xref: /linux-6.15/kernel/sched/debug.c (revision 1c36432b)
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  */
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 
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 
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 
85 static void sched_feat_disable(int i)
86 {
87 	static_key_disable_cpuslocked(&sched_feat_keys[i]);
88 }
89 
90 static void sched_feat_enable(int i)
91 {
92 	static_key_enable_cpuslocked(&sched_feat_keys[i]);
93 }
94 #else
95 static void sched_feat_disable(int i) { };
96 static void sched_feat_enable(int i) { };
97 #endif /* CONFIG_JUMP_LABEL */
98 
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
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 
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 
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 
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 
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 
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 
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 
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 static const struct seq_operations sched_debug_sops;
285 
286 static int sched_debug_open(struct inode *inode, struct file *filp)
287 {
288 	return seq_open(filp, &sched_debug_sops);
289 }
290 
291 static const struct file_operations sched_debug_fops = {
292 	.open		= sched_debug_open,
293 	.read		= seq_read,
294 	.llseek		= seq_lseek,
295 	.release	= seq_release,
296 };
297 
298 static struct dentry *debugfs_sched;
299 
300 static __init int sched_init_debug(void)
301 {
302 	struct dentry __maybe_unused *numa;
303 
304 	debugfs_sched = debugfs_create_dir("sched", NULL);
305 
306 	debugfs_create_file("features", 0644, debugfs_sched, NULL, &sched_feat_fops);
307 	debugfs_create_bool("verbose", 0644, debugfs_sched, &sched_debug_verbose);
308 #ifdef CONFIG_PREEMPT_DYNAMIC
309 	debugfs_create_file("preempt", 0644, debugfs_sched, NULL, &sched_dynamic_fops);
310 #endif
311 
312 	debugfs_create_u32("latency_ns", 0644, debugfs_sched, &sysctl_sched_latency);
313 	debugfs_create_u32("min_granularity_ns", 0644, debugfs_sched, &sysctl_sched_min_granularity);
314 	debugfs_create_u32("idle_min_granularity_ns", 0644, debugfs_sched, &sysctl_sched_idle_min_granularity);
315 	debugfs_create_u32("wakeup_granularity_ns", 0644, debugfs_sched, &sysctl_sched_wakeup_granularity);
316 
317 	debugfs_create_u32("latency_warn_ms", 0644, debugfs_sched, &sysctl_resched_latency_warn_ms);
318 	debugfs_create_u32("latency_warn_once", 0644, debugfs_sched, &sysctl_resched_latency_warn_once);
319 
320 #ifdef CONFIG_SMP
321 	debugfs_create_file("tunable_scaling", 0644, debugfs_sched, NULL, &sched_scaling_fops);
322 	debugfs_create_u32("migration_cost_ns", 0644, debugfs_sched, &sysctl_sched_migration_cost);
323 	debugfs_create_u32("nr_migrate", 0644, debugfs_sched, &sysctl_sched_nr_migrate);
324 
325 	mutex_lock(&sched_domains_mutex);
326 	update_sched_domain_debugfs();
327 	mutex_unlock(&sched_domains_mutex);
328 #endif
329 
330 #ifdef CONFIG_NUMA_BALANCING
331 	numa = debugfs_create_dir("numa_balancing", debugfs_sched);
332 
333 	debugfs_create_u32("scan_delay_ms", 0644, numa, &sysctl_numa_balancing_scan_delay);
