xref: /openbmc/linux/kernel/sched/debug.c (revision 2fa49589)
1 /*
2  * kernel/sched/debug.c
3  *
4  * Print the CFS rbtree and other debugging details
5  *
6  * Copyright(C) 2007, Red Hat, Inc., Ingo Molnar
7  *
8  * This program is free software; you can redistribute it and/or modify
9  * it under the terms of the GNU General Public License version 2 as
10  * published by the Free Software Foundation.
11  */
12 #include "sched.h"
13 
14 static DEFINE_SPINLOCK(sched_debug_lock);
15 
16 /*
17  * This allows printing both to /proc/sched_debug and
18  * to the console
19  */
20 #define SEQ_printf(m, x...)			\
21  do {						\
22 	if (m)					\
23 		seq_printf(m, x);		\
24 	else					\
25 		pr_cont(x);			\
26  } while (0)
27 
28 /*
29  * Ease the printing of nsec fields:
30  */
31 static long long nsec_high(unsigned long long nsec)
32 {
33 	if ((long long)nsec < 0) {
34 		nsec = -nsec;
35 		do_div(nsec, 1000000);
36 		return -nsec;
37 	}
38 	do_div(nsec, 1000000);
39 
40 	return nsec;
41 }
42 
43 static unsigned long nsec_low(unsigned long long nsec)
44 {
45 	if ((long long)nsec < 0)
46 		nsec = -nsec;
47 
48 	return do_div(nsec, 1000000);
49 }
50 
51 #define SPLIT_NS(x) nsec_high(x), nsec_low(x)
52 
53 #define SCHED_FEAT(name, enabled)	\
54 	#name ,
55 
56 static const char * const sched_feat_names[] = {
57 #include "features.h"
58 };
59 
60 #undef SCHED_FEAT
61 
62 static int sched_feat_show(struct seq_file *m, void *v)
63 {
64 	int i;
65 
66 	for (i = 0; i < __SCHED_FEAT_NR; i++) {
67 		if (!(sysctl_sched_features & (1UL << i)))
68 			seq_puts(m, "NO_");
69 		seq_printf(m, "%s ", sched_feat_names[i]);
70 	}
71 	seq_puts(m, "\n");
72 
73 	return 0;
74 }
75 
76 #ifdef CONFIG_JUMP_LABEL
77 
78 #define jump_label_key__true  STATIC_KEY_INIT_TRUE
79 #define jump_label_key__false STATIC_KEY_INIT_FALSE
80 
81 #define SCHED_FEAT(name, enabled)	\
82 	jump_label_key__##enabled ,
83 
84 struct static_key sched_feat_keys[__SCHED_FEAT_NR] = {
85 #include "features.h"
86 };
87 
88 #undef SCHED_FEAT
89 
90 static void sched_feat_disable(int i)
91 {
92 	static_key_disable_cpuslocked(&sched_feat_keys[i]);
93 }
94 
95 static void sched_feat_enable(int i)
96 {
97 	static_key_enable_cpuslocked(&sched_feat_keys[i]);
98 }
99 #else
100 static void sched_feat_disable(int i) { };
101 static void sched_feat_enable(int i) { };
102 #endif /* CONFIG_JUMP_LABEL */
103 
104 static int sched_feat_set(char *cmp)
105 {
106 	int i;
107 	int neg = 0;
108 
109 	if (strncmp(cmp, "NO_", 3) == 0) {
110 		neg = 1;
111 		cmp += 3;
112 	}
113 
114 	i = match_string(sched_feat_names, __SCHED_FEAT_NR, cmp);
115 	if (i < 0)
116 		return i;
117 
118 	if (neg) {
119 		sysctl_sched_features &= ~(1UL << i);
120 		sched_feat_disable(i);
121 	} else {
122 		sysctl_sched_features |= (1UL << i);
123 		sched_feat_enable(i);
124 	}
125 
126 	return 0;
127 }
128 
129 static ssize_t
130 sched_feat_write(struct file *filp, const char __user *ubuf,
131 		size_t cnt, loff_t *ppos)
132 {
133 	char buf[64];
134 	char *cmp;
135 	int ret;
136 	struct inode *inode;
137 
138 	if (cnt > 63)
139 		cnt = 63;
140 
141 	if (copy_from_user(&buf, ubuf, cnt))
142 		return -EFAULT;
143 
144 	buf[cnt] = 0;
145 	cmp = strstrip(buf);
146 
147 	/* Ensure the static_key remains in a consistent state */
148 	inode = file_inode(filp);
149 	cpus_read_lock();
150 	inode_lock(inode);
151 	ret = sched_feat_set(cmp);
152 	inode_unlock(inode);
153 	cpus_read_unlock();
154 	if (ret < 0)
155 		return ret;
156 
157 	*ppos += cnt;
158 
159 	return cnt;
160 }
161 
162 static int sched_feat_open(struct inode *inode, struct file *filp)
163 {
164 	return single_open(filp, sched_feat_show, NULL);
165 }
166 
