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 "sched.h"
10
11 /*
12 * This allows printing both to /proc/sched_debug and
13 * to the console
14 */
15 #define SEQ_printf(m, x...) \
16 do { \
17 if (m) \
18 seq_printf(m, x); \
19 else \
20 pr_cont(x); \
21 } while (0)
22
23 /*
24 * Ease the printing of nsec fields:
25 */
nsec_high(unsigned long long nsec)26 static long long nsec_high(unsigned long long nsec)
27 {
28 if ((long long)nsec < 0) {
29 nsec = -nsec;
30 do_div(nsec, 1000000);
31 return -nsec;
32 }
33 do_div(nsec, 1000000);
34
35 return nsec;
36 }
37
nsec_low(unsigned long long nsec)38 static unsigned long nsec_low(unsigned long long nsec)
39 {
40 if ((long long)nsec < 0)
41 nsec = -nsec;
42
43 return do_div(nsec, 1000000);
44 }
45
46 #define SPLIT_NS(x) nsec_high(x), nsec_low(x)
47
48 #define SCHED_FEAT(name, enabled) \
49 #name ,
50
51 static const char * const sched_feat_names[] = {
52 #include "features.h"
53 };
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
83 #undef SCHED_FEAT
84
sched_feat_disable(int i)85 static void sched_feat_disable(int i)
86 {
87 static_key_disable_cpuslocked(&sched_feat_keys[i]);
88 }
89
sched_feat_enable(int i)90 static void sched_feat_enable(int i)
91 {
92 static_key_enable_cpuslocked(&sched_feat_keys[i]);
93 }
94 #else
sched_feat_disable(int i)95 static void sched_feat_disable(int i) { };
sched_feat_enable(int i)96 static void sched_feat_enable(int i) { };
97 #endif /* CONFIG_JUMP_LABEL */
98
sched_feat_set(char * cmp)99 static int sched_feat_set(char *cmp)
100 {
101 int i;
102 int neg = 0;
103
104 if (strncmp(cmp, "NO_", 3) == 0) {
105 neg = 1;
106 cmp += 3;
107 }
108
109 i = match_string(sched_feat_names, __SCHED_FEAT_NR, cmp);
110 if (i < 0)
111 return i;
112
113 if (neg) {
114 sysctl_sched_features &= ~(1UL << i);
115 sched_feat_disable(i);
116 } else {
117 sysctl_sched_features |= (1UL << i);
118 sched_feat_enable(i);
119 }
120
121 return 0;
122 }
123
124 static ssize_t
sched_feat_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)125 sched_feat_write(struct file *filp, const char __user *ubuf,
126 size_t cnt, loff_t *ppos)
127 {
128 char buf[64];
129 char *cmp;
130 int ret;
131 struct inode *inode;
132
133 if (cnt > 63)
134 cnt = 63;
135
136 if (copy_from_user(&buf, ubuf, cnt))
137 return -EFAULT;
138
139 buf[cnt] = 0;
140 cmp = strstrip(buf);
141
142 /* Ensure the static_key remains in a consistent state */
143 inode = file_inode(filp);
144 cpus_read_lock();
145 inode_lock(inode);
146 ret = sched_feat_set(cmp);
147 inode_unlock(inode);
148 cpus_read_unlock();
149 if (ret < 0)
150 return ret;
151
152 *ppos += cnt;
153
154 return cnt;
155 }
156
sched_feat_open(struct inode * inode,struct file * filp)157 static int sched_feat_open(struct inode *inode, struct file *filp)
158 {
159 return single_open(filp, sched_feat_show, NULL);
160 }
161
162 static const struct file_operations sched_feat_fops = {
163 .open = sched_feat_open,
164 .write = sched_feat_write,
165 .read = seq_read,
166 .llseek = seq_lseek,
167 .release = single_release,
168 };
169
170 #ifdef CONFIG_SMP
171
sched_scaling_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)172 static 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(&buf, ubuf, cnt))
182 return -EFAULT;
183 buf[cnt] = '\0';
184
185 if (kstrtouint(buf, 10, &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
sched_scaling_show(struct seq_file * m,void * v)199 static int sched_scaling_show(struct seq_file *m, void *v)
200 {
201 seq_printf(m, "%d\n", sysctl_sched_tunable_scaling);
202 return 0;
203 }
204
sched_scaling_open(struct inode * inode,struct file * filp)205 static int sched_scaling_open(struct inode *inode, struct file *filp)
206 {
207 return single_open(filp, sched_scaling_show, NULL);
208 }
209
210 static 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 #endif /* SMP */
219
220 #ifdef CONFIG_PREEMPT_DYNAMIC
221
sched_dynamic_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)222 static ssize_t sched_dynamic_write(struct file *filp, const char __user *ubuf,
223 size_t cnt, loff_t *ppos)
224 {
225 char buf[16];
226 int mode;
227
228 if (cnt > 15)
229 cnt = 15;
230
231 if (copy_from_user(&buf, ubuf, cnt))
232 return -EFAULT;
233
234 buf[cnt] = 0;
235 mode = sched_dynamic_mode(strstrip(buf));
236 if (mode < 0)
237 return mode;
238
239 sched_dynamic_update(mode);
240
241 *ppos += cnt;
242
243 return cnt;
244 }
245
sched_dynamic_show(struct seq_file * m,void * v)246 static int sched_dynamic_show(struct seq_file *m, void *v)
247 {
248 static const char * preempt_modes[] = {
249 "none", "voluntary", "full"
250 };
251 int i;
252
