xref: /linux/kernel/sched/debug.c (revision e80a48bade619ec5a92230b3d4ae84bfc2746822)
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  */
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 
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 static const char * const sched_feat_names[] = {
51 #include "features.h"
52 };
53 
54 #undef SCHED_FEAT
55 
56 static int sched_feat_show(struct seq_file *m, void *v)
57 {
58 	int i;
59 
60 	for (i = 0; i < __SCHED_FEAT_NR; i++) {
61 		if (!(sysctl_sched_features & (1UL << i)))
62 			seq_puts(m, "NO_");
63 		seq_printf(m, "%s ", sched_feat_names[i]);
64 	}
65 	seq_puts(m, "\n");
66 
67 	return 0;
68 }
69 
70 #ifdef CONFIG_JUMP_LABEL
71 
72 #define jump_label_key__true  STATIC_KEY_INIT_TRUE
73 #define jump_label_key__false STATIC_KEY_INIT_FALSE
74 
75 #define SCHED_FEAT(name, enabled)	\
76 	jump_label_key__##enabled ,
77 
78 struct static_key sched_feat_keys[__SCHED_FEAT_NR] = {
79 #include "features.h"
80 };
81 
82 #undef SCHED_FEAT
83 
84 static void sched_feat_disable(int i)
85 {
86 	static_key_disable_cpuslocked(&sched_feat_keys[i]);
87 }
88 
89 static void sched_feat_enable(int i)
90 {
91 	static_key_enable_cpuslocked(&sched_feat_keys[i]);
92 }
93 #else
94 static void sched_feat_disable(int i) { };
95 static void sched_feat_enable(int i) { };
96 #endif /* CONFIG_JUMP_LABEL */
97 
98 static int sched_feat_set(char *cmp)
99 {
100 	int i;
101 	int neg = 0;
102 
103 	if (strncmp(cmp, "NO_", 3) == 0) {
104 		neg = 1;
105 		cmp += 3;
106 	}
107 
108 	i = match_string(sched_feat_names, __SCHED_FEAT_NR, cmp);
109 	if (i < 0)
110 		return i;
111 
112 	if (neg) {
113 		sysctl_sched_features &= ~(1UL << i);
114 		sched_feat_disable(i);
115 	} else {
116 		sysctl_sched_features |= (1UL << i);
117 		sched_feat_enable(i);
118 	}
119 
120 	return 0;
121 }
122 
123 static ssize_t
124 sched_feat_write(struct file *filp, const char __user *ubuf,
125 		size_t cnt, loff_t *ppos)
126 {
127 	char buf[64];
128 	char *cmp;
129 	int ret;
130 	struct inode *inode;
131 
132 	if (cnt > 63)
133 		cnt = 63;
134 
135 	if (copy_from_user(&buf, ubuf, cnt))
136 		return -EFAULT;
137 
138 	buf[cnt] = 0;
139 	cmp = strstrip(buf);
140 
141 	/* Ensure the static_key remains in a consistent state */
142 	inode = file_inode(filp);
143 	cpus_read_lock();
144 	inode_lock(inode);
145 	ret = sched_feat_set(cmp);
146 	inode_unlock(inode);
147 	cpus_read_unlock();
148 	if (ret < 0)
149 		return ret;
150 
151 	*ppos += cnt;
152 
153 	return cnt;
154 }
155 
156 static int sched_feat_open(struct inode *inode, struct file *filp)
157 {
158 	return single_open(filp, sched_feat_show, NULL);
159 }
160 
161 static const struct file_operations sched_feat_fops = {
162 	.open		= sched_feat_open,
163 	.write		= sched_feat_write,
164 	.read		= seq_read,
165 	.llseek		= seq_lseek,
166 	.release	= single_release,
167 };
168 
169 #ifdef CONFIG_SMP
170 
171 static ssize_t sched_scaling_write(struct file *filp, const char __user *ubuf,
172 				   size_t cnt, loff_t *ppos)
173 {
174 	char buf[16];
175 	unsigned int scaling;
176 
177 	if (cnt > 15)
178 		cnt = 15;
179 
180 	if (copy_from_user(&buf, ubuf, cnt))
181 		return -EFAULT;
182 	buf[cnt] = '\0';
183 
184 	if (kstrtouint(buf, 10, &scaling))
185 		return -EINVAL;
186 
187 	if (scaling >= SCHED_TUNABLESCALING_END)
188 		return -EINVAL;
189 
190 	sysctl_sched_tunable_scaling = scaling;
191 	if (sched_update_scaling())
192 		return -EINVAL;
193 
194 	*ppos += cnt;
195 	return cnt;
196 }
197 
198 static int sched_scaling_show(struct seq_file *m, void *v)
199 {
200 	seq_printf(m, "%d\n", sysctl_sched_tunable_scaling);
201 	return 0;
202 }
203 
204 static int sched_scaling_open(struct inode *inode, struct file *filp)
205 {
206 	return single_open(filp, sched_scaling_show, NULL);
207 }
208 
209 static const struct file_operations sched_scaling_fops = {
210 	.open		= sched_scaling_open,
