xref: /openbmc/linux/kernel/sched/debug.c (revision ceeadb83)
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",	\
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 =  __schedstats_from_se(se);
467 
468 		PN_SCHEDSTAT(wait_start);
469 		PN_SCHEDSTAT(sleep_start);
470 		PN_SCHEDSTAT(block_start);
471 		PN_SCHEDSTAT(sleep_max);
472 		PN_SCHEDSTAT(block_max);
473 		PN_SCHEDSTAT(exec_max);
474 		PN_SCHEDSTAT(slice_max);
475 		PN_SCHEDSTAT(wait_max);
476 		PN_SCHEDSTAT(wait_sum);
477 		P_SCHEDSTAT(wait_count);
478 	}
479 
480 	P(se->load.weight);
481 #ifdef CONFIG_SMP
482 	P(se->avg.load_avg);
483 	P(se->avg.util_avg);
484 	P(se->avg.runnable_avg);
485 #endif
486 
487 #undef PN_SCHEDSTAT
488 #undef PN
489 #undef P_SCHEDSTAT
490 #undef P
491 }
492 #endif
493 
494 #ifdef CONFIG_CGROUP_SCHED
495 static DEFINE_SPINLOCK(sched_debug_lock);
496 static char group_path[PATH_MAX];
497 
498 static void task_group_path(struct task_group *tg, char *path, int plen)
499 {
500 	if (autogroup_path(tg, path, plen))
501 		return;
502 
503 	cgroup_path(tg->css.cgroup, path, plen);
504 }
505 
506 /*
507  * Only 1 SEQ_printf_task_group_path() caller can use the full length
508  * group_path[] for cgroup path. Other simultaneous callers will have
509  * to use a shorter stack buffer. A "..." suffix is appended at the end
510  * of the stack buffer so that it will show up in case the output length
511  * matches the given buffer size to indicate possible path name truncation.
512  */
513 #define SEQ_printf_task_group_path(m, tg, fmt...)			\
514 {									\
515 	if (spin_trylock(&sched_debug_lock)) {				\
516 		task_group_path(tg, group_path, sizeof(group_path));	\
517 		SEQ_printf(m, fmt, group_path);				\
518 		spin_unlock(&sched_debug_lock);				\
519 	} else {							\
520 		char buf[128];						\
521 		char *bufend = buf + sizeof(buf) - 3;			\
522 		task_group_path(tg, buf, bufend - buf);			\
523 		strcpy(bufend - 1, "...");				\
524 		SEQ_printf(m, fmt, buf);				\
525 	}								\
526 }
527 #endif
528 
529 static void
530 print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
531 {
532 	if (task_current(rq, p))
533 		SEQ_printf(m, ">R");
534 	else
535 		SEQ_printf(m, " %c", task_state_to_char(p));
536 
537 	SEQ_printf(m, " %15s %5d %9Ld.%06ld %9Ld %5d ",
538 		p->comm, task_pid_nr(p),
539 		SPLIT_NS(p->se.vruntime),
540 		(long long)(p->nvcsw + p->nivcsw),
541 		p->prio);
542 
543 	SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
544 		SPLIT_NS(schedstat_val_or_zero(p->stats.wait_sum)),
545 		SPLIT_NS(p->se.sum_exec_runtime),
546 		SPLIT_NS(schedstat_val_or_zero(p->stats.sum_sleep_runtime)));
547 
548 #ifdef CONFIG_NUMA_BALANCING
549 	SEQ_printf(m, " %d %d", task_node(p), task_numa_group_id(p));
550 #endif
551 #ifdef CONFIG_CGROUP_SCHED
552 	SEQ_printf_task_group_path(m, task_group(p), " %s")
553 #endif
554 
555 	SEQ_printf(m, "\n");
556 }
557 
558 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
559 {
560 	struct task_struct *g, *p;
561 
562 	SEQ_printf(m, "\n");
563 	SEQ_printf(m, "runnable tasks:\n");