253 for (i = 0; i < ARRAY_SIZE(preempt_modes); 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
sched_dynamic_open(struct inode * inode,struct file * filp)267 static int sched_dynamic_open(struct inode *inode, struct file *filp)
268 {
269 return single_open(filp, sched_dynamic_show, NULL);
270 }
271
272 static 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
285 static struct dentry *sd_dentry;
286
287
sched_verbose_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)288 static 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
316 static 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
323 static const struct seq_operations sched_debug_sops;
324
sched_debug_open(struct inode * inode,struct file * filp)325 static int sched_debug_open(struct inode *inode, struct file *filp)
326 {
327 return seq_open(filp, &sched_debug_sops);
328 }
329
330 static 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
337 static struct dentry *debugfs_sched;
338
sched_init_debug(void)339 static __init int sched_init_debug(void)
340 {
341 struct dentry __maybe_unused *numa;
342
343 debugfs_sched = debugfs_create_dir("sched", NULL);
344
345 debugfs_create_file("features", 0644, debugfs_sched, NULL, &sched_feat_fops);
346 debugfs_create_file_unsafe("verbose", 0644, debugfs_sched, &sched_debug_verbose, &sched_verbose_fops);
347 #ifdef CONFIG_PREEMPT_DYNAMIC
348 debugfs_create_file("preempt", 0644, debugfs_sched, NULL, &sched_dynamic_fops);
349 #endif
350
351 debugfs_create_u32("base_slice_ns", 0644, debugfs_sched, &sysctl_sched_base_slice);
352
353 debugfs_create_u32("latency_warn_ms", 0644, debugfs_sched, &sysctl_resched_latency_warn_ms);
354 debugfs_create_u32("latency_warn_once", 0644, debugfs_sched, &sysctl_resched_latency_warn_once);
355
356 #ifdef CONFIG_SMP
357 debugfs_create_file("tunable_scaling", 0644, debugfs_sched, NULL, &sched_scaling_fops);
358 debugfs_create_u32("migration_cost_ns", 0644, debugfs_sched, &sysctl_sched_migration_cost);
359 debugfs_create_u32("nr_migrate", 0644, debugfs_sched, &sysctl_sched_nr_migrate);
360
361 mutex_lock(&sched_domains_mutex);
362 update_sched_domain_debugfs();
363 mutex_unlock(&sched_domains_mutex);
364 #endif
365
366 #ifdef CONFIG_NUMA_BALANCING
367 numa = debugfs_create_dir("numa_balancing", debugfs_sched);
368
369 debugfs_create_u32("scan_delay_ms", 0644, numa, &sysctl_numa_balancing_scan_delay);
370 debugfs_create_u32("scan_period_min_ms", 0644, numa, &sysctl_numa_balancing_scan_period_min);
371 debugfs_create_u32("scan_period_max_ms", 0644, numa, &sysctl_numa_balancing_scan_period_max);
372 debugfs_create_u32("scan_size_mb", 0644, numa, &sysctl_numa_balancing_scan_size);
373 debugfs_create_u32("hot_threshold_ms", 0644, numa, &sysctl_numa_balancing_hot_threshold);
374 #endif
375
376 debugfs_create_file("debug", 0444, debugfs_sched, NULL, &sched_debug_fops);
377
378 return 0;
379 }
380 late_initcall(sched_init_debug);
381
382 #ifdef CONFIG_SMP
383
384 static cpumask_var_t sd_sysctl_cpus;
385
sd_flags_show(struct seq_file * m,void * v)386 static int sd_flags_show(struct seq_file *m, void *v)
387 {
388 unsigned long flags = *(unsigned int *)m->private;
389 int idx;
390
391 for_each_set_bit(idx, &flags, __SD_FLAG_CNT) {
392 seq_puts(m, sd_flag_debug[idx].name);
393 seq_puts(m, " ");
394 }
395 seq_puts(m, "\n");
396
397 return 0;
398 }
399
sd_flags_open(struct inode * inode,struct file * file)400 static int sd_flags_open(struct inode *inode, struct file *file)
401 {
402 return single_open(file, sd_flags_show, inode->i_private);
403 }
404
405 static const struct file_operations sd_flags_fops = {
406 .open = sd_flags_open,
407 .read = seq_read,
408 .llseek = seq_lseek,
409 .release = single_release,
410 };
411
register_sd(struct sched_domain * sd,struct dentry * parent)412 static void register_sd(struct sched_domain *sd, struct dentry *parent)
413 {
414 #define SDM(type, mode, member) \
415 debugfs_create_##type(#member, mode, parent, &sd->member)
416
417 SDM(ulong, 0644, min_interval);
418 SDM(ulong, 0644, max_interval);
419 SDM(u64, 0644, max_newidle_lb_cost);
420 SDM(u32, 0644, busy_factor);
421 SDM(u32, 0644, imbalance_pct);
422 SDM(u32, 0644, cache_nice_tries);
423 SDM(str, 0444, name);
424
425 #undef SDM
426
427 debugfs_create_file("flags", 0444, parent, &sd->flags, &sd_flags_fops);
428 debugfs_create_file("groups_flags", 0444, parent, &sd->groups->flags, &sd_flags_fops);
429 }
430
update_sched_domain_debugfs(void)431 void update_sched_domain_debugfs(void)
432 {
433 int cpu, i;
434
435 /*
436 * This can unfortunately be invoked before sched_debug_init() creates
437 * the debug directory. Don't touch sd_sysctl_cpus until then.
