_scan_period_max);
335 debugfs_create_u32("scan_size_mb", 0644, numa, &sysctl_numa_balancing_scan_size);
336#endif
337
338 debugfs_create_file("debug", 0444, debugfs_sched, NULL, &sched_debug_fops);
339
340 return 0;
341}
342late_initcall(sched_init_debug);
343
344#ifdef CONFIG_SMP
345
346static cpumask_var_t sd_sysctl_cpus;
347static struct dentry *sd_dentry;
348
349static int sd_flags_show(struct seq_file *m, void *v)
350{
351 unsigned long flags = *(unsigned int *)m->private;
352 int idx;
353
354 for_each_set_bit(idx, &flags, __SD_FLAG_CNT) {
355 seq_puts(m, sd_flag_debug[idx].name);
356 seq_puts(m, " ");
357 }
358 seq_puts(m, "\n");
359
360 return 0;
361}
362
363static int sd_flags_open(struct inode *inode, struct file *file)
364{
365 return single_open(file, sd_flags_show, inode->i_private);
366}
367
368static const struct file_operations sd_flags_fops = {
369 .open = sd_flags_open,
370 .read = seq_read,
371 .llseek = seq_lseek,
372 .release = single_release,
373};
374
375static void register_sd(struct sched_domain *sd, struct dentry *parent)
376{
377#define SDM(type, mode, member) \
378 debugfs_create_##type(#member, mode, parent, &sd->member)
379
380 SDM(ulong, 0644, min_interval);
381 SDM(ulong, 0644, max_interval);
382 SDM(u64, 0644, max_newidle_lb_cost);
383 SDM(u32, 0644, busy_factor);
384 SDM(u32, 0644, imbalance_pct);
385 SDM(u32, 0644, cache_nice_tries);
386 SDM(str, 0444, name);
387
388#undef SDM
389
390 debugfs_create_file("flags", 0444, parent, &sd->flags, &sd_flags_fops);
391}
392
393void update_sched_domain_debugfs(void)
394{
395 int cpu, i;
396
397 /*
398 * This can unfortunately be invoked before sched_debug_init() creates
399 * the debug directory. Don't touch sd_sysctl_cpus until then.
400 */
401 if (!debugfs_sched)
402 return;
403
404 if (!cpumask_available(sd_sysctl_cpus)) {
405 if (!alloc_cpumask_var(&sd_sysctl_cpus, GFP_KERNEL))
406 return;
407 cpumask_copy(sd_sysctl_cpus, cpu_possible_mask);
408 }
409
410 if (!sd_dentry)
411 sd_dentry = debugfs_create_dir("domains", debugfs_sched);
412
413 for_each_cpu(cpu, sd_sysctl_cpus) {
414 struct sched_domain *sd;
415 struct dentry *d_cpu;
416 char buf[32];
417
418 snprintf(buf, sizeof(buf), "cpu%d", cpu);
419 debugfs_remove(debugfs_lookup(buf, sd_dentry));
420 d_cpu = debugfs_create_dir(buf, sd_dentry);
421
422 i = 0;
423 for_each_domain(cpu, sd) {
424 struct dentry *d_sd;
425
426 snprintf(buf, sizeof(buf), "domain%d", i);
427 d_sd = debugfs_create_dir(buf, d_cpu);
428
429 register_sd(sd, d_sd);
430 i++;
431 }
432
433 __cpumask_clear_cpu(cpu, sd_sysctl_cpus);
434 }
435}
436
437void dirty_sched_domain_sysctl(int cpu)
438{
439 if (cpumask_available(sd_sysctl_cpus))
440 __cpumask_set_cpu(cpu, sd_sysctl_cpus);
441}
442
443#endif /* CONFIG_SMP */
444
445#ifdef CONFIG_FAIR_GROUP_SCHED
446static void print_cfs_group_stats(struct seq_file *m, int cpu, struct task_group *tg)
447{
448 struct sched_entity *se = tg->se[cpu];
449
