1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * kernel/sched/debug.c
4 *
5 * Print the CFS rbtree and other debugging details
6 *
7 * Copyright(C) 2007, Red Hat, Inc., Ingo Molnar
8 */
9
10 /*
11 * This allows printing both to /proc/sched_debug and
12 * to the console
13 */
14 #define SEQ_printf(m, x...) \
15 do { \
16 if (m) \
17 seq_printf(m, x); \
18 else \
19 pr_cont(x); \
20 } while (0)
21
22 /*
23 * Ease the printing of nsec fields:
24 */
nsec_high(unsigned long long nsec)25 static 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
nsec_low(unsigned long long nsec)37 static 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
50 const char * const sched_feat_names[] = {
51 #include "features.h"
52 };
53 EXPORT_SYMBOL_GPL(sched_feat_names);
54
55 #undef SCHED_FEAT
56
sched_feat_show(struct seq_file * m,void * v)57 static int sched_feat_show(struct seq_file *m, void *v)
58 {
59 int i;
60
61 for (i = 0; i < __SCHED_FEAT_NR; i++) {
62 if (!(sysctl_sched_features & (1UL << i)))
63 seq_puts(m, "NO_");
64 seq_printf(m, "%s ", sched_feat_names[i]);
65 }
66 seq_puts(m, "\n");
67
68 return 0;
69 }
70
71 #ifdef CONFIG_JUMP_LABEL
72
73 #define jump_label_key__true STATIC_KEY_INIT_TRUE
74 #define jump_label_key__false STATIC_KEY_INIT_FALSE
75
76 #define SCHED_FEAT(name, enabled) \
77 jump_label_key__##enabled ,
78
79 struct static_key sched_feat_keys[__SCHED_FEAT_NR] = {
80 #include "features.h"
81 };
82 EXPORT_SYMBOL_GPL(sched_feat_keys);
83
84 #undef SCHED_FEAT
85
sched_feat_disable(int i)86 static void sched_feat_disable(int i)
87 {
88 static_key_disable_cpuslocked(&sched_feat_keys[i]);
89 }
90
sched_feat_enable(int i)91 static void sched_feat_enable(int i)
92 {
93 static_key_enable_cpuslocked(&sched_feat_keys[i]);
94 }
95 #else
sched_feat_disable(int i)96 static void sched_feat_disable(int i) { };
sched_feat_enable(int i)97 static void sched_feat_enable(int i) { };
98 #endif /* CONFIG_JUMP_LABEL */
99
sched_feat_set(char * cmp)100 static int sched_feat_set(char *cmp)
101 {
102 int i;
103 int neg = 0;
104
105 if (strncmp(cmp, "NO_", 3) == 0) {
106 neg = 1;
107 cmp += 3;
108 }
109
110 i = match_string(sched_feat_names, __SCHED_FEAT_NR, cmp);
111 if (i < 0)
112 return i;
113
114 if (neg) {
115 sysctl_sched_features &= ~(1UL << i);
116 sched_feat_disable(i);
117 } else {
118 sysctl_sched_features |= (1UL << i);
119 sched_feat_enable(i);
120 }
121
122 return 0;
123 }
124
125 static ssize_t
sched_feat_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)126 sched_feat_write(struct file *filp, const char __user *ubuf,
127 size_t cnt, loff_t *ppos)
128 {
129 char buf[64];
130 char *cmp;
131 int ret;
132 struct inode *inode;
133
134 if (cnt > 63)
135 cnt = 63;
136
137 if (copy_from_user(&buf, ubuf, cnt))
138 return -EFAULT;
139
140 buf[cnt] = 0;
141 cmp = strstrip(buf);
142
143 /* Ensure the static_key remains in a consistent state */
144 inode = file_inode(filp);
145 cpus_read_lock();
146 inode_lock(inode);
147 ret = sched_feat_set(cmp);
148 inode_unlock(inode);
149 cpus_read_unlock();
150 if (ret < 0)
151 return ret;
152
153 *ppos += cnt;
154
155 return cnt;
156 }
157
sched_feat_open(struct inode * inode,struct file * filp)158 static int sched_feat_open(struct inode *inode, struct file *filp)
159 {
160 return single_open(filp, sched_feat_show, NULL);
161 }
162
163 static const struct file_operations sched_feat_fops = {
164 .open = sched_feat_open,
165 .write = sched_feat_write,
166 .read = seq_read,
167 .llseek = seq_lseek,
168 .release = single_release,
169 };
170
171 #ifdef CONFIG_SMP
172
sched_scaling_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)173 static ssize_t sched_scaling_write(struct file *filp, const char __user *ubuf,
174 size_t cnt, loff_t *ppos)
175 {
176 char buf[16];
177 unsigned int scaling;
178
179 if (cnt > 15)
180 cnt = 15;
181
182 if (copy_from_user(&buf, ubuf, cnt))
183 return -EFAULT;
184 buf[cnt] = '\0';
185
186 if (kstrtouint(buf, 10, &scaling))
187 return -EINVAL;
188
189 if (scaling >= SCHED_TUNABLESCALING_END)
190 return -EINVAL;
191
192 sysctl_sched_tunable_scaling = scaling;
193 if (sched_update_scaling())
194 return -EINVAL;
195
196 *ppos += cnt;
197 return cnt;
198 }
199
