• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * kernel/sched/debug.c
3  *
4  * Print the CFS rbtree
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 
13 #include <linux/proc_fs.h>
14 #include <linux/sched.h>
15 #include <linux/seq_file.h>
16 #include <linux/kallsyms.h>
17 #include <linux/utsname.h>
18 #include <linux/mempolicy.h>
19 #include <linux/debugfs.h>
20 
21 #include "sched.h"
22 
23 static DEFINE_SPINLOCK(sched_debug_lock);
24 
25 /*
26  * This allows printing both to /proc/sched_debug and
27  * to the console
28  */
29 #define SEQ_printf(m, x...)			\
30  do {						\
31 	if (m)					\
32 		seq_printf(m, x);		\
33 	else					\
34 		printk(x);			\
35  } while (0)
36 
37 /*
38  * Ease the printing of nsec fields:
39  */
nsec_high(unsigned long long nsec)40 static long long nsec_high(unsigned long long nsec)
41 {
42 	if ((long long)nsec < 0) {
43 		nsec = -nsec;
44 		do_div(nsec, 1000000);
45 		return -nsec;
46 	}
47 	do_div(nsec, 1000000);
48 
49 	return nsec;
50 }
51 
nsec_low(unsigned long long nsec)52 static unsigned long nsec_low(unsigned long long nsec)
53 {
54 	if ((long long)nsec < 0)
55 		nsec = -nsec;
56 
57 	return do_div(nsec, 1000000);
58 }
59 
60 #define SPLIT_NS(x) nsec_high(x), nsec_low(x)
61 
62 #define SCHED_FEAT(name, enabled)	\
63 	#name ,
64 
65 static const char * const sched_feat_names[] = {
66 #include "features.h"
67 };
68 
69 #undef SCHED_FEAT
70 
sched_feat_show(struct seq_file * m,void * v)71 static int sched_feat_show(struct seq_file *m, void *v)
72 {
73 	int i;
74 
75 	for (i = 0; i < __SCHED_FEAT_NR; i++) {
76 		if (!(sysctl_sched_features & (1UL << i)))
77 			seq_puts(m, "NO_");
78 		seq_printf(m, "%s ", sched_feat_names[i]);
79 	}
80 	seq_puts(m, "\n");
81 
82 	return 0;
83 }
84 
85 #ifdef HAVE_JUMP_LABEL
86 
87 #define jump_label_key__true  STATIC_KEY_INIT_TRUE
88 #define jump_label_key__false STATIC_KEY_INIT_FALSE
89 
90 #define SCHED_FEAT(name, enabled)	\
91 	jump_label_key__##enabled ,
92 
93 struct static_key sched_feat_keys[__SCHED_FEAT_NR] = {
94 #include "features.h"
95 };
96 
97 #undef SCHED_FEAT
98 
sched_feat_disable(int i)99 static void sched_feat_disable(int i)
100 {
101 	static_key_disable(&sched_feat_keys[i]);
102 }
103 
sched_feat_enable(int i)104 static void sched_feat_enable(int i)
105 {
106 	static_key_enable(&sched_feat_keys[i]);
107 }
108 #else
sched_feat_disable(int i)109 static void sched_feat_disable(int i) { };
sched_feat_enable(int i)110 static void sched_feat_enable(int i) { };
111 #endif /* HAVE_JUMP_LABEL */
112 
sched_feat_set(char * cmp)113 static int sched_feat_set(char *cmp)
114 {
115 	int i;
116 	int neg = 0;
117 
118 	if (strncmp(cmp, "NO_", 3) == 0) {
119 		neg = 1;
120 		cmp += 3;
121 	}
122 
123 	for (i = 0; i < __SCHED_FEAT_NR; i++) {
124 		if (strcmp(cmp, sched_feat_names[i]) == 0) {
125 			if (neg) {
126 				sysctl_sched_features &= ~(1UL << i);
127 				sched_feat_disable(i);
128 			} else {
129 				sysctl_sched_features |= (1UL << i);
130 				sched_feat_enable(i);
131 			}
132 			break;
133 		}
134 	}
135 
136 	return i;
137 }
138 
139 static ssize_t
sched_feat_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)140 sched_feat_write(struct file *filp, const char __user *ubuf,
141 		size_t cnt, loff_t *ppos)
142 {
143 	char buf[64];
144 	char *cmp;
145 	int i;
146 	struct inode *inode;
147 
148 	if (cnt > 63)
149 		cnt = 63;
150 
151 	if (copy_from_user(&buf, ubuf, cnt))
152 		return -EFAULT;
153 
154 	buf[cnt] = 0;
155 	cmp = strstrip(buf);
156 
157 	/* Ensure the static_key remains in a consistent state */
158 	inode = file_inode(filp);
159 	inode_lock(inode);
160 	i = sched_feat_set(cmp);
161 	inode_unlock(inode);
162 	if (i == __SCHED_FEAT_NR)
163 		return -EINVAL;
164 
165 	*ppos += cnt;
166 
167 	return cnt;
168 }
169 
sched_feat_open(struct inode * inode,struct file * filp)170 static int sched_feat_open(struct inode *inode, struct file *filp)
171 {
172 	return single_open(filp, sched_feat_show, NULL);
