• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 /sys/kernel/debug/sched/debug and
12  * to the console
13  */
14 #define SEQ_printf(m, x...)			\
15  do {						\
16 	if (m)					\
17 		seq_printf(m, x);		\
18 	else					\
19 		pr_cont(x);			\
20  } while (0)
21 
22 /*
23  * Ease the printing of nsec fields:
24  */
nsec_high(unsigned long long nsec)25 static long long nsec_high(unsigned long long nsec)
26 {
27 	if ((long long)nsec < 0) {
28 		nsec = -nsec;
29 		do_div(nsec, 1000000);
30 		return -nsec;
31 	}
32 	do_div(nsec, 1000000);
33 
34 	return nsec;
35 }
36 
nsec_low(unsigned long long nsec)37 static unsigned long nsec_low(unsigned long long nsec)
38 {
39 	if ((long long)nsec < 0)
40 		nsec = -nsec;
41 
42 	return do_div(nsec, 1000000);
43 }
44 
45 #define SPLIT_NS(x) nsec_high(x), nsec_low(x)
46 
47 #define SCHED_FEAT(name, enabled)	\
48 	#name ,
49 
50 const char * const sched_feat_names[] = {
51 #include "features.h"
52 };
53 EXPORT_SYMBOL_GPL(sched_feat_names);
54 
55 #undef SCHED_FEAT
56 
sched_feat_show(struct seq_file * m,void * v)57 static int sched_feat_show(struct seq_file *m, void *v)
58 {
59 	int i;
60 
61 	for (i = 0; i < __SCHED_FEAT_NR; i++) {
62 		if (!(sysctl_sched_features & (1UL << i)))
63 			seq_puts(m, "NO_");
64 		seq_printf(m, "%s ", sched_feat_names[i]);
65 	}
66 	seq_puts(m, "\n");
67 
68 	return 0;
69 }
70 
71 #ifdef CONFIG_JUMP_LABEL
72 
73 #define jump_label_key__true  STATIC_KEY_INIT_TRUE
74 #define jump_label_key__false STATIC_KEY_INIT_FALSE
75 
76 #define SCHED_FEAT(name, enabled)	\
77 	jump_label_key__##enabled ,
78 
79 struct static_key sched_feat_keys[__SCHED_FEAT_NR] = {
80 #include "features.h"
81 };
82 EXPORT_SYMBOL_GPL(sched_feat_keys);
83 
84 #undef SCHED_FEAT
85 
sched_feat_disable(int i)86 static void sched_feat_disable(int i)
87 {
88 	static_key_disable_cpuslocked(&sched_feat_keys[i]);
89 }
90 
sched_feat_enable(int i)91 static void sched_feat_enable(int i)
92 {
93 	static_key_enable_cpuslocked(&sched_feat_keys[i]);
94 }
95 #else
sched_feat_disable(int i)96 static void sched_feat_disable(int i) { };
sched_feat_enable(int i)97 static void sched_feat_enable(int i) { };
98 #endif /* CONFIG_JUMP_LABEL */
99 
sched_feat_set(char * cmp)100 static int sched_feat_set(char *cmp)
101 {
102 	int i;
103 	int neg = 0;
104 
105 	if (strncmp(cmp, "NO_", 3) == 0) {
106 		neg = 1;
107 		cmp += 3;
108 	}
109 
110 	i = match_string(sched_feat_names, __SCHED_FEAT_NR, cmp);
111 	if (i < 0)
112 		return i;
113 
114 	if (neg) {
115 		sysctl_sched_features &= ~(1UL << i);
116 		sched_feat_disable(i);
117 	} else {
118 		sysctl_sched_features |= (1UL << i);
119 		sched_feat_enable(i);
120 	}
121 
122 	return 0;
123 }
124 
125 static ssize_t
sched_feat_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)126 sched_feat_write(struct file *filp, const char __user *ubuf,
127 		size_t cnt, loff_t *ppos)
128 {
129 	char buf[64];
130 	char *cmp;
131 	int ret;
132 	struct inode *inode;
133 
134 	if (cnt > 63)
135 		cnt = 63;
136 
137 	if (copy_from_user(&buf, ubuf, cnt))
138 		return -EFAULT;
139 
140 	buf[cnt] = 0;
141 	cmp = strstrip(buf);
142 
143 	/* Ensure the static_key remains in a consistent state */
144 	inode = file_inode(filp);
145 	cpus_read_lock();
146 	inode_lock(inode);
147 	ret = sched_feat_set(cmp);
148 	inode_unlock(inode);
149 	cpus_read_unlock();
150 	if (ret < 0)
151 		return ret;
152 
153 	*ppos += cnt;
154 
155 	return cnt;
156 }
157 
sched_feat_open(struct inode * inode,struct file * filp)158 static int sched_feat_open(struct inode *inode, struct file *filp)
159 {
160 	return single_open(filp, sched_feat_show, NULL);
161 }
162 
163 static const struct file_operations sched_feat_fops = {
164 	.open		= sched_feat_open,
165 	.write		= sched_feat_write,
166 	.read		= seq_read,
167 	.llseek		= seq_lseek,
168 	.release	= single_release,
169 };
170 
171 #ifdef CONFIG_SMP
172 
sched_scaling_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)173 static ssize_t sched_scaling_write(struct file *filp, const char __user *ubuf,
174 				   size_t cnt, loff_t *ppos)
175 {
176 	char buf[16];
177 	unsigned int scaling;
178 
179 	if (cnt > 15)
180 		cnt = 15;
181 
182 	if (copy_from_user(&buf, ubuf, cnt))
183 		return -EFAULT;
184 	buf[cnt] = '\0';
185 
186 	if (kstrtouint(buf, 10, &scaling))
187 		return -EINVAL;
188 
189 	if (scaling >= SCHED_TUNABLESCALING_END)
190 		return -EINVAL;
191 
192 	sysctl_sched_tunable_scaling = scaling;
193 	if (sched_update_scaling())
194 		return -EINVAL;
195 
196 	*ppos += cnt;
197 	return cnt;
198 }
199 
sched_scaling_show(struct seq_file * m,void * v)200 static int sched_scaling_show(struct seq_file *m, void *v)
201 {
202 	seq_printf(m, "%d\n", sysctl_sched_tunable_scaling);
203 	return 0;
204 }
205 
sched_scaling_open(struct inode * inode,struct file * filp)206 static int sched_scaling_open(struct inode *inode, struct file *filp)
207 {
208 	return single_open(filp, sched_scaling_show, NULL);
209 }
210 
211 static const struct file_operations sched_scaling_fops = {
212 	.open		= sched_scaling_open,
213 	.write		= sched_scaling_write,
214 	.read		= seq_read,
215 	.llseek		= seq_lseek,
216 	.release	= single_release,
217 };
218 
219 #endif /* SMP */
220 
