Loading...
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, "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", "lazy",
249 };
250 int j = ARRAY_SIZE(preempt_modes) - !IS_ENABLED(CONFIG_ARCH_HAS_PREEMPT_LAZY);
251 int i = IS_ENABLED(CONFIG_PREEMPT_RT) * 2;
252
253 for (; i < j; 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
284#ifdef CONFIG_SMP
285static struct dentry *sd_dentry;
286
287
288static ssize_t sched_verbose_write(struct file *filp, const char __user *ubuf,
289 size_t cnt, loff_t *ppos)
290{
291 ssize_t result;
292 bool orig;
293
294 cpus_read_lock();
295 mutex_lock(&sched_domains_mutex);
296
297 orig = sched_debug_verbose;
298 result = debugfs_write_file_bool(filp, ubuf, cnt, ppos);
299
300 if (sched_debug_verbose && !orig)
301 update_sched_domain_debugfs();
302 else if (!sched_debug_verbose && orig) {
303 debugfs_remove(sd_dentry);
304 sd_dentry = NULL;
305 }
306
307 mutex_unlock(&sched_domains_mutex);
308 cpus_read_unlock();
309
310 return result;
311}
312#else
313#define sched_verbose_write debugfs_write_file_bool
314#endif
315
316static const struct file_operations sched_verbose_fops = {
317 .read = debugfs_read_file_bool,
318 .write = sched_verbose_write,
319 .open = simple_open,
320 .llseek = default_llseek,
321};
322
323static const struct seq_operations sched_debug_sops;
324
325static int sched_debug_open(struct inode *inode, struct file *filp)
326{
327 return seq_open(filp, &sched_debug_sops);
328}
329
330static const struct file_operations sched_debug_fops = {
331 .open = sched_debug_open,
332 .read = seq_read,
333 .llseek = seq_lseek,
334 .release = seq_release,
335};
336
337enum dl_param {
338 DL_RUNTIME = 0,
339 DL_PERIOD,
340};
341
342static unsigned long fair_server_period_max = (1UL << 22) * NSEC_PER_USEC; /* ~4 seconds */
343static unsigned long fair_server_period_min = (100) * NSEC_PER_USEC; /* 100 us */
344
345static ssize_t sched_fair_server_write(struct file *filp, const char __user *ubuf,
346 size_t cnt, loff_t *ppos, enum dl_param param)
347{
348 long cpu = (long) ((struct seq_file *) filp->private_data)->private;
349 struct rq *rq = cpu_rq(cpu);
350 u64 runtime, period;
351 size_t err;
352 int retval;
353 u64 value;
354
355 err = kstrtoull_from_user(ubuf, cnt, 10, &value);
356 if (err)
357 return err;
358
359 scoped_guard (rq_lock_irqsave, rq) {
360 runtime = rq->fair_server.dl_runtime;
361 period = rq->fair_server.dl_period;
362
363 switch (param) {
364 case DL_RUNTIME:
365 if (runtime == value)
366 break;
367 runtime = value;
368 break;
369 case DL_PERIOD:
370 if (value == period)
371 break;
372 period = value;
373 break;
374 }
375
376 if (runtime > period ||
377 period > fair_server_period_max ||
378 period < fair_server_period_min) {
379 return -EINVAL;
380 }
381
382 if (rq->cfs.h_nr_running) {
383 update_rq_clock(rq);
384 dl_server_stop(&rq->fair_server);
385 }
386
387 retval = dl_server_apply_params(&rq->fair_server, runtime, period, 0);
388 if (retval)
389 cnt = retval;
390
391 if (!runtime)
392 printk_deferred("Fair server disabled in CPU %d, system may crash due to starvation.\n",
393 cpu_of(rq));
394
395 if (rq->cfs.h_nr_running)
396 dl_server_start(&rq->fair_server);
397 }
398
399 *ppos += cnt;
400 return cnt;
401}
402
403static size_t sched_fair_server_show(struct seq_file *m, void *v, enum dl_param param)
404{
405 unsigned long cpu = (unsigned long) m->private;
406 struct rq *rq = cpu_rq(cpu);
407 u64 value;
408
409 switch (param) {
410 case DL_RUNTIME:
411 value = rq->fair_server.dl_runtime;
412 break;
413 case DL_PERIOD:
414 value = rq->fair_server.dl_period;
415 break;
416 }
417
418 seq_printf(m, "%llu\n", value);
419 return 0;
420
421}
422
423static ssize_t
424sched_fair_server_runtime_write(struct file *filp, const char __user *ubuf,
425 size_t cnt, loff_t *ppos)
426{
427 return sched_fair_server_write(filp, ubuf, cnt, ppos, DL_RUNTIME);
428}
429
430static int sched_fair_server_runtime_show(struct seq_file *m, void *v)
431{
432 return sched_fair_server_show(m, v, DL_RUNTIME);
433}
434
435static int sched_fair_server_runtime_open(struct inode *inode, struct file *filp)
436{
437 return single_open(filp, sched_fair_server_runtime_show, inode->i_private);
438}
439
440static const struct file_operations fair_server_runtime_fops = {
441 .open = sched_fair_server_runtime_open,
442 .write = sched_fair_server_runtime_write,
443 .read = seq_read,
444 .llseek = seq_lseek,
445 .release = single_release,
446};
447
448static ssize_t
449sched_fair_server_period_write(struct file *filp, const char __user *ubuf,
450 size_t cnt, loff_t *ppos)
451{
452 return sched_fair_server_write(filp, ubuf, cnt, ppos, DL_PERIOD);
453}
454
455static int sched_fair_server_period_show(struct seq_file *m, void *v)
456{
457 return sched_fair_server_show(m, v, DL_PERIOD);
458}
459
460static int sched_fair_server_period_open(struct inode *inode, struct file *filp)
461{
462 return single_open(filp, sched_fair_server_period_show, inode->i_private);
463}
464
465static const struct file_operations fair_server_period_fops = {
466 .open = sched_fair_server_period_open,
467 .write = sched_fair_server_period_write,
468 .read = seq_read,
469 .llseek = seq_lseek,
470 .release = single_release,
471};
472
473static struct dentry *debugfs_sched;
474
475static void debugfs_fair_server_init(void)
476{
477 struct dentry *d_fair;
478 unsigned long cpu;
479
480 d_fair = debugfs_create_dir("fair_server", debugfs_sched);
481 if (!d_fair)
482 return;
483
484 for_each_possible_cpu(cpu) {
485 struct dentry *d_cpu;
486 char buf[32];
487
488 snprintf(buf, sizeof(buf), "cpu%lu", cpu);
489 d_cpu = debugfs_create_dir(buf, d_fair);
490
491 debugfs_create_file("runtime", 0644, d_cpu, (void *) cpu, &fair_server_runtime_fops);
492 debugfs_create_file("period", 0644, d_cpu, (void *) cpu, &fair_server_period_fops);
493 }
494}
495
496static __init int sched_init_debug(void)
497{
498 struct dentry __maybe_unused *numa;
499
500 debugfs_sched = debugfs_create_dir("sched", NULL);
501
502 debugfs_create_file("features", 0644, debugfs_sched, NULL, &sched_feat_fops);
503 debugfs_create_file_unsafe("verbose", 0644, debugfs_sched, &sched_debug_verbose, &sched_verbose_fops);
504#ifdef CONFIG_PREEMPT_DYNAMIC
505 debugfs_create_file("preempt", 0644, debugfs_sched, NULL, &sched_dynamic_fops);
506#endif
507
