xref: /linux/kernel/sched/debug.c (revision adc4fb9c814b5d5cc6021022900fd5eb0b3c8165)
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 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 	int i = IS_ENABLED(CONFIG_PREEMPT_RT) * 2;
248 	int j;
249 
250 	/* Count entries in NULL terminated preempt_modes */
251 	for (j = 0; preempt_modes[j]; j++)
252 		;
253 	j -= !IS_ENABLED(CONFIG_ARCH_HAS_PREEMPT_LAZY);
254 
255 	for (; i < j; i++) {
256 		if (preempt_dynamic_mode == i)
257 			seq_puts(m, "(");
258 		seq_puts(m, preempt_modes[i]);
259 		if (preempt_dynamic_mode == i)
260 			seq_puts(m, ")");
261 
262 		seq_puts(m, " ");
263 	}
264 
265 	seq_puts(m, "\n");
266 	return 0;
267 }
268 
sched_dynamic_open(struct inode * inode,struct file * filp)269 static int sched_dynamic_open(struct inode *inode, struct file *filp)
270 {
271 	return single_open(filp, sched_dynamic_show, NULL);
272 }
273 
274 static const struct file_operations sched_dynamic_fops = {
275 	.open		= sched_dynamic_open,
276 	.write		= sched_dynamic_write,
277 	.read		= seq_read,
278 	.llseek		= seq_lseek,
279 	.release	= single_release,
280 };
281 
282 #endif /* CONFIG_PREEMPT_DYNAMIC */
283 
284 __read_mostly bool sched_debug_verbose;
285 
286 #ifdef CONFIG_SMP
287 static struct dentry           *sd_dentry;
288 
289 
sched_verbose_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)290 static ssize_t sched_verbose_write(struct file *filp, const char __user *ubuf,
291 				  size_t cnt, loff_t *ppos)
292 {
293 	ssize_t result;
294 	bool orig;
295 
296 	cpus_read_lock();
297 	sched_domains_mutex_lock();
298 
299 	orig = sched_debug_verbose;
300 	result = debugfs_write_file_bool(filp, ubuf, cnt, ppos);
301 
302 	if (sched_debug_verbose && !orig)
303 		update_sched_domain_debugfs();
304 	else if (!sched_debug_verbose && orig) {
305 		debugfs_remove(sd_dentry);
306 		sd_dentry = NULL;
307 	}
308 
309 	sched_domains_mutex_unlock();
310 	cpus_read_unlock();
311 
312 	return result;
313 }
314 #else
315 #define sched_verbose_write debugfs_write_file_bool
316 #endif
317 
318 static const struct file_operations sched_verbose_fops = {
319 	.read =         debugfs_read_file_bool,
320 	.write =        sched_verbose_write,
321 	.open =         simple_open,
322 	.llseek =       default_llseek,
323 };
324 
325 static const struct seq_operations sched_debug_sops;
326 
sched_debug_open(struct inode * inode,struct file * filp)327 static int sched_debug_open(struct inode *inode, struct file *filp)
328 {
329 	return seq_open(filp, &sched_debug_sops);
330 }
331 
332 static const struct file_operations sched_debug_fops = {
333 	.open		= sched_debug_open,
334 	.read		= seq_read,
335 	.llseek		= seq_lseek,
336 	.release	= seq_release,
337 };
338 
339 enum dl_param {
340 	DL_RUNTIME = 0,
341 	DL_PERIOD,
342 };
343 
344 static unsigned long fair_server_period_max = (1UL << 22) * NSEC_PER_USEC; /* ~4 seconds */
345 static unsigned long fair_server_period_min = (100) * NSEC_PER_USEC;     /* 100 us */
346 
sched_fair_server_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos,enum dl_param param)347 static ssize_t sched_fair_server_write(struct file *filp, const char __user *ubuf,
348 				       size_t cnt, loff_t *ppos, enum dl_param param)
349 {
350 	long cpu = (long) ((struct seq_file *) filp->private_data)->private;
351 	struct rq *rq = cpu_rq(cpu);
352 	u64 runtime, period;
353 	size_t err;
354 	int retval;
355 	u64 value;
356 
357 	err = kstrtoull_from_user(ubuf, cnt, 10, &value);
358 	if (err)
359 		return err;
360 
361 	scoped_guard (rq_lock_irqsave, rq) {
362 		runtime  = rq->fair_server.dl_runtime;
363 		period = rq->fair_server.dl_period;
364 
365 		switch (param) {
366 		case DL_RUNTIME:
367 			if (runtime == value)
368 				break;
369 			runtime = value;
370 			break;
371 		case DL_PERIOD:
372 			if (value == period)
373 				break;
374 			period = value;
375 			break;