221 #ifdef CONFIG_PREEMPT_DYNAMIC
222 
sched_dynamic_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)223 static ssize_t sched_dynamic_write(struct file *filp, const char __user *ubuf,
224 				   size_t cnt, loff_t *ppos)
225 {
226 	char buf[16];
227 	int mode;
228 
229 	if (cnt > 15)
230 		cnt = 15;
231 
232 	if (copy_from_user(&buf, ubuf, cnt))
233 		return -EFAULT;
234 
235 	buf[cnt] = 0;
236 	mode = sched_dynamic_mode(strstrip(buf));
237 	if (mode < 0)
238 		return mode;
239 
240 	sched_dynamic_update(mode);
241 
242 	*ppos += cnt;
243 
244 	return cnt;
245 }
246 
sched_dynamic_show(struct seq_file * m,void * v)247 static int sched_dynamic_show(struct seq_file *m, void *v)
248 {
249 	static const char * preempt_modes[] = {
250 		"none", "voluntary", "full"
251 	};
252 	int i;
253 
254 	for (i = 0; i < ARRAY_SIZE(preempt_modes); i++) {
255 		if (preempt_dynamic_mode == i)
256 			seq_puts(m, "(");
257 		seq_puts(m, preempt_modes[i]);
258 		if (preempt_dynamic_mode == i)
259 			seq_puts(m, ")");
260 
261 		seq_puts(m, " ");
262 	}
263 
264 	seq_puts(m, "\n");
265 	return 0;
266 }
267 
sched_dynamic_open(struct inode * inode,struct file * filp)268 static int sched_dynamic_open(struct inode *inode, struct file *filp)
269 {
270 	return single_open(filp, sched_dynamic_show, NULL);
271 }
272 
273 static const struct file_operations sched_dynamic_fops = {
274 	.open		= sched_dynamic_open,
275 	.write		= sched_dynamic_write,
276 	.read		= seq_read,
277 	.llseek		= seq_lseek,
278 	.release	= single_release,
279 };
280 
281 #endif /* CONFIG_PREEMPT_DYNAMIC */
282 
283 __read_mostly bool sched_debug_verbose;
284 
285 #ifdef CONFIG_SMP
286 static struct dentry           *sd_dentry;
287 
288 
sched_verbose_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)289 static ssize_t sched_verbose_write(struct file *filp, const char __user *ubuf,
290 				  size_t cnt, loff_t *ppos)
291 {
292 	ssize_t result;
293 	bool orig;
294 
295 	cpus_read_lock();
296 	mutex_lock(&sched_domains_mutex);
297 
298 	orig = sched_debug_verbose;
299 	result = debugfs_write_file_bool(filp, ubuf, cnt, ppos);
300 
301 	if (sched_debug_verbose && !orig)
302 		update_sched_domain_debugfs();
303 	else if (!sched_debug_verbose && orig) {
304 		debugfs_remove(sd_dentry);
305 		sd_dentry = NULL;
306 	}
307 
308 	mutex_unlock(&sched_domains_mutex);
309 	cpus_read_unlock();
310 
311 	return result;
312 }
313 #else
314 #define sched_verbose_write debugfs_write_file_bool
315 #endif
316 
317 static const struct file_operations sched_verbose_fops = {
318 	.read =         debugfs_read_file_bool,
319 	.write =        sched_verbose_write,
320 	.open =         simple_open,
321 	.llseek =       default_llseek,
322 };
323 
324 static const struct seq_operations sched_debug_sops;
325 
sched_debug_open(struct inode * inode,struct file * filp)326 static int sched_debug_open(struct inode *inode, struct file *filp)
327 {
328 	return seq_open(filp, &sched_debug_sops);
329 }
330 
331 static const struct file_operations sched_debug_fops = {
332 	.open		= sched_debug_open,
333 	.read		= seq_read,
334 	.llseek		= seq_lseek,
335 	.release	= seq_release,
336 };
337 
338 enum dl_param {
339 	DL_RUNTIME = 0,
340 	DL_PERIOD,
341 };
342 
343 static unsigned long fair_server_period_max = (1UL << 22) * NSEC_PER_USEC; /* ~4 seconds */
344 static unsigned long fair_server_period_min = (100) * NSEC_PER_USEC;     /* 100 us */
345 
sched_fair_server_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos,enum dl_param param)346 static ssize_t sched_fair_server_write(struct file *filp, const char __user *ubuf,
347 				       size_t cnt, loff_t *ppos, enum dl_param param)
348 {
349 	long cpu = (long) ((struct seq_file *) filp->private_data)->private;
350 	struct rq *rq = cpu_rq(cpu);
351 	u64 runtime, period;
352 	size_t err;
353 	int retval;
354 	u64 value;
355 
356 	err = kstrtoull_from_user(ubuf, cnt, 10, &value);
357 	if (err)
358 		return err;
359 
360 	scoped_guard (rq_lock_irqsave, rq) {
361 		runtime  = rq->fair_server.dl_runtime;
362 		period = rq->fair_server.dl_period;
363 
364 		switch (param) {
365 		case DL_RUNTIME:
366 			if (runtime == value)
367 				break;
368 			runtime = value;
369 			break;
370 		case DL_PERIOD:
371 			if (value == period)
372 				break;
373 			period = value;
374 			break;
375 		}
376 
377 		if (runtime > period ||
378 		    period > fair_server_period_max ||
379 		    period < fair_server_period_min) {
380 			return  -EINVAL;
381 		}
382 
383 		update_rq_clock(rq);
384 		dl_server_stop(&rq->fair_server);
385 
386 		retval = dl_server_apply_params(&rq->fair_server, runtime, period, 0);
387 		if (retval)
388 			cnt = retval;
389 
390 		if (!runtime)
391 			printk_deferred("Fair server disabled in CPU %d, system may crash due to starvation.\n",
392 					cpu_of(rq));
393 
394 		if (rq->cfs.h_nr_running)
395 			dl_server_start(&rq->fair_server);
396 	}
397 
398 	*ppos += cnt;
399 	return cnt;
400 }
401 
sched_fair_server_show(struct seq_file * m,void * v,enum dl_param param)402 static size_t sched_fair_server_show(struct seq_file *m, void *v, enum dl_param param)
403 {
404 	unsigned long cpu = (unsigned long) m->private;
405 	struct rq *rq = cpu_rq(cpu);
406 	u64 value;
