1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * kernel/sched/debug.c
4 *
5 * Print the CFS rbtree and other debugging details
6 *
7 * Copyright(C) 2007, Red Hat, Inc., Ingo Molnar
8 */
9#include <linux/debugfs.h>
10#include <linux/nmi.h>
11#include "sched.h"
12
13/*
14 * This allows printing both to /sys/kernel/debug/sched/debug and
15 * to the console
16 */
17#define SEQ_printf(m, x...) \
18 do { \
19 if (m) \
20 seq_printf(m, x); \
21 else \
22 pr_cont(x); \
23 } while (0)
24
25/*
26 * Ease the printing of nsec fields:
27 */
28static long long nsec_high(unsigned long long nsec)
29{
30 if ((long long)nsec < 0) {
31 nsec = -nsec;
32 do_div(nsec, 1000000);
33 return -nsec;
34 }
35 do_div(nsec, 1000000);
36
37 return nsec;
38}
39
40static unsigned long nsec_low(unsigned long long nsec)
41{
42 if ((long long)nsec < 0)
43 nsec = -nsec;
44
45 return do_div(nsec, 1000000);
46}
47
48#define SPLIT_NS(x) nsec_high(x), nsec_low(x)
49
50#define SCHED_FEAT(name, enabled) \
51 #name ,
52
53static const char * const sched_feat_names[] = {
54#include "features.h"
55};
56
57#undef SCHED_FEAT
58
59static int sched_feat_show(struct seq_file *m, void *v)
60{
61 int i;
62
63 for (i = 0; i < __SCHED_FEAT_NR; i++) {
64 if (!(sysctl_sched_features & (1UL << i)))
65 seq_puts(m, s: "NO_");
66 seq_printf(m, fmt: "%s ", sched_feat_names[i]);
67 }
68 seq_puts(m, s: "\n");
69
70 return 0;
71}
72
73#ifdef CONFIG_JUMP_LABEL
74
75#define jump_label_key__true STATIC_KEY_INIT_TRUE
76#define jump_label_key__false STATIC_KEY_INIT_FALSE
77
78#define SCHED_FEAT(name, enabled) \
79 jump_label_key__##enabled ,
80
81struct static_key sched_feat_keys[__SCHED_FEAT_NR] = {
82#include "features.h"
83};
84
85#undef SCHED_FEAT
86
87static void sched_feat_disable(int i)
88{
89 static_key_disable_cpuslocked(key: &sched_feat_keys[i]);
90}
91
92static void sched_feat_enable(int i)
93{
94 static_key_enable_cpuslocked(key: &sched_feat_keys[i]);
95}
96#else /* !CONFIG_JUMP_LABEL: */
97static void sched_feat_disable(int i) { };
98static void sched_feat_enable(int i) { };
99#endif /* !CONFIG_JUMP_LABEL */
100
101static int sched_feat_set(char *cmp)
102{
103 int i;
104 int neg = 0;
105
106 if (strncmp(cmp, "NO_", 3) == 0) {
107 neg = 1;
108 cmp += 3;
109 }
110
111 i = match_string(array: sched_feat_names, n: __SCHED_FEAT_NR, string: cmp);
112 if (i < 0)
113 return i;
114
115 if (neg) {
116 sysctl_sched_features &= ~(1UL << i);
117 sched_feat_disable(i);
118 } else {
119 sysctl_sched_features |= (1UL << i);
120 sched_feat_enable(i);
121 }
122
123 return 0;
124}
125
126static ssize_t
127sched_feat_write(struct file *filp, const char __user *ubuf,
128 size_t cnt, loff_t *ppos)
129{
130 char buf[64];
131 char *cmp;
132 int ret;
133 struct inode *inode;
134
135 if (cnt > 63)
136 cnt = 63;
137
138 if (copy_from_user(to: &buf, from: ubuf, n: cnt))
139 return -EFAULT;
140
141 buf[cnt] = 0;
142 cmp = strstrip(str: buf);
143
144 /* Ensure the static_key remains in a consistent state */
145 inode = file_inode(f: filp);
146 cpus_read_lock();
147 inode_lock(inode);
148 ret = sched_feat_set(cmp);
149 inode_unlock(inode);
150 cpus_read_unlock();
151 if (ret < 0)
152 return ret;
153
154 *ppos += cnt;
155
156 return cnt;
157}
158
159static int sched_feat_open(struct inode *inode, struct file *filp)
160{
161 return single_open(filp, sched_feat_show, NULL);
162}
163
164static const struct file_operations sched_feat_fops = {
165 .open = sched_feat_open,
166 .write = sched_feat_write,
167 .read = seq_read,
168 .llseek = seq_lseek,
169 .release = single_release,
170};
171
172static ssize_t sched_scaling_write(struct file *filp, const char __user *ubuf,
173 size_t cnt, loff_t *ppos)
174{
175 char buf[16];
176 unsigned int scaling;
177
178 if (cnt > 15)
179 cnt = 15;
180
181 if (copy_from_user(to: &buf, from: ubuf, n: cnt))
182 return -EFAULT;
183 buf[cnt] = '\0';
184
185 if (kstrtouint(s: buf, base: 10, res: &scaling))
186 return -EINVAL;
187
188 if (scaling >= SCHED_TUNABLESCALING_END)
189 return -EINVAL;
190
191 sysctl_sched_tunable_scaling = scaling;
192 if (sched_update_scaling())
193 return -EINVAL;
194
195 *ppos += cnt;
196 return cnt;
197}
198
199static int sched_scaling_show(struct seq_file *m, void *v)
200{
201 seq_printf(m, fmt: "%d\n", sysctl_sched_tunable_scaling);
202 return 0;
203}
204
205static int sched_scaling_open(struct inode *inode, struct file *filp)
206{
207 return single_open(filp, sched_scaling_show, NULL);
208}
209
210static const struct file_operations sched_scaling_fops = {
211 .open = sched_scaling_open,
212 .write = sched_scaling_write,
213 .read = seq_read,
214 .llseek = seq_lseek,
215 .release = single_release,
216};
217
218#ifdef CONFIG_PREEMPT_DYNAMIC
