xref: /linux/kernel/sched/debug.c (revision 4f58e6dceb0e44ca8f21568ed81e1df24e55964c)
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  */
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 
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 
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 
99 static void sched_feat_disable(int i)
100 {
101 	static_key_disable(&sched_feat_keys[i]);
102 }
103 
104 static void sched_feat_enable(int i)
105 {
106 	static_key_enable(&sched_feat_keys[i]);
107 }
108 #else
109 static void sched_feat_disable(int i) { };
110 static void sched_feat_enable(int i) { };
111 #endif /* HAVE_JUMP_LABEL */
112 
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
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 
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 
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 
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 
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
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 *
264 sd_alloc_ctl_domain_table(struct sched_domain *sd)
265 {
266 	struct ctl_table *table = sd_alloc_ctl_entry(14);
267 
268 	if (table == NULL)
269 		return NULL;
270 
271 	set_table_entry(&table[0], "min_interval", &sd->min_interval,
272 		sizeof(long), 0644, proc_doulongvec_minmax, false);
273 	set_table_entry(&table[1], "max_interval", &sd->max_interval,
274 		sizeof(long), 0644, proc_doulongvec_minmax, false);
275 	set_table_entry(&table[2], "busy_idx", &sd->busy_idx,
276 		sizeof(int), 0644, proc_dointvec_minmax, true);
277 	set_table_entry(&table[3], "idle_idx", &sd->idle_idx,
278 		sizeof(int), 0644, proc_dointvec_minmax, true);
279 	set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx,
280 		sizeof(int), 0644, proc_dointvec_minmax, true);
281 	set_table_entry(&table[5], "wake_idx", &sd->wake_idx,
282 		sizeof(int), 0644, proc_dointvec_minmax, true);
283 	set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx,
284 		sizeof(int), 0644, proc_dointvec_minmax, true);
285 	set_table_entry(&table[7], "busy_factor", &sd->busy_factor,
286 		sizeof(int), 0644, proc_dointvec_minmax, false);
287 	set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct,
288 		sizeof(int), 0644, proc_dointvec_minmax, false);
289 	set_table_entry(&table[9], "cache_nice_tries",
290 		&sd->cache_nice_tries,
291 		sizeof(int), 0644, proc_dointvec_minmax, false);
292 	set_table_entry(&table[10], "flags", &sd->flags,
293 		sizeof(int), 0644, proc_dointvec_minmax, false);
294 	set_table_entry(&table[11], "max_newidle_lb_cost",
295 		&sd->max_newidle_lb_cost,
296 		sizeof(long), 0644, proc_doulongvec_minmax, false);
297 	set_table_entry(&table[12], "name", sd->name,
298 		CORENAME_MAX_SIZE, 0444, proc_dostring, false);
299 	/* &table[13] is terminator */
300 
301 	return table;
302 }
303 
304 static struct ctl_table *sd_alloc_ctl_cpu_table(int cpu)
305 {
306 	struct ctl_table *entry, *table;
307 	struct sched_domain *sd;
308 	int domain_num = 0, i;
309 	char buf[32];
310 
311 	for_each_domain(cpu, sd)
312 		domain_num++;
313 	entry = table = sd_alloc_ctl_entry(domain_num + 1);
314 	if (table == NULL)
315 		return NULL;
316 
317 	i = 0;
