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/*
2 * kernel/sched/debug.c
3 *
4 * Print the CFS rbtree
5 *
6 * Copyright(C) 2007, Red Hat, Inc., Ingo Molnar
7 *
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 as
10 * published by the Free Software Foundation.
11 */
12
13#include <linux/proc_fs.h>
14#include <linux/sched.h>
15#include <linux/seq_file.h>
16#include <linux/kallsyms.h>
17#include <linux/utsname.h>
18#include <linux/mempolicy.h>
19
20#include "sched.h"
21
22static DEFINE_SPINLOCK(sched_debug_lock);
23
24/*
25 * This allows printing both to /proc/sched_debug and
26 * to the console
27 */
28#define SEQ_printf(m, x...) \
29 do { \
30 if (m) \
31 seq_printf(m, x); \
32 else \
33 printk(x); \
34 } while (0)
35
36/*
37 * Ease the printing of nsec fields:
38 */
39static long long nsec_high(unsigned long long nsec)
40{
41 if ((long long)nsec < 0) {
42 nsec = -nsec;
43 do_div(nsec, 1000000);
44 return -nsec;
45 }
46 do_div(nsec, 1000000);
47
48 return nsec;
49}
50
51static unsigned long nsec_low(unsigned long long nsec)
52{
53 if ((long long)nsec < 0)
54 nsec = -nsec;
55
56 return do_div(nsec, 1000000);
57}
58
59#define SPLIT_NS(x) nsec_high(x), nsec_low(x)
60
61#ifdef CONFIG_FAIR_GROUP_SCHED
62static void print_cfs_group_stats(struct seq_file *m, int cpu, struct task_group *tg)
63{
64 struct sched_entity *se = tg->se[cpu];
65
66#define P(F) \
67 SEQ_printf(m, " .%-30s: %lld\n", #F, (long long)F)
68#define PN(F) \
69 SEQ_printf(m, " .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F))
70
71 if (!se) {
72 struct sched_avg *avg = &cpu_rq(cpu)->avg;
73 P(avg->runnable_avg_sum);
74 P(avg->runnable_avg_period);
75 return;
76 }
77
78
79 PN(se->exec_start);
80 PN(se->vruntime);
81 PN(se->sum_exec_runtime);
82#ifdef CONFIG_SCHEDSTATS
83 PN(se->statistics.wait_start);
84 PN(se->statistics.sleep_start);
85 PN(se->statistics.block_start);
86 PN(se->statistics.sleep_max);
87 PN(se->statistics.block_max);
88 PN(se->statistics.exec_max);
89 PN(se->statistics.slice_max);
90 PN(se->statistics.wait_max);
91 PN(se->statistics.wait_sum);
92 P(se->statistics.wait_count);
93#endif
94 P(se->load.weight);
95#ifdef CONFIG_SMP
96 P(se->avg.runnable_avg_sum);
97 P(se->avg.runnable_avg_period);
98 P(se->avg.load_avg_contrib);
99 P(se->avg.decay_count);
100#endif
101#undef PN
102#undef P
103}
104#endif
105
106#ifdef CONFIG_CGROUP_SCHED
107static char group_path[PATH_MAX];
108
109static char *task_group_path(struct task_group *tg)
110{
111 if (autogroup_path(tg, group_path, PATH_MAX))
112 return group_path;
113
114 return cgroup_path(tg->css.cgroup, group_path, PATH_MAX);
115}
116#endif
117
118static void
119print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
120{
121 if (rq->curr == p)
122 SEQ_printf(m, "R");
123 else
124 SEQ_printf(m, " ");
125
126 SEQ_printf(m, "%15s %5d %9Ld.%06ld %9Ld %5d ",
127 p->comm, task_pid_nr(p),
128 SPLIT_NS(p->se.vruntime),
129 (long long)(p->nvcsw + p->nivcsw),
130 p->prio);
131#ifdef CONFIG_SCHEDSTATS
132 SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
133 SPLIT_NS(p->se.vruntime),
134 SPLIT_NS(p->se.sum_exec_runtime),
135 SPLIT_NS(p->se.statistics.sum_sleep_runtime));
136#else
137 SEQ_printf(m, "%15Ld %15Ld %15Ld.%06ld %15Ld.%06ld %15Ld.%06ld",
138 0LL, 0LL, 0LL, 0L, 0LL, 0L, 0LL, 0L);
139#endif
140#ifdef CONFIG_NUMA_BALANCING
141 SEQ_printf(m, " %d", task_node(p));
142#endif
143#ifdef CONFIG_CGROUP_SCHED
144 SEQ_printf(m, " %s", task_group_path(task_group(p)));
145#endif
146
147 SEQ_printf(m, "\n");
148}
149
150static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
151{
152 struct task_struct *g, *p;
153 unsigned long flags;
154
155 SEQ_printf(m,
156 "\nrunnable tasks:\n"