376 		}
377 
378 		if (runtime > period ||
379 		    period > fair_server_period_max ||
380 		    period < fair_server_period_min) {
381 			return  -EINVAL;
382 		}
383 
384 		if (rq->cfs.h_nr_queued) {
385 			update_rq_clock(rq);
386 			dl_server_stop(&rq->fair_server);
387 		}
388 
389 		retval = dl_server_apply_params(&rq->fair_server, runtime, period, 0);
390 		if (retval)
391 			cnt = retval;
392 
393 		if (!runtime)
394 			printk_deferred("Fair server disabled in CPU %d, system may crash due to starvation.\n",
395 					cpu_of(rq));
396 
397 		if (rq->cfs.h_nr_queued)
398 			dl_server_start(&rq->fair_server);
399 	}
400 
401 	*ppos += cnt;
402 	return cnt;
403 }
404 
sched_fair_server_show(struct seq_file * m,void * v,enum dl_param param)405 static size_t sched_fair_server_show(struct seq_file *m, void *v, enum dl_param param)
406 {
407 	unsigned long cpu = (unsigned long) m->private;
408 	struct rq *rq = cpu_rq(cpu);
409 	u64 value;
410 
411 	switch (param) {
412 	case DL_RUNTIME:
413 		value = rq->fair_server.dl_runtime;
414 		break;
415 	case DL_PERIOD:
416 		value = rq->fair_server.dl_period;
417 		break;
418 	}
419 
420 	seq_printf(m, "%llu\n", value);
421 	return 0;
422 
423 }
424 
425 static ssize_t
sched_fair_server_runtime_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)426 sched_fair_server_runtime_write(struct file *filp, const char __user *ubuf,
427 				size_t cnt, loff_t *ppos)
428 {
429 	return sched_fair_server_write(filp, ubuf, cnt, ppos, DL_RUNTIME);
430 }
431 
sched_fair_server_runtime_show(struct seq_file * m,void * v)432 static int sched_fair_server_runtime_show(struct seq_file *m, void *v)
433 {
434 	return sched_fair_server_show(m, v, DL_RUNTIME);
435 }
436 
sched_fair_server_runtime_open(struct inode * inode,struct file * filp)437 static int sched_fair_server_runtime_open(struct inode *inode, struct file *filp)
438 {
439 	return single_open(filp, sched_fair_server_runtime_show, inode->i_private);
440 }
441 
442 static const struct file_operations fair_server_runtime_fops = {
443 	.open		= sched_fair_server_runtime_open,
444 	.write		= sched_fair_server_runtime_write,
445 	.read		= seq_read,
446 	.llseek		= seq_lseek,
447 	.release	= single_release,
448 };
449 
450 static ssize_t
sched_fair_server_period_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)451 sched_fair_server_period_write(struct file *filp, const char __user *ubuf,
452 			       size_t cnt, loff_t *ppos)
453 {
454 	return sched_fair_server_write(filp, ubuf, cnt, ppos, DL_PERIOD);
455 }
456 
sched_fair_server_period_show(struct seq_file * m,void * v)457 static int sched_fair_server_period_show(struct seq_file *m, void *v)
458 {
459 	return sched_fair_server_show(m, v, DL_PERIOD);
460 }
461 
sched_fair_server_period_open(struct inode * inode,struct file * filp)462 static int sched_fair_server_period_open(struct inode *inode, struct file *filp)
463 {
464 	return single_open(filp, sched_fair_server_period_show, inode->i_private);
465 }
466 
467 static const struct file_operations fair_server_period_fops = {
468 	.open		= sched_fair_server_period_open,
469 	.write		= sched_fair_server_period_write,
470 	.read		= seq_read,
471 	.llseek		= seq_lseek,
472 	.release	= single_release,
473 };
474 
475 static struct dentry *debugfs_sched;
476 
debugfs_fair_server_init(void)477 static void debugfs_fair_server_init(void)
478 {
479 	struct dentry *d_fair;
480 	unsigned long cpu;
481 
482 	d_fair = debugfs_create_dir("fair_server", debugfs_sched);
483 	if (!d_fair)
484 		return;
485 
486 	for_each_possible_cpu(cpu) {
487 		struct dentry *d_cpu;
488 		char buf[32];
489 
490 		snprintf(buf, sizeof(buf), "cpu%lu", cpu);
491 		d_cpu = debugfs_create_dir(buf, d_fair);
492 
493 		debugfs_create_file("runtime", 0644, d_cpu, (void *) cpu, &fair_server_runtime_fops);
494 		debugfs_create_file("period", 0644, d_cpu, (void *) cpu, &fair_server_period_fops);
