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 /sys/kernel/debug/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, s: "NO_");
63 seq_printf(m, fmt: "%s ", sched_feat_names[i]);
64 }
65 seq_puts(m, s: "\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(key: &sched_feat_keys[i]);
87}
88
89static void sched_feat_enable(int i)
90{
91 static_key_enable_cpuslocked(key: &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(array: sched_feat_names, n: __SCHED_FEAT_NR, string: 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(to: &buf, from: ubuf, n: cnt))
136 return -EFAULT;
137
138 buf[cnt] = 0;
139 cmp = strstrip(str: buf);
140
141 /* Ensure the static_key remains in a consistent state */
142 inode = file_inode(f: 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(to: &buf, from: ubuf, n: cnt))
181 return -EFAULT;
182 buf[cnt] = '\0';
183
184 if (kstrtouint(s: buf, base: 10, res: &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, fmt: "%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(to: &buf, from: ubuf, n: cnt))
231 return -EFAULT;
232
233 buf[cnt] = 0;
234 mode = sched_dynamic_mode(str: strstrip(str: 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, s: "(");
255 seq_puts(m, s: preempt_modes[i]);
256 if (preempt_dynamic_mode == i)
257 seq_puts(m, s: ")");
258
259 seq_puts(m, s: " ");
260 }
261
262 seq_puts(m, s: "\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(file: filp, user_buf: ubuf, count: cnt, ppos);
298
299 if (sched_debug_verbose && !orig)
300 update_sched_domain_debugfs();
301 else if (!sched_debug_verbose && orig) {
302 debugfs_remove(dentry: sd_dentry);
303 sd_dentry = NULL;
304 }
305
306 mutex_unlock(lock: &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(name: "sched", NULL);
343
344 debugfs_create_file(name: "features", mode: 0644, parent: debugfs_sched, NULL, fops: &sched_feat_fops);
345 debugfs_create_file_unsafe(name: "verbose", mode: 0644, parent: debugfs_sched, data: &sched_debug_verbose, fops: &sched_verbose_fops);
346#ifdef CONFIG_PREEMPT_DYNAMIC
347 debugfs_create_file(name: "preempt", mode: 0644, parent: debugfs_sched, NULL, fops: &sched_dynamic_fops);
348#endif
349
350 debugfs_create_u32(name: "base_slice_ns", mode: 0644, parent: debugfs_sched, value: &sysctl_sched_base_slice);
351
352 debugfs_create_u32(name: "latency_warn_ms", mode: 0644, parent: debugfs_sched, value: &sysctl_resched_latency_warn_ms);
353 debugfs_create_u32(name: "latency_warn_once", mode: 0644, parent: debugfs_sched, value: &sysctl_resched_latency_warn_once);
354
355#ifdef CONFIG_SMP
356 debugfs_create_file(name: "tunable_scaling", mode: 0644, parent: debugfs_sched, NULL, fops: &sched_scaling_fops);
357 debugfs_create_u32(name: "migration_cost_ns", mode: 0644, parent: debugfs_sched, value: &sysctl_sched_migration_cost);
358 debugfs_create_u32(name: "nr_migrate", mode: 0644, parent: debugfs_sched, value: &sysctl_sched_nr_migrate);
359
360 mutex_lock(&sched_domains_mutex);
361 update_sched_domain_debugfs();
362 mutex_unlock(lock: &sched_domains_mutex);
363#endif
364
365#ifdef CONFIG_NUMA_BALANCING
366 numa = debugfs_create_dir(name: "numa_balancing", parent: debugfs_sched);
367
368 debugfs_create_u32(name: "scan_delay_ms", mode: 0644, parent: numa, value: &sysctl_numa_balancing_scan_delay);
369 debugfs_create_u32(name: "scan_period_min_ms", mode: 0644, parent: numa, value: &sysctl_numa_balancing_scan_period_min);