173 }
174 
175 static const struct file_operations sched_feat_fops = {
176 	.open		= sched_feat_open,
177 	.write		= sched_feat_write,
178 	.read		= seq_read,
179 	.llseek		= seq_lseek,
180 	.release	= single_release,
181 };
182 
sched_init_debug(void)183 static __init int sched_init_debug(void)
184 {
185 	debugfs_create_file("sched_features", 0644, NULL, NULL,
186 			&sched_feat_fops);
187 
188 	return 0;
189 }
190 late_initcall(sched_init_debug);
191 
192 #ifdef CONFIG_SMP
193 
194 #ifdef CONFIG_SYSCTL
195 
196 static struct ctl_table sd_ctl_dir[] = {
197 	{
198 		.procname	= "sched_domain",
199 		.mode		= 0555,
200 	},
201 	{}
202 };
203 
204 static struct ctl_table sd_ctl_root[] = {
205 	{
206 		.procname	= "kernel",
207 		.mode		= 0555,
208 		.child		= sd_ctl_dir,
209 	},
210 	{}
211 };
212 
sd_alloc_ctl_entry(int n)213 static struct ctl_table *sd_alloc_ctl_entry(int n)
214 {
215 	struct ctl_table *entry =
216 		kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL);
217 
218 	return entry;
219 }
220 
sd_free_ctl_entry(struct ctl_table ** tablep)221 static void sd_free_ctl_entry(struct ctl_table **tablep)
222 {
223 	struct ctl_table *entry;
224 
225 	/*
226 	 * In the intermediate directories, both the child directory and
227 	 * procname are dynamically allocated and could fail but the mode
228 	 * will always be set. In the lowest directory the names are
229 	 * static strings and all have proc handlers.
230 	 */
231 	for (entry = *tablep; entry->mode; entry++) {
232 		if (entry->child)
233 			sd_free_ctl_entry(&entry->child);
234 		if (entry->proc_handler == NULL)
235 			kfree(entry->procname);
236 	}
237 
238 	kfree(*tablep);
239 	*tablep = NULL;
240 }
241 
242 static int min_load_idx = 0;
243 static int max_load_idx = CPU_LOAD_IDX_MAX-1;
244 
245 static void
set_table_entry(struct ctl_table * entry,const char * procname,void * data,int maxlen,umode_t mode,proc_handler * proc_handler,bool load_idx)246 set_table_entry(struct ctl_table *entry,
247 		const char *procname, void *data, int maxlen,
248 		umode_t mode, proc_handler *proc_handler,
249 		bool load_idx)
250 {
251 	entry->procname = procname;
252 	entry->data = data;
253 	entry->maxlen = maxlen;
254 	entry->mode = mode;
255 	entry->proc_handler = proc_handler;
256 
257 	if (load_idx) {
258 		entry->extra1 = &min_load_idx;
259 		entry->extra2 = &max_load_idx;
260 	}
261 }
262 
263 static struct ctl_table *
sd_alloc_ctl_energy_table(struct sched_group_energy * sge)264 sd_alloc_ctl_energy_table(struct sched_group_energy *sge)
265 {
266 	struct ctl_table *table = sd_alloc_ctl_entry(5);
267 
268 	if (table == NULL)
269 		return NULL;
270 
271 	set_table_entry(&table[0], "nr_idle_states", &sge->nr_idle_states,
272 			sizeof(int), 0644, proc_dointvec_minmax, false);
273 	set_table_entry(&table[1], "idle_states", &sge->idle_states[0].power,
274 			sge->nr_idle_states*sizeof(struct idle_state), 0644,
275 			proc_doulongvec_minmax, false);
276 	set_table_entry(&table[2], "nr_cap_states", &sge->nr_cap_states,
277 			sizeof(int), 0644, proc_dointvec_minmax, false);
278 	set_table_entry(&table[3], "cap_states", &sge->cap_states[0].cap,
279 			sge->nr_cap_states*sizeof(struct capacity_state), 0644,
280 			proc_doulongvec_minmax, false);
281 
282 	return table;
283 }
284 
285 static struct ctl_table *
sd_alloc_ctl_group_table(struct sched_group * sg)286 sd_alloc_ctl_group_table(struct sched_group *sg)
287 {
288 	struct ctl_table *table = sd_alloc_ctl_entry(2);
289 
290 	if (table == NULL)
291 		return NULL;
292 
293 	table->procname = kstrdup("energy", GFP_KERNEL);
294 	table->mode = 0555;
295 	table->child = sd_alloc_ctl_energy_table((struct sched_group_energy *)sg->sge);
296 
297 	return table;
298 }
299 
300 static struct ctl_table *
sd_alloc_ctl_domain_table(struct sched_domain * sd)301 sd_alloc_ctl_domain_table(struct sched_domain *sd)
302 {
303 	struct ctl_table *table;
304 	unsigned int nr_entries = 14;