450#define P(F) SEQ_printf(m, " .%-30s: %lld\n", #F, (long long)F)
451#define P_SCHEDSTAT(F) SEQ_printf(m, " .%-30s: %lld\n", #F, (long long)schedstat_val(F))
452#define PN(F) SEQ_printf(m, " .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F))
453#define PN_SCHEDSTAT(F) SEQ_printf(m, " .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)schedstat_val(F)))
454
455 if (!se)
456 return;
457
458 PN(se->exec_start);
459 PN(se->vruntime);
460 PN(se->sum_exec_runtime);
461
462 if (schedstat_enabled()) {
463 PN_SCHEDSTAT(se->statistics.wait_start);
464 PN_SCHEDSTAT(se->statistics.sleep_start);
465 PN_SCHEDSTAT(se->statistics.block_start);
466 PN_SCHEDSTAT(se->statistics.sleep_max);
467 PN_SCHEDSTAT(se->statistics.block_max);
468 PN_SCHEDSTAT(se->statistics.exec_max);
469 PN_SCHEDSTAT(se->statistics.slice_max);
470 PN_SCHEDSTAT(se->statistics.wait_max);
471 PN_SCHEDSTAT(se->statistics.wait_sum);
472 P_SCHEDSTAT(se->statistics.wait_count);
473 }
474
475 P(se->load.weight);
476#ifdef CONFIG_SMP
477 P(se->avg.load_avg);
478 P(se->avg.util_avg);
479 P(se->avg.runnable_avg);
480#endif
481
482#undef PN_SCHEDSTAT
483#undef PN
484#undef P_SCHEDSTAT
485#undef P
486}
487#endif
488
489#ifdef CONFIG_CGROUP_SCHED
490static DEFINE_SPINLOCK(sched_debug_lock);
491static char group_path[PATH_MAX];
492
493static void task_group_path(struct task_group *tg, char *path, int plen)
494{
495 if (autogroup_path(tg, path, plen))
496 return;
497
498 cgroup_path(tg->css.cgroup, path, plen);
499}
500
501/*
502 * Only 1 SEQ_printf_task_group_path() caller can use the full length
503 * group_path[] for cgroup path. Other simultaneous callers will have
504 * to use a shorter stack buffer. A "..." suffix is appended at the end
505 * of the stack buffer so that it will show up in case the output length
506 * matches the given buffer size to indicate possible path name truncation.
507 */
508#define SEQ_printf_task_group_path(m, tg, fmt...) \
509{ \
510 if (spin_trylock(&sched_debug_lock)) { \
511 task_group_path(tg, group_path, sizeof(group_path)); \
512 SEQ_printf(m, fmt, group_path); \
513 spin_unlock(&sched_debug_lock); \
514 } else { \
515 char buf[128]; \
516 char *bufend = buf + sizeof(buf) - 3; \
517 task_group_path(tg, buf, bufend - buf); \
518 strcpy(bufend - 1, "..."); \
519 SEQ_printf(m, fmt, buf); \
520 } \
521}
522#endif
523
524static void
525print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
526{
527 if (task_current(rq, p))
528 SEQ_printf(m, ">R");
529 else
530 SEQ_printf(m, " %c", task_state_to_char(p));
531
532 SEQ_printf(m, " %15s %5d %9Ld.%06ld %9Ld %5d ",
533 p->comm, task_pid_nr(p),
534 SPLIT_NS(p->se.vruntime),
535 (long long)(p->nvcsw + p->nivcsw),
536 p->prio);
537
538 SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
539 SPLIT_NS(schedstat_val_or_zero(p->se.statistics.wait_sum)),
540 SPLIT_NS(p->se.sum_exec_runtime),
541 SPLIT_NS(schedstat_val_or_zero(p->se.statistics.sum_sleep_runtime)));