318 	for_each_domain(cpu, sd) {
319 		snprintf(buf, 32, "domain%d", i);
320 		entry->procname = kstrdup(buf, GFP_KERNEL);
321 		entry->mode = 0555;
322 		entry->child = sd_alloc_ctl_domain_table(sd);
323 		entry++;
324 		i++;
325 	}
326 	return table;
327 }
328 
329 static struct ctl_table_header *sd_sysctl_header;
330 void register_sched_domain_sysctl(void)
331 {
332 	int i, cpu_num = num_possible_cpus();
333 	struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1);
334 	char buf[32];
335 
336 	WARN_ON(sd_ctl_dir[0].child);
337 	sd_ctl_dir[0].child = entry;
338 
339 	if (entry == NULL)
340 		return;
341 
342 	for_each_possible_cpu(i) {
343 		snprintf(buf, 32, "cpu%d", i);
344 		entry->procname = kstrdup(buf, GFP_KERNEL);
345 		entry->mode = 0555;
346 		entry->child = sd_alloc_ctl_cpu_table(i);
347 		entry++;
348 	}
349 
350 	WARN_ON(sd_sysctl_header);
351 	sd_sysctl_header = register_sysctl_table(sd_ctl_root);
352 }
353 
354 /* may be called multiple times per register */
355 void unregister_sched_domain_sysctl(void)
356 {
357 	unregister_sysctl_table(sd_sysctl_header);
358 	sd_sysctl_header = NULL;
359 	if (sd_ctl_dir[0].child)
360 		sd_free_ctl_entry(&sd_ctl_dir[0].child);
361 }
362 #endif /* CONFIG_SYSCTL */
363 #endif /* CONFIG_SMP */
364 
365 #ifdef CONFIG_FAIR_GROUP_SCHED
366 static void print_cfs_group_stats(struct seq_file *m, int cpu, struct task_group *tg)
367 {
368 	struct sched_entity *se = tg->se[cpu];
369 
370 #define P(F) \
371 	SEQ_printf(m, "  .%-30s: %lld\n", #F, (long long)F)
372 #define P_SCHEDSTAT(F) \
373 	SEQ_printf(m, "  .%-30s: %lld\n", #F, (long long)schedstat_val(F))
374 #define PN(F) \
375 	SEQ_printf(m, "  .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F))
376 #define PN_SCHEDSTAT(F) \
377 	SEQ_printf(m, "  .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)schedstat_val(F)))
378 
379 	if (!se)
380 		return;
381 
382 	PN(se->exec_start);
383 	PN(se->vruntime);
384 	PN(se->sum_exec_runtime);
385 	if (schedstat_enabled()) {
386 		PN_SCHEDSTAT(se->statistics.wait_start);
387 		PN_SCHEDSTAT(se->statistics.sleep_start);
388 		PN_SCHEDSTAT(se->statistics.block_start);
389 		PN_SCHEDSTAT(se->statistics.sleep_max);
390 		PN_SCHEDSTAT(se->statistics.block_max);
391 		PN_SCHEDSTAT(se->statistics.exec_max);
392 		PN_SCHEDSTAT(se->statistics.slice_max);
393 		PN_SCHEDSTAT(se->statistics.wait_max);
394 		PN_SCHEDSTAT(se->statistics.wait_sum);
395 		P_SCHEDSTAT(se->statistics.wait_count);
396 	}
397 	P(se->load.weight);
398 #ifdef CONFIG_SMP
399 	P(se->avg.load_avg);
400 	P(se->avg.util_avg);
401 #endif
402 
403 #undef PN_SCHEDSTAT
404 #undef PN
405 #undef P_SCHEDSTAT
406 #undef P
407 }
408 #endif
409 
410 #ifdef CONFIG_CGROUP_SCHED
411 static char group_path[PATH_MAX];
412 
413 static char *task_group_path(struct task_group *tg)
414 {
415 	if (autogroup_path(tg, group_path, PATH_MAX))
416 		return group_path;
417 
418 	return cgroup_path(tg->css.cgroup, group_path, PATH_MAX);
419 }
420 #endif
421 
422 static void
423 print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
424 {
425 	if (rq->curr == p)
426 		SEQ_printf(m, "R");
427 	else
428 		SEQ_printf(m, " ");
429 
430 	SEQ_printf(m, "%15s %5d %9Ld.%06ld %9Ld %5d ",