564 	SEQ_printf(m, " S            task   PID         tree-key  switches  prio"
565 		   "     wait-time             sum-exec        sum-sleep\n");
566 	SEQ_printf(m, "-------------------------------------------------------"
567 		   "------------------------------------------------------\n");
568 
569 	rcu_read_lock();
570 	for_each_process_thread(g, p) {
571 		if (task_cpu(p) != rq_cpu)
572 			continue;
573 
574 		print_task(m, rq, p);
575 	}
576 	rcu_read_unlock();
577 }
578 
579 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
580 {
581 	s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
582 		spread, rq0_min_vruntime, spread0;
583 	struct rq *rq = cpu_rq(cpu);
584 	struct sched_entity *last;
585 	unsigned long flags;
586 
587 #ifdef CONFIG_FAIR_GROUP_SCHED
588 	SEQ_printf(m, "\n");
589 	SEQ_printf_task_group_path(m, cfs_rq->tg, "cfs_rq[%d]:%s\n", cpu);
590 #else
591 	SEQ_printf(m, "\n");
592 	SEQ_printf(m, "cfs_rq[%d]:\n", cpu);
593 #endif
594 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "exec_clock",
595 			SPLIT_NS(cfs_rq->exec_clock));
596 
597 	raw_spin_rq_lock_irqsave(rq, flags);
598 	if (rb_first_cached(&cfs_rq->tasks_timeline))
599 		MIN_vruntime = (__pick_first_entity(cfs_rq))->vruntime;
600 	last = __pick_last_entity(cfs_rq);
601 	if (last)
602 		max_vruntime = last->vruntime;
603 	min_vruntime = cfs_rq->min_vruntime;
604 	rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
605 	raw_spin_rq_unlock_irqrestore(rq, flags);
606 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "MIN_vruntime",
607 			SPLIT_NS(MIN_vruntime));
608 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "min_vruntime",
609 			SPLIT_NS(min_vruntime));
610 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "max_vruntime",
611 			SPLIT_NS(max_vruntime));
612 	spread = max_vruntime - MIN_vruntime;
613 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread",
614 			SPLIT_NS(spread));
615 	spread0 = min_vruntime - rq0_min_vruntime;
616 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread0",
617 			SPLIT_NS(spread0));
618 	SEQ_printf(m, "  .%-30s: %d\n", "nr_spread_over",
619 			cfs_rq->nr_spread_over);
620 	SEQ_printf(m, "  .%-30s: %d\n", "nr_running", cfs_rq->nr_running);
621 	SEQ_printf(m, "  .%-30s: %d\n", "h_nr_running", cfs_rq->h_nr_running);
622 	SEQ_printf(m, "  .%-30s: %d\n", "idle_nr_running",
623 			cfs_rq->idle_nr_running);
624 	SEQ_printf(m, "  .%-30s: %d\n", "idle_h_nr_running",
625 			cfs_rq->idle_h_nr_running);
626 	SEQ_printf(m, "  .%-30s: %ld\n", "load", cfs_rq->load.weight);
627 #ifdef CONFIG_SMP
628 	SEQ_printf(m, "  .%-30s: %lu\n", "load_avg",
629 			cfs_rq->avg.load_avg);
630 	SEQ_printf(m, "  .%-30s: %lu\n", "runnable_avg",
631 			cfs_rq->avg.runnable_avg);
632 	SEQ_printf(m, "  .%-30s: %lu\n", "util_avg",
633 			cfs_rq->avg.util_avg);
634 	SEQ_printf(m, "  .%-30s: %u\n", "util_est_enqueued",
635 			cfs_rq->avg.util_est.enqueued);
636 	SEQ_printf(m, "  .%-30s: %ld\n", "removed.load_avg",
637 			cfs_rq->removed.load_avg);
638 	SEQ_printf(m, "  .%-30s: %ld\n", "removed.util_avg",