211 	.write		= sched_scaling_write,
212 	.read		= seq_read,
213 	.llseek		= seq_lseek,
214 	.release	= single_release,
215 };
216 
217 #endif /* SMP */
218 
219 #ifdef CONFIG_PREEMPT_DYNAMIC
220 
221 static ssize_t sched_dynamic_write(struct file *filp, const char __user *ubuf,
222 				   size_t cnt, loff_t *ppos)
223 {
224 	char buf[16];
225 	int mode;
226 
227 	if (cnt > 15)
228 		cnt = 15;
229 
230 	if (copy_from_user(&buf, ubuf, cnt))
231 		return -EFAULT;
232 
233 	buf[cnt] = 0;
234 	mode = sched_dynamic_mode(strstrip(buf));
235 	if (mode < 0)
236 		return mode;
237 
238 	sched_dynamic_update(mode);
239 
240 	*ppos += cnt;
241 
242 	return cnt;
243 }
244 
245 static int sched_dynamic_show(struct seq_file *m, void *v)
246 {
247 	static const char * preempt_modes[] = {
248 		"none", "voluntary", "full"
249 	};
250 	int i;
251 
252 	for (i = 0; i < ARRAY_SIZE(preempt_modes); i++) {
253 		if (preempt_dynamic_mode == i)
254 			seq_puts(m, "(");
255 		seq_puts(m, preempt_modes[i]);
256 		if (preempt_dynamic_mode == i)
257 			seq_puts(m, ")");
258 
259 		seq_puts(m, " ");
260 	}
261 
262 	seq_puts(m, "\n");
263 	return 0;
264 }
265 
266 static int sched_dynamic_open(struct inode *inode, struct file *filp)
267 {
268 	return single_open(filp, sched_dynamic_show, NULL);
269 }
270 
271 static const struct file_operations sched_dynamic_fops = {
272 	.open		= sched_dynamic_open,
273 	.write		= sched_dynamic_write,
274 	.read		= seq_read,
275 	.llseek		= seq_lseek,
276 	.release	= single_release,
277 };
278 
279 #endif /* CONFIG_PREEMPT_DYNAMIC */
280 
281 __read_mostly bool sched_debug_verbose;
282 
283 static const struct seq_operations sched_debug_sops;
284 
285 static int sched_debug_open(struct inode *inode, struct file *filp)
286 {
287 	return seq_open(filp, &sched_debug_sops);
288 }
289 
290 static const struct file_operations sched_debug_fops = {
291 	.open		= sched_debug_open,
292 	.read		= seq_read,
293 	.llseek		= seq_lseek,
294 	.release	= seq_release,
295 };
296 
297 static struct dentry *debugfs_sched;
298 
299 static __init int sched_init_debug(void)
300 {
301 	struct dentry __maybe_unused *numa;
302 
303 	debugfs_sched = debugfs_create_dir("sched", NULL);
304 
305 	debugfs_create_file("features", 0644, debugfs_sched, NULL, &sched_feat_fops);
306 	debugfs_create_bool("verbose", 0644, debugfs_sched, &sched_debug_verbose);
307 #ifdef CONFIG_PREEMPT_DYNAMIC
308 	debugfs_create_file("preempt", 0644, debugfs_sched, NULL, &sched_dynamic_fops);
309 #endif
310 
311 	debugfs_create_u32("latency_ns", 0644, debugfs_sched, &sysctl_sched_latency);
312 	debugfs_create_u32("min_granularity_ns", 0644, debugfs_sched, &sysctl_sched_min_granularity);
313 	debugfs_create_u32("idle_min_granularity_ns", 0644, debugfs_sched, &sysctl_sched_idle_min_granularity);
314 	debugfs_create_u32("wakeup_granularity_ns", 0644, debugfs_sched, &sysctl_sched_wakeup_granularity);
315 
316 	debugfs_create_u32("latency_warn_ms", 0644, debugfs_sched, &sysctl_resched_latency_warn_ms);
317 	debugfs_create_u32("latency_warn_once", 0644, debugfs_sched, &sysctl_resched_latency_warn_once);
318 
319 #ifdef CONFIG_SMP
320 	debugfs_create_file("tunable_scaling", 0644, debugfs_sched, NULL, &sched_scaling_fops);
321 	debugfs_create_u32("migration_cost_ns", 0644, debugfs_sched, &sysctl_sched_migration_cost);
322 	debugfs_create_u32("nr_migrate", 0644, debugfs_sched, &sysctl_sched_nr_migrate);
323 
324 	mutex_lock(&sched_domains_mutex);
325 	update_sched_domain_debugfs();
326 	mutex_unlock(&sched_domains_mutex);
327 #endif
328 
329 #ifdef CONFIG_NUMA_BALANCING
330 	numa = debugfs_create_dir("numa_balancing", debugfs_sched);
331 
332 	debugfs_create_u32("scan_delay_ms", 0644, numa, &sysctl_numa_balancing_scan_delay);
333 	debugfs_create_u32("scan_period_min_ms", 0644, numa, &sysctl_numa_balancing_scan_period_min);
334 	debugfs_create_u32("scan_period_max_ms", 0644, numa, &sysctl_numa_balancing_scan_period_max);