508 debugfs_create_u32("base_slice_ns", 0644, debugfs_sched, &sysctl_sched_base_slice);
509
510 debugfs_create_u32("latency_warn_ms", 0644, debugfs_sched, &sysctl_resched_latency_warn_ms);
511 debugfs_create_u32("latency_warn_once", 0644, debugfs_sched, &sysctl_resched_latency_warn_once);
512
513#ifdef CONFIG_SMP
514 debugfs_create_file("tunable_scaling", 0644, debugfs_sched, NULL, &sched_scaling_fops);
515 debugfs_create_u32("migration_cost_ns", 0644, debugfs_sched, &sysctl_sched_migration_cost);
516 debugfs_create_u32("nr_migrate", 0644, debugfs_sched, &sysctl_sched_nr_migrate);
517
518 mutex_lock(&sched_domains_mutex);
519 update_sched_domain_debugfs();
520 mutex_unlock(&sched_domains_mutex);
521#endif
522
523#ifdef CONFIG_NUMA_BALANCING
524 numa = debugfs_create_dir("numa_balancing", debugfs_sched);
525
526 debugfs_create_u32("scan_delay_ms", 0644, numa, &sysctl_numa_balancing_scan_delay);
527 debugfs_create_u32("scan_period_min_ms", 0644, numa, &sysctl_numa_balancing_scan_period_min);
528 debugfs_create_u32("scan_period_max_ms", 0644, numa, &sysctl_numa_balancing_scan_period_max);
529 debugfs_create_u32("scan_size_mb", 0644, numa, &sysctl_numa_balancing_scan_size);
530 debugfs_create_u32("hot_threshold_ms", 0644, numa, &sysctl_numa_balancing_hot_threshold);
531#endif
532
533 debugfs_create_file("debug", 0444, debugfs_sched, NULL, &sched_debug_fops);
534
535 debugfs_fair_server_init();
536
537 return 0;
538}
539late_initcall(sched_init_debug);
540
541#ifdef CONFIG_SMP
542
543static cpumask_var_t sd_sysctl_cpus;
544
545static int sd_flags_show(struct seq_file *m, void *v)
546{
547 unsigned long flags = *(unsigned int *)m->private;
548 int idx;
549
550 for_each_set_bit(idx, &flags, __SD_FLAG_CNT) {
551 seq_puts(m, sd_flag_debug[idx].name);
552 seq_puts(m, " ");
553 }
554 seq_puts(m, "\n");
555
556 return 0;
557}
558
559static int sd_flags_open(struct inode *inode, struct file *file)
560{
561 return single_open(file, sd_flags_show, inode->i_private);
562}
563
564static const struct file_operations sd_flags_fops = {
565 .open = sd_flags_open,
566 .read = seq_read,
567 .llseek = seq_lseek,
568 .release = single_release,
569};
570
571static void register_sd(struct sched_domain *sd, struct dentry *parent)
572{
573#define SDM(type, mode, member) \
574 debugfs_create_##type(#member, mode, parent, &sd->member)
575
576 SDM(ulong, 0644, min_interval);
577 SDM(ulong, 0644, max_interval);
578 SDM(u64, 0644, max_newidle_lb_cost);
579 SDM(u32, 0644, busy_factor);
580 SDM(u32, 0644, imbalance_pct);
581 SDM(u32, 0644, cache_nice_tries);
582 SDM(str, 0444, name);
583
584#undef SDM
585
586 debugfs_create_file("flags", 0444, parent, &sd->flags, &sd_flags_fops);
587 debugfs_create_file("groups_flags", 0444, parent, &sd->groups->flags, &sd_flags_fops);
588 debugfs_create_u32("level", 0444, parent, (u32 *)&sd->level);
589}
590
591void update_sched_domain_debugfs(void)
592{
593 int cpu, i;
594
595 /*
596 * This can unfortunately be invoked before sched_debug_init() creates
597 * the debug directory. Don't touch sd_sysctl_cpus until then.
598 */
599 if (!debugfs_sched)
600 return;
601
602 if (!sched_debug_verbose)
603 return;
604
605 if (!cpumask_available(sd_sysctl_cpus)) {
606 if (!alloc_cpumask_var(&sd_sysctl_cpus, GFP_KERNEL))
607 return;
608 cpumask_copy(sd_sysctl_cpus, cpu_possible_mask);
609 }
610
611 if (!sd_dentry) {
612 sd_dentry = debugfs_create_dir("domains", debugfs_sched);
613
614 /* rebuild sd_sysctl_cpus if empty since it gets cleared below */
615 if (cpumask_empty(sd_sysctl_cpus))
616 cpumask_copy(sd_sysctl_cpus, cpu_online_mask);
617 }
618
619 for_each_cpu(cpu, sd_sysctl_cpus) {
620 struct sched_domain *sd;
621 struct dentry *d_cpu;
622 char buf[32];
623
624 snprintf(buf, sizeof(buf), "cpu%d", cpu);
625 debugfs_lookup_and_remove(buf, sd_dentry);
626 d_cpu = debugfs_create_dir(buf, sd_dentry);
627
628 i = 0;
629 for_each_domain(cpu, sd) {
630 struct dentry *d_sd;
631
632 snprintf(buf, sizeof(buf), "domain%d", i);
633 d_sd = debugfs_create_dir(buf, d_cpu);
634
635 register_sd(sd, d_sd);
636 i++;
637 }
638
639 __cpumask_clear_cpu(cpu, sd_sysctl_cpus);
640 }
641}
642
643void dirty_sched_domain_sysctl(int cpu)
644{
645 if (cpumask_available(sd_sysctl_cpus))
646 __cpumask_set_cpu(cpu, sd_sysctl_cpus);
647}
648
649#endif /* CONFIG_SMP */
650
651#ifdef CONFIG_FAIR_GROUP_SCHED
652static void print_cfs_group_stats(struct seq_file *m, int cpu, struct task_group *tg)
653{
654 struct sched_entity *se = tg->se[cpu];
655
656#define P(F) SEQ_printf(m, " .%-30s: %lld\n", #F, (long long)F)
657#define P_SCHEDSTAT(F) SEQ_printf(m, " .%-30s: %lld\n", \
658 #F, (long long)schedstat_val(stats->F))
659#define PN(F) SEQ_printf(m, " .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F))
660#define PN_SCHEDSTAT(F) SEQ_printf(m, " .%-30s: %lld.%06ld\n", \
661 #F, SPLIT_NS((long long)schedstat_val(stats->F)))
662
663 if (!se)
664 return;
665
666 PN(se->exec_start);
667 PN(se->vruntime);
668 PN(se->sum_exec_runtime);
669
670 if (schedstat_enabled()) {
671 struct sched_statistics *stats;
672 stats = __schedstats_from_se(se);
673
674 PN_SCHEDSTAT(wait_start);
675 PN_SCHEDSTAT(sleep_start);
676 PN_SCHEDSTAT(block_start);
677 PN_SCHEDSTAT(sleep_max);
678 PN_SCHEDSTAT(block_max);
679 PN_SCHEDSTAT(exec_max);
680 PN_SCHEDSTAT(slice_max);
681 PN_SCHEDSTAT(wait_max);
682 PN_SCHEDSTAT(wait_sum);
683 P_SCHEDSTAT(wait_count);
684 }
685
686 P(se->load.weight);
687#ifdef CONFIG_SMP
688 P(se->avg.load_avg);
689 P(se->avg.util_avg);
690 P(se->avg.runnable_avg);
691#endif
692
693#undef PN_SCHEDSTAT
694#undef PN
695#undef P_SCHEDSTAT
696#undef P
697}
698#endif
699
700#ifdef CONFIG_CGROUP_SCHED
701static DEFINE_SPINLOCK(sched_debug_lock);
702static char group_path[PATH_MAX];
703
704static void task_group_path(struct task_group *tg, char *path, int plen)
705{
706 if (autogroup_path(tg, path, plen))
707 return;
708
709 cgroup_path(tg->css.cgroup, path, plen);
710}
711
712/*
713 * Only 1 SEQ_printf_task_group_path() caller can use the full length
714 * group_path[] for cgroup path. Other simultaneous callers will have
715 * to use a shorter stack buffer. A "..." suffix is appended at the end
716 * of the stack buffer so that it will show up in case the output length
717 * matches the given buffer size to indicate possible path name truncation.
