projects/linux·blob v6.5 · refs

1126 lines · 26.6 KiB

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
10/*
11 * This allows printing both to /proc/sched_debug and
12 * to the console
13 */
14#define SEQ_printf(m, x...) \
15 do { \
16 if (m) \
17 seq_printf(m, x); \
18 else \
19 pr_cont(x); \
20 } while (0)
21
22/*
23 * Ease the printing of nsec fields:
24 */
25static long long nsec_high(unsigned long long nsec)
26{
27 if ((long long)nsec < 0) {
28 nsec = -nsec;
29 do_div(nsec, 1000000);
30 return -nsec;
31 }
32 do_div(nsec, 1000000);
33
34 return nsec;
35}
36
37static unsigned long nsec_low(unsigned long long nsec)
38{
39 if ((long long)nsec < 0)
40 nsec = -nsec;
41
42 return do_div(nsec, 1000000);
43}
44
45#define SPLIT_NS(x) nsec_high(x), nsec_low(x)
46
47#define SCHED_FEAT(name, enabled) \
48 #name ,
49
50static const char * const sched_feat_names[] = {
51#include "features.h"
52};
53
54#undef SCHED_FEAT
55
56static int sched_feat_show(struct seq_file *m, void *v)
57{
58 int i;
59
60 for (i = 0; i < __SCHED_FEAT_NR; i++) {
61 if (!(sysctl_sched_features & (1UL << i)))
62 seq_puts(m, "NO_");
63 seq_printf(m, "%s ", sched_feat_names[i]);
64 }
65 seq_puts(m, "\n");
66
67 return 0;
68}
69
70#ifdef CONFIG_JUMP_LABEL
71
72#define jump_label_key__true STATIC_KEY_INIT_TRUE
73#define jump_label_key__false STATIC_KEY_INIT_FALSE
74
75#define SCHED_FEAT(name, enabled) \
76 jump_label_key__##enabled ,
77
78struct static_key sched_feat_keys[__SCHED_FEAT_NR] = {
79#include "features.h"
80};
81
82#undef SCHED_FEAT
83
84static void sched_feat_disable(int i)
85{
86 static_key_disable_cpuslocked(&sched_feat_keys[i]);
87}
88
89static void sched_feat_enable(int i)
90{
91 static_key_enable_cpuslocked(&sched_feat_keys[i]);
92}
93#else
94static void sched_feat_disable(int i) { };
95static void sched_feat_enable(int i) { };
96#endif /* CONFIG_JUMP_LABEL */
97
98static int sched_feat_set(char *cmp)
99{
100 int i;
101 int neg = 0;
102
103 if (strncmp(cmp, "NO_", 3) == 0) {
104 neg = 1;
105 cmp += 3;
106 }
107
108 i = match_string(sched_feat_names, __SCHED_FEAT_NR, cmp);
109 if (i < 0)
110 return i;
111
112 if (neg) {
113 sysctl_sched_features &= ~(1UL << i);
114 sched_feat_disable(i);
115 } else {
116 sysctl_sched_features |= (1UL << i);
117 sched_feat_enable(i);
118 }
119
120 return 0;
121}
122
123static ssize_t
124sched_feat_write(struct file *filp, const char __user *ubuf,
125 size_t cnt, loff_t *ppos)
126{
127 char buf[64];
128 char *cmp;
129 int ret;
130 struct inode *inode;
131
132 if (cnt > 63)
133 cnt = 63;
134
135 if (copy_from_user(&buf, ubuf, cnt))
136 return -EFAULT;
137
138 buf[cnt] = 0;
139 cmp = strstrip(buf);
140
141 /* Ensure the static_key remains in a consistent state */
142 inode = file_inode(filp);
143 cpus_read_lock();
144 inode_lock(inode);
145 ret = sched_feat_set(cmp);
146 inode_unlock(inode);
147 cpus_read_unlock();
148 if (ret < 0)
149 return ret;
150
151 *ppos += cnt;
152
153 return cnt;
154}
155
156static int sched_feat_open(struct inode *inode, struct file *filp)
157{
158 return single_open(filp, sched_feat_show, NULL);
159}
160
161static const struct file_operations sched_feat_fops = {
162 .open = sched_feat_open,
163 .write = sched_feat_write,
164 .read = seq_read,
165 .llseek = seq_lseek,
166 .release = single_release,
167};
168
169#ifdef CONFIG_SMP
170
171static ssize_t sched_scaling_write(struct file *filp, const char __user *ubuf,
172 size_t cnt, loff_t *ppos)
173{
174 char buf[16];
175 unsigned int scaling;
176
177 if (cnt > 15)
178 cnt = 15;
179
180 if (copy_from_user(&buf, ubuf, cnt))
181 return -EFAULT;
