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 static const char * const sched_feat_names[] = {
51 #include "features.h"
52 };
53 
54 #undef SCHED_FEAT
55 
sched_feat_show(struct seq_file * m,void * v)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 
sched_feat_disable(int i)84 static void sched_feat_disable(int i)
85 {
86 	static_key_disable_cpuslocked(&sched_feat_keys[i]);
87 }
88 
sched_feat_enable(int i)89 static void sched_feat_enable(int i)
90 {
91 	static_key_enable_cpuslocked(&sched_feat_keys[i]);
92 }
93 #else
sched_feat_disable(int i)94 static void sched_feat_disable(int i) { };
sched_feat_enable(int i)95 static void sched_feat_enable(int i) { };
96 #endif /* CONFIG_JUMP_LABEL */
97 
sched_feat_set(char * cmp)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
sched_feat_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)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 
sched_feat_open(struct inode * inode,struct file * filp)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 
sched_scaling_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)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 
sched_scaling_show(struct seq_file * m,void * v)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 
sched_scaling_open(struct inode * inode,struct file * filp)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 
sched_dynamic_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)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 
sched_dynamic_show(struct seq_file * m,void * v)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 
sched_dynamic_open(struct inode * inode,struct file * filp)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 #ifdef CONFIG_SMP
284 static struct dentry           *sd_dentry;
285 
286 
sched_verbose_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)287 static ssize_t sched_verbose_write(struct file *filp, const char __user *ubuf,
288 				  size_t cnt, loff_t *ppos)
289 {
290 	ssize_t result;
291 	bool orig;
292 
293 	cpus_read_lock();
294 	mutex_lock(&sched_domains_mutex);
295 
296 	orig = sched_debug_verbose;
297 	result = debugfs_write_file_bool(filp, ubuf, cnt, ppos);
298 
299 	if (sched_debug_verbose && !orig)
300 		update_sched_domain_debugfs();
301 	else if (!sched_debug_verbose && orig) {
302 		debugfs_remove(sd_dentry);
303 		sd_dentry = NULL;
304 	}
305 
306 	mutex_unlock(&sched_domains_mutex);
307 	cpus_read_unlock();
308 
309 	return result;
310 }
311 #else
312 #define sched_verbose_write debugfs_write_file_bool
313 #endif
314 
315 static const struct file_operations sched_verbose_fops = {
316 	.read =         debugfs_read_file_bool,
317 	.write =        sched_verbose_write,
318 	.open =         simple_open,
319 	.llseek =       default_llseek,
320 };
321 
322 static const struct seq_operations sched_debug_sops;
323 
sched_debug_open(struct inode * inode,struct file * filp)324 static int sched_debug_open(struct inode *inode, struct file *filp)
325 {
326 	return seq_open(filp, &sched_debug_sops);
327 }
328 
329 static const struct file_operations sched_debug_fops = {
330 	.open		= sched_debug_open,
331 	.read		= seq_read,
332 	.llseek		= seq_lseek,
333 	.release	= seq_release,
334 };
335 
336 static struct dentry *debugfs_sched;
337 
sched_init_debug(void)338 static __init int sched_init_debug(void)
339 {
340 	struct dentry __maybe_unused *numa;
341 
342 	debugfs_sched = debugfs_create_dir("sched", NULL);
343 
344 	debugfs_create_file("features", 0644, debugfs_sched, NULL, &sched_feat_fops);