sched_scaling_show(struct seq_file * m,void * v)200 static int sched_scaling_show(struct seq_file *m, void *v)
201 {
202 seq_printf(m, "%d\n", sysctl_sched_tunable_scaling);
203 return 0;
204 }
205
sched_scaling_open(struct inode * inode,struct file * filp)206 static int sched_scaling_open(struct inode *inode, struct file *filp)
207 {
208 return single_open(filp, sched_scaling_show, NULL);
209 }
210
211 static const struct file_operations sched_scaling_fops = {
212 .open = sched_scaling_open,
213 .write = sched_scaling_write,
214 .read = seq_read,
215 .llseek = seq_lseek,
216 .release = single_release,
217 };
218
219 #endif /* SMP */
220
221 #ifdef CONFIG_PREEMPT_DYNAMIC
222
sched_dynamic_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)223 static ssize_t sched_dynamic_write(struct file *filp, const char __user *ubuf,
224 size_t cnt, loff_t *ppos)
225 {
226 char buf[16];
227 int mode;
228
229 if (cnt > 15)
230 cnt = 15;
231
232 if (copy_from_user(&buf, ubuf, cnt))
233 return -EFAULT;
234
235 buf[cnt] = 0;
236 mode = sched_dynamic_mode(strstrip(buf));
237 if (mode < 0)
238 return mode;
239
240 sched_dynamic_update(mode);
241
242 *ppos += cnt;
243
244 return cnt;
245 }
246
sched_dynamic_show(struct seq_file * m,void * v)247 static int sched_dynamic_show(struct seq_file *m, void *v)
248 {
249 static const char * preempt_modes[] = {
250 "none", "voluntary", "full"
251 };
252 int i;
253
254 for (i = 0; i < ARRAY_SIZE(preempt_modes); i++) {
255 if (preempt_dynamic_mode == i)
256 seq_puts(m, "(");
257 seq_puts(m, preempt_modes[i]);
258 if (preempt_dynamic_mode == i)
259 seq_puts(m, ")");
260
261 seq_puts(m, " ");
262 }
263
264 seq_puts(m, "\n");
265 return 0;
266 }
267
sched_dynamic_open(struct inode * inode,struct file * filp)268 static int sched_dynamic_open(struct inode *inode, struct file *filp)
269 {
270 return single_open(filp, sched_dynamic_show, NULL);
271 }
272
273 static 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
285 #ifdef CONFIG_SMP
286 static struct dentry *sd_dentry;
287
288
sched_verbose_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)289 static ssize_t sched_verbose_write(struct file *filp, const char __user *ubuf,
290 size_t cnt, loff_t *ppos)
291 {
292 ssize_t result;
293 bool orig;
294
295 cpus_read_lock();
296 mutex_lock(&sched_domains_mutex);
297
298 orig = sched_debug_verbose;
299 result = debugfs_write_file_bool(filp, ubuf, cnt, ppos);
300
301 if (sched_debug_verbose && !orig)
302 update_sched_domain_debugfs();
303 else if (!sched_debug_verbose && orig) {
304 debugfs_remove(sd_dentry);
305 sd_dentry = NULL;
306 }
307
308 mutex_unlock(&sched_domains_mutex);
309 cpus_read_unlock();
310
311 return result;
312 }
313 #else
314 #define sched_verbose_write debugfs_write_file_bool
315 #endif
316
317 static const struct file_operations sched_verbose_fops = {
318 .read = debugfs_read_file_bool,
319 .write = sched_verbose_write,
320 .open = simple_open,
321 .llseek = default_llseek,
322 };
323
324 static const struct seq_operations sched_debug_sops;
325
sched_debug_open(struct inode * inode,struct file * filp)326 static int sched_debug_open(struct inode *inode, struct file *filp)
327 {
328 return seq_open(filp, &sched_debug_sops);
329 }
330
331 static const struct file_operations sched_debug_fops = {
332 .open = sched_debug_open,
333 .read = seq_read,
334 .llseek = seq_lseek,
335 .release = seq_release,
336 };
337
338 static struct dentry *debugfs_sched;
339
sched_init_debug(void)340 static __init int sched_init_debug(void)
341 {
342 struct dentry __maybe_unused *numa;
343
344 debugfs_sched = debugfs_create_dir("sched", NULL);
345
346 debugfs_create_file("features", 0644, debugfs_sched, NULL, &sched_feat_fops);
347 debugfs_create_file_unsafe("verbose", 0644, debugfs_sched, &sched_debug_verbose, &sched_verbose_fops);
348 #ifdef CONFIG_PREEMPT_DYNAMIC
349 debugfs_create_file("preempt", 0644, debugfs_sched, NULL, &sched_dynamic_fops);
350 #endif
351
352 debugfs_create_u32("base_slice_ns", 0644, debugfs_sched, &sysctl_sched_base_slice);
353
354 debugfs_create_u32("latency_warn_ms", 0644, debugfs_sched, &sysctl_resched_latency_warn_ms);
355 debugfs_create_u32("latency_warn_once", 0644, debugfs_sched, &sysctl_resched_latency_warn_once);
356