182 buf[cnt] = '\0';
183
184 if (kstrtouint(buf, 10, &scaling))
185 return -EINVAL;
186
187 if (scaling >= SCHED_TUNABLESCALING_END)
188 return -EINVAL;
189
190 sysctl_sched_tunable_scaling = scaling;
191 if (sched_update_scaling())
192 return -EINVAL;
193
194 *ppos += cnt;
195 return cnt;
196}
197
198static int sched_scaling_show(struct seq_file *m, void *v)
199{
200 seq_printf(m, "%d\n", sysctl_sched_tunable_scaling);
201 return 0;
202}
203
204static int sched_scaling_open(struct inode *inode, struct file *filp)
205{
206 return single_open(filp, sched_scaling_show, NULL);
207}
208
209static const struct file_operations sched_scaling_fops = {
210 .open = sched_scaling_open,
211 .write = sched_scaling_write,
212 .read = seq_read,
213 .llseek = seq_lseek,
214 .release = single_release,
215};
216
217#endif /* SMP */
218
219#ifdef CONFIG_PREEMPT_DYNAMIC
220
221static ssize_t sched_dynamic_write(struct file *filp, const char __user *ubuf,
222 size_t cnt, loff_t *ppos)
223{
224 char buf[16];
225 int mode;
226
227 if (cnt > 15)
228 cnt = 15;
229
230 if (copy_from_user(&buf, ubuf, cnt))
231 return -EFAULT;
232
233 buf[cnt] = 0;
234 mode = sched_dynamic_mode(strstrip(buf));
235 if (mode < 0)
236 return mode;
237
238 sched_dynamic_update(mode);
239
240 *ppos += cnt;
241
242 return cnt;
243}
244
245static int sched_dynamic_show(struct seq_file *m, void *v)
246{
247 static const char * preempt_modes[] = {
248 "none", "voluntary", "full"
249 };
250 int i;
251
252 for (i = 0; i < ARRAY_SIZE(preempt_modes); i++) {
253 if (preempt_dynamic_mode == i)
254 seq_puts(m, "(");
255 seq_puts(m, preempt_modes[i]);
256 if (preempt_dynamic_mode == i)
257 seq_puts(m, ")");
258
259 seq_puts(m, " ");
260 }
261
262 seq_puts(m, "\n");
263 return 0;
264}
265
266static int sched_dynamic_open(struct inode *inode, struct file *filp)
267{
268 return single_open(filp, sched_dynamic_show, NULL);
269}
270
271static const struct file_operations sched_dynamic_fops = {
272 .open = sched_dynamic_open,
273 .write = sched_dynamic_write,
274 .read = seq_read,
275 .llseek = seq_lseek,
276 .release = single_release,
277};
278
279#endif /* CONFIG_PREEMPT_DYNAMIC */
280
281__read_mostly bool sched_debug_verbose;
282
283#ifdef CONFIG_SMP
284static struct dentry *sd_dentry;
285
286
287static ssize_t sched_verbose_write(struct file *filp, const char __user *ubuf,
288 size_t cnt, loff_t *ppos)
289{
290 ssize_t result;
291 bool orig;
292
293 cpus_read_lock();
294 mutex_lock(&sched_domains_mutex);
295
296 orig = sched_debug_verbose;
297 result = debugfs_write_file_bool(filp, ubuf, cnt, ppos);
298
299 if (sched_debug_verbose && !orig)
300 update_sched_domain_debugfs();
301 else if (!sched_debug_verbose && orig) {
302 debugfs_remove(sd_dentry);
303 sd_dentry = NULL;
304 }
305
306 mutex_unlock(&sched_domains_mutex);
307 cpus_read_unlock();
308
309 return result;
310}
311#else
312#define sched_verbose_write debugfs_write_file_bool
313#endif
314
315static const struct file_operations sched_verbose_fops = {
316 .read = debugfs_read_file_bool,
317 .write = sched_verbose_write,
318 .open = simple_open,
319 .llseek = default_llseek,
320};
321
322static const struct seq_operations sched_debug_sops;
323
324static int sched_debug_open(struct inode *inode, struct file *filp)
325{
326 return seq_open(filp, &sched_debug_sops);
327}
328
329static const struct file_operations sched_debug_fops = {
330 .open = sched_debug_open,
331 .read = seq_read,
332 .llseek = seq_lseek,
333 .release = seq_release,
334};
335
336static struct dentry *debugfs_sched;
337
338static __init int sched_init_debug(void)
339{