305 
306 	int i = 0;
307 	struct sched_group *sg = sd->groups;
308 
309 	if (sg->sge) {
310 		int nr_sgs = 0;
311 
312 		do {} while (nr_sgs++, sg = sg->next, sg != sd->groups);
313 
314 		nr_entries += nr_sgs;
315 	}
316 
317 	table = sd_alloc_ctl_entry(nr_entries);
318 
319 	if (table == NULL)
320 		return NULL;
321 
322 	set_table_entry(&table[0], "min_interval", &sd->min_interval,
323 		sizeof(long), 0644, proc_doulongvec_minmax, false);
324 	set_table_entry(&table[1], "max_interval", &sd->max_interval,
325 		sizeof(long), 0644, proc_doulongvec_minmax, false);
326 	set_table_entry(&table[2], "busy_idx", &sd->busy_idx,
327 		sizeof(int), 0644, proc_dointvec_minmax, true);
328 	set_table_entry(&table[3], "idle_idx", &sd->idle_idx,
329 		sizeof(int), 0644, proc_dointvec_minmax, true);
330 	set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx,
331 		sizeof(int), 0644, proc_dointvec_minmax, true);
332 	set_table_entry(&table[5], "wake_idx", &sd->wake_idx,
333 		sizeof(int), 0644, proc_dointvec_minmax, true);
334 	set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx,
335 		sizeof(int), 0644, proc_dointvec_minmax, true);
336 	set_table_entry(&table[7], "busy_factor", &sd->busy_factor,
337 		sizeof(int), 0644, proc_dointvec_minmax, false);
338 	set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct,
339 		sizeof(int), 0644, proc_dointvec_minmax, false);
340 	set_table_entry(&table[9], "cache_nice_tries",
341 		&sd->cache_nice_tries,
342 		sizeof(int), 0644, proc_dointvec_minmax, false);
343 	set_table_entry(&table[10], "flags", &sd->flags,
344 		sizeof(int), 0644, proc_dointvec_minmax, false);
345 	set_table_entry(&table[11], "max_newidle_lb_cost",
346 		&sd->max_newidle_lb_cost,
347 		sizeof(long), 0644, proc_doulongvec_minmax, false);
348 	set_table_entry(&table[12], "name", sd->name,
349 		CORENAME_MAX_SIZE, 0444, proc_dostring, false);
350 	sg = sd->groups;
351 	if (sg->sge) {
352 		char buf[32];
353 		struct ctl_table *entry = &table[13];
354 
355 		do {
356 			snprintf(buf, 32, "group%d", i);
357 			entry->procname = kstrdup(buf, GFP_KERNEL);
358 			entry->mode = 0555;
359 			entry->child = sd_alloc_ctl_group_table(sg);
360 		} while (entry++, i++, sg = sg->next, sg != sd->groups);
361 	}
362 	/* &table[nr_entries-1] is terminator */
363 
364 	return table;
365 }
366 
sd_alloc_ctl_cpu_table(int cpu)367 static struct ctl_table *sd_alloc_ctl_cpu_table(int cpu)
368 {
369 	struct ctl_table *entry, *table;
370 	struct sched_domain *sd;
371 	int domain_num = 0, i;
372 	char buf[32];
373 
374 	for_each_domain(cpu, sd)
375 		domain_num++;
376 	entry = table = sd_alloc_ctl_entry(domain_num + 1);
377 	if (table == NULL)
378 		return NULL;
379 
380 	i = 0;
381 	for_each_domain(cpu, sd) {
382 		snprintf(buf, 32, "domain%d", i);
383 		entry->procname = kstrdup(buf, GFP_KERNEL);
384 		entry->mode = 0555;
385 		entry->child = sd_alloc_ctl_domain_table(sd);
386 		entry++;
387 		i++;
388 	}
389 	return table;
390 }
391 
392 static struct ctl_table_header *sd_sysctl_header;
register_sched_domain_sysctl(void)393 void register_sched_domain_sysctl(void)
394 {
395 	int i, cpu_num = num_possible_cpus();
396 	struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1);
397 	char buf[32];
398 
399 	WARN_ON(sd_ctl_dir[0].child);
400 	sd_ctl_dir[0].child = entry;
401 
402 	if (entry == NULL)
403 		return;
404 
405 	for_each_possible_cpu(i) {
406 		snprintf(buf, 32, "cpu%d", i);
407 		entry->procname = kstrdup(buf, GFP_KERNEL);
408 		entry->mode = 0555;
409 		entry->child = sd_alloc_ctl_cpu_table(i);
410 		entry++;
411 	}
412 
413 	WARN_ON(sd_sysctl_header);
414 	sd_sysctl_header = register_sysctl_table(sd_ctl_root);
415 }
416 
417 /* may be called multiple times per register */
unregister_sched_domain_sysctl(void)418 void unregister_sched_domain_sysctl(void)
419 {
420 	unregister_sysctl_table(sd_sysctl_header);
421 	sd_sysctl_header = NULL;
422 	if (sd_ctl_dir[0].child)