495 	}
496 }
497 
sched_init_debug(void)498 static __init int sched_init_debug(void)
499 {
500 	struct dentry __maybe_unused *numa;
501 
502 	debugfs_sched = debugfs_create_dir("sched", NULL);
503 
504 	debugfs_create_file("features", 0644, debugfs_sched, NULL, &sched_feat_fops);
505 	debugfs_create_file_unsafe("verbose", 0644, debugfs_sched, &sched_debug_verbose, &sched_verbose_fops);
506 #ifdef CONFIG_PREEMPT_DYNAMIC
507 	debugfs_create_file("preempt", 0644, debugfs_sched, NULL, &sched_dynamic_fops);
508 #endif
509 
510 	debugfs_create_u32("base_slice_ns", 0644, debugfs_sched, &sysctl_sched_base_slice);
511 
512 	debugfs_create_u32("latency_warn_ms", 0644, debugfs_sched, &sysctl_resched_latency_warn_ms);
513 	debugfs_create_u32("latency_warn_once", 0644, debugfs_sched, &sysctl_resched_latency_warn_once);
514 
515 #ifdef CONFIG_SMP
516 	debugfs_create_file("tunable_scaling", 0644, debugfs_sched, NULL, &sched_scaling_fops);
517 	debugfs_create_u32("migration_cost_ns", 0644, debugfs_sched, &sysctl_sched_migration_cost);
518 	debugfs_create_u32("nr_migrate", 0644, debugfs_sched, &sysctl_sched_nr_migrate);
519 
520 	sched_domains_mutex_lock();
521 	update_sched_domain_debugfs();
522 	sched_domains_mutex_unlock();
523 #endif
524 
525 #ifdef CONFIG_NUMA_BALANCING
526 	numa = debugfs_create_dir("numa_balancing", debugfs_sched);
527 
528 	debugfs_create_u32("scan_delay_ms", 0644, numa, &sysctl_numa_balancing_scan_delay);
529 	debugfs_create_u32("scan_period_min_ms", 0644, numa, &sysctl_numa_balancing_scan_period_min);
530 	debugfs_create_u32("scan_period_max_ms", 0644, numa, &sysctl_numa_balancing_scan_period_max);
531 	debugfs_create_u32("scan_size_mb", 0644, numa, &sysctl_numa_balancing_scan_size);
532 	debugfs_create_u32("hot_threshold_ms", 0644, numa, &sysctl_numa_balancing_hot_threshold);
533 #endif
534 
535 	debugfs_create_file("debug", 0444, debugfs_sched, NULL, &sched_debug_fops);
536 
537 	debugfs_fair_server_init();
538 
539 	return 0;
540 }
541 late_initcall(sched_init_debug);
542 
543 #ifdef CONFIG_SMP
544 
545 static cpumask_var_t		sd_sysctl_cpus;
546 
sd_flags_show(struct seq_file * m,void * v)547 static int sd_flags_show(struct seq_file *m, void *v)
548 {
549 	unsigned long flags = *(unsigned int *)m->private;
550 	int idx;
551 
552 	for_each_set_bit(idx, &flags, __SD_FLAG_CNT) {
553 		seq_puts(m, sd_flag_debug[idx].name);
554 		seq_puts(m, " ");
555 	}
556 	seq_puts(m, "\n");
557 
558 	return 0;
559 }
560 
sd_flags_open(struct inode * inode,struct file * file)561 static int sd_flags_open(struct inode *inode, struct file *file)
562 {
563 	return single_open(file, sd_flags_show, inode->i_private);
564 }
565 
566 static const struct file_operations sd_flags_fops = {
567 	.open		= sd_flags_open,
568 	.read		= seq_read,
569 	.llseek		= seq_lseek,
570 	.release	= single_release,
571 };
572 
register_sd(struct sched_domain * sd,struct dentry * parent)573 static void register_sd(struct sched_domain *sd, struct dentry *parent)
574 {
575 #define SDM(type, mode, member)	\
576 	debugfs_create_##type(#member, mode, parent, &sd->member)
577 
578 	SDM(ulong, 0644, min_interval);
579 	SDM(ulong, 0644, max_interval);
580 	SDM(u64,   0644, max_newidle_lb_cost);
581 	SDM(u32,   0644, busy_factor);
582 	SDM(u32,   0644, imbalance_pct);
583 	SDM(u32,   0644, cache_nice_tries);
584 	SDM(str,   0444, name);
585 
586 #undef SDM
587 
588 	debugfs_create_file("flags", 0444, parent, &sd->flags, &sd_flags_fops);
589 	debugfs_create_file("groups_flags", 0444, parent, &sd->groups->flags, &sd_flags_fops);
590 	debugfs_create_u32("level", 0444, parent, (u32 *)&sd->level);
591 }
592 
update_sched_domain_debugfs(void)593 void update_sched_domain_debugfs(void)
594 {
595 	int cpu, i;
596 
597 	/*
598 	 * This can unfortunately be invoked before sched_debug_init() creates
599 	 * the debug directory. Don't touch sd_sysctl_cpus until then.