340 struct dentry __maybe_unused *numa;
341
342 debugfs_sched = debugfs_create_dir("sched", NULL);
343
344 debugfs_create_file("features", 0644, debugfs_sched, NULL, &sched_feat_fops);
345 debugfs_create_file_unsafe("verbose", 0644, debugfs_sched, &sched_debug_verbose, &sched_verbose_fops);
346#ifdef CONFIG_PREEMPT_DYNAMIC
347 debugfs_create_file("preempt", 0644, debugfs_sched, NULL, &sched_dynamic_fops);
348#endif
349
350 debugfs_create_u32("latency_ns", 0644, debugfs_sched, &sysctl_sched_latency);
351 debugfs_create_u32("min_granularity_ns", 0644, debugfs_sched, &sysctl_sched_min_granularity);
352 debugfs_create_u32("idle_min_granularity_ns", 0644, debugfs_sched, &sysctl_sched_idle_min_granularity);
353 debugfs_create_u32("wakeup_granularity_ns", 0644, debugfs_sched, &sysctl_sched_wakeup_granularity);
354
355 debugfs_create_u32("latency_warn_ms", 0644, debugfs_sched, &sysctl_resched_latency_warn_ms);
356 debugfs_create_u32("latency_warn_once", 0644, debugfs_sched, &sysctl_resched_latency_warn_once);
357
358#ifdef CONFIG_SMP
359 debugfs_create_file("tunable_scaling", 0644, debugfs_sched, NULL, &sched_scaling_fops);
360 debugfs_create_u32("migration_cost_ns", 0644, debugfs_sched, &sysctl_sched_migration_cost);
361 debugfs_create_u32("nr_migrate", 0644, debugfs_sched, &sysctl_sched_nr_migrate);
362
363 mutex_lock(&sched_domains_mutex);
364 update_sched_domain_debugfs();
365 mutex_unlock(&sched_domains_mutex);
366#endif
367
368#ifdef CONFIG_NUMA_BALANCING
369 numa = debugfs_create_dir("numa_balancing", debugfs_sched);
370
371 debugfs_create_u32("scan_delay_ms", 0644, numa, &sysctl_numa_balancing_scan_delay);
372 debugfs_create_u32("scan_period_min_ms", 0644, numa, &sysctl_numa_balancing_scan_period_min);
373 debugfs_create_u32("scan_period_max_ms", 0644, numa, &sysctl_numa_balancing_scan_period_max);
374 debugfs_create_u32("scan_size_mb", 0644, numa, &sysctl_numa_balancing_scan_size);
375 debugfs_create_u32("hot_threshold_ms", 0644, numa, &sysctl_numa_balancing_hot_threshold);
376#endif
377
378 debugfs_create_file("debug", 0444, debugfs_sched, NULL, &sched_debug_fops);
379
380 return 0;
381}
382late_initcall(sched_init_debug);
383
384#ifdef CONFIG_SMP
385
386static cpumask_var_t sd_sysctl_cpus;
387
388static int sd_flags_show(struct seq_file *m, void *v)
389{
390 unsigned long flags = *(unsigned int *)m->private;
391 int idx;
392
393 for_each_set_bit(idx, &flags, __SD_FLAG_CNT) {
394 seq_puts(m, sd_flag_debug[idx].name);
395 seq_puts(m, " ");
396 }
397 seq_puts(m, "\n");
398
399 return 0;
400}
401
402static int sd_flags_open(struct inode *inode, struct file *file)
403{
404 return single_open(file, sd_flags_show, inode->i_private);
405}
406
407static const struct file_operations sd_flags_fops = {
408 .open = sd_flags_open,
409 .read = seq_read,
410 .llseek = seq_lseek,
411 .release = single_release,
412};
413
414static void register_sd(struct sched_domain *sd, struct dentry *parent)
415{
416#define SDM(type, mode, member) \
417 debugfs_create_##type(#member, mode, parent, &sd->member)
418
419 SDM(ulong, 0644, min_interval);
420 SDM(ulong, 0644, max_interval);
421 SDM(u64, 0644, max_newidle_lb_cost);
422 SDM(u32, 0644, busy_factor);
423 SDM(u32, 0644, imbalance_pct);
424 SDM(u32, 0644, cache_nice_tries);
425 SDM(str, 0444, name);
426
427#undef SDM
428
429 debugfs_create_file("flags", 0444, parent, &sd->flags, &sd_flags_fops);
430}
431
432void update_sched_domain_debugfs(void)
433{
434 int cpu, i;
435
436 /*
437 * This can unfortunately be invoked before sched_debug_init() creates
438 * the debug directory. Don't touch sd_sysctl_cpus until then.