407 
408 	switch (param) {
409 	case DL_RUNTIME:
410 		value = rq->fair_server.dl_runtime;
411 		break;
412 	case DL_PERIOD:
413 		value = rq->fair_server.dl_period;
414 		break;
415 	}
416 
417 	seq_printf(m, "%llu\n", value);
418 	return 0;
419 
420 }
421 
422 static ssize_t
sched_fair_server_runtime_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)423 sched_fair_server_runtime_write(struct file *filp, const char __user *ubuf,
424 				size_t cnt, loff_t *ppos)
425 {
426 	return sched_fair_server_write(filp, ubuf, cnt, ppos, DL_RUNTIME);
427 }
428 
sched_fair_server_runtime_show(struct seq_file * m,void * v)429 static int sched_fair_server_runtime_show(struct seq_file *m, void *v)
430 {
431 	return sched_fair_server_show(m, v, DL_RUNTIME);
432 }
433 
sched_fair_server_runtime_open(struct inode * inode,struct file * filp)434 static int sched_fair_server_runtime_open(struct inode *inode, struct file *filp)
435 {
436 	return single_open(filp, sched_fair_server_runtime_show, inode->i_private);
437 }
438 
439 static const struct file_operations fair_server_runtime_fops = {
440 	.open		= sched_fair_server_runtime_open,
441 	.write		= sched_fair_server_runtime_write,
442 	.read		= seq_read,
443 	.llseek		= seq_lseek,
444 	.release	= single_release,
445 };
446 
447 static ssize_t
sched_fair_server_period_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)448 sched_fair_server_period_write(struct file *filp, const char __user *ubuf,
449 			       size_t cnt, loff_t *ppos)
450 {
451 	return sched_fair_server_write(filp, ubuf, cnt, ppos, DL_PERIOD);
452 }
453 
sched_fair_server_period_show(struct seq_file * m,void * v)454 static int sched_fair_server_period_show(struct seq_file *m, void *v)
455 {
456 	return sched_fair_server_show(m, v, DL_PERIOD);
457 }
458 
sched_fair_server_period_open(struct inode * inode,struct file * filp)459 static int sched_fair_server_period_open(struct inode *inode, struct file *filp)
460 {
461 	return single_open(filp, sched_fair_server_period_show, inode->i_private);
462 }
463 
464 static const struct file_operations fair_server_period_fops = {
465 	.open		= sched_fair_server_period_open,
466 	.write		= sched_fair_server_period_write,
467 	.read		= seq_read,
468 	.llseek		= seq_lseek,
469 	.release	= single_release,
470 };
471 
472 static struct dentry *debugfs_sched;
473 
debugfs_fair_server_init(void)474 static void debugfs_fair_server_init(void)
475 {
476 	struct dentry *d_fair;
477 	unsigned long cpu;
478 
479 	d_fair = debugfs_create_dir("fair_server", debugfs_sched);
480 	if (!d_fair)
481 		return;
482 
483 	for_each_possible_cpu(cpu) {
484 		struct dentry *d_cpu;
485 		char buf[32];
486 
487 		snprintf(buf, sizeof(buf), "cpu%lu", cpu);
488 		d_cpu = debugfs_create_dir(buf, d_fair);
489 
490 		debugfs_create_file("runtime", 0644, d_cpu, (void *) cpu, &fair_server_runtime_fops);
491 		debugfs_create_file("period", 0644, d_cpu, (void *) cpu, &fair_server_period_fops);
492 	}
493 }
494 
sched_init_debug(void)495 static __init int sched_init_debug(void)
496 {
497 	struct dentry __maybe_unused *numa;
498 
499 	debugfs_sched = debugfs_create_dir("sched", NULL);
500 
501 	debugfs_create_file("features", 0644, debugfs_sched, NULL, &sched_feat_fops);
502 	debugfs_create_file_unsafe("verbose", 0644, debugfs_sched, &sched_debug_verbose, &sched_verbose_fops);
503 #ifdef CONFIG_PREEMPT_DYNAMIC
504 	debugfs_create_file("preempt", 0644, debugfs_sched, NULL, &sched_dynamic_fops);
505 #endif
506 
507 	debugfs_create_u32("base_slice_ns", 0644, debugfs_sched, &sysctl_sched_base_slice);
508 
509 	debugfs_create_u32("latency_warn_ms", 0644, debugfs_sched, &sysctl_resched_latency_warn_ms);
510 	debugfs_create_u32("latency_warn_once", 0644, debugfs_sched, &sysctl_resched_latency_warn_once);
511 
512 #ifdef CONFIG_SMP
513 	debugfs_create_file("tunable_scaling", 0644, debugfs_sched, NULL, &sched_scaling_fops);
514 	debugfs_create_u32("migration_cost_ns", 0644, debugfs_sched, &sysctl_sched_migration_cost);
515 	debugfs_create_u32("nr_migrate", 0644, debugfs_sched, &sysctl_sched_nr_migrate);
516 
517 	mutex_lock(&sched_domains_mutex);
518 	update_sched_domain_debugfs();
519 	mutex_unlock(&sched_domains_mutex);
520 #endif
521 
522 #ifdef CONFIG_NUMA_BALANCING
523 	numa = debugfs_create_dir("numa_balancing", debugfs_sched);
524 
525 	debugfs_create_u32("scan_delay_ms", 0644, numa, &sysctl_numa_balancing_scan_delay);
526 	debugfs_create_u32("scan_period_min_ms", 0644, numa, &sysctl_numa_balancing_scan_period_min);
527 	debugfs_create_u32("scan_period_max_ms", 0644, numa, &sysctl_numa_balancing_scan_period_max);
528 	debugfs_create_u32("scan_size_mb", 0644, numa, &sysctl_numa_balancing_scan_size);
529 	debugfs_create_u32("hot_threshold_ms", 0644, numa, &sysctl_numa_balancing_hot_threshold);
530 #endif
531 
532 	debugfs_create_file("debug", 0444, debugfs_sched, NULL, &sched_debug_fops);
533 
534 	debugfs_fair_server_init();
535 
536 	return 0;
537 }
538 late_initcall(sched_init_debug);
539 
540 #ifdef CONFIG_SMP
541 
542 static cpumask_var_t		sd_sysctl_cpus;
543 
sd_flags_show(struct seq_file * m,void * v)544 static int sd_flags_show(struct seq_file *m, void *v)
545 {
546 	unsigned long flags = *(unsigned int *)m->private;