370 debugfs_create_u32(name: "scan_period_max_ms", mode: 0644, parent: numa, value: &sysctl_numa_balancing_scan_period_max);
371 debugfs_create_u32(name: "scan_size_mb", mode: 0644, parent: numa, value: &sysctl_numa_balancing_scan_size);
372 debugfs_create_u32(name: "hot_threshold_ms", mode: 0644, parent: numa, value: &sysctl_numa_balancing_hot_threshold);
373#endif
374
375 debugfs_create_file(name: "debug", mode: 0444, parent: debugfs_sched, NULL, fops: &sched_debug_fops);
376
377 return 0;
378}
379late_initcall(sched_init_debug);
380
381#ifdef CONFIG_SMP
382
383static cpumask_var_t sd_sysctl_cpus;
384
385static int sd_flags_show(struct seq_file *m, void *v)
386{
387 unsigned long flags = *(unsigned int *)m->private;
388 int idx;
389
390 for_each_set_bit(idx, &flags, __SD_FLAG_CNT) {
391 seq_puts(m, s: sd_flag_debug[idx].name);
392 seq_puts(m, s: " ");
393 }
394 seq_puts(m, s: "\n");
395
396 return 0;
397}
398
399static int sd_flags_open(struct inode *inode, struct file *file)
400{
401 return single_open(file, sd_flags_show, inode->i_private);
402}
403
404static const struct file_operations sd_flags_fops = {
405 .open = sd_flags_open,
406 .read = seq_read,
407 .llseek = seq_lseek,
408 .release = single_release,
409};
410
411static void register_sd(struct sched_domain *sd, struct dentry *parent)
412{
413#define SDM(type, mode, member) \
414 debugfs_create_##type(#member, mode, parent, &sd->member)
415
416 SDM(ulong, 0644, min_interval);
417 SDM(ulong, 0644, max_interval);
418 SDM(u64, 0644, max_newidle_lb_cost);
419 SDM(u32, 0644, busy_factor);
420 SDM(u32, 0644, imbalance_pct);
421 SDM(u32, 0644, cache_nice_tries);
422 SDM(str, 0444, name);
423
424#undef SDM
425
426 debugfs_create_file(name: "flags", mode: 0444, parent, data: &sd->flags, fops: &sd_flags_fops);
427 debugfs_create_file(name: "groups_flags", mode: 0444, parent, data: &sd->groups->flags, fops: &sd_flags_fops);
428}
429
430void update_sched_domain_debugfs(void)
431{
432 int cpu, i;
433
434 /*
435 * This can unfortunately be invoked before sched_debug_init() creates
436 * the debug directory. Don't touch sd_sysctl_cpus until then.
437 */
438 if (!debugfs_sched)
439 return;
440
441 if (!sched_debug_verbose)
442 return;
443
444 if (!cpumask_available(mask: sd_sysctl_cpus)) {
445 if (!alloc_cpumask_var(mask: &sd_sysctl_cpus, GFP_KERNEL))
446 return;
447 cpumask_copy(dstp: sd_sysctl_cpus, cpu_possible_mask);
448 }
449
450 if (!sd_dentry) {
451 sd_dentry = debugfs_create_dir(name: "domains", parent: debugfs_sched);
452
453 /* rebuild sd_sysctl_cpus if empty since it gets cleared below */
454 if (cpumask_empty(srcp: sd_sysctl_cpus))
455 cpumask_copy(dstp: sd_sysctl_cpus, cpu_online_mask);
456 }
457
458 for_each_cpu(cpu, sd_sysctl_cpus) {
459 struct sched_domain *sd;
460 struct dentry *d_cpu;
461 char buf[32];
462
463 snprintf(buf, size: sizeof(buf), fmt: "cpu%d", cpu);
464 debugfs_lookup_and_remove(name: buf, parent: sd_dentry);
465 d_cpu = debugfs_create_dir(name: buf, parent: sd_dentry);
466
467 i = 0;
468 for_each_domain(cpu, sd) {
469 struct dentry *d_sd;
470
471 snprintf(buf, size: sizeof(buf), fmt: "domain%d", i);
472 d_sd = debugfs_create_dir(name: buf, parent: d_cpu);
473
474 register_sd(sd, parent: d_sd);
475 i++;
476 }
477
478 __cpumask_clear_cpu(cpu, dstp: sd_sysctl_cpus);
479 }
480}
481
482void dirty_sched_domain_sysctl(int cpu)
483{