423 		sd_free_ctl_entry(&sd_ctl_dir[0].child);
424 }
425 #endif /* CONFIG_SYSCTL */
426 #endif /* CONFIG_SMP */
427 
428 #ifdef CONFIG_FAIR_GROUP_SCHED
print_cfs_group_stats(struct seq_file * m,int cpu,struct task_group * tg)429 static void print_cfs_group_stats(struct seq_file *m, int cpu, struct task_group *tg)
430 {
431 	struct sched_entity *se = tg->se[cpu];
432 
433 #define P(F) \
434 	SEQ_printf(m, "  .%-30s: %lld\n", #F, (long long)F)
435 #define P_SCHEDSTAT(F) \
436 	SEQ_printf(m, "  .%-30s: %lld\n", #F, (long long)schedstat_val(F))
437 #define PN(F) \
438 	SEQ_printf(m, "  .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F))
439 #define PN_SCHEDSTAT(F) \
440 	SEQ_printf(m, "  .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)schedstat_val(F)))
441 
442 	if (!se)
443 		return;
444 
445 	PN(se->exec_start);
446 	PN(se->vruntime);
447 	PN(se->sum_exec_runtime);
448 	if (schedstat_enabled()) {
449 		PN_SCHEDSTAT(se->statistics.wait_start);
450 		PN_SCHEDSTAT(se->statistics.sleep_start);
451 		PN_SCHEDSTAT(se->statistics.block_start);
452 		PN_SCHEDSTAT(se->statistics.sleep_max);
453 		PN_SCHEDSTAT(se->statistics.block_max);
454 		PN_SCHEDSTAT(se->statistics.exec_max);
455 		PN_SCHEDSTAT(se->statistics.slice_max);
456 		PN_SCHEDSTAT(se->statistics.wait_max);
457 		PN_SCHEDSTAT(se->statistics.wait_sum);
458 		P_SCHEDSTAT(se->statistics.wait_count);
459 	}
460 	P(se->load.weight);
461 #ifdef CONFIG_SMP
462 	P(se->avg.load_avg);
463 	P(se->avg.util_avg);
464 #endif
465 
466 #undef PN_SCHEDSTAT
467 #undef PN
468 #undef P_SCHEDSTAT
469 #undef P
470 }
471 #endif
472 
473 #ifdef CONFIG_CGROUP_SCHED
474 static char group_path[PATH_MAX];
475 
task_group_path(struct task_group * tg)476 static char *task_group_path(struct task_group *tg)
477 {
478 	if (autogroup_path(tg, group_path, PATH_MAX))
479 		return group_path;
480 
481 	cgroup_path(tg->css.cgroup, group_path, PATH_MAX);
482 	return group_path;
483 }
484 #endif
485 
486 static void
print_task(struct seq_file * m,struct rq * rq,struct task_struct * p)487 print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
488 {
489 	if (rq->curr == p)
490 		SEQ_printf(m, "R");
491 	else
492 		SEQ_printf(m, " ");
493 
494 	SEQ_printf(m, "%15s %5d %9Ld.%06ld %9Ld %5d ",
495 		p->comm, task_pid_nr(p),
496 		SPLIT_NS(p->se.vruntime),
497 		(long long)(p->nvcsw + p->nivcsw),
498 		p->prio);
499 
500 	SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
501 		SPLIT_NS(schedstat_val_or_zero(p->se.statistics.wait_sum)),
502 		SPLIT_NS(p->se.sum_exec_runtime),
503 		SPLIT_NS(schedstat_val_or_zero(p->se.statistics.sum_sleep_runtime)));
504 
505 #ifdef CONFIG_NUMA_BALANCING
506 	SEQ_printf(m, " %d %d", task_node(p), task_numa_group_id(p));
507 #endif
508 #ifdef CONFIG_CGROUP_SCHED
509 	SEQ_printf(m, " %s", task_group_path(task_group(p)));
510 #endif
511 
512 	SEQ_printf(m, "\n");
513 }
514 
print_rq(struct seq_file * m,struct rq * rq,int rq_cpu)515 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
516 {
517 	struct task_struct *g, *p;
518 
519 	SEQ_printf(m,
520 	"\nrunnable tasks:\n"
521 	"            task   PID         tree-key  switches  prio"
522 	"     wait-time             sum-exec        sum-sleep\n"
523 	"------------------------------------------------------"
524 	"----------------------------------------------------\n");
525 
526 	rcu_read_lock();
527 	for_each_process_thread(g, p) {
528 		if (task_cpu(p) != rq_cpu)
529 			continue;
530 
531 		print_task(m, rq, p);
532 	}
533 	rcu_read_unlock();
534 }
535 
print_cfs_rq(struct seq_file * m,int cpu,struct cfs_rq * cfs_rq)536 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
537 {
538 	s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
539 		spread, rq0_min_vruntime, spread0;
540 	struct rq *rq = cpu_rq(cpu);
541 	struct sched_entity *last;
542 	unsigned long flags;
543 
544 #ifdef CONFIG_FAIR_GROUP_SCHED
545 	SEQ_printf(m, "\ncfs_rq[%d]:%s\n", cpu, task_group_path(cfs_rq->tg));
546 #else