345 	debugfs_create_file_unsafe("verbose", 0644, debugfs_sched, &sched_debug_verbose, &sched_verbose_fops);
346 #ifdef CONFIG_PREEMPT_DYNAMIC
347 	debugfs_create_file("preempt", 0644, debugfs_sched, NULL, &sched_dynamic_fops);
348 #endif
349 
350 	debugfs_create_u32("base_slice_ns", 0644, debugfs_sched, &sysctl_sched_base_slice);
351 
352 	debugfs_create_u32("latency_warn_ms", 0644, debugfs_sched, &sysctl_resched_latency_warn_ms);
353 	debugfs_create_u32("latency_warn_once", 0644, debugfs_sched, &sysctl_resched_latency_warn_once);
354 
355 #ifdef CONFIG_SMP
356 	debugfs_create_file("tunable_scaling", 0644, debugfs_sched, NULL, &sched_scaling_fops);
357 	debugfs_create_u32("migration_cost_ns", 0644, debugfs_sched, &sysctl_sched_migration_cost);
358 	debugfs_create_u32("nr_migrate", 0644, debugfs_sched, &sysctl_sched_nr_migrate);
359 
360 	mutex_lock(&sched_domains_mutex);
361 	update_sched_domain_debugfs();
362 	mutex_unlock(&sched_domains_mutex);
363 #endif
364 
365 #ifdef CONFIG_NUMA_BALANCING
366 	numa = debugfs_create_dir("numa_balancing", debugfs_sched);
367 
368 	debugfs_create_u32("scan_delay_ms", 0644, numa, &sysctl_numa_balancing_scan_delay);
369 	debugfs_create_u32("scan_period_min_ms", 0644, numa, &sysctl_numa_balancing_scan_period_min);
370 	debugfs_create_u32("scan_period_max_ms", 0644, numa, &sysctl_numa_balancing_scan_period_max);
371 	debugfs_create_u32("scan_size_mb", 0644, numa, &sysctl_numa_balancing_scan_size);
372 	debugfs_create_u32("hot_threshold_ms", 0644, numa, &sysctl_numa_balancing_hot_threshold);
373 #endif
374 
375 	debugfs_create_file("debug", 0444, debugfs_sched, NULL, &sched_debug_fops);
376 
377 	return 0;
378 }
379 late_initcall(sched_init_debug);
380 
381 #ifdef CONFIG_SMP
382 
383 static cpumask_var_t		sd_sysctl_cpus;
384 
sd_flags_show(struct seq_file * m,void * v)385 static int sd_flags_show(struct seq_file *m, void *v)
386 {
387 	unsigned long flags = *(unsigned int *)m->private;
388 	int idx;
389 
390 	for_each_set_bit(idx, &flags, __SD_FLAG_CNT) {
391 		seq_puts(m, sd_flag_debug[idx].name);
392 		seq_puts(m, " ");
393 	}
394 	seq_puts(m, "\n");
395 
396 	return 0;
397 }
398 
sd_flags_open(struct inode * inode,struct file * file)399 static int sd_flags_open(struct inode *inode, struct file *file)
400 {
401 	return single_open(file, sd_flags_show, inode->i_private);
402 }
403 
404 static const struct file_operations sd_flags_fops = {
405 	.open		= sd_flags_open,
406 	.read		= seq_read,
407 	.llseek		= seq_lseek,
408 	.release	= single_release,
409 };
410 
register_sd(struct sched_domain * sd,struct dentry * parent)411 static void register_sd(struct sched_domain *sd, struct dentry *parent)
412 {
413 #define SDM(type, mode, member)	\
414 	debugfs_create_##type(#member, mode, parent, &sd->member)
415 
416 	SDM(ulong, 0644, min_interval);
417 	SDM(ulong, 0644, max_interval);
418 	SDM(u64,   0644, max_newidle_lb_cost);
419 	SDM(u32,   0644, busy_factor);
420 	SDM(u32,   0644, imbalance_pct);
421 	SDM(u32,   0644, cache_nice_tries);
422 	SDM(str,   0444, name);
423 
424 #undef SDM
425 
426 	debugfs_create_file("flags", 0444, parent, &sd->flags, &sd_flags_fops);
427 	debugfs_create_file("groups_flags", 0444, parent, &sd->groups->flags, &sd_flags_fops);
428 }
429 
update_sched_domain_debugfs(void)430 void update_sched_domain_debugfs(void)
431 {
432 	int cpu, i;
433 
434 	/*
435 	 * This can unfortunately be invoked before sched_debug_init() creates
436 	 * the debug directory. Don't touch sd_sysctl_cpus until then.