718 */
719#define SEQ_printf_task_group_path(m, tg, fmt...) \
720{ \
721 if (spin_trylock(&sched_debug_lock)) { \
722 task_group_path(tg, group_path, sizeof(group_path)); \
723 SEQ_printf(m, fmt, group_path); \
724 spin_unlock(&sched_debug_lock); \
725 } else { \
726 char buf[128]; \
727 char *bufend = buf + sizeof(buf) - 3; \
728 task_group_path(tg, buf, bufend - buf); \
729 strcpy(bufend - 1, "..."); \
730 SEQ_printf(m, fmt, buf); \
731 } \
732}
733#endif
734
735static void
736print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
737{
738 if (task_current(rq, p))
739 SEQ_printf(m, ">R");
740 else
741 SEQ_printf(m, " %c", task_state_to_char(p));
742
743 SEQ_printf(m, " %15s %5d %9Ld.%06ld %c %9Ld.%06ld %c %9Ld.%06ld %9Ld.%06ld %9Ld %5d ",
744 p->comm, task_pid_nr(p),
745 SPLIT_NS(p->se.vruntime),
746 entity_eligible(cfs_rq_of(&p->se), &p->se) ? 'E' : 'N',
747 SPLIT_NS(p->se.deadline),
748 p->se.custom_slice ? 'S' : ' ',
749 SPLIT_NS(p->se.slice),
750 SPLIT_NS(p->se.sum_exec_runtime),
751 (long long)(p->nvcsw + p->nivcsw),
752 p->prio);
753
754 SEQ_printf(m, "%9lld.%06ld %9lld.%06ld %9lld.%06ld",
755 SPLIT_NS(schedstat_val_or_zero(p->stats.wait_sum)),
756 SPLIT_NS(schedstat_val_or_zero(p->stats.sum_sleep_runtime)),
757 SPLIT_NS(schedstat_val_or_zero(p->stats.sum_block_runtime)));
758
759#ifdef CONFIG_NUMA_BALANCING
760 SEQ_printf(m, " %d %d", task_node(p), task_numa_group_id(p));
761#endif
762#ifdef CONFIG_CGROUP_SCHED
763 SEQ_printf_task_group_path(m, task_group(p), " %s")
764#endif
765
766 SEQ_printf(m, "\n");
767}
768
769static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
770{
771 struct task_struct *g, *p;
772
773 SEQ_printf(m, "\n");
774 SEQ_printf(m, "runnable tasks:\n");
775 SEQ_printf(m, " S task PID vruntime eligible "
776 "deadline slice sum-exec switches "
777 "prio wait-time sum-sleep sum-block"
778#ifdef CONFIG_NUMA_BALANCING
779 " node group-id"
780#endif
781#ifdef CONFIG_CGROUP_SCHED
782 " group-path"
783#endif
784 "\n");
785 SEQ_printf(m, "-------------------------------------------------------"
786 "------------------------------------------------------"
787 "------------------------------------------------------"
788#ifdef CONFIG_NUMA_BALANCING
789 "--------------"
790#endif
791#ifdef CONFIG_CGROUP_SCHED
792 "--------------"
793#endif
794 "\n");
795
796 rcu_read_lock();
797 for_each_process_thread(g, p) {
798 if (task_cpu(p) != rq_cpu)
799 continue;
800
801 print_task(m, rq, p);
802 }
803 rcu_read_unlock();
804}
805
806void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
807{
808 s64 left_vruntime = -1, min_vruntime, right_vruntime = -1, left_deadline = -1, spread;
809 struct sched_entity *last, *first, *root;
810 struct rq *rq = cpu_rq(cpu);
811 unsigned long flags;
812
813#ifdef CONFIG_FAIR_GROUP_SCHED
814 SEQ_printf(m, "\n");
815 SEQ_printf_task_group_path(m, cfs_rq->tg, "cfs_rq[%d]:%s\n", cpu);
816#else
817 SEQ_printf(m, "\n");
818 SEQ_printf(m, "cfs_rq[%d]:\n", cpu);
819#endif
820
821 raw_spin_rq_lock_irqsave(rq, flags);
822 root = __pick_root_entity(cfs_rq);
823 if (root)
824 left_vruntime = root->min_vruntime;
825 first = __pick_first_entity(cfs_rq);
826 if (first)
827 left_deadline = first->deadline;
828 last = __pick_last_entity(cfs_rq);
829 if (last)
830 right_vruntime = last->vruntime;
831 min_vruntime = cfs_rq->min_vruntime;
832 raw_spin_rq_unlock_irqrestore(rq, flags);
833
834 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "left_deadline",
835 SPLIT_NS(left_deadline));
836 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "left_vruntime",
837 SPLIT_NS(left_vruntime));
838 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "min_vruntime",
839 SPLIT_NS(min_vruntime));
840 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "avg_vruntime",
841 SPLIT_NS(avg_vruntime(cfs_rq)));
842 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "right_vruntime",
843 SPLIT_NS(right_vruntime));
844 spread = right_vruntime - left_vruntime;
845 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread", SPLIT_NS(spread));
846 SEQ_printf(m, " .%-30s: %d\n", "nr_running", cfs_rq->nr_running);
847 SEQ_printf(m, " .%-30s: %d\n", "h_nr_running", cfs_rq->h_nr_running);
848 SEQ_printf(m, " .%-30s: %d\n", "h_nr_delayed", cfs_rq->h_nr_delayed);
849 SEQ_printf(m, " .%-30s: %d\n", "idle_nr_running",
850 cfs_rq->idle_nr_running);
851 SEQ_printf(m, " .%-30s: %d\n", "idle_h_nr_running",
852 cfs_rq->idle_h_nr_running);
853 SEQ_printf(m, " .%-30s: %ld\n", "load", cfs_rq->load.weight);
854#ifdef CONFIG_SMP
855 SEQ_printf(m, " .%-30s: %lu\n", "load_avg",
856 cfs_rq->avg.load_avg);
857 SEQ_printf(m, " .%-30s: %lu\n", "runnable_avg",
858 cfs_rq->avg.runnable_avg);
859 SEQ_printf(m, " .%-30s: %lu\n", "util_avg",
860 cfs_rq->avg.util_avg);
861 SEQ_printf(m, " .%-30s: %u\n", "util_est",
862 cfs_rq->avg.util_est);
863 SEQ_printf(m, " .%-30s: %ld\n", "removed.load_avg",
864 cfs_rq->removed.load_avg);
865 SEQ_printf(m, " .%-30s: %ld\n", "removed.util_avg",
866 cfs_rq->removed.util_avg);
867 SEQ_printf(m, " .%-30s: %ld\n", "removed.runnable_avg",
868 cfs_rq->removed.runnable_avg);
869#ifdef CONFIG_FAIR_GROUP_SCHED
870 SEQ_printf(m, " .%-30s: %lu\n", "tg_load_avg_contrib",
871 cfs_rq->tg_load_avg_contrib);
872 SEQ_printf(m, " .%-30s: %ld\n", "tg_load_avg",
873 atomic_long_read(&cfs_rq->tg->load_avg));
874#endif