600 	 */
601 	if (!debugfs_sched)
602 		return;
603 
604 	if (!sched_debug_verbose)
605 		return;
606 
607 	if (!cpumask_available(sd_sysctl_cpus)) {
608 		if (!alloc_cpumask_var(&sd_sysctl_cpus, GFP_KERNEL))
609 			return;
610 		cpumask_copy(sd_sysctl_cpus, cpu_possible_mask);
611 	}
612 
613 	if (!sd_dentry) {
614 		sd_dentry = debugfs_create_dir("domains", debugfs_sched);
615 
616 		/* rebuild sd_sysctl_cpus if empty since it gets cleared below */
617 		if (cpumask_empty(sd_sysctl_cpus))
618 			cpumask_copy(sd_sysctl_cpus, cpu_online_mask);
619 	}
620 
621 	for_each_cpu(cpu, sd_sysctl_cpus) {
622 		struct sched_domain *sd;
623 		struct dentry *d_cpu;
624 		char buf[32];
625 
626 		snprintf(buf, sizeof(buf), "cpu%d", cpu);
627 		debugfs_lookup_and_remove(buf, sd_dentry);
628 		d_cpu = debugfs_create_dir(buf, sd_dentry);
629 
630 		i = 0;
631 		for_each_domain(cpu, sd) {
632 			struct dentry *d_sd;
633 
634 			snprintf(buf, sizeof(buf), "domain%d", i);
635 			d_sd = debugfs_create_dir(buf, d_cpu);
636 
637 			register_sd(sd, d_sd);
638 			i++;
639 		}
640 
641 		__cpumask_clear_cpu(cpu, sd_sysctl_cpus);
642 	}
643 }
644 
dirty_sched_domain_sysctl(int cpu)645 void dirty_sched_domain_sysctl(int cpu)
646 {
647 	if (cpumask_available(sd_sysctl_cpus))
648 		__cpumask_set_cpu(cpu, sd_sysctl_cpus);
649 }
650 
651 #endif /* CONFIG_SMP */
652 
653 #ifdef CONFIG_FAIR_GROUP_SCHED
print_cfs_group_stats(struct seq_file * m,int cpu,struct task_group * tg)654 static void print_cfs_group_stats(struct seq_file *m, int cpu, struct task_group *tg)
655 {
656 	struct sched_entity *se = tg->se[cpu];
657 
658 #define P(F)		SEQ_printf(m, "  .%-30s: %lld\n",	#F, (long long)F)
659 #define P_SCHEDSTAT(F)	SEQ_printf(m, "  .%-30s: %lld\n",	\
660 		#F, (long long)schedstat_val(stats->F))
661 #define PN(F)		SEQ_printf(m, "  .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F))
662 #define PN_SCHEDSTAT(F)	SEQ_printf(m, "  .%-30s: %lld.%06ld\n", \
663 		#F, SPLIT_NS((long long)schedstat_val(stats->F)))
664 
665 	if (!se)
666 		return;
667 
668 	PN(se->exec_start);
669 	PN(se->vruntime);
670 	PN(se->sum_exec_runtime);
671 
672 	if (schedstat_enabled()) {
673 		struct sched_statistics *stats;
674 		stats = __schedstats_from_se(se);
675 
676 		PN_SCHEDSTAT(wait_start);
677 		PN_SCHEDSTAT(sleep_start);
678 		PN_SCHEDSTAT(block_start);
679 		PN_SCHEDSTAT(sleep_max);
680 		PN_SCHEDSTAT(block_max);
681 		PN_SCHEDSTAT(exec_max);
682 		PN_SCHEDSTAT(slice_max);
683 		PN_SCHEDSTAT(wait_max);
684 		PN_SCHEDSTAT(wait_sum);
685 		P_SCHEDSTAT(wait_count);
686 	}
687 
688 	P(se->load.weight);
689 #ifdef CONFIG_SMP
690 	P(se->avg.load_avg);
691 	P(se->avg.util_avg);
692 	P(se->avg.runnable_avg);
693 #endif
694 
695 #undef PN_SCHEDSTAT
696 #undef PN
697 #undef P_SCHEDSTAT
698 #undef P
699 }
700 #endif
701 
702 #ifdef CONFIG_CGROUP_SCHED
703 static DEFINE_SPINLOCK(sched_debug_lock);
704 static char group_path[PATH_MAX];
705 
task_group_path(struct task_group * tg,char * path,int plen)706 static void task_group_path(struct task_group *tg, char *path, int plen)
707 {
708 	if (autogroup_path(tg, path, plen))
709 		return;
710 
711 	cgroup_path(tg->css.cgroup, path, plen);
712 }
713 
714 /*
715  * Only 1 SEQ_printf_task_group_path() caller can use the full length
716  * group_path[] for cgroup path. Other simultaneous callers will have
717  * to use a shorter stack buffer. A "..." suffix is appended at the end
718  * of the stack buffer so that it will show up in case the output length
719  * matches the given buffer size to indicate possible path name truncation.