547 	int idx;
548 
549 	for_each_set_bit(idx, &flags, __SD_FLAG_CNT) {
550 		seq_puts(m, sd_flag_debug[idx].name);
551 		seq_puts(m, " ");
552 	}
553 	seq_puts(m, "\n");
554 
555 	return 0;
556 }
557 
sd_flags_open(struct inode * inode,struct file * file)558 static int sd_flags_open(struct inode *inode, struct file *file)
559 {
560 	return single_open(file, sd_flags_show, inode->i_private);
561 }
562 
563 static const struct file_operations sd_flags_fops = {
564 	.open		= sd_flags_open,
565 	.read		= seq_read,
566 	.llseek		= seq_lseek,
567 	.release	= single_release,
568 };
569 
register_sd(struct sched_domain * sd,struct dentry * parent)570 static void register_sd(struct sched_domain *sd, struct dentry *parent)
571 {
572 #define SDM(type, mode, member)	\
573 	debugfs_create_##type(#member, mode, parent, &sd->member)
574 
575 	SDM(ulong, 0644, min_interval);
576 	SDM(ulong, 0644, max_interval);
577 	SDM(u64,   0644, max_newidle_lb_cost);
578 	SDM(u32,   0644, busy_factor);
579 	SDM(u32,   0644, imbalance_pct);
580 	SDM(u32,   0644, cache_nice_tries);
581 	SDM(str,   0444, name);
582 
583 #undef SDM
584 
585 	debugfs_create_file("flags", 0444, parent, &sd->flags, &sd_flags_fops);
586 	debugfs_create_file("groups_flags", 0444, parent, &sd->groups->flags, &sd_flags_fops);
587 	debugfs_create_u32("level", 0444, parent, (u32 *)&sd->level);
588 }
589 
update_sched_domain_debugfs(void)590 void update_sched_domain_debugfs(void)
591 {
592 	int cpu, i;
593 
594 	/*
595 	 * This can unfortunately be invoked before sched_debug_init() creates
596 	 * the debug directory. Don't touch sd_sysctl_cpus until then.
597 	 */
598 	if (!debugfs_sched)
599 		return;
600 
601 	if (!sched_debug_verbose)
602 		return;
603 
604 	if (!cpumask_available(sd_sysctl_cpus)) {
605 		if (!alloc_cpumask_var(&sd_sysctl_cpus, GFP_KERNEL))
606 			return;
607 		cpumask_copy(sd_sysctl_cpus, cpu_possible_mask);
608 	}
609 
610 	if (!sd_dentry) {
611 		sd_dentry = debugfs_create_dir("domains", debugfs_sched);
612 
613 		/* rebuild sd_sysctl_cpus if empty since it gets cleared below */
614 		if (cpumask_empty(sd_sysctl_cpus))
615 			cpumask_copy(sd_sysctl_cpus, cpu_online_mask);
616 	}
617 
618 	for_each_cpu(cpu, sd_sysctl_cpus) {
619 		struct sched_domain *sd;
620 		struct dentry *d_cpu;
621 		char buf[32];
622 
623 		snprintf(buf, sizeof(buf), "cpu%d", cpu);
624 		debugfs_lookup_and_remove(buf, sd_dentry);
625 		d_cpu = debugfs_create_dir(buf, sd_dentry);
626 
627 		i = 0;
628 		for_each_domain(cpu, sd) {
629 			struct dentry *d_sd;
630 
631 			snprintf(buf, sizeof(buf), "domain%d", i);
632 			d_sd = debugfs_create_dir(buf, d_cpu);
633 
634 			register_sd(sd, d_sd);
635 			i++;
636 		}
637 
638 		__cpumask_clear_cpu(cpu, sd_sysctl_cpus);
639 	}
640 }
641 
dirty_sched_domain_sysctl(int cpu)642 void dirty_sched_domain_sysctl(int cpu)
643 {
644 	if (cpumask_available(sd_sysctl_cpus))
645 		__cpumask_set_cpu(cpu, sd_sysctl_cpus);
646 }
647 
648 #endif /* CONFIG_SMP */
649 
650 #ifdef CONFIG_FAIR_GROUP_SCHED
print_cfs_group_stats(struct seq_file * m,int cpu,struct task_group * tg)651 static void print_cfs_group_stats(struct seq_file *m, int cpu, struct task_group *tg)
652 {
653 	struct sched_entity *se = tg->se[cpu];
654 
655 #define P(F)		SEQ_printf(m, "  .%-30s: %lld\n",	#F, (long long)F)
656 #define P_SCHEDSTAT(F)	SEQ_printf(m, "  .%-30s: %lld\n",	\
657 		#F, (long long)schedstat_val(stats->F))
658 #define PN(F)		SEQ_printf(m, "  .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F))
659 #define PN_SCHEDSTAT(F)	SEQ_printf(m, "  .%-30s: %lld.%06ld\n", \
660 		#F, SPLIT_NS((long long)schedstat_val(stats->F)))
661 
662 	if (!se)
663 		return;
664 
665 	PN(se->exec_start);
666 	PN(se->vruntime);
667 	PN(se->sum_exec_runtime);
668 
669 	if (schedstat_enabled()) {
670 		struct sched_statistics *stats;
671 		stats = __schedstats_from_se(se);
672 
673 		PN_SCHEDSTAT(wait_start);
674 		PN_SCHEDSTAT(sleep_start);
675 		PN_SCHEDSTAT(block_start);
676 		PN_SCHEDSTAT(sleep_max);
677 		PN_SCHEDSTAT(block_max);
678 		PN_SCHEDSTAT(exec_max);
679 		PN_SCHEDSTAT(slice_max);
680 		PN_SCHEDSTAT(wait_max);
681 		PN_SCHEDSTAT(wait_sum);
682 		P_SCHEDSTAT(wait_count);
683 	}
684 
685 	P(se->load.weight);
686 #ifdef CONFIG_SMP
687 	P(se->avg.load_avg);
688 	P(se->avg.util_avg);
689 	P(se->avg.runnable_avg);
690 #endif
691 
692 #undef PN_SCHEDSTAT
693 #undef PN
694 #undef P_SCHEDSTAT
695 #undef P
696 }
697 #endif
698 
699 #ifdef CONFIG_CGROUP_SCHED
700 static DEFINE_SPINLOCK(sched_debug_lock);
701 static char group_path[PATH_MAX];
702 
task_group_path(struct task_group * tg,char * path,int plen)703 static void task_group_path(struct task_group *tg, char *path, int plen)
704 {
705 	if (autogroup_path(tg, path, plen))
706 		return;
707 
708 	cgroup_path(tg->css.cgroup, path, plen);
709 }
710 
711 /*
712  * Only 1 SEQ_printf_task_group_path() caller can use the full length
713  * group_path[] for cgroup path. Other simultaneous callers will have
714  * to use a shorter stack buffer. A "..." suffix is appended at the end
715  * of the stack buffer so that it will show up in case the output length
716  * matches the given buffer size to indicate possible path name truncation.