542
543#ifdef CONFIG_NUMA_BALANCING
544 SEQ_printf(m, " %d %d", task_node(p), task_numa_group_id(p));
545#endif
546#ifdef CONFIG_CGROUP_SCHED
547 SEQ_printf_task_group_path(m, task_group(p), " %s")
548#endif
549
550 SEQ_printf(m, "\n");
551}
552
553static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
554{
555 struct task_struct *g, *p;
556
557 SEQ_printf(m, "\n");
558 SEQ_printf(m, "runnable tasks:\n");
559 SEQ_printf(m, " S task PID tree-key switches prio"
560 " wait-time sum-exec sum-sleep\n");
561 SEQ_printf(m, "-------------------------------------------------------"
562 "------------------------------------------------------\n");
563
564 rcu_read_lock();
565 for_each_process_thread(g, p) {
566 if (task_cpu(p) != rq_cpu)
567 continue;
568
569 print_task(m, rq, p);
570 }
571 rcu_read_unlock();
572}
573
574void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
575{
576 s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
577 spread, rq0_min_vruntime, spread0;
578 struct rq *rq = cpu_rq(cpu);
579 struct sched_entity *last;
580 unsigned long flags;
581
582#ifdef CONFIG_FAIR_GROUP_SCHED
583 SEQ_printf(m, "\n");
584 SEQ_printf_task_group_path(m, cfs_rq->tg, "cfs_rq[%d]:%s\n", cpu);
585#else
586 SEQ_printf(m, "\n");
587 SEQ_printf(m, "cfs_rq[%d]:\n", cpu);
588#endif
589 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "exec_clock",
590 SPLIT_NS(cfs_rq->exec_clock));
591
592 raw_spin_rq_lock_irqsave(rq, flags);
593 if (rb_first_cached(&cfs_rq->tasks_timeline))
594 MIN_vruntime = (__pick_first_entity(cfs_rq))->vruntime;
595 last = __pick_last_entity(cfs_rq);
596 if (last)
597 max_vruntime = last->vruntime;
598 min_vruntime = cfs_rq->min_vruntime;
599 rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
600 raw_spin_rq_unlock_irqrestore(rq, flags);
601 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "MIN_vruntime",
602 SPLIT_NS(MIN_vruntime));
603 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "min_vruntime",
604 SPLIT_NS(min_vruntime));
605 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "max_vruntime",
606 SPLIT_NS(max_vruntime));
607 spread = max_vruntime - MIN_vruntime;
608 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread",
609 SPLIT_NS(spread));
610 spread0 = min_vruntime - rq0_min_vruntime;
611 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread0",
612 SPLIT_NS(spread0));
613 SEQ_printf(m, " .%-30s: %d\n", "nr_spread_over",
614 cfs_rq->nr_spread_over);
615 SEQ_printf(m, " .%-30s: %d\n", "nr_running", cfs_rq->nr_running);
616 SEQ_printf(m, " .%-30s: %ld\n", "load", cfs_rq->load.weight);
617#ifdef CONFIG_SMP
618 SEQ_printf(m, " .%-30s: %lu\n", "load_avg",
619 cfs_rq->avg.load_avg);
620 SEQ_printf(m, " .%-30s: %lu\n", "runnable_avg",
621 cfs_rq->avg.runnable_avg);
622 SEQ_printf(m, " .%-30s: %lu\n", "util_avg",
623 cfs_rq->avg.util_avg);
624 SEQ_printf(m, " .%-30s: %u\n", "util_est_enqueued",
625 cfs_rq->avg.util_est.enqueued);
