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 static const struct seq_operations sched_debug_sops;
286
sched_debug_open(struct inode * inode,struct file * filp)287 static int sched_debug_open(struct inode *inode, struct file *filp)
288 {
289 return seq_open(filp, &sched_debug_sops);
290 }
291
292 static const struct file_operations sched_debug_fops = {
293 .open = sched_debug_open,
294 .read = seq_read,
295 .llseek = seq_lseek,
296 .release = seq_release,
297 };
298
299 static struct dentry *debugfs_sched;
300
sched_init_debug(void)301 static __init int sched_init_debug(void)
302 {
303 struct dentry __maybe_unused *numa;
304
305 debugfs_sched = debugfs_create_dir("sched", NULL);
306
307 debugfs_create_file("features", 0644, debugfs_sched, NULL, &sched_feat_fops);
308 debugfs_create_bool("verbose", 0644, debugfs_sched, &sched_debug_verbose);
309 #ifdef CONFIG_PREEMPT_DYNAMIC
310 debugfs_create_file("preempt", 0644, debugfs_sched, NULL, &sched_dynamic_fops);
311 #endif
312
313 debugfs_create_u32("latency_ns", 0644, debugfs_sched, &sysctl_sched_latency);
314 debugfs_create_u32("min_granularity_ns", 0644, debugfs_sched, &sysctl_sched_min_granularity);
315 debugfs_create_u32("idle_min_granularity_ns", 0644, debugfs_sched, &sysctl_sched_idle_min_granularity);
316 debugfs_create_u32("wakeup_granularity_ns", 0644, debugfs_sched, &sysctl_sched_wakeup_granularity);
317
318 debugfs_create_u32("latency_warn_ms", 0644, debugfs_sched, &sysctl_resched_latency_warn_ms);
319 debugfs_create_u32("latency_warn_once", 0644, debugfs_sched, &sysctl_resched_latency_warn_once);
320
321 #ifdef CONFIG_SMP
322 debugfs_create_file("tunable_scaling", 0644, debugfs_sched, NULL, &sched_scaling_fops);
323 debugfs_create_u32("migration_cost_ns", 0644, debugfs_sched, &sysctl_sched_migration_cost);
324 debugfs_create_u32("nr_migrate", 0644, debugfs_sched, &sysctl_sched_nr_migrate);
325
326 mutex_lock(&sched_domains_mutex);
327 update_sched_domain_debugfs();
328 mutex_unlock(&sched_domains_mutex);
329 #endif
330
331 #ifdef CONFIG_NUMA_BALANCING
332 numa = debugfs_create_dir("numa_balancing", debugfs_sched);
333
334 debugfs_create_u32("scan_delay_ms", 0644, numa, &sysctl_numa_balancing_scan_delay);
335 debugfs_create_u32("scan_period_min_ms", 0644, numa, &sysctl_numa_balancing_scan_period_min);
336 debugfs_create_u32("scan_period_max_ms", 0644, numa, &sysctl_numa_balancing_scan_period_max);
337 debugfs_create_u32("scan_size_mb", 0644, numa, &sysctl_numa_balancing_scan_size);
338 debugfs_create_u32("hot_threshold_ms", 0644, numa, &sysctl_numa_balancing_hot_threshold);
339 #endif
340
341 debugfs_create_file("debug", 0444, debugfs_sched, NULL, &sched_debug_fops);
342
343 return 0;
344 }
345 late_initcall(sched_init_debug);
346
347 #ifdef CONFIG_SMP
348
349 static cpumask_var_t sd_sysctl_cpus;
350 static struct dentry *sd_dentry;
351
sd_flags_show(struct seq_file * m,void * v)352 static int sd_flags_show(struct seq_file *m, void *v)
353 {
354 unsigned long flags = *(unsigned int *)m->private;
355 int idx;
356
357 for_each_set_bit(idx, &flags, __SD_FLAG_CNT) {