717  */
718 #define SEQ_printf_task_group_path(m, tg, fmt...)			\
719 {									\
720 	if (spin_trylock(&sched_debug_lock)) {				\
721 		task_group_path(tg, group_path, sizeof(group_path));	\
722 		SEQ_printf(m, fmt, group_path);				\
723 		spin_unlock(&sched_debug_lock);				\
724 	} else {							\
725 		char buf[128];						\
726 		char *bufend = buf + sizeof(buf) - 3;			\
727 		task_group_path(tg, buf, bufend - buf);			\
728 		strcpy(bufend - 1, "...");				\
729 		SEQ_printf(m, fmt, buf);				\
730 	}								\
731 }
732 #endif
733 
734 static void
print_task(struct seq_file * m,struct rq * rq,struct task_struct * p)735 print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
736 {
737 	if (task_current(rq, p))
738 		SEQ_printf(m, ">R");
739 	else
740 		SEQ_printf(m, " %c", task_state_to_char(p));
741 
742 	SEQ_printf(m, " %15s %5d %9Ld.%06ld   %c   %9Ld.%06ld %c %9Ld.%06ld %9Ld.%06ld %9Ld   %5d ",
743 		p->comm, task_pid_nr(p),
744 		SPLIT_NS(p->se.vruntime),
745 		entity_eligible(cfs_rq_of(&p->se), &p->se) ? 'E' : 'N',
746 		SPLIT_NS(p->se.deadline),
747 		p->se.custom_slice ? 'S' : ' ',
748 		SPLIT_NS(p->se.slice),
749 		SPLIT_NS(p->se.sum_exec_runtime),
750 		(long long)(p->nvcsw + p->nivcsw),
751 		p->prio);
752 
753 	SEQ_printf(m, "%9lld.%06ld %9lld.%06ld %9lld.%06ld",
754 		SPLIT_NS(schedstat_val_or_zero(p->stats.wait_sum)),
755 		SPLIT_NS(schedstat_val_or_zero(p->stats.sum_sleep_runtime)),
756 		SPLIT_NS(schedstat_val_or_zero(p->stats.sum_block_runtime)));
757 
758 #ifdef CONFIG_NUMA_BALANCING
759 	SEQ_printf(m, "   %d      %d", task_node(p), task_numa_group_id(p));
760 #endif
761 #ifdef CONFIG_CGROUP_SCHED
762 	SEQ_printf_task_group_path(m, task_group(p), "        %s")
763 #endif
764 
765 	SEQ_printf(m, "\n");
766 }
767 
print_rq(struct seq_file * m,struct rq * rq,int rq_cpu)768 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
769 {
770 	struct task_struct *g, *p;
771 
772 	SEQ_printf(m, "\n");
773 	SEQ_printf(m, "runnable tasks:\n");
774 	SEQ_printf(m, " S            task   PID       vruntime   eligible    "
775 		   "deadline             slice          sum-exec      switches  "
776 		   "prio         wait-time        sum-sleep       sum-block"
777 #ifdef CONFIG_NUMA_BALANCING
778 		   "  node   group-id"
779 #endif
780 #ifdef CONFIG_CGROUP_SCHED
781 		   "  group-path"
782 #endif
783 		   "\n");
784 	SEQ_printf(m, "-------------------------------------------------------"
785 		   "------------------------------------------------------"
786 		   "------------------------------------------------------"
787 #ifdef CONFIG_NUMA_BALANCING
788 		   "--------------"
789 #endif
790 #ifdef CONFIG_CGROUP_SCHED
791 		   "--------------"
792 #endif
793 		   "\n");
794 
795 	rcu_read_lock();
796 	for_each_process_thread(g, p) {
797 		if (task_cpu(p) != rq_cpu)
798 			continue;
799 
800 		print_task(m, rq, p);
801 	}
802 	rcu_read_unlock();
803 }
804 
print_cfs_rq(struct seq_file * m,int cpu,struct cfs_rq * cfs_rq)805 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
806 {
807 	s64 left_vruntime = -1, min_vruntime, right_vruntime = -1, left_deadline = -1, spread;
808 	struct sched_entity *last, *first, *root;
809 	struct rq *rq = cpu_rq(cpu);
810 	unsigned long flags;
811 
812 #ifdef CONFIG_FAIR_GROUP_SCHED
813 	SEQ_printf(m, "\n");
814 	SEQ_printf_task_group_path(m, cfs_rq->tg, "cfs_rq[%d]:%s\n", cpu);
815 #else
816 	SEQ_printf(m, "\n");
817 	SEQ_printf(m, "cfs_rq[%d]:\n", cpu);
818 #endif
819 
820 	raw_spin_rq_lock_irqsave(rq, flags);
821 	root = __pick_root_entity(cfs_rq);
822 	if (root)
823 		left_vruntime = root->min_vruntime;
824 	first = __pick_first_entity(cfs_rq);
825 	if (first)
826 		left_deadline = first->deadline;
827 	last = __pick_last_entity(cfs_rq);
828 	if (last)
829 		right_vruntime = last->vruntime;
830 	min_vruntime = cfs_rq->min_vruntime;
831 	raw_spin_rq_unlock_irqrestore(rq, flags);
832 
833 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "left_deadline",
834 			SPLIT_NS(left_deadline));
835 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "left_vruntime",
836 			SPLIT_NS(left_vruntime));
837 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "min_vruntime",
838 			SPLIT_NS(min_vruntime));
839 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "avg_vruntime",
840 			SPLIT_NS(avg_vruntime(cfs_rq)));
841 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "right_vruntime",
842 			SPLIT_NS(right_vruntime));
843 	spread = right_vruntime - left_vruntime;
844 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread", SPLIT_NS(spread));
845 	SEQ_printf(m, "  .%-30s: %d\n", "nr_running", cfs_rq->nr_running);