334 	debugfs_create_u32("scan_period_min_ms", 0644, numa, &sysctl_numa_balancing_scan_period_min);
335 	debugfs_create_u32("scan_period_max_ms", 0644, numa, &sysctl_numa_balancing_scan_period_max);
336 	debugfs_create_u32("scan_size_mb", 0644, numa, &sysctl_numa_balancing_scan_size);
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_remove(debugfs_lookup(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",	#F, (long long)schedstat_val(F))
453 #define PN(F)		SEQ_printf(m, "  .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F))
454 #define PN_SCHEDSTAT(F)	SEQ_printf(m, "  .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)schedstat_val(F)))
455 
456 	if (!se)
457 		return;
458 
459 	PN(se->exec_start);
460 	PN(se->vruntime);
461 	PN(se->sum_exec_runtime);
462 
463 	if (schedstat_enabled()) {
464 		PN_SCHEDSTAT(se->statistics.wait_start);
465 		PN_SCHEDSTAT(se->statistics.sleep_start);
466 		PN_SCHEDSTAT(se->statistics.block_start);
467 		PN_SCHEDSTAT(se->statistics.sleep_max);
468 		PN_SCHEDSTAT(se->statistics.block_max);
469 		PN_SCHEDSTAT(se->statistics.exec_max);
470 		PN_SCHEDSTAT(se->statistics.slice_max);
471 		PN_SCHEDSTAT(se->statistics.wait_max);
472 		PN_SCHEDSTAT(se->statistics.wait_sum);
473 		P_SCHEDSTAT(se->statistics.wait_count);
474 	}
475 
476 	P(se->load.weight);
477 #ifdef CONFIG_SMP
478 	P(se->avg.load_avg);
479 	P(se->avg.util_avg);
480 	P(se->avg.runnable_avg);
481 #endif
482 
483 #undef PN_SCHEDSTAT
484 #undef PN
485 #undef P_SCHEDSTAT
486 #undef P
487 }
488 #endif
489 
490 #ifdef CONFIG_CGROUP_SCHED
491 static DEFINE_SPINLOCK(sched_debug_lock);
492 static char group_path[PATH_MAX];
493 
494 static void task_group_path(struct task_group *tg, char *path, int plen)
495 {
496 	if (autogroup_path(tg, path, plen))
497 		return;
498 
499 	cgroup_path(tg->css.cgroup, path, plen);
500 }
501 
502 /*
503  * Only 1 SEQ_printf_task_group_path() caller can use the full length
504  * group_path[] for cgroup path. Other simultaneous callers will have
505  * to use a shorter stack buffer. A "..." suffix is appended at the end
506  * of the stack buffer so that it will show up in case the output length
507  * matches the given buffer size to indicate possible path name truncation.
508  */
509 #define SEQ_printf_task_group_path(m, tg, fmt...)			\
510 {									\
511 	if (spin_trylock(&sched_debug_lock)) {				\
512 		task_group_path(tg, group_path, sizeof(group_path));	\
513 		SEQ_printf(m, fmt, group_path);				\
514 		spin_unlock(&sched_debug_lock);				\
515 	} else {							\
516 		char buf[128];						\
517 		char *bufend = buf + sizeof(buf) - 3;			\
518 		task_group_path(tg, buf, bufend - buf);			\
519 		strcpy(bufend - 1, "...");				\
520 		SEQ_printf(m, fmt, buf);				\
521 	}								\
522 }
523 #endif
524 
525 static void
526 print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
527 {
528 	if (task_current(rq, p))
529 		SEQ_printf(m, ">R");
530 	else
531 		SEQ_printf(m, " %c", task_state_to_char(p));
532 
533 	SEQ_printf(m, " %15s %5d %9Ld.%06ld %9Ld %5d ",
534 		p->comm, task_pid_nr(p),
535 		SPLIT_NS(p->se.vruntime),
536 		(long long)(p->nvcsw + p->nivcsw),
537 		p->prio);
538 
539 	SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
540 		SPLIT_NS(schedstat_val_or_zero(p->se.statistics.wait_sum)),