484 if (cpumask_available(mask: sd_sysctl_cpus))
485 __cpumask_set_cpu(cpu, dstp: sd_sysctl_cpus);
486}
487
488#endif /* CONFIG_SMP */
489
490#ifdef CONFIG_FAIR_GROUP_SCHED
491static void print_cfs_group_stats(struct seq_file *m, int cpu, struct task_group *tg)
492{
493 struct sched_entity *se = tg->se[cpu];
494
495#define P(F) SEQ_printf(m, " .%-30s: %lld\n", #F, (long long)F)
496#define P_SCHEDSTAT(F) SEQ_printf(m, " .%-30s: %lld\n", \
497 #F, (long long)schedstat_val(stats->F))
498#define PN(F) SEQ_printf(m, " .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F))
499#define PN_SCHEDSTAT(F) SEQ_printf(m, " .%-30s: %lld.%06ld\n", \
500 #F, SPLIT_NS((long long)schedstat_val(stats->F)))
501
502 if (!se)
503 return;
504
505 PN(se->exec_start);
506 PN(se->vruntime);
507 PN(se->sum_exec_runtime);
508
509 if (schedstat_enabled()) {
510 struct sched_statistics *stats;
511 stats = __schedstats_from_se(se);
512
513 PN_SCHEDSTAT(wait_start);
514 PN_SCHEDSTAT(sleep_start);
515 PN_SCHEDSTAT(block_start);
516 PN_SCHEDSTAT(sleep_max);
517 PN_SCHEDSTAT(block_max);
518 PN_SCHEDSTAT(exec_max);
519 PN_SCHEDSTAT(slice_max);
520 PN_SCHEDSTAT(wait_max);
521 PN_SCHEDSTAT(wait_sum);
522 P_SCHEDSTAT(wait_count);
523 }
524
525 P(se->load.weight);
526#ifdef CONFIG_SMP
527 P(se->avg.load_avg);
528 P(se->avg.util_avg);
529 P(se->avg.runnable_avg);
530#endif
531
532#undef PN_SCHEDSTAT
533#undef PN
534#undef P_SCHEDSTAT
535#undef P
536}
537#endif
538
539#ifdef CONFIG_CGROUP_SCHED
540static DEFINE_SPINLOCK(sched_debug_lock);
541static char group_path[PATH_MAX];
542
543static void task_group_path(struct task_group *tg, char *path, int plen)
544{
545 if (autogroup_path(tg, buf: path, buflen: plen))
546 return;
547
548 cgroup_path(cgrp: tg->css.cgroup, buf: path, buflen: plen);
549}
550
551/*
552 * Only 1 SEQ_printf_task_group_path() caller can use the full length
553 * group_path[] for cgroup path. Other simultaneous callers will have
554 * to use a shorter stack buffer. A "..." suffix is appended at the end
555 * of the stack buffer so that it will show up in case the output length
556 * matches the given buffer size to indicate possible path name truncation.
557 */
558#define SEQ_printf_task_group_path(m, tg, fmt...) \
559{ \
560 if (spin_trylock(&sched_debug_lock)) { \
561 task_group_path(tg, group_path, sizeof(group_path)); \
562 SEQ_printf(m, fmt, group_path); \
563 spin_unlock(&sched_debug_lock); \
564 } else { \
565 char buf[128]; \
566 char *bufend = buf + sizeof(buf) - 3; \
567 task_group_path(tg, buf, bufend - buf); \
568 strcpy(bufend - 1, "..."); \
569 SEQ_printf(m, fmt, buf); \
570 } \
571}
572#endif
573
574static void
575print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
576{
577 if (task_current(rq, p))
578 SEQ_printf(m, ">R");
579 else
580 SEQ_printf(m, " %c", task_state_to_char(p));
581
582 SEQ_printf(m, "%15s %5d %9Ld.%06ld %c %9Ld.%06ld %9Ld.%06ld %9Ld.%06ld %9Ld %5d ",
583 p->comm, task_pid_nr(p),
584 SPLIT_NS(p->se.vruntime),
585 entity_eligible(cfs_rq_of(&p->se), &p->se) ? 'E' : 'N',
586 SPLIT_NS(p->se.deadline),
587 SPLIT_NS(p->se.slice),
588 SPLIT_NS(p->se.sum_exec_runtime),
589 (long long)(p->nvcsw + p->nivcsw),
590 p->prio);
591
592 SEQ_printf(m, "%9lld.%06ld %9lld.%06ld %9lld.%06ld %9lld.%06ld",
593 SPLIT_NS(schedstat_val_or_zero(p->stats.wait_sum)),
594 SPLIT_NS(p->se.sum_exec_runtime),
595 SPLIT_NS(schedstat_val_or_zero(p->stats.sum_sleep_runtime)),