438 */
439 if (!debugfs_sched)
440 return;
441
442 if (!sched_debug_verbose)
443 return;
444
445 if (!cpumask_available(sd_sysctl_cpus)) {
446 if (!alloc_cpumask_var(&sd_sysctl_cpus, GFP_KERNEL))
447 return;
448 cpumask_copy(sd_sysctl_cpus, cpu_possible_mask);
449 }
450
451 if (!sd_dentry) {
452 sd_dentry = debugfs_create_dir("domains", debugfs_sched);
453
454 /* rebuild sd_sysctl_cpus if empty since it gets cleared below */
455 if (cpumask_empty(sd_sysctl_cpus))
456 cpumask_copy(sd_sysctl_cpus, cpu_online_mask);
457 }
458
459 for_each_cpu(cpu, sd_sysctl_cpus) {
460 struct sched_domain *sd;
461 struct dentry *d_cpu;
462 char buf[32];
463
464 snprintf(buf, sizeof(buf), "cpu%d", cpu);
465 debugfs_lookup_and_remove(buf, sd_dentry);
466 d_cpu = debugfs_create_dir(buf, sd_dentry);
467
468 i = 0;
469 for_each_domain(cpu, sd) {
470 struct dentry *d_sd;
471
472 snprintf(buf, sizeof(buf), "domain%d", i);
473 d_sd = debugfs_create_dir(buf, d_cpu);
474
475 register_sd(sd, d_sd);
476 i++;
477 }
478
479 __cpumask_clear_cpu(cpu, sd_sysctl_cpus);
480 }
481 }
482
dirty_sched_domain_sysctl(int cpu)483 void dirty_sched_domain_sysctl(int cpu)
484 {
485 if (cpumask_available(sd_sysctl_cpus))
486 __cpumask_set_cpu(cpu, sd_sysctl_cpus);
487 }
488
489 #endif /* CONFIG_SMP */
490
491 #ifdef CONFIG_FAIR_GROUP_SCHED
print_cfs_group_stats(struct seq_file * m,int cpu,struct task_group * tg)492 static void print_cfs_group_stats(struct seq_file *m, int cpu, struct task_group *tg)
493 {
494 struct sched_entity *se = tg->se[cpu];
495
496 #define P(F) SEQ_printf(m, " .%-30s: %lld\n", #F, (long long)F)
497 #define P_SCHEDSTAT(F) SEQ_printf(m, " .%-30s: %lld\n", \
498 #F, (long long)schedstat_val(stats->F))
499 #define PN(F) SEQ_printf(m, " .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F))
500 #define PN_SCHEDSTAT(F) SEQ_printf(m, " .%-30s: %lld.%06ld\n", \
501 #F, SPLIT_NS((long long)schedstat_val(stats->F)))
502
503 if (!se)
504 return;
505
506 PN(se->exec_start);
507 PN(se->vruntime);
508 PN(se->sum_exec_runtime);
509
510 if (schedstat_enabled()) {
511 struct sched_statistics *stats;
512 stats = __schedstats_from_se(se);
513
514 PN_SCHEDSTAT(wait_start);
515 PN_SCHEDSTAT(sleep_start);
516 PN_SCHEDSTAT(block_start);
517 PN_SCHEDSTAT(sleep_max);
518 PN_SCHEDSTAT(block_max);
519 PN_SCHEDSTAT(exec_max);
520 PN_SCHEDSTAT(slice_max);
521 PN_SCHEDSTAT(wait_max);
522 PN_SCHEDSTAT(wait_sum);
523 P_SCHEDSTAT(wait_count);
524 }
525
526 P(se->load.weight);
527 #ifdef CONFIG_SMP
528 P(se->avg.load_avg);
529 P(se->avg.util_avg);
530 P(se->avg.runnable_avg);
531 #endif
532
533 #undef PN_SCHEDSTAT
534 #undef PN
535 #undef P_SCHEDSTAT
536 #undef P
537 }
538 #endif
539
540 #ifdef CONFIG_CGROUP_SCHED
541 static DEFINE_SPINLOCK(sched_debug_lock);
542 static char group_path[PATH_MAX];
543
task_group_path(struct task_group * tg,char * path,int plen)544 static void task_group_path(struct task_group *tg, char *path, int plen)
545 {
546 if (autogroup_path(tg, path, plen))
547 return;
548
549 cgroup_path(tg->css.cgroup, path, plen);
550 }
551
552 /*
553 * Only 1 SEQ_printf_task_group_path() caller can use the full length
554 * group_path[] for cgroup path. Other simultaneous callers will have
555 * to use a shorter stack buffer. A "..." suffix is appended at the end
556 * of the stack buffer so that it will show up in case the output length
557 * matches the given buffer size to indicate possible path name truncation.