541 		SPLIT_NS(p->se.sum_exec_runtime),
542 		SPLIT_NS(schedstat_val_or_zero(p->se.statistics.sum_sleep_runtime)));
543 
544 #ifdef CONFIG_NUMA_BALANCING
545 	SEQ_printf(m, " %d %d", task_node(p), task_numa_group_id(p));
546 #endif
547 #ifdef CONFIG_CGROUP_SCHED
548 	SEQ_printf_task_group_path(m, task_group(p), " %s")
549 #endif
550 
551 	SEQ_printf(m, "\n");
552 }
553 
554 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
555 {
556 	struct task_struct *g, *p;
557 
558 	SEQ_printf(m, "\n");
559 	SEQ_printf(m, "runnable tasks:\n");
560 	SEQ_printf(m, " S            task   PID         tree-key  switches  prio"
561 		   "     wait-time             sum-exec        sum-sleep\n");
562 	SEQ_printf(m, "-------------------------------------------------------"
563 		   "------------------------------------------------------\n");
564 
565 	rcu_read_lock();
566 	for_each_process_thread(g, p) {
567 		if (task_cpu(p) != rq_cpu)
568 			continue;
569 
570 		print_task(m, rq, p);
571 	}
572 	rcu_read_unlock();
573 }
574 
575 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
576 {
577 	s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
578 		spread, rq0_min_vruntime, spread0;
579 	struct rq *rq = cpu_rq(cpu);
580 	struct sched_entity *last;
581 	unsigned long flags;
582 
583 #ifdef CONFIG_FAIR_GROUP_SCHED
584 	SEQ_printf(m, "\n");
585 	SEQ_printf_task_group_path(m, cfs_rq->tg, "cfs_rq[%d]:%s\n", cpu);
586 #else
587 	SEQ_printf(m, "\n");
588 	SEQ_printf(m, "cfs_rq[%d]:\n", cpu);
589 #endif
590 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "exec_clock",
591 			SPLIT_NS(cfs_rq->exec_clock));
592 
593 	raw_spin_rq_lock_irqsave(rq, flags);
594 	if (rb_first_cached(&cfs_rq->tasks_timeline))
595 		MIN_vruntime = (__pick_first_entity(cfs_rq))->vruntime;
596 	last = __pick_last_entity(cfs_rq);
597 	if (last)
598 		max_vruntime = last->vruntime;
599 	min_vruntime = cfs_rq->min_vruntime;
600 	rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
601 	raw_spin_rq_unlock_irqrestore(rq, flags);
602 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "MIN_vruntime",
603 			SPLIT_NS(MIN_vruntime));
604 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "min_vruntime",
605 			SPLIT_NS(min_vruntime));
606 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "max_vruntime",
607 			SPLIT_NS(max_vruntime));
608 	spread = max_vruntime - MIN_vruntime;
609 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread",
610 			SPLIT_NS(spread));
611 	spread0 = min_vruntime - rq0_min_vruntime;
612 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread0",
613 			SPLIT_NS(spread0));
614 	SEQ_printf(m, "  .%-30s: %d\n", "nr_spread_over",
615 			cfs_rq->nr_spread_over);
616 	SEQ_printf(m, "  .%-30s: %d\n", "nr_running", cfs_rq->nr_running);
617 	SEQ_printf(m, "  .%-30s: %d\n", "h_nr_running", cfs_rq->h_nr_running);
618 	SEQ_printf(m, "  .%-30s: %d\n", "idle_nr_running",
619 			cfs_rq->idle_nr_running);
620 	SEQ_printf(m, "  .%-30s: %d\n", "idle_h_nr_running",
621 			cfs_rq->idle_h_nr_running);
622 	SEQ_printf(m, "  .%-30s: %ld\n", "load", cfs_rq->load.weight);
623 #ifdef CONFIG_SMP
624 	SEQ_printf(m, "  .%-30s: %lu\n", "load_avg",
625 			cfs_rq->avg.load_avg);
626 	SEQ_printf(m, "  .%-30s: %lu\n", "runnable_avg",