357 #ifdef CONFIG_SMP
358 debugfs_create_file("tunable_scaling", 0644, debugfs_sched, NULL, &sched_scaling_fops);
359 debugfs_create_u32("migration_cost_ns", 0644, debugfs_sched, &sysctl_sched_migration_cost);
360 debugfs_create_u32("nr_migrate", 0644, debugfs_sched, &sysctl_sched_nr_migrate);
361
362 mutex_lock(&sched_domains_mutex);
363 update_sched_domain_debugfs();
364 mutex_unlock(&sched_domains_mutex);
365 #endif
366
367 #ifdef CONFIG_NUMA_BALANCING
368 numa = debugfs_create_dir("numa_balancing", debugfs_sched);
369
370 debugfs_create_u32("scan_delay_ms", 0644, numa, &sysctl_numa_balancing_scan_delay);
371 debugfs_create_u32("scan_period_min_ms", 0644, numa, &sysctl_numa_balancing_scan_period_min);
372 debugfs_create_u32("scan_period_max_ms", 0644, numa, &sysctl_numa_balancing_scan_period_max);
373 debugfs_create_u32("scan_size_mb", 0644, numa, &sysctl_numa_balancing_scan_size);
374 debugfs_create_u32("hot_threshold_ms", 0644, numa, &sysctl_numa_balancing_hot_threshold);
375 #endif
376
377 debugfs_create_file("debug", 0444, debugfs_sched, NULL, &sched_debug_fops);
378
379 return 0;
380 }
381 late_initcall(sched_init_debug);
382
383 #ifdef CONFIG_SMP
384
385 static cpumask_var_t sd_sysctl_cpus;
386
sd_flags_show(struct seq_file * m,void * v)387 static int sd_flags_show(struct seq_file *m, void *v)
388 {
389 unsigned long flags = *(unsigned int *)m->private;
390 int idx;
391
392 for_each_set_bit(idx, &flags, __SD_FLAG_CNT) {
393 seq_puts(m, sd_flag_debug[idx].name);
394 seq_puts(m, " ");
395 }
396 seq_puts(m, "\n");
397
398 return 0;
399 }
400
sd_flags_open(struct inode * inode,struct file * file)401 static int sd_flags_open(struct inode *inode, struct file *file)
402 {
403 return single_open(file, sd_flags_show, inode->i_private);
404 }
405
406 static const struct file_operations sd_flags_fops = {
407 .open = sd_flags_open,
408 .read = seq_read,
409 .llseek = seq_lseek,
410 .release = single_release,
411 };
412
register_sd(struct sched_domain * sd,struct dentry * parent)413 static void register_sd(struct sched_domain *sd, struct dentry *parent)
414 {
415 #define SDM(type, mode, member) \
416 debugfs_create_##type(#member, mode, parent, &sd->member)
417
418 SDM(ulong, 0644, min_interval);
419 SDM(ulong, 0644, max_interval);
420 SDM(u64, 0644, max_newidle_lb_cost);
421 SDM(u32, 0644, busy_factor);
422 SDM(u32, 0644, imbalance_pct);
423 SDM(u32, 0644, cache_nice_tries);
424 SDM(str, 0444, name);
425
426 #undef SDM
427
428 debugfs_create_file("flags", 0444, parent, &sd->flags, &sd_flags_fops);
429 debugfs_create_file("groups_flags", 0444, parent, &sd->groups->flags, &sd_flags_fops);
430 }
431
update_sched_domain_debugfs(void)432 void update_sched_domain_debugfs(void)
433 {
434 int cpu, i;
435
436 /*
437 * This can unfortunately be invoked before sched_debug_init() creates
438 * the debug directory. Don't touch sd_sysctl_cpus until then.
439 */
440 if (!debugfs_sched)
441 return;
442
443 if (!sched_debug_verbose)
444 return;
445
446 if (!cpumask_available(sd_sysctl_cpus)) {
447 if (!alloc_cpumask_var(&sd_sysctl_cpus, GFP_KERNEL))
448 return;
449 cpumask_copy(sd_sysctl_cpus, cpu_possible_mask);
450 }
451
452 if (!sd_dentry) {
453 sd_dentry = debugfs_create_dir("domains", debugfs_sched);
454
455 /* rebuild sd_sysctl_cpus if empty since it gets cleared below */
456 if (cpumask_empty(sd_sysctl_cpus))
457 cpumask_copy(sd_sysctl_cpus, cpu_online_mask);
458 }
459
460 for_each_cpu(cpu, sd_sysctl_cpus) {
461 struct sched_domain *sd;
462 struct dentry *d_cpu;
463 char buf[32];
464
465 snprintf(buf, sizeof(buf), "cpu%d", cpu);
466 debugfs_lookup_and_remove(buf, sd_dentry);
467 d_cpu = debugfs_create_dir(buf, sd_dentry);
468
469 i = 0;
470 for_each_domain(cpu, sd) {
471 struct dentry *d_sd;
472
473 snprintf(buf, sizeof(buf), "domain%d", i);
474 d_sd = debugfs_create_dir(buf, d_cpu);
475
476 register_sd(sd, d_sd);
477 i++;
478 }
479
480 __cpumask_clear_cpu(cpu, sd_sysctl_cpus);
481 }
482 }
483
dirty_sched_domain_sysctl(int cpu)484 void dirty_sched_domain_sysctl(int cpu)
485 {
486 if (cpumask_available(sd_sysctl_cpus))
487 __cpumask_set_cpu(cpu, sd_sysctl_cpus);
488 }
489
490 #endif /* CONFIG_SMP */
491
492 #ifdef CONFIG_FAIR_GROUP_SCHED