437 	 */
438 	if (!debugfs_sched)
439 		return;
440 
441 	if (!sched_debug_verbose)
442 		return;
443 
444 	if (!cpumask_available(sd_sysctl_cpus)) {
445 		if (!alloc_cpumask_var(&sd_sysctl_cpus, GFP_KERNEL))
446 			return;
447 		cpumask_copy(sd_sysctl_cpus, cpu_possible_mask);
448 	}
449 
450 	if (!sd_dentry) {
451 		sd_dentry = debugfs_create_dir("domains", debugfs_sched);
452 
453 		/* rebuild sd_sysctl_cpus if empty since it gets cleared below */
454 		if (cpumask_empty(sd_sysctl_cpus))
455 			cpumask_copy(sd_sysctl_cpus, cpu_online_mask);
456 	}
457 
458 	for_each_cpu(cpu, sd_sysctl_cpus) {
459 		struct sched_domain *sd;
460 		struct dentry *d_cpu;
461 		char buf[32];
462 
463 		snprintf(buf, sizeof(buf), "cpu%d", cpu);
464 		debugfs_lookup_and_remove(buf, sd_dentry);
465 		d_cpu = debugfs_create_dir(buf, sd_dentry);
466 
467 		i = 0;
468 		for_each_domain(cpu, sd) {
469 			struct dentry *d_sd;
470 
471 			snprintf(buf, sizeof(buf), "domain%d", i);
472 			d_sd = debugfs_create_dir(buf, d_cpu);
473 
474 			register_sd(sd, d_sd);
475 			i++;
476 		}
477 
478 		__cpumask_clear_cpu(cpu, sd_sysctl_cpus);
479 	}
480 }
481 
dirty_sched_domain_sysctl(int cpu)482 void dirty_sched_domain_sysctl(int cpu)
483 {
484 	if (cpumask_available(sd_sysctl_cpus))
485 		__cpumask_set_cpu(cpu, sd_sysctl_cpus);
486 }
487 
488 #endif /* CONFIG_SMP */
489 
490 #ifdef CONFIG_FAIR_GROUP_SCHED
print_cfs_group_stats(struct seq_file * m,int cpu,struct task_group * tg)491 static void print_cfs_group_stats(struct seq_file *m, int cpu, struct task_group *tg)
492 {
493 	struct sched_entity *se = tg->se[cpu];
494 
495 #define P(F)		SEQ_printf(m, "  .%-30s: %lld\n",	#F, (long long)F)
496 #define P_SCHEDSTAT(F)	SEQ_printf(m, "  .%-30s: %lld\n",	\
497 		#F, (long long)schedstat_val(stats->F))
498 #define PN(F)		SEQ_printf(m, "  .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F))
499 #define PN_SCHEDSTAT(F)	SEQ_printf(m, "  .%-30s: %lld.%06ld\n", \
500 		#F, SPLIT_NS((long long)schedstat_val(stats->F)))
501 
502 	if (!se)
503 		return;
504 
505 	PN(se->exec_start);
506 	PN(se->vruntime);
507 	PN(se->sum_exec_runtime);
508 
509 	if (schedstat_enabled()) {
510 		struct sched_statistics *stats;
511 		stats = __schedstats_from_se(se);
512 
513 		PN_SCHEDSTAT(wait_start);
514 		PN_SCHEDSTAT(sleep_start);
515 		PN_SCHEDSTAT(block_start);
516 		PN_SCHEDSTAT(sleep_max);
517 		PN_SCHEDSTAT(block_max);
518 		PN_SCHEDSTAT(exec_max);
519 		PN_SCHEDSTAT(slice_max);
520 		PN_SCHEDSTAT(wait_max);
521 		PN_SCHEDSTAT(wait_sum);
522 		P_SCHEDSTAT(wait_count);
523 	}
524 
525 	P(se->load.weight);
526 #ifdef CONFIG_SMP
527 	P(se->avg.load_avg);
528 	P(se->avg.util_avg);
529 	P(se->avg.runnable_avg);
530 #endif
531 
532 #undef PN_SCHEDSTAT
533 #undef PN
534 #undef P_SCHEDSTAT
535 #undef P
536 }
537 #endif
538 
539 #ifdef CONFIG_CGROUP_SCHED
540 static DEFINE_SPINLOCK(sched_debug_lock);
541 static char group_path[PATH_MAX];
542 
task_group_path(struct task_group * tg,char * path,int plen)543 static void task_group_path(struct task_group *tg, char *path, int plen)
544 {
545 	if (autogroup_path(tg, path, plen))
546 		return;
547 
548 	cgroup_path(tg->css.cgroup, path, plen);
549 }
550 
551 /*
552  * Only 1 SEQ_printf_task_group_path() caller can use the full length
553  * group_path[] for cgroup path. Other simultaneous callers will have
554  * to use a shorter stack buffer. A "..." suffix is appended at the end
555  * of the stack buffer so that it will show up in case the output length
556  * matches the given buffer size to indicate possible path name truncation.