596 SPLIT_NS(schedstat_val_or_zero(p->stats.sum_block_runtime)));
597
598#ifdef CONFIG_NUMA_BALANCING
599 SEQ_printf(m, " %d %d", task_node(p), task_numa_group_id(p));
600#endif
601#ifdef CONFIG_CGROUP_SCHED
602 SEQ_printf_task_group_path(m, task_group(p), " %s")
603#endif
604
605 SEQ_printf(m, "\n");
606}
607
608static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
609{
610 struct task_struct *g, *p;
611
612 SEQ_printf(m, "\n");
613 SEQ_printf(m, "runnable tasks:\n");
614 SEQ_printf(m, " S task PID tree-key switches prio"
615 " wait-time sum-exec sum-sleep\n");
616 SEQ_printf(m, "-------------------------------------------------------"
617 "------------------------------------------------------\n");
618
619 rcu_read_lock();
620 for_each_process_thread(g, p) {
621 if (task_cpu(p) != rq_cpu)
622 continue;
623
624 print_task(m, rq, p);
625 }
626 rcu_read_unlock();
627}
628
629void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
630{
631 s64 left_vruntime = -1, min_vruntime, right_vruntime = -1, spread;
632 struct sched_entity *last, *first;
633 struct rq *rq = cpu_rq(cpu);
634 unsigned long flags;
635
636#ifdef CONFIG_FAIR_GROUP_SCHED
637 SEQ_printf(m, "\n");
638 SEQ_printf_task_group_path(m, cfs_rq->tg, "cfs_rq[%d]:%s\n", cpu);
639#else
640 SEQ_printf(m, "\n");
641 SEQ_printf(m, "cfs_rq[%d]:\n", cpu);
642#endif
643 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "exec_clock",
644 SPLIT_NS(cfs_rq->exec_clock));
645
646 raw_spin_rq_lock_irqsave(rq, flags);
647 first = __pick_first_entity(cfs_rq);
648 if (first)
649 left_vruntime = first->vruntime;
650 last = __pick_last_entity(cfs_rq);
651 if (last)
652 right_vruntime = last->vruntime;
653 min_vruntime = cfs_rq->min_vruntime;
654 raw_spin_rq_unlock_irqrestore(rq, flags);
655
656 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "left_vruntime",
657 SPLIT_NS(left_vruntime));
658 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "min_vruntime",
659 SPLIT_NS(min_vruntime));
660 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "avg_vruntime",
661 SPLIT_NS(avg_vruntime(cfs_rq)));
662 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "right_vruntime",
663 SPLIT_NS(right_vruntime));
664 spread = right_vruntime - left_vruntime;
665 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread", SPLIT_NS(spread));
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, tg: 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 P(rt_throttled);
728 PN(rt_time);
729 PN(rt_runtime);
730
731#undef PN
732#undef PU
733#undef P
734}
735
736void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq)
737{
738 struct dl_bw *dl_bw;
739
740 SEQ_printf(m, "\n");
741 SEQ_printf(m, "dl_rq[%d]:\n", cpu);
742
743#define PU(x) \
744 SEQ_printf(m, " .%-30s: %lu\n", #x, (unsigned long)(dl_rq->x))
745
746 PU(dl_nr_running);
747#ifdef CONFIG_SMP
748 dl_bw = &cpu_rq(cpu)->rd->dl_bw;
749#else
750 dl_bw = &dl_rq->dl_bw;
751#endif
752 SEQ_printf(m, " .%-30s: %lld\n", "dl_bw->bw", dl_bw->bw);
753 SEQ_printf(m, " .%-30s: %lld\n", "dl_bw->total_bw", dl_bw->total_bw);
754
755#undef PU
756}
757
758static void print_cpu(struct seq_file *m, int cpu)
759{
760 struct rq *rq = cpu_rq(cpu);
761
762#ifdef CONFIG_X86
763 {
764 unsigned int freq = cpu_khz ? : 1;
765
766 SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
767 cpu, freq / 1000, (freq % 1000));
768 }
769#else
770 SEQ_printf(m, "cpu#%d\n", cpu);