558 */
559 #define SEQ_printf_task_group_path(m, tg, fmt...) \
560 { \
561 if (spin_trylock(&sched_debug_lock)) { \
562 task_group_path(tg, group_path, sizeof(group_path)); \
563 SEQ_printf(m, fmt, group_path); \
564 spin_unlock(&sched_debug_lock); \
565 } else { \
566 char buf[128]; \
567 char *bufend = buf + sizeof(buf) - 3; \
568 task_group_path(tg, buf, bufend - buf); \
569 strcpy(bufend - 1, "..."); \
570 SEQ_printf(m, fmt, buf); \
571 } \
572 }
573 #endif
574
575 static void
print_task(struct seq_file * m,struct rq * rq,struct task_struct * p)576 print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
577 {
578 if (task_current(rq, p))
579 SEQ_printf(m, ">R");
580 else
581 SEQ_printf(m, " %c", task_state_to_char(p));
582
583 SEQ_printf(m, "%15s %5d %9Ld.%06ld %c %9Ld.%06ld %9Ld.%06ld %9Ld.%06ld %9Ld %5d ",
584 p->comm, task_pid_nr(p),
585 SPLIT_NS(p->se.vruntime),
586 entity_eligible(cfs_rq_of(&p->se), &p->se) ? 'E' : 'N',
587 SPLIT_NS(p->se.deadline),
588 SPLIT_NS(p->se.slice),
589 SPLIT_NS(p->se.sum_exec_runtime),
590 (long long)(p->nvcsw + p->nivcsw),
591 p->prio);
592
593 SEQ_printf(m, "%9lld.%06ld %9lld.%06ld %9lld.%06ld %9lld.%06ld",
594 SPLIT_NS(schedstat_val_or_zero(p->stats.wait_sum)),
595 SPLIT_NS(p->se.sum_exec_runtime),
596 SPLIT_NS(schedstat_val_or_zero(p->stats.sum_sleep_runtime)),
597 SPLIT_NS(schedstat_val_or_zero(p->stats.sum_block_runtime)));
598
599 #ifdef CONFIG_NUMA_BALANCING
600 SEQ_printf(m, " %d %d", task_node(p), task_numa_group_id(p));
601 #endif
602 #ifdef CONFIG_CGROUP_SCHED
603 SEQ_printf_task_group_path(m, task_group(p), " %s")
604 #endif
605
606 SEQ_printf(m, "\n");
607 }
608
print_rq(struct seq_file * m,struct rq * rq,int rq_cpu)609 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
610 {
611 struct task_struct *g, *p;
612
613 SEQ_printf(m, "\n");
614 SEQ_printf(m, "runnable tasks:\n");
615 SEQ_printf(m, " S task PID tree-key switches prio"
616 " wait-time sum-exec sum-sleep\n");
617 SEQ_printf(m, "-------------------------------------------------------"
618 "------------------------------------------------------\n");
619
620 rcu_read_lock();
621 for_each_process_thread(g, p) {
622 if (task_cpu(p) != rq_cpu)
623 continue;
624
625 print_task(m, rq, p);
626 }
627 rcu_read_unlock();
628 }
629
print_cfs_rq(struct seq_file * m,int cpu,struct cfs_rq * cfs_rq)630 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
631 {
632 s64 left_vruntime = -1, min_vruntime, right_vruntime = -1, spread;
633 struct sched_entity *last, *first;
634 struct rq *rq = cpu_rq(cpu);
635 unsigned long flags;
636
637 #ifdef CONFIG_FAIR_GROUP_SCHED
638 SEQ_printf(m, "\n");
639 SEQ_printf_task_group_path(m, cfs_rq->tg, "cfs_rq[%d]:%s\n", cpu);
640 #else
641 SEQ_printf(m, "\n");
642 SEQ_printf(m, "cfs_rq[%d]:\n", cpu);
643 #endif
644 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "exec_clock",
645 SPLIT_NS(cfs_rq->exec_clock));
646
647 raw_spin_rq_lock_irqsave(rq, flags);
648 first = __pick_first_entity(cfs_rq);
649 if (first)
650 left_vruntime = first->vruntime;
651 last = __pick_last_entity(cfs_rq);
652 if (last)
653 right_vruntime = last->vruntime;
654 min_vruntime = cfs_rq->min_vruntime;
655 raw_spin_rq_unlock_irqrestore(rq, flags);
656
657 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "left_vruntime",
658 SPLIT_NS(left_vruntime));
659 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "min_vruntime",
660 SPLIT_NS(min_vruntime));