875#endif
876#ifdef CONFIG_CFS_BANDWIDTH
877 SEQ_printf(m, " .%-30s: %d\n", "throttled",
878 cfs_rq->throttled);
879 SEQ_printf(m, " .%-30s: %d\n", "throttle_count",
880 cfs_rq->throttle_count);
881#endif
882
883#ifdef CONFIG_FAIR_GROUP_SCHED
884 print_cfs_group_stats(m, cpu, cfs_rq->tg);
885#endif
886}
887
888void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
889{
890#ifdef CONFIG_RT_GROUP_SCHED
891 SEQ_printf(m, "\n");
892 SEQ_printf_task_group_path(m, rt_rq->tg, "rt_rq[%d]:%s\n", cpu);
893#else
894 SEQ_printf(m, "\n");
895 SEQ_printf(m, "rt_rq[%d]:\n", cpu);
896#endif
897
898#define P(x) \
899 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
900#define PU(x) \
901 SEQ_printf(m, " .%-30s: %lu\n", #x, (unsigned long)(rt_rq->x))
902#define PN(x) \
903 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
904
905 PU(rt_nr_running);
906
907#ifdef CONFIG_RT_GROUP_SCHED
908 P(rt_throttled);
909 PN(rt_time);
910 PN(rt_runtime);
911#endif
912
913#undef PN
914#undef PU
915#undef P
916}
917
918void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq)
919{
920 struct dl_bw *dl_bw;
921
922 SEQ_printf(m, "\n");
923 SEQ_printf(m, "dl_rq[%d]:\n", cpu);
924
925#define PU(x) \
926 SEQ_printf(m, " .%-30s: %lu\n", #x, (unsigned long)(dl_rq->x))
927
928 PU(dl_nr_running);
929#ifdef CONFIG_SMP
930 dl_bw = &cpu_rq(cpu)->rd->dl_bw;
931#else
932 dl_bw = &dl_rq->dl_bw;
933#endif
934 SEQ_printf(m, " .%-30s: %lld\n", "dl_bw->bw", dl_bw->bw);
935 SEQ_printf(m, " .%-30s: %lld\n", "dl_bw->total_bw", dl_bw->total_bw);
936
937#undef PU
938}
939
940static void print_cpu(struct seq_file *m, int cpu)
941{
942 struct rq *rq = cpu_rq(cpu);
943
944#ifdef CONFIG_X86
945 {
946 unsigned int freq = cpu_khz ? : 1;
947
948 SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
949 cpu, freq / 1000, (freq % 1000));
950 }
951#else
952 SEQ_printf(m, "cpu#%d\n", cpu);
953#endif
954
955#define P(x) \
956do { \
957 if (sizeof(rq->x) == 4) \
958 SEQ_printf(m, " .%-30s: %d\n", #x, (int)(rq->x)); \
959 else \
960 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rq->x));\
961} while (0)
962
963#define PN(x) \
964 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
965
966 P(nr_running);
967 P(nr_switches);
968 P(nr_uninterruptible);
969 PN(next_balance);
970 SEQ_printf(m, " .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr)));
971 PN(clock);
972 PN(clock_task);
973#undef P
974#undef PN
975
976#ifdef CONFIG_SMP
977#define P64(n) SEQ_printf(m, " .%-30s: %Ld\n", #n, rq->n);
978 P64(avg_idle);
979 P64(max_idle_balance_cost);
980#undef P64
981#endif
982
983#define P(n) SEQ_printf(m, " .%-30s: %d\n", #n, schedstat_val(rq->n));
984 if (schedstat_enabled()) {
985 P(yld_count);
986 P(sched_count);
987 P(sched_goidle);
988 P(ttwu_count);
989 P(ttwu_local);
990 }
991#undef P
992
993 print_cfs_stats(m, cpu);
994 print_rt_stats(m, cpu);
995 print_dl_stats(m, cpu);
996
997 print_rq(m, rq, cpu);
998 SEQ_printf(m, "\n");
999}
1000
1001static const char *sched_tunable_scaling_names[] = {
1002 "none",
1003 "logarithmic",
1004 "linear"
1005};
1006
1007static void sched_debug_header(struct seq_file *m)
1008{
1009 u64 ktime, sched_clk, cpu_clk;
1010 unsigned long flags;
1011
1012 local_irq_save(flags);
1013 ktime = ktime_to_ns(ktime_get());
1014 sched_clk = sched_clock();
1015 cpu_clk = local_clock();
1016 local_irq_restore(flags);
1017
1018 SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n",
1019 init_utsname()->release,
1020 (int)strcspn(init_utsname()->version, " "),
1021 init_utsname()->version);
1022
1023#define P(x) \
1024 SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
1025#define PN(x) \
1026 SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
1027 PN(ktime);
1028 PN(sched_clk);
1029 PN(cpu_clk);
1030 P(jiffies);
1031#ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
1032 P(sched_clock_stable());
1033#endif
1034#undef PN
1035#undef P
1036
1037 SEQ_printf(m, "\n");
1038 SEQ_printf(m, "sysctl_sched\n");
1039
1040#define P(x) \
1041 SEQ_printf(m, " .%-40s: %Ld\n", #x, (long long)(x))
1042#define PN(x) \
1043 SEQ_printf(m, " .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
1044 PN(sysctl_sched_base_slice);
1045 P(sysctl_sched_features);
1046#undef PN
1047#undef P
1048
1049 SEQ_printf(m, " .%-40s: %d (%s)\n",
1050 "sysctl_sched_tunable_scaling",
1051 sysctl_sched_tunable_scaling,
1052 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
1053 SEQ_printf(m, "\n");
1054}
1055
1056static int sched_debug_show(struct seq_file *m, void *v)
1057{
1058 int cpu = (unsigned long)(v - 2);
1059
1060 if (cpu != -1)
1061 print_cpu(m, cpu);
1062 else
1063 sched_debug_header(m);
1064
1065 return 0;
1066}
1067
1068void sysrq_sched_debug_show(void)
1069{
1070 int cpu;
1071
1072 sched_debug_header(NULL);
1073 for_each_online_cpu(cpu) {
1074 /*
1075 * Need to reset softlockup watchdogs on all CPUs, because
1076 * another CPU might be blocked waiting for us to process
1077 * an IPI or stop_machine.
1078 */
1079 touch_nmi_watchdog();
1080 touch_all_softlockup_watchdogs();
1081 print_cpu(NULL, cpu);
1082 }
1083}
1084
1085/*
1086 * This iterator needs some explanation.
1087 * It returns 1 for the header position.
1088 * This means 2 is CPU 0.
1089 * In a hotplugged system some CPUs, including CPU 0, may be missing so we have
1090 * to use cpumask_* to iterate over the CPUs.