358 seq_puts(m, sd_flag_debug[idx].name);
359 seq_puts(m, " ");
360 }
361 seq_puts(m, "\n");
362
363 return 0;
364 }
365
sd_flags_open(struct inode * inode,struct file * file)366 static int sd_flags_open(struct inode *inode, struct file *file)
367 {
368 return single_open(file, sd_flags_show, inode->i_private);
369 }
370
371 static const struct file_operations sd_flags_fops = {
372 .open = sd_flags_open,
373 .read = seq_read,
374 .llseek = seq_lseek,
375 .release = single_release,
376 };
377
register_sd(struct sched_domain * sd,struct dentry * parent)378 static void register_sd(struct sched_domain *sd, struct dentry *parent)
379 {
380 #define SDM(type, mode, member) \
381 debugfs_create_##type(#member, mode, parent, &sd->member)
382
383 SDM(ulong, 0644, min_interval);
384 SDM(ulong, 0644, max_interval);
385 SDM(u64, 0644, max_newidle_lb_cost);
386 SDM(u32, 0644, busy_factor);
387 SDM(u32, 0644, imbalance_pct);
388 SDM(u32, 0644, cache_nice_tries);
389 SDM(str, 0444, name);
390
391 #undef SDM
392
393 debugfs_create_file("flags", 0444, parent, &sd->flags, &sd_flags_fops);
394 }
395
update_sched_domain_debugfs(void)396 void update_sched_domain_debugfs(void)
397 {
398 int cpu, i;
399
400 /*
401 * This can unfortunately be invoked before sched_debug_init() creates
402 * the debug directory. Don't touch sd_sysctl_cpus until then.
403 */
404 if (!debugfs_sched)
405 return;
406
407 if (!cpumask_available(sd_sysctl_cpus)) {
408 if (!alloc_cpumask_var(&sd_sysctl_cpus, GFP_KERNEL))
409 return;
410 cpumask_copy(sd_sysctl_cpus, cpu_possible_mask);
411 }
412
413 if (!sd_dentry)
414 sd_dentry = debugfs_create_dir("domains", debugfs_sched);
415
416 for_each_cpu(cpu, sd_sysctl_cpus) {
417 struct sched_domain *sd;
418 struct dentry *d_cpu;
419 char buf[32];
420
421 snprintf(buf, sizeof(buf), "cpu%d", cpu);
422 debugfs_lookup_and_remove(buf, sd_dentry);
423 d_cpu = debugfs_create_dir(buf, sd_dentry);
424
425 i = 0;
426 for_each_domain(cpu, sd) {
427 struct dentry *d_sd;
428
429 snprintf(buf, sizeof(buf), "domain%d", i);
430 d_sd = debugfs_create_dir(buf, d_cpu);
431
432 register_sd(sd, d_sd);
433 i++;
434 }
435
436 __cpumask_clear_cpu(cpu, sd_sysctl_cpus);
437 }
438 }
439
dirty_sched_domain_sysctl(int cpu)440 void dirty_sched_domain_sysctl(int cpu)
441 {
442 if (cpumask_available(sd_sysctl_cpus))
443 __cpumask_set_cpu(cpu, sd_sysctl_cpus);
444 }
445
446 #endif /* CONFIG_SMP */
447
448 #ifdef CONFIG_FAIR_GROUP_SCHED
print_cfs_group_stats(struct seq_file * m,int cpu,struct task_group * tg)449 static void print_cfs_group_stats(struct seq_file *m, int cpu, struct task_group *tg)
450 {
451 struct sched_entity *se = tg->se[cpu];
452
453 #define P(F) SEQ_printf(m, " .%-30s: %lld\n", #F, (long long)F)
454 #define P_SCHEDSTAT(F) SEQ_printf(m, " .%-30s: %lld\n", \
455 #F, (long long)schedstat_val(stats->F))
456 #define PN(F) SEQ_printf(m, " .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F))
457 #define PN_SCHEDSTAT(F) SEQ_printf(m, " .%-30s: %lld.%06ld\n", \
458 #F, SPLIT_NS((long long)schedstat_val(stats->F)))
459
460 if (!se)