627 			cfs_rq->avg.runnable_avg);
628 	SEQ_printf(m, "  .%-30s: %lu\n", "util_avg",
629 			cfs_rq->avg.util_avg);
630 	SEQ_printf(m, "  .%-30s: %u\n", "util_est_enqueued",
631 			cfs_rq->avg.util_est.enqueued);
632 	SEQ_printf(m, "  .%-30s: %ld\n", "removed.load_avg",
633 			cfs_rq->removed.load_avg);
634 	SEQ_printf(m, "  .%-30s: %ld\n", "removed.util_avg",
635 			cfs_rq->removed.util_avg);
636 	SEQ_printf(m, "  .%-30s: %ld\n", "removed.runnable_avg",
637 			cfs_rq->removed.runnable_avg);
638 #ifdef CONFIG_FAIR_GROUP_SCHED
639 	SEQ_printf(m, "  .%-30s: %lu\n", "tg_load_avg_contrib",
640 			cfs_rq->tg_load_avg_contrib);
641 	SEQ_printf(m, "  .%-30s: %ld\n", "tg_load_avg",
642 			atomic_long_read(&cfs_rq->tg->load_avg));
643 #endif
644 #endif
645 #ifdef CONFIG_CFS_BANDWIDTH
646 	SEQ_printf(m, "  .%-30s: %d\n", "throttled",
647 			cfs_rq->throttled);
648 	SEQ_printf(m, "  .%-30s: %d\n", "throttle_count",
649 			cfs_rq->throttle_count);
650 #endif
651 
652 #ifdef CONFIG_FAIR_GROUP_SCHED
653 	print_cfs_group_stats(m, cpu, cfs_rq->tg);
654 #endif
655 }
656 
657 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
658 {
659 #ifdef CONFIG_RT_GROUP_SCHED
660 	SEQ_printf(m, "\n");
661 	SEQ_printf_task_group_path(m, rt_rq->tg, "rt_rq[%d]:%s\n", cpu);
662 #else
663 	SEQ_printf(m, "\n");
664 	SEQ_printf(m, "rt_rq[%d]:\n", cpu);
665 #endif
666 
667 #define P(x) \
668 	SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
669 #define PU(x) \
670 	SEQ_printf(m, "  .%-30s: %lu\n", #x, (unsigned long)(rt_rq->x))
671 #define PN(x) \
672 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
673 
674 	PU(rt_nr_running);
675 #ifdef CONFIG_SMP
676 	PU(rt_nr_migratory);
677 #endif
678 	P(rt_throttled);
679 	PN(rt_time);
680 	PN(rt_runtime);
681 
682 #undef PN
683 #undef PU
684 #undef P
685 }
686 
687 void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq)
688 {
689 	struct dl_bw *dl_bw;
690 
691 	SEQ_printf(m, "\n");
692 	SEQ_printf(m, "dl_rq[%d]:\n", cpu);
693 
694 #define PU(x) \
695 	SEQ_printf(m, "  .%-30s: %lu\n", #x, (unsigned long)(dl_rq->x))
696 
697 	PU(dl_nr_running);
698 #ifdef CONFIG_SMP
699 	PU(dl_nr_migratory);
700 	dl_bw = &cpu_rq(cpu)->rd->dl_bw;
701 #else
702 	dl_bw = &dl_rq->dl_bw;
703 #endif
704 	SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->bw", dl_bw->bw);
705 	SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->total_bw", dl_bw->total_bw);
706 
707 #undef PU
708 }
709 
710 static void print_cpu(struct seq_file *m, int cpu)
711 {
712 	struct rq *rq = cpu_rq(cpu);
713 
714 #ifdef CONFIG_X86
715 	{
716 		unsigned int freq = cpu_khz ? : 1;
717 
718 		SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
719 			   cpu, freq / 1000, (freq % 1000));
720 	}
721 #else
722 	SEQ_printf(m, "cpu#%d\n", cpu);
723 #endif
724 
725 #define P(x)								\
726 do {									\
727 	if (sizeof(rq->x) == 4)						\
728 		SEQ_printf(m, "  .%-30s: %ld\n", #x, (long)(rq->x));	\
729 	else								\
730 		SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rq->x));\
731 } while (0)
732 
733 #define PN(x) \
734 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
735 
736 	P(nr_running);
737 	P(nr_switches);
738 	P(nr_uninterruptible);
739 	PN(next_balance);