557  */
558 #define SEQ_printf_task_group_path(m, tg, fmt...)			\
559 {									\
560 	if (spin_trylock(&sched_debug_lock)) {				\
561 		task_group_path(tg, group_path, sizeof(group_path));	\
562 		SEQ_printf(m, fmt, group_path);				\
563 		spin_unlock(&sched_debug_lock);				\
564 	} else {							\
565 		char buf[128];						\
566 		char *bufend = buf + sizeof(buf) - 3;			\
567 		task_group_path(tg, buf, bufend - buf);			\
568 		strcpy(bufend - 1, "...");				\
569 		SEQ_printf(m, fmt, buf);				\
570 	}								\
571 }
572 #endif
573 
574 static void
print_task(struct seq_file * m,struct rq * rq,struct task_struct * p)575 print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
576 {
577 	if (task_current(rq, p))
578 		SEQ_printf(m, ">R");
579 	else
580 		SEQ_printf(m, " %c", task_state_to_char(p));
581 
582 	SEQ_printf(m, "%15s %5d %9Ld.%06ld %c %9Ld.%06ld %9Ld.%06ld %9Ld.%06ld %9Ld %5d ",
583 		p->comm, task_pid_nr(p),
584 		SPLIT_NS(p->se.vruntime),
585 		entity_eligible(cfs_rq_of(&p->se), &p->se) ? 'E' : 'N',
586 		SPLIT_NS(p->se.deadline),
587 		SPLIT_NS(p->se.slice),
588 		SPLIT_NS(p->se.sum_exec_runtime),
589 		(long long)(p->nvcsw + p->nivcsw),
590 		p->prio);
591 
592 	SEQ_printf(m, "%9lld.%06ld %9lld.%06ld %9lld.%06ld %9lld.%06ld",
593 		SPLIT_NS(schedstat_val_or_zero(p->stats.wait_sum)),
594 		SPLIT_NS(p->se.sum_exec_runtime),
595 		SPLIT_NS(schedstat_val_or_zero(p->stats.sum_sleep_runtime)),
596 		SPLIT_NS(schedstat_val_or_zero(p->stats.sum_block_runtime)));
597 
598 #ifdef CONFIG_NUMA_BALANCING
599 	SEQ_printf(m, " %d %d", task_node(p), task_numa_group_id(p));
600 #endif
601 #ifdef CONFIG_CGROUP_SCHED
602 	SEQ_printf_task_group_path(m, task_group(p), " %s")
603 #endif
604 
605 	SEQ_printf(m, "\n");
606 }
607 
print_rq(struct seq_file * m,struct rq * rq,int rq_cpu)608 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
609 {
610 	struct task_struct *g, *p;
611 
612 	SEQ_printf(m, "\n");
613 	SEQ_printf(m, "runnable tasks:\n");
614 	SEQ_printf(m, " S            task   PID         tree-key  switches  prio"
615 		   "     wait-time             sum-exec        sum-sleep\n");
616 	SEQ_printf(m, "-------------------------------------------------------"
617 		   "------------------------------------------------------\n");
618 
619 	rcu_read_lock();
620 	for_each_process_thread(g, p) {
621 		if (task_cpu(p) != rq_cpu)
622 			continue;
623 
624 		print_task(m, rq, p);
625 	}
626 	rcu_read_unlock();
627 }
628 
print_cfs_rq(struct seq_file * m,int cpu,struct cfs_rq * cfs_rq)629 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
630 {
631 	s64 left_vruntime = -1, min_vruntime, right_vruntime = -1, spread;
632 	struct sched_entity *last, *first;
633 	struct rq *rq = cpu_rq(cpu);
634 	unsigned long flags;
635 
636 #ifdef CONFIG_FAIR_GROUP_SCHED
637 	SEQ_printf(m, "\n");
638 	SEQ_printf_task_group_path(m, cfs_rq->tg, "cfs_rq[%d]:%s\n", cpu);
639 #else
640 	SEQ_printf(m, "\n");
641 	SEQ_printf(m, "cfs_rq[%d]:\n", cpu);
642 #endif
643 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "exec_clock",
644 			SPLIT_NS(cfs_rq->exec_clock));
645 
646 	raw_spin_rq_lock_irqsave(rq, flags);
647 	first = __pick_first_entity(cfs_rq);
648 	if (first)
649 		left_vruntime = first->vruntime;
650 	last = __pick_last_entity(cfs_rq);
651 	if (last)
652 		right_vruntime = last->vruntime;
653 	min_vruntime = cfs_rq->min_vruntime;
654 	raw_spin_rq_unlock_irqrestore(rq, flags);
655 
656 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "left_vruntime",
657 			SPLIT_NS(left_vruntime));
658 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "min_vruntime",
659 			SPLIT_NS(min_vruntime));
660 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "avg_vruntime",
661 			SPLIT_NS(avg_vruntime(cfs_rq)));