157 " task PID tree-key switches prio"
158 " exec-runtime sum-exec sum-sleep\n"
159 "------------------------------------------------------"
160 "----------------------------------------------------\n");
161
162 read_lock_irqsave(&tasklist_lock, flags);
163
164 do_each_thread(g, p) {
165 if (task_cpu(p) != rq_cpu)
166 continue;
167
168 print_task(m, rq, p);
169 } while_each_thread(g, p);
170
171 read_unlock_irqrestore(&tasklist_lock, flags);
172}
173
174void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
175{
176 s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
177 spread, rq0_min_vruntime, spread0;
178 struct rq *rq = cpu_rq(cpu);
179 struct sched_entity *last;
180 unsigned long flags;
181
182#ifdef CONFIG_FAIR_GROUP_SCHED
183 SEQ_printf(m, "\ncfs_rq[%d]:%s\n", cpu, task_group_path(cfs_rq->tg));
184#else
185 SEQ_printf(m, "\ncfs_rq[%d]:\n", cpu);
186#endif
187 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "exec_clock",
188 SPLIT_NS(cfs_rq->exec_clock));
189
190 raw_spin_lock_irqsave(&rq->lock, flags);
191 if (cfs_rq->rb_leftmost)
192 MIN_vruntime = (__pick_first_entity(cfs_rq))->vruntime;
193 last = __pick_last_entity(cfs_rq);
194 if (last)
195 max_vruntime = last->vruntime;
196 min_vruntime = cfs_rq->min_vruntime;
197 rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
198 raw_spin_unlock_irqrestore(&rq->lock, flags);
199 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "MIN_vruntime",
200 SPLIT_NS(MIN_vruntime));
201 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "min_vruntime",
202 SPLIT_NS(min_vruntime));
203 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "max_vruntime",
204 SPLIT_NS(max_vruntime));
205 spread = max_vruntime - MIN_vruntime;
206 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread",
207 SPLIT_NS(spread));
208 spread0 = min_vruntime - rq0_min_vruntime;
209 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread0",
210 SPLIT_NS(spread0));
211 SEQ_printf(m, " .%-30s: %d\n", "nr_spread_over",
212 cfs_rq->nr_spread_over);
213 SEQ_printf(m, " .%-30s: %d\n", "nr_running", cfs_rq->nr_running);
214 SEQ_printf(m, " .%-30s: %ld\n", "load", cfs_rq->load.weight);
215#ifdef CONFIG_SMP
216 SEQ_printf(m, " .%-30s: %ld\n", "runnable_load_avg",
217 cfs_rq->runnable_load_avg);
218 SEQ_printf(m, " .%-30s: %ld\n", "blocked_load_avg",
219 cfs_rq->blocked_load_avg);
220#ifdef CONFIG_FAIR_GROUP_SCHED
221 SEQ_printf(m, " .%-30s: %ld\n", "tg_load_contrib",
222 cfs_rq->tg_load_contrib);
223 SEQ_printf(m, " .%-30s: %d\n", "tg_runnable_contrib",
224 cfs_rq->tg_runnable_contrib);
225 SEQ_printf(m, " .%-30s: %ld\n", "tg_load_avg",
226 atomic_long_read(&cfs_rq->tg->load_avg));
227 SEQ_printf(m, " .%-30s: %d\n", "tg->runnable_avg",
228 atomic_read(&cfs_rq->tg->runnable_avg));
229#endif
230#endif
231#ifdef CONFIG_CFS_BANDWIDTH
232 SEQ_printf(m, " .%-30s: %d\n", "tg->cfs_bandwidth.timer_active",
233 cfs_rq->tg->cfs_bandwidth.timer_active);
234 SEQ_printf(m, " .%-30s: %d\n", "throttled",
235 cfs_rq->throttled);
236 SEQ_printf(m, " .%-30s: %d\n", "throttle_count",
237 cfs_rq->throttle_count);
238#endif
239
240#ifdef CONFIG_FAIR_GROUP_SCHED
241 print_cfs_group_stats(m, cpu, cfs_rq->tg);
242#endif
243}
244
245void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
246{
247#ifdef CONFIG_RT_GROUP_SCHED
248 SEQ_printf(m, "\nrt_rq[%d]:%s\n", cpu, task_group_path(rt_rq->tg));
249#else
250 SEQ_printf(m, "\nrt_rq[%d]:\n", cpu);
251#endif
252
253#define P(x) \
254 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
255#define PN(x) \
256 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
257
258 P(rt_nr_running);
259 P(rt_throttled);
260 PN(rt_time);
261 PN(rt_runtime);
262
263#undef PN
264#undef P
265}
266