1091 */
1092static void *sched_debug_start(struct seq_file *file, loff_t *offset)
1093{
1094 unsigned long n = *offset;
1095
1096 if (n == 0)
1097 return (void *) 1;
1098
1099 n--;
1100
1101 if (n > 0)
1102 n = cpumask_next(n - 1, cpu_online_mask);
1103 else
1104 n = cpumask_first(cpu_online_mask);
1105
1106 *offset = n + 1;
1107
1108 if (n < nr_cpu_ids)
1109 return (void *)(unsigned long)(n + 2);
1110
1111 return NULL;
1112}
1113
1114static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
1115{
1116 (*offset)++;
1117 return sched_debug_start(file, offset);
1118}
1119
1120static void sched_debug_stop(struct seq_file *file, void *data)
1121{
1122}
1123
1124static const struct seq_operations sched_debug_sops = {
1125 .start = sched_debug_start,
1126 .next = sched_debug_next,
1127 .stop = sched_debug_stop,
1128 .show = sched_debug_show,
1129};
1130
1131#define __PS(S, F) SEQ_printf(m, "%-45s:%21Ld\n", S, (long long)(F))
1132#define __P(F) __PS(#F, F)
1133#define P(F) __PS(#F, p->F)
1134#define PM(F, M) __PS(#F, p->F & (M))
1135#define __PSN(S, F) SEQ_printf(m, "%-45s:%14Ld.%06ld\n", S, SPLIT_NS((long long)(F)))
1136#define __PN(F) __PSN(#F, F)
1137#define PN(F) __PSN(#F, p->F)
1138
1139
1140#ifdef CONFIG_NUMA_BALANCING
1141void print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
1142 unsigned long tpf, unsigned long gsf, unsigned long gpf)
1143{
1144 SEQ_printf(m, "numa_faults node=%d ", node);
1145 SEQ_printf(m, "task_private=%lu task_shared=%lu ", tpf, tsf);
1146 SEQ_printf(m, "group_private=%lu group_shared=%lu\n", gpf, gsf);
1147}
1148#endif
1149
1150
1151static void sched_show_numa(struct task_struct *p, struct seq_file *m)
1152{
1153#ifdef CONFIG_NUMA_BALANCING
1154 if (p->mm)
1155 P(mm->numa_scan_seq);
1156
1157 P(numa_pages_migrated);
1158 P(numa_preferred_nid);
1159 P(total_numa_faults);
1160 SEQ_printf(m, "current_node=%d, numa_group_id=%d\n",
1161 task_node(p), task_numa_group_id(p));
1162 show_numa_stats(p, m);
1163#endif
1164}
1165
1166void proc_sched_show_task(struct task_struct *p, struct pid_namespace *ns,
1167 struct seq_file *m)
1168{
1169 unsigned long nr_switches;
1170
1171 SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr_ns(p, ns),
1172 get_nr_threads(p));
1173 SEQ_printf(m,
1174 "---------------------------------------------------------"
1175 "----------\n");
1176
1177#define P_SCHEDSTAT(F) __PS(#F, schedstat_val(p->stats.F))
1178#define PN_SCHEDSTAT(F) __PSN(#F, schedstat_val(p->stats.F))
1179
1180 PN(se.exec_start);
1181 PN(se.vruntime);
1182 PN(se.sum_exec_runtime);
1183
1184 nr_switches = p->nvcsw + p->nivcsw;
1185
1186 P(se.nr_migrations);
1187
1188 if (schedstat_enabled()) {
1189 u64 avg_atom, avg_per_cpu;
1190
1191 PN_SCHEDSTAT(sum_sleep_runtime);
1192 PN_SCHEDSTAT(sum_block_runtime);
1193 PN_SCHEDSTAT(wait_start);
1194 PN_SCHEDSTAT(sleep_start);
1195 PN_SCHEDSTAT(block_start);
1196 PN_SCHEDSTAT(sleep_max);
1197 PN_SCHEDSTAT(block_max);
1198 PN_SCHEDSTAT(exec_max);
1199 PN_SCHEDSTAT(slice_max);
1200 PN_SCHEDSTAT(wait_max);
1201 PN_SCHEDSTAT(wait_sum);
1202 P_SCHEDSTAT(wait_count);
1203 PN_SCHEDSTAT(iowait_sum);
1204 P_SCHEDSTAT(iowait_count);
1205 P_SCHEDSTAT(nr_migrations_cold);
1206 P_SCHEDSTAT(nr_failed_migrations_affine);
1207 P_SCHEDSTAT(nr_failed_migrations_running);
1208 P_SCHEDSTAT(nr_failed_migrations_hot);
1209 P_SCHEDSTAT(nr_forced_migrations);
1210 P_SCHEDSTAT(nr_wakeups);
1211 P_SCHEDSTAT(nr_wakeups_sync);
1212 P_SCHEDSTAT(nr_wakeups_migrate);
1213 P_SCHEDSTAT(nr_wakeups_local);
1214 P_SCHEDSTAT(nr_wakeups_remote);
1215 P_SCHEDSTAT(nr_wakeups_affine);
1216 P_SCHEDSTAT(nr_wakeups_affine_attempts);
1217 P_SCHEDSTAT(nr_wakeups_passive);
1218 P_SCHEDSTAT(nr_wakeups_idle);
1219
1220 avg_atom = p->se.sum_exec_runtime;
1221 if (nr_switches)
1222 avg_atom = div64_ul(avg_atom, nr_switches);
1223 else
1224 avg_atom = -1LL;
1225
1226 avg_per_cpu = p->se.sum_exec_runtime;
1227 if (p->se.nr_migrations) {
1228 avg_per_cpu = div64_u64(avg_per_cpu,
1229 p->se.nr_migrations);
1230 } else {
1231 avg_per_cpu = -1LL;
1232 }
1233
1234 __PN(avg_atom);
1235 __PN(avg_per_cpu);
1236
1237#ifdef CONFIG_SCHED_CORE
1238 PN_SCHEDSTAT(core_forceidle_sum);
1239#endif
1240 }
1241
1242 __P(nr_switches);
1243 __PS("nr_voluntary_switches", p->nvcsw);
1244 __PS("nr_involuntary_switches", p->nivcsw);
1245
1246 P(se.load.weight);
1247#ifdef CONFIG_SMP
1248 P(se.avg.load_sum);
1249 P(se.avg.runnable_sum);
1250 P(se.avg.util_sum);
1251 P(se.avg.load_avg);
1252 P(se.avg.runnable_avg);
1253 P(se.avg.util_avg);
1254 P(se.avg.last_update_time);
1255 PM(se.avg.util_est, ~UTIL_AVG_UNCHANGED);
1256#endif
1257#ifdef CONFIG_UCLAMP_TASK
1258 __PS("uclamp.min", p->uclamp_req[UCLAMP_MIN].value);
1259 __PS("uclamp.max", p->uclamp_req[UCLAMP_MAX].value);
1260 __PS("effective uclamp.min", uclamp_eff_value(p, UCLAMP_MIN));
1261 __PS("effective uclamp.max", uclamp_eff_value(p, UCLAMP_MAX));
1262#endif
1263 P(policy);
1264 P(prio);
1265 if (task_has_dl_policy(p)) {
1266 P(dl.runtime);
1267 P(dl.deadline);
1268 }
1269#ifdef CONFIG_SCHED_CLASS_EXT
1270 __PS("ext.enabled", task_on_scx(p));
1271#endif
1272#undef PN_SCHEDSTAT
1273#undef P_SCHEDSTAT
1274
1275 {
1276 unsigned int this_cpu = raw_smp_processor_id();
1277 u64 t0, t1;
1278
1279 t0 = cpu_clock(this_cpu);
1280 t1 = cpu_clock(this_cpu);
1281 __PS("clock-delta", t1-t0);
1282 }
1283
1284 sched_show_numa(p, m);
1285}
1286
1287void proc_sched_set_task(struct task_struct *p)
1288{
1289#ifdef CONFIG_SCHEDSTATS