461 return;
462
463 PN(se->exec_start);
464 PN(se->vruntime);
465 PN(se->sum_exec_runtime);
466
467 if (schedstat_enabled()) {
468 struct sched_statistics *stats;
469 stats = __schedstats_from_se(se);
470
471 PN_SCHEDSTAT(wait_start);
472 PN_SCHEDSTAT(sleep_start);
473 PN_SCHEDSTAT(block_start);
474 PN_SCHEDSTAT(sleep_max);
475 PN_SCHEDSTAT(block_max);
476 PN_SCHEDSTAT(exec_max);
477 PN_SCHEDSTAT(slice_max);
478 PN_SCHEDSTAT(wait_max);
479 PN_SCHEDSTAT(wait_sum);
480 P_SCHEDSTAT(wait_count);
481 }
482
483 P(se->load.weight);
484 #ifdef CONFIG_SMP
485 P(se->avg.load_avg);
486 P(se->avg.util_avg);
487 P(se->avg.runnable_avg);
488 #endif
489
490 #undef PN_SCHEDSTAT
491 #undef PN
492 #undef P_SCHEDSTAT
493 #undef P
494 }
495 #endif
496
497 #ifdef CONFIG_CGROUP_SCHED
498 static DEFINE_SPINLOCK(sched_debug_lock);
499 static char group_path[PATH_MAX];
500
task_group_path(struct task_group * tg,char * path,int plen)501 static void task_group_path(struct task_group *tg, char *path, int plen)
502 {
503 if (autogroup_path(tg, path, plen))
504 return;
505
506 cgroup_path(tg->css.cgroup, path, plen);
507 }
508
509 /*
510 * Only 1 SEQ_printf_task_group_path() caller can use the full length
511 * group_path[] for cgroup path. Other simultaneous callers will have
512 * to use a shorter stack buffer. A "..." suffix is appended at the end
513 * of the stack buffer so that it will show up in case the output length
514 * matches the given buffer size to indicate possible path name truncation.
515 */
516 #define SEQ_printf_task_group_path(m, tg, fmt...) \
517 { \
518 if (spin_trylock(&sched_debug_lock)) { \
519 task_group_path(tg, group_path, sizeof(group_path)); \
520 SEQ_printf(m, fmt, group_path); \
521 spin_unlock(&sched_debug_lock); \
522 } else { \
523 char buf[128]; \
524 char *bufend = buf + sizeof(buf) - 3; \
525 task_group_path(tg, buf, bufend - buf); \
526 strcpy(bufend - 1, "..."); \
527 SEQ_printf(m, fmt, buf); \
528 } \
529 }
530 #endif
531
532 static void
print_task(struct seq_file * m,struct rq * rq,struct task_struct * p)533 print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
534 {
535 if (task_current(rq, p))
536 SEQ_printf(m, ">R");
537 else
538 SEQ_printf(m, " %c", task_state_to_char(p));
539
540 SEQ_printf(m, " %15s %5d %9Ld.%06ld %9Ld %5d ",
541 p->comm, task_pid_nr(p),
542 SPLIT_NS(p->se.vruntime),
543 (long long)(p->nvcsw + p->nivcsw),
544 p->prio);
545
546 SEQ_printf(m, "%9lld.%06ld %9lld.%06ld %9lld.%06ld %9lld.%06ld",
547 SPLIT_NS(schedstat_val_or_zero(p->stats.wait_sum)),
548 SPLIT_NS(p->se.sum_exec_runtime),
549 SPLIT_NS(schedstat_val_or_zero(p->stats.sum_sleep_runtime)),
550 SPLIT_NS(schedstat_val_or_zero(p->stats.sum_block_runtime)));
551
552 #ifdef CONFIG_NUMA_BALANCING
553 SEQ_printf(m, " %d %d", task_node(p), task_numa_group_id(p));
554 #endif
555 #ifdef CONFIG_CGROUP_SCHED
556 SEQ_printf_task_group_path(m, task_group(p), " %s")
557 #endif
558
559 SEQ_printf(m, "\n");
560 }
561
print_rq(struct seq_file * m,struct rq * rq,int rq_cpu)562 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