740 	SEQ_printf(m, "  .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr)));
741 	PN(clock);
742 	PN(clock_task);
743 #undef P
744 #undef PN
745 
746 #ifdef CONFIG_SMP
747 #define P64(n) SEQ_printf(m, "  .%-30s: %Ld\n", #n, rq->n);
748 	P64(avg_idle);
749 	P64(max_idle_balance_cost);
750 #undef P64
751 #endif
752 
753 #define P(n) SEQ_printf(m, "  .%-30s: %d\n", #n, schedstat_val(rq->n));
754 	if (schedstat_enabled()) {
755 		P(yld_count);
756 		P(sched_count);
757 		P(sched_goidle);
758 		P(ttwu_count);
759 		P(ttwu_local);
760 	}
761 #undef P
762 
763 	print_cfs_stats(m, cpu);
764 	print_rt_stats(m, cpu);
765 	print_dl_stats(m, cpu);
766 
767 	print_rq(m, rq, cpu);
768 	SEQ_printf(m, "\n");
769 }
770 
771 static const char *sched_tunable_scaling_names[] = {
772 	"none",
773 	"logarithmic",
774 	"linear"
775 };
776 
777 static void sched_debug_header(struct seq_file *m)
778 {
779 	u64 ktime, sched_clk, cpu_clk;
780 	unsigned long flags;
781 
782 	local_irq_save(flags);
783 	ktime = ktime_to_ns(ktime_get());
784 	sched_clk = sched_clock();
785 	cpu_clk = local_clock();
786 	local_irq_restore(flags);
787 
788 	SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n",
789 		init_utsname()->release,
790 		(int)strcspn(init_utsname()->version, " "),
791 		init_utsname()->version);
792 
793 #define P(x) \
794 	SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
795 #define PN(x) \
796 	SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
797 	PN(ktime);
798 	PN(sched_clk);
799 	PN(cpu_clk);
800 	P(jiffies);
801 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
802 	P(sched_clock_stable());
803 #endif
804 #undef PN
805 #undef P
806 
807 	SEQ_printf(m, "\n");
808 	SEQ_printf(m, "sysctl_sched\n");
809 
810 #define P(x) \
811 	SEQ_printf(m, "  .%-40s: %Ld\n", #x, (long long)(x))
812 #define PN(x) \
813 	SEQ_printf(m, "  .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
814 	PN(sysctl_sched_latency);
815 	PN(sysctl_sched_min_granularity);
816 	PN(sysctl_sched_idle_min_granularity);
817 	PN(sysctl_sched_wakeup_granularity);
818 	P(sysctl_sched_child_runs_first);
819 	P(sysctl_sched_features);
820 #undef PN
821 #undef P
822 
823 	SEQ_printf(m, "  .%-40s: %d (%s)\n",
824 		"sysctl_sched_tunable_scaling",
825 		sysctl_sched_tunable_scaling,
826 		sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
827 	SEQ_printf(m, "\n");
828 }
829 
830 static int sched_debug_show(struct seq_file *m, void *v)
831 {
832 	int cpu = (unsigned long)(v - 2);
833 
834 	if (cpu != -1)
835 		print_cpu(m, cpu);
836 	else
837 		sched_debug_header(m);
838 
839 	return 0;
840 }
841 
842 void sysrq_sched_debug_show(void)
843 {
844 	int cpu;
845 
846 	sched_debug_header(NULL);
847 	for_each_online_cpu(cpu) {
848 		/*
849 		 * Need to reset softlockup watchdogs on all CPUs, because
850 		 * another CPU might be blocked waiting for us to process
851 		 * an IPI or stop_machine.
852 		 */
853 		touch_nmi_watchdog();
854 		touch_all_softlockup_watchdogs();
855 		print_cpu(NULL, cpu);
856 	}
857 }
858 
859 /*
860  * This iterator needs some explanation.
861  * It returns 1 for the header position.
862  * This means 2 is CPU 0.
863  * In a hotplugged system some CPUs, including CPU 0, may be missing so we have
864  * to use cpumask_* to iterate over the CPUs.