547 	SEQ_printf(m, "\ncfs_rq[%d]:\n", cpu);
548 #endif
549 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "exec_clock",
550 			SPLIT_NS(cfs_rq->exec_clock));
551 
552 	raw_spin_lock_irqsave(&rq->lock, flags);
553 	if (cfs_rq->rb_leftmost)
554 		MIN_vruntime = (__pick_first_entity(cfs_rq))->vruntime;
555 	last = __pick_last_entity(cfs_rq);
556 	if (last)
557 		max_vruntime = last->vruntime;
558 	min_vruntime = cfs_rq->min_vruntime;
559 	rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
560 	raw_spin_unlock_irqrestore(&rq->lock, flags);
561 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "MIN_vruntime",
562 			SPLIT_NS(MIN_vruntime));
563 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "min_vruntime",
564 			SPLIT_NS(min_vruntime));
565 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "max_vruntime",
566 			SPLIT_NS(max_vruntime));
567 	spread = max_vruntime - MIN_vruntime;
568 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread",
569 			SPLIT_NS(spread));
570 	spread0 = min_vruntime - rq0_min_vruntime;
571 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread0",
572 			SPLIT_NS(spread0));
573 	SEQ_printf(m, "  .%-30s: %d\n", "nr_spread_over",
574 			cfs_rq->nr_spread_over);
575 	SEQ_printf(m, "  .%-30s: %d\n", "nr_running", cfs_rq->nr_running);
576 	SEQ_printf(m, "  .%-30s: %ld\n", "load", cfs_rq->load.weight);
577 #ifdef CONFIG_SMP
578 	SEQ_printf(m, "  .%-30s: %lu\n", "load_avg",
579 			cfs_rq->avg.load_avg);
580 	SEQ_printf(m, "  .%-30s: %lu\n", "runnable_load_avg",
581 			cfs_rq->runnable_load_avg);
582 	SEQ_printf(m, "  .%-30s: %lu\n", "util_avg",
583 			cfs_rq->avg.util_avg);
584 	SEQ_printf(m, "  .%-30s: %ld\n", "removed_load_avg",
585 			atomic_long_read(&cfs_rq->removed_load_avg));
586 	SEQ_printf(m, "  .%-30s: %ld\n", "removed_util_avg",
587 			atomic_long_read(&cfs_rq->removed_util_avg));
588 #ifdef CONFIG_FAIR_GROUP_SCHED
589 	SEQ_printf(m, "  .%-30s: %lu\n", "tg_load_avg_contrib",
590 			cfs_rq->tg_load_avg_contrib);
591 	SEQ_printf(m, "  .%-30s: %ld\n", "tg_load_avg",
592 			atomic_long_read(&cfs_rq->tg->load_avg));
593 #endif
594 #endif
595 #ifdef CONFIG_CFS_BANDWIDTH
596 	SEQ_printf(m, "  .%-30s: %d\n", "throttled",
597 			cfs_rq->throttled);
598 	SEQ_printf(m, "  .%-30s: %d\n", "throttle_count",
599 			cfs_rq->throttle_count);
600 #endif
601 
602 #ifdef CONFIG_FAIR_GROUP_SCHED
603 	print_cfs_group_stats(m, cpu, cfs_rq->tg);
604 #endif
605 }
606 
print_rt_rq(struct seq_file * m,int cpu,struct rt_rq * rt_rq)607 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
608 {
609 #ifdef CONFIG_RT_GROUP_SCHED
610 	SEQ_printf(m, "\nrt_rq[%d]:%s\n", cpu, task_group_path(rt_rq->tg));
611 #else
612 	SEQ_printf(m, "\nrt_rq[%d]:\n", cpu);
613 #endif
614 
615 #define P(x) \
616 	SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
617 #define PN(x) \
618 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
619 
620 	P(rt_nr_running);
621 	P(rt_throttled);
622 	PN(rt_time);
623 	PN(rt_runtime);
624 
625 #undef PN
626 #undef P
627 }
628 
print_dl_rq(struct seq_file * m,int cpu,struct dl_rq * dl_rq)629 void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq)
630 {
631 	struct dl_bw *dl_bw;
632 
633 	SEQ_printf(m, "\ndl_rq[%d]:\n", cpu);
634 	SEQ_printf(m, "  .%-30s: %ld\n", "dl_nr_running", dl_rq->dl_nr_running);
635 #ifdef CONFIG_SMP
636 	dl_bw = &cpu_rq(cpu)->rd->dl_bw;
637 #else
638 	dl_bw = &dl_rq->dl_bw;
639 #endif
640 	SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->bw", dl_bw->bw);
641 	SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->total_bw", dl_bw->total_bw);
642 }
643 
644 extern __read_mostly int sched_clock_running;
645 
print_cpu(struct seq_file * m,int cpu)646 static void print_cpu(struct seq_file *m, int cpu)
647 {
648 	struct rq *rq = cpu_rq(cpu);
649 	unsigned long flags;
650 
651 #ifdef CONFIG_X86
652 	{
653 		unsigned int freq = cpu_khz ? : 1;
654 
655 		SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