335 	debugfs_create_u32("scan_size_mb", 0644, numa, &sysctl_numa_balancing_scan_size);
336 	debugfs_create_u32("hot_threshold_ms", 0644, numa, &sysctl_numa_balancing_hot_threshold);
337 #endif
338 
339 	debugfs_create_file("debug", 0444, debugfs_sched, NULL, &sched_debug_fops);
340 
341 	return 0;
342 }
343 late_initcall(sched_init_debug);
344 
345 #ifdef CONFIG_SMP
346 
347 static cpumask_var_t		sd_sysctl_cpus;
348 static struct dentry		*sd_dentry;
349 
350 static int sd_flags_show(struct seq_file *m, void *v)
351 {
352 	unsigned long flags = *(unsigned int *)m->private;
353 	int idx;
354 
355 	for_each_set_bit(idx, &flags, __SD_FLAG_CNT) {
356 		seq_puts(m, sd_flag_debug[idx].name);
357 		seq_puts(m, " ");
358 	}
359 	seq_puts(m, "\n");
360 
361 	return 0;
362 }
363 
364 static int sd_flags_open(struct inode *inode, struct file *file)
365 {
366 	return single_open(file, sd_flags_show, inode->i_private);
367 }
368 
369 static const struct file_operations sd_flags_fops = {
370 	.open		= sd_flags_open,
371 	.read		= seq_read,
372 	.llseek		= seq_lseek,
373 	.release	= single_release,
374 };
375 
376 static void register_sd(struct sched_domain *sd, struct dentry *parent)
377 {
378 #define SDM(type, mode, member)	\
379 	debugfs_create_##type(#member, mode, parent, &sd->member)
380 
381 	SDM(ulong, 0644, min_interval);
382 	SDM(ulong, 0644, max_interval);
383 	SDM(u64,   0644, max_newidle_lb_cost);
384 	SDM(u32,   0644, busy_factor);
385 	SDM(u32,   0644, imbalance_pct);
386 	SDM(u32,   0644, cache_nice_tries);
387 	SDM(str,   0444, name);
388 
389 #undef SDM
390 
391 	debugfs_create_file("flags", 0444, parent, &sd->flags, &sd_flags_fops);
392 }
393 
394 void update_sched_domain_debugfs(void)
395 {
396 	int cpu, i;
397 
398 	/*
399 	 * This can unfortunately be invoked before sched_debug_init() creates
400 	 * the debug directory. Don't touch sd_sysctl_cpus until then.
401 	 */
402 	if (!debugfs_sched)
403 		return;
404 
405 	if (!cpumask_available(sd_sysctl_cpus)) {
406 		if (!alloc_cpumask_var(&sd_sysctl_cpus, GFP_KERNEL))
407 			return;
408 		cpumask_copy(sd_sysctl_cpus, cpu_possible_mask);
409 	}
410 
411 	if (!sd_dentry)
412 		sd_dentry = debugfs_create_dir("domains", debugfs_sched);
413 
414 	for_each_cpu(cpu, sd_sysctl_cpus) {
415 		struct sched_domain *sd;
416 		struct dentry *d_cpu;
417 		char buf[32];
418 
419 		snprintf(buf, sizeof(buf), "cpu%d", cpu);
420 		debugfs_lookup_and_remove(buf, sd_dentry);
421 		d_cpu = debugfs_create_dir(buf, sd_dentry);
422 
423 		i = 0;
424 		for_each_domain(cpu, sd) {
425 			struct dentry *d_sd;
426 
427 			snprintf(buf, sizeof(buf), "domain%d", i);
428 			d_sd = debugfs_create_dir(buf, d_cpu);
429 
430 			register_sd(sd, d_sd);
431 			i++;
432 		}
433 
434 		__cpumask_clear_cpu(cpu, sd_sysctl_cpus);
435 	}
436 }
437 
438 void dirty_sched_domain_sysctl(int cpu)
439 {
440 	if (cpumask_available(sd_sysctl_cpus))
441 		__cpumask_set_cpu(cpu, sd_sysctl_cpus);
442 }
443 
444 #endif /* CONFIG_SMP */
445 
446 #ifdef CONFIG_FAIR_GROUP_SCHED
447 static void print_cfs_group_stats(struct seq_file *m, int cpu, struct task_group *tg)
448 {
449 	struct sched_entity *se = tg->se[cpu];
450 
451 #define P(F)		SEQ_printf(m, "  .%-30s: %lld\n",	#F, (long long)F)
452 #define P_SCHEDSTAT(F)	SEQ_printf(m, "  .%-30s: %lld\n",	\
453 		#F, (long long)schedstat_val(stats->F))
454 #define PN(F)		SEQ_printf(m, "  .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F))
455 #define PN_SCHEDSTAT(F)	SEQ_printf(m, "  .%-30s: %lld.%06ld\n", \
456 		#F, SPLIT_NS((long long)schedstat_val(stats->F)))
457 
458 	if (!se)
459 		return;
460 
461 	PN(se->exec_start);
462 	PN(se->vruntime);
463 	PN(se->sum_exec_runtime);
464 
465 	if (schedstat_enabled()) {
466 		struct sched_statistics *stats;
467 		stats = __schedstats_from_se(se);
468 
469 		PN_SCHEDSTAT(wait_start);
470 		PN_SCHEDSTAT(sleep_start);