639 			cfs_rq->removed.util_avg);
640 	SEQ_printf(m, "  .%-30s: %ld\n", "removed.runnable_avg",
641 			cfs_rq->removed.runnable_avg);
642 #ifdef CONFIG_FAIR_GROUP_SCHED
643 	SEQ_printf(m, "  .%-30s: %lu\n", "tg_load_avg_contrib",
644 			cfs_rq->tg_load_avg_contrib);
645 	SEQ_printf(m, "  .%-30s: %ld\n", "tg_load_avg",
646 			atomic_long_read(&cfs_rq->tg->load_avg));
647 #endif
648 #endif
649 #ifdef CONFIG_CFS_BANDWIDTH
650 	SEQ_printf(m, "  .%-30s: %d\n", "throttled",
651 			cfs_rq->throttled);
652 	SEQ_printf(m, "  .%-30s: %d\n", "throttle_count",
653 			cfs_rq->throttle_count);
654 #endif
655 
656 #ifdef CONFIG_FAIR_GROUP_SCHED
657 	print_cfs_group_stats(m, cpu, cfs_rq->tg);
658 #endif
659 }
660 
661 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
662 {
663 #ifdef CONFIG_RT_GROUP_SCHED
664 	SEQ_printf(m, "\n");
665 	SEQ_printf_task_group_path(m, rt_rq->tg, "rt_rq[%d]:%s\n", cpu);
666 #else
667 	SEQ_printf(m, "\n");
668 	SEQ_printf(m, "rt_rq[%d]:\n", cpu);
669 #endif
670 
671 #define P(x) \
672 	SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
673 #define PU(x) \
674 	SEQ_printf(m, "  .%-30s: %lu\n", #x, (unsigned long)(rt_rq->x))
675 #define PN(x) \
676 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
677 
678 	PU(rt_nr_running);
679 #ifdef CONFIG_SMP
680 	PU(rt_nr_migratory);
681 #endif
682 	P(rt_throttled);
683 	PN(rt_time);
684 	PN(rt_runtime);
685 
686 #undef PN
687 #undef PU
688 #undef P
689 }
690 
691 void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq)
692 {
693 	struct dl_bw *dl_bw;
694 
695 	SEQ_printf(m, "\n");
696 	SEQ_printf(m, "dl_rq[%d]:\n", cpu);
697 
698 #define PU(x) \
699 	SEQ_printf(m, "  .%-30s: %lu\n", #x, (unsigned long)(dl_rq->x))
700 
701 	PU(dl_nr_running);
702 #ifdef CONFIG_SMP
703 	PU(dl_nr_migratory);
704 	dl_bw = &cpu_rq(cpu)->rd->dl_bw;
705 #else
706 	dl_bw = &dl_rq->dl_bw;
707 #endif
708 	SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->bw", dl_bw->bw);
709 	SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->total_bw", dl_bw->total_bw);
710 
711 #undef PU
712 }
713 
714 static void print_cpu(struct seq_file *m, int cpu)
715 {
716 	struct rq *rq = cpu_rq(cpu);
717 
718 #ifdef CONFIG_X86
719 	{
720 		unsigned int freq = cpu_khz ? : 1;
721 
722 		SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
723 			   cpu, freq / 1000, (freq % 1000));
724 	}
725 #else
726 	SEQ_printf(m, "cpu#%d\n", cpu);
727 #endif
728 
729 #define P(x)								\
730 do {									\
731 	if (sizeof(rq->x) == 4)						\
732 		SEQ_printf(m, "  .%-30s: %ld\n", #x, (long)(rq->x));	\
733 	else								\
734 		SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rq->x));\
735 } while (0)
736 
737 #define PN(x) \
738 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
739 
740 	P(nr_running);
741 	P(nr_switches);
742 	P(nr_uninterruptible);
743 	PN(next_balance);
744 	SEQ_printf(m, "  .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr)));
745 	PN(clock);
746 	PN(clock_task);
747 #undef P
748 #undef PN
749 
750 #ifdef CONFIG_SMP
751 #define P64(n) SEQ_printf(m, "  .%-30s: %Ld\n", #n, rq->n);