167 static const struct file_operations sched_feat_fops = {
168 	.open		= sched_feat_open,
169 	.write		= sched_feat_write,
170 	.read		= seq_read,
171 	.llseek		= seq_lseek,
172 	.release	= single_release,
173 };
174 
175 __read_mostly bool sched_debug_enabled;
176 
177 static __init int sched_init_debug(void)
178 {
179 	debugfs_create_file("sched_features", 0644, NULL, NULL,
180 			&sched_feat_fops);
181 
182 	debugfs_create_bool("sched_debug", 0644, NULL,
183 			&sched_debug_enabled);
184 
185 	return 0;
186 }
187 late_initcall(sched_init_debug);
188 
189 #ifdef CONFIG_SMP
190 
191 #ifdef CONFIG_SYSCTL
192 
193 static struct ctl_table sd_ctl_dir[] = {
194 	{
195 		.procname	= "sched_domain",
196 		.mode		= 0555,
197 	},
198 	{}
199 };
200 
201 static struct ctl_table sd_ctl_root[] = {
202 	{
203 		.procname	= "kernel",
204 		.mode		= 0555,
205 		.child		= sd_ctl_dir,
206 	},
207 	{}
208 };
209 
210 static struct ctl_table *sd_alloc_ctl_entry(int n)
211 {
212 	struct ctl_table *entry =
213 		kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL);
214 
215 	return entry;
216 }
217 
218 static void sd_free_ctl_entry(struct ctl_table **tablep)
219 {
220 	struct ctl_table *entry;
221 
222 	/*
223 	 * In the intermediate directories, both the child directory and
224 	 * procname are dynamically allocated and could fail but the mode
225 	 * will always be set. In the lowest directory the names are
226 	 * static strings and all have proc handlers.
227 	 */
228 	for (entry = *tablep; entry->mode; entry++) {
229 		if (entry->child)
230 			sd_free_ctl_entry(&entry->child);
231 		if (entry->proc_handler == NULL)
232 			kfree(entry->procname);
233 	}
234 
235 	kfree(*tablep);
236 	*tablep = NULL;
237 }
238 
239 static int min_load_idx = 0;
240 static int max_load_idx = CPU_LOAD_IDX_MAX-1;
241 
242 static void
243 set_table_entry(struct ctl_table *entry,
244 		const char *procname, void *data, int maxlen,
245 		umode_t mode, proc_handler *proc_handler,
246 		bool load_idx)
247 {
248 	entry->procname = procname;
249 	entry->data = data;
250 	entry->maxlen = maxlen;
251 	entry->mode = mode;
252 	entry->proc_handler = proc_handler;
253 
254 	if (load_idx) {
255 		entry->extra1 = &min_load_idx;
256 		entry->extra2 = &max_load_idx;
257 	}
258 }
259 
260 static struct ctl_table *
261 sd_alloc_ctl_domain_table(struct sched_domain *sd)
262 {
263 	struct ctl_table *table = sd_alloc_ctl_entry(14);
264 
265 	if (table == NULL)
266 		return NULL;
267 
268 	set_table_entry(&table[0] , "min_interval",	   &sd->min_interval,	     sizeof(long), 0644, proc_doulongvec_minmax, false);
269 	set_table_entry(&table[1] , "max_interval",	   &sd->max_interval,	     sizeof(long), 0644, proc_doulongvec_minmax, false);
270 	set_table_entry(&table[2] , "busy_idx",		   &sd->busy_idx,	     sizeof(int) , 0644, proc_dointvec_minmax,   true );
271 	set_table_entry(&table[3] , "idle_idx",		   &sd->idle_idx,	     sizeof(int) , 0644, proc_dointvec_minmax,   true );
272 	set_table_entry(&table[4] , "newidle_idx",	   &sd->newidle_idx,	     sizeof(int) , 0644, proc_dointvec_minmax,   true );
273 	set_table_entry(&table[5] , "wake_idx",		   &sd->wake_idx,	     sizeof(int) , 0644, proc_dointvec_minmax,   true );
274 	set_table_entry(&table[6] , "forkexec_idx",	   &sd->forkexec_idx,	     sizeof(int) , 0644, proc_dointvec_minmax,   true );
275 	set_table_entry(&table[7] , "busy_factor",	   &sd->busy_factor,	     sizeof(int) , 0644, proc_dointvec_minmax,   false);
276 	set_table_entry(&table[8] , "imbalance_pct",	   &sd->imbalance_pct,	     sizeof(int) , 0644, proc_dointvec_minmax,   false);
277 	set_table_entry(&table[9] , "cache_nice_tries",	   &sd->cache_nice_tries,    sizeof(int) , 0644, proc_dointvec_minmax,   false);