431 		p->comm, task_pid_nr(p),
432 		SPLIT_NS(p->se.vruntime),
433 		(long long)(p->nvcsw + p->nivcsw),
434 		p->prio);
435 
436 	SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
437 		SPLIT_NS(schedstat_val_or_zero(p->se.statistics.wait_sum)),
438 		SPLIT_NS(p->se.sum_exec_runtime),
439 		SPLIT_NS(schedstat_val_or_zero(p->se.statistics.sum_sleep_runtime)));
440 
441 #ifdef CONFIG_NUMA_BALANCING
442 	SEQ_printf(m, " %d %d", task_node(p), task_numa_group_id(p));
443 #endif
444 #ifdef CONFIG_CGROUP_SCHED
445 	SEQ_printf(m, " %s", task_group_path(task_group(p)));
446 #endif
447 
448 	SEQ_printf(m, "\n");
449 }
450 
451 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
452 {
453 	struct task_struct *g, *p;
454 
455 	SEQ_printf(m,
456 	"\nrunnable tasks:\n"
457 	"            task   PID         tree-key  switches  prio"
458 	"     wait-time             sum-exec        sum-sleep\n"
459 	"------------------------------------------------------"
460 	"----------------------------------------------------\n");
461 
462 	rcu_read_lock();
463 	for_each_process_thread(g, p) {
464 		if (task_cpu(p) != rq_cpu)
465 			continue;
466 
467 		print_task(m, rq, p);
468 	}
469 	rcu_read_unlock();
470 }
471 
472 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
473 {
474 	s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
475 		spread, rq0_min_vruntime, spread0;
476 	struct rq *rq = cpu_rq(cpu);
477 	struct sched_entity *last;
478 	unsigned long flags;
479 
480 #ifdef CONFIG_FAIR_GROUP_SCHED
481 	SEQ_printf(m, "\ncfs_rq[%d]:%s\n", cpu, task_group_path(cfs_rq->tg));
482 #else
483 	SEQ_printf(m, "\ncfs_rq[%d]:\n", cpu);
484 #endif
485 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "exec_clock",
486 			SPLIT_NS(cfs_rq->exec_clock));
487 
488 	raw_spin_lock_irqsave(&rq->lock, flags);
489 	if (cfs_rq->rb_leftmost)
490 		MIN_vruntime = (__pick_first_entity(cfs_rq))->vruntime;
491 	last = __pick_last_entity(cfs_rq);
492 	if (last)
493 		max_vruntime = last->vruntime;
494 	min_vruntime = cfs_rq->min_vruntime;
495 	rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
496 	raw_spin_unlock_irqrestore(&rq->lock, flags);
497 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "MIN_vruntime",
498 			SPLIT_NS(MIN_vruntime));
499 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "min_vruntime",
500 			SPLIT_NS(min_vruntime));
501 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "max_vruntime",
502 			SPLIT_NS(max_vruntime));
503 	spread = max_vruntime - MIN_vruntime;
504 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread",
505 			SPLIT_NS(spread));
506 	spread0 = min_vruntime - rq0_min_vruntime;
507 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread0",
508 			SPLIT_NS(spread0));
509 	SEQ_printf(m, "  .%-30s: %d\n", "nr_spread_over",
510 			cfs_rq->nr_spread_over);
511 	SEQ_printf(m, "  .%-30s: %d\n", "nr_running", cfs_rq->nr_running);
512 	SEQ_printf(m, "  .%-30s: %ld\n", "load", cfs_rq->load.weight);
513 #ifdef CONFIG_SMP
514 	SEQ_printf(m, "  .%-30s: %lu\n", "load_avg",
515 			cfs_rq->avg.load_avg);
516 	SEQ_printf(m, "  .%-30s: %lu\n", "runnable_load_avg",
517 			cfs_rq->runnable_load_avg);