752 	P64(avg_idle);
753 	P64(max_idle_balance_cost);
754 #undef P64
755 #endif
756 
757 #define P(n) SEQ_printf(m, "  .%-30s: %d\n", #n, schedstat_val(rq->n));
758 	if (schedstat_enabled()) {
759 		P(yld_count);
760 		P(sched_count);
761 		P(sched_goidle);
762 		P(ttwu_count);
763 		P(ttwu_local);
764 	}
765 #undef P
766 
767 	print_cfs_stats(m, cpu);
768 	print_rt_stats(m, cpu);
769 	print_dl_stats(m, cpu);
770 
771 	print_rq(m, rq, cpu);
772 	SEQ_printf(m, "\n");
773 }
774 
775 static const char *sched_tunable_scaling_names[] = {
776 	"none",
777 	"logarithmic",
778 	"linear"
779 };
780 
781 static void sched_debug_header(struct seq_file *m)
782 {
783 	u64 ktime, sched_clk, cpu_clk;
784 	unsigned long flags;
785 
786 	local_irq_save(flags);
787 	ktime = ktime_to_ns(ktime_get());
788 	sched_clk = sched_clock();
789 	cpu_clk = local_clock();
790 	local_irq_restore(flags);
791 
792 	SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n",
793 		init_utsname()->release,
794 		(int)strcspn(init_utsname()->version, " "),
795 		init_utsname()->version);
796 
797 #define P(x) \
798 	SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
799 #define PN(x) \
800 	SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
801 	PN(ktime);
802 	PN(sched_clk);
803 	PN(cpu_clk);
804 	P(jiffies);
805 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
806 	P(sched_clock_stable());
807 #endif
808 #undef PN
809 #undef P
810 
811 	SEQ_printf(m, "\n");
812 	SEQ_printf(m, "sysctl_sched\n");
813 
814 #define P(x) \
815 	SEQ_printf(m, "  .%-40s: %Ld\n", #x, (long long)(x))
816 #define PN(x) \
817 	SEQ_printf(m, "  .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
818 	PN(sysctl_sched_latency);
819 	PN(sysctl_sched_min_granularity);
820 	PN(sysctl_sched_idle_min_granularity);
821 	PN(sysctl_sched_wakeup_granularity);
822 	P(sysctl_sched_child_runs_first);
823 	P(sysctl_sched_features);
824 #undef PN
825 #undef P
826 
827 	SEQ_printf(m, "  .%-40s: %d (%s)\n",
828 		"sysctl_sched_tunable_scaling",
829 		sysctl_sched_tunable_scaling,
830 		sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
831 	SEQ_printf(m, "\n");
832 }
833 
834 static int sched_debug_show(struct seq_file *m, void *v)
835 {
836 	int cpu = (unsigned long)(v - 2);
837 
838 	if (cpu != -1)
839 		print_cpu(m, cpu);
840 	else
841 		sched_debug_header(m);
842 
843 	return 0;
844 }
845 
846 void sysrq_sched_debug_show(void)
847 {
848 	int cpu;
849 
850 	sched_debug_header(NULL);
851 	for_each_online_cpu(cpu) {
852 		/*
853 		 * Need to reset softlockup watchdogs on all CPUs, because
854 		 * another CPU might be blocked waiting for us to process
855 		 * an IPI or stop_machine.
856 		 */
857 		touch_nmi_watchdog();
858 		touch_all_softlockup_watchdogs();
859 		print_cpu(NULL, cpu);
860 	}
861 }
862 
863 /*
864  * This iterator needs some explanation.
865  * It returns 1 for the header position.
866  * This means 2 is CPU 0.
867  * In a hotplugged system some CPUs, including CPU 0, may be missing so we have
868  * to use cpumask_* to iterate over the CPUs.