471 		PN_SCHEDSTAT(block_start);
472 		PN_SCHEDSTAT(sleep_max);
473 		PN_SCHEDSTAT(block_max);
474 		PN_SCHEDSTAT(exec_max);
475 		PN_SCHEDSTAT(slice_max);
476 		PN_SCHEDSTAT(wait_max);
477 		PN_SCHEDSTAT(wait_sum);
478 		P_SCHEDSTAT(wait_count);
479 	}
480 
481 	P(se->load.weight);
482 #ifdef CONFIG_SMP
483 	P(se->avg.load_avg);
484 	P(se->avg.util_avg);
485 	P(se->avg.runnable_avg);
486 #endif
487 
488 #undef PN_SCHEDSTAT
489 #undef PN
490 #undef P_SCHEDSTAT
491 #undef P
492 }
493 #endif
494 
495 #ifdef CONFIG_CGROUP_SCHED
496 static DEFINE_SPINLOCK(sched_debug_lock);
497 static char group_path[PATH_MAX];
498 
499 static void task_group_path(struct task_group *tg, char *path, int plen)
500 {
501 	if (autogroup_path(tg, path, plen))
502 		return;
503 
504 	cgroup_path(tg->css.cgroup, path, plen);
505 }
506 
507 /*
508  * Only 1 SEQ_printf_task_group_path() caller can use the full length
509  * group_path[] for cgroup path. Other simultaneous callers will have
510  * to use a shorter stack buffer. A "..." suffix is appended at the end
511  * of the stack buffer so that it will show up in case the output length
512  * matches the given buffer size to indicate possible path name truncation.
513  */
514 #define SEQ_printf_task_group_path(m, tg, fmt...)			\
515 {									\
516 	if (spin_trylock(&sched_debug_lock)) {				\
517 		task_group_path(tg, group_path, sizeof(group_path));	\
518 		SEQ_printf(m, fmt, group_path);				\
519 		spin_unlock(&sched_debug_lock);				\
520 	} else {							\
521 		char buf[128];						\
522 		char *bufend = buf + sizeof(buf) - 3;			\
523 		task_group_path(tg, buf, bufend - buf);			\
524 		strcpy(bufend - 1, "...");				\
525 		SEQ_printf(m, fmt, buf);				\
526 	}								\
527 }
528 #endif
529 
530 static void
531 print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
532 {
533 	if (task_current(rq, p))
534 		SEQ_printf(m, ">R");
535 	else
536 		SEQ_printf(m, " %c", task_state_to_char(p));
537 
538 	SEQ_printf(m, " %15s %5d %9Ld.%06ld %9Ld %5d ",
539 		p->comm, task_pid_nr(p),
540 		SPLIT_NS(p->se.vruntime),
541 		(long long)(p->nvcsw + p->nivcsw),
542 		p->prio);
543 
544 	SEQ_printf(m, "%9lld.%06ld %9lld.%06ld %9lld.%06ld %9lld.%06ld",
545 		SPLIT_NS(schedstat_val_or_zero(p->stats.wait_sum)),
546 		SPLIT_NS(p->se.sum_exec_runtime),
547 		SPLIT_NS(schedstat_val_or_zero(p->stats.sum_sleep_runtime)),
548 		SPLIT_NS(schedstat_val_or_zero(p->stats.sum_block_runtime)));
549 
550 #ifdef CONFIG_NUMA_BALANCING
551 	SEQ_printf(m, " %d %d", task_node(p), task_numa_group_id(p));
552 #endif
553 #ifdef CONFIG_CGROUP_SCHED
554 	SEQ_printf_task_group_path(m, task_group(p), " %s")
555 #endif
556 
557 	SEQ_printf(m, "\n");
558 }
559 
560 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
561 {
562 	struct task_struct *g, *p;
563 
564 	SEQ_printf(m, "\n");
565 	SEQ_printf(m, "runnable tasks:\n");
566 	SEQ_printf(m, " S            task   PID         tree-key  switches  prio"
567 		   "     wait-time             sum-exec        sum-sleep\n");
568 	SEQ_printf(m, "-------------------------------------------------------"
569 		   "------------------------------------------------------\n");
570 
571 	rcu_read_lock();
572 	for_each_process_thread(g, p) {
573 		if (task_cpu(p) != rq_cpu)
574 			continue;
575 
576 		print_task(m, rq, p);
577 	}
578 	rcu_read_unlock();
579 }
580 
581 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
582 {
583 	s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
584 		spread, rq0_min_vruntime, spread0;
585 	struct rq *rq = cpu_rq(cpu);
586 	struct sched_entity *last;
587 	unsigned long flags;
588 
589 #ifdef CONFIG_FAIR_GROUP_SCHED
590 	SEQ_printf(m, "\n");
591 	SEQ_printf_task_group_path(m, cfs_rq->tg, "cfs_rq[%d]:%s\n", cpu);
592 #else
593 	SEQ_printf(m, "\n");
594 	SEQ_printf(m, "cfs_rq[%d]:\n", cpu);
595 #endif
596 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "exec_clock",