219
220static ssize_t sched_dynamic_write(struct file *filp, const char __user *ubuf,
221 size_t cnt, loff_t *ppos)
222{
223 char buf[16];
224 int mode;
225
226 if (cnt > 15)
227 cnt = 15;
228
229 if (copy_from_user(to: &buf, from: ubuf, n: cnt))
230 return -EFAULT;
231
232 buf[cnt] = 0;
233 mode = sched_dynamic_mode(str: strstrip(str: buf));
234 if (mode < 0)
235 return mode;
236
237 sched_dynamic_update(mode);
238
239 *ppos += cnt;
240
241 return cnt;
242}
243
244static int sched_dynamic_show(struct seq_file *m, void *v)
245{
246 int i = IS_ENABLED(CONFIG_PREEMPT_RT) * 2;
247 int j;
248
249 /* Count entries in NULL terminated preempt_modes */
250 for (j = 0; preempt_modes[j]; j++)
251 ;
252 j -= !IS_ENABLED(CONFIG_ARCH_HAS_PREEMPT_LAZY);
253
254 for (; i < j; i++) {
255 if (preempt_dynamic_mode == i)
256 seq_puts(m, s: "(");
257 seq_puts(m, s: preempt_modes[i]);
258 if (preempt_dynamic_mode == i)
259 seq_puts(m, s: ")");
260
261 seq_puts(m, s: " ");
262 }
263
264 seq_puts(m, s: "\n");
265 return 0;
266}
267
268static int sched_dynamic_open(struct inode *inode, struct file *filp)
269{
270 return single_open(filp, sched_dynamic_show, NULL);
271}
272
273static const struct file_operations sched_dynamic_fops = {
274 .open = sched_dynamic_open,
275 .write = sched_dynamic_write,
276 .read = seq_read,
277 .llseek = seq_lseek,
278 .release = single_release,
279};
280
281#endif /* CONFIG_PREEMPT_DYNAMIC */
282
283__read_mostly bool sched_debug_verbose;
284
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 sched_domains_mutex_lock();
296
297 orig = sched_debug_verbose;
298 result = debugfs_write_file_bool(file: filp, user_buf: ubuf, count: cnt, ppos);
299
300 if (sched_debug_verbose && !orig)
301 update_sched_domain_debugfs();
302 else if (!sched_debug_verbose && orig) {
303 debugfs_remove(dentry: sd_dentry);
304 sd_dentry = NULL;
305 }
306
307 sched_domains_mutex_unlock();
308 cpus_read_unlock();
309
310 return result;
311}
312
313static const struct file_operations sched_verbose_fops = {
314 .read = debugfs_read_file_bool,
315 .write = sched_verbose_write,
316 .open = simple_open,
317 .llseek = default_llseek,
318};
319
320static const struct seq_operations sched_debug_sops;
321
322static int sched_debug_open(struct inode *inode, struct file *filp)
323{
324 return seq_open(filp, &sched_debug_sops);
325}
326
327static const struct file_operations sched_debug_fops = {
328 .open = sched_debug_open,
329 .read = seq_read,
330 .llseek = seq_lseek,
331 .release = seq_release,
332};
333
334enum dl_param {
335 DL_RUNTIME = 0,
336 DL_PERIOD,
337};
338
339static unsigned long fair_server_period_max = (1UL << 22) * NSEC_PER_USEC; /* ~4 seconds */
340static unsigned long fair_server_period_min = (100) * NSEC_PER_USEC; /* 100 us */
341
342static ssize_t sched_fair_server_write(struct file *filp, const char __user *ubuf,
343 size_t cnt, loff_t *ppos, enum dl_param param)
344{
345 long cpu = (long) ((struct seq_file *) filp->private_data)->private;
346 struct rq *rq = cpu_rq(cpu);
347 u64 runtime, period;
348 size_t err;
349 int retval;
350 u64 value;
351
352 err = kstrtoull_from_user(s: ubuf, count: cnt, base: 10, res: &value);
353 if (err)
354 return err;
355
356 scoped_guard (rq_lock_irqsave, rq) {
357 runtime = rq->fair_server.dl_runtime;
358 period = rq->fair_server.dl_period;
359
360 switch (param) {
361 case DL_RUNTIME:
362 if (runtime == value)
363 break;
364 runtime = value;
365 break;
366 case DL_PERIOD:
367 if (value == period)
368 break;
369 period = value;
370 break;
371 }
372
373 if (runtime > period ||
374 period > fair_server_period_max ||
375 period < fair_server_period_min) {
376 return -EINVAL;
377 }
378
379 update_rq_clock(rq);
380 dl_server_stop(dl_se: &rq->fair_server);
381
382 retval = dl_server_apply_params(dl_se: &rq->fair_server, runtime, period, init: 0);
383 if (retval)
384 cnt = retval;
385
386 if (!runtime)
387 printk_deferred("Fair server disabled in CPU %d, system may crash due to starvation.\n",
388 cpu_of(rq));
389
390 if (rq->cfs.h_nr_queued)
391 dl_server_start(dl_se: &rq->fair_server);
392 }
393
394 *ppos += cnt;
395 return cnt;
396}
397
398static size_t sched_fair_server_show(struct seq_file *m, void *v, enum dl_param param)
399{
400 unsigned long cpu = (unsigned long) m->private;
401 struct rq *rq = cpu_rq(cpu);
402 u64 value;
403
404 switch (param) {
405 case DL_RUNTIME:
406 value = rq->fair_server.dl_runtime;
407 break;
408 case DL_PERIOD:
409 value = rq->fair_server.dl_period;
410 break;
411 }
412
413 seq_printf(m, fmt: "%llu\n", value);
414 return 0;
415
416}
417
418static ssize_t
419sched_fair_server_runtime_write(struct file *filp, const char __user *ubuf,