439 */
440 if (!debugfs_sched)
441 return;
442
443 if (!sched_debug_verbose)
444 return;
445
446 if (!cpumask_available(sd_sysctl_cpus)) {
447 if (!alloc_cpumask_var(&sd_sysctl_cpus, GFP_KERNEL))
448 return;
449 cpumask_copy(sd_sysctl_cpus, cpu_possible_mask);
450 }
451
452 if (!sd_dentry) {
453 sd_dentry = debugfs_create_dir("domains", debugfs_sched);
454
455 /* rebuild sd_sysctl_cpus if empty since it gets cleared below */
456 if (cpumask_empty(sd_sysctl_cpus))
457 cpumask_copy(sd_sysctl_cpus, cpu_online_mask);
458 }
459
460 for_each_cpu(cpu, sd_sysctl_cpus) {
461 struct sched_domain *sd;
462 struct dentry *d_cpu;
463 char buf[32];
464
465 snprintf(buf, sizeof(buf), "cpu%d", cpu);
466 debugfs_lookup_and_remove(buf, sd_dentry);
467 d_cpu = debugfs_create_dir(buf, sd_dentry);
468
469 i = 0;
470 for_each_domain(cpu, sd) {
471 struct dentry *d_sd;
472
473 snprintf(buf, sizeof(buf), "domain%d", i);
474 d_sd = debugfs_create_dir(buf, d_cpu);
475
476 register_sd(sd, d_sd);
477 i++;
478 }
479
480 __cpumask_clear_cpu(cpu, sd_sysctl_cpus);
481 }
482}
483
484void dirty_sched_domain_sysctl(int cpu)
485{
486 if (cpumask_available(sd_sysctl_cpus))
487 __cpumask_set_cpu(cpu, sd_sysctl_cpus);
488}
489
490#endif /* CONFIG_SMP */
491
492#ifdef CONFIG_FAIR_GROUP_SCHED
493static void print_cfs_group_stats(struct seq_file *m, int cpu, struct task_group *tg)
494{
495 struct sched_entity *se = tg->se[cpu];
496
497#define P(F) SEQ_printf(m, " .%-30s: %lld\n", #F, (long long)F)
498#define P_SCHEDSTAT(F) SEQ_printf(m, " .%-30s: %lld\n", \
499 #F, (long long)schedstat_val(stats->F))
500#define PN(F) SEQ_printf(m, " .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F))
501#define PN_SCHEDSTAT(F) SEQ_printf(m, " .%-30s: %lld.%06ld\n", \
502 #F, SPLIT_NS((long long)schedstat_val(stats->F)))
503
504 if (!se)
505 return;
506
507 PN(se->exec_start);
508 PN(se->vruntime);
509 PN(se->sum_exec_runtime);
510
511 if (schedstat_enabled()) {
512 struct sched_statistics *stats;
513 stats = __schedstats_from_se(se);
514
515 PN_SCHEDSTAT(wait_start);
516 PN_SCHEDSTAT(sleep_start);
517 PN_SCHEDSTAT(block_start);
518 PN_SCHEDSTAT(sleep_max);
519 PN_SCHEDSTAT(block_max);
520 PN_SCHEDSTAT(exec_max);
521 PN_SCHEDSTAT(slice_max);
522 PN_SCHEDSTAT(wait_max);
523 PN_SCHEDSTAT(wait_sum);
524 P_SCHEDSTAT(wait_count);
525 }
526
527 P(se->load.weight);
528#ifdef CONFIG_SMP
529 P(se->avg.load_avg);
530 P(se->avg.util_avg);
531 P(se->avg.runnable_avg);
532#endif
533
534#undef PN_SCHEDSTAT
535#undef PN
536#undef P_SCHEDSTAT
537#undef P
538}
539#endif
540
541#ifdef CONFIG_CGROUP_SCHED
542static DEFINE_SPINLOCK(sched_debug_lock);
543static char group_path[PATH_MAX];
544
545static void task_group_path(struct task_group *tg, char *path, int plen)
546{
547 if (autogroup_path(tg, path, plen))
548 return;
549
550 cgroup_path(tg->css.cgroup, path, plen);
551}
552
553/*
554 * Only 1 SEQ_printf_task_group_path() caller can use the full length
555 * group_path[] for cgroup path. Other simultaneous callers will have
556 * to use a shorter stack buffer. A "..." suffix is appended at the end
557 * of the stack buffer so that it will show up in case the output length
558 * matches the given buffer size to indicate possible path name truncation.
559 */
560#define SEQ_printf_task_group_path(m, tg, fmt...) \
561{ \
562 if (spin_trylock(&sched_debug_lock)) { \
563 task_group_path(tg, group_path, sizeof(group_path)); \
564 SEQ_printf(m, fmt, group_path); \
565 spin_unlock(&sched_debug_lock); \
566 } else { \
567 char buf[128]; \
568 char *bufend = buf + sizeof(buf) - 3; \
569 task_group_path(tg, buf, bufend - buf); \
570 strcpy(bufend - 1, "..."); \
571 SEQ_printf(m, fmt, buf); \
572 } \
573}
574#endif
575
576static void
577print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
578{
579 if (task_current(rq, p))
580 SEQ_printf(m, ">R");
581 else
582 SEQ_printf(m, " %c", task_state_to_char(p));
583
584 SEQ_printf(m, " %15s %5d %9Ld.%06ld %9Ld %5d ",
585 p->comm, task_pid_nr(p),
586 SPLIT_NS(p->se.vruntime),
587 (long long)(p->nvcsw + p->nivcsw),
588 p->prio);
589
590 SEQ_printf(m, "%9lld.%06ld %9lld.%06ld %9lld.%06ld %9lld.%06ld",
591 SPLIT_NS(schedstat_val_or_zero(p->stats.wait_sum)),
592 SPLIT_NS(p->se.sum_exec_runtime),
593 SPLIT_NS(schedstat_val_or_zero(p->stats.sum_sleep_runtime)),
594 SPLIT_NS(schedstat_val_or_zero(p->stats.sum_block_runtime)));
595