869  */
870 static void *sched_debug_start(struct seq_file *file, loff_t *offset)
871 {
872 	unsigned long n = *offset;
873 
874 	if (n == 0)
875 		return (void *) 1;
876 
877 	n--;
878 
879 	if (n > 0)
880 		n = cpumask_next(n - 1, cpu_online_mask);
881 	else
882 		n = cpumask_first(cpu_online_mask);
883 
884 	*offset = n + 1;
885 
886 	if (n < nr_cpu_ids)
887 		return (void *)(unsigned long)(n + 2);
888 
889 	return NULL;
890 }
891 
892 static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
893 {
894 	(*offset)++;
895 	return sched_debug_start(file, offset);
896 }
897 
898 static void sched_debug_stop(struct seq_file *file, void *data)
899 {
900 }
901 
902 static const struct seq_operations sched_debug_sops = {
903 	.start		= sched_debug_start,
904 	.next		= sched_debug_next,
905 	.stop		= sched_debug_stop,
906 	.show		= sched_debug_show,
907 };
908 
909 #define __PS(S, F) SEQ_printf(m, "%-45s:%21Ld\n", S, (long long)(F))
910 #define __P(F) __PS(#F, F)
911 #define   P(F) __PS(#F, p->F)
912 #define   PM(F, M) __PS(#F, p->F & (M))
913 #define __PSN(S, F) SEQ_printf(m, "%-45s:%14Ld.%06ld\n", S, SPLIT_NS((long long)(F)))
914 #define __PN(F) __PSN(#F, F)
915 #define   PN(F) __PSN(#F, p->F)
916 
917 
918 #ifdef CONFIG_NUMA_BALANCING
919 void print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
920 		unsigned long tpf, unsigned long gsf, unsigned long gpf)
921 {
922 	SEQ_printf(m, "numa_faults node=%d ", node);
923 	SEQ_printf(m, "task_private=%lu task_shared=%lu ", tpf, tsf);
924 	SEQ_printf(m, "group_private=%lu group_shared=%lu\n", gpf, gsf);
925 }
926 #endif
927 
928 
929 static void sched_show_numa(struct task_struct *p, struct seq_file *m)
930 {
931 #ifdef CONFIG_NUMA_BALANCING
932 	struct mempolicy *pol;
933 
934 	if (p->mm)
935 		P(mm->numa_scan_seq);
936 
937 	task_lock(p);
938 	pol = p->mempolicy;
939 	if (pol && !(pol->flags & MPOL_F_MORON))
940 		pol = NULL;
941 	mpol_get(pol);
942 	task_unlock(p);
943 
944 	P(numa_pages_migrated);
945 	P(numa_preferred_nid);
946 	P(total_numa_faults);
947 	SEQ_printf(m, "current_node=%d, numa_group_id=%d\n",
948 			task_node(p), task_numa_group_id(p));
949 	show_numa_stats(p, m);
950 	mpol_put(pol);
951 #endif
952 }
953 
954 void proc_sched_show_task(struct task_struct *p, struct pid_namespace *ns,
955 						  struct seq_file *m)
956 {
957 	unsigned long nr_switches;
958 
959 	SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr_ns(p, ns),
960 						get_nr_threads(p));
961 	SEQ_printf(m,
962 		"---------------------------------------------------------"
963 		"----------\n");
964 
965 #define P_SCHEDSTAT(F)  __PS(#F, schedstat_val(p->stats.F))
966 #define PN_SCHEDSTAT(F) __PSN(#F, schedstat_val(p->stats.F))
967 
968 	PN(se.exec_start);
969 	PN(se.vruntime);
970 	PN(se.sum_exec_runtime);
971 
972 	nr_switches = p->nvcsw + p->nivcsw;
973 
974 	P(se.nr_migrations);
975 
976 	if (schedstat_enabled()) {
977 		u64 avg_atom, avg_per_cpu;
978 
979 		PN_SCHEDSTAT(sum_sleep_runtime);
980 		PN_SCHEDSTAT(wait_start);
981 		PN_SCHEDSTAT(sleep_start);