662 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "right_vruntime",
663 			SPLIT_NS(right_vruntime));
664 	spread = right_vruntime - left_vruntime;
665 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread", SPLIT_NS(spread));
666 	SEQ_printf(m, "  .%-30s: %d\n", "nr_spread_over",
667 			cfs_rq->nr_spread_over);
668 	SEQ_printf(m, "  .%-30s: %d\n", "nr_running", cfs_rq->nr_running);
669 	SEQ_printf(m, "  .%-30s: %d\n", "h_nr_running", cfs_rq->h_nr_running);
670 	SEQ_printf(m, "  .%-30s: %d\n", "idle_nr_running",
671 			cfs_rq->idle_nr_running);
672 	SEQ_printf(m, "  .%-30s: %d\n", "idle_h_nr_running",
673 			cfs_rq->idle_h_nr_running);
674 	SEQ_printf(m, "  .%-30s: %ld\n", "load", cfs_rq->load.weight);
675 #ifdef CONFIG_SMP
676 	SEQ_printf(m, "  .%-30s: %lu\n", "load_avg",
677 			cfs_rq->avg.load_avg);
678 	SEQ_printf(m, "  .%-30s: %lu\n", "runnable_avg",
679 			cfs_rq->avg.runnable_avg);
680 	SEQ_printf(m, "  .%-30s: %lu\n", "util_avg",
681 			cfs_rq->avg.util_avg);
682 	SEQ_printf(m, "  .%-30s: %u\n", "util_est_enqueued",
683 			cfs_rq->avg.util_est.enqueued);
684 	SEQ_printf(m, "  .%-30s: %ld\n", "removed.load_avg",
685 			cfs_rq->removed.load_avg);
686 	SEQ_printf(m, "  .%-30s: %ld\n", "removed.util_avg",
687 			cfs_rq->removed.util_avg);
688 	SEQ_printf(m, "  .%-30s: %ld\n", "removed.runnable_avg",
689 			cfs_rq->removed.runnable_avg);
690 #ifdef CONFIG_FAIR_GROUP_SCHED
691 	SEQ_printf(m, "  .%-30s: %lu\n", "tg_load_avg_contrib",
692 			cfs_rq->tg_load_avg_contrib);
693 	SEQ_printf(m, "  .%-30s: %ld\n", "tg_load_avg",
694 			atomic_long_read(&cfs_rq->tg->load_avg));
695 #endif
696 #endif
697 #ifdef CONFIG_CFS_BANDWIDTH
698 	SEQ_printf(m, "  .%-30s: %d\n", "throttled",
699 			cfs_rq->throttled);
700 	SEQ_printf(m, "  .%-30s: %d\n", "throttle_count",
701 			cfs_rq->throttle_count);
702 #endif
703 
704 #ifdef CONFIG_FAIR_GROUP_SCHED
705 	print_cfs_group_stats(m, cpu, cfs_rq->tg);
706 #endif
707 }
708 
print_rt_rq(struct seq_file * m,int cpu,struct rt_rq * rt_rq)709 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
710 {
711 #ifdef CONFIG_RT_GROUP_SCHED
712 	SEQ_printf(m, "\n");
713 	SEQ_printf_task_group_path(m, rt_rq->tg, "rt_rq[%d]:%s\n", cpu);
714 #else
715 	SEQ_printf(m, "\n");
716 	SEQ_printf(m, "rt_rq[%d]:\n", cpu);
717 #endif
718 
719 #define P(x) \
720 	SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
721 #define PU(x) \
722 	SEQ_printf(m, "  .%-30s: %lu\n", #x, (unsigned long)(rt_rq->x))
723 #define PN(x) \
724 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
725 
726 	PU(rt_nr_running);
727 #ifdef CONFIG_SMP
728 	PU(rt_nr_migratory);
729 #endif
730 	P(rt_throttled);
731 	PN(rt_time);
732 	PN(rt_runtime);
733 
734 #undef PN
735 #undef PU
736 #undef P
737 }
738 
print_dl_rq(struct seq_file * m,int cpu,struct dl_rq * dl_rq)739 void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq)
740 {
741 	struct dl_bw *dl_bw;
742 
743 	SEQ_printf(m, "\n");
744 	SEQ_printf(m, "dl_rq[%d]:\n", cpu);
745 
746 #define PU(x) \
747 	SEQ_printf(m, "  .%-30s: %lu\n", #x, (unsigned long)(dl_rq->x))
748 
749 	PU(dl_nr_running);
750 #ifdef CONFIG_SMP
751 	PU(dl_nr_migratory);
752 	dl_bw = &cpu_rq(cpu)->rd->dl_bw;
753 #else
754 	dl_bw = &dl_rq->dl_bw;
755 #endif
756 	SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->bw", dl_bw->bw);
757 	SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->total_bw", dl_bw->total_bw);
758 
759 #undef PU
760 }
761 
print_cpu(struct seq_file * m,int cpu)762 static void print_cpu(struct seq_file *m, int cpu)
763 {
764 	struct rq *rq = cpu_rq(cpu);
765 
766 #ifdef CONFIG_X86
767 	{
768 		unsigned int freq = cpu_khz ? : 1;
769 
770 		SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
771 			   cpu, freq / 1000, (freq % 1000));
772 	}
773 #else