278 	set_table_entry(&table[10], "flags",		   &sd->flags,		     sizeof(int) , 0644, proc_dointvec_minmax,   false);
279 	set_table_entry(&table[11], "max_newidle_lb_cost", &sd->max_newidle_lb_cost, sizeof(long), 0644, proc_doulongvec_minmax, false);
280 	set_table_entry(&table[12], "name",		   sd->name,		CORENAME_MAX_SIZE, 0444, proc_dostring,		 false);
281 	/* &table[13] is terminator */
282 
283 	return table;
284 }
285 
286 static struct ctl_table *sd_alloc_ctl_cpu_table(int cpu)
287 {
288 	struct ctl_table *entry, *table;
289 	struct sched_domain *sd;
290 	int domain_num = 0, i;
291 	char buf[32];
292 
293 	for_each_domain(cpu, sd)
294 		domain_num++;
295 	entry = table = sd_alloc_ctl_entry(domain_num + 1);
296 	if (table == NULL)
297 		return NULL;
298 
299 	i = 0;
300 	for_each_domain(cpu, sd) {
301 		snprintf(buf, 32, "domain%d", i);
302 		entry->procname = kstrdup(buf, GFP_KERNEL);
303 		entry->mode = 0555;
304 		entry->child = sd_alloc_ctl_domain_table(sd);
305 		entry++;
306 		i++;
307 	}
308 	return table;
309 }
310 
311 static cpumask_var_t		sd_sysctl_cpus;
312 static struct ctl_table_header	*sd_sysctl_header;
313 
314 void register_sched_domain_sysctl(void)
315 {
316 	static struct ctl_table *cpu_entries;
317 	static struct ctl_table **cpu_idx;
318 	char buf[32];
319 	int i;
320 
321 	if (!cpu_entries) {
322 		cpu_entries = sd_alloc_ctl_entry(num_possible_cpus() + 1);
323 		if (!cpu_entries)
324 			return;
325 
326 		WARN_ON(sd_ctl_dir[0].child);
327 		sd_ctl_dir[0].child = cpu_entries;
328 	}
329 
330 	if (!cpu_idx) {
331 		struct ctl_table *e = cpu_entries;
332 
333 		cpu_idx = kcalloc(nr_cpu_ids, sizeof(struct ctl_table*), GFP_KERNEL);
334 		if (!cpu_idx)
335 			return;
336 
337 		/* deal with sparse possible map */
338 		for_each_possible_cpu(i) {
339 			cpu_idx[i] = e;
340 			e++;
341 		}
342 	}
343 
344 	if (!cpumask_available(sd_sysctl_cpus)) {
345 		if (!alloc_cpumask_var(&sd_sysctl_cpus, GFP_KERNEL))
346 			return;
347 
348 		/* init to possible to not have holes in @cpu_entries */
349 		cpumask_copy(sd_sysctl_cpus, cpu_possible_mask);
350 	}
351 
352 	for_each_cpu(i, sd_sysctl_cpus) {
353 		struct ctl_table *e = cpu_idx[i];
354 
355 		if (e->child)
356 			sd_free_ctl_entry(&e->child);
357 
358 		if (!e->procname) {
359 			snprintf(buf, 32, "cpu%d", i);
360 			e->procname = kstrdup(buf, GFP_KERNEL);
361 		}
362 		e->mode = 0555;
363 		e->child = sd_alloc_ctl_cpu_table(i);
364 
365 		__cpumask_clear_cpu(i, sd_sysctl_cpus);
366 	}
367 
368 	WARN_ON(sd_sysctl_header);
369 	sd_sysctl_header = register_sysctl_table(sd_ctl_root);
370 }
371 
372 void dirty_sched_domain_sysctl(int cpu)
373 {
374 	if (cpumask_available(sd_sysctl_cpus))
375 		__cpumask_set_cpu(cpu, sd_sysctl_cpus);
376 }
377 
378 /* may be called multiple times per register */
379 void unregister_sched_domain_sysctl(void)
380 {
381 	unregister_sysctl_table(sd_sysctl_header);
382 	sd_sysctl_header = NULL;
383 }
384 #endif /* CONFIG_SYSCTL */
385 #endif /* CONFIG_SMP */
386 
387 #ifdef CONFIG_FAIR_GROUP_SCHED
388 static void print_cfs_group_stats(struct seq_file *m, int cpu, struct task_group *tg)
389 {
390 	struct sched_entity *se = tg->se[cpu];
391 
392 #define P(F)		SEQ_printf(m, "  .%-30s: %lld\n",	#F, (long long)F)
393 #define P_SCHEDSTAT(F)	SEQ_printf(m, "  .%-30s: %lld\n",	#F, (long long)schedstat_val(F))
394 #define PN(F)		SEQ_printf(m, "  .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F))
395 #define PN_SCHEDSTAT(F)	SEQ_printf(m, "  .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)schedstat_val(F)))
396 
397 	if (!se)
398 		return;
399 
400 	PN(se->exec_start);
401 	PN(se->vruntime);