865  */
866 static void *sched_debug_start(struct seq_file *file, loff_t *offset)
867 {
868 	unsigned long n = *offset;
869 
870 	if (n == 0)
871 		return (void *) 1;
872 
873 	n--;
874 
875 	if (n > 0)
876 		n = cpumask_next(n - 1, cpu_online_mask);
877 	else
878 		n = cpumask_first(cpu_online_mask);
879 
880 	*offset = n + 1;
881 
882 	if (n < nr_cpu_ids)
883 		return (void *)(unsigned long)(n + 2);
884 
885 	return NULL;
886 }
887 
888 static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
889 {
890 	(*offset)++;
891 	return sched_debug_start(file, offset);
892 }
893 
894 static void sched_debug_stop(struct seq_file *file, void *data)
895 {
896 }
897 
898 static const struct seq_operations sched_debug_sops = {
899 	.start		= sched_debug_start,
900 	.next		= sched_debug_next,
901 	.stop		= sched_debug_stop,
902 	.show		= sched_debug_show,
903 };
904 
905 #define __PS(S, F) SEQ_printf(m, "%-45s:%21Ld\n", S, (long long)(F))
906 #define __P(F) __PS(#F, F)
907 #define   P(F) __PS(#F, p->F)
908 #define   PM(F, M) __PS(#F, p->F & (M))
909 #define __PSN(S, F) SEQ_printf(m, "%-45s:%14Ld.%06ld\n", S, SPLIT_NS((long long)(F)))
910 #define __PN(F) __PSN(#F, F)
911 #define   PN(F) __PSN(#F, p->F)
912 
913 
914 #ifdef CONFIG_NUMA_BALANCING
915 void print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
916 		unsigned long tpf, unsigned long gsf, unsigned long gpf)
917 {
918 	SEQ_printf(m, "numa_faults node=%d ", node);
919 	SEQ_printf(m, "task_private=%lu task_shared=%lu ", tpf, tsf);
920 	SEQ_printf(m, "group_private=%lu group_shared=%lu\n", gpf, gsf);
921 }
922 #endif
923 
924 
925 static void sched_show_numa(struct task_struct *p, struct seq_file *m)
926 {
927 #ifdef CONFIG_NUMA_BALANCING
928 	struct mempolicy *pol;
929 
930 	if (p->mm)
931 		P(mm->numa_scan_seq);
932 
933 	task_lock(p);
934 	pol = p->mempolicy;
935 	if (pol && !(pol->flags & MPOL_F_MORON))
936 		pol = NULL;
937 	mpol_get(pol);
938 	task_unlock(p);
939 
940 	P(numa_pages_migrated);
941 	P(numa_preferred_nid);
942 	P(total_numa_faults);
943 	SEQ_printf(m, "current_node=%d, numa_group_id=%d\n",
944 			task_node(p), task_numa_group_id(p));
945 	show_numa_stats(p, m);
946 	mpol_put(pol);
947 #endif
948 }
949 
950 void proc_sched_show_task(struct task_struct *p, struct pid_namespace *ns,
951 						  struct seq_file *m)
952 {
953 	unsigned long nr_switches;
954 
955 	SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr_ns(p, ns),
956 						get_nr_threads(p));
957 	SEQ_printf(m,
958 		"---------------------------------------------------------"
959 		"----------\n");
960 
961 #define P_SCHEDSTAT(F)  __PS(#F, schedstat_val(p->F))
962 #define PN_SCHEDSTAT(F) __PSN(#F, schedstat_val(p->F))
963 
964 	PN(se.exec_start);
965 	PN(se.vruntime);
966 	PN(se.sum_exec_runtime);
967 
968 	nr_switches = p->nvcsw + p->nivcsw;
969 
970 	P(se.nr_migrations);
971 
972 	if (schedstat_enabled()) {
973 		u64 avg_atom, avg_per_cpu;
974 
975 		PN_SCHEDSTAT(se.statistics.sum_sleep_runtime);
976 		PN_SCHEDSTAT(se.statistics.wait_start);
977 		PN_SCHEDSTAT(se.statistics.sleep_start);
978 		PN_SCHEDSTAT(se.statistics.block_start);
979 		PN_SCHEDSTAT(se.statistics.sleep_max);
980 		PN_SCHEDSTAT(se.statistics.block_max);
981 		PN_SCHEDSTAT(se.statistics.exec_max);