print_cfs_group_stats(struct seq_file * m,int cpu,struct task_group * tg)493 static void print_cfs_group_stats(struct seq_file *m, int cpu, struct task_group *tg)
494 {
495 struct sched_entity *se = tg->se[cpu];
496
497 #define P(F) SEQ_printf(m, " .%-30s: %lld\n", #F, (long long)F)
498 #define P_SCHEDSTAT(F) SEQ_printf(m, " .%-30s: %lld\n", \
499 #F, (long long)schedstat_val(stats->F))
500 #define PN(F) SEQ_printf(m, " .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F))
501 #define PN_SCHEDSTAT(F) SEQ_printf(m, " .%-30s: %lld.%06ld\n", \
502 #F, SPLIT_NS((long long)schedstat_val(stats->F)))
503
504 if (!se)
505 return;
506
507 PN(se->exec_start);
508 PN(se->vruntime);
509 PN(se->sum_exec_runtime);
510
511 if (schedstat_enabled()) {
512 struct sched_statistics *stats;
513 stats = __schedstats_from_se(se);
514
515 PN_SCHEDSTAT(wait_start);
516 PN_SCHEDSTAT(sleep_start);
517 PN_SCHEDSTAT(block_start);
518 PN_SCHEDSTAT(sleep_max);
519 PN_SCHEDSTAT(block_max);
520 PN_SCHEDSTAT(exec_max);
521 PN_SCHEDSTAT(slice_max);
522 PN_SCHEDSTAT(wait_max);
523 PN_SCHEDSTAT(wait_sum);
524 P_SCHEDSTAT(wait_count);
525 }
526
527 P(se->load.weight);
528 #ifdef CONFIG_SMP
529 P(se->avg.load_avg);
530 P(se->avg.util_avg);
531 P(se->avg.runnable_avg);
532 #endif
533
534 #undef PN_SCHEDSTAT
535 #undef PN
536 #undef P_SCHEDSTAT
537 #undef P
538 }
539 #endif
540
541 #ifdef CONFIG_CGROUP_SCHED
542 static DEFINE_SPINLOCK(sched_debug_lock);
543 static char group_path[PATH_MAX];
544
task_group_path(struct task_group * tg,char * path,int plen)545 static void task_group_path(struct task_group *tg, char *path, int plen)
546 {
547 if (autogroup_path(tg, path, plen))
548 return;
549
550 cgroup_path(tg->css.cgroup, path, plen);
551 }
552
553 /*
554 * Only 1 SEQ_printf_task_group_path() caller can use the full length
555 * group_path[] for cgroup path. Other simultaneous callers will have
556 * to use a shorter stack buffer. A "..." suffix is appended at the end
557 * of the stack buffer so that it will show up in case the output length
558 * matches the given buffer size to indicate possible path name truncation.
559 */
560 #define SEQ_printf_task_group_path(m, tg, fmt...) \
561 { \
562 if (spin_trylock(&sched_debug_lock)) { \
563 task_group_path(tg, group_path, sizeof(group_path)); \
564 SEQ_printf(m, fmt, group_path); \
565 spin_unlock(&sched_debug_lock); \
566 } else { \
567 char buf[128]; \
568 char *bufend = buf + sizeof(buf) - 3; \
569 task_group_path(tg, buf, bufend - buf); \
570 strcpy(bufend - 1, "..."); \
571 SEQ_printf(m, fmt, buf); \
572 } \
573 }
574 #endif
575
576 static void
print_task(struct seq_file * m,struct rq * rq,struct task_struct * p)577 print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
578 {
579 if (task_current(rq, p))
580 SEQ_printf(m, ">R");
581 else
582 SEQ_printf(m, " %c", task_state_to_char(p));
583
584 SEQ_printf(m, "%15s %5d %9Ld.%06ld %c %9Ld.%06ld %9Ld.%06ld %9Ld.%06ld %9Ld %5d ",
585 p->comm, task_pid_nr(p),
586 SPLIT_NS(p->se.vruntime),
587 entity_eligible(cfs_rq_of(&p->se), &p->se) ? 'E' : 'N',
588 SPLIT_NS(p->se.deadline),
589 SPLIT_NS(p->se.slice),
590 SPLIT_NS(p->se.sum_exec_runtime),
591 (long long)(p->nvcsw + p->nivcsw),
592 p->prio);
593
594 SEQ_printf(m, "%9lld.%06ld %9lld.%06ld %9lld.%06ld %9lld.%06ld",
595 SPLIT_NS(schedstat_val_or_zero(p->stats.wait_sum)),
596 SPLIT_NS(p->se.sum_exec_runtime),
597 SPLIT_NS(schedstat_val_or_zero(p->stats.sum_sleep_runtime)),
598 SPLIT_NS(schedstat_val_or_zero(p->stats.sum_block_runtime)));
599
600 #ifdef CONFIG_NUMA_BALANCING
601 SEQ_printf(m, " %d %d", task_node(p), task_numa_group_id(p));
602 #endif
603 #ifdef CONFIG_CGROUP_SCHED
604 SEQ_printf_task_group_path(m, task_group(p), " %s")
605 #endif
606
607 SEQ_printf(m, "\n");
608 }
609
print_rq(struct seq_file * m,struct rq * rq,int rq_cpu)610 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
611 {
612 struct task_struct *g, *p;
613
614 SEQ_printf(m, "\n");
615 SEQ_printf(m, "runnable tasks:\n");
616 SEQ_printf(m, " S task PID tree-key switches prio"
617 " wait-time sum-exec sum-sleep\n");
618 SEQ_printf(m, "-------------------------------------------------------"
619 "------------------------------------------------------\n");
620
621 rcu_read_lock();