563 {
564 struct task_struct *g, *p;
565
566 SEQ_printf(m, "\n");
567 SEQ_printf(m, "runnable tasks:\n");
568 SEQ_printf(m, " S task PID tree-key switches prio"
569 " wait-time sum-exec sum-sleep\n");
570 SEQ_printf(m, "-------------------------------------------------------"
571 "------------------------------------------------------\n");
572
573 rcu_read_lock();
574 for_each_process_thread(g, p) {
575 if (task_cpu(p) != rq_cpu)
576 continue;
577
578 print_task(m, rq, p);
579 }
580 rcu_read_unlock();
581 }
582
print_cfs_rq(struct seq_file * m,int cpu,struct cfs_rq * cfs_rq)583 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
584 {
585 s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
586 spread, rq0_min_vruntime, spread0;
587 struct rq *rq = cpu_rq(cpu);
588 struct sched_entity *last;
589 unsigned long flags;
590
591 #ifdef CONFIG_FAIR_GROUP_SCHED
592 SEQ_printf(m, "\n");
593 SEQ_printf_task_group_path(m, cfs_rq->tg, "cfs_rq[%d]:%s\n", cpu);
594 #else
595 SEQ_printf(m, "\n");
596 SEQ_printf(m, "cfs_rq[%d]:\n", cpu);
597 #endif
598 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "exec_clock",
599 SPLIT_NS(cfs_rq->exec_clock));
600
601 raw_spin_rq_lock_irqsave(rq, flags);
602 if (rb_first_cached(&cfs_rq->tasks_timeline))
603 MIN_vruntime = (__pick_first_entity(cfs_rq))->vruntime;
604 last = __pick_last_entity(cfs_rq);
605 if (last)
606 max_vruntime = last->vruntime;
607 min_vruntime = cfs_rq->min_vruntime;
608 rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
609 raw_spin_rq_unlock_irqrestore(rq, flags);
610 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "MIN_vruntime",
611 SPLIT_NS(MIN_vruntime));
612 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "min_vruntime",
613 SPLIT_NS(min_vruntime));
614 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "max_vruntime",
615 SPLIT_NS(max_vruntime));
616 spread = max_vruntime - MIN_vruntime;
617 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread",
618 SPLIT_NS(spread));
619 spread0 = min_vruntime - rq0_min_vruntime;
620 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread0",
621 SPLIT_NS(spread0));
622 SEQ_printf(m, " .%-30s: %d\n", "nr_spread_over",
623 cfs_rq->nr_spread_over);
624 SEQ_printf(m, " .%-30s: %d\n", "nr_running", cfs_rq->nr_running);
625 SEQ_printf(m, " .%-30s: %d\n", "h_nr_running", cfs_rq->h_nr_running);
626 SEQ_printf(m, " .%-30s: %d\n", "idle_nr_running",
627 cfs_rq->idle_nr_running);
628 SEQ_printf(m, " .%-30s: %d\n", "idle_h_nr_running",
629 cfs_rq->idle_h_nr_running);
630 SEQ_printf(m, " .%-30s: %ld\n", "load", cfs_rq->load.weight);
631 #ifdef CONFIG_SMP
632 SEQ_printf(m, " .%-30s: %lu\n", "load_avg",
633 cfs_rq->avg.load_avg);
634 SEQ_printf(m, " .%-30s: %lu\n", "runnable_avg",
635 cfs_rq->avg.runnable_avg);
636 SEQ_printf(m, " .%-30s: %lu\n", "util_avg",
637 cfs_rq->avg.util_avg);
638 SEQ_printf(m, " .%-30s: %u\n", "util_est_enqueued",
639 cfs_rq->avg.util_est.enqueued);
640 SEQ_printf(m, " .%-30s: %ld\n", "removed.load_avg",
641 cfs_rq->removed.load_avg);
642 SEQ_printf(m, " .%-30s: %ld\n", "removed.util_avg",