846 	SEQ_printf(m, "  .%-30s: %d\n", "h_nr_running", cfs_rq->h_nr_running);
847 	SEQ_printf(m, "  .%-30s: %d\n", "h_nr_delayed", cfs_rq->h_nr_delayed);
848 	SEQ_printf(m, "  .%-30s: %d\n", "idle_nr_running",
849 			cfs_rq->idle_nr_running);
850 	SEQ_printf(m, "  .%-30s: %d\n", "idle_h_nr_running",
851 			cfs_rq->idle_h_nr_running);
852 	SEQ_printf(m, "  .%-30s: %ld\n", "load", cfs_rq->load.weight);
853 #ifdef CONFIG_SMP
854 	SEQ_printf(m, "  .%-30s: %lu\n", "load_avg",
855 			cfs_rq->avg.load_avg);
856 	SEQ_printf(m, "  .%-30s: %lu\n", "runnable_avg",
857 			cfs_rq->avg.runnable_avg);
858 	SEQ_printf(m, "  .%-30s: %lu\n", "util_avg",
859 			cfs_rq->avg.util_avg);
860 	SEQ_printf(m, "  .%-30s: %u\n", "util_est",
861 			cfs_rq->avg.util_est);
862 	SEQ_printf(m, "  .%-30s: %ld\n", "removed.load_avg",
863 			cfs_rq->removed.load_avg);
864 	SEQ_printf(m, "  .%-30s: %ld\n", "removed.util_avg",
865 			cfs_rq->removed.util_avg);
866 	SEQ_printf(m, "  .%-30s: %ld\n", "removed.runnable_avg",
867 			cfs_rq->removed.runnable_avg);
868 #ifdef CONFIG_FAIR_GROUP_SCHED
869 	SEQ_printf(m, "  .%-30s: %lu\n", "tg_load_avg_contrib",
870 			cfs_rq->tg_load_avg_contrib);
871 	SEQ_printf(m, "  .%-30s: %ld\n", "tg_load_avg",
872 			atomic_long_read(&cfs_rq->tg->load_avg));
873 #endif
874 #endif
875 #ifdef CONFIG_CFS_BANDWIDTH
876 	SEQ_printf(m, "  .%-30s: %d\n", "throttled",
877 			cfs_rq->throttled);
878 	SEQ_printf(m, "  .%-30s: %d\n", "throttle_count",
879 			cfs_rq->throttle_count);
880 #endif
881 
882 #ifdef CONFIG_FAIR_GROUP_SCHED
883 	print_cfs_group_stats(m, cpu, cfs_rq->tg);
884 #endif
885 }
886 
print_rt_rq(struct seq_file * m,int cpu,struct rt_rq * rt_rq)887 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
888 {
889 #ifdef CONFIG_RT_GROUP_SCHED
890 	SEQ_printf(m, "\n");
891 	SEQ_printf_task_group_path(m, rt_rq->tg, "rt_rq[%d]:%s\n", cpu);
892 #else
893 	SEQ_printf(m, "\n");
894 	SEQ_printf(m, "rt_rq[%d]:\n", cpu);
895 #endif
896 
897 #define P(x) \
898 	SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
899 #define PU(x) \
900 	SEQ_printf(m, "  .%-30s: %lu\n", #x, (unsigned long)(rt_rq->x))
901 #define PN(x) \
902 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
903 
904 	PU(rt_nr_running);
905 
906 #ifdef CONFIG_RT_GROUP_SCHED
907 	P(rt_throttled);
908 	PN(rt_time);
909 	PN(rt_runtime);
910 #endif
911 
912 #undef PN
913 #undef PU
914 #undef P
915 }
916 
print_dl_rq(struct seq_file * m,int cpu,struct dl_rq * dl_rq)917 void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq)
918 {
919 	struct dl_bw *dl_bw;
920 
921 	SEQ_printf(m, "\n");
922 	SEQ_printf(m, "dl_rq[%d]:\n", cpu);
923 
924 #define PU(x) \
925 	SEQ_printf(m, "  .%-30s: %lu\n", #x, (unsigned long)(dl_rq->x))
926 
927 	PU(dl_nr_running);
928 #ifdef CONFIG_SMP
929 	dl_bw = &cpu_rq(cpu)->rd->dl_bw;
930 #else
931 	dl_bw = &dl_rq->dl_bw;
932 #endif
933 	SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->bw", dl_bw->bw);
934 	SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->total_bw", dl_bw->total_bw);
935 
936 #undef PU
937 }
938 
print_cpu(struct seq_file * m,int cpu)939 static void print_cpu(struct seq_file *m, int cpu)
940 {
941 	struct rq *rq = cpu_rq(cpu);
942 
943 #ifdef CONFIG_X86
944 	{
945 		unsigned int freq = cpu_khz ? : 1;
946 
947 		SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
948 			   cpu, freq / 1000, (freq % 1000));
949 	}
950 #else
951 	SEQ_printf(m, "cpu#%d\n", cpu);
952 #endif
953 
954 #define P(x)								\
955 do {									\
956 	if (sizeof(rq->x) == 4)						\
957 		SEQ_printf(m, "  .%-30s: %d\n", #x, (int)(rq->x));	\
958 	else								\
959 		SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rq->x));\
960 } while (0)
961 
962 #define PN(x) \
963 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
964 
965 	P(nr_running);
966 	P(nr_switches);
967 	P(nr_uninterruptible);
968 	PN(next_balance);
969 	SEQ_printf(m, "  .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr)));
970 	PN(clock);
971 	PN(clock_task);
972 #undef P
973 #undef PN
974 
975 #ifdef CONFIG_SMP
976 #define P64(n) SEQ_printf(m, "  .%-30s: %Ld\n", #n, rq->n);
977 	P64(avg_idle);
978 	P64(max_idle_balance_cost);
979 #undef P64
980 #endif
981 
982 #define P(n) SEQ_printf(m, "  .%-30s: %d\n", #n, schedstat_val(rq->n));
983 	if (schedstat_enabled()) {
984 		P(yld_count);
985 		P(sched_count);
986 		P(sched_goidle);
987 		P(ttwu_count);
988 		P(ttwu_local);
989 	}
990 #undef P
991 
992 	print_cfs_stats(m, cpu);
993 	print_rt_stats(m, cpu);
994 	print_dl_stats(m, cpu);
995 
996 	print_rq(m, rq, cpu);
997 	SEQ_printf(m, "\n");
998 }
999 
1000 static const char *sched_tunable_scaling_names[] = {