622 for_each_process_thread(g, p) {
623 if (task_cpu(p) != rq_cpu)
624 continue;
625
626 print_task(m, rq, p);
627 }
628 rcu_read_unlock();
629 }
630
print_cfs_rq(struct seq_file * m,int cpu,struct cfs_rq * cfs_rq)631 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
632 {
633 s64 left_vruntime = -1, min_vruntime, right_vruntime = -1, left_deadline = -1, spread;
634 struct sched_entity *last, *first, *root;
635 struct rq *rq = cpu_rq(cpu);
636 unsigned long flags;
637
638 #ifdef CONFIG_FAIR_GROUP_SCHED
639 SEQ_printf(m, "\n");
640 SEQ_printf_task_group_path(m, cfs_rq->tg, "cfs_rq[%d]:%s\n", cpu);
641 #else
642 SEQ_printf(m, "\n");
643 SEQ_printf(m, "cfs_rq[%d]:\n", cpu);
644 #endif
645 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "exec_clock",
646 SPLIT_NS(cfs_rq->exec_clock));
647
648 raw_spin_rq_lock_irqsave(rq, flags);
649 root = __pick_root_entity(cfs_rq);
650 if (root)
651 left_vruntime = root->min_vruntime;
652 first = __pick_first_entity(cfs_rq);
653 if (first)
654 left_deadline = first->deadline;
655 last = __pick_last_entity(cfs_rq);
656 if (last)
657 right_vruntime = last->vruntime;
658 min_vruntime = cfs_rq->min_vruntime;
659 raw_spin_rq_unlock_irqrestore(rq, flags);
660
661 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "left_deadline",
662 SPLIT_NS(left_deadline));
663 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "left_vruntime",
664 SPLIT_NS(left_vruntime));
665 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "min_vruntime",
666 SPLIT_NS(min_vruntime));
667 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "avg_vruntime",
668 SPLIT_NS(avg_vruntime(cfs_rq)));
669 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "right_vruntime",
670 SPLIT_NS(right_vruntime));
671 spread = right_vruntime - left_vruntime;
672 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread", SPLIT_NS(spread));
673 SEQ_printf(m, " .%-30s: %d\n", "nr_spread_over",
674 cfs_rq->nr_spread_over);
675 SEQ_printf(m, " .%-30s: %d\n", "nr_running", cfs_rq->nr_running);
676 SEQ_printf(m, " .%-30s: %d\n", "h_nr_running", cfs_rq->h_nr_running);
677 SEQ_printf(m, " .%-30s: %d\n", "idle_nr_running",
678 cfs_rq->idle_nr_running);
679 SEQ_printf(m, " .%-30s: %d\n", "idle_h_nr_running",
680 cfs_rq->idle_h_nr_running);
681 SEQ_printf(m, " .%-30s: %ld\n", "load", cfs_rq->load.weight);
682 #ifdef CONFIG_SMP
683 SEQ_printf(m, " .%-30s: %lu\n", "load_avg",
684 cfs_rq->avg.load_avg);
685 SEQ_printf(m, " .%-30s: %lu\n", "runnable_avg",
686 cfs_rq->avg.runnable_avg);
687 SEQ_printf(m, " .%-30s: %lu\n", "util_avg",
688 cfs_rq->avg.util_avg);
689 SEQ_printf(m, " .%-30s: %u\n", "util_est",
690 cfs_rq->avg.util_est);
691 SEQ_printf(m, " .%-30s: %ld\n", "removed.load_avg",
692 cfs_rq->removed.load_avg);
693 SEQ_printf(m, " .%-30s: %ld\n", "removed.util_avg",
694 cfs_rq->removed.util_avg);
695 SEQ_printf(m, " .%-30s: %ld\n", "removed.runnable_avg",
696 cfs_rq->removed.runnable_avg);
697 #ifdef CONFIG_FAIR_GROUP_SCHED
698 SEQ_printf(m, " .%-30s: %lu\n", "tg_load_avg_contrib",
699 cfs_rq->tg_load_avg_contrib);
700 SEQ_printf(m, " .%-30s: %ld\n", "tg_load_avg",
701 atomic_long_read(&cfs_rq->tg->load_avg));
702 #endif
703 #endif
704 #ifdef CONFIG_CFS_BANDWIDTH
705 SEQ_printf(m, " .%-30s: %d\n", "throttled",
706 cfs_rq->throttled);
707 SEQ_printf(m, " .%-30s: %d\n", "throttle_count",
708 cfs_rq->throttle_count);
709 #endif
710
711 #ifdef CONFIG_FAIR_GROUP_SCHED
712 print_cfs_group_stats(m, cpu, cfs_rq->tg);
713 #endif
714 }
715
print_rt_rq(struct seq_file * m,int cpu,struct rt_rq * rt_rq)716 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
717 {
718 #ifdef CONFIG_RT_GROUP_SCHED
719 SEQ_printf(m, "\n");
720 SEQ_printf_task_group_path(m, rt_rq->tg, "rt_rq[%d]:%s\n", cpu);
721 #else
722 SEQ_printf(m, "\n");
723 SEQ_printf(m, "rt_rq[%d]:\n", cpu);
724 #endif
725
726 #define P(x) \
727 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
728 #define PU(x) \
729 SEQ_printf(m, " .%-30s: %lu\n", #x, (unsigned long)(rt_rq->x))
730 #define PN(x) \
731 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
732
733 PU(rt_nr_running);
734 #ifdef CONFIG_SMP
735 PU(rt_nr_migratory);
736 #endif
737 P(rt_throttled);
738 PN(rt_time);
739 PN(rt_runtime);
740
741 #undef PN