661 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "avg_vruntime",
662 SPLIT_NS(avg_vruntime(cfs_rq)));
663 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "right_vruntime",
664 SPLIT_NS(right_vruntime));
665 spread = right_vruntime - left_vruntime;
666 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread", SPLIT_NS(spread));
667 SEQ_printf(m, " .%-30s: %d\n", "nr_spread_over",
668 cfs_rq->nr_spread_over);
669 SEQ_printf(m, " .%-30s: %d\n", "nr_running", cfs_rq->nr_running);
670 SEQ_printf(m, " .%-30s: %d\n", "h_nr_running", cfs_rq->h_nr_running);
671 SEQ_printf(m, " .%-30s: %d\n", "idle_nr_running",
672 cfs_rq->idle_nr_running);
673 SEQ_printf(m, " .%-30s: %d\n", "idle_h_nr_running",
674 cfs_rq->idle_h_nr_running);
675 SEQ_printf(m, " .%-30s: %ld\n", "load", cfs_rq->load.weight);
676 #ifdef CONFIG_SMP
677 SEQ_printf(m, " .%-30s: %lu\n", "load_avg",
678 cfs_rq->avg.load_avg);
679 SEQ_printf(m, " .%-30s: %lu\n", "runnable_avg",
680 cfs_rq->avg.runnable_avg);
681 SEQ_printf(m, " .%-30s: %lu\n", "util_avg",
682 cfs_rq->avg.util_avg);
683 SEQ_printf(m, " .%-30s: %u\n", "util_est_enqueued",
684 cfs_rq->avg.util_est.enqueued);
685 SEQ_printf(m, " .%-30s: %ld\n", "removed.load_avg",
686 cfs_rq->removed.load_avg);
687 SEQ_printf(m, " .%-30s: %ld\n", "removed.util_avg",
688 cfs_rq->removed.util_avg);
689 SEQ_printf(m, " .%-30s: %ld\n", "removed.runnable_avg",
690 cfs_rq->removed.runnable_avg);
691 #ifdef CONFIG_FAIR_GROUP_SCHED
692 SEQ_printf(m, " .%-30s: %lu\n", "tg_load_avg_contrib",
693 cfs_rq->tg_load_avg_contrib);
694 SEQ_printf(m, " .%-30s: %ld\n", "tg_load_avg",
695 atomic_long_read(&cfs_rq->tg->load_avg));
696 #endif
697 #endif
698 #ifdef CONFIG_CFS_BANDWIDTH
699 SEQ_printf(m, " .%-30s: %d\n", "throttled",
700 cfs_rq->throttled);
701 SEQ_printf(m, " .%-30s: %d\n", "throttle_count",
702 cfs_rq->throttle_count);
703 #endif
704
705 #ifdef CONFIG_FAIR_GROUP_SCHED
706 print_cfs_group_stats(m, cpu, cfs_rq->tg);
707 #endif
708 }
709
print_rt_rq(struct seq_file * m,int cpu,struct rt_rq * rt_rq)710 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
711 {
712 #ifdef CONFIG_RT_GROUP_SCHED
713 SEQ_printf(m, "\n");
714 SEQ_printf_task_group_path(m, rt_rq->tg, "rt_rq[%d]:%s\n", cpu);
715 #else
716 SEQ_printf(m, "\n");
717 SEQ_printf(m, "rt_rq[%d]:\n", cpu);
718 #endif
719
720 #define P(x) \
721 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
722 #define PU(x) \
723 SEQ_printf(m, " .%-30s: %lu\n", #x, (unsigned long)(rt_rq->x))
724 #define PN(x) \
725 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
726
727 PU(rt_nr_running);
728 #ifdef CONFIG_SMP
729 PU(rt_nr_migratory);
730 #endif
731 P(rt_throttled);
732 PN(rt_time);
733 PN(rt_runtime);
734
735 #undef PN
736 #undef PU
737 #undef P
738 }
739
print_dl_rq(struct seq_file * m,int cpu,struct dl_rq * dl_rq)740 void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq)
741 {
742 struct dl_bw *dl_bw;
743
744 SEQ_printf(m, "\n");
745 SEQ_printf(m, "dl_rq[%d]:\n", cpu);
746
747 #define PU(x) \
748 SEQ_printf(m, " .%-30s: %lu\n", #x, (unsigned long)(dl_rq->x))
749
750 PU(dl_nr_running);
751 #ifdef CONFIG_SMP
752 PU(dl_nr_migratory);
753 dl_bw = &cpu_rq(cpu)->rd->dl_bw;
754 #else
755 dl_bw = &dl_rq->dl_bw;
756 #endif
757 SEQ_printf(m, " .%-30s: %lld\n", "dl_bw->bw", dl_bw->bw);
758 SEQ_printf(m, " .%-30s: %lld\n", "dl_bw->total_bw", dl_bw->total_bw);
759
760 #undef PU
761 }
762
print_cpu(struct seq_file * m,int cpu)763 static void print_cpu(struct seq_file *m, int cpu)