267extern __read_mostly int sched_clock_running;
268
269static void print_cpu(struct seq_file *m, int cpu)
270{
271 struct rq *rq = cpu_rq(cpu);
272 unsigned long flags;
273
274#ifdef CONFIG_X86
275 {
276 unsigned int freq = cpu_khz ? : 1;
277
278 SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
279 cpu, freq / 1000, (freq % 1000));
280 }
281#else
282 SEQ_printf(m, "cpu#%d\n", cpu);
283#endif
284
285#define P(x) \
286do { \
287 if (sizeof(rq->x) == 4) \
288 SEQ_printf(m, " .%-30s: %ld\n", #x, (long)(rq->x)); \
289 else \
290 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rq->x));\
291} while (0)
292
293#define PN(x) \
294 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
295
296 P(nr_running);
297 SEQ_printf(m, " .%-30s: %lu\n", "load",
298 rq->load.weight);
299 P(nr_switches);
300 P(nr_load_updates);
301 P(nr_uninterruptible);
302 PN(next_balance);
303 SEQ_printf(m, " .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr)));
304 PN(clock);
305 P(cpu_load[0]);
306 P(cpu_load[1]);
307 P(cpu_load[2]);
308 P(cpu_load[3]);
309 P(cpu_load[4]);
310#undef P
311#undef PN
312
313#ifdef CONFIG_SCHEDSTATS
314#define P(n) SEQ_printf(m, " .%-30s: %d\n", #n, rq->n);
315#define P64(n) SEQ_printf(m, " .%-30s: %Ld\n", #n, rq->n);
316
317 P(yld_count);
318
319 P(sched_count);
320 P(sched_goidle);
321#ifdef CONFIG_SMP
322 P64(avg_idle);
323 P64(max_idle_balance_cost);
324#endif
325
326 P(ttwu_count);
327 P(ttwu_local);
328
329#undef P
330#undef P64
331#endif
332 spin_lock_irqsave(&sched_debug_lock, flags);
333 print_cfs_stats(m, cpu);
334 print_rt_stats(m, cpu);
335
336 rcu_read_lock();
337 print_rq(m, rq, cpu);
338 rcu_read_unlock();
339 spin_unlock_irqrestore(&sched_debug_lock, flags);
340 SEQ_printf(m, "\n");
341}
342
343static const char *sched_tunable_scaling_names[] = {
344 "none",
345 "logaritmic",
346 "linear"
347};
348
349static void sched_debug_header(struct seq_file *m)
350{
351 u64 ktime, sched_clk, cpu_clk;
352 unsigned long flags;
353
354 local_irq_save(flags);
355 ktime = ktime_to_ns(ktime_get());
356 sched_clk = sched_clock();
357 cpu_clk = local_clock();
358 local_irq_restore(flags);
359
360 SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n",
361 init_utsname()->release,
362 (int)strcspn(init_utsname()->version, " "),
363 init_utsname()->version);
364
365#define P(x) \
366 SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
367#define PN(x) \
368 SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
369 PN(ktime);
370 PN(sched_clk);
371 PN(cpu_clk);
372 P(jiffies);
373#ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
374 P(sched_clock_stable());
375#endif
376#undef PN
377#undef P
378
379 SEQ_printf(m, "\n");
380 SEQ_printf(m, "sysctl_sched\n");
381
382#define P(x) \
383 SEQ_printf(m, " .%-40s: %Ld\n", #x, (long long)(x))
384#define PN(x) \
385 SEQ_printf(m, " .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
386 PN(sysctl_sched_latency);
387 PN(sysctl_sched_min_granularity);
388 PN(sysctl_sched_wakeup_granularity);
389 P(sysctl_sched_child_runs_first);
390 P(sysctl_sched_features);
391#undef PN
392#undef P
393
394 SEQ_printf(m, " .%-40s: %d (%s)\n",
395 "sysctl_sched_tunable_scaling",
396 sysctl_sched_tunable_scaling,
397 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
398 SEQ_printf(m, "\n");
399}
400
401static int sched_debug_show(struct seq_file *m, void *v)
402{
403 int cpu = (unsigned long)(v - 2);
404
405 if (cpu != -1)
406 print_cpu(m, cpu);
407 else
408 sched_debug_header(m);
409
410 return 0;
411}
412
413void sysrq_sched_debug_show(void)
414{
415 int cpu;
416
417 sched_debug_header(NULL);
418 for_each_online_cpu(cpu)
419 print_cpu(NULL, cpu);
420
421}
422
423/*
424 * This itererator needs some explanation.