742 #undef PU
743 #undef P
744 }
745
print_dl_rq(struct seq_file * m,int cpu,struct dl_rq * dl_rq)746 void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq)
747 {
748 struct dl_bw *dl_bw;
749
750 SEQ_printf(m, "\n");
751 SEQ_printf(m, "dl_rq[%d]:\n", cpu);
752
753 #define PU(x) \
754 SEQ_printf(m, " .%-30s: %lu\n", #x, (unsigned long)(dl_rq->x))
755
756 PU(dl_nr_running);
757 #ifdef CONFIG_SMP
758 PU(dl_nr_migratory);
759 dl_bw = &cpu_rq(cpu)->rd->dl_bw;
760 #else
761 dl_bw = &dl_rq->dl_bw;
762 #endif
763 SEQ_printf(m, " .%-30s: %lld\n", "dl_bw->bw", dl_bw->bw);
764 SEQ_printf(m, " .%-30s: %lld\n", "dl_bw->total_bw", dl_bw->total_bw);
765
766 #undef PU
767 }
768
print_cpu(struct seq_file * m,int cpu)769 static void print_cpu(struct seq_file *m, int cpu)
770 {
771 struct rq *rq = cpu_rq(cpu);
772
773 #ifdef CONFIG_X86
774 {
775 unsigned int freq = cpu_khz ? : 1;
776
777 SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
778 cpu, freq / 1000, (freq % 1000));
779 }
780 #else
781 SEQ_printf(m, "cpu#%d\n", cpu);
782 #endif
783
784 #define P(x) \
785 do { \
786 if (sizeof(rq->x) == 4) \
787 SEQ_printf(m, " .%-30s: %d\n", #x, (int)(rq->x)); \
788 else \
789 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rq->x));\
790 } while (0)
791
792 #define PN(x) \
793 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
794
795 P(nr_running);
796 P(nr_switches);
797 P(nr_uninterruptible);
798 PN(next_balance);
799 SEQ_printf(m, " .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr)));
800 PN(clock);
801 PN(clock_task);
802 #undef P
803 #undef PN
804
805 #ifdef CONFIG_SMP
806 #define P64(n) SEQ_printf(m, " .%-30s: %Ld\n", #n, rq->n);
807 P64(avg_idle);
808 P64(max_idle_balance_cost);
809 #undef P64
810 #endif
811
812 #define P(n) SEQ_printf(m, " .%-30s: %d\n", #n, schedstat_val(rq->n));
813 if (schedstat_enabled()) {
814 P(yld_count);
815 P(sched_count);
816 P(sched_goidle);
817 P(ttwu_count);
818 P(ttwu_local);
819 }
820 #undef P
821
822 print_cfs_stats(m, cpu);
823 print_rt_stats(m, cpu);
824 print_dl_stats(m, cpu);
825
826 print_rq(m, rq, cpu);
827 SEQ_printf(m, "\n");
828 }
829
830 static const char *sched_tunable_scaling_names[] = {
831 "none",
832 "logarithmic",
833 "linear"
834 };
835
sched_debug_header(struct seq_file * m)836 static void sched_debug_header(struct seq_file *m)
837 {
838 u64 ktime, sched_clk, cpu_clk;
839 unsigned long flags;
840
841 local_irq_save(flags);
842 ktime = ktime_to_ns(ktime_get());
843 sched_clk = sched_clock();
844 cpu_clk = local_clock();
845 local_irq_restore(flags);
846
847 SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n",
848 init_utsname()->release,
849 (int)strcspn(init_utsname()->version, " "),
850 init_utsname()->version);
851
852 #define P(x) \
853 SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
854 #define PN(x) \
855 SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
856 PN(ktime);
857 PN(sched_clk);
858 PN(cpu_clk);
859 P(jiffies);
860 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
861 P(sched_clock_stable());
862 #endif
863 #undef PN
864 #undef P
865
866 SEQ_printf(m, "\n");
867 SEQ_printf(m, "sysctl_sched\n");
868
869 #define P(x) \
870 SEQ_printf(m, " .%-40s: %Ld\n", #x, (long long)(x))
871 #define PN(x) \
872 SEQ_printf(m, " .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
873 PN(sysctl_sched_base_slice);
874 P(sysctl_sched_child_runs_first);
875 P(sysctl_sched_features);
876 #undef PN
877 #undef P
878
879 SEQ_printf(m, " .%-40s: %d (%s)\n",
880 "sysctl_sched_tunable_scaling",
881 sysctl_sched_tunable_scaling,
882 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
883 SEQ_printf(m, "\n");
884 }
885
sched_debug_show(struct seq_file * m,void * v)886 static int sched_debug_show(struct seq_file *m, void *v)
887 {
888 int cpu = (unsigned long)(v - 2);
889
890 if (cpu != -1)
891 print_cpu(m, cpu);
892 else
893 sched_debug_header(m);
894
895 return 0;
896 }
897
sysrq_sched_debug_show(void)898 void sysrq_sched_debug_show(void)
899 {
900 int cpu;
901
902 sched_debug_header(NULL);
903 for_each_online_cpu(cpu) {
904 /*
905 * Need to reset softlockup watchdogs on all CPUs, because
906 * another CPU might be blocked waiting for us to process
907 * an IPI or stop_machine.