1290 memset(&p->stats, 0, sizeof(p->stats));
1291#endif
1292}
1293
1294void resched_latency_warn(int cpu, u64 latency)
1295{
1296 static DEFINE_RATELIMIT_STATE(latency_check_ratelimit, 60 * 60 * HZ, 1);
1297
1298 WARN(__ratelimit(&latency_check_ratelimit),
1299 "sched: CPU %d need_resched set for > %llu ns (%d ticks) "
1300 "without schedule\n",
1301 cpu, latency, cpu_rq(cpu)->ticks_without_resched);
1302}
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, "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("base_slice_ns", 0644, debugfs_sched, &sysctl_sched_base_slice);
351
352 debugfs_create_u32("latency_warn_ms", 0644, debugfs_sched, &sysctl_resched_latency_warn_ms);
353 debugfs_create_u32("latency_warn_once", 0644, debugfs_sched, &sysctl_resched_latency_warn_once);
354
355#ifdef CONFIG_SMP
356 debugfs_create_file("tunable_scaling", 0644, debugfs_sched, NULL, &sched_scaling_fops);
357 debugfs_create_u32("migration_cost_ns", 0644, debugfs_sched, &sysctl_sched_migration_cost);
358 debugfs_create_u32("nr_migrate", 0644, debugfs_sched, &sysctl_sched_nr_migrate);
359
360 mutex_lock(&sched_domains_mutex);
361 update_sched_domain_debugfs();
362 mutex_unlock(&sched_domains_mutex);
363#endif
364
365#ifdef CONFIG_NUMA_BALANCING
366 numa = debugfs_create_dir("numa_balancing", debugfs_sched);
367
368 debugfs_create_u32("scan_delay_ms", 0644, numa, &sysctl_numa_balancing_scan_delay);
369 debugfs_create_u32("scan_period_min_ms", 0644, numa, &sysctl_numa_balancing_scan_period_min);
370 debugfs_create_u32("scan_period_max_ms", 0644, numa, &sysctl_numa_balancing_scan_period_max);
371 debugfs_create_u32("scan_size_mb", 0644, numa, &sysctl_numa_balancing_scan_size);
372 debugfs_create_u32("hot_threshold_ms", 0644, numa, &sysctl_numa_balancing_hot_threshold);
373#endif
374
375 debugfs_create_file("debug", 0444, debugfs_sched, NULL, &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, sd_flag_debug[idx].name);
392 seq_puts(m, " ");
393 }
394 seq_puts(m, "\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("flags", 0444, parent, &sd->flags, &sd_flags_fops);
427 debugfs_create_file("groups_flags", 0444, parent, &sd->groups->flags, &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(sd_sysctl_cpus)) {
445 if (!alloc_cpumask_var(&sd_sysctl_cpus, GFP_KERNEL))
446 return;
447 cpumask_copy(sd_sysctl_cpus, cpu_possible_mask);
448 }
449
450 if (!sd_dentry) {
451 sd_dentry = debugfs_create_dir("domains", debugfs_sched);
452
453 /* rebuild sd_sysctl_cpus if empty since it gets cleared below */
454 if (cpumask_empty(sd_sysctl_cpus))
455 cpumask_copy(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, sizeof(buf), "cpu%d", cpu);
464 debugfs_lookup_and_remove(buf, sd_dentry);
465 d_cpu = debugfs_create_dir(buf, sd_dentry);
466
467 i = 0;
468 for_each_domain(cpu, sd) {
469 struct dentry *d_sd;
470
471 snprintf(buf, sizeof(buf), "domain%d", i);
472 d_sd = debugfs_create_dir(buf, d_cpu);
473
474 register_sd(sd, d_sd);
475 i++;
476 }
477
478 __cpumask_clear_cpu(cpu, sd_sysctl_cpus);
479 }
480}
481
482void dirty_sched_domain_sysctl(int cpu)
483{
484 if (cpumask_available(sd_sysctl_cpus))
485 __cpumask_set_cpu(cpu, 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, path, plen))
546 return;
547
548 cgroup_path(tg->css.cgroup, path, 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, left_deadline = -1, spread;
632 struct sched_entity *last, *first, *root;
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 root = __pick_root_entity(cfs_rq);
648 if (root)
649 left_vruntime = root->min_vruntime;
650 first = __pick_first_entity(cfs_rq);
651 if (first)
652 left_deadline = first->deadline;
653 last = __pick_last_entity(cfs_rq);
654 if (last)
655 right_vruntime = last->vruntime;
656 min_vruntime = cfs_rq->min_vruntime;
657 raw_spin_rq_unlock_irqrestore(rq, flags);
658
659 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "left_deadline",
660 SPLIT_NS(left_deadline));
661 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "left_vruntime",
662 SPLIT_NS(left_vruntime));
663 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "min_vruntime",
664 SPLIT_NS(min_vruntime));
665 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "avg_vruntime",
666 SPLIT_NS(avg_vruntime(cfs_rq)));
667 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "right_vruntime",
668 SPLIT_NS(right_vruntime));
669 spread = right_vruntime - left_vruntime;
670 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread", SPLIT_NS(spread));
671 SEQ_printf(m, " .%-30s: %d\n", "nr_spread_over",
672 cfs_rq->nr_spread_over);
673 SEQ_printf(m, " .%-30s: %d\n", "nr_running", cfs_rq->nr_running);
674 SEQ_printf(m, " .%-30s: %d\n", "h_nr_running", cfs_rq->h_nr_running);
675 SEQ_printf(m, " .%-30s: %d\n", "idle_nr_running",
676 cfs_rq->idle_nr_running);
677 SEQ_printf(m, " .%-30s: %d\n", "idle_h_nr_running",
678 cfs_rq->idle_h_nr_running);
679 SEQ_printf(m, " .%-30s: %ld\n", "load", cfs_rq->load.weight);
680#ifdef CONFIG_SMP
681 SEQ_printf(m, " .%-30s: %lu\n", "load_avg",
682 cfs_rq->avg.load_avg);
683 SEQ_printf(m, " .%-30s: %lu\n", "runnable_avg",
684 cfs_rq->avg.runnable_avg);
685 SEQ_printf(m, " .%-30s: %lu\n", "util_avg",
686 cfs_rq->avg.util_avg);
687 SEQ_printf(m, " .%-30s: %u\n", "util_est",
688 cfs_rq->avg.util_est);
689 SEQ_printf(m, " .%-30s: %ld\n", "removed.load_avg",
690 cfs_rq->removed.load_avg);
691 SEQ_printf(m, " .%-30s: %ld\n", "removed.util_avg",
692 cfs_rq->removed.util_avg);
693 SEQ_printf(m, " .%-30s: %ld\n", "removed.runnable_avg",