402 	PN(se->sum_exec_runtime);
403 
404 	if (schedstat_enabled()) {
405 		PN_SCHEDSTAT(se->statistics.wait_start);
406 		PN_SCHEDSTAT(se->statistics.sleep_start);
407 		PN_SCHEDSTAT(se->statistics.block_start);
408 		PN_SCHEDSTAT(se->statistics.sleep_max);
409 		PN_SCHEDSTAT(se->statistics.block_max);
410 		PN_SCHEDSTAT(se->statistics.exec_max);
411 		PN_SCHEDSTAT(se->statistics.slice_max);
412 		PN_SCHEDSTAT(se->statistics.wait_max);
413 		PN_SCHEDSTAT(se->statistics.wait_sum);
414 		P_SCHEDSTAT(se->statistics.wait_count);
415 	}
416 
417 	P(se->load.weight);
418 	P(se->runnable_weight);
419 #ifdef CONFIG_SMP
420 	P(se->avg.load_avg);
421 	P(se->avg.util_avg);
422 	P(se->avg.runnable_load_avg);
423 #endif
424 
425 #undef PN_SCHEDSTAT
426 #undef PN
427 #undef P_SCHEDSTAT
428 #undef P
429 }
430 #endif
431 
432 #ifdef CONFIG_CGROUP_SCHED
433 static char group_path[PATH_MAX];
434 
435 static char *task_group_path(struct task_group *tg)
436 {
437 	if (autogroup_path(tg, group_path, PATH_MAX))
438 		return group_path;
439 
440 	cgroup_path(tg->css.cgroup, group_path, PATH_MAX);
441 
442 	return group_path;
443 }
444 #endif
445 
446 static void
447 print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
448 {
449 	if (rq->curr == p)
450 		SEQ_printf(m, ">R");
451 	else
452 		SEQ_printf(m, " %c", task_state_to_char(p));
453 
454 	SEQ_printf(m, "%15s %5d %9Ld.%06ld %9Ld %5d ",
455 		p->comm, task_pid_nr(p),
456 		SPLIT_NS(p->se.vruntime),
457 		(long long)(p->nvcsw + p->nivcsw),
458 		p->prio);
459 
460 	SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
461 		SPLIT_NS(schedstat_val_or_zero(p->se.statistics.wait_sum)),
462 		SPLIT_NS(p->se.sum_exec_runtime),
463 		SPLIT_NS(schedstat_val_or_zero(p->se.statistics.sum_sleep_runtime)));
464 
465 #ifdef CONFIG_NUMA_BALANCING
466 	SEQ_printf(m, " %d %d", task_node(p), task_numa_group_id(p));
467 #endif
468 #ifdef CONFIG_CGROUP_SCHED
469 	SEQ_printf(m, " %s", task_group_path(task_group(p)));
470 #endif
471 
472 	SEQ_printf(m, "\n");
473 }
474 
475 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
476 {
477 	struct task_struct *g, *p;
478 
479 	SEQ_printf(m, "\n");
480 	SEQ_printf(m, "runnable tasks:\n");
481 	SEQ_printf(m, " S           task   PID         tree-key  switches  prio"
482 		   "     wait-time             sum-exec        sum-sleep\n");
483 	SEQ_printf(m, "-------------------------------------------------------"
484 		   "----------------------------------------------------\n");
485 
486 	rcu_read_lock();
487 	for_each_process_thread(g, p) {
488 		if (task_cpu(p) != rq_cpu)
489 			continue;
490 
491 		print_task(m, rq, p);
492 	}
493 	rcu_read_unlock();
494 }
495 
496 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
497 {
498 	s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
499 		spread, rq0_min_vruntime, spread0;
500 	struct rq *rq = cpu_rq(cpu);
501 	struct sched_entity *last;
502 	unsigned long flags;
503 
504 #ifdef CONFIG_FAIR_GROUP_SCHED
505 	SEQ_printf(m, "\n");
506 	SEQ_printf(m, "cfs_rq[%d]:%s\n", cpu, task_group_path(cfs_rq->tg));
507 #else
508 	SEQ_printf(m, "\n");
509 	SEQ_printf(m, "cfs_rq[%d]:\n", cpu);
510 #endif
511 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "exec_clock",
512 			SPLIT_NS(cfs_rq->exec_clock));
513 
514 	raw_spin_lock_irqsave(&rq->lock, flags);
515 	if (rb_first_cached(&cfs_rq->tasks_timeline))
516 		MIN_vruntime = (__pick_first_entity(cfs_rq))->vruntime;
517 	last = __pick_last_entity(cfs_rq);
518 	if (last)
519 		max_vruntime = last->vruntime;
520 	min_vruntime = cfs_rq->min_vruntime;
521 	rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
522 	raw_spin_unlock_irqrestore(&rq->lock, flags);