720  */
721 #define SEQ_printf_task_group_path(m, tg, fmt...)			\
722 {									\
723 	if (spin_trylock(&sched_debug_lock)) {				\
724 		task_group_path(tg, group_path, sizeof(group_path));	\
725 		SEQ_printf(m, fmt, group_path);				\
726 		spin_unlock(&sched_debug_lock);				\
727 	} else {							\
728 		char buf[128];						\
729 		char *bufend = buf + sizeof(buf) - 3;			\
730 		task_group_path(tg, buf, bufend - buf);			\
731 		strcpy(bufend - 1, "...");				\
732 		SEQ_printf(m, fmt, buf);				\
733 	}								\
734 }
735 #endif
736 
737 static void
print_task(struct seq_file * m,struct rq * rq,struct task_struct * p)738 print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
739 {
740 	if (task_current(rq, p))
741 		SEQ_printf(m, ">R");
742 	else
743 		SEQ_printf(m, " %c", task_state_to_char(p));
744 
745 	SEQ_printf(m, " %15s %5d %9Ld.%06ld   %c   %9Ld.%06ld %c %9Ld.%06ld %9Ld.%06ld %9Ld   %5d ",
746 		p->comm, task_pid_nr(p),
747 		SPLIT_NS(p->se.vruntime),
748 		entity_eligible(cfs_rq_of(&p->se), &p->se) ? 'E' : 'N',
749 		SPLIT_NS(p->se.deadline),
750 		p->se.custom_slice ? 'S' : ' ',
751 		SPLIT_NS(p->se.slice),
752 		SPLIT_NS(p->se.sum_exec_runtime),
753 		(long long)(p->nvcsw + p->nivcsw),
754 		p->prio);
755 
756 	SEQ_printf(m, "%9lld.%06ld %9lld.%06ld %9lld.%06ld",
757 		SPLIT_NS(schedstat_val_or_zero(p->stats.wait_sum)),
758 		SPLIT_NS(schedstat_val_or_zero(p->stats.sum_sleep_runtime)),
759 		SPLIT_NS(schedstat_val_or_zero(p->stats.sum_block_runtime)));
760 
761 #ifdef CONFIG_NUMA_BALANCING
762 	SEQ_printf(m, "   %d      %d", task_node(p), task_numa_group_id(p));
763 #endif
764 #ifdef CONFIG_CGROUP_SCHED
765 	SEQ_printf_task_group_path(m, task_group(p), "        %s")
766 #endif
767 
768 	SEQ_printf(m, "\n");
769 }
770 
print_rq(struct seq_file * m,struct rq * rq,int rq_cpu)771 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
772 {
773 	struct task_struct *g, *p;
774 
775 	SEQ_printf(m, "\n");
776 	SEQ_printf(m, "runnable tasks:\n");
777 	SEQ_printf(m, " S            task   PID       vruntime   eligible    "
778 		   "deadline             slice          sum-exec      switches  "
779 		   "prio         wait-time        sum-sleep       sum-block"
780 #ifdef CONFIG_NUMA_BALANCING
781 		   "  node   group-id"
782 #endif
783 #ifdef CONFIG_CGROUP_SCHED
784 		   "  group-path"
785 #endif
786 		   "\n");
787 	SEQ_printf(m, "-------------------------------------------------------"
788 		   "------------------------------------------------------"
789 		   "------------------------------------------------------"
790 #ifdef CONFIG_NUMA_BALANCING
791 		   "--------------"
792 #endif
793 #ifdef CONFIG_CGROUP_SCHED
794 		   "--------------"
795 #endif
796 		   "\n");
797 
798 	rcu_read_lock();
799 	for_each_process_thread(g, p) {
800 		if (task_cpu(p) != rq_cpu)
801 			continue;
802 
803 		print_task(m, rq, p);
804 	}
805 	rcu_read_unlock();
806 }
807 
print_cfs_rq(struct seq_file * m,int cpu,struct cfs_rq * cfs_rq)808 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
809 {
810 	s64 left_vruntime = -1, min_vruntime, right_vruntime = -1, left_deadline = -1, spread;