764 {
765 struct rq *rq = cpu_rq(cpu);
766
767 #ifdef CONFIG_X86
768 {
769 unsigned int freq = cpu_khz ? : 1;
770
771 SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
772 cpu, freq / 1000, (freq % 1000));
773 }
774 #else
775 SEQ_printf(m, "cpu#%d\n", cpu);
776 #endif
777
778 #define P(x) \
779 do { \
780 if (sizeof(rq->x) == 4) \
781 SEQ_printf(m, " .%-30s: %d\n", #x, (int)(rq->x)); \
782 else \
783 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rq->x));\
784 } while (0)
785
786 #define PN(x) \
787 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
788
789 P(nr_running);
790 P(nr_switches);
791 P(nr_uninterruptible);
792 PN(next_balance);
793 SEQ_printf(m, " .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr)));
794 PN(clock);
795 PN(clock_task);
796 #undef P
797 #undef PN
798
799 #ifdef CONFIG_SMP
800 #define P64(n) SEQ_printf(m, " .%-30s: %Ld\n", #n, rq->n);
801 P64(avg_idle);
802 P64(max_idle_balance_cost);
803 #undef P64
804 #endif
805
806 #define P(n) SEQ_printf(m, " .%-30s: %d\n", #n, schedstat_val(rq->n));
807 if (schedstat_enabled()) {
808 P(yld_count);
809 P(sched_count);
810 P(sched_goidle);
811 P(ttwu_count);
812 P(ttwu_local);
813 }
814 #undef P
815
816 print_cfs_stats(m, cpu);
817 print_rt_stats(m, cpu);
818 print_dl_stats(m, cpu);
819
820 print_rq(m, rq, cpu);
821 SEQ_printf(m, "\n");
822 }
823
824 static const char *sched_tunable_scaling_names[] = {
825 "none",
826 "logarithmic",
827 "linear"
828 };
829
sched_debug_header(struct seq_file * m)830 static void sched_debug_header(struct seq_file *m)
831 {
832 u64 ktime, sched_clk, cpu_clk;
833 unsigned long flags;
834
835 local_irq_save(flags);
836 ktime = ktime_to_ns(ktime_get());
837 sched_clk = sched_clock();
838 cpu_clk = local_clock();
839 local_irq_restore(flags);
840
841 SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n",
842 init_utsname()->release,
843 (int)strcspn(init_utsname()->version, " "),
844 init_utsname()->version);
845
846 #define P(x) \
847 SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
848 #define PN(x) \
849 SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
850 PN(ktime);
851 PN(sched_clk);
852 PN(cpu_clk);
853 P(jiffies);
854 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
855 P(sched_clock_stable());
856 #endif
857 #undef PN
858 #undef P
859
860 SEQ_printf(m, "\n");
861 SEQ_printf(m, "sysctl_sched\n");
862
863 #define P(x) \
864 SEQ_printf(m, " .%-40s: %Ld\n", #x, (long long)(x))
865 #define PN(x) \
866 SEQ_printf(m, " .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
867 PN(sysctl_sched_base_slice);
868 P(sysctl_sched_child_runs_first);
869 P(sysctl_sched_features);
870 #undef PN
871 #undef P
872
873 SEQ_printf(m, " .%-40s: %d (%s)\n",
874 "sysctl_sched_tunable_scaling",
875 sysctl_sched_tunable_scaling,
876 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
877 SEQ_printf(m, "\n");
878 }
879
sched_debug_show(struct seq_file * m,void * v)880 static int sched_debug_show(struct seq_file *m, void *v)
881 {
882 int cpu = (unsigned long)(v - 2);
883
884 if (cpu != -1)
885 print_cpu(m, cpu);
886 else
887 sched_debug_header(m);
888
889 return 0;
890 }
891
sysrq_sched_debug_show(void)892 void sysrq_sched_debug_show(void)
893 {
894 int cpu;
895
896 sched_debug_header(NULL);
897 for_each_online_cpu(cpu) {
898 /*
899 * Need to reset softlockup watchdogs on all CPUs, because
900 * another CPU might be blocked waiting for us to process
901 * an IPI or stop_machine.