908 */
909 touch_nmi_watchdog();
910 touch_all_softlockup_watchdogs();
911 print_cpu(NULL, cpu);
912 }
913 }
914
915 /*
916 * This iterator needs some explanation.
917 * It returns 1 for the header position.
918 * This means 2 is CPU 0.
919 * In a hotplugged system some CPUs, including CPU 0, may be missing so we have
920 * to use cpumask_* to iterate over the CPUs.
921 */
sched_debug_start(struct seq_file * file,loff_t * offset)922 static void *sched_debug_start(struct seq_file *file, loff_t *offset)
923 {
924 unsigned long n = *offset;
925
926 if (n == 0)
927 return (void *) 1;
928
929 n--;
930
931 if (n > 0)
932 n = cpumask_next(n - 1, cpu_online_mask);
933 else
934 n = cpumask_first(cpu_online_mask);
935
936 *offset = n + 1;
937
938 if (n < nr_cpu_ids)
939 return (void *)(unsigned long)(n + 2);
940
941 return NULL;
942 }
943
sched_debug_next(struct seq_file * file,void * data,loff_t * offset)944 static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
945 {
946 (*offset)++;
947 return sched_debug_start(file, offset);
948 }
949
sched_debug_stop(struct seq_file * file,void * data)950 static void sched_debug_stop(struct seq_file *file, void *data)
951 {
952 }
953
954 static const struct seq_operations sched_debug_sops = {
955 .start = sched_debug_start,
956 .next = sched_debug_next,
957 .stop = sched_debug_stop,
958 .show = sched_debug_show,
959 };
960
961 #define __PS(S, F) SEQ_printf(m, "%-45s:%21Ld\n", S, (long long)(F))
962 #define __P(F) __PS(#F, F)
963 #define P(F) __PS(#F, p->F)
964 #define PM(F, M) __PS(#F, p->F & (M))
965 #define __PSN(S, F) SEQ_printf(m, "%-45s:%14Ld.%06ld\n", S, SPLIT_NS((long long)(F)))
966 #define __PN(F) __PSN(#F, F)
967 #define PN(F) __PSN(#F, p->F)
968
969
970 #ifdef CONFIG_NUMA_BALANCING
print_numa_stats(struct seq_file * m,int node,unsigned long tsf,unsigned long tpf,unsigned long gsf,unsigned long gpf)971 void print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
972 unsigned long tpf, unsigned long gsf, unsigned long gpf)
973 {
974 SEQ_printf(m, "numa_faults node=%d ", node);
975 SEQ_printf(m, "task_private=%lu task_shared=%lu ", tpf, tsf);
976 SEQ_printf(m, "group_private=%lu group_shared=%lu\n", gpf, gsf);
977 }
978 #endif
979
980
sched_show_numa(struct task_struct * p,struct seq_file * m)981 static void sched_show_numa(struct task_struct *p, struct seq_file *m)
982 {
983 #ifdef CONFIG_NUMA_BALANCING
984 if (p->mm)
985 P(mm->numa_scan_seq);
986
987 P(numa_pages_migrated);
988 P(numa_preferred_nid);
989 P(total_numa_faults);
990 SEQ_printf(m, "current_node=%d, numa_group_id=%d\n",
991 task_node(p), task_numa_group_id(p));
992 show_numa_stats(p, m);
993 #endif
994 }
995
proc_sched_show_task(struct task_struct * p,struct pid_namespace * ns,struct seq_file * m)996 void proc_sched_show_task(struct task_struct *p, struct pid_namespace *ns,
997 struct seq_file *m)
998 {
999 unsigned long nr_switches;
1000
1001 SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr_ns(p, ns),
1002 get_nr_threads(p));
1003 SEQ_printf(m,
1004 "---------------------------------------------------------"
1005 "----------\n");
1006
1007 #define P_SCHEDSTAT(F) __PS(#F, schedstat_val(p->stats.F))
1008 #define PN_SCHEDSTAT(F) __PSN(#F, schedstat_val(p->stats.F))
1009
1010 PN(se.exec_start);
1011 PN(se.vruntime);
1012 PN(se.sum_exec_runtime);
1013