420 size_t cnt, loff_t *ppos)
421{
422 return sched_fair_server_write(filp, ubuf, cnt, ppos, param: DL_RUNTIME);
423}
424
425static int sched_fair_server_runtime_show(struct seq_file *m, void *v)
426{
427 return sched_fair_server_show(m, v, param: DL_RUNTIME);
428}
429
430static int sched_fair_server_runtime_open(struct inode *inode, struct file *filp)
431{
432 return single_open(filp, sched_fair_server_runtime_show, inode->i_private);
433}
434
435static const struct file_operations fair_server_runtime_fops = {
436 .open = sched_fair_server_runtime_open,
437 .write = sched_fair_server_runtime_write,
438 .read = seq_read,
439 .llseek = seq_lseek,
440 .release = single_release,
441};
442
443static ssize_t
444sched_fair_server_period_write(struct file *filp, const char __user *ubuf,
445 size_t cnt, loff_t *ppos)
446{
447 return sched_fair_server_write(filp, ubuf, cnt, ppos, param: DL_PERIOD);
448}
449
450static int sched_fair_server_period_show(struct seq_file *m, void *v)
451{
452 return sched_fair_server_show(m, v, param: DL_PERIOD);
453}
454
455static int sched_fair_server_period_open(struct inode *inode, struct file *filp)
456{
457 return single_open(filp, sched_fair_server_period_show, inode->i_private);
458}
459
460static const struct file_operations fair_server_period_fops = {
461 .open = sched_fair_server_period_open,
462 .write = sched_fair_server_period_write,
463 .read = seq_read,
464 .llseek = seq_lseek,
465 .release = single_release,
466};
467
468static struct dentry *debugfs_sched;
469
470static void debugfs_fair_server_init(void)
471{
472 struct dentry *d_fair;
473 unsigned long cpu;
474
475 d_fair = debugfs_create_dir(name: "fair_server", parent: debugfs_sched);
476 if (!d_fair)
477 return;
478
479 for_each_possible_cpu(cpu) {
480 struct dentry *d_cpu;
481 char buf[32];
482
483 snprintf(buf, size: sizeof(buf), fmt: "cpu%lu", cpu);
484 d_cpu = debugfs_create_dir(name: buf, parent: d_fair);
485
486 debugfs_create_file("runtime", 0644, d_cpu, (void *) cpu, &fair_server_runtime_fops);
487 debugfs_create_file("period", 0644, d_cpu, (void *) cpu, &fair_server_period_fops);
488 }
489}
490
491static __init int sched_init_debug(void)
492{
493 struct dentry __maybe_unused *numa;
494
495 debugfs_sched = debugfs_create_dir(name: "sched", NULL);
496
497 debugfs_create_file("features", 0644, debugfs_sched, NULL, &sched_feat_fops);
498 debugfs_create_file_unsafe(name: "verbose", mode: 0644, parent: debugfs_sched, data: &sched_debug_verbose, fops: &sched_verbose_fops);
499#ifdef CONFIG_PREEMPT_DYNAMIC
500 debugfs_create_file("preempt", 0644, debugfs_sched, NULL, &sched_dynamic_fops);
501#endif
502
503 debugfs_create_u32(name: "base_slice_ns", mode: 0644, parent: debugfs_sched, value: &sysctl_sched_base_slice);
504
505 debugfs_create_u32(name: "latency_warn_ms", mode: 0644, parent: debugfs_sched, value: &sysctl_resched_latency_warn_ms);
506 debugfs_create_u32(name: "latency_warn_once", mode: 0644, parent: debugfs_sched, value: &sysctl_resched_latency_warn_once);
507
508 debugfs_create_file("tunable_scaling", 0644, debugfs_sched, NULL, &sched_scaling_fops);
509 debugfs_create_u32(name: "migration_cost_ns", mode: 0644, parent: debugfs_sched, value: &sysctl_sched_migration_cost);
510 debugfs_create_u32(name: "nr_migrate", mode: 0644, parent: debugfs_sched, value: &sysctl_sched_nr_migrate);
511
512 sched_domains_mutex_lock();
513 update_sched_domain_debugfs();
514 sched_domains_mutex_unlock();
515
516#ifdef CONFIG_NUMA_BALANCING
517 numa = debugfs_create_dir("numa_balancing", debugfs_sched);
518
519 debugfs_create_u32("scan_delay_ms", 0644, numa, &sysctl_numa_balancing_scan_delay);
520 debugfs_create_u32("scan_period_min_ms", 0644, numa, &sysctl_numa_balancing_scan_period_min);
521 debugfs_create_u32("scan_period_max_ms", 0644, numa, &sysctl_numa_balancing_scan_period_max);
522 debugfs_create_u32("scan_size_mb", 0644, numa, &sysctl_numa_balancing_scan_size);
523 debugfs_create_u32("hot_threshold_ms", 0644, numa, &sysctl_numa_balancing_hot_threshold);
524#endif /* CONFIG_NUMA_BALANCING */
525
526 debugfs_create_file("debug", 0444, debugfs_sched, NULL, &sched_debug_fops);
527
528 debugfs_fair_server_init();
529
530 return 0;
531}
532late_initcall(sched_init_debug);
533
534static cpumask_var_t sd_sysctl_cpus;
535
536static int sd_flags_show(struct seq_file *m, void *v)
537{
538 unsigned long flags = *(unsigned int *)m->private;
539 int idx;
540
541 for_each_set_bit(idx, &flags, __SD_FLAG_CNT) {
542 seq_puts(m, s: sd_flag_debug[idx].name);
543 seq_puts(m, s: " ");
544 }
545 seq_puts(m, s: "\n");
546
547 return 0;
548}
549
550static int sd_flags_open(struct inode *inode, struct file *file)
551{