643 cfs_rq->removed.util_avg);
644 SEQ_printf(m, " .%-30s: %ld\n", "removed.runnable_avg",
645 cfs_rq->removed.runnable_avg);
646 #ifdef CONFIG_FAIR_GROUP_SCHED
647 SEQ_printf(m, " .%-30s: %lu\n", "tg_load_avg_contrib",
648 cfs_rq->tg_load_avg_contrib);
649 SEQ_printf(m, " .%-30s: %ld\n", "tg_load_avg",
650 atomic_long_read(&cfs_rq->tg->load_avg));
651 #endif
652 #endif
653 #ifdef CONFIG_CFS_BANDWIDTH
654 SEQ_printf(m, " .%-30s: %d\n", "throttled",
655 cfs_rq->throttled);
656 SEQ_printf(m, " .%-30s: %d\n", "throttle_count",
657 cfs_rq->throttle_count);
658 #endif
659
660 #ifdef CONFIG_FAIR_GROUP_SCHED
661 print_cfs_group_stats(m, cpu, cfs_rq->tg);
662 #endif
663 }
664
print_rt_rq(struct seq_file * m,int cpu,struct rt_rq * rt_rq)665 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
666 {
667 #ifdef CONFIG_RT_GROUP_SCHED
668 SEQ_printf(m, "\n");
669 SEQ_printf_task_group_path(m, rt_rq->tg, "rt_rq[%d]:%s\n", cpu);
670 #else
671 SEQ_printf(m, "\n");
672 SEQ_printf(m, "rt_rq[%d]:\n", cpu);
673 #endif
674
675 #define P(x) \
676 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
677 #define PU(x) \
678 SEQ_printf(m, " .%-30s: %lu\n", #x, (unsigned long)(rt_rq->x))
679 #define PN(x) \
680 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
681
682 PU(rt_nr_running);
683 #ifdef CONFIG_SMP
684 PU(rt_nr_migratory);
685 #endif
686 P(rt_throttled);
687 PN(rt_time);
688 PN(rt_runtime);
689
690 #undef PN
691 #undef PU
692 #undef P
693 }
694
print_dl_rq(struct seq_file * m,int cpu,struct dl_rq * dl_rq)695 void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq)
696 {
697 struct dl_bw *dl_bw;
698
699 SEQ_printf(m, "\n");
700 SEQ_printf(m, "dl_rq[%d]:\n", cpu);
701
702 #define PU(x) \
703 SEQ_printf(m, " .%-30s: %lu\n", #x, (unsigned long)(dl_rq->x))
704
705 PU(dl_nr_running);
706 #ifdef CONFIG_SMP
707 PU(dl_nr_migratory);
708 dl_bw = &cpu_rq(cpu)->rd->dl_bw;
709 #else
710 dl_bw = &dl_rq->dl_bw;
711 #endif
712 SEQ_printf(m, " .%-30s: %lld\n", "dl_bw->bw", dl_bw->bw);
713 SEQ_printf(m, " .%-30s: %lld\n", "dl_bw->total_bw", dl_bw->total_bw);
714
715 #undef PU
716 }
717
print_cpu(struct seq_file * m,int cpu)718 static void print_cpu(struct seq_file *m, int cpu)
719 {
720 struct rq *rq = cpu_rq(cpu);
721
722 #ifdef CONFIG_X86
723 {
724 unsigned int freq = cpu_khz ? : 1;
725
726 SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
727 cpu, freq / 1000, (freq % 1000));
728 }
729 #else
730 SEQ_printf(m, "cpu#%d\n", cpu);
731 #endif
732
733 #define P(x) \
734 do { \
735 if (sizeof(rq->x) == 4) \
736 SEQ_printf(m, " .%-30s: %ld\n", #x, (long)(rq->x)); \
737 else \
738 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rq->x));\
739 } while (0)
740
741 #define PN(x) \
742 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
743
744 P(nr_running);
745 P(nr_switches);
746 P(nr_uninterruptible);
747 PN(next_balance);
748 SEQ_printf(m, " .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr)));
749 PN(clock);
750 PN(clock_task);
751 #undef P
752 #undef PN
753
754 #ifdef CONFIG_SMP
755 #define P64(n) SEQ_printf(m, " .%-30s: %Ld\n", #n, rq->n);