694 cfs_rq->removed.runnable_avg);
695#ifdef CONFIG_FAIR_GROUP_SCHED
696 SEQ_printf(m, " .%-30s: %lu\n", "tg_load_avg_contrib",
697 cfs_rq->tg_load_avg_contrib);
698 SEQ_printf(m, " .%-30s: %ld\n", "tg_load_avg",
699 atomic_long_read(&cfs_rq->tg->load_avg));
700#endif
701#endif
702#ifdef CONFIG_CFS_BANDWIDTH
703 SEQ_printf(m, " .%-30s: %d\n", "throttled",
704 cfs_rq->throttled);
705 SEQ_printf(m, " .%-30s: %d\n", "throttle_count",
706 cfs_rq->throttle_count);
707#endif
708
709#ifdef CONFIG_FAIR_GROUP_SCHED
710 print_cfs_group_stats(m, cpu, cfs_rq->tg);
711#endif
712}
713
714void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
715{
716#ifdef CONFIG_RT_GROUP_SCHED
717 SEQ_printf(m, "\n");
718 SEQ_printf_task_group_path(m, rt_rq->tg, "rt_rq[%d]:%s\n", cpu);
719#else
720 SEQ_printf(m, "\n");
721 SEQ_printf(m, "rt_rq[%d]:\n", cpu);
722#endif
723
724#define P(x) \
725 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
726#define PU(x) \
727 SEQ_printf(m, " .%-30s: %lu\n", #x, (unsigned long)(rt_rq->x))
728#define PN(x) \
729 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
730
731 PU(rt_nr_running);
732 P(rt_throttled);
733 PN(rt_time);
734 PN(rt_runtime);
735
736#undef PN
737#undef PU
738#undef P
739}
740
741void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq)
742{
743 struct dl_bw *dl_bw;
744
745 SEQ_printf(m, "\n");
746 SEQ_printf(m, "dl_rq[%d]:\n", cpu);
747
748#define PU(x) \
749 SEQ_printf(m, " .%-30s: %lu\n", #x, (unsigned long)(dl_rq->x))
750
751 PU(dl_nr_running);
752#ifdef CONFIG_SMP
753 dl_bw = &cpu_rq(cpu)->rd->dl_bw;
754#else
755 dl_bw = &dl_rq->dl_bw;
756#endif
757 SEQ_printf(m, " .%-30s: %lld\n", "dl_bw->bw", dl_bw->bw);
758 SEQ_printf(m, " .%-30s: %lld\n", "dl_bw->total_bw", dl_bw->total_bw);
759
760#undef PU
761}
762
763static void print_cpu(struct seq_file *m, int cpu)
764{
765 struct rq *rq = cpu_rq(cpu);
766
767#ifdef CONFIG_X86
768 {
769 unsigned int freq = cpu_khz ? : 1;
770
771 SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
772 cpu, freq / 1000, (freq % 1000));
773 }
774#else
775 SEQ_printf(m, "cpu#%d\n", cpu);
776#endif
777
778#define P(x) \
779do { \
780 if (sizeof(rq->x) == 4) \
781 SEQ_printf(m, " .%-30s: %d\n", #x, (int)(rq->x)); \
782 else \
783 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rq->x));\
784} while (0)
785
786#define PN(x) \
787 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
788
789 P(nr_running);
790 P(nr_switches);
791 P(nr_uninterruptible);
792 PN(next_balance);
793 SEQ_printf(m, " .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr)));
794 PN(clock);
795 PN(clock_task);
796#undef P
797#undef PN
798
799#ifdef CONFIG_SMP
800#define P64(n) SEQ_printf(m, " .%-30s: %Ld\n", #n, rq->n);
801 P64(avg_idle);
802 P64(max_idle_balance_cost);
803#undef P64
804#endif
805
806#define P(n) SEQ_printf(m, " .%-30s: %d\n", #n, schedstat_val(rq->n));
807 if (schedstat_enabled()) {
808 P(yld_count);
809 P(sched_count);
810 P(sched_goidle);
811 P(ttwu_count);
812 P(ttwu_local);
813 }
814#undef P
815
816 print_cfs_stats(m, cpu);
817 print_rt_stats(m, cpu);
818 print_dl_stats(m, cpu);
819
820 print_rq(m, rq, cpu);
821 SEQ_printf(m, "\n");
822}
823
824static const char *sched_tunable_scaling_names[] = {
825 "none",
826 "logarithmic",
827 "linear"
828};
829
830static void sched_debug_header(struct seq_file *m)
831{
832 u64 ktime, sched_clk, cpu_clk;
833 unsigned long flags;
834
835 local_irq_save(flags);
836 ktime = ktime_to_ns(ktime_get());
837 sched_clk = sched_clock();
838 cpu_clk = local_clock();
839 local_irq_restore(flags);
840
841 SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n",
842 init_utsname()->release,
843 (int)strcspn(init_utsname()->version, " "),
844 init_utsname()->version);
845
846#define P(x) \
847 SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
848#define PN(x) \
849 SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
850 PN(ktime);
851 PN(sched_clk);
852 PN(cpu_clk);
853 P(jiffies);
854#ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
855 P(sched_clock_stable());
856#endif
857#undef PN
858#undef P
859
860 SEQ_printf(m, "\n");
861 SEQ_printf(m, "sysctl_sched\n");
862
863#define P(x) \
864 SEQ_printf(m, " .%-40s: %Ld\n", #x, (long long)(x))
865#define PN(x) \
866 SEQ_printf(m, " .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
867 PN(sysctl_sched_base_slice);
868 P(sysctl_sched_features);
869#undef PN
870#undef P
871
872 SEQ_printf(m, " .%-40s: %d (%s)\n",
873 "sysctl_sched_tunable_scaling",
874 sysctl_sched_tunable_scaling,
875 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
876 SEQ_printf(m, "\n");
877}
878
879static int sched_debug_show(struct seq_file *m, void *v)
880{
881 int cpu = (unsigned long)(v - 2);
882
883 if (cpu != -1)
884 print_cpu(m, cpu);
885 else
886 sched_debug_header(m);
887
888 return 0;
889}
890
891void sysrq_sched_debug_show(void)
892{
893 int cpu;
894
895 sched_debug_header(NULL);
896 for_each_online_cpu(cpu) {
897 /*
898 * Need to reset softlockup watchdogs on all CPUs, because
899 * another CPU might be blocked waiting for us to process
900 * an IPI or stop_machine.
901 */
902 touch_nmi_watchdog();
903 touch_all_softlockup_watchdogs();
904 print_cpu(NULL, cpu);
905 }
906}
907
908/*
909 * This iterator needs some explanation.
910 * It returns 1 for the header position.
911 * This means 2 is CPU 0.
912 * In a hotplugged system some CPUs, including CPU 0, may be missing so we have
913 * to use cpumask_* to iterate over the CPUs.