552 return single_open(file, sd_flags_show, inode->i_private);
553}
554
555static const struct file_operations sd_flags_fops = {
556 .open = sd_flags_open,
557 .read = seq_read,
558 .llseek = seq_lseek,
559 .release = single_release,
560};
561
562static void register_sd(struct sched_domain *sd, struct dentry *parent)
563{
564#define SDM(type, mode, member) \
565 debugfs_create_##type(#member, mode, parent, &sd->member)
566
567 SDM(ulong, 0644, min_interval);
568 SDM(ulong, 0644, max_interval);
569 SDM(u64, 0644, max_newidle_lb_cost);
570 SDM(u32, 0644, busy_factor);
571 SDM(u32, 0644, imbalance_pct);
572 SDM(u32, 0644, cache_nice_tries);
573 SDM(str, 0444, name);
574
575#undef SDM
576
577 debugfs_create_file("flags", 0444, parent, &sd->flags, &sd_flags_fops);
578 debugfs_create_file("groups_flags", 0444, parent, &sd->groups->flags, &sd_flags_fops);
579 debugfs_create_u32(name: "level", mode: 0444, parent, value: (u32 *)&sd->level);
580
581 if (sd->flags & SD_ASYM_PACKING)
582 debugfs_create_u32(name: "group_asym_prefer_cpu", mode: 0444, parent,
583 value: (u32 *)&sd->groups->asym_prefer_cpu);
584}
585
586void update_sched_domain_debugfs(void)
587{
588 int cpu, i;
589
590 /*
591 * This can unfortunately be invoked before sched_debug_init() creates
592 * the debug directory. Don't touch sd_sysctl_cpus until then.
593 */
594 if (!debugfs_sched)
595 return;
596
597 if (!sched_debug_verbose)
598 return;
599
600 if (!cpumask_available(mask: sd_sysctl_cpus)) {
601 if (!alloc_cpumask_var(mask: &sd_sysctl_cpus, GFP_KERNEL))
602 return;
603 cpumask_copy(dstp: sd_sysctl_cpus, cpu_possible_mask);
604 }
605
606 if (!sd_dentry) {
607 sd_dentry = debugfs_create_dir(name: "domains", parent: debugfs_sched);
608
609 /* rebuild sd_sysctl_cpus if empty since it gets cleared below */
610 if (cpumask_empty(srcp: sd_sysctl_cpus))
611 cpumask_copy(dstp: sd_sysctl_cpus, cpu_online_mask);
612 }
613
614 for_each_cpu(cpu, sd_sysctl_cpus) {
615 struct sched_domain *sd;
616 struct dentry *d_cpu;
617 char buf[32];
618
619 snprintf(buf, size: sizeof(buf), fmt: "cpu%d", cpu);
620 debugfs_lookup_and_remove(name: buf, parent: sd_dentry);
621 d_cpu = debugfs_create_dir(name: buf, parent: sd_dentry);
622
623 i = 0;
624 for_each_domain(cpu, sd) {
625 struct dentry *d_sd;
626
627 snprintf(buf, size: sizeof(buf), fmt: "domain%d", i);
628 d_sd = debugfs_create_dir(name: buf, parent: d_cpu);
629
630 register_sd(sd, parent: d_sd);
631 i++;
632 }
633
634 __cpumask_clear_cpu(cpu, dstp: sd_sysctl_cpus);
635 }
636}
637
638void dirty_sched_domain_sysctl(int cpu)
639{
640 if (cpumask_available(mask: sd_sysctl_cpus))
641 __cpumask_set_cpu(cpu, dstp: sd_sysctl_cpus);
642}
643
644#ifdef CONFIG_FAIR_GROUP_SCHED
645static void print_cfs_group_stats(struct seq_file *m, int cpu, struct task_group *tg)
646{
647 struct sched_entity *se = tg->se[cpu];
648
649#define P(F) SEQ_printf(m, " .%-30s: %lld\n", #F, (long long)F)
650#define P_SCHEDSTAT(F) SEQ_printf(m, " .%-30s: %lld\n", \
651 #F, (long long)schedstat_val(stats->F))
652#define PN(F) SEQ_printf(m, " .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F))
653#define PN_SCHEDSTAT(F) SEQ_printf(m, " .%-30s: %lld.%06ld\n", \
654 #F, SPLIT_NS((long long)schedstat_val(stats->F)))
655
656 if (!se)
657 return;
658
659 PN(se->exec_start);
660 PN(se->vruntime);
661 PN(se->sum_exec_runtime);
662
663 if (schedstat_enabled()) {
664 struct sched_statistics *stats;
665 stats = __schedstats_from_se(se);
666
667 PN_SCHEDSTAT(wait_start);
668 PN_SCHEDSTAT(sleep_start);
669 PN_SCHEDSTAT(block_start);
670 PN_SCHEDSTAT(sleep_max);
671 PN_SCHEDSTAT(block_max);
672 PN_SCHEDSTAT(exec_max);
673 PN_SCHEDSTAT(slice_max);
674 PN_SCHEDSTAT(wait_max);
675 PN_SCHEDSTAT(wait_sum);
676 P_SCHEDSTAT(wait_count);
677 }
678
679 P(se->load.weight);
680 P(se->avg.load_avg);
681 P(se->avg.util_avg);
682 P(se->avg.runnable_avg);
683
684#undef PN_SCHEDSTAT
685#undef PN
686#undef P_SCHEDSTAT
687#undef P
688}
689#endif /* CONFIG_FAIR_GROUP_SCHED */
690
691#ifdef CONFIG_CGROUP_SCHED
692static DEFINE_SPINLOCK(sched_debug_lock);
693static char group_path[PATH_MAX];
694
695static void task_group_path(struct task_group *tg, char *path, int plen)
696{
697 if (autogroup_path(tg, buf: path, buflen: plen))
698 return;
699
700 cgroup_path(cgrp: tg->css.cgroup, buf: path, buflen: plen);
701}
702
703/*
704 * Only 1 SEQ_printf_task_group_path() caller can use the full length
705 * group_path[] for cgroup path. Other simultaneous callers will have
706 * to use a shorter stack buffer. A "..." suffix is appended at the end
707 * of the stack buffer so that it will show up in case the output length
708 * matches the given buffer size to indicate possible path name truncation.