756 P64(avg_idle);
757 P64(max_idle_balance_cost);
758 #undef P64
759 #endif
760
761 #define P(n) SEQ_printf(m, " .%-30s: %d\n", #n, schedstat_val(rq->n));
762 if (schedstat_enabled()) {
763 P(yld_count);
764 P(sched_count);
765 P(sched_goidle);
766 P(ttwu_count);
767 P(ttwu_local);
768 }
769 #undef P
770
771 print_cfs_stats(m, cpu);
772 print_rt_stats(m, cpu);
773 print_dl_stats(m, cpu);
774
775 print_rq(m, rq, cpu);
776 SEQ_printf(m, "\n");
777 }
778
779 static const char *sched_tunable_scaling_names[] = {
780 "none",
781 "logarithmic",
782 "linear"
783 };
784
sched_debug_header(struct seq_file * m)785 static void sched_debug_header(struct seq_file *m)
786 {
787 u64 ktime, sched_clk, cpu_clk;
788 unsigned long flags;
789
790 local_irq_save(flags);
791 ktime = ktime_to_ns(ktime_get());
792 sched_clk = sched_clock();
793 cpu_clk = local_clock();
794 local_irq_restore(flags);
795
796 SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n",
797 init_utsname()->release,
798 (int)strcspn(init_utsname()->version, " "),
799 init_utsname()->version);
800
801 #define P(x) \
802 SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
803 #define PN(x) \
804 SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
805 PN(ktime);
806 PN(sched_clk);
807 PN(cpu_clk);
808 P(jiffies);
809 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
810 P(sched_clock_stable());
811 #endif
812 #undef PN
813 #undef P
814
815 SEQ_printf(m, "\n");
816 SEQ_printf(m, "sysctl_sched\n");
817
818 #define P(x) \
819 SEQ_printf(m, " .%-40s: %Ld\n", #x, (long long)(x))
820 #define PN(x) \
821 SEQ_printf(m, " .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
822 PN(sysctl_sched_latency);
823 PN(sysctl_sched_min_granularity);
824 PN(sysctl_sched_idle_min_granularity);
825 PN(sysctl_sched_wakeup_granularity);
826 P(sysctl_sched_child_runs_first);
827 P(sysctl_sched_features);
828 #undef PN
829 #undef P
830
831 SEQ_printf(m, " .%-40s: %d (%s)\n",
832 "sysctl_sched_tunable_scaling",
833 sysctl_sched_tunable_scaling,
834 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
835 SEQ_printf(m, "\n");
836 }
837
sched_debug_show(struct seq_file * m,void * v)838 static int sched_debug_show(struct seq_file *m, void *v)
839 {
840 int cpu = (unsigned long)(v - 2);
841
842 if (cpu != -1)
843 print_cpu(m, cpu);
844 else
845 sched_debug_header(m);
846
847 return 0;
848 }
849
sysrq_sched_debug_show(void)850 void sysrq_sched_debug_show(void)
851 {
852 int cpu;
853
854 sched_debug_header(NULL);
855 for_each_online_cpu(cpu) {
856 /*
857 * Need to reset softlockup watchdogs on all CPUs, because
858 * another CPU might be blocked waiting for us to process
859 * an IPI or stop_machine.
860 */
861 touch_nmi_watchdog();
862 touch_all_softlockup_watchdogs();
863 print_cpu(NULL, cpu);
864 }
865 }
866
867 /*
868 * This iterator needs some explanation.
869 * It returns 1 for the header position.
870 * This means 2 is CPU 0.
871 * In a hotplugged system some CPUs, including CPU 0, may be missing so we have
872 * to use cpumask_* to iterate over the CPUs.