596#ifdef CONFIG_NUMA_BALANCING
597 SEQ_printf(m, " %d %d", task_node(p), task_numa_group_id(p));
598#endif
599#ifdef CONFIG_CGROUP_SCHED
600 SEQ_printf_task_group_path(m, task_group(p), " %s")
601#endif
602
603 SEQ_printf(m, "\n");
604}
605
606static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
607{
608 struct task_struct *g, *p;
609
610 SEQ_printf(m, "\n");
611 SEQ_printf(m, "runnable tasks:\n");
612 SEQ_printf(m, " S task PID tree-key switches prio"
613 " wait-time sum-exec sum-sleep\n");
614 SEQ_printf(m, "-------------------------------------------------------"
615 "------------------------------------------------------\n");
616
617 rcu_read_lock();
618 for_each_process_thread(g, p) {
619 if (task_cpu(p) != rq_cpu)
620 continue;
621
622 print_task(m, rq, p);
623 }
624 rcu_read_unlock();
625}
626
627void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
628{
629 s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
630 spread, rq0_min_vruntime, spread0;
631 struct rq *rq = cpu_rq(cpu);
632 struct sched_entity *last;
633 unsigned long flags;
634
635#ifdef CONFIG_FAIR_GROUP_SCHED
636 SEQ_printf(m, "\n");
637 SEQ_printf_task_group_path(m, cfs_rq->tg, "cfs_rq[%d]:%s\n", cpu);
638#else
639 SEQ_printf(m, "\n");
640 SEQ_printf(m, "cfs_rq[%d]:\n", cpu);
641#endif
642 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "exec_clock",
643 SPLIT_NS(cfs_rq->exec_clock));
644
645 raw_spin_rq_lock_irqsave(rq, flags);
646 if (rb_first_cached(&cfs_rq->tasks_timeline))
647 MIN_vruntime = (__pick_first_entity(cfs_rq))->vruntime;
648 last = __pick_last_entity(cfs_rq);
649 if (last)
650 max_vruntime = last->vruntime;
651 min_vruntime = cfs_rq->min_vruntime;
652 rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
653 raw_spin_rq_unlock_irqrestore(rq, flags);
654 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "MIN_vruntime",
655 SPLIT_NS(MIN_vruntime));
656 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "min_vruntime",
657 SPLIT_NS(min_vruntime));
658 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "max_vruntime",
659 SPLIT_NS(max_vruntime));
660 spread = max_vruntime - MIN_vruntime;
661 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread",
662 SPLIT_NS(spread));
663 spread0 = min_vruntime - rq0_min_vruntime;
664 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread0",
665 SPLIT_NS(spread0));
666 SEQ_printf(m, " .%-30s: %d\n", "nr_spread_over",
667 cfs_rq->nr_spread_over);
668 SEQ_printf(m, " .%-30s: %d\n", "nr_running", cfs_rq->nr_running);
669 SEQ_printf(m, " .%-30s: %d\n", "h_nr_running", cfs_rq->h_nr_running);
670 SEQ_printf(m, " .%-30s: %d\n", "idle_nr_running",
671 cfs_rq->idle_nr_running);
672 SEQ_printf(m, " .%-30s: %d\n", "idle_h_nr_running",
673 cfs_rq->idle_h_nr_running);
674 SEQ_printf(m, " .%-30s: %ld\n", "load", cfs_rq->load.weight);
675#ifdef CONFIG_SMP
676 SEQ_printf(m, " .%-30s: %lu\n", "load_avg",
677 cfs_rq->avg.load_avg);
678 SEQ_printf(m, " .%-30s: %lu\n", "runnable_avg",
679 cfs_rq->avg.runnable_avg);
680 SEQ_printf(m, " .%-30s: %lu\n", "util_avg",
681 cfs_rq->avg.util_avg);
682 SEQ_printf(m, " .%-30s: %u\n", "util_est_enqueued",
683 cfs_rq->avg.util_est.enqueued);
684 SEQ_printf(m, " .%-30s: %ld\n", "removed.load_avg",
685 cfs_rq->removed.load_avg);
686 SEQ_printf(m, " .%-30s: %ld\n", "removed.util_avg",
687 cfs_rq->removed.util_avg);
688 SEQ_printf(m, " .%-30s: %ld\n", "removed.runnable_avg",
689 cfs_rq->removed.runnable_avg);
690#ifdef CONFIG_FAIR_GROUP_SCHED
691 SEQ_printf(m, " .%-30s: %lu\n", "tg_load_avg_contrib",
692 cfs_rq->tg_load_avg_contrib);
693 SEQ_printf(m, " .%-30s: %ld\n", "tg_load_avg",
694 atomic_long_read(&cfs_rq->tg->load_avg));
695#endif
696#endif
697#ifdef CONFIG_CFS_BANDWIDTH
698 SEQ_printf(m, " .%-30s: %d\n", "throttled",
699 cfs_rq->throttled);
700 SEQ_printf(m, " .%-30s: %d\n", "throttle_count",
701 cfs_rq->throttle_count);
702#endif
703
704#ifdef CONFIG_FAIR_GROUP_SCHED
705 print_cfs_group_stats(m, cpu, cfs_rq->tg);
706#endif
707}
708
709void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
710{
711#ifdef CONFIG_RT_GROUP_SCHED
712 SEQ_printf(m, "\n");
713 SEQ_printf_task_group_path(m, rt_rq->tg, "rt_rq[%d]:%s\n", cpu);
714#else
715 SEQ_printf(m, "\n");
716 SEQ_printf(m, "rt_rq[%d]:\n", cpu);
717#endif
718
719#define P(x) \
720 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
721#define PU(x) \