425 * It returns 1 for the header position.
426 * This means 2 is cpu 0.
427 * In a hotplugged system some cpus, including cpu 0, may be missing so we have
428 * to use cpumask_* to iterate over the cpus.
429 */
430static void *sched_debug_start(struct seq_file *file, loff_t *offset)
431{
432 unsigned long n = *offset;
433
434 if (n == 0)
435 return (void *) 1;
436
437 n--;
438
439 if (n > 0)
440 n = cpumask_next(n - 1, cpu_online_mask);
441 else
442 n = cpumask_first(cpu_online_mask);
443
444 *offset = n + 1;
445
446 if (n < nr_cpu_ids)
447 return (void *)(unsigned long)(n + 2);
448 return NULL;
449}
450
451static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
452{
453 (*offset)++;
454 return sched_debug_start(file, offset);
455}
456
457static void sched_debug_stop(struct seq_file *file, void *data)
458{
459}
460
461static const struct seq_operations sched_debug_sops = {
462 .start = sched_debug_start,
463 .next = sched_debug_next,
464 .stop = sched_debug_stop,
465 .show = sched_debug_show,
466};
467
468static int sched_debug_release(struct inode *inode, struct file *file)
469{
470 seq_release(inode, file);
471
472 return 0;
473}
474
475static int sched_debug_open(struct inode *inode, struct file *filp)
476{
477 int ret = 0;
478
479 ret = seq_open(filp, &sched_debug_sops);
480
481 return ret;
482}
483
484static const struct file_operations sched_debug_fops = {
485 .open = sched_debug_open,
486 .read = seq_read,
487 .llseek = seq_lseek,
488 .release = sched_debug_release,
489};
490
491static int __init init_sched_debug_procfs(void)
492{
493 struct proc_dir_entry *pe;
494
495 pe = proc_create("sched_debug", 0444, NULL, &sched_debug_fops);
496 if (!pe)
497 return -ENOMEM;
498 return 0;
499}
500
501__initcall(init_sched_debug_procfs);
502
503#define __P(F) \
504 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)F)
505#define P(F) \
506 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)p->F)
507#define __PN(F) \
508 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
509#define PN(F) \
510 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
511
512
513static void sched_show_numa(struct task_struct *p, struct seq_file *m)
514{
515#ifdef CONFIG_NUMA_BALANCING
516 struct mempolicy *pol;
517 int node, i;
518
519 if (p->mm)
520 P(mm->numa_scan_seq);
521
522 task_lock(p);
523 pol = p->mempolicy;
524 if (pol && !(pol->flags & MPOL_F_MORON))
525 pol = NULL;
526 mpol_get(pol);
527 task_unlock(p);
528
529 SEQ_printf(m, "numa_migrations, %ld\n", xchg(&p->numa_pages_migrated, 0));
530
531 for_each_online_node(node) {
532 for (i = 0; i < 2; i++) {
533 unsigned long nr_faults = -1;
534 int cpu_current, home_node;
535
536 if (p->numa_faults_memory)
537 nr_faults = p->numa_faults_memory[2*node + i];
538
539 cpu_current = !i ? (task_node(p) == node) :
540 (pol && node_isset(node, pol->v.nodes));
541
542 home_node = (p->numa_preferred_nid == node);
543
544 SEQ_printf(m, "numa_faults_memory, %d, %d, %d, %d, %ld\n",
545 i, node, cpu_current, home_node, nr_faults);
546 }
547 }
548
549 mpol_put(pol);
550#endif
551}
552