771#endif
772
773#define P(x) \
774do { \
775 if (sizeof(rq->x) == 4) \
776 SEQ_printf(m, " .%-30s: %d\n", #x, (int)(rq->x)); \
777 else \
778 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rq->x));\
779} while (0)
780
781#define PN(x) \
782 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
783
784 P(nr_running);
785 P(nr_switches);
786 P(nr_uninterruptible);
787 PN(next_balance);
788 SEQ_printf(m, " .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr)));
789 PN(clock);
790 PN(clock_task);
791#undef P
792#undef PN
793
794#ifdef CONFIG_SMP
795#define P64(n) SEQ_printf(m, " .%-30s: %Ld\n", #n, rq->n);
796 P64(avg_idle);
797 P64(max_idle_balance_cost);
798#undef P64
799#endif
800
801#define P(n) SEQ_printf(m, " .%-30s: %d\n", #n, schedstat_val(rq->n));
802 if (schedstat_enabled()) {
803 P(yld_count);
804 P(sched_count);
805 P(sched_goidle);
806 P(ttwu_count);
807 P(ttwu_local);
808 }
809#undef P
810
811 print_cfs_stats(m, cpu);
812 print_rt_stats(m, cpu);
813 print_dl_stats(m, cpu);
814
815 print_rq(m, rq, rq_cpu: cpu);
816 SEQ_printf(m, "\n");
817}
818
819static const char *sched_tunable_scaling_names[] = {
820 "none",
821 "logarithmic",
822 "linear"
823};
824
825static void sched_debug_header(struct seq_file *m)
826{
827 u64 ktime, sched_clk, cpu_clk;
828 unsigned long flags;
829
830 local_irq_save(flags);
831 ktime = ktime_to_ns(kt: ktime_get());
832 sched_clk = sched_clock();
833 cpu_clk = local_clock();
834 local_irq_restore(flags);
835
836 SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n",
837 init_utsname()->release,
838 (int)strcspn(init_utsname()->version, " "),
839 init_utsname()->version);
840
841#define P(x) \
842 SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
843#define PN(x) \
844 SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
845 PN(ktime);
846 PN(sched_clk);
847 PN(cpu_clk);
848 P(jiffies);
849#ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
850 P(sched_clock_stable());
851#endif
852#undef PN
853#undef P
854
855 SEQ_printf(m, "\n");
856 SEQ_printf(m, "sysctl_sched\n");
857
858#define P(x) \
859 SEQ_printf(m, " .%-40s: %Ld\n", #x, (long long)(x))
860#define PN(x) \
861 SEQ_printf(m, " .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
862 PN(sysctl_sched_base_slice);
863 P(sysctl_sched_features);
864#undef PN
865#undef P
866
867 SEQ_printf(m, " .%-40s: %d (%s)\n",
868 "sysctl_sched_tunable_scaling",
869 sysctl_sched_tunable_scaling,
870 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
871 SEQ_printf(m, "\n");
872}
873
874static int sched_debug_show(struct seq_file *m, void *v)
875{
876 int cpu = (unsigned long)(v - 2);
877
878 if (cpu != -1)
879 print_cpu(m, cpu);
880 else
881 sched_debug_header(m);
882
883 return 0;
884}
885
886void sysrq_sched_debug_show(void)
887{
888 int cpu;
889
890 sched_debug_header(NULL);
891 for_each_online_cpu(cpu) {
892 /*
893 * Need to reset softlockup watchdogs on all CPUs, because
894 * another CPU might be blocked waiting for us to process
895 * an IPI or stop_machine.
896 */
897 touch_nmi_watchdog();
898 touch_all_softlockup_watchdogs();
899 print_cpu(NULL, cpu);
900 }
901}
902
903/*
904 * This iterator needs some explanation.
905 * It returns 1 for the header position.
906 * This means 2 is CPU 0.
907 * In a hotplugged system some CPUs, including CPU 0, may be missing so we have
908 * to use cpumask_* to iterate over the CPUs.