811 	struct sched_entity *last, *first, *root;
812 	struct rq *rq = cpu_rq(cpu);
813 	unsigned long flags;
814 
815 #ifdef CONFIG_FAIR_GROUP_SCHED
816 	SEQ_printf(m, "\n");
817 	SEQ_printf_task_group_path(m, cfs_rq->tg, "cfs_rq[%d]:%s\n", cpu);
818 #else
819 	SEQ_printf(m, "\n");
820 	SEQ_printf(m, "cfs_rq[%d]:\n", cpu);
821 #endif
822 
823 	raw_spin_rq_lock_irqsave(rq, flags);
824 	root = __pick_root_entity(cfs_rq);
825 	if (root)
826 		left_vruntime = root->min_vruntime;
827 	first = __pick_first_entity(cfs_rq);
828 	if (first)
829 		left_deadline = first->deadline;
830 	last = __pick_last_entity(cfs_rq);
831 	if (last)
832 		right_vruntime = last->vruntime;
833 	min_vruntime = cfs_rq->min_vruntime;
834 	raw_spin_rq_unlock_irqrestore(rq, flags);
835 
836 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "left_deadline",
837 			SPLIT_NS(left_deadline));
838 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "left_vruntime",
839 			SPLIT_NS(left_vruntime));
840 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "min_vruntime",
841 			SPLIT_NS(min_vruntime));
842 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "avg_vruntime",
843 			SPLIT_NS(avg_vruntime(cfs_rq)));
844 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "right_vruntime",
845 			SPLIT_NS(right_vruntime));
846 	spread = right_vruntime - left_vruntime;
847 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread", SPLIT_NS(spread));
848 	SEQ_printf(m, "  .%-30s: %d\n", "nr_queued", cfs_rq->nr_queued);
849 	SEQ_printf(m, "  .%-30s: %d\n", "h_nr_runnable", cfs_rq->h_nr_runnable);
850 	SEQ_printf(m, "  .%-30s: %d\n", "h_nr_queued", cfs_rq->h_nr_queued);
851 	SEQ_printf(m, "  .%-30s: %d\n", "h_nr_idle", cfs_rq->h_nr_idle);
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_features);
1045 #undef PN
1046 #undef P
1047 
1048 	SEQ_printf(m, "  .%-40s: %d (%s)\n",
1049 		"sysctl_sched_tunable_scaling",
1050 		sysctl_sched_tunable_scaling,
1051 		sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
1052 	SEQ_printf(m, "\n");
1053 }
1054 
sched_debug_show(struct seq_file * m,void * v)1055 static int sched_debug_show(struct seq_file *m, void *v)
1056 {
1057 	int cpu = (unsigned long)(v - 2);
1058 
1059 	if (cpu != -1)
1060 		print_cpu(m, cpu);
1061 	else
1062 		sched_debug_header(m);
1063 
1064 	return 0;
1065 }
1066 
sysrq_sched_debug_show(void)1067 void sysrq_sched_debug_show(void)
1068 {
1069 	int cpu;
1070 
1071 	sched_debug_header(NULL);
1072 	for_each_online_cpu(cpu) {
1073 		/*
1074 		 * Need to reset softlockup watchdogs on all CPUs, because
1075 		 * another CPU might be blocked waiting for us to process
1076 		 * an IPI or stop_machine.
1077 		 */
1078 		touch_nmi_watchdog();
1079 		touch_all_softlockup_watchdogs();
1080 		print_cpu(NULL, cpu);
1081 	}
1082 }
1083 
1084 /*
1085  * This iterator needs some explanation.
1086  * It returns 1 for the header position.
1087  * This means 2 is CPU 0.
1088  * In a hotplugged system some CPUs, including CPU 0, may be missing so we have
1089  * to use cpumask_* to iterate over the CPUs.