873 */
sched_debug_start(struct seq_file * file,loff_t * offset)874 static void *sched_debug_start(struct seq_file *file, loff_t *offset)
875 {
876 unsigned long n = *offset;
877
878 if (n == 0)
879 return (void *) 1;
880
881 n--;
882
883 if (n > 0)
884 n = cpumask_next(n - 1, cpu_online_mask);
885 else
886 n = cpumask_first(cpu_online_mask);
887
888 *offset = n + 1;
889
890 if (n < nr_cpu_ids)
891 return (void *)(unsigned long)(n + 2);
892
893 return NULL;
894 }
895
sched_debug_next(struct seq_file * file,void * data,loff_t * offset)896 static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
897 {
898 (*offset)++;
899 return sched_debug_start(file, offset);
900 }
901
sched_debug_stop(struct seq_file * file,void * data)902 static void sched_debug_stop(struct seq_file *file, void *data)
903 {
904 }
905
906 static const struct seq_operations sched_debug_sops = {
907 .start = sched_debug_start,
908 .next = sched_debug_next,
909 .stop = sched_debug_stop,
910 .show = sched_debug_show,
911 };
912
913 #define __PS(S, F) SEQ_printf(m, "%-45s:%21Ld\n", S, (long long)(F))
914 #define __P(F) __PS(#F, F)
915 #define P(F) __PS(#F, p->F)
916 #define PM(F, M) __PS(#F, p->F & (M))
917 #define __PSN(S, F) SEQ_printf(m, "%-45s:%14Ld.%06ld\n", S, SPLIT_NS((long long)(F)))
918 #define __PN(F) __PSN(#F, F)
919 #define PN(F) __PSN(#F, p->F)
920
921
922 #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)923 void print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
924 unsigned long tpf, unsigned long gsf, unsigned long gpf)
925 {
926 SEQ_printf(m, "numa_faults node=%d ", node);
927 SEQ_printf(m, "task_private=%lu task_shared=%lu ", tpf, tsf);
928 SEQ_printf(m, "group_private=%lu group_shared=%lu\n", gpf, gsf);
929 }
930 #endif
931
932
sched_show_numa(struct task_struct * p,struct seq_file * m)933 static void sched_show_numa(struct task_struct *p, struct seq_file *m)
934 {
935 #ifdef CONFIG_NUMA_BALANCING
936 if (p->mm)
937 P(mm->numa_scan_seq);
938
939 P(numa_pages_migrated);
940 P(numa_preferred_nid);
941 P(total_numa_faults);
942 SEQ_printf(m, "current_node=%d, numa_group_id=%d\n",
943 task_node(p), task_numa_group_id(p));
944 show_numa_stats(p, m);
945 #endif
946 }
947
proc_sched_show_task(struct task_struct * p,struct pid_namespace * ns,struct seq_file * m)948 void proc_sched_show_task(struct task_struct *p, struct pid_namespace *ns,
949 struct seq_file *m)
950 {
951 unsigned long nr_switches;
952
953 SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr_ns(p, ns),
954 get_nr_threads(p));
955 SEQ_printf(m,
956 "---------------------------------------------------------"
957 "----------\n");
958
959 #define P_SCHEDSTAT(F) __PS(#F, schedstat_val(p->stats.F))
960 #define PN_SCHEDSTAT(F) __PSN(#F, schedstat_val(p->stats.F))
961
962 PN(se.exec_start);
963 PN(se.vruntime);
964 PN(se.sum_exec_runtime);
965
966 nr_switches = p->nvcsw + p->nivcsw;
967
968 P(se.nr_migrations);
969
970 if (schedstat_enabled()) {
971 u64 avg_atom, avg_per_cpu;
972
973 PN_SCHEDSTAT(sum_sleep_runtime);
974 PN_SCHEDSTAT(sum_block_runtime);
975 PN_SCHEDSTAT(wait_start);