902 */
903 touch_nmi_watchdog();
904 touch_all_softlockup_watchdogs();
905 print_cpu(NULL, cpu);
906 }
907 }
908
909 /*
910 * This iterator needs some explanation.
911 * It returns 1 for the header position.
912 * This means 2 is CPU 0.
913 * In a hotplugged system some CPUs, including CPU 0, may be missing so we have
914 * to use cpumask_* to iterate over the CPUs.
915 */
sched_debug_start(struct seq_file * file,loff_t * offset)916 static void *sched_debug_start(struct seq_file *file, loff_t *offset)
917 {
918 unsigned long n = *offset;
919
920 if (n == 0)
921 return (void *) 1;
922
923 n--;
924
925 if (n > 0)
926 n = cpumask_next(n - 1, cpu_online_mask);
927 else
928 n = cpumask_first(cpu_online_mask);
929
930 *offset = n + 1;
931
932 if (n < nr_cpu_ids)
933 return (void *)(unsigned long)(n + 2);
934
935 return NULL;
936 }
937
sched_debug_next(struct seq_file * file,void * data,loff_t * offset)938 static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
939 {
940 (*offset)++;
941 return sched_debug_start(file, offset);
942 }
943
sched_debug_stop(struct seq_file * file,void * data)944 static void sched_debug_stop(struct seq_file *file, void *data)
945 {
946 }
947
948 static const struct seq_operations sched_debug_sops = {
949 .start = sched_debug_start,
950 .next = sched_debug_next,
951 .stop = sched_debug_stop,
952 .show = sched_debug_show,
953 };
954
955 #define __PS(S, F) SEQ_printf(m, "%-45s:%21Ld\n", S, (long long)(F))
956 #define __P(F) __PS(#F, F)
957 #define P(F) __PS(#F, p->F)
958 #define PM(F, M) __PS(#F, p->F & (M))
959 #define __PSN(S, F) SEQ_printf(m, "%-45s:%14Ld.%06ld\n", S, SPLIT_NS((long long)(F)))
960 #define __PN(F) __PSN(#F, F)
961 #define PN(F) __PSN(#F, p->F)
962
963
964 #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)965 void print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
966 unsigned long tpf, unsigned long gsf, unsigned long gpf)
967 {
968 SEQ_printf(m, "numa_faults node=%d ", node);
969 SEQ_printf(m, "task_private=%lu task_shared=%lu ", tpf, tsf);
970 SEQ_printf(m, "group_private=%lu group_shared=%lu\n", gpf, gsf);
971 }
972 #endif
973
974
sched_show_numa(struct task_struct * p,struct seq_file * m)975 static void sched_show_numa(struct task_struct *p, struct seq_file *m)
976 {
977 #ifdef CONFIG_NUMA_BALANCING
978 if (p->mm)
979 P(mm->numa_scan_seq);
980
981 P(numa_pages_migrated);
982 P(numa_preferred_nid);
983 P(total_numa_faults);
984 SEQ_printf(m, "current_node=%d, numa_group_id=%d\n",
985 task_node(p), task_numa_group_id(p));
986 show_numa_stats(p, m);
987 #endif
988 }
989
proc_sched_show_task(struct task_struct * p,struct pid_namespace * ns,struct seq_file * m)990 void proc_sched_show_task(struct task_struct *p, struct pid_namespace *ns,
991 struct seq_file *m)
992 {
993 unsigned long nr_switches;
994
995 SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr_ns(p, ns),
996 get_nr_threads(p));
997 SEQ_printf(m,
998 "---------------------------------------------------------"
999 "----------\n");
1000
1001 #define P_SCHEDSTAT(F) __PS(#F, schedstat_val(p->stats.F))
1002 #define PN_SCHEDSTAT(F) __PSN(#F, schedstat_val(p->stats.F))
1003
1004 PN(se.exec_start);
1005 PN(se.vruntime);
1006 PN(se.sum_exec_runtime);
1007
1008 nr_switches = p->nvcsw + p->nivcsw;
1009
1010 P(se.nr_migrations);
1011
1012 if (schedstat_enabled()) {