656 			   cpu, freq / 1000, (freq % 1000));
657 	}
658 #else
659 	SEQ_printf(m, "cpu#%d\n", cpu);
660 #endif
661 
662 #define P(x)								\
663 do {									\
664 	if (sizeof(rq->x) == 4)						\
665 		SEQ_printf(m, "  .%-30s: %ld\n", #x, (long)(rq->x));	\
666 	else								\
667 		SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rq->x));\
668 } while (0)
669 
670 #define PN(x) \
671 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
672 
673 	P(nr_running);
674 	SEQ_printf(m, "  .%-30s: %lu\n", "load",
675 		   rq->load.weight);
676 	P(nr_switches);
677 	P(nr_load_updates);
678 	P(nr_uninterruptible);
679 	PN(next_balance);
680 	SEQ_printf(m, "  .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr)));
681 	PN(clock);
682 	PN(clock_task);
683 	P(cpu_load[0]);
684 	P(cpu_load[1]);
685 	P(cpu_load[2]);
686 	P(cpu_load[3]);
687 	P(cpu_load[4]);
688 #undef P
689 #undef PN
690 
691 #ifdef CONFIG_SMP
692 #define P64(n) SEQ_printf(m, "  .%-30s: %Ld\n", #n, rq->n);
693 	P64(avg_idle);
694 	P64(max_idle_balance_cost);
695 #undef P64
696 #endif
697 
698 #define P(n) SEQ_printf(m, "  .%-30s: %d\n", #n, schedstat_val(rq->n));
699 	if (schedstat_enabled()) {
700 		P(yld_count);
701 		P(sched_count);
702 		P(sched_goidle);
703 		P(ttwu_count);
704 		P(ttwu_local);
705 	}
706 #undef P
707 
708 	spin_lock_irqsave(&sched_debug_lock, flags);
709 	print_cfs_stats(m, cpu);
710 	print_rt_stats(m, cpu);
711 	print_dl_stats(m, cpu);
712 
713 	print_rq(m, rq, cpu);
714 	spin_unlock_irqrestore(&sched_debug_lock, flags);
715 	SEQ_printf(m, "\n");
716 }
717 
718 static const char *sched_tunable_scaling_names[] = {
719 	"none",
720 	"logaritmic",
721 	"linear"
722 };
723 
sched_debug_header(struct seq_file * m)724 static void sched_debug_header(struct seq_file *m)
725 {
726 	u64 ktime, sched_clk, cpu_clk;
727 	unsigned long flags;
728 
729 	local_irq_save(flags);
730 	ktime = ktime_to_ns(ktime_get());
731 	sched_clk = sched_clock();
732 	cpu_clk = local_clock();
733 	local_irq_restore(flags);
734 
735 	SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n",
736 		init_utsname()->release,
737 		(int)strcspn(init_utsname()->version, " "),
738 		init_utsname()->version);
739 
740 #define P(x) \
741 	SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
742 #define PN(x) \
743 	SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
744 	PN(ktime);
745 	PN(sched_clk);
746 	PN(cpu_clk);
747 	P(jiffies);
748 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
749 	P(sched_clock_stable());
750 #endif
751 #undef PN
752 #undef P
753 
754 	SEQ_printf(m, "\n");
755 	SEQ_printf(m, "sysctl_sched\n");
756 
757 #define P(x) \
758 	SEQ_printf(m, "  .%-40s: %Ld\n", #x, (long long)(x))
759 #define PN(x) \
760 	SEQ_printf(m, "  .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
761 	PN(sysctl_sched_latency);
762 	PN(sysctl_sched_min_granularity);
763 	PN(sysctl_sched_wakeup_granularity);
764 	P(sysctl_sched_child_runs_first);
765 	P(sysctl_sched_features);
766 #undef PN
767 #undef P
768 
769 	SEQ_printf(m, "  .%-40s: %d (%s)\n",
770 		"sysctl_sched_tunable_scaling",
771 		sysctl_sched_tunable_scaling,
772 		sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
773 	SEQ_printf(m, "\n");
774 }
775 
sched_debug_show(struct seq_file * m,void * v)776 static int sched_debug_show(struct seq_file *m, void *v)
777 {
778 	int cpu = (unsigned long)(v - 2);
779 
780 	if (cpu != -1)
781 		print_cpu(m, cpu);
782 	else
783 		sched_debug_header(m);
784 
785 	return 0;
786 }
787 
sysrq_sched_debug_show(void)788 void sysrq_sched_debug_show(void)
789 {
790 	int cpu;
791 
792 	sched_debug_header(NULL);
793 	for_each_online_cpu(cpu)
794 		print_cpu(NULL, cpu);
795 
796 }
797 
798 /*
799  * This itererator needs some explanation.
800  * It returns 1 for the header position.
801  * This means 2 is cpu 0.
802  * In a hotplugged system some cpus, including cpu 0, may be missing so we have
803  * to use cpumask_* to iterate over the cpus.