774 	SEQ_printf(m, "cpu#%d\n", cpu);
775 #endif
776 
777 #define P(x)								\
778 do {									\
779 	if (sizeof(rq->x) == 4)						\
780 		SEQ_printf(m, "  .%-30s: %d\n", #x, (int)(rq->x));	\
781 	else								\
782 		SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rq->x));\
783 } while (0)
784 
785 #define PN(x) \
786 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
787 
788 	P(nr_running);
789 	P(nr_switches);
790 	P(nr_uninterruptible);
791 	PN(next_balance);
792 	SEQ_printf(m, "  .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr)));
793 	PN(clock);
794 	PN(clock_task);
795 #undef P
796 #undef PN
797 
798 #ifdef CONFIG_SMP
799 #define P64(n) SEQ_printf(m, "  .%-30s: %Ld\n", #n, rq->n);
800 	P64(avg_idle);
801 	P64(max_idle_balance_cost);
802 #undef P64
803 #endif
804 
805 #define P(n) SEQ_printf(m, "  .%-30s: %d\n", #n, schedstat_val(rq->n));
806 	if (schedstat_enabled()) {
807 		P(yld_count);
808 		P(sched_count);
809 		P(sched_goidle);
810 		P(ttwu_count);
811 		P(ttwu_local);
812 	}
813 #undef P
814 
815 	print_cfs_stats(m, cpu);
816 	print_rt_stats(m, cpu);
817 	print_dl_stats(m, cpu);
818 
819 	print_rq(m, rq, cpu);
820 	SEQ_printf(m, "\n");
821 }
822 
823 static const char *sched_tunable_scaling_names[] = {
824 	"none",
825 	"logarithmic",
826 	"linear"
827 };
828 
sched_debug_header(struct seq_file * m)829 static void sched_debug_header(struct seq_file *m)
830 {
831 	u64 ktime, sched_clk, cpu_clk;
832 	unsigned long flags;
833 
834 	local_irq_save(flags);
835 	ktime = ktime_to_ns(ktime_get());
836 	sched_clk = sched_clock();
837 	cpu_clk = local_clock();
838 	local_irq_restore(flags);
839 
840 	SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n",
841 		init_utsname()->release,
842 		(int)strcspn(init_utsname()->version, " "),
843 		init_utsname()->version);
844 
845 #define P(x) \
846 	SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
847 #define PN(x) \
848 	SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
849 	PN(ktime);
850 	PN(sched_clk);
851 	PN(cpu_clk);
852 	P(jiffies);
853 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
854 	P(sched_clock_stable());
855 #endif
856 #undef PN
857 #undef P
858 
859 	SEQ_printf(m, "\n");
860 	SEQ_printf(m, "sysctl_sched\n");
861 
862 #define P(x) \
863 	SEQ_printf(m, "  .%-40s: %Ld\n", #x, (long long)(x))
864 #define PN(x) \
865 	SEQ_printf(m, "  .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
866 	PN(sysctl_sched_base_slice);
867 	P(sysctl_sched_child_runs_first);
868 	P(sysctl_sched_features);
869 #undef PN
870 #undef P
871 
872 	SEQ_printf(m, "  .%-40s: %d (%s)\n",
873 		"sysctl_sched_tunable_scaling",
874 		sysctl_sched_tunable_scaling,
875 		sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
876 	SEQ_printf(m, "\n");
877 }
878 
sched_debug_show(struct seq_file * m,void * v)879 static int sched_debug_show(struct seq_file *m, void *v)
880 {
881 	int cpu = (unsigned long)(v - 2);
882 
883 	if (cpu != -1)
884 		print_cpu(m, cpu);
885 	else
886 		sched_debug_header(m);
887 
888 	return 0;
889 }
890 
sysrq_sched_debug_show(void)891 void sysrq_sched_debug_show(void)
892 {
893 	int cpu;
894 
895 	sched_debug_header(NULL);
896 	for_each_online_cpu(cpu) {
897 		/*
898 		 * Need to reset softlockup watchdogs on all CPUs, because
899 		 * another CPU might be blocked waiting for us to process
900 		 * an IPI or stop_machine.
901 		 */
902 		touch_nmi_watchdog();
903 		touch_all_softlockup_watchdogs();
904 		print_cpu(NULL, cpu);
905 	}
906 }
907 
908 /*
909  * This iterator needs some explanation.
910  * It returns 1 for the header position.
911  * This means 2 is CPU 0.
912  * In a hotplugged system some CPUs, including CPU 0, may be missing so we have
913  * to use cpumask_* to iterate over the CPUs.