523 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "MIN_vruntime",
524 			SPLIT_NS(MIN_vruntime));
525 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "min_vruntime",
526 			SPLIT_NS(min_vruntime));
527 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "max_vruntime",
528 			SPLIT_NS(max_vruntime));
529 	spread = max_vruntime - MIN_vruntime;
530 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread",
531 			SPLIT_NS(spread));
532 	spread0 = min_vruntime - rq0_min_vruntime;
533 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread0",
534 			SPLIT_NS(spread0));
535 	SEQ_printf(m, "  .%-30s: %d\n", "nr_spread_over",
536 			cfs_rq->nr_spread_over);
537 	SEQ_printf(m, "  .%-30s: %d\n", "nr_running", cfs_rq->nr_running);
538 	SEQ_printf(m, "  .%-30s: %ld\n", "load", cfs_rq->load.weight);
539 #ifdef CONFIG_SMP
540 	SEQ_printf(m, "  .%-30s: %ld\n", "runnable_weight", cfs_rq->runnable_weight);
541 	SEQ_printf(m, "  .%-30s: %lu\n", "load_avg",
542 			cfs_rq->avg.load_avg);
543 	SEQ_printf(m, "  .%-30s: %lu\n", "runnable_load_avg",
544 			cfs_rq->avg.runnable_load_avg);
545 	SEQ_printf(m, "  .%-30s: %lu\n", "util_avg",
546 			cfs_rq->avg.util_avg);
547 	SEQ_printf(m, "  .%-30s: %u\n", "util_est_enqueued",
548 			cfs_rq->avg.util_est.enqueued);
549 	SEQ_printf(m, "  .%-30s: %ld\n", "removed.load_avg",
550 			cfs_rq->removed.load_avg);
551 	SEQ_printf(m, "  .%-30s: %ld\n", "removed.util_avg",
552 			cfs_rq->removed.util_avg);
553 	SEQ_printf(m, "  .%-30s: %ld\n", "removed.runnable_sum",
554 			cfs_rq->removed.runnable_sum);
555 #ifdef CONFIG_FAIR_GROUP_SCHED
556 	SEQ_printf(m, "  .%-30s: %lu\n", "tg_load_avg_contrib",
557 			cfs_rq->tg_load_avg_contrib);
558 	SEQ_printf(m, "  .%-30s: %ld\n", "tg_load_avg",
559 			atomic_long_read(&cfs_rq->tg->load_avg));
560 #endif
561 #endif
562 #ifdef CONFIG_CFS_BANDWIDTH
563 	SEQ_printf(m, "  .%-30s: %d\n", "throttled",
564 			cfs_rq->throttled);
565 	SEQ_printf(m, "  .%-30s: %d\n", "throttle_count",
566 			cfs_rq->throttle_count);
567 #endif
568 
569 #ifdef CONFIG_FAIR_GROUP_SCHED
570 	print_cfs_group_stats(m, cpu, cfs_rq->tg);
571 #endif
572 }
573 
574 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
575 {
576 #ifdef CONFIG_RT_GROUP_SCHED
577 	SEQ_printf(m, "\n");
578 	SEQ_printf(m, "rt_rq[%d]:%s\n", cpu, task_group_path(rt_rq->tg));
579 #else
580 	SEQ_printf(m, "\n");
581 	SEQ_printf(m, "rt_rq[%d]:\n", cpu);
582 #endif
583 
584 #define P(x) \
585 	SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
586 #define PU(x) \
587 	SEQ_printf(m, "  .%-30s: %lu\n", #x, (unsigned long)(rt_rq->x))
588 #define PN(x) \
589 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
590 
591 	PU(rt_nr_running);
592 #ifdef CONFIG_SMP
593 	PU(rt_nr_migratory);
594 #endif
595 	P(rt_throttled);
596 	PN(rt_time);
597 	PN(rt_runtime);
598 
599 #undef PN
600 #undef PU
601 #undef P
602 }
603 
604 void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq)
605 {
606 	struct dl_bw *dl_bw;
607 
608 	SEQ_printf(m, "\n");
609 	SEQ_printf(m, "dl_rq[%d]:\n", cpu);
610 
611 #define PU(x) \
612 	SEQ_printf(m, "  .%-30s: %lu\n", #x, (unsigned long)(dl_rq->x))
613 
614 	PU(dl_nr_running);
615 #ifdef CONFIG_SMP
616 	PU(dl_nr_migratory);
617 	dl_bw = &cpu_rq(cpu)->rd->dl_bw;
618 #else
619 	dl_bw = &dl_rq->dl_bw;
620 #endif
621 	SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->bw", dl_bw->bw);
622 	SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->total_bw", dl_bw->total_bw);
623 
624 #undef PU
625 }
626 
627 static void print_cpu(struct seq_file *m, int cpu)
628 {
629 	struct rq *rq = cpu_rq(cpu);
630 	unsigned long flags;
631 
632 #ifdef CONFIG_X86
633 	{
634 		unsigned int freq = cpu_khz ? : 1;
635 