626 SEQ_printf(m, " .%-30s: %ld\n", "removed.load_avg",
627 cfs_rq->removed.load_avg);
628 SEQ_printf(m, " .%-30s: %ld\n", "removed.util_avg",
629 cfs_rq->removed.util_avg);
630 SEQ_printf(m, " .%-30s: %ld\n", "removed.runnable_avg",
631 cfs_rq->removed.runnable_avg);
632#ifdef CONFIG_FAIR_GROUP_SCHED
633 SEQ_printf(m, " .%-30s: %lu\n", "tg_load_avg_contrib",
634 cfs_rq->tg_load_avg_contrib);
635 SEQ_printf(m, " .%-30s: %ld\n", "tg_load_avg",
636 atomic_long_read(&cfs_rq->tg->load_avg));
637#endif
638#endif
639#ifdef CONFIG_CFS_BANDWIDTH
640 SEQ_printf(m, " .%-30s: %d\n", "throttled",
641 cfs_rq->throttled);
642 SEQ_printf(m, " .%-30s: %d\n", "throttle_count",
643 cfs_rq->throttle_count);
644#endif
645
646#ifdef CONFIG_FAIR_GROUP_SCHED
647 print_cfs_group_stats(m, cpu, cfs_rq->tg);
648#endif
649}
650
651void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
652{
653#ifdef CONFIG_RT_GROUP_SCHED
654 SEQ_printf(m, "\n");
655 SEQ_printf_task_group_path(m, rt_rq->tg, "rt_rq[%d]:%s\n", cpu);
656#else
657 SEQ_printf(m, "\n");
658 SEQ_printf(m, "rt_rq[%d]:\n", cpu);
659#endif
660
661#define P(x) \
662 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
663#define PU(x) \
664 SEQ_printf(m, " .%-30s: %lu\n", #x, (unsigned long)(rt_rq->x))
665#define PN(x) \
666 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
667
668 PU(rt_nr_running);
669#ifdef CONFIG_SMP
670 PU(rt_nr_migratory);
671#endif
672 P(rt_throttled);
673 PN(rt_time);
674 PN(rt_runtime);
675
676#undef PN
677#undef PU
678#undef P
679}
680
681void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq)
682{
683 struct dl_bw *dl_bw;
684
685 SEQ_printf(m, "\n");
686 SEQ_printf(m, "dl_rq[%d]:\n", cpu);
687
688#define PU(x) \
689 SEQ_printf(m, " .%-30s: %lu\n", #x, (unsigned long)(dl_rq->x))
690
691 PU(dl_nr_running);
692#ifdef CONFIG_SMP
693 PU(dl_nr_migratory);
694 dl_bw = &cpu_rq(cpu)->rd->dl_bw;
695#else
696 dl_bw = &dl_rq->dl_bw;
697#endif
698 SEQ_printf(m, " .%-30s: %lld\n", "dl_bw->bw", dl_bw->bw);
699 SEQ_printf(m, " .%-30s: %lld\n", "dl_bw->total_bw", dl_bw->total_bw);
700
701#undef PU
702}
703
704static void print_cpu(struct seq_file *m, int cpu)
705{
706 struct rq *rq = cpu_rq(cpu);
707
708#ifdef CONFIG_X86
709 {
710 unsigned int freq = cpu_khz ? : 1;
711
712 SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
713 cpu, freq / 1000, (freq % 1000));
714 }
715#else
716 SEQ_printf(m, "cpu#%d\n", cpu);
717#endif
718
719#define P(x) \
720do { \
721 if (sizeof(rq->x) == 4) \
722 SEQ_printf(m, " .%-30s: %ld\n", #x, (long)(rq->x)); \
723 else \
724 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rq->x));\
725} while (0)
726
727#define PN(x) \
728 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
729
730 P(nr_running);
731 P(nr_switches);
732 P(nr_uninterruptible);
733 PN(next_balance);
734 SEQ_printf(m, " .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr)));
735 PN(clock);
736 PN(clock_task);
737#undef P
738#undef PN
739