804  */
sched_debug_start(struct seq_file * file,loff_t * offset)805 static void *sched_debug_start(struct seq_file *file, loff_t *offset)
806 {
807 	unsigned long n = *offset;
808 
809 	if (n == 0)
810 		return (void *) 1;
811 
812 	n--;
813 
814 	if (n > 0)
815 		n = cpumask_next(n - 1, cpu_online_mask);
816 	else
817 		n = cpumask_first(cpu_online_mask);
818 
819 	*offset = n + 1;
820 
821 	if (n < nr_cpu_ids)
822 		return (void *)(unsigned long)(n + 2);
823 	return NULL;
824 }
825 
sched_debug_next(struct seq_file * file,void * data,loff_t * offset)826 static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
827 {
828 	(*offset)++;
829 	return sched_debug_start(file, offset);
830 }
831 
sched_debug_stop(struct seq_file * file,void * data)832 static void sched_debug_stop(struct seq_file *file, void *data)
833 {
834 }
835 
836 static const struct seq_operations sched_debug_sops = {
837 	.start = sched_debug_start,
838 	.next = sched_debug_next,
839 	.stop = sched_debug_stop,
840 	.show = sched_debug_show,
841 };
842 
sched_debug_release(struct inode * inode,struct file * file)843 static int sched_debug_release(struct inode *inode, struct file *file)
844 {
845 	seq_release(inode, file);
846 
847 	return 0;
848 }
849 
sched_debug_open(struct inode * inode,struct file * filp)850 static int sched_debug_open(struct inode *inode, struct file *filp)
851 {
852 	int ret = 0;
853 
854 	ret = seq_open(filp, &sched_debug_sops);
855 
856 	return ret;
857 }
858 
859 static const struct file_operations sched_debug_fops = {
860 	.open		= sched_debug_open,
861 	.read		= seq_read,
862 	.llseek		= seq_lseek,
863 	.release	= sched_debug_release,
864 };
865 
init_sched_debug_procfs(void)866 static int __init init_sched_debug_procfs(void)
867 {
868 	struct proc_dir_entry *pe;
869 
870 	pe = proc_create("sched_debug", 0444, NULL, &sched_debug_fops);
871 	if (!pe)
872 		return -ENOMEM;
873 	return 0;
874 }
875 
876 __initcall(init_sched_debug_procfs);
877 
878 #define __P(F) \
879 	SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)F)
880 #define P(F) \
881 	SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)p->F)
882 #define __PN(F) \
883 	SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
884 #define PN(F) \
885 	SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
886 
887 
888 #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)889 void print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
890 		unsigned long tpf, unsigned long gsf, unsigned long gpf)
891 {
892 	SEQ_printf(m, "numa_faults node=%d ", node);
893 	SEQ_printf(m, "task_private=%lu task_shared=%lu ", tsf, tpf);
894 	SEQ_printf(m, "group_private=%lu group_shared=%lu\n", gsf, gpf);
895 }
896 #endif
897 
898 
sched_show_numa(struct task_struct * p,struct seq_file * m)899 static void sched_show_numa(struct task_struct *p, struct seq_file *m)
900 {
901 #ifdef CONFIG_NUMA_BALANCING
902 	struct mempolicy *pol;
903 
904 	if (p->mm)
905 		P(mm->numa_scan_seq);
906 
907 	task_lock(p);
908 	pol = p->mempolicy;
909 	if (pol && !(pol->flags & MPOL_F_MORON))
910 		pol = NULL;
911 	mpol_get(pol);
912 	task_unlock(p);
913 
914 	P(numa_pages_migrated);
915 	P(numa_preferred_nid);
916 	P(total_numa_faults);
917 	SEQ_printf(m, "current_node=%d, numa_group_id=%d\n",
918 			task_node(p), task_numa_group_id(p));
919 	show_numa_stats(p, m);
920 	mpol_put(pol);
921 #endif
922 }
923 
proc_sched_show_task(struct task_struct * p,struct seq_file * m)924 void proc_sched_show_task(struct task_struct *p, struct seq_file *m)
925 {
926 	unsigned long nr_switches;
927 
928 	SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr(p),
929 						get_nr_threads(p));
930 	SEQ_printf(m,
931 		"---------------------------------------------------------"
932 		"----------\n");
933 #define __P(F) \
934 	SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)F)
935 #define P(F) \
936 	SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)p->F)
937 #define P_SCHEDSTAT(F) \
938 	SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)schedstat_val(p->F))
939 #define __PN(F) \
940 	SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
941 #define PN(F) \
942 	SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
943 #define PN_SCHEDSTAT(F) \
944 	SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)schedstat_val(p->F)))
945 
946 	PN(se.exec_start);
947 	PN(se.vruntime);