982 		PN_SCHEDSTAT(se.statistics.slice_max);
983 		PN_SCHEDSTAT(se.statistics.wait_max);
984 		PN_SCHEDSTAT(se.statistics.wait_sum);
985 		P_SCHEDSTAT(se.statistics.wait_count);
986 		PN_SCHEDSTAT(se.statistics.iowait_sum);
987 		P_SCHEDSTAT(se.statistics.iowait_count);
988 		P_SCHEDSTAT(se.statistics.nr_migrations_cold);
989 		P_SCHEDSTAT(se.statistics.nr_failed_migrations_affine);
990 		P_SCHEDSTAT(se.statistics.nr_failed_migrations_running);
991 		P_SCHEDSTAT(se.statistics.nr_failed_migrations_hot);
992 		P_SCHEDSTAT(se.statistics.nr_forced_migrations);
993 		P_SCHEDSTAT(se.statistics.nr_wakeups);
994 		P_SCHEDSTAT(se.statistics.nr_wakeups_sync);
995 		P_SCHEDSTAT(se.statistics.nr_wakeups_migrate);
996 		P_SCHEDSTAT(se.statistics.nr_wakeups_local);
997 		P_SCHEDSTAT(se.statistics.nr_wakeups_remote);
998 		P_SCHEDSTAT(se.statistics.nr_wakeups_affine);
999 		P_SCHEDSTAT(se.statistics.nr_wakeups_affine_attempts);
1000 		P_SCHEDSTAT(se.statistics.nr_wakeups_passive);
1001 		P_SCHEDSTAT(se.statistics.nr_wakeups_idle);
1002 
1003 		avg_atom = p->se.sum_exec_runtime;
1004 		if (nr_switches)
1005 			avg_atom = div64_ul(avg_atom, nr_switches);
1006 		else
1007 			avg_atom = -1LL;
1008 
1009 		avg_per_cpu = p->se.sum_exec_runtime;
1010 		if (p->se.nr_migrations) {
1011 			avg_per_cpu = div64_u64(avg_per_cpu,
1012 						p->se.nr_migrations);
1013 		} else {
1014 			avg_per_cpu = -1LL;
1015 		}
1016 
1017 		__PN(avg_atom);
1018 		__PN(avg_per_cpu);
1019 	}
1020 
1021 	__P(nr_switches);
1022 	__PS("nr_voluntary_switches", p->nvcsw);
1023 	__PS("nr_involuntary_switches", p->nivcsw);
1024 
1025 	P(se.load.weight);
1026 #ifdef CONFIG_SMP
1027 	P(se.avg.load_sum);
1028 	P(se.avg.runnable_sum);
1029 	P(se.avg.util_sum);
1030 	P(se.avg.load_avg);
1031 	P(se.avg.runnable_avg);
1032 	P(se.avg.util_avg);
1033 	P(se.avg.last_update_time);
1034 	P(se.avg.util_est.ewma);
1035 	PM(se.avg.util_est.enqueued, ~UTIL_AVG_UNCHANGED);
1036 #endif
1037 #ifdef CONFIG_UCLAMP_TASK
1038 	__PS("uclamp.min", p->uclamp_req[UCLAMP_MIN].value);
1039 	__PS("uclamp.max", p->uclamp_req[UCLAMP_MAX].value);
1040 	__PS("effective uclamp.min", uclamp_eff_value(p, UCLAMP_MIN));
1041 	__PS("effective uclamp.max", uclamp_eff_value(p, UCLAMP_MAX));
1042 #endif
1043 	P(policy);
1044 	P(prio);
1045 	if (task_has_dl_policy(p)) {
1046 		P(dl.runtime);
1047 		P(dl.deadline);
1048 	}
1049 #undef PN_SCHEDSTAT
1050 #undef P_SCHEDSTAT
1051 
1052 	{
1053 		unsigned int this_cpu = raw_smp_processor_id();
1054 		u64 t0, t1;
1055 
1056 		t0 = cpu_clock(this_cpu);
1057 		t1 = cpu_clock(this_cpu);
1058 		__PS("clock-delta", t1-t0);
1059 	}
1060 
1061 	sched_show_numa(p, m);
1062 }
1063 
1064 void proc_sched_set_task(struct task_struct *p)
1065 {
1066 #ifdef CONFIG_SCHEDSTATS
1067 	memset(&p->se.statistics, 0, sizeof(p->se.statistics));
1068 #endif
1069 }
1070 
1071 void resched_latency_warn(int cpu, u64 latency)
1072 {
1073 	static DEFINE_RATELIMIT_STATE(latency_check_ratelimit, 60 * 60 * HZ, 1);
1074 
1075 	WARN(__ratelimit(&latency_check_ratelimit),
1076 	     "sched: CPU %d need_resched set for > %llu ns (%d ticks) "
1077 	     "without schedule\n",
1078 	     cpu, latency, cpu_rq(cpu)->ticks_without_resched);
1079 }
1080