740#ifdef CONFIG_SMP
741#define P64(n) SEQ_printf(m, " .%-30s: %Ld\n", #n, rq->n);
742 P64(avg_idle);
743 P64(max_idle_balance_cost);
744#undef P64
745#endif
746
747#define P(n) SEQ_printf(m, " .%-30s: %d\n", #n, schedstat_val(rq->n));
748 if (schedstat_enabled()) {
749 P(yld_count);
750 P(sched_count);
751 P(sched_goidle);
752 P(ttwu_count);
753 P(ttwu_local);
754 }
755#undef P
756
757 print_cfs_stats(m, cpu);
758 print_rt_stats(m, cpu);
759 print_dl_stats(m, cpu);
760
761 print_rq(m, rq, cpu);
762 SEQ_printf(m, "\n");
763}
764
765static const char *sched_tunable_scaling_names[] = {
766 "none",
767 "logarithmic",
768 "linear"
769};
770
771static void sched_debug_header(struct seq_file *m)
772{
773 u64 ktime, sched_clk, cpu_clk;
774 unsigned long flags;
775
776 local_irq_save(flags);
777 ktime = ktime_to_ns(ktime_get());
778 sched_clk = sched_clock();
779 cpu_clk = local_clock();
780 local_irq_restore(flags);
781
782 SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n",
783 init_utsname()->release,
784 (int)strcspn(init_utsname()->version, " "),
785 init_utsname()->version);
786
787#define P(x) \
788 SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
789#define PN(x) \
790 SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
791 PN(ktime);
792 PN(sched_clk);
793 PN(cpu_clk);
794 P(jiffies);
795#ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
796 P(sched_clock_stable());
797#endif
798#undef PN
799#undef P
800
801 SEQ_printf(m, "\n");
802 SEQ_printf(m, "sysctl_sched\n");
803
804#define P(x) \
805 SEQ_printf(m, " .%-40s: %Ld\n", #x, (long long)(x))
806#define PN(x) \
807 SEQ_printf(m, " .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
808 PN(sysctl_sched_latency);
809 PN(sysctl_sched_min_granularity);
810 PN(sysctl_sched_wakeup_granularity);
811 P(sysctl_sched_child_runs_first);
812 P(sysctl_sched_features);
813#undef PN
814#undef P
815
816 SEQ_printf(m, " .%-40s: %d (%s)\n",
817 "sysctl_sched_tunable_scaling",
818 sysctl_sched_tunable_scaling,
819 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
820 SEQ_printf(m, "\n");
821}
822
823static int sched_debug_show(struct seq_file *m, void *v)
824{
825 int cpu = (unsigned long)(v - 2);
826
827 if (cpu != -1)
828 print_cpu(m, cpu);
829 else
830 sched_debug_header(m);
831
832 return 0;
833}
834
835void sysrq_sched_debug_show(void)
836{
837 int cpu;
838
839 sched_debug_header(NULL);
840 for_each_online_cpu(cpu) {
841 /*
842 * Need to reset softlockup watchdogs on all CPUs, because
843 * another CPU might be blocked waiting for us to process
844 * an IPI or stop_machine.
845 */
846 touch_nmi_watchdog();
847 touch_all_softlockup_watchdogs();
848 print_cpu(NULL, cpu);
849 }
850}
851
852/*
853 * This iterator needs some explanation.
854 * It returns 1 for the header position.
855 * This means 2 is CPU 0.
856 * In a hotplugged system some CPUs, including CPU 0, may be missing so we have
857 * to use cpumask_* to iterate over the CPUs.