1014 nr_switches = p->nvcsw + p->nivcsw;
1015
1016 P(se.nr_migrations);
1017
1018 if (schedstat_enabled()) {
1019 u64 avg_atom, avg_per_cpu;
1020
1021 PN_SCHEDSTAT(sum_sleep_runtime);
1022 PN_SCHEDSTAT(sum_block_runtime);
1023 PN_SCHEDSTAT(wait_start);
1024 PN_SCHEDSTAT(sleep_start);
1025 PN_SCHEDSTAT(block_start);
1026 PN_SCHEDSTAT(sleep_max);
1027 PN_SCHEDSTAT(block_max);
1028 PN_SCHEDSTAT(exec_max);
1029 PN_SCHEDSTAT(slice_max);
1030 PN_SCHEDSTAT(wait_max);
1031 PN_SCHEDSTAT(wait_sum);
1032 P_SCHEDSTAT(wait_count);
1033 PN_SCHEDSTAT(iowait_sum);
1034 P_SCHEDSTAT(iowait_count);
1035 P_SCHEDSTAT(nr_migrations_cold);
1036 P_SCHEDSTAT(nr_failed_migrations_affine);
1037 P_SCHEDSTAT(nr_failed_migrations_running);
1038 P_SCHEDSTAT(nr_failed_migrations_hot);
1039 P_SCHEDSTAT(nr_forced_migrations);
1040 P_SCHEDSTAT(nr_wakeups);
1041 P_SCHEDSTAT(nr_wakeups_sync);
1042 P_SCHEDSTAT(nr_wakeups_migrate);
1043 P_SCHEDSTAT(nr_wakeups_local);
1044 P_SCHEDSTAT(nr_wakeups_remote);
1045 P_SCHEDSTAT(nr_wakeups_affine);
1046 P_SCHEDSTAT(nr_wakeups_affine_attempts);
1047 P_SCHEDSTAT(nr_wakeups_passive);
1048 P_SCHEDSTAT(nr_wakeups_idle);
1049
1050 avg_atom = p->se.sum_exec_runtime;
1051 if (nr_switches)
1052 avg_atom = div64_ul(avg_atom, nr_switches);
1053 else
1054 avg_atom = -1LL;
1055
1056 avg_per_cpu = p->se.sum_exec_runtime;
1057 if (p->se.nr_migrations) {
1058 avg_per_cpu = div64_u64(avg_per_cpu,
1059 p->se.nr_migrations);
1060 } else {
1061 avg_per_cpu = -1LL;
1062 }
1063
1064 __PN(avg_atom);
1065 __PN(avg_per_cpu);
1066
1067 #ifdef CONFIG_SCHED_CORE
1068 PN_SCHEDSTAT(core_forceidle_sum);
1069 #endif
1070 }
1071
1072 __P(nr_switches);
1073 __PS("nr_voluntary_switches", p->nvcsw);
1074 __PS("nr_involuntary_switches", p->nivcsw);
1075
1076 P(se.load.weight);
1077 #ifdef CONFIG_SMP
1078 P(se.avg.load_sum);
1079 P(se.avg.runnable_sum);
1080 P(se.avg.util_sum);
1081 P(se.avg.load_avg);
1082 P(se.avg.runnable_avg);
1083 P(se.avg.util_avg);
1084 P(se.avg.last_update_time);
1085 PM(se.avg.util_est, ~UTIL_AVG_UNCHANGED);
1086 #endif
1087 #ifdef CONFIG_UCLAMP_TASK
1088 __PS("uclamp.min", p->uclamp_req[UCLAMP_MIN].value);
1089 __PS("uclamp.max", p->uclamp_req[UCLAMP_MAX].value);
1090 __PS("effective uclamp.min", uclamp_eff_value(p, UCLAMP_MIN));
1091 __PS("effective uclamp.max", uclamp_eff_value(p, UCLAMP_MAX));
1092 #endif
1093 P(policy);
1094 P(prio);
1095 if (task_has_dl_policy(p)) {
1096 P(dl.runtime);
1097 P(dl.deadline);
1098 }
1099 #undef PN_SCHEDSTAT
1100 #undef P_SCHEDSTAT
1101
1102 {
1103 unsigned int this_cpu = raw_smp_processor_id();
1104 u64 t0, t1;
1105
1106 t0 = cpu_clock(this_cpu);
1107 t1 = cpu_clock(this_cpu);
1108 __PS("clock-delta", t1-t0);
1109 }
1110
1111 sched_show_numa(p, m);
1112 }
1113
proc_sched_set_task(struct task_struct * p)1114 void proc_sched_set_task(struct task_struct *p)
1115 {
1116 #ifdef CONFIG_SCHEDSTATS
1117 memset(&p->stats, 0, sizeof(p->stats));
1118 #endif
1119 }
1120
resched_latency_warn(int cpu,u64 latency)1121 void resched_latency_warn(int cpu, u64 latency)
1122 {
1123 static DEFINE_RATELIMIT_STATE(latency_check_ratelimit, 60 * 60 * HZ, 1);
1124
1125 WARN(__ratelimit(&latency_check_ratelimit),
1126 "sched: CPU %d need_resched set for > %llu ns (%d ticks) "
1127 "without schedule\n",
1128 cpu, latency, cpu_rq(cpu)->ticks_without_resched);
1129 }
1130