597 			SPLIT_NS(cfs_rq->exec_clock));
598 
599 	raw_spin_rq_lock_irqsave(rq, flags);
600 	if (rb_first_cached(&cfs_rq->tasks_timeline))
601 		MIN_vruntime = (__pick_first_entity(cfs_rq))->vruntime;
602 	last = __pick_last_entity(cfs_rq);
603 	if (last)
604 		max_vruntime = last->vruntime;
605 	min_vruntime = cfs_rq->min_vruntime;
606 	rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
607 	raw_spin_rq_unlock_irqrestore(rq, flags);
608 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "MIN_vruntime",
609 			SPLIT_NS(MIN_vruntime));
610 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "min_vruntime",
611 			SPLIT_NS(min_vruntime));
612 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "max_vruntime",
613 			SPLIT_NS(max_vruntime));
614 	spread = max_vruntime - MIN_vruntime;
615 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread",
616 			SPLIT_NS(spread));
617 	spread0 = min_vruntime - rq0_min_vruntime;
618 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread0",
619 			SPLIT_NS(spread0));
620 	SEQ_printf(m, "  .%-30s: %d\n", "nr_spread_over",
621 			cfs_rq->nr_spread_over);
622 	SEQ_printf(m, "  .%-30s: %d\n", "nr_running", cfs_rq->nr_running);
623 	SEQ_printf(m, "  .%-30s: %d\n", "h_nr_running", cfs_rq->h_nr_running);
624 	SEQ_printf(m, "  .%-30s: %d\n", "idle_nr_running",
625 			cfs_rq->idle_nr_running);
626 	SEQ_printf(m, "  .%-30s: %d\n", "idle_h_nr_running",
627 			cfs_rq->idle_h_nr_running);
628 	SEQ_printf(m, "  .%-30s: %ld\n", "load", cfs_rq->load.weight);
629 #ifdef CONFIG_SMP
630 	SEQ_printf(m, "  .%-30s: %lu\n", "load_avg",
631 			cfs_rq->avg.load_avg);
632 	SEQ_printf(m, "  .%-30s: %lu\n", "runnable_avg",
633 			cfs_rq->avg.runnable_avg);
634 	SEQ_printf(m, "  .%-30s: %lu\n", "util_avg",
635 			cfs_rq->avg.util_avg);
636 	SEQ_printf(m, "  .%-30s: %u\n", "util_est_enqueued",
637 			cfs_rq->avg.util_est.enqueued);
638 	SEQ_printf(m, "  .%-30s: %ld\n", "removed.load_avg",
639 			cfs_rq->removed.load_avg);
640 	SEQ_printf(m, "  .%-30s: %ld\n", "removed.util_avg",
641 			cfs_rq->removed.util_avg);
642 	SEQ_printf(m, "  .%-30s: %ld\n", "removed.runnable_avg",
643 			cfs_rq->removed.runnable_avg);
644 #ifdef CONFIG_FAIR_GROUP_SCHED
645 	SEQ_printf(m, "  .%-30s: %lu\n", "tg_load_avg_contrib",
646 			cfs_rq->tg_load_avg_contrib);
647 	SEQ_printf(m, "  .%-30s: %ld\n", "tg_load_avg",
648 			atomic_long_read(&cfs_rq->tg->load_avg));
649 #endif
650 #endif
651 #ifdef CONFIG_CFS_BANDWIDTH
652 	SEQ_printf(m, "  .%-30s: %d\n", "throttled",
653 			cfs_rq->throttled);
654 	SEQ_printf(m, "  .%-30s: %d\n", "throttle_count",
655 			cfs_rq->throttle_count);
656 #endif
657 
658 #ifdef CONFIG_FAIR_GROUP_SCHED
659 	print_cfs_group_stats(m, cpu, cfs_rq->tg);
660 #endif
661 }
662 
663 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
664 {
665 #ifdef CONFIG_RT_GROUP_SCHED
666 	SEQ_printf(m, "\n");
667 	SEQ_printf_task_group_path(m, rt_rq->tg, "rt_rq[%d]:%s\n", cpu);
668 #else
669 	SEQ_printf(m, "\n");
670 	SEQ_printf(m, "rt_rq[%d]:\n", cpu);
671 #endif
672 
673 #define P(x) \
674 	SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
675 #define PU(x) \
676 	SEQ_printf(m, "  .%-30s: %lu\n", #x, (unsigned long)(rt_rq->x))
677 #define PN(x) \
678 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
679 
680 	PU(rt_nr_running);
681 #ifdef CONFIG_SMP
682 	PU(rt_nr_migratory);
683 #endif
684 	P(rt_throttled);
685 	PN(rt_time);
686 	PN(rt_runtime);
687 
688 #undef PN
689 #undef PU
690 #undef P
691 }
692 
693 void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq)
694 {
695 	struct dl_bw *dl_bw;
696 
697 	SEQ_printf(m, "\n");
698 	SEQ_printf(m, "dl_rq[%d]:\n", cpu);
699 
700 #define PU(x) \
701 	SEQ_printf(m, "  .%-30s: %lu\n", #x, (unsigned long)(dl_rq->x))
702 
703 	PU(dl_nr_running);
704 #ifdef CONFIG_SMP
705 	PU(dl_nr_migratory);
706 	dl_bw = &cpu_rq(cpu)->rd->dl_bw;
707 #else
708 	dl_bw = &dl_rq->dl_bw;