1013 u64 avg_atom, avg_per_cpu;
1014
1015 PN_SCHEDSTAT(sum_sleep_runtime);
1016 PN_SCHEDSTAT(sum_block_runtime);
1017 PN_SCHEDSTAT(wait_start);
1018 PN_SCHEDSTAT(sleep_start);
1019 PN_SCHEDSTAT(block_start);
1020 PN_SCHEDSTAT(sleep_max);
1021 PN_SCHEDSTAT(block_max);
1022 PN_SCHEDSTAT(exec_max);
1023 PN_SCHEDSTAT(slice_max);
1024 PN_SCHEDSTAT(wait_max);
1025 PN_SCHEDSTAT(wait_sum);
1026 P_SCHEDSTAT(wait_count);
1027 PN_SCHEDSTAT(iowait_sum);
1028 P_SCHEDSTAT(iowait_count);
1029 P_SCHEDSTAT(nr_migrations_cold);
1030 P_SCHEDSTAT(nr_failed_migrations_affine);
1031 P_SCHEDSTAT(nr_failed_migrations_running);
1032 P_SCHEDSTAT(nr_failed_migrations_hot);
1033 P_SCHEDSTAT(nr_forced_migrations);
1034 P_SCHEDSTAT(nr_wakeups);
1035 P_SCHEDSTAT(nr_wakeups_sync);
1036 P_SCHEDSTAT(nr_wakeups_migrate);
1037 P_SCHEDSTAT(nr_wakeups_local);
1038 P_SCHEDSTAT(nr_wakeups_remote);
1039 P_SCHEDSTAT(nr_wakeups_affine);
1040 P_SCHEDSTAT(nr_wakeups_affine_attempts);
1041 P_SCHEDSTAT(nr_wakeups_passive);
1042 P_SCHEDSTAT(nr_wakeups_idle);
1043
1044 avg_atom = p->se.sum_exec_runtime;
1045 if (nr_switches)
1046 avg_atom = div64_ul(avg_atom, nr_switches);
1047 else
1048 avg_atom = -1LL;
1049
1050 avg_per_cpu = p->se.sum_exec_runtime;
1051 if (p->se.nr_migrations) {
1052 avg_per_cpu = div64_u64(avg_per_cpu,
1053 p->se.nr_migrations);
1054 } else {
1055 avg_per_cpu = -1LL;
1056 }
1057
1058 __PN(avg_atom);
1059 __PN(avg_per_cpu);
1060
1061 #ifdef CONFIG_SCHED_CORE
1062 PN_SCHEDSTAT(core_forceidle_sum);
1063 #endif
1064 }
1065
1066 __P(nr_switches);
1067 __PS("nr_voluntary_switches", p->nvcsw);
1068 __PS("nr_involuntary_switches", p->nivcsw);
1069
1070 P(se.load.weight);
1071 #ifdef CONFIG_SMP
1072 P(se.avg.load_sum);
1073 P(se.avg.runnable_sum);
1074 P(se.avg.util_sum);
1075 P(se.avg.load_avg);
1076 P(se.avg.runnable_avg);
1077 P(se.avg.util_avg);
1078 P(se.avg.last_update_time);
1079 P(se.avg.util_est.ewma);
1080 PM(se.avg.util_est.enqueued, ~UTIL_AVG_UNCHANGED);
1081 #endif
1082 #ifdef CONFIG_UCLAMP_TASK
1083 __PS("uclamp.min", p->uclamp_req[UCLAMP_MIN].value);
1084 __PS("uclamp.max", p->uclamp_req[UCLAMP_MAX].value);
1085 __PS("effective uclamp.min", uclamp_eff_value(p, UCLAMP_MIN));
1086 __PS("effective uclamp.max", uclamp_eff_value(p, UCLAMP_MAX));
1087 #endif
1088 P(policy);
1089 P(prio);
1090 #ifdef CONFIG_SCHED_LATENCY_NICE
1091 P(latency_prio);
1092 #endif
1093 if (task_has_dl_policy(p)) {
1094 P(dl.runtime);
1095 P(dl.deadline);
1096 }
1097 #undef PN_SCHEDSTAT
1098 #undef P_SCHEDSTAT
1099
1100 {
1101 unsigned int this_cpu = raw_smp_processor_id();
1102 u64 t0, t1;
1103
1104 t0 = cpu_clock(this_cpu);
1105 t1 = cpu_clock(this_cpu);
1106 __PS("clock-delta", t1-t0);
1107 }
1108
1109 sched_show_numa(p, m);
1110 }
1111
proc_sched_set_task(struct task_struct * p)1112 void proc_sched_set_task(struct task_struct *p)
1113 {
1114 #ifdef CONFIG_SCHEDSTATS
1115 memset(&p->stats, 0, sizeof(p->stats));
1116 #endif
1117 }
1118
resched_latency_warn(int cpu,u64 latency)1119 void resched_latency_warn(int cpu, u64 latency)
1120 {
1121 static DEFINE_RATELIMIT_STATE(latency_check_ratelimit, 60 * 60 * HZ, 1);
1122
1123 WARN(__ratelimit(&latency_check_ratelimit),
1124 "sched: CPU %d need_resched set for > %llu ns (%d ticks) "
1125 "without schedule\n",
1126 cpu, latency, cpu_rq(cpu)->ticks_without_resched);
1127 }
1128