1090  */
sched_debug_start(struct seq_file * file,loff_t * offset)1091 static void *sched_debug_start(struct seq_file *file, loff_t *offset)
1092 {
1093 	unsigned long n = *offset;
1094 
1095 	if (n == 0)
1096 		return (void *) 1;
1097 
1098 	n--;
1099 
1100 	if (n > 0)
1101 		n = cpumask_next(n - 1, cpu_online_mask);
1102 	else
1103 		n = cpumask_first(cpu_online_mask);
1104 
1105 	*offset = n + 1;
1106 
1107 	if (n < nr_cpu_ids)
1108 		return (void *)(unsigned long)(n + 2);
1109 
1110 	return NULL;
1111 }
1112 
sched_debug_next(struct seq_file * file,void * data,loff_t * offset)1113 static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
1114 {
1115 	(*offset)++;
1116 	return sched_debug_start(file, offset);
1117 }
1118 
sched_debug_stop(struct seq_file * file,void * data)1119 static void sched_debug_stop(struct seq_file *file, void *data)
1120 {
1121 }
1122 
1123 static const struct seq_operations sched_debug_sops = {
1124 	.start		= sched_debug_start,
1125 	.next		= sched_debug_next,
1126 	.stop		= sched_debug_stop,
1127 	.show		= sched_debug_show,
1128 };
1129 
1130 #define __PS(S, F) SEQ_printf(m, "%-45s:%21Ld\n", S, (long long)(F))
1131 #define __P(F) __PS(#F, F)
1132 #define   P(F) __PS(#F, p->F)
1133 #define   PM(F, M) __PS(#F, p->F & (M))
1134 #define __PSN(S, F) SEQ_printf(m, "%-45s:%14Ld.%06ld\n", S, SPLIT_NS((long long)(F)))
1135 #define __PN(F) __PSN(#F, F)
1136 #define   PN(F) __PSN(#F, p->F)
1137 
1138 
1139 #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)1140 void print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
1141 		unsigned long tpf, unsigned long gsf, unsigned long gpf)
1142 {
1143 	SEQ_printf(m, "numa_faults node=%d ", node);
1144 	SEQ_printf(m, "task_private=%lu task_shared=%lu ", tpf, tsf);
1145 	SEQ_printf(m, "group_private=%lu group_shared=%lu\n", gpf, gsf);
1146 }
1147 #endif
1148 
1149 
sched_show_numa(struct task_struct * p,struct seq_file * m)1150 static void sched_show_numa(struct task_struct *p, struct seq_file *m)
1151 {
1152 #ifdef CONFIG_NUMA_BALANCING
1153 	if (p->mm)
1154 		P(mm->numa_scan_seq);
1155 
1156 	P(numa_pages_migrated);
1157 	P(numa_preferred_nid);
1158 	P(total_numa_faults);
1159 	SEQ_printf(m, "current_node=%d, numa_group_id=%d\n",
1160 			task_node(p), task_numa_group_id(p));
1161 	show_numa_stats(p, m);
1162 #endif
1163 }
1164 
proc_sched_show_task(struct task_struct * p,struct pid_namespace * ns,struct seq_file * m)1165 void proc_sched_show_task(struct task_struct *p, struct pid_namespace *ns,
1166 						  struct seq_file *m)
1167 {
1168 	unsigned long nr_switches;
1169 
1170 	SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr_ns(p, ns),
1171 						get_nr_threads(p));
1172 	SEQ_printf(m,
1173 		"---------------------------------------------------------"
1174 		"----------\n");
1175 
1176 #define P_SCHEDSTAT(F)  __PS(#F, schedstat_val(p->stats.F))
1177 #define PN_SCHEDSTAT(F) __PSN(#F, schedstat_val(p->stats.F))
1178 
1179 	PN(se.exec_start);
1180 	PN(se.vruntime);
1181 	PN(se.sum_exec_runtime);
1182 
1183 	nr_switches = p->nvcsw + p->nivcsw;
1184 
1185 	P(se.nr_migrations);
1186 
1187 	if (schedstat_enabled()) {
1188 		u64 avg_atom, avg_per_cpu;
1189 
1190 		PN_SCHEDSTAT(sum_sleep_runtime);
1191 		PN_SCHEDSTAT(sum_block_runtime);
1192 		PN_SCHEDSTAT(wait_start);
1193 		PN_SCHEDSTAT(sleep_start);
1194 		PN_SCHEDSTAT(block_start);