722 SEQ_printf(m, " .%-30s: %lu\n", #x, (unsigned long)(rt_rq->x))
723#define PN(x) \
724 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
725
726 PU(rt_nr_running);
727#ifdef CONFIG_SMP
728 PU(rt_nr_migratory);
729#endif
730 P(rt_throttled);
731 PN(rt_time);
732 PN(rt_runtime);
733
734#undef PN
735#undef PU
736#undef P
737}
738
739void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq)
740{
741 struct dl_bw *dl_bw;
742
743 SEQ_printf(m, "\n");
744 SEQ_printf(m, "dl_rq[%d]:\n", cpu);
745
746#define PU(x) \
747 SEQ_printf(m, " .%-30s: %lu\n", #x, (unsigned long)(dl_rq->x))
748
749 PU(dl_nr_running);
750#ifdef CONFIG_SMP
751 PU(dl_nr_migratory);
752 dl_bw = &cpu_rq(cpu)->rd->dl_bw;
753#else
754 dl_bw = &dl_rq->dl_bw;
755#endif
756 SEQ_printf(m, " .%-30s: %lld\n", "dl_bw->bw", dl_bw->bw);
757 SEQ_printf(m, " .%-30s: %lld\n", "dl_bw->total_bw", dl_bw->total_bw);
758
759#undef PU
760}
761
762static void print_cpu(struct seq_file *m, int cpu)
763{
764 struct rq *rq = cpu_rq(cpu);
765
766#ifdef CONFIG_X86
767 {
768 unsigned int freq = cpu_khz ? : 1;
769
770 SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
771 cpu, freq / 1000, (freq % 1000));
772 }
773#else
774 SEQ_printf(m, "cpu#%d\n", cpu);
775#endif
776
777#define P(x) \
778do { \
779 if (sizeof(rq->x) == 4) \
780 SEQ_printf(m, " .%-30s: %d\n", #x, (int)(rq->x)); \
781 else \
782 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rq->x));\
783} while (0)
784
785#define PN(x) \
786 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
787
788 P(nr_running);
789 P(nr_switches);
790 P(nr_uninterruptible);
791 PN(next_balance);
792 SEQ_printf(m, " .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr)));
793 PN(clock);
794 PN(clock_task);
795#undef P
796#undef PN
797
798#ifdef CONFIG_SMP
799#define P64(n) SEQ_printf(m, " .%-30s: %Ld\n", #n, rq->n);
800 P64(avg_idle);
801 P64(max_idle_balance_cost);
802#undef P64
803#endif
804
805#define P(n) SEQ_printf(m, " .%-30s: %d\n", #n, schedstat_val(rq->n));
806 if (schedstat_enabled()) {
807 P(yld_count);
808 P(sched_count);
809 P(sched_goidle);
810 P(ttwu_count);
811 P(ttwu_local);
812 }
813#undef P
814
815 print_cfs_stats(m, cpu);
816 print_rt_stats(m, cpu);
817 print_dl_stats(m, cpu);
818
819 print_rq(m, rq, cpu);
820 SEQ_printf(m, "\n");
821}
822
823static const char *sched_tunable_scaling_names[] = {
824 "none",
825 "logarithmic",
826 "linear"
827};
828
829static void sched_debug_header(struct seq_file *m)
830{
831 u64 ktime, sched_clk, cpu_clk;
832 unsigned long flags;
833
834 local_irq_save(flags);
835 ktime = ktime_to_ns(ktime_get());
836 sched_clk = sched_clock();
837 cpu_clk = local_clock();
838 local_irq_restore(flags);
839
840 SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n",
841 init_utsname()->release,
842 (int)strcspn(init_utsname()->version, " "),
843 init_utsname()->version);
844
845#define P(x) \
846 SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
847#define PN(x) \
848 SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
849 PN(ktime);
850 PN(sched_clk);
851 PN(cpu_clk);
852 P(jiffies);
853#ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
854 P(sched_clock_stable());
855#endif
856#undef PN
857#undef P
858
859 SEQ_printf(m, "\n");
860 SEQ_printf(m, "sysctl_sched\n");
861
862#define P(x) \
863 SEQ_printf(m, " .%-40s: %Ld\n", #x, (long long)(x))
864#define PN(x) \
865 SEQ_printf(m, " .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
866 PN(sysctl_sched_latency);
867 PN(sysctl_sched_min_granularity);
868 PN(sysctl_sched_idle_min_granularity);
869 PN(sysctl_sched_wakeup_granularity);
870 P(sysctl_sched_child_runs_first);
871 P(sysctl_sched_features);
872#undef PN
873#undef P
874
875 SEQ_printf(m, " .%-40s: %d (%s)\n",
876 "sysctl_sched_tunable_scaling",
877 sysctl_sched_tunable_scaling,
878 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
879 SEQ_printf(m, "\n");
880}
881
882static int sched_debug_show(struct seq_file *m, void *v)
883{
884 int cpu = (unsigned long)(v - 2);
885
886 if (cpu != -1)
887 print_cpu(m, cpu);
888 else
889 sched_debug_header(m);
890
891 return 0;
892}
893
894void sysrq_sched_debug_show(void)
895{
896 int cpu;
897
898 sched_debug_header(NULL);
899 for_each_online_cpu(cpu) {
900 /*
901 * Need to reset softlockup watchdogs on all CPUs, because
902 * another CPU might be blocked waiting for us to process
903 * an IPI or stop_machine.