518 	SEQ_printf(m, "  .%-30s: %lu\n", "util_avg",
519 			cfs_rq->avg.util_avg);
520 	SEQ_printf(m, "  .%-30s: %ld\n", "removed_load_avg",
521 			atomic_long_read(&cfs_rq->removed_load_avg));
522 	SEQ_printf(m, "  .%-30s: %ld\n", "removed_util_avg",
523 			atomic_long_read(&cfs_rq->removed_util_avg));
524 #ifdef CONFIG_FAIR_GROUP_SCHED
525 	SEQ_printf(m, "  .%-30s: %lu\n", "tg_load_avg_contrib",
526 			cfs_rq->tg_load_avg_contrib);
527 	SEQ_printf(m, "  .%-30s: %ld\n", "tg_load_avg",
528 			atomic_long_read(&cfs_rq->tg->load_avg));
529 #endif
530 #endif
531 #ifdef CONFIG_CFS_BANDWIDTH
532 	SEQ_printf(m, "  .%-30s: %d\n", "throttled",
533 			cfs_rq->throttled);
534 	SEQ_printf(m, "  .%-30s: %d\n", "throttle_count",
535 			cfs_rq->throttle_count);
536 #endif
537 
538 #ifdef CONFIG_FAIR_GROUP_SCHED
539 	print_cfs_group_stats(m, cpu, cfs_rq->tg);
540 #endif
541 }
542 
543 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
544 {
545 #ifdef CONFIG_RT_GROUP_SCHED
546 	SEQ_printf(m, "\nrt_rq[%d]:%s\n", cpu, task_group_path(rt_rq->tg));
547 #else
548 	SEQ_printf(m, "\nrt_rq[%d]:\n", cpu);
549 #endif
550 
551 #define P(x) \
552 	SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
553 #define PN(x) \
554 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
555 
556 	P(rt_nr_running);
557 	P(rt_throttled);
558 	PN(rt_time);
559 	PN(rt_runtime);
560 
561 #undef PN
562 #undef P
563 }
564 
565 void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq)
566 {
567 	struct dl_bw *dl_bw;
568 
569 	SEQ_printf(m, "\ndl_rq[%d]:\n", cpu);
570 	SEQ_printf(m, "  .%-30s: %ld\n", "dl_nr_running", dl_rq->dl_nr_running);
571 #ifdef CONFIG_SMP
572 	dl_bw = &cpu_rq(cpu)->rd->dl_bw;
573 #else
574 	dl_bw = &dl_rq->dl_bw;
575 #endif
576 	SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->bw", dl_bw->bw);
577 	SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->total_bw", dl_bw->total_bw);
578 }
579 
580 extern __read_mostly int sched_clock_running;
581 
582 static void print_cpu(struct seq_file *m, int cpu)
583 {
584 	struct rq *rq = cpu_rq(cpu);
585 	unsigned long flags;
586 
587 #ifdef CONFIG_X86
588 	{
589 		unsigned int freq = cpu_khz ? : 1;
590 
591 		SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
592 			   cpu, freq / 1000, (freq % 1000));
593 	}
594 #else
595 	SEQ_printf(m, "cpu#%d\n", cpu);
596 #endif
597 
598 #define P(x)								\
599 do {									\
600 	if (sizeof(rq->x) == 4)						\
601 		SEQ_printf(m, "  .%-30s: %ld\n", #x, (long)(rq->x));	\
602 	else								\
603 		SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rq->x));\
604 } while (0)
605 
606 #define PN(x) \
607 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
608 
609 	P(nr_running);
610 	SEQ_printf(m, "  .%-30s: %lu\n", "load",
611 		   rq->load.weight);
612 	P(nr_switches);
613 	P(nr_load_updates);
614 	P(nr_uninterruptible);
615 	PN(next_balance);
616 	SEQ_printf(m, "  .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr)));
617 	PN(clock);
618 	PN(clock_task);
619 	P(cpu_load[0]);
620 	P(cpu_load[1]);
621 	P(cpu_load[2]);
622 	P(cpu_load[3]);
623 	P(cpu_load[4]);
624 #undef P
625 #undef PN
626 
627 #ifdef CONFIG_SMP
628 #define P64(n) SEQ_printf(m, "  .%-30s: %Ld\n", #n, rq->n);