914  */
sched_debug_start(struct seq_file * file,loff_t * offset)915 static void *sched_debug_start(struct seq_file *file, loff_t *offset)
916 {
917 	unsigned long n = *offset;
918 
919 	if (n == 0)
920 		return (void *) 1;
921 
922 	n--;
923 
924 	if (n > 0)
925 		n = cpumask_next(n - 1, cpu_online_mask);
926 	else
927 		n = cpumask_first(cpu_online_mask);
928 
929 	*offset = n + 1;
930 
931 	if (n < nr_cpu_ids)
932 		return (void *)(unsigned long)(n + 2);
933 
934 	return NULL;
935 }
936 
sched_debug_next(struct seq_file * file,void * data,loff_t * offset)937 static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
938 {
939 	(*offset)++;
940 	return sched_debug_start(file, offset);
941 }
942 
sched_debug_stop(struct seq_file * file,void * data)943 static void sched_debug_stop(struct seq_file *file, void *data)
944 {
945 }
946 
947 static const struct seq_operations sched_debug_sops = {
948 	.start		= sched_debug_start,
949 	.next		= sched_debug_next,
950 	.stop		= sched_debug_stop,
951 	.show		= sched_debug_show,
952 };
953 
954 #define __PS(S, F) SEQ_printf(m, "%-45s:%21Ld\n", S, (long long)(F))
955 #define __P(F) __PS(#F, F)
956 #define   P(F) __PS(#F, p->F)
957 #define   PM(F, M) __PS(#F, p->F & (M))
958 #define __PSN(S, F) SEQ_printf(m, "%-45s:%14Ld.%06ld\n", S, SPLIT_NS((long long)(F)))
959 #define __PN(F) __PSN(#F, F)
960 #define   PN(F) __PSN(#F, p->F)
961 
962 
963 #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)964 void print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
965 		unsigned long tpf, unsigned long gsf, unsigned long gpf)
966 {
967 	SEQ_printf(m, "numa_faults node=%d ", node);
968 	SEQ_printf(m, "task_private=%lu task_shared=%lu ", tpf, tsf);
969 	SEQ_printf(m, "group_private=%lu group_shared=%lu\n", gpf, gsf);
970 }
971 #endif
972 
973 
sched_show_numa(struct task_struct * p,struct seq_file * m)974 static void sched_show_numa(struct task_struct *p, struct seq_file *m)
975 {
976 #ifdef CONFIG_NUMA_BALANCING
977 	if (p->mm)
978 		P(mm->numa_scan_seq);
979 
980 	P(numa_pages_migrated);
981 	P(numa_preferred_nid);
982 	P(total_numa_faults);
983 	SEQ_printf(m, "current_node=%d, numa_group_id=%d\n",
984 			task_node(p), task_numa_group_id(p));
985 	show_numa_stats(p, m);
986 #endif
987 }
988 
proc_sched_show_task(struct task_struct * p,struct pid_namespace * ns,struct seq_file * m)989 void proc_sched_show_task(struct task_struct *p, struct pid_namespace *ns,
990 						  struct seq_file *m)
991 {
992 	unsigned long nr_switches;
993 
994 	SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr_ns(p, ns),
995 						get_nr_threads(p));
996 	SEQ_printf(m,
997 		"---------------------------------------------------------"
998 		"----------\n");
999 
1000 #define P_SCHEDSTAT(F)  __PS(#F, schedstat_val(p->stats.F))
1001 #define PN_SCHEDSTAT(F) __PSN(#F, schedstat_val(p->stats.F))
1002 
1003 	PN(se.exec_start);
1004 	PN(se.vruntime);
1005 	PN(se.sum_exec_runtime);
1006 
1007 	nr_switches = p->nvcsw + p->nivcsw;
1008 
1009 	P(se.nr_migrations);
1010 
1011 	if (schedstat_enabled()) {
1012 		u64 avg_atom, avg_per_cpu;
1013 
1014 		PN_SCHEDSTAT(sum_sleep_runtime);
1015 		PN_SCHEDSTAT(sum_block_runtime);
1016 		PN_SCHEDSTAT(wait_start);
1017 		PN_SCHEDSTAT(sleep_start);