1001 	"none",
1002 	"logarithmic",
1003 	"linear"
1004 };
1005 
sched_debug_header(struct seq_file * m)1006 static void sched_debug_header(struct seq_file *m)
1007 {
1008 	u64 ktime, sched_clk, cpu_clk;
1009 	unsigned long flags;
1010 
1011 	local_irq_save(flags);
1012 	ktime = ktime_to_ns(ktime_get());
1013 	sched_clk = sched_clock();
1014 	cpu_clk = local_clock();
1015 	local_irq_restore(flags);
1016 
1017 	SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n",
1018 		init_utsname()->release,
1019 		(int)strcspn(init_utsname()->version, " "),
1020 		init_utsname()->version);
1021 
1022 #define P(x) \
1023 	SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
1024 #define PN(x) \
1025 	SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
1026 	PN(ktime);
1027 	PN(sched_clk);
1028 	PN(cpu_clk);
1029 	P(jiffies);
1030 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
1031 	P(sched_clock_stable());
1032 #endif
1033 #undef PN
1034 #undef P
1035 
1036 	SEQ_printf(m, "\n");
1037 	SEQ_printf(m, "sysctl_sched\n");
1038 
1039 #define P(x) \
1040 	SEQ_printf(m, "  .%-40s: %Ld\n", #x, (long long)(x))
1041 #define PN(x) \
1042 	SEQ_printf(m, "  .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
1043 	PN(sysctl_sched_base_slice);
1044 	P(sysctl_sched_child_runs_first);
1045 	P(sysctl_sched_features);
1046 #undef PN
1047 #undef P
1048 
1049 	SEQ_printf(m, "  .%-40s: %d (%s)\n",
1050 		"sysctl_sched_tunable_scaling",
1051 		sysctl_sched_tunable_scaling,
1052 		sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
1053 	SEQ_printf(m, "\n");
1054 }
1055 
sched_debug_show(struct seq_file * m,void * v)1056 static int sched_debug_show(struct seq_file *m, void *v)
1057 {
1058 	int cpu = (unsigned long)(v - 2);
1059 
1060 	if (cpu != -1)
1061 		print_cpu(m, cpu);
1062 	else
1063 		sched_debug_header(m);
1064 
1065 	return 0;
1066 }
1067 
sysrq_sched_debug_show(void)1068 void sysrq_sched_debug_show(void)
1069 {
1070 	int cpu;
1071 
1072 	sched_debug_header(NULL);
1073 	for_each_online_cpu(cpu) {
1074 		/*
1075 		 * Need to reset softlockup watchdogs on all CPUs, because
1076 		 * another CPU might be blocked waiting for us to process
1077 		 * an IPI or stop_machine.
1078 		 */
1079 		touch_nmi_watchdog();
1080 		touch_all_softlockup_watchdogs();
1081 		print_cpu(NULL, cpu);
1082 	}
1083 }
1084 
1085 /*
1086  * This iterator needs some explanation.
1087  * It returns 1 for the header position.
1088  * This means 2 is CPU 0.
1089  * In a hotplugged system some CPUs, including CPU 0, may be missing so we have
1090  * to use cpumask_* to iterate over the CPUs.
1091  */
sched_debug_start(struct seq_file * file,loff_t * offset)1092 static void *sched_debug_start(struct seq_file *file, loff_t *offset)
1093 {
1094 	unsigned long n = *offset;
1095 
1096 	if (n == 0)
1097 		return (void *) 1;
1098 
1099 	n--;
1100 
1101 	if (n > 0)
1102 		n = cpumask_next(n - 1, cpu_online_mask);
1103 	else
1104 		n = cpumask_first(cpu_online_mask);
1105 
1106 	*offset = n + 1;
1107 
1108 	if (n < nr_cpu_ids)
1109 		return (void *)(unsigned long)(n + 2);
1110 
1111 	return NULL;
1112 }
1113 
sched_debug_next(struct seq_file * file,void * data,loff_t * offset)1114 static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
1115 {
1116 	(*offset)++;
1117 	return sched_debug_start(file, offset);
1118 }
1119 
sched_debug_stop(struct seq_file * file,void * data)1120 static void sched_debug_stop(struct seq_file *file, void *data)
1121 {
1122 }
1123 
1124 static const struct seq_operations sched_debug_sops = {
1125 	.start		= sched_debug_start,
1126 	.next		= sched_debug_next,
1127 	.stop		= sched_debug_stop,
1128 	.show		= sched_debug_show,
1129 };
1130 
1131 #define __PS(S, F) SEQ_printf(m, "%-45s:%21Ld\n", S, (long long)(F))
1132 #define __P(F) __PS(#F, F)
1133 #define   P(F) __PS(#F, p->F)
1134 #define   PM(F, M) __PS(#F, p->F & (M))
1135 #define __PSN(S, F) SEQ_printf(m, "%-45s:%14Ld.%06ld\n", S, SPLIT_NS((long long)(F)))
1136 #define __PN(F) __PSN(#F, F)
1137 #define   PN(F) __PSN(#F, p->F)
1138 
1139 
1140 #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)1141 void print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
1142 		unsigned long tpf, unsigned long gsf, unsigned long gpf)
1143 {
1144 	SEQ_printf(m, "numa_faults node=%d ", node);
1145 	SEQ_printf(m, "task_private=%lu task_shared=%lu ", tpf, tsf);
1146 	SEQ_printf(m, "group_private=%lu group_shared=%lu\n", gpf, gsf);
1147 }
1148 #endif
1149 
1150 
sched_show_numa(struct task_struct * p,struct seq_file * m)1151 static void sched_show_numa(struct task_struct *p, struct seq_file *m)
1152 {
1153 #ifdef CONFIG_NUMA_BALANCING