914 */
915static void *sched_debug_start(struct seq_file *file, loff_t *offset)
916{
917 unsigned long n = *offset;
918
919 if (n == 0)
920 return (void *) 1;
921
922 n--;
923
924 if (n > 0)
925 n = cpumask_next(n - 1, cpu_online_mask);
926 else
927 n = cpumask_first(cpu_online_mask);
928
929 *offset = n + 1;
930
931 if (n < nr_cpu_ids)
932 return (void *)(unsigned long)(n + 2);
933
934 return NULL;
935}
936
937static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
938{
939 (*offset)++;
940 return sched_debug_start(file, offset);
941}
942
943static void sched_debug_stop(struct seq_file *file, void *data)
944{
945}
946
947static const struct seq_operations sched_debug_sops = {
948 .start = sched_debug_start,
949 .next = sched_debug_next,
950 .stop = sched_debug_stop,
951 .show = sched_debug_show,
952};
953
954#define __PS(S, F) SEQ_printf(m, "%-45s:%21Ld\n", S, (long long)(F))
955#define __P(F) __PS(#F, F)
956#define P(F) __PS(#F, p->F)
957#define PM(F, M) __PS(#F, p->F & (M))
958#define __PSN(S, F) SEQ_printf(m, "%-45s:%14Ld.%06ld\n", S, SPLIT_NS((long long)(F)))
959#define __PN(F) __PSN(#F, F)
960#define PN(F) __PSN(#F, p->F)
961
962
963#ifdef CONFIG_NUMA_BALANCING
964void print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
965 unsigned long tpf, unsigned long gsf, unsigned long gpf)
966{
967 SEQ_printf(m, "numa_faults node=%d ", node);
968 SEQ_printf(m, "task_private=%lu task_shared=%lu ", tpf, tsf);
969 SEQ_printf(m, "group_private=%lu group_shared=%lu\n", gpf, gsf);
970}
971#endif
972
973
974static void sched_show_numa(struct task_struct *p, struct seq_file *m)
975{
976#ifdef CONFIG_NUMA_BALANCING
977 if (p->mm)
978 P(mm->numa_scan_seq);
979
980 P(numa_pages_migrated);
981 P(numa_preferred_nid);
982 P(total_numa_faults);
983 SEQ_printf(m, "current_node=%d, numa_group_id=%d\n",
984 task_node(p), task_numa_group_id(p));
985 show_numa_stats(p, m);
986#endif
987}
988
989void proc_sched_show_task(struct task_struct *p, struct pid_namespace *ns,
990 struct seq_file *m)
991{
992 unsigned long nr_switches;
993
994 SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr_ns(p, ns),
995 get_nr_threads(p));
996 SEQ_printf(m,
997 "---------------------------------------------------------"
998 "----------\n");
999
1000#define P_SCHEDSTAT(F) __PS(#F, schedstat_val(p->stats.F))
1001#define PN_SCHEDSTAT(F) __PSN(#F, schedstat_val(p->stats.F))
1002
1003 PN(se.exec_start);
1004 PN(se.vruntime);
1005 PN(se.sum_exec_runtime);
1006
1007 nr_switches = p->nvcsw + p->nivcsw;
1008
1009 P(se.nr_migrations);
1010
1011 if (schedstat_enabled()) {
1012 u64 avg_atom, avg_per_cpu;
1013
1014 PN_SCHEDSTAT(sum_sleep_runtime);
1015 PN_SCHEDSTAT(sum_block_runtime);
1016 PN_SCHEDSTAT(wait_start);
1017 PN_SCHEDSTAT(sleep_start);
1018 PN_SCHEDSTAT(block_start);
1019 PN_SCHEDSTAT(sleep_max);
1020 PN_SCHEDSTAT(block_max);
1021 PN_SCHEDSTAT(exec_max);
1022 PN_SCHEDSTAT(slice_max);
1023 PN_SCHEDSTAT(wait_max);
1024 PN_SCHEDSTAT(wait_sum);
1025 P_SCHEDSTAT(wait_count);
1026 PN_SCHEDSTAT(iowait_sum);
1027 P_SCHEDSTAT(iowait_count);
1028 P_SCHEDSTAT(nr_migrations_cold);
1029 P_SCHEDSTAT(nr_failed_migrations_affine);
1030 P_SCHEDSTAT(nr_failed_migrations_running);
1031 P_SCHEDSTAT(nr_failed_migrations_hot);
1032 P_SCHEDSTAT(nr_forced_migrations);
1033 P_SCHEDSTAT(nr_wakeups);
1034 P_SCHEDSTAT(nr_wakeups_sync);
1035 P_SCHEDSTAT(nr_wakeups_migrate);
1036 P_SCHEDSTAT(nr_wakeups_local);
1037 P_SCHEDSTAT(nr_wakeups_remote);
1038 P_SCHEDSTAT(nr_wakeups_affine);
1039 P_SCHEDSTAT(nr_wakeups_affine_attempts);
1040 P_SCHEDSTAT(nr_wakeups_passive);
1041 P_SCHEDSTAT(nr_wakeups_idle);
1042
1043 avg_atom = p->se.sum_exec_runtime;
1044 if (nr_switches)
1045 avg_atom = div64_ul(avg_atom, nr_switches);
1046 else
1047 avg_atom = -1LL;
1048
1049 avg_per_cpu = p->se.sum_exec_runtime;
1050 if (p->se.nr_migrations) {
1051 avg_per_cpu = div64_u64(avg_per_cpu,
1052 p->se.nr_migrations);
1053 } else {
1054 avg_per_cpu = -1LL;
1055 }
1056
1057 __PN(avg_atom);
1058 __PN(avg_per_cpu);
1059
1060#ifdef CONFIG_SCHED_CORE
1061 PN_SCHEDSTAT(core_forceidle_sum);
1062#endif
1063 }
1064
1065 __P(nr_switches);
1066 __PS("nr_voluntary_switches", p->nvcsw);
1067 __PS("nr_involuntary_switches", p->nivcsw);
1068
1069 P(se.load.weight);
1070#ifdef CONFIG_SMP
1071 P(se.avg.load_sum);
1072 P(se.avg.runnable_sum);
1073 P(se.avg.util_sum);
1074 P(se.avg.load_avg);
1075 P(se.avg.runnable_avg);
1076 P(se.avg.util_avg);
1077 P(se.avg.last_update_time);
1078 PM(se.avg.util_est, ~UTIL_AVG_UNCHANGED);
1079#endif
1080#ifdef CONFIG_UCLAMP_TASK
1081 __PS("uclamp.min", p->uclamp_req[UCLAMP_MIN].value);
1082 __PS("uclamp.max", p->uclamp_req[UCLAMP_MAX].value);
1083 __PS("effective uclamp.min", uclamp_eff_value(p, UCLAMP_MIN));
1084 __PS("effective uclamp.max", uclamp_eff_value(p, UCLAMP_MAX));
1085#endif
1086 P(policy);
1087 P(prio);
1088 if (task_has_dl_policy(p)) {
1089 P(dl.runtime);
1090 P(dl.deadline);
1091 }
1092#undef PN_SCHEDSTAT
1093#undef P_SCHEDSTAT
1094
1095 {
1096 unsigned int this_cpu = raw_smp_processor_id();
1097 u64 t0, t1;
1098
1099 t0 = cpu_clock(this_cpu);
1100 t1 = cpu_clock(this_cpu);
1101 __PS("clock-delta", t1-t0);
1102 }
1103
1104 sched_show_numa(p, m);
1105}
1106
1107void proc_sched_set_task(struct task_struct *p)
1108{
1109#ifdef CONFIG_SCHEDSTATS
1110 memset(&p->stats, 0, sizeof(p->stats));
1111#endif
1112}
1113
1114void resched_latency_warn(int cpu, u64 latency)
1115{
1116 static DEFINE_RATELIMIT_STATE(latency_check_ratelimit, 60 * 60 * HZ, 1);
1117
1118 WARN(__ratelimit(&latency_check_ratelimit),
1119 "sched: CPU %d need_resched set for > %llu ns (%d ticks) "
1120 "without schedule\n",
1121 cpu, latency, cpu_rq(cpu)->ticks_without_resched);
1122}