1018 		PN_SCHEDSTAT(block_start);
1019 		PN_SCHEDSTAT(sleep_max);
1020 		PN_SCHEDSTAT(block_max);
1021 		PN_SCHEDSTAT(exec_max);
1022 		PN_SCHEDSTAT(slice_max);
1023 		PN_SCHEDSTAT(wait_max);
1024 		PN_SCHEDSTAT(wait_sum);
1025 		P_SCHEDSTAT(wait_count);
1026 		PN_SCHEDSTAT(iowait_sum);
1027 		P_SCHEDSTAT(iowait_count);
1028 		P_SCHEDSTAT(nr_migrations_cold);
1029 		P_SCHEDSTAT(nr_failed_migrations_affine);
1030 		P_SCHEDSTAT(nr_failed_migrations_running);
1031 		P_SCHEDSTAT(nr_failed_migrations_hot);
1032 		P_SCHEDSTAT(nr_forced_migrations);
1033 		P_SCHEDSTAT(nr_wakeups);
1034 		P_SCHEDSTAT(nr_wakeups_sync);
1035 		P_SCHEDSTAT(nr_wakeups_migrate);
1036 		P_SCHEDSTAT(nr_wakeups_local);
1037 		P_SCHEDSTAT(nr_wakeups_remote);
1038 		P_SCHEDSTAT(nr_wakeups_affine);
1039 		P_SCHEDSTAT(nr_wakeups_affine_attempts);
1040 		P_SCHEDSTAT(nr_wakeups_passive);
1041 		P_SCHEDSTAT(nr_wakeups_idle);
1042 
1043 		avg_atom = p->se.sum_exec_runtime;
1044 		if (nr_switches)
1045 			avg_atom = div64_ul(avg_atom, nr_switches);
1046 		else
1047 			avg_atom = -1LL;
1048 
1049 		avg_per_cpu = p->se.sum_exec_runtime;
1050 		if (p->se.nr_migrations) {
1051 			avg_per_cpu = div64_u64(avg_per_cpu,
1052 						p->se.nr_migrations);
1053 		} else {
1054 			avg_per_cpu = -1LL;
1055 		}
1056 
1057 		__PN(avg_atom);
1058 		__PN(avg_per_cpu);
1059 
1060 #ifdef CONFIG_SCHED_CORE
1061 		PN_SCHEDSTAT(core_forceidle_sum);
1062 #endif
1063 	}
1064 
1065 	__P(nr_switches);
1066 	__PS("nr_voluntary_switches", p->nvcsw);
1067 	__PS("nr_involuntary_switches", p->nivcsw);
1068 
1069 	P(se.load.weight);
1070 #ifdef CONFIG_SMP
1071 	P(se.avg.load_sum);
1072 	P(se.avg.runnable_sum);
1073 	P(se.avg.util_sum);
1074 	P(se.avg.load_avg);
1075 	P(se.avg.runnable_avg);
1076 	P(se.avg.util_avg);
1077 	P(se.avg.last_update_time);
1078 	P(se.avg.util_est.ewma);
1079 	PM(se.avg.util_est.enqueued, ~UTIL_AVG_UNCHANGED);
1080 #endif
1081 #ifdef CONFIG_UCLAMP_TASK
1082 	__PS("uclamp.min", p->uclamp_req[UCLAMP_MIN].value);
1083 	__PS("uclamp.max", p->uclamp_req[UCLAMP_MAX].value);
1084 	__PS("effective uclamp.min", uclamp_eff_value(p, UCLAMP_MIN));
1085 	__PS("effective uclamp.max", uclamp_eff_value(p, UCLAMP_MAX));
1086 #endif
1087 	P(policy);
1088 	P(prio);
1089 	if (task_has_dl_policy(p)) {
1090 		P(dl.runtime);
1091 		P(dl.deadline);
1092 	}
1093 #undef PN_SCHEDSTAT
1094 #undef P_SCHEDSTAT
1095 
1096 	{
1097 		unsigned int this_cpu = raw_smp_processor_id();
1098 		u64 t0, t1;
1099 
1100 		t0 = cpu_clock(this_cpu);
1101 		t1 = cpu_clock(this_cpu);
1102 		__PS("clock-delta", t1-t0);
1103 	}
1104 
1105 	sched_show_numa(p, m);
1106 }
1107 
proc_sched_set_task(struct task_struct * p)1108 void proc_sched_set_task(struct task_struct *p)
1109 {
1110 #ifdef CONFIG_SCHEDSTATS
1111 	memset(&p->stats, 0, sizeof(p->stats));
1112 #endif
1113 }
1114 
resched_latency_warn(int cpu,u64 latency)1115 void resched_latency_warn(int cpu, u64 latency)
1116 {
1117 	static DEFINE_RATELIMIT_STATE(latency_check_ratelimit, 60 * 60 * HZ, 1);
1118 
1119 	WARN(__ratelimit(&latency_check_ratelimit),
1120 	     "sched: CPU %d need_resched set for > %llu ns (%d ticks) "
1121 	     "without schedule\n",
1122 	     cpu, latency, cpu_rq(cpu)->ticks_without_resched);
1123 }
1124