909 */
910static void *sched_debug_start(struct seq_file *file, loff_t *offset)
911{
912 unsigned long n = *offset;
913
914 if (n == 0)
915 return (void *) 1;
916
917 n--;
918
919 if (n > 0)
920 n = cpumask_next(n: n - 1, cpu_online_mask);
921 else
922 n = cpumask_first(cpu_online_mask);
923
924 *offset = n + 1;
925
926 if (n < nr_cpu_ids)
927 return (void *)(unsigned long)(n + 2);
928
929 return NULL;
930}
931
932static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
933{
934 (*offset)++;
935 return sched_debug_start(file, offset);
936}
937
938static void sched_debug_stop(struct seq_file *file, void *data)
939{
940}
941
942static const struct seq_operations sched_debug_sops = {
943 .start = sched_debug_start,
944 .next = sched_debug_next,
945 .stop = sched_debug_stop,
946 .show = sched_debug_show,
947};
948
949#define __PS(S, F) SEQ_printf(m, "%-45s:%21Ld\n", S, (long long)(F))
950#define __P(F) __PS(#F, F)
951#define P(F) __PS(#F, p->F)
952#define PM(F, M) __PS(#F, p->F & (M))
953#define __PSN(S, F) SEQ_printf(m, "%-45s:%14Ld.%06ld\n", S, SPLIT_NS((long long)(F)))
954#define __PN(F) __PSN(#F, F)
955#define PN(F) __PSN(#F, p->F)
956
957
958#ifdef CONFIG_NUMA_BALANCING
959void print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
960 unsigned long tpf, unsigned long gsf, unsigned long gpf)
961{
962 SEQ_printf(m, "numa_faults node=%d ", node);
963 SEQ_printf(m, "task_private=%lu task_shared=%lu ", tpf, tsf);
964 SEQ_printf(m, "group_private=%lu group_shared=%lu\n", gpf, gsf);
965}
966#endif
967
968
969static void sched_show_numa(struct task_struct *p, struct seq_file *m)
970{
971#ifdef CONFIG_NUMA_BALANCING
972 if (p->mm)
973 P(mm->numa_scan_seq);
974
975 P(numa_pages_migrated);
976 P(numa_preferred_nid);
977 P(total_numa_faults);
978 SEQ_printf(m, "current_node=%d, numa_group_id=%d\n",
979 task_node(p), task_numa_group_id(p));
980 show_numa_stats(p, m);
981#endif
982}
983
984void proc_sched_show_task(struct task_struct *p, struct pid_namespace *ns,
985 struct seq_file *m)
986{
987 unsigned long nr_switches;
988
989 SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr_ns(p, ns),
990 get_nr_threads(p));
991 SEQ_printf(m,
992 "---------------------------------------------------------"
993 "----------\n");
994
995#define P_SCHEDSTAT(F) __PS(#F, schedstat_val(p->stats.F))
996#define PN_SCHEDSTAT(F) __PSN(#F, schedstat_val(p->stats.F))
997
998 PN(se.exec_start);
999 PN(se.vruntime);
1000 PN(se.sum_exec_runtime);
1001
1002 nr_switches = p->nvcsw + p->nivcsw;
1003
1004 P(se.nr_migrations);
1005
1006 if (schedstat_enabled()) {
1007 u64 avg_atom, avg_per_cpu;
1008
1009 PN_SCHEDSTAT(sum_sleep_runtime);
1010 PN_SCHEDSTAT(sum_block_runtime);
1011 PN_SCHEDSTAT(wait_start);
1012 PN_SCHEDSTAT(sleep_start);
1013 PN_SCHEDSTAT(block_start);
1014 PN_SCHEDSTAT(sleep_max);
1015 PN_SCHEDSTAT(block_max);
1016 PN_SCHEDSTAT(exec_max);
1017 PN_SCHEDSTAT(slice_max);