904 */
905 touch_nmi_watchdog();
906 touch_all_softlockup_watchdogs();
907 print_cpu(NULL, cpu);
908 }
909}
910
911/*
912 * This iterator needs some explanation.
913 * It returns 1 for the header position.
914 * This means 2 is CPU 0.
915 * In a hotplugged system some CPUs, including CPU 0, may be missing so we have
916 * to use cpumask_* to iterate over the CPUs.
917 */
918static void *sched_debug_start(struct seq_file *file, loff_t *offset)
919{
920 unsigned long n = *offset;
921
922 if (n == 0)
923 return (void *) 1;
924
925 n--;
926
927 if (n > 0)
928 n = cpumask_next(n - 1, cpu_online_mask);
929 else
930 n = cpumask_first(cpu_online_mask);
931
932 *offset = n + 1;
933
934 if (n < nr_cpu_ids)
935 return (void *)(unsigned long)(n + 2);
936
937 return NULL;
938}
939
940static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
941{
942 (*offset)++;
943 return sched_debug_start(file, offset);
944}
945
946static void sched_debug_stop(struct seq_file *file, void *data)
947{
948}
949
950static const struct seq_operations sched_debug_sops = {
951 .start = sched_debug_start,
952 .next = sched_debug_next,
953 .stop = sched_debug_stop,
954 .show = sched_debug_show,
955};
956
957#define __PS(S, F) SEQ_printf(m, "%-45s:%21Ld\n", S, (long long)(F))
958#define __P(F) __PS(#F, F)
959#define P(F) __PS(#F, p->F)
960#define PM(F, M) __PS(#F, p->F & (M))
961#define __PSN(S, F) SEQ_printf(m, "%-45s:%14Ld.%06ld\n", S, SPLIT_NS((long long)(F)))
962#define __PN(F) __PSN(#F, F)
963#define PN(F) __PSN(#F, p->F)
964
965
966#ifdef CONFIG_NUMA_BALANCING
967void print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
968 unsigned long tpf, unsigned long gsf, unsigned long gpf)
969{
970 SEQ_printf(m, "numa_faults node=%d ", node);
971 SEQ_printf(m, "task_private=%lu task_shared=%lu ", tpf, tsf);
972 SEQ_printf(m, "group_private=%lu group_shared=%lu\n", gpf, gsf);
973}
974#endif
975
976
977static void sched_show_numa(struct task_struct *p, struct seq_file *m)
978{
979#ifdef CONFIG_NUMA_BALANCING
980 if (p->mm)
981 P(mm->numa_scan_seq);
982
983 P(numa_pages_migrated);
984 P(numa_preferred_nid);
985 P(total_numa_faults);
986 SEQ_printf(m, "current_node=%d, numa_group_id=%d\n",
987 task_node(p), task_numa_group_id(p));
988 show_numa_stats(p, m);
989#endif
990}
991
992void proc_sched_show_task(struct task_struct *p, struct pid_namespace *ns,
993 struct seq_file *m)
994{
995 unsigned long nr_switches;
996
997 SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr_ns(p, ns),
998 get_nr_threads(p));
999 SEQ_printf(m,
1000 "---------------------------------------------------------"
1001 "----------\n");
1002
1003#define P_SCHEDSTAT(F) __PS(#F, schedstat_val(p->stats.F))
1004#define PN_SCHEDSTAT(F) __PSN(#F, schedstat_val(p->stats.F))
1005
1006 PN(se.exec_start);
1007 PN(se.vruntime);
1008 PN(se.sum_exec_runtime);
1009
1010 nr_switches = p->nvcsw + p->nivcsw;
1011
1012 P(se.nr_migrations);
1013
1014 if (schedstat_enabled()) {
1015 u64 avg_atom, avg_per_cpu;
1016
1017 PN_SCHEDSTAT(sum_sleep_runtime);
1018 PN_SCHEDSTAT(sum_block_runtime);