858 */
859static void *sched_debug_start(struct seq_file *file, loff_t *offset)
860{
861 unsigned long n = *offset;
862
863 if (n == 0)
864 return (void *) 1;
865
866 n--;
867
868 if (n > 0)
869 n = cpumask_next(n - 1, cpu_online_mask);
870 else
871 n = cpumask_first(cpu_online_mask);
872
873 *offset = n + 1;
874
875 if (n < nr_cpu_ids)
876 return (void *)(unsigned long)(n + 2);
877
878 return NULL;
879}
880
881static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
882{
883 (*offset)++;
884 return sched_debug_start(file, offset);
885}
886
887static void sched_debug_stop(struct seq_file *file, void *data)
888{
889}
890
891static const struct seq_operations sched_debug_sops = {
892 .start = sched_debug_start,
893 .next = sched_debug_next,
894 .stop = sched_debug_stop,
895 .show = sched_debug_show,
896};
897
898#define __PS(S, F) SEQ_printf(m, "%-45s:%21Ld\n", S, (long long)(F))
899#define __P(F) __PS(#F, F)
900#define P(F) __PS(#F, p->F)
901#define PM(F, M) __PS(#F, p->F & (M))
902#define __PSN(S, F) SEQ_printf(m, "%-45s:%14Ld.%06ld\n", S, SPLIT_NS((long long)(F)))
903#define __PN(F) __PSN(#F, F)
904#define PN(F) __PSN(#F, p->F)
905
906
907#ifdef CONFIG_NUMA_BALANCING
908void print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
909 unsigned long tpf, unsigned long gsf, unsigned long gpf)
910{
911 SEQ_printf(m, "numa_faults node=%d ", node);
912 SEQ_printf(m, "task_private=%lu task_shared=%lu ", tpf, tsf);
913 SEQ_printf(m, "group_private=%lu group_shared=%lu\n", gpf, gsf);
914}
915#endif
916
917
918static void sched_show_numa(struct task_struct *p, struct seq_file *m)
919{
920#ifdef CONFIG_NUMA_BALANCING
921 struct mempolicy *pol;
922
923 if (p->mm)
924 P(mm->numa_scan_seq);
925
926 task_lock(p);
927 pol = p->mempolicy;
928 if (pol && !(pol->flags & MPOL_F_MORON))
929 pol = NULL;
930 mpol_get(pol);
931 task_unlock(p);
932
933 P(numa_pages_migrated);
934 P(numa_preferred_nid);
935 P(total_numa_faults);
936 SEQ_printf(m, "current_node=%d, numa_group_id=%d\n",
937 task_node(p), task_numa_group_id(p));
938 show_numa_stats(p, m);
939 mpol_put(pol);
940#endif
941}
942
943void proc_sched_show_task(struct task_struct *p, struct pid_namespace *ns,
944 struct seq_file *m)
945{
946 unsigned long nr_switches;
947
948 SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr_ns(p, ns),
949 get_nr_threads(p));
950 SEQ_printf(m,
951 "---------------------------------------------------------"
952 "----------\n");
953
954#define P_SCHEDSTAT(F) __PS(#F, schedstat_val(p->F))
955#define PN_SCHEDSTAT(F) __PSN(#F, schedstat_val(p->F))
956
957 PN(se.exec_start);
958 PN(se.vruntime);
959 PN(se.sum_exec_runtime);
960
961 nr_switches = p->nvcsw + p->nivcsw;
962
963 P(se.nr_migrations);
964
965 if (schedstat_enabled()) {
966 u64 avg_atom, avg_per_cpu;
967
968 PN_SCHEDSTAT(se.statistics.sum_sleep_runtime);
969 PN_SCHEDSTAT(se.statistics.wait_start);
970 PN_SCHEDSTAT(se.statistics.sleep_start);
971 PN_SCHEDSTAT(se.statistics.block_start);
972 PN_SCHEDSTAT(se.statistics.sleep_max);
973 PN_SCHEDSTAT(se.statistics.block_max);