709 */
710#define SEQ_printf_task_group_path(m, tg, fmt...) \
711{ \
712 if (spin_trylock(&sched_debug_lock)) { \
713 task_group_path(tg, group_path, sizeof(group_path)); \
714 SEQ_printf(m, fmt, group_path); \
715 spin_unlock(&sched_debug_lock); \
716 } else { \
717 char buf[128]; \
718 char *bufend = buf + sizeof(buf) - 3; \
719 task_group_path(tg, buf, bufend - buf); \
720 strcpy(bufend - 1, "..."); \
721 SEQ_printf(m, fmt, buf); \
722 } \
723}
724#endif
725
726static void
727print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
728{
729 if (task_current(rq, p))
730 SEQ_printf(m, ">R");
731 else
732 SEQ_printf(m, " %c", task_state_to_char(p));
733
734 SEQ_printf(m, " %15s %5d %9Ld.%06ld %c %9Ld.%06ld %c %9Ld.%06ld %9Ld.%06ld %9Ld %5d ",
735 p->comm, task_pid_nr(p),
736 SPLIT_NS(p->se.vruntime),
737 entity_eligible(cfs_rq_of(&p->se), &p->se) ? 'E' : 'N',
738 SPLIT_NS(p->se.deadline),
739 p->se.custom_slice ? 'S' : ' ',
740 SPLIT_NS(p->se.slice),
741 SPLIT_NS(p->se.sum_exec_runtime),
742 (long long)(p->nvcsw + p->nivcsw),
743 p->prio);
744
745 SEQ_printf(m, "%9lld.%06ld %9lld.%06ld %9lld.%06ld",
746 SPLIT_NS(schedstat_val_or_zero(p->stats.wait_sum)),
747 SPLIT_NS(schedstat_val_or_zero(p->stats.sum_sleep_runtime)),
748 SPLIT_NS(schedstat_val_or_zero(p->stats.sum_block_runtime)));
749
750#ifdef CONFIG_NUMA_BALANCING
751 SEQ_printf(m, " %d %d", task_node(p), task_numa_group_id(p));
752#endif
753#ifdef CONFIG_CGROUP_SCHED
754 SEQ_printf_task_group_path(m, task_group(p), " %s")
755#endif
756
757 SEQ_printf(m, "\n");
758}
759
760static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
761{
762 struct task_struct *g, *p;
763
764 SEQ_printf(m, "\n");
765 SEQ_printf(m, "runnable tasks:\n");
766 SEQ_printf(m, " S task PID vruntime eligible "
767 "deadline slice sum-exec switches "
768 "prio wait-time sum-sleep sum-block"
769#ifdef CONFIG_NUMA_BALANCING
770 " node group-id"
771#endif
772#ifdef CONFIG_CGROUP_SCHED
773 " group-path"
774#endif
775 "\n");
776 SEQ_printf(m, "-------------------------------------------------------"
777 "------------------------------------------------------"
778 "------------------------------------------------------"
779#ifdef CONFIG_NUMA_BALANCING
780 "--------------"
781#endif
782#ifdef CONFIG_CGROUP_SCHED
783 "--------------"
784#endif
785 "\n");
786
787 rcu_read_lock();
788 for_each_process_thread(g, p) {
789 if (task_cpu(p) != rq_cpu)
790 continue;
791
792 print_task(m, rq, p);
793 }
794 rcu_read_unlock();
795}
796
797void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
798{
799 s64 left_vruntime = -1, min_vruntime, right_vruntime = -1, left_deadline = -1, spread;
800 struct sched_entity *last, *first, *root;
801 struct rq *rq = cpu_rq(cpu);
802 unsigned long flags;
803
804#ifdef CONFIG_FAIR_GROUP_SCHED
805 SEQ_printf(m, "\n");
806 SEQ_printf_task_group_path(m, cfs_rq->tg, "cfs_rq[%d]:%s\n", cpu);
807#else
808 SEQ_printf(m, "\n");
809 SEQ_printf(m, "cfs_rq[%d]:\n", cpu);
810#endif
811
812 raw_spin_rq_lock_irqsave(rq, flags);
813 root = __pick_root_entity(cfs_rq);
814 if (root)
815 left_vruntime = root->min_vruntime;
816 first = __pick_first_entity(cfs_rq);
817 if (first)
818 left_deadline = first->deadline;
819 last = __pick_last_entity(cfs_rq);
820 if (last)
821 right_vruntime = last->vruntime;
822 min_vruntime = cfs_rq->min_vruntime;
823 raw_spin_rq_unlock_irqrestore(rq, flags);
824
825 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "left_deadline",
826 SPLIT_NS(left_deadline));
827 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "left_vruntime",
828 SPLIT_NS(left_vruntime));
829 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "min_vruntime",
830 SPLIT_NS(min_vruntime));
831 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "avg_vruntime",
832 SPLIT_NS(avg_vruntime(cfs_rq)));
833 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "right_vruntime",
834 SPLIT_NS(right_vruntime));
835 spread = right_vruntime - left_vruntime;
836 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread", SPLIT_NS(spread));
837 SEQ_printf(m, " .%-30s: %d\n", "nr_queued", cfs_rq->nr_queued);