982 		PN_SCHEDSTAT(block_start);
983 		PN_SCHEDSTAT(sleep_max);
984 		PN_SCHEDSTAT(block_max);
985 		PN_SCHEDSTAT(exec_max);
986 		PN_SCHEDSTAT(slice_max);
987 		PN_SCHEDSTAT(wait_max);
988 		PN_SCHEDSTAT(wait_sum);
989 		P_SCHEDSTAT(wait_count);
990 		PN_SCHEDSTAT(iowait_sum);
991 		P_SCHEDSTAT(iowait_count);
992 		P_SCHEDSTAT(nr_migrations_cold);
993 		P_SCHEDSTAT(nr_failed_migrations_affine);
994 		P_SCHEDSTAT(nr_failed_migrations_running);
995 		P_SCHEDSTAT(nr_failed_migrations_hot);
996 		P_SCHEDSTAT(nr_forced_migrations);
997 		P_SCHEDSTAT(nr_wakeups);
998 		P_SCHEDSTAT(nr_wakeups_sync);
999 		P_SCHEDSTAT(nr_wakeups_migrate);
1000 		P_SCHEDSTAT(nr_wakeups_local);
1001 		P_SCHEDSTAT(nr_wakeups_remote);
1002 		P_SCHEDSTAT(nr_wakeups_affine);
1003 		P_SCHEDSTAT(nr_wakeups_affine_attempts);
1004 		P_SCHEDSTAT(nr_wakeups_passive);
1005 		P_SCHEDSTAT(nr_wakeups_idle);
1006 
1007 		avg_atom = p->se.sum_exec_runtime;
1008 		if (nr_switches)
1009 			avg_atom = div64_ul(avg_atom, nr_switches);
1010 		else
1011 			avg_atom = -1LL;
1012 
1013 		avg_per_cpu = p->se.sum_exec_runtime;
1014 		if (p->se.nr_migrations) {
1015 			avg_per_cpu = div64_u64(avg_per_cpu,
1016 						p->se.nr_migrations);
1017 		} else {
1018 			avg_per_cpu = -1LL;
1019 		}
1020 
1021 		__PN(avg_atom);
1022 		__PN(avg_per_cpu);
1023 	}
1024 
1025 	__P(nr_switches);
1026 	__PS("nr_voluntary_switches", p->nvcsw);
1027 	__PS("nr_involuntary_switches", p->nivcsw);
1028 
1029 	P(se.load.weight);
1030 #ifdef CONFIG_SMP
1031 	P(se.avg.load_sum);
1032 	P(se.avg.runnable_sum);
1033 	P(se.avg.util_sum);
1034 	P(se.avg.load_avg);
1035 	P(se.avg.runnable_avg);
1036 	P(se.avg.util_avg);
1037 	P(se.avg.last_update_time);
1038 	P(se.avg.util_est.ewma);
1039 	PM(se.avg.util_est.enqueued, ~UTIL_AVG_UNCHANGED);
1040 #endif
1041 #ifdef CONFIG_UCLAMP_TASK
1042 	__PS("uclamp.min", p->uclamp_req[UCLAMP_MIN].value);
1043 	__PS("uclamp.max", p->uclamp_req[UCLAMP_MAX].value);
1044 	__PS("effective uclamp.min", uclamp_eff_value(p, UCLAMP_MIN));
1045 	__PS("effective uclamp.max", uclamp_eff_value(p, UCLAMP_MAX));
1046 #endif
1047 	P(policy);
1048 	P(prio);
1049 	if (task_has_dl_policy(p)) {
1050 		P(dl.runtime);
1051 		P(dl.deadline);
1052 	}
1053 #undef PN_SCHEDSTAT
1054 #undef P_SCHEDSTAT
1055 
1056 	{
1057 		unsigned int this_cpu = raw_smp_processor_id();
1058 		u64 t0, t1;
1059 
1060 		t0 = cpu_clock(this_cpu);
1061 		t1 = cpu_clock(this_cpu);
1062 		__PS("clock-delta", t1-t0);
1063 	}
1064 
1065 	sched_show_numa(p, m);
1066 }
1067 
1068 void proc_sched_set_task(struct task_struct *p)
1069 {
1070 #ifdef CONFIG_SCHEDSTATS
1071 	memset(&p->stats, 0, sizeof(p->stats));
1072 #endif
1073 }
1074 
1075 void resched_latency_warn(int cpu, u64 latency)
1076 {
1077 	static DEFINE_RATELIMIT_STATE(latency_check_ratelimit, 60 * 60 * HZ, 1);
1078 
1079 	WARN(__ratelimit(&latency_check_ratelimit),
1080 	     "sched: CPU %d need_resched set for > %llu ns (%d ticks) "
1081 	     "without schedule\n",
1082 	     cpu, latency, cpu_rq(cpu)->ticks_without_resched);
1083 }
1084