553void proc_sched_show_task(struct task_struct *p, struct seq_file *m)
554{
555 unsigned long nr_switches;
556
557 SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr(p),
558 get_nr_threads(p));
559 SEQ_printf(m,
560 "---------------------------------------------------------"
561 "----------\n");
562#define __P(F) \
563 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)F)
564#define P(F) \
565 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)p->F)
566#define __PN(F) \
567 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
568#define PN(F) \
569 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
570
571 PN(se.exec_start);
572 PN(se.vruntime);
573 PN(se.sum_exec_runtime);
574
575 nr_switches = p->nvcsw + p->nivcsw;
576
577#ifdef CONFIG_SCHEDSTATS
578 PN(se.statistics.wait_start);
579 PN(se.statistics.sleep_start);
580 PN(se.statistics.block_start);
581 PN(se.statistics.sleep_max);
582 PN(se.statistics.block_max);
583 PN(se.statistics.exec_max);
584 PN(se.statistics.slice_max);
585 PN(se.statistics.wait_max);
586 PN(se.statistics.wait_sum);
587 P(se.statistics.wait_count);
588 PN(se.statistics.iowait_sum);
589 P(se.statistics.iowait_count);
590 P(se.nr_migrations);
591 P(se.statistics.nr_migrations_cold);
592 P(se.statistics.nr_failed_migrations_affine);
593 P(se.statistics.nr_failed_migrations_running);
594 P(se.statistics.nr_failed_migrations_hot);
595 P(se.statistics.nr_forced_migrations);
596 P(se.statistics.nr_wakeups);
597 P(se.statistics.nr_wakeups_sync);
598 P(se.statistics.nr_wakeups_migrate);
599 P(se.statistics.nr_wakeups_local);
600 P(se.statistics.nr_wakeups_remote);
601 P(se.statistics.nr_wakeups_affine);
602 P(se.statistics.nr_wakeups_affine_attempts);
603 P(se.statistics.nr_wakeups_passive);
604 P(se.statistics.nr_wakeups_idle);
605
606 {
607 u64 avg_atom, avg_per_cpu;
608
609 avg_atom = p->se.sum_exec_runtime;
610 if (nr_switches)
611 do_div(avg_atom, nr_switches);
612 else
613 avg_atom = -1LL;
614
615 avg_per_cpu = p->se.sum_exec_runtime;
616 if (p->se.nr_migrations) {
617 avg_per_cpu = div64_u64(avg_per_cpu,
618 p->se.nr_migrations);
619 } else {
620 avg_per_cpu = -1LL;
621 }
622
623 __PN(avg_atom);
624 __PN(avg_per_cpu);
625 }
626#endif
627 __P(nr_switches);
628 SEQ_printf(m, "%-45s:%21Ld\n",
629 "nr_voluntary_switches", (long long)p->nvcsw);
630 SEQ_printf(m, "%-45s:%21Ld\n",
631 "nr_involuntary_switches", (long long)p->nivcsw);
632
633 P(se.load.weight);
634#ifdef CONFIG_SMP
635 P(se.avg.runnable_avg_sum);
636 P(se.avg.runnable_avg_period);
637 P(se.avg.load_avg_contrib);
638 P(se.avg.decay_count);
639#endif
640 P(policy);
641 P(prio);
642#undef PN
643#undef __PN
644#undef P
645#undef __P
646
647 {
648 unsigned int this_cpu = raw_smp_processor_id();
649 u64 t0, t1;
650
651 t0 = cpu_clock(this_cpu);
652 t1 = cpu_clock(this_cpu);
653 SEQ_printf(m, "%-45s:%21Ld\n",
654 "clock-delta", (long long)(t1-t0));
655 }
656
657 sched_show_numa(p, m);
658}
659
660void proc_sched_set_task(struct task_struct *p)
661{
662#ifdef CONFIG_SCHEDSTATS
663 memset(&p->se.statistics, 0, sizeof(p->se.statistics));
664#endif
665}