838 SEQ_printf(m, " .%-30s: %d\n", "h_nr_runnable", cfs_rq->h_nr_runnable);
839 SEQ_printf(m, " .%-30s: %d\n", "h_nr_queued", cfs_rq->h_nr_queued);
840 SEQ_printf(m, " .%-30s: %d\n", "h_nr_idle", cfs_rq->h_nr_idle);
841 SEQ_printf(m, " .%-30s: %ld\n", "load", cfs_rq->load.weight);
842 SEQ_printf(m, " .%-30s: %lu\n", "load_avg",
843 cfs_rq->avg.load_avg);
844 SEQ_printf(m, " .%-30s: %lu\n", "runnable_avg",
845 cfs_rq->avg.runnable_avg);
846 SEQ_printf(m, " .%-30s: %lu\n", "util_avg",
847 cfs_rq->avg.util_avg);
848 SEQ_printf(m, " .%-30s: %u\n", "util_est",
849 cfs_rq->avg.util_est);
850 SEQ_printf(m, " .%-30s: %ld\n", "removed.load_avg",
851 cfs_rq->removed.load_avg);
852 SEQ_printf(m, " .%-30s: %ld\n", "removed.util_avg",
853 cfs_rq->removed.util_avg);
854 SEQ_printf(m, " .%-30s: %ld\n", "removed.runnable_avg",
855 cfs_rq->removed.runnable_avg);
856#ifdef CONFIG_FAIR_GROUP_SCHED
857 SEQ_printf(m, " .%-30s: %lu\n", "tg_load_avg_contrib",
858 cfs_rq->tg_load_avg_contrib);
859 SEQ_printf(m, " .%-30s: %ld\n", "tg_load_avg",
860 atomic_long_read(&cfs_rq->tg->load_avg));
861#endif /* CONFIG_FAIR_GROUP_SCHED */
862#ifdef CONFIG_CFS_BANDWIDTH
863 SEQ_printf(m, " .%-30s: %d\n", "throttled",
864 cfs_rq->throttled);
865 SEQ_printf(m, " .%-30s: %d\n", "throttle_count",
866 cfs_rq->throttle_count);
867#endif
868
869#ifdef CONFIG_FAIR_GROUP_SCHED
870 print_cfs_group_stats(m, cpu, tg: cfs_rq->tg);
871#endif
872}
873
874void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
875{
876#ifdef CONFIG_RT_GROUP_SCHED
877 SEQ_printf(m, "\n");
878 SEQ_printf_task_group_path(m, rt_rq->tg, "rt_rq[%d]:%s\n", cpu);
879#else
880 SEQ_printf(m, "\n");
881 SEQ_printf(m, "rt_rq[%d]:\n", cpu);
882#endif
883
884#define P(x) \
885 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
886#define PU(x) \
887 SEQ_printf(m, " .%-30s: %lu\n", #x, (unsigned long)(rt_rq->x))
888#define PN(x) \
889 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
890
891 PU(rt_nr_running);
892
893#ifdef CONFIG_RT_GROUP_SCHED
894 P(rt_throttled);
895 PN(rt_time);
896 PN(rt_runtime);
897#endif
898
899#undef PN
900#undef PU
901#undef P
902}
903
904void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq)
905{
906 struct dl_bw *dl_bw;
907
908 SEQ_printf(m, "\n");
909 SEQ_printf(m, "dl_rq[%d]:\n", cpu);
910
911#define PU(x) \
912 SEQ_printf(m, " .%-30s: %lu\n", #x, (unsigned long)(dl_rq->x))
913
914 PU(dl_nr_running);
915 dl_bw = &cpu_rq(cpu)->rd->dl_bw;
916 SEQ_printf(m, " .%-30s: %lld\n", "dl_bw->bw", dl_bw->bw);
917 SEQ_printf(m, " .%-30s: %lld\n", "dl_bw->total_bw", dl_bw->total_bw);
918
919#undef PU
920}
921
922static void print_cpu(struct seq_file *m, int cpu)
923{
924 struct rq *rq = cpu_rq(cpu);
925
926#ifdef CONFIG_X86
927 {
928 unsigned int freq = cpu_khz ? : 1;
929
930 SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
931 cpu, freq / 1000, (freq % 1000));
932 }
933#else /* !CONFIG_X86: */
934 SEQ_printf(m, "cpu#%d\n", cpu);
935#endif /* !CONFIG_X86 */
936
937#define P(x) \
938do { \
939 if (sizeof(rq->x) == 4) \
940 SEQ_printf(m, " .%-30s: %d\n", #x, (int)(rq->x)); \
941 else \
942 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rq->x));\
943} while (0)
944
945#define PN(x) \
946 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
947
948 P(nr_running);
949 P(nr_switches);
950 P(nr_uninterruptible);
951 PN(next_balance);
952 SEQ_printf(m, " .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr)));
953 PN(clock);
954 PN(clock_task);
955#undef P
956#undef PN
957
958#define P64(n) SEQ_printf(m, " .%-30s: %Ld\n", #n, rq->n);
959 P64(avg_idle);
960 P64(max_idle_balance_cost);
961#undef P64
962
963#define P(n) SEQ_printf(m, " .%-30s: %d\n", #n, schedstat_val(rq->n));
964 if (schedstat_enabled()) {
965 P(yld_count);
966 P(sched_count);
967 P(sched_goidle);
968 P(ttwu_count);
969 P(ttwu_local);
970 }
971#undef P
972
973 print_cfs_stats(m, cpu);
974 print_rt_stats(m, cpu);
975 print_dl_stats(m, cpu);
976
977 print_rq(m, rq, rq_cpu: cpu);
978 SEQ_printf(m, "\n");
979}
980
981static const char *sched_tunable_scaling_names[] = {
982 "none",
983 "logarithmic",
984 "linear"
985};
986
987static void sched_debug_header(struct seq_file *m)