636 		SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
637 			   cpu, freq / 1000, (freq % 1000));
638 	}
639 #else
640 	SEQ_printf(m, "cpu#%d\n", cpu);
641 #endif
642 
643 #define P(x)								\
644 do {									\
645 	if (sizeof(rq->x) == 4)						\
646 		SEQ_printf(m, "  .%-30s: %ld\n", #x, (long)(rq->x));	\
647 	else								\
648 		SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rq->x));\
649 } while (0)
650 
651 #define PN(x) \
652 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
653 
654 	P(nr_running);
655 	SEQ_printf(m, "  .%-30s: %lu\n", "load",
656 		   rq->load.weight);
657 	P(nr_switches);
658 	P(nr_load_updates);
659 	P(nr_uninterruptible);
660 	PN(next_balance);
661 	SEQ_printf(m, "  .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr)));
662 	PN(clock);
663 	PN(clock_task);
664 	P(cpu_load[0]);
665 	P(cpu_load[1]);
666 	P(cpu_load[2]);
667 	P(cpu_load[3]);
668 	P(cpu_load[4]);
669 #undef P
670 #undef PN
671 
672 #ifdef CONFIG_SMP
673 #define P64(n) SEQ_printf(m, "  .%-30s: %Ld\n", #n, rq->n);
674 	P64(avg_idle);
675 	P64(max_idle_balance_cost);
676 #undef P64
677 #endif
678 
679 #define P(n) SEQ_printf(m, "  .%-30s: %d\n", #n, schedstat_val(rq->n));
680 	if (schedstat_enabled()) {
681 		P(yld_count);
682 		P(sched_count);
683 		P(sched_goidle);
684 		P(ttwu_count);
685 		P(ttwu_local);
686 	}
687 #undef P
688 
689 	spin_lock_irqsave(&sched_debug_lock, flags);
690 	print_cfs_stats(m, cpu);
691 	print_rt_stats(m, cpu);
692 	print_dl_stats(m, cpu);
693 
694 	print_rq(m, rq, cpu);
695 	spin_unlock_irqrestore(&sched_debug_lock, flags);
696 	SEQ_printf(m, "\n");
697 }
698 
699 static const char *sched_tunable_scaling_names[] = {
700 	"none",
701 	"logaritmic",
702 	"linear"
703 };
704 
705 static void sched_debug_header(struct seq_file *m)
706 {
707 	u64 ktime, sched_clk, cpu_clk;
708 	unsigned long flags;
709 
710 	local_irq_save(flags);
711 	ktime = ktime_to_ns(ktime_get());
712 	sched_clk = sched_clock();
713 	cpu_clk = local_clock();
714 	local_irq_restore(flags);
715 
716 	SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n",
717 		init_utsname()->release,
718 		(int)strcspn(init_utsname()->version, " "),
719 		init_utsname()->version);
720 
721 #define P(x) \
722 	SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
723 #define PN(x) \
724 	SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
725 	PN(ktime);
726 	PN(sched_clk);
727 	PN(cpu_clk);
728 	P(jiffies);
729 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
730 	P(sched_clock_stable());
731 #endif
732 #undef PN
733 #undef P
734 
735 	SEQ_printf(m, "\n");
736 	SEQ_printf(m, "sysctl_sched\n");
737 
738 #define P(x) \
739 	SEQ_printf(m, "  .%-40s: %Ld\n", #x, (long long)(x))
740 #define PN(x) \
741 	SEQ_printf(m, "  .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
742 	PN(sysctl_sched_latency);
743 	PN(sysctl_sched_min_granularity);
744 	PN(sysctl_sched_wakeup_granularity);
745 	P(sysctl_sched_child_runs_first);
746 	P(sysctl_sched_features);
747 #undef PN
748 #undef P
749 
750 	SEQ_printf(m, "  .%-40s: %d (%s)\n",
751 		"sysctl_sched_tunable_scaling",
752 		sysctl_sched_tunable_scaling,
753 		sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
754 	SEQ_printf(m, "\n");
755 }
756 
757 static int sched_debug_show(struct seq_file *m, void *v)
758 {
759 	int cpu = (unsigned long)(v - 2);
760 
761 	if (cpu != -1)
762 		print_cpu(m, cpu);
763 	else
764 		sched_debug_header(m);
765 
766 	return 0;
767 }
768 
769 void sysrq_sched_debug_show(void)
770 {
771 	int cpu;
772 
773 	sched_debug_header(NULL);
774 	for_each_online_cpu(cpu)
775 		print_cpu(NULL, cpu);
776 
777 }
778 
779 /*
780  * This itererator needs some explanation.