1195 		PN_SCHEDSTAT(sleep_max);
1196 		PN_SCHEDSTAT(block_max);
1197 		PN_SCHEDSTAT(exec_max);
1198 		PN_SCHEDSTAT(slice_max);
1199 		PN_SCHEDSTAT(wait_max);
1200 		PN_SCHEDSTAT(wait_sum);
1201 		P_SCHEDSTAT(wait_count);
1202 		PN_SCHEDSTAT(iowait_sum);
1203 		P_SCHEDSTAT(iowait_count);
1204 		P_SCHEDSTAT(nr_migrations_cold);
1205 		P_SCHEDSTAT(nr_failed_migrations_affine);
1206 		P_SCHEDSTAT(nr_failed_migrations_running);
1207 		P_SCHEDSTAT(nr_failed_migrations_hot);
1208 		P_SCHEDSTAT(nr_forced_migrations);
1209 		P_SCHEDSTAT(nr_wakeups);
1210 		P_SCHEDSTAT(nr_wakeups_sync);
1211 		P_SCHEDSTAT(nr_wakeups_migrate);
1212 		P_SCHEDSTAT(nr_wakeups_local);
1213 		P_SCHEDSTAT(nr_wakeups_remote);
1214 		P_SCHEDSTAT(nr_wakeups_affine);
1215 		P_SCHEDSTAT(nr_wakeups_affine_attempts);
1216 		P_SCHEDSTAT(nr_wakeups_passive);
1217 		P_SCHEDSTAT(nr_wakeups_idle);
1218 
1219 		avg_atom = p->se.sum_exec_runtime;
1220 		if (nr_switches)
1221 			avg_atom = div64_ul(avg_atom, nr_switches);
1222 		else
1223 			avg_atom = -1LL;
1224 
1225 		avg_per_cpu = p->se.sum_exec_runtime;
1226 		if (p->se.nr_migrations) {
1227 			avg_per_cpu = div64_u64(avg_per_cpu,
1228 						p->se.nr_migrations);
1229 		} else {
1230 			avg_per_cpu = -1LL;
1231 		}
1232 
1233 		__PN(avg_atom);
1234 		__PN(avg_per_cpu);
1235 
1236 #ifdef CONFIG_SCHED_CORE
1237 		PN_SCHEDSTAT(core_forceidle_sum);
1238 #endif
1239 	}
1240 
1241 	__P(nr_switches);
1242 	__PS("nr_voluntary_switches", p->nvcsw);
1243 	__PS("nr_involuntary_switches", p->nivcsw);
1244 
1245 	P(se.load.weight);
1246 #ifdef CONFIG_SMP
1247 	P(se.avg.load_sum);
1248 	P(se.avg.runnable_sum);
1249 	P(se.avg.util_sum);
1250 	P(se.avg.load_avg);
1251 	P(se.avg.runnable_avg);
1252 	P(se.avg.util_avg);
1253 	P(se.avg.last_update_time);
1254 	PM(se.avg.util_est, ~UTIL_AVG_UNCHANGED);
1255 #endif
1256 #ifdef CONFIG_UCLAMP_TASK
1257 	__PS("uclamp.min", p->uclamp_req[UCLAMP_MIN].value);
1258 	__PS("uclamp.max", p->uclamp_req[UCLAMP_MAX].value);
1259 	__PS("effective uclamp.min", uclamp_eff_value(p, UCLAMP_MIN));
1260 	__PS("effective uclamp.max", uclamp_eff_value(p, UCLAMP_MAX));
1261 #endif
1262 	P(policy);
1263 	P(prio);
1264 	if (task_has_dl_policy(p)) {
1265 		P(dl.runtime);
1266 		P(dl.deadline);
1267 	} else if (fair_policy(p->policy)) {
1268 		P(se.slice);
1269 	}
1270 #ifdef CONFIG_SCHED_CLASS_EXT
1271 	__PS("ext.enabled", task_on_scx(p));
1272 #endif
1273 #undef PN_SCHEDSTAT
1274 #undef P_SCHEDSTAT
1275 
1276 	{
1277 		unsigned int this_cpu = raw_smp_processor_id();
1278 		u64 t0, t1;
1279 
1280 		t0 = cpu_clock(this_cpu);
1281 		t1 = cpu_clock(this_cpu);
1282 		__PS("clock-delta", t1-t0);
1283 	}
1284 
1285 	sched_show_numa(p, m);
1286 }
1287 
proc_sched_set_task(struct task_struct * p)1288 void proc_sched_set_task(struct task_struct *p)
1289 {
1290 #ifdef CONFIG_SCHEDSTATS
1291 	memset(&p->stats, 0, sizeof(p->stats));
1292 #endif
1293 }
1294 
resched_latency_warn(int cpu,u64 latency)1295 void resched_latency_warn(int cpu, u64 latency)
1296 {
1297 	static DEFINE_RATELIMIT_STATE(latency_check_ratelimit, 60 * 60 * HZ, 1);
1298 
1299 	if (likely(!__ratelimit(&latency_check_ratelimit)))
1300 		return;
1301 
1302 	pr_err("sched: CPU %d need_resched set for > %llu ns (%d ticks) without schedule\n",
1303 	       cpu, latency, cpu_rq(cpu)->ticks_without_resched);
1304 	dump_stack();
1305 }
1306