948 	PN(se.sum_exec_runtime);
949 
950 	nr_switches = p->nvcsw + p->nivcsw;
951 
952 	P(se.nr_migrations);
953 
954 	if (schedstat_enabled()) {
955 		u64 avg_atom, avg_per_cpu;
956 
957 		PN_SCHEDSTAT(se.statistics.sum_sleep_runtime);
958 		PN_SCHEDSTAT(se.statistics.wait_start);
959 		PN_SCHEDSTAT(se.statistics.sleep_start);
960 		PN_SCHEDSTAT(se.statistics.block_start);
961 		PN_SCHEDSTAT(se.statistics.sleep_max);
962 		PN_SCHEDSTAT(se.statistics.block_max);
963 		PN_SCHEDSTAT(se.statistics.exec_max);
964 		PN_SCHEDSTAT(se.statistics.slice_max);
965 		PN_SCHEDSTAT(se.statistics.wait_max);
966 		PN_SCHEDSTAT(se.statistics.wait_sum);
967 		P_SCHEDSTAT(se.statistics.wait_count);
968 		PN_SCHEDSTAT(se.statistics.iowait_sum);
969 		P_SCHEDSTAT(se.statistics.iowait_count);
970 		P_SCHEDSTAT(se.statistics.nr_migrations_cold);
971 		P_SCHEDSTAT(se.statistics.nr_failed_migrations_affine);
972 		P_SCHEDSTAT(se.statistics.nr_failed_migrations_running);
973 		P_SCHEDSTAT(se.statistics.nr_failed_migrations_hot);
974 		P_SCHEDSTAT(se.statistics.nr_forced_migrations);
975 		P_SCHEDSTAT(se.statistics.nr_wakeups);
976 		P_SCHEDSTAT(se.statistics.nr_wakeups_sync);
977 		P_SCHEDSTAT(se.statistics.nr_wakeups_migrate);
978 		P_SCHEDSTAT(se.statistics.nr_wakeups_local);
979 		P_SCHEDSTAT(se.statistics.nr_wakeups_remote);
980 		P_SCHEDSTAT(se.statistics.nr_wakeups_affine);
981 		P_SCHEDSTAT(se.statistics.nr_wakeups_affine_attempts);
982 		P_SCHEDSTAT(se.statistics.nr_wakeups_passive);
983 		P_SCHEDSTAT(se.statistics.nr_wakeups_idle);
984 		/* eas */
985 		/* select_idle_sibling() */
986 		P_SCHEDSTAT(se.statistics.nr_wakeups_sis_attempts);
987 		P_SCHEDSTAT(se.statistics.nr_wakeups_sis_idle);
988 		P_SCHEDSTAT(se.statistics.nr_wakeups_sis_cache_affine);
989 		P_SCHEDSTAT(se.statistics.nr_wakeups_sis_suff_cap);
990 		P_SCHEDSTAT(se.statistics.nr_wakeups_sis_idle_cpu);
991 		P_SCHEDSTAT(se.statistics.nr_wakeups_sis_count);
992 		/* select_energy_cpu_brute() */
993 		P_SCHEDSTAT(se.statistics.nr_wakeups_secb_attempts);
994 		P_SCHEDSTAT(se.statistics.nr_wakeups_secb_sync);
995 		P_SCHEDSTAT(se.statistics.nr_wakeups_secb_idle_bt);
996 		P_SCHEDSTAT(se.statistics.nr_wakeups_secb_insuff_cap);
997 		P_SCHEDSTAT(se.statistics.nr_wakeups_secb_no_nrg_sav);
998 		P_SCHEDSTAT(se.statistics.nr_wakeups_secb_nrg_sav);
999 		P_SCHEDSTAT(se.statistics.nr_wakeups_secb_count);
1000 		/* find_best_target() */
1001 		P_SCHEDSTAT(se.statistics.nr_wakeups_fbt_attempts);
1002 		P_SCHEDSTAT(se.statistics.nr_wakeups_fbt_no_cpu);
1003 		P_SCHEDSTAT(se.statistics.nr_wakeups_fbt_no_sd);
1004 		P_SCHEDSTAT(se.statistics.nr_wakeups_fbt_pref_idle);
1005 		P_SCHEDSTAT(se.statistics.nr_wakeups_fbt_count);
1006 		/* cas */
1007 		/* select_task_rq_fair() */
1008 		P_SCHEDSTAT(se.statistics.nr_wakeups_cas_attempts);
1009 		P_SCHEDSTAT(se.statistics.nr_wakeups_cas_count);
1010 
1011 		avg_atom = p->se.sum_exec_runtime;
1012 		if (nr_switches)
1013 			avg_atom = div64_ul(avg_atom, nr_switches);
1014 		else
1015 			avg_atom = -1LL;
1016 
1017 		avg_per_cpu = p->se.sum_exec_runtime;
1018 		if (p->se.nr_migrations) {
1019 			avg_per_cpu = div64_u64(avg_per_cpu,
1020 						p->se.nr_migrations);
1021 		} else {
1022 			avg_per_cpu = -1LL;
1023 		}
1024 
1025 		__PN(avg_atom);
1026 		__PN(avg_per_cpu);
1027 	}
1028 
1029 	__P(nr_switches);
1030 	SEQ_printf(m, "%-45s:%21Ld\n",
1031 		   "nr_voluntary_switches", (long long)p->nvcsw);
1032 	SEQ_printf(m, "%-45s:%21Ld\n",
1033 		   "nr_involuntary_switches", (long long)p->nivcsw);
1034 
1035 	P(se.load.weight);
1036 #ifdef CONFIG_SMP
1037 	P(se.avg.load_sum);
1038 	P(se.avg.util_sum);
1039 	P(se.avg.load_avg);
1040 	P(se.avg.util_avg);
1041 	P(se.avg.last_update_time);
1042 #endif
1043 	P(policy);
1044 	P(prio);
1045 #undef PN_SCHEDSTAT
1046 #undef PN
1047 #undef __PN
1048 #undef P_SCHEDSTAT
1049 #undef P
1050 #undef __P
1051 
1052 	{
1053 		unsigned int this_cpu = raw_smp_processor_id();
1054 		u64 t0, t1;
1055 
1056 		t0 = cpu_clock(this_cpu);
1057 		t1 = cpu_clock(this_cpu);
1058 		SEQ_printf(m, "%-45s:%21Ld\n",
1059 			   "clock-delta", (long long)(t1-t0));
1060 	}
1061 
1062 	sched_show_numa(p, m);
1063 }
1064 
proc_sched_set_task(struct task_struct * p)1065 void proc_sched_set_task(struct task_struct *p)
1066 {
1067 #ifdef CONFIG_SCHEDSTATS
1068 	memset(&p->se.statistics, 0, sizeof(p->se.statistics));
1069 #endif
1070 }
1071