781  * It returns 1 for the header position.
782  * This means 2 is CPU 0.
783  * In a hotplugged system some CPUs, including CPU 0, may be missing so we have
784  * to use cpumask_* to iterate over the CPUs.
785  */
786 static void *sched_debug_start(struct seq_file *file, loff_t *offset)
787 {
788 	unsigned long n = *offset;
789 
790 	if (n == 0)
791 		return (void *) 1;
792 
793 	n--;
794 
795 	if (n > 0)
796 		n = cpumask_next(n - 1, cpu_online_mask);
797 	else
798 		n = cpumask_first(cpu_online_mask);
799 
800 	*offset = n + 1;
801 
802 	if (n < nr_cpu_ids)
803 		return (void *)(unsigned long)(n + 2);
804 
805 	return NULL;
806 }
807 
808 static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
809 {
810 	(*offset)++;
811 	return sched_debug_start(file, offset);
812 }
813 
814 static void sched_debug_stop(struct seq_file *file, void *data)
815 {
816 }
817 
818 static const struct seq_operations sched_debug_sops = {
819 	.start		= sched_debug_start,
820 	.next		= sched_debug_next,
821 	.stop		= sched_debug_stop,
822 	.show		= sched_debug_show,
823 };
824 
825 static int __init init_sched_debug_procfs(void)
826 {
827 	if (!proc_create_seq("sched_debug", 0444, NULL, &sched_debug_sops))
828 		return -ENOMEM;
829 	return 0;
830 }
831 
832 __initcall(init_sched_debug_procfs);
833 
834 #define __P(F)	SEQ_printf(m, "%-45s:%21Ld\n",	     #F, (long long)F)
835 #define   P(F)	SEQ_printf(m, "%-45s:%21Ld\n",	     #F, (long long)p->F)
836 #define __PN(F)	SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
837 #define   PN(F)	SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
838 
839 
840 #ifdef CONFIG_NUMA_BALANCING
841 void print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
842 		unsigned long tpf, unsigned long gsf, unsigned long gpf)
843 {
844 	SEQ_printf(m, "numa_faults node=%d ", node);
845 	SEQ_printf(m, "task_private=%lu task_shared=%lu ", tpf, tsf);
846 	SEQ_printf(m, "group_private=%lu group_shared=%lu\n", gpf, gsf);
847 }
848 #endif
849 
850 
851 static void sched_show_numa(struct task_struct *p, struct seq_file *m)
852 {
853 #ifdef CONFIG_NUMA_BALANCING
854 	struct mempolicy *pol;
855 
856 	if (p->mm)
857 		P(mm->numa_scan_seq);
858 
859 	task_lock(p);
860 	pol = p->mempolicy;
861 	if (pol && !(pol->flags & MPOL_F_MORON))
862 		pol = NULL;
863 	mpol_get(pol);
864 	task_unlock(p);
865 
866 	P(numa_pages_migrated);
867 	P(numa_preferred_nid);
868 	P(total_numa_faults);
869 	SEQ_printf(m, "current_node=%d, numa_group_id=%d\n",
870 			task_node(p), task_numa_group_id(p));
871 	show_numa_stats(p, m);
872 	mpol_put(pol);
873 #endif
874 }
875 
876 void proc_sched_show_task(struct task_struct *p, struct pid_namespace *ns,
877 						  struct seq_file *m)
878 {
879 	unsigned long nr_switches;
880 
881 	SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr_ns(p, ns),
882 						get_nr_threads(p));
883 	SEQ_printf(m,
884 		"---------------------------------------------------------"
885 		"----------\n");
886 #define __P(F) \
887 	SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)F)
888 #define P(F) \
889 	SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)p->F)
890 #define P_SCHEDSTAT(F) \
891 	SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)schedstat_val(p->F))
892 #define __PN(F) \
893 	SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
894 #define PN(F) \
895 	SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