1019 PN_SCHEDSTAT(wait_start);
1020 PN_SCHEDSTAT(sleep_start);
1021 PN_SCHEDSTAT(block_start);
1022 PN_SCHEDSTAT(sleep_max);
1023 PN_SCHEDSTAT(block_max);
1024 PN_SCHEDSTAT(exec_max);
1025 PN_SCHEDSTAT(slice_max);
1026 PN_SCHEDSTAT(wait_max);
1027 PN_SCHEDSTAT(wait_sum);
1028 P_SCHEDSTAT(wait_count);
1029 PN_SCHEDSTAT(iowait_sum);
1030 P_SCHEDSTAT(iowait_count);
1031 P_SCHEDSTAT(nr_migrations_cold);
1032 P_SCHEDSTAT(nr_failed_migrations_affine);
1033 P_SCHEDSTAT(nr_failed_migrations_running);
1034 P_SCHEDSTAT(nr_failed_migrations_hot);
1035 P_SCHEDSTAT(nr_forced_migrations);
1036 P_SCHEDSTAT(nr_wakeups);
1037 P_SCHEDSTAT(nr_wakeups_sync);
1038 P_SCHEDSTAT(nr_wakeups_migrate);
1039 P_SCHEDSTAT(nr_wakeups_local);
1040 P_SCHEDSTAT(nr_wakeups_remote);
1041 P_SCHEDSTAT(nr_wakeups_affine);
1042 P_SCHEDSTAT(nr_wakeups_affine_attempts);
1043 P_SCHEDSTAT(nr_wakeups_passive);
1044 P_SCHEDSTAT(nr_wakeups_idle);
1045
1046 avg_atom = p->se.sum_exec_runtime;
1047 if (nr_switches)
1048 avg_atom = div64_ul(avg_atom, nr_switches);
1049 else
1050 avg_atom = -1LL;
1051
1052 avg_per_cpu = p->se.sum_exec_runtime;
1053 if (p->se.nr_migrations) {
1054 avg_per_cpu = div64_u64(avg_per_cpu,
1055 p->se.nr_migrations);
1056 } else {
1057 avg_per_cpu = -1LL;
1058 }
1059
1060 __PN(avg_atom);
1061 __PN(avg_per_cpu);
1062
1063#ifdef CONFIG_SCHED_CORE
1064 PN_SCHEDSTAT(core_forceidle_sum);
1065#endif
1066 }
1067
1068 __P(nr_switches);
1069 __PS("nr_voluntary_switches", p->nvcsw);
1070 __PS("nr_involuntary_switches", p->nivcsw);
1071
1072 P(se.load.weight);
1073#ifdef CONFIG_SMP
1074 P(se.avg.load_sum);
1075 P(se.avg.runnable_sum);
1076 P(se.avg.util_sum);
1077 P(se.avg.load_avg);
1078 P(se.avg.runnable_avg);
1079 P(se.avg.util_avg);
1080 P(se.avg.last_update_time);
1081 P(se.avg.util_est.ewma);
1082 PM(se.avg.util_est.enqueued, ~UTIL_AVG_UNCHANGED);
1083#endif
1084#ifdef CONFIG_UCLAMP_TASK
1085 __PS("uclamp.min", p->uclamp_req[UCLAMP_MIN].value);
1086 __PS("uclamp.max", p->uclamp_req[UCLAMP_MAX].value);
1087 __PS("effective uclamp.min", uclamp_eff_value(p, UCLAMP_MIN));
1088 __PS("effective uclamp.max", uclamp_eff_value(p, UCLAMP_MAX));
1089#endif
1090 P(policy);
1091 P(prio);
1092 if (task_has_dl_policy(p)) {
1093 P(dl.runtime);
1094 P(dl.deadline);
1095 }
1096#undef PN_SCHEDSTAT
1097#undef P_SCHEDSTAT
1098
1099 {
1100 unsigned int this_cpu = raw_smp_processor_id();
1101 u64 t0, t1;
1102
1103 t0 = cpu_clock(this_cpu);
1104 t1 = cpu_clock(this_cpu);
1105 __PS("clock-delta", t1-t0);
1106 }
1107
1108 sched_show_numa(p, m);
1109}
1110
1111void proc_sched_set_task(struct task_struct *p)
1112{
1113#ifdef CONFIG_SCHEDSTATS
1114 memset(&p->stats, 0, sizeof(p->stats));
1115#endif
1116}
1117
1118void resched_latency_warn(int cpu, u64 latency)
1119{
1120 static DEFINE_RATELIMIT_STATE(latency_check_ratelimit, 60 * 60 * HZ, 1);
1121
1122 WARN(__ratelimit(&latency_check_ratelimit),
1123 "sched: CPU %d need_resched set for > %llu ns (%d ticks) "
1124 "without schedule\n",
1125 cpu, latency, cpu_rq(cpu)->ticks_without_resched);
1126}