974 PN_SCHEDSTAT(se.statistics.exec_max);
975 PN_SCHEDSTAT(se.statistics.slice_max);
976 PN_SCHEDSTAT(se.statistics.wait_max);
977 PN_SCHEDSTAT(se.statistics.wait_sum);
978 P_SCHEDSTAT(se.statistics.wait_count);
979 PN_SCHEDSTAT(se.statistics.iowait_sum);
980 P_SCHEDSTAT(se.statistics.iowait_count);
981 P_SCHEDSTAT(se.statistics.nr_migrations_cold);
982 P_SCHEDSTAT(se.statistics.nr_failed_migrations_affine);
983 P_SCHEDSTAT(se.statistics.nr_failed_migrations_running);
984 P_SCHEDSTAT(se.statistics.nr_failed_migrations_hot);
985 P_SCHEDSTAT(se.statistics.nr_forced_migrations);
986 P_SCHEDSTAT(se.statistics.nr_wakeups);
987 P_SCHEDSTAT(se.statistics.nr_wakeups_sync);
988 P_SCHEDSTAT(se.statistics.nr_wakeups_migrate);
989 P_SCHEDSTAT(se.statistics.nr_wakeups_local);
990 P_SCHEDSTAT(se.statistics.nr_wakeups_remote);
991 P_SCHEDSTAT(se.statistics.nr_wakeups_affine);
992 P_SCHEDSTAT(se.statistics.nr_wakeups_affine_attempts);
993 P_SCHEDSTAT(se.statistics.nr_wakeups_passive);
994 P_SCHEDSTAT(se.statistics.nr_wakeups_idle);
995
996 avg_atom = p->se.sum_exec_runtime;
997 if (nr_switches)
998 avg_atom = div64_ul(avg_atom, nr_switches);
999 else
1000 avg_atom = -1LL;
1001
1002 avg_per_cpu = p->se.sum_exec_runtime;
1003 if (p->se.nr_migrations) {
1004 avg_per_cpu = div64_u64(avg_per_cpu,
1005 p->se.nr_migrations);
1006 } else {
1007 avg_per_cpu = -1LL;
1008 }
1009
1010 __PN(avg_atom);
1011 __PN(avg_per_cpu);
1012 }
1013
1014 __P(nr_switches);
1015 __PS("nr_voluntary_switches", p->nvcsw);
1016 __PS("nr_involuntary_switches", p->nivcsw);
1017
1018 P(se.load.weight);
1019#ifdef CONFIG_SMP
1020 P(se.avg.load_sum);
1021 P(se.avg.runnable_sum);
1022 P(se.avg.util_sum);
1023 P(se.avg.load_avg);
1024 P(se.avg.runnable_avg);
1025 P(se.avg.util_avg);
1026 P(se.avg.last_update_time);
1027 P(se.avg.util_est.ewma);
1028 PM(se.avg.util_est.enqueued, ~UTIL_AVG_UNCHANGED);
1029#endif
1030#ifdef CONFIG_UCLAMP_TASK
1031 __PS("uclamp.min", p->uclamp_req[UCLAMP_MIN].value);
1032 __PS("uclamp.max", p->uclamp_req[UCLAMP_MAX].value);
1033 __PS("effective uclamp.min", uclamp_eff_value(p, UCLAMP_MIN));
1034 __PS("effective uclamp.max", uclamp_eff_value(p, UCLAMP_MAX));
1035#endif
1036 P(policy);
1037 P(prio);
1038 if (task_has_dl_policy(p)) {
1039 P(dl.runtime);
1040 P(dl.deadline);
1041 }
1042#undef PN_SCHEDSTAT
1043#undef P_SCHEDSTAT
1044
1045 {
1046 unsigned int this_cpu = raw_smp_processor_id();
1047 u64 t0, t1;
1048
1049 t0 = cpu_clock(this_cpu);
1050 t1 = cpu_clock(this_cpu);
1051 __PS("clock-delta", t1-t0);
1052 }
1053
1054 sched_show_numa(p, m);
1055}
1056
1057void proc_sched_set_task(struct task_struct *p)
1058{
1059#ifdef CONFIG_SCHEDSTATS
1060 memset(&p->se.statistics, 0, sizeof(p->se.statistics));
1061#endif
1062}
1063
1064void resched_latency_warn(int cpu, u64 latency)
1065{
1066 static DEFINE_RATELIMIT_STATE(latency_check_ratelimit, 60 * 60 * HZ, 1);
1067
1068 WARN(__ratelimit(&latency_check_ratelimit),
1069 "sched: CPU %d need_resched set for > %llu ns (%d ticks) "
1070 "without schedule\n",
1071 cpu, latency, cpu_rq(cpu)->ticks_without_resched);
1072}