988{
989 u64 ktime, sched_clk, cpu_clk;
990 unsigned long flags;
991
992 local_irq_save(flags);
993 ktime = ktime_to_ns(kt: ktime_get());
994 sched_clk = sched_clock();
995 cpu_clk = local_clock();
996 local_irq_restore(flags);
997
998 SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n",
999 init_utsname()->release,
1000 (int)strcspn(init_utsname()->version, " "),
1001 init_utsname()->version);
1002
1003#define P(x) \
1004 SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
1005#define PN(x) \
1006 SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
1007 PN(ktime);
1008 PN(sched_clk);
1009 PN(cpu_clk);
1010 P(jiffies);
1011#ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
1012 P(sched_clock_stable());
1013#endif
1014#undef PN
1015#undef P
1016
1017 SEQ_printf(m, "\n");
1018 SEQ_printf(m, "sysctl_sched\n");
1019
1020#define P(x) \
1021 SEQ_printf(m, " .%-40s: %Ld\n", #x, (long long)(x))
1022#define PN(x) \
1023 SEQ_printf(m, " .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
1024 PN(sysctl_sched_base_slice);
1025 P(sysctl_sched_features);
1026#undef PN
1027#undef P
1028
1029 SEQ_printf(m, " .%-40s: %d (%s)\n",
1030 "sysctl_sched_tunable_scaling",
1031 sysctl_sched_tunable_scaling,
1032 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
1033 SEQ_printf(m, "\n");
1034}
1035
1036static int sched_debug_show(struct seq_file *m, void *v)
1037{
1038 int cpu = (unsigned long)(v - 2);
1039
1040 if (cpu != -1)
1041 print_cpu(m, cpu);
1042 else
1043 sched_debug_header(m);
1044
1045 return 0;
1046}
1047
1048void sysrq_sched_debug_show(void)
1049{
1050 int cpu;
1051
1052 sched_debug_header(NULL);
1053 for_each_online_cpu(cpu) {
1054 /*
1055 * Need to reset softlockup watchdogs on all CPUs, because
1056 * another CPU might be blocked waiting for us to process
1057 * an IPI or stop_machine.
1058 */
1059 touch_nmi_watchdog();
1060 touch_all_softlockup_watchdogs();
1061 print_cpu(NULL, cpu);
1062 }
1063}
1064
1065/*
1066 * This iterator needs some explanation.
1067 * It returns 1 for the header position.
1068 * This means 2 is CPU 0.
1069 * In a hotplugged system some CPUs, including CPU 0, may be missing so we have
1070 * to use cpumask_* to iterate over the CPUs.
1071 */
1072static void *sched_debug_start(struct seq_file *file, loff_t *offset)
1073{
1074 unsigned long n = *offset;
1075
1076 if (n == 0)
1077 return (void *) 1;
1078
1079 n--;
1080
1081 if (n > 0)
1082 n = cpumask_next(n: n - 1, cpu_online_mask);
1083 else
1084 n = cpumask_first(cpu_online_mask);
1085
1086 *offset = n + 1;
1087
1088 if (n < nr_cpu_ids)
1089 return (void *)(unsigned long)(n + 2);
1090
1091 return NULL;
1092}
1093
1094static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
1095{
1096 (*offset)++;
1097 return sched_debug_start(file, offset);
1098}
1099
1100static void sched_debug_stop(struct seq_file *file, void *data)
1101{
1102}
1103
1104static const struct seq_operations sched_debug_sops = {
1105 .start = sched_debug_start,
1106 .next = sched_debug_next,
1107 .stop = sched_debug_stop,
1108 .show = sched_debug_show,
1109};
1110
1111#define __PS(S, F) SEQ_printf(m, "%-45s:%21Ld\n", S, (long long)(F))
1112#define __P(F) __PS(#F, F)
1113#define P(F) __PS(#F, p->F)
1114#define PM(F, M) __PS(#F, p->F & (M))
1115#define __PSN(S, F) SEQ_printf(m, "%-45s:%14Ld.%06ld\n", S, SPLIT_NS((long long)(F)))
1116#define __PN(F) __PSN(#F, F)
1117#define PN(F) __PSN(#F, p->F)
1118
1119
1120#ifdef CONFIG_NUMA_BALANCING
1121void print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
1122 unsigned long tpf, unsigned long gsf, unsigned long gpf)
1123{
1124 SEQ_printf(m, "numa_faults node=%d ", node);
1125 SEQ_printf(m, "task_private=%lu task_shared=%lu ", tpf, tsf);
1126 SEQ_printf(m, "group_private=%lu group_shared=%lu\n", gpf, gsf);
1127}
1128#endif
1129
1130
1131static void sched_show_numa(struct task_struct *p, struct seq_file *m)
1132{
1133#ifdef CONFIG_NUMA_BALANCING
1134 if (p->mm)
1135 P(mm->numa_scan_seq);
1136
1137 P(numa_pages_migrated);
1138 P(numa_preferred_nid);
1139 P(total_numa_faults);
1140 SEQ_printf(m, "current_node=%d, numa_group_id=%d\n",