1018 PN_SCHEDSTAT(wait_max);
1019 PN_SCHEDSTAT(wait_sum);
1020 P_SCHEDSTAT(wait_count);
1021 PN_SCHEDSTAT(iowait_sum);
1022 P_SCHEDSTAT(iowait_count);
1023 P_SCHEDSTAT(nr_migrations_cold);
1024 P_SCHEDSTAT(nr_failed_migrations_affine);
1025 P_SCHEDSTAT(nr_failed_migrations_running);
1026 P_SCHEDSTAT(nr_failed_migrations_hot);
1027 P_SCHEDSTAT(nr_forced_migrations);
1028 P_SCHEDSTAT(nr_wakeups);
1029 P_SCHEDSTAT(nr_wakeups_sync);
1030 P_SCHEDSTAT(nr_wakeups_migrate);
1031 P_SCHEDSTAT(nr_wakeups_local);
1032 P_SCHEDSTAT(nr_wakeups_remote);
1033 P_SCHEDSTAT(nr_wakeups_affine);
1034 P_SCHEDSTAT(nr_wakeups_affine_attempts);
1035 P_SCHEDSTAT(nr_wakeups_passive);
1036 P_SCHEDSTAT(nr_wakeups_idle);
1037
1038 avg_atom = p->se.sum_exec_runtime;
1039 if (nr_switches)
1040 avg_atom = div64_ul(avg_atom, nr_switches);
1041 else
1042 avg_atom = -1LL;
1043
1044 avg_per_cpu = p->se.sum_exec_runtime;
1045 if (p->se.nr_migrations) {
1046 avg_per_cpu = div64_u64(dividend: avg_per_cpu,
1047 divisor: p->se.nr_migrations);
1048 } else {
1049 avg_per_cpu = -1LL;
1050 }
1051
1052 __PN(avg_atom);
1053 __PN(avg_per_cpu);
1054
1055#ifdef CONFIG_SCHED_CORE
1056 PN_SCHEDSTAT(core_forceidle_sum);
1057#endif
1058 }
1059
1060 __P(nr_switches);
1061 __PS("nr_voluntary_switches", p->nvcsw);
1062 __PS("nr_involuntary_switches", p->nivcsw);
1063
1064 P(se.load.weight);
1065#ifdef CONFIG_SMP
1066 P(se.avg.load_sum);
1067 P(se.avg.runnable_sum);
1068 P(se.avg.util_sum);
1069 P(se.avg.load_avg);
1070 P(se.avg.runnable_avg);
1071 P(se.avg.util_avg);
1072 P(se.avg.last_update_time);
1073 P(se.avg.util_est.ewma);
1074 PM(se.avg.util_est.enqueued, ~UTIL_AVG_UNCHANGED);
1075#endif
1076#ifdef CONFIG_UCLAMP_TASK
1077 __PS("uclamp.min", p->uclamp_req[UCLAMP_MIN].value);
1078 __PS("uclamp.max", p->uclamp_req[UCLAMP_MAX].value);
1079 __PS("effective uclamp.min", uclamp_eff_value(p, UCLAMP_MIN));
1080 __PS("effective uclamp.max", uclamp_eff_value(p, UCLAMP_MAX));
1081#endif
1082 P(policy);
1083 P(prio);
1084 if (task_has_dl_policy(p)) {
1085 P(dl.runtime);
1086 P(dl.deadline);
1087 }
1088#undef PN_SCHEDSTAT
1089#undef P_SCHEDSTAT
1090
1091 {
1092 unsigned int this_cpu = raw_smp_processor_id();
1093 u64 t0, t1;
1094
1095 t0 = cpu_clock(cpu: this_cpu);
1096 t1 = cpu_clock(cpu: this_cpu);
1097 __PS("clock-delta", t1-t0);
1098 }
1099
1100 sched_show_numa(p, m);
1101}
1102
1103void proc_sched_set_task(struct task_struct *p)
1104{
1105#ifdef CONFIG_SCHEDSTATS
1106 memset(&p->stats, 0, sizeof(p->stats));
1107#endif
1108}
1109
1110void resched_latency_warn(int cpu, u64 latency)
1111{
1112 static DEFINE_RATELIMIT_STATE(latency_check_ratelimit, 60 * 60 * HZ, 1);
1113
1114 WARN(__ratelimit(&latency_check_ratelimit),
1115 "sched: CPU %d need_resched set for > %llu ns (%d ticks) "
1116 "without schedule\n",
1117 cpu, latency, cpu_rq(cpu)->ticks_without_resched);
1118}
1119

source code of linux/kernel/sched/debug.c