709 #endif
710 	SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->bw", dl_bw->bw);
711 	SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->total_bw", dl_bw->total_bw);
712 
713 #undef PU
714 }
715 
716 static void print_cpu(struct seq_file *m, int cpu)
717 {
718 	struct rq *rq = cpu_rq(cpu);
719 
720 #ifdef CONFIG_X86
721 	{
722 		unsigned int freq = cpu_khz ? : 1;
723 
724 		SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
725 			   cpu, freq / 1000, (freq % 1000));
726 	}
727 #else
728 	SEQ_printf(m, "cpu#%d\n", cpu);
729 #endif
730 
731 #define P(x)								\
732 do {									\
733 	if (sizeof(rq->x) == 4)						\
734 		SEQ_printf(m, "  .%-30s: %ld\n", #x, (long)(rq->x));	\
735 	else								\
736 		SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rq->x));\
737 } while (0)
738 
739 #define PN(x) \
740 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
741 
742 	P(nr_running);
743 	P(nr_switches);
744 	P(nr_uninterruptible);
745 	PN(next_balance);
746 	SEQ_printf(m, "  .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr)));
747 	PN(clock);
748 	PN(clock_task);
749 #undef P
750 #undef PN
751 
752 #ifdef CONFIG_SMP
753 #define P64(n) SEQ_printf(m, "  .%-30s: %Ld\n", #n, rq->n);
754 	P64(avg_idle);
755 	P64(max_idle_balance_cost);
756 #undef P64
757 #endif
758 
759 #define P(n) SEQ_printf(m, "  .%-30s: %d\n", #n, schedstat_val(rq->n));
760 	if (schedstat_enabled()) {
761 		P(yld_count);
762 		P(sched_count);
763 		P(sched_goidle);
764 		P(ttwu_count);
765 		P(ttwu_local);
766 	}
767 #undef P
768 
769 	print_cfs_stats(m, cpu);
770 	print_rt_stats(m, cpu);
771 	print_dl_stats(m, cpu);
772 
773 	print_rq(m, rq, cpu);
774 	SEQ_printf(m, "\n");
775 }
776 
777 static const char *sched_tunable_scaling_names[] = {
778 	"none",
779 	"logarithmic",
780 	"linear"
781 };
782 
783 static void sched_debug_header(struct seq_file *m)
784 {
785 	u64 ktime, sched_clk, cpu_clk;
786 	unsigned long flags;
787 
788 	local_irq_save(flags);
789 	ktime = ktime_to_ns(ktime_get());
790 	sched_clk = sched_clock();
791 	cpu_clk = local_clock();
792 	local_irq_restore(flags);
793 
794 	SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n",
795 		init_utsname()->release,
796 		(int)strcspn(init_utsname()->version, " "),
797 		init_utsname()->version);
798 
799 #define P(x) \
800 	SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
801 #define PN(x) \
802 	SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
803 	PN(ktime);
804 	PN(sched_clk);
805 	PN(cpu_clk);
806 	P(jiffies);
807 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
808 	P(sched_clock_stable());
809 #endif
810 #undef PN
811 #undef P
812 
813 	SEQ_printf(m, "\n");
814 	SEQ_printf(m, "sysctl_sched\n");
815 
816 #define P(x) \
817 	SEQ_printf(m, "  .%-40s: %Ld\n", #x, (long long)(x))
818 #define PN(x) \
819 	SEQ_printf(m, "  .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
820 	PN(sysctl_sched_latency);
821 	PN(sysctl_sched_min_granularity);
822 	PN(sysctl_sched_idle_min_granularity);
823 	PN(sysctl_sched_wakeup_granularity);
824 	P(sysctl_sched_child_runs_first);
825 	P(sysctl_sched_features);
826 #undef PN
827 #undef P
828 
829 	SEQ_printf(m, "  .%-40s: %d (%s)\n",
830 		"sysctl_sched_tunable_scaling",
831 		sysctl_sched_tunable_scaling,
832 		sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
833 	SEQ_printf(m, "\n");
834 }
835 
836 static int sched_debug_show(struct seq_file *m, void *v)
837 {
838 	int cpu = (unsigned long)(v - 2);
839 
840 	if (cpu != -1)
841 		print_cpu(m, cpu);
842 	else
843 		sched_debug_header(m);
844 
845 	return 0;
846 }
847 
848 void sysrq_sched_debug_show(void)
849 {
850 	int cpu;
851 
852 	sched_debug_header(NULL);
853 	for_each_online_cpu(cpu) {
854 		/*
855 		 * Need to reset softlockup watchdogs on all CPUs, because
856 		 * another CPU might be blocked waiting for us to process
857 		 * an IPI or stop_machine.