1154 	if (p->mm)
1155 		P(mm->numa_scan_seq);
1156 
1157 	P(numa_pages_migrated);
1158 	P(numa_preferred_nid);
1159 	P(total_numa_faults);
1160 	SEQ_printf(m, "current_node=%d, numa_group_id=%d\n",
1161 			task_node(p), task_numa_group_id(p));
1162 	show_numa_stats(p, m);
1163 #endif
1164 }
1165 
proc_sched_show_task(struct task_struct * p,struct pid_namespace * ns,struct seq_file * m)1166 void proc_sched_show_task(struct task_struct *p, struct pid_namespace *ns,
1167 						  struct seq_file *m)
1168 {
1169 	unsigned long nr_switches;
1170 
1171 	SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr_ns(p, ns),
1172 						get_nr_threads(p));
1173 	SEQ_printf(m,
1174 		"---------------------------------------------------------"
1175 		"----------\n");
1176 
1177 #define P_SCHEDSTAT(F)  __PS(#F, schedstat_val(p->stats.F))
1178 #define PN_SCHEDSTAT(F) __PSN(#F, schedstat_val(p->stats.F))
1179 
1180 	PN(se.exec_start);
1181 	PN(se.vruntime);
1182 	PN(se.sum_exec_runtime);
1183 
1184 	nr_switches = p->nvcsw + p->nivcsw;
1185 
1186 	P(se.nr_migrations);
1187 
1188 	if (schedstat_enabled()) {
1189 		u64 avg_atom, avg_per_cpu;
1190 
1191 		PN_SCHEDSTAT(sum_sleep_runtime);
1192 		PN_SCHEDSTAT(sum_block_runtime);
1193 		PN_SCHEDSTAT(wait_start);
1194 		PN_SCHEDSTAT(sleep_start);
1195 		PN_SCHEDSTAT(block_start);
1196 		PN_SCHEDSTAT(sleep_max);
1197 		PN_SCHEDSTAT(block_max);
1198 		PN_SCHEDSTAT(exec_max);
1199 		PN_SCHEDSTAT(slice_max);
1200 		PN_SCHEDSTAT(wait_max);
1201 		PN_SCHEDSTAT(wait_sum);
1202 		P_SCHEDSTAT(wait_count);
1203 		PN_SCHEDSTAT(iowait_sum);
1204 		P_SCHEDSTAT(iowait_count);
1205 		P_SCHEDSTAT(nr_migrations_cold);
1206 		P_SCHEDSTAT(nr_failed_migrations_affine);
1207 		P_SCHEDSTAT(nr_failed_migrations_running);
1208 		P_SCHEDSTAT(nr_failed_migrations_hot);
1209 		P_SCHEDSTAT(nr_forced_migrations);
1210 		P_SCHEDSTAT(nr_wakeups);
1211 		P_SCHEDSTAT(nr_wakeups_sync);
1212 		P_SCHEDSTAT(nr_wakeups_migrate);
1213 		P_SCHEDSTAT(nr_wakeups_local);
1214 		P_SCHEDSTAT(nr_wakeups_remote);
1215 		P_SCHEDSTAT(nr_wakeups_affine);
1216 		P_SCHEDSTAT(nr_wakeups_affine_attempts);
1217 		P_SCHEDSTAT(nr_wakeups_passive);
1218 		P_SCHEDSTAT(nr_wakeups_idle);
1219 
1220 		avg_atom = p->se.sum_exec_runtime;
1221 		if (nr_switches)
1222 			avg_atom = div64_ul(avg_atom, nr_switches);
1223 		else
1224 			avg_atom = -1LL;
1225 
1226 		avg_per_cpu = p->se.sum_exec_runtime;
1227 		if (p->se.nr_migrations) {
1228 			avg_per_cpu = div64_u64(avg_per_cpu,
1229 						p->se.nr_migrations);
1230 		} else {
1231 			avg_per_cpu = -1LL;
1232 		}
1233 
1234 		__PN(avg_atom);
1235 		__PN(avg_per_cpu);
1236 
1237 #ifdef CONFIG_SCHED_CORE
1238 		PN_SCHEDSTAT(core_forceidle_sum);
1239 #endif
1240 	}
1241 
1242 	__P(nr_switches);
1243 	__PS("nr_voluntary_switches", p->nvcsw);
1244 	__PS("nr_involuntary_switches", p->nivcsw);
1245 
1246 	P(se.load.weight);
1247 #ifdef CONFIG_SMP
1248 	P(se.avg.load_sum);
1249 	P(se.avg.runnable_sum);
1250 	P(se.avg.util_sum);
1251 	P(se.avg.load_avg);
1252 	P(se.avg.runnable_avg);
1253 	P(se.avg.util_avg);
1254 	P(se.avg.last_update_time);
1255 	PM(se.avg.util_est, ~UTIL_AVG_UNCHANGED);
1256 #endif
1257 #ifdef CONFIG_UCLAMP_TASK
1258 	__PS("uclamp.min", p->uclamp_req[UCLAMP_MIN].value);
1259 	__PS("uclamp.max", p->uclamp_req[UCLAMP_MAX].value);
1260 	__PS("effective uclamp.min", uclamp_eff_value(p, UCLAMP_MIN));
1261 	__PS("effective uclamp.max", uclamp_eff_value(p, UCLAMP_MAX));
1262 #endif
1263 	P(policy);
1264 	P(prio);
1265 	if (task_has_dl_policy(p)) {
1266 		P(dl.runtime);
1267 		P(dl.deadline);
1268 	} else if (fair_policy(p->policy)) {
1269 		P(se.slice);
1270 	}
1271 #ifdef CONFIG_SCHED_CLASS_EXT
1272 	__PS("ext.enabled", task_on_scx(p));
1273 #endif
1274 #undef PN_SCHEDSTAT
1275 #undef P_SCHEDSTAT
1276 
1277 	{
1278 		unsigned int this_cpu = raw_smp_processor_id();
1279 		u64 t0, t1;
1280 
1281 		t0 = cpu_clock(this_cpu);
1282 		t1 = cpu_clock(this_cpu);
1283 		__PS("clock-delta", t1-t0);
1284 	}
1285 
1286 	sched_show_numa(p, m);
1287 }
1288 
proc_sched_set_task(struct task_struct * p)1289 void proc_sched_set_task(struct task_struct *p)
1290 {
1291 #ifdef CONFIG_SCHEDSTATS
1292 	memset(&p->stats, 0, sizeof(p->stats));
1293 #endif
1294 }
1295 
resched_latency_warn(int cpu,u64 latency)1296 void resched_latency_warn(int cpu, u64 latency)
1297 {
1298 	static DEFINE_RATELIMIT_STATE(latency_check_ratelimit, 60 * 60 * HZ, 1);
1299 
1300 	WARN(__ratelimit(&latency_check_ratelimit),
1301 	     "sched: CPU %d need_resched set for > %llu ns (%d ticks) "
1302 	     "without schedule\n",
1303 	     cpu, latency, cpu_rq(cpu)->ticks_without_resched);
1304 }
1305