1141 task_node(p), task_numa_group_id(p));
1142 show_numa_stats(p, m);
1143#endif /* CONFIG_NUMA_BALANCING */
1144}
1145
1146void proc_sched_show_task(struct task_struct *p, struct pid_namespace *ns,
1147 struct seq_file *m)
1148{
1149 unsigned long nr_switches;
1150
1151 SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr_ns(p, ns),
1152 get_nr_threads(p));
1153 SEQ_printf(m,
1154 "---------------------------------------------------------"
1155 "----------\n");
1156
1157#define P_SCHEDSTAT(F) __PS(#F, schedstat_val(p->stats.F))
1158#define PN_SCHEDSTAT(F) __PSN(#F, schedstat_val(p->stats.F))
1159
1160 PN(se.exec_start);
1161 PN(se.vruntime);
1162 PN(se.sum_exec_runtime);
1163
1164 nr_switches = p->nvcsw + p->nivcsw;
1165
1166 P(se.nr_migrations);
1167
1168 if (schedstat_enabled()) {
1169 u64 avg_atom, avg_per_cpu;
1170
1171 PN_SCHEDSTAT(sum_sleep_runtime);
1172 PN_SCHEDSTAT(sum_block_runtime);
1173 PN_SCHEDSTAT(wait_start);
1174 PN_SCHEDSTAT(sleep_start);
1175 PN_SCHEDSTAT(block_start);
1176 PN_SCHEDSTAT(sleep_max);
1177 PN_SCHEDSTAT(block_max);
1178 PN_SCHEDSTAT(exec_max);
1179 PN_SCHEDSTAT(slice_max);
1180 PN_SCHEDSTAT(wait_max);
1181 PN_SCHEDSTAT(wait_sum);
1182 P_SCHEDSTAT(wait_count);
1183 PN_SCHEDSTAT(iowait_sum);
1184 P_SCHEDSTAT(iowait_count);
1185 P_SCHEDSTAT(nr_migrations_cold);
1186 P_SCHEDSTAT(nr_failed_migrations_affine);
1187 P_SCHEDSTAT(nr_failed_migrations_running);
1188 P_SCHEDSTAT(nr_failed_migrations_hot);
1189 P_SCHEDSTAT(nr_forced_migrations);
1190 P_SCHEDSTAT(nr_wakeups);
1191 P_SCHEDSTAT(nr_wakeups_sync);
1192 P_SCHEDSTAT(nr_wakeups_migrate);
1193 P_SCHEDSTAT(nr_wakeups_local);
1194 P_SCHEDSTAT(nr_wakeups_remote);
1195 P_SCHEDSTAT(nr_wakeups_affine);
1196 P_SCHEDSTAT(nr_wakeups_affine_attempts);
1197 P_SCHEDSTAT(nr_wakeups_passive);
1198 P_SCHEDSTAT(nr_wakeups_idle);
1199
1200 avg_atom = p->se.sum_exec_runtime;
1201 if (nr_switches)
1202 avg_atom = div64_ul(avg_atom, nr_switches);
1203 else
1204 avg_atom = -1LL;
1205
1206 avg_per_cpu = p->se.sum_exec_runtime;
1207 if (p->se.nr_migrations) {
1208 avg_per_cpu = div64_u64(dividend: avg_per_cpu,
1209 divisor: p->se.nr_migrations);
1210 } else {
1211 avg_per_cpu = -1LL;
1212 }
1213
1214 __PN(avg_atom);
1215 __PN(avg_per_cpu);
1216
1217#ifdef CONFIG_SCHED_CORE
1218 PN_SCHEDSTAT(core_forceidle_sum);
1219#endif
1220 }
1221
1222 __P(nr_switches);
1223 __PS("nr_voluntary_switches", p->nvcsw);
1224 __PS("nr_involuntary_switches", p->nivcsw);
1225
1226 P(se.load.weight);
1227 P(se.avg.load_sum);
1228 P(se.avg.runnable_sum);
1229 P(se.avg.util_sum);
1230 P(se.avg.load_avg);
1231 P(se.avg.runnable_avg);
1232 P(se.avg.util_avg);
1233 P(se.avg.last_update_time);
1234 PM(se.avg.util_est, ~UTIL_AVG_UNCHANGED);
1235#ifdef CONFIG_UCLAMP_TASK
1236 __PS("uclamp.min", p->uclamp_req[UCLAMP_MIN].value);
1237 __PS("uclamp.max", p->uclamp_req[UCLAMP_MAX].value);
1238 __PS("effective uclamp.min", uclamp_eff_value(p, UCLAMP_MIN));
1239 __PS("effective uclamp.max", uclamp_eff_value(p, UCLAMP_MAX));
1240#endif /* CONFIG_UCLAMP_TASK */
1241 P(policy);
1242 P(prio);
1243 if (task_has_dl_policy(p)) {
1244 P(dl.runtime);
1245 P(dl.deadline);
1246 } else if (fair_policy(policy: p->policy)) {
1247 P(se.slice);
1248 }
1249#ifdef CONFIG_SCHED_CLASS_EXT
1250 __PS("ext.enabled", task_on_scx(p));
1251#endif
1252#undef PN_SCHEDSTAT
1253#undef P_SCHEDSTAT
1254
1255 {
1256 unsigned int this_cpu = raw_smp_processor_id();
1257 u64 t0, t1;
1258
1259 t0 = cpu_clock(cpu: this_cpu);
1260 t1 = cpu_clock(cpu: this_cpu);
1261 __PS("clock-delta", t1-t0);
1262 }
1263
1264 sched_show_numa(p, m);
1265}
1266
1267void proc_sched_set_task(struct task_struct *p)
1268{
1269#ifdef CONFIG_SCHEDSTATS
1270 memset(s: &p->stats, c: 0, n: sizeof(p->stats));
1271#endif
1272}
1273
1274void resched_latency_warn(int cpu, u64 latency)
1275{
1276 static DEFINE_RATELIMIT_STATE(latency_check_ratelimit, 60 * 60 * HZ, 1);
1277
1278 if (likely(!__ratelimit(&latency_check_ratelimit)))
1279 return;
1280
1281 pr_err("sched: CPU %d need_resched set for > %llu ns (%d ticks) without schedule\n",
1282 cpu, latency, cpu_rq(cpu)->ticks_without_resched);
1283 dump_stack();
1284}
1285