858 		 */
859 		touch_nmi_watchdog();
860 		touch_all_softlockup_watchdogs();
861 		print_cpu(NULL, cpu);
862 	}
863 }
864 
865 /*
866  * This iterator needs some explanation.
867  * It returns 1 for the header position.
868  * This means 2 is CPU 0.
869  * In a hotplugged system some CPUs, including CPU 0, may be missing so we have
870  * to use cpumask_* to iterate over the CPUs.
871  */
872 static void *sched_debug_start(struct seq_file *file, loff_t *offset)
873 {
874 	unsigned long n = *offset;
875 
876 	if (n == 0)
877 		return (void *) 1;
878 
879 	n--;
880 
881 	if (n > 0)
882 		n = cpumask_next(n - 1, cpu_online_mask);
883 	else
884 		n = cpumask_first(cpu_online_mask);
885 
886 	*offset = n + 1;
887 
888 	if (n < nr_cpu_ids)
889 		return (void *)(unsigned long)(n + 2);
890 
891 	return NULL;
892 }
893 
894 static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
895 {
896 	(*offset)++;
897 	return sched_debug_start(file, offset);
898 }
899 
900 static void sched_debug_stop(struct seq_file *file, void *data)
901 {
902 }
903 
904 static const struct seq_operations sched_debug_sops = {
905 	.start		= sched_debug_start,
906 	.next		= sched_debug_next,
907 	.stop		= sched_debug_stop,
908 	.show		= sched_debug_show,
909 };
910 
911 #define __PS(S, F) SEQ_printf(m, "%-45s:%21Ld\n", S, (long long)(F))
912 #define __P(F) __PS(#F, F)
913 #define   P(F) __PS(#F, p->F)
914 #define   PM(F, M) __PS(#F, p->F & (M))
915 #define __PSN(S, F) SEQ_printf(m, "%-45s:%14Ld.%06ld\n", S, SPLIT_NS((long long)(F)))
916 #define __PN(F) __PSN(#F, F)
917 #define   PN(F) __PSN(#F, p->F)
918 
919 
920 #ifdef CONFIG_NUMA_BALANCING
921 void print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
922 		unsigned long tpf, unsigned long gsf, unsigned long gpf)
923 {
924 	SEQ_printf(m, "numa_faults node=%d ", node);
925 	SEQ_printf(m, "task_private=%lu task_shared=%lu ", tpf, tsf);
926 	SEQ_printf(m, "group_private=%lu group_shared=%lu\n", gpf, gsf);
927 }
928 #endif
929 
930 
931 static void sched_show_numa(struct task_struct *p, struct seq_file *m)
932 {
933 #ifdef CONFIG_NUMA_BALANCING
934 	if (p->mm)
935 		P(mm->numa_scan_seq);
936 
937 	P(numa_pages_migrated);
938 	P(numa_preferred_nid);
939 	P(total_numa_faults);
940 	SEQ_printf(m, "current_node=%d, numa_group_id=%d\n",
941 			task_node(p), task_numa_group_id(p));
942 	show_numa_stats(p, m);
943 #endif
944 }
945 
946 void proc_sched_show_task(struct task_struct *p, struct pid_namespace *ns,
947 						  struct seq_file *m)
948 {
949 	unsigned long nr_switches;
950 
951 	SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr_ns(p, ns),
952 						get_nr_threads(p));
953 	SEQ_printf(m,
954 		"---------------------------------------------------------"
955 		"----------\n");
956 
957 #define P_SCHEDSTAT(F)  __PS(#F, schedstat_val(p->stats.F))
958 #define PN_SCHEDSTAT(F) __PSN(#F, schedstat_val(p->stats.F))
959 
960 	PN(se.exec_start);
961 	PN(se.vruntime);
962 	PN(se.sum_exec_runtime);
963 
964 	nr_switches = p->nvcsw + p->nivcsw;
965 
966 	P(se.nr_migrations);
967 
968 	if (schedstat_enabled()) {
969 		u64 avg_atom, avg_per_cpu;
970 
971 		PN_SCHEDSTAT(sum_sleep_runtime);
972 		PN_SCHEDSTAT(sum_block_runtime);