976 PN_SCHEDSTAT(sleep_start);
977 PN_SCHEDSTAT(block_start);
978 PN_SCHEDSTAT(sleep_max);
979 PN_SCHEDSTAT(block_max);
980 PN_SCHEDSTAT(exec_max);
981 PN_SCHEDSTAT(slice_max);
982 PN_SCHEDSTAT(wait_max);
983 PN_SCHEDSTAT(wait_sum);
984 P_SCHEDSTAT(wait_count);
985 PN_SCHEDSTAT(iowait_sum);
986 P_SCHEDSTAT(iowait_count);
987 P_SCHEDSTAT(nr_migrations_cold);
988 P_SCHEDSTAT(nr_failed_migrations_affine);
989 P_SCHEDSTAT(nr_failed_migrations_running);
990 P_SCHEDSTAT(nr_failed_migrations_hot);
991 P_SCHEDSTAT(nr_forced_migrations);
992 P_SCHEDSTAT(nr_wakeups);
993 P_SCHEDSTAT(nr_wakeups_sync);
994 P_SCHEDSTAT(nr_wakeups_migrate);
995 P_SCHEDSTAT(nr_wakeups_local);
996 P_SCHEDSTAT(nr_wakeups_remote);
997 P_SCHEDSTAT(nr_wakeups_affine);
998 P_SCHEDSTAT(nr_wakeups_affine_attempts);
999 P_SCHEDSTAT(nr_wakeups_passive);
1000 P_SCHEDSTAT(nr_wakeups_idle);
1001
1002 avg_atom = p->se.sum_exec_runtime;
1003 if (nr_switches)
1004 avg_atom = div64_ul(avg_atom, nr_switches);
1005 else
1006 avg_atom = -1LL;
1007
1008 avg_per_cpu = p->se.sum_exec_runtime;
1009 if (p->se.nr_migrations) {
1010 avg_per_cpu = div64_u64(avg_per_cpu,
1011 p->se.nr_migrations);
1012 } else {
1013 avg_per_cpu = -1LL;
1014 }
1015
1016 __PN(avg_atom);
1017 __PN(avg_per_cpu);
1018
1019 #ifdef CONFIG_SCHED_CORE
1020 PN_SCHEDSTAT(core_forceidle_sum);
1021 #endif
1022 }
1023
1024 __P(nr_switches);
1025 __PS("nr_voluntary_switches", p->nvcsw);
1026 __PS("nr_involuntary_switches", p->nivcsw);
1027
1028 P(se.load.weight);
1029 #ifdef CONFIG_SMP
1030 P(se.avg.load_sum);
1031 P(se.avg.runnable_sum);
1032 P(se.avg.util_sum);
1033 P(se.avg.load_avg);
1034 P(se.avg.runnable_avg);
1035 P(se.avg.util_avg);
1036 P(se.avg.last_update_time);
1037 P(se.avg.util_est.ewma);
1038 PM(se.avg.util_est.enqueued, ~UTIL_AVG_UNCHANGED);
1039 #endif
1040 #ifdef CONFIG_UCLAMP_TASK
1041 __PS("uclamp.min", p->uclamp_req[UCLAMP_MIN].value);
1042 __PS("uclamp.max", p->uclamp_req[UCLAMP_MAX].value);
1043 __PS("effective uclamp.min", uclamp_eff_value(p, UCLAMP_MIN));
1044 __PS("effective uclamp.max", uclamp_eff_value(p, UCLAMP_MAX));
1045 #endif
1046 P(policy);
1047 P(prio);
1048 if (task_has_dl_policy(p)) {
1049 P(dl.runtime);
1050 P(dl.deadline);
1051 }
1052 #undef PN_SCHEDSTAT
1053 #undef P_SCHEDSTAT
1054
1055 {
1056 unsigned int this_cpu = raw_smp_processor_id();
1057 u64 t0, t1;
1058
1059 t0 = cpu_clock(this_cpu);
1060 t1 = cpu_clock(this_cpu);
1061 __PS("clock-delta", t1-t0);
1062 }
1063
1064 sched_show_numa(p, m);
1065 }
1066
proc_sched_set_task(struct task_struct * p)1067 void proc_sched_set_task(struct task_struct *p)
1068 {
1069 #ifdef CONFIG_SCHEDSTATS
1070 memset(&p->stats, 0, sizeof(p->stats));
1071 #endif
1072 }
1073
resched_latency_warn(int cpu,u64 latency)1074 void resched_latency_warn(int cpu, u64 latency)
1075 {
1076 static DEFINE_RATELIMIT_STATE(latency_check_ratelimit, 60 * 60 * HZ, 1);
1077
1078 WARN(__ratelimit(&latency_check_ratelimit),
1079 "sched: CPU %d need_resched set for > %llu ns (%d ticks) "
1080 "without schedule\n",
1081 cpu, latency, cpu_rq(cpu)->ticks_without_resched);
1082 }
1083