896 #define PN_SCHEDSTAT(F) \
897 	SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)schedstat_val(p->F)))
898 
899 	PN(se.exec_start);
900 	PN(se.vruntime);
901 	PN(se.sum_exec_runtime);
902 
903 	nr_switches = p->nvcsw + p->nivcsw;
904 
905 	P(se.nr_migrations);
906 
907 	if (schedstat_enabled()) {
908 		u64 avg_atom, avg_per_cpu;
909 
910 		PN_SCHEDSTAT(se.statistics.sum_sleep_runtime);
911 		PN_SCHEDSTAT(se.statistics.wait_start);
912 		PN_SCHEDSTAT(se.statistics.sleep_start);
913 		PN_SCHEDSTAT(se.statistics.block_start);
914 		PN_SCHEDSTAT(se.statistics.sleep_max);
915 		PN_SCHEDSTAT(se.statistics.block_max);
916 		PN_SCHEDSTAT(se.statistics.exec_max);
917 		PN_SCHEDSTAT(se.statistics.slice_max);
918 		PN_SCHEDSTAT(se.statistics.wait_max);
919 		PN_SCHEDSTAT(se.statistics.wait_sum);
920 		P_SCHEDSTAT(se.statistics.wait_count);
921 		PN_SCHEDSTAT(se.statistics.iowait_sum);
922 		P_SCHEDSTAT(se.statistics.iowait_count);
923 		P_SCHEDSTAT(se.statistics.nr_migrations_cold);
924 		P_SCHEDSTAT(se.statistics.nr_failed_migrations_affine);
925 		P_SCHEDSTAT(se.statistics.nr_failed_migrations_running);
926 		P_SCHEDSTAT(se.statistics.nr_failed_migrations_hot);
927 		P_SCHEDSTAT(se.statistics.nr_forced_migrations);
928 		P_SCHEDSTAT(se.statistics.nr_wakeups);
929 		P_SCHEDSTAT(se.statistics.nr_wakeups_sync);
930 		P_SCHEDSTAT(se.statistics.nr_wakeups_migrate);
931 		P_SCHEDSTAT(se.statistics.nr_wakeups_local);
932 		P_SCHEDSTAT(se.statistics.nr_wakeups_remote);
933 		P_SCHEDSTAT(se.statistics.nr_wakeups_affine);
934 		P_SCHEDSTAT(se.statistics.nr_wakeups_affine_attempts);
935 		P_SCHEDSTAT(se.statistics.nr_wakeups_passive);
936 		P_SCHEDSTAT(se.statistics.nr_wakeups_idle);
937 
938 		avg_atom = p->se.sum_exec_runtime;
939 		if (nr_switches)
940 			avg_atom = div64_ul(avg_atom, nr_switches);
941 		else
942 			avg_atom = -1LL;
943 
944 		avg_per_cpu = p->se.sum_exec_runtime;
945 		if (p->se.nr_migrations) {
946 			avg_per_cpu = div64_u64(avg_per_cpu,
947 						p->se.nr_migrations);
948 		} else {
949 			avg_per_cpu = -1LL;
950 		}
951 
952 		__PN(avg_atom);
953 		__PN(avg_per_cpu);
954 	}
955 
956 	__P(nr_switches);
957 	SEQ_printf(m, "%-45s:%21Ld\n",
958 		   "nr_voluntary_switches", (long long)p->nvcsw);
959 	SEQ_printf(m, "%-45s:%21Ld\n",
960 		   "nr_involuntary_switches", (long long)p->nivcsw);
961 
962 	P(se.load.weight);
963 	P(se.runnable_weight);
964 #ifdef CONFIG_SMP
965 	P(se.avg.load_sum);
966 	P(se.avg.runnable_load_sum);
967 	P(se.avg.util_sum);
968 	P(se.avg.load_avg);
969 	P(se.avg.runnable_load_avg);
970 	P(se.avg.util_avg);
971 	P(se.avg.last_update_time);
972 	P(se.avg.util_est.ewma);
973 	P(se.avg.util_est.enqueued);
974 #endif
975 	P(policy);
976 	P(prio);
977 	if (task_has_dl_policy(p)) {
978 		P(dl.runtime);
979 		P(dl.deadline);
980 	}
981 #undef PN_SCHEDSTAT
982 #undef PN
983 #undef __PN
984 #undef P_SCHEDSTAT
985 #undef P
986 #undef __P
987 
988 	{
989 		unsigned int this_cpu = raw_smp_processor_id();
990 		u64 t0, t1;
991 
992 		t0 = cpu_clock(this_cpu);
993 		t1 = cpu_clock(this_cpu);
994 		SEQ_printf(m, "%-45s:%21Ld\n",
995 			   "clock-delta", (long long)(t1-t0));
996 	}
997 
998 	sched_show_numa(p, m);
999 }
1000 
1001 void proc_sched_set_task(struct task_struct *p)
1002 {
1003 #ifdef CONFIG_SCHEDSTATS
1004 	memset(&p->se.statistics, 0, sizeof(p->se.statistics));
1005 #endif
1006 }
1007