973 		PN_SCHEDSTAT(wait_start);
974 		PN_SCHEDSTAT(sleep_start);
975 		PN_SCHEDSTAT(block_start);
976 		PN_SCHEDSTAT(sleep_max);
977 		PN_SCHEDSTAT(block_max);
978 		PN_SCHEDSTAT(exec_max);
979 		PN_SCHEDSTAT(slice_max);
980 		PN_SCHEDSTAT(wait_max);
981 		PN_SCHEDSTAT(wait_sum);
982 		P_SCHEDSTAT(wait_count);
983 		PN_SCHEDSTAT(iowait_sum);
984 		P_SCHEDSTAT(iowait_count);
985 		P_SCHEDSTAT(nr_migrations_cold);
986 		P_SCHEDSTAT(nr_failed_migrations_affine);
987 		P_SCHEDSTAT(nr_failed_migrations_running);
988 		P_SCHEDSTAT(nr_failed_migrations_hot);
989 		P_SCHEDSTAT(nr_forced_migrations);
990 		P_SCHEDSTAT(nr_wakeups);
991 		P_SCHEDSTAT(nr_wakeups_sync);
992 		P_SCHEDSTAT(nr_wakeups_migrate);
993 		P_SCHEDSTAT(nr_wakeups_local);
994 		P_SCHEDSTAT(nr_wakeups_remote);
995 		P_SCHEDSTAT(nr_wakeups_affine);
996 		P_SCHEDSTAT(nr_wakeups_affine_attempts);
997 		P_SCHEDSTAT(nr_wakeups_passive);
998 		P_SCHEDSTAT(nr_wakeups_idle);
999 
1000 		avg_atom = p->se.sum_exec_runtime;
1001 		if (nr_switches)
1002 			avg_atom = div64_ul(avg_atom, nr_switches);
1003 		else
1004 			avg_atom = -1LL;
1005 
1006 		avg_per_cpu = p->se.sum_exec_runtime;
1007 		if (p->se.nr_migrations) {
1008 			avg_per_cpu = div64_u64(avg_per_cpu,
1009 						p->se.nr_migrations);
1010 		} else {
1011 			avg_per_cpu = -1LL;
1012 		}
1013 
1014 		__PN(avg_atom);
1015 		__PN(avg_per_cpu);
1016 
1017 #ifdef CONFIG_SCHED_CORE
1018 		PN_SCHEDSTAT(core_forceidle_sum);
1019 #endif
1020 	}
1021 
1022 	__P(nr_switches);
1023 	__PS("nr_voluntary_switches", p->nvcsw);
1024 	__PS("nr_involuntary_switches", p->nivcsw);
1025 
1026 	P(se.load.weight);
1027 #ifdef CONFIG_SMP
1028 	P(se.avg.load_sum);
1029 	P(se.avg.runnable_sum);
1030 	P(se.avg.util_sum);
1031 	P(se.avg.load_avg);
1032 	P(se.avg.runnable_avg);
1033 	P(se.avg.util_avg);
1034 	P(se.avg.last_update_time);
1035 	P(se.avg.util_est.ewma);
1036 	PM(se.avg.util_est.enqueued, ~UTIL_AVG_UNCHANGED);
1037 #endif
1038 #ifdef CONFIG_UCLAMP_TASK
1039 	__PS("uclamp.min", p->uclamp_req[UCLAMP_MIN].value);
1040 	__PS("uclamp.max", p->uclamp_req[UCLAMP_MAX].value);
1041 	__PS("effective uclamp.min", uclamp_eff_value(p, UCLAMP_MIN));
1042 	__PS("effective uclamp.max", uclamp_eff_value(p, UCLAMP_MAX));
1043 #endif
1044 	P(policy);
1045 	P(prio);
1046 	if (task_has_dl_policy(p)) {
1047 		P(dl.runtime);
1048 		P(dl.deadline);
1049 	}
1050 #undef PN_SCHEDSTAT
1051 #undef P_SCHEDSTAT
1052 
1053 	{
1054 		unsigned int this_cpu = raw_smp_processor_id();
1055 		u64 t0, t1;
1056 
1057 		t0 = cpu_clock(this_cpu);
1058 		t1 = cpu_clock(this_cpu);
1059 		__PS("clock-delta", t1-t0);
1060 	}
1061 
1062 	sched_show_numa(p, m);
1063 }
1064 
1065 void proc_sched_set_task(struct task_struct *p)
1066 {
1067 #ifdef CONFIG_SCHEDSTATS
1068 	memset(&p->stats, 0, sizeof(p->stats));
1069 #endif
1070 }
1071 
1072 void resched_latency_warn(int cpu, u64 latency)
1073 {
1074 	static DEFINE_RATELIMIT_STATE(latency_check_ratelimit, 60 * 60 * HZ, 1);
1075 
1076 	WARN(__ratelimit(&latency_check_ratelimit),
1077 	     "sched: CPU %d need_resched set for > %llu ns (%d ticks) "
1078 	     "without schedule\n",
1079 	     cpu, latency, cpu_rq(cpu)->ticks_without_resched);
1080 }
1081