629 	P64(avg_idle);
630 	P64(max_idle_balance_cost);
631 #undef P64
632 #endif
633 
634 #define P(n) SEQ_printf(m, "  .%-30s: %d\n", #n, schedstat_val(rq->n));
635 	if (schedstat_enabled()) {
636 		P(yld_count);
637 		P(sched_count);
638 		P(sched_goidle);
639 		P(ttwu_count);
640 		P(ttwu_local);
641 	}
642 #undef P
643 
644 	spin_lock_irqsave(&sched_debug_lock, flags);
645 	print_cfs_stats(m, cpu);
646 	print_rt_stats(m, cpu);
647 	print_dl_stats(m, cpu);
648 
649 	print_rq(m, rq, cpu);
650 	spin_unlock_irqrestore(&sched_debug_lock, flags);
651 	SEQ_printf(m, "\n");
652 }
653 
654 static const char *sched_tunable_scaling_names[] = {
655 	"none",
656 	"logaritmic",
657 	"linear"
658 };
659 
660 static void sched_debug_header(struct seq_file *m)
661 {
662 	u64 ktime, sched_clk, cpu_clk;
663 	unsigned long flags;
664 
665 	local_irq_save(flags);
666 	ktime = ktime_to_ns(ktime_get());
667 	sched_clk = sched_clock();
668 	cpu_clk = local_clock();
669 	local_irq_restore(flags);
670 
671 	SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n",
672 		init_utsname()->release,
673 		(int)strcspn(init_utsname()->version, " "),
674 		init_utsname()->version);
675 
676 #define P(x) \
677 	SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
678 #define PN(x) \
679 	SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
680 	PN(ktime);
681 	PN(sched_clk);
682 	PN(cpu_clk);
683 	P(jiffies);
684 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
685 	P(sched_clock_stable());
686 #endif
687 #undef PN
688 #undef P
689 
690 	SEQ_printf(m, "\n");
691 	SEQ_printf(m, "sysctl_sched\n");
692 
693 #define P(x) \
694 	SEQ_printf(m, "  .%-40s: %Ld\n", #x, (long long)(x))
695 #define PN(x) \
696 	SEQ_printf(m, "  .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
697 	PN(sysctl_sched_latency);
698 	PN(sysctl_sched_min_granularity);
699 	PN(sysctl_sched_wakeup_granularity);
700 	P(sysctl_sched_child_runs_first);
701 	P(sysctl_sched_features);
702 #undef PN
703 #undef P
704 
705 	SEQ_printf(m, "  .%-40s: %d (%s)\n",
706 		"sysctl_sched_tunable_scaling",
707 		sysctl_sched_tunable_scaling,
708 		sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
709 	SEQ_printf(m, "\n");
710 }
711 
712 static int sched_debug_show(struct seq_file *m, void *v)
713 {
714 	int cpu = (unsigned long)(v - 2);
715 
716 	if (cpu != -1)
717 		print_cpu(m, cpu);
718 	else
719 		sched_debug_header(m);
720 
721 	return 0;
722 }
723 
724 void sysrq_sched_debug_show(void)
725 {
726 	int cpu;
727 
728 	sched_debug_header(NULL);
729 	for_each_online_cpu(cpu)
730 		print_cpu(NULL, cpu);
731 
732 }
733 
734 /*
735  * This itererator needs some explanation.
736  * It returns 1 for the header position.
737  * This means 2 is cpu 0.
738  * In a hotplugged system some cpus, including cpu 0, may be missing so we have
739  * to use cpumask_* to iterate over the cpus.
740  */
741 static void *sched_debug_start(struct seq_file *file, loff_t *offset)
742 {
743 	unsigned long n = *offset;
744 
745 	if (n == 0)
746 		return (void *) 1;
747 
748 	n--;
749 
750 	if (n > 0)
751 		n = cpumask_next(n - 1, cpu_online_mask);
752 	else
753 		n = cpumask_first(cpu_online_mask);
754 
755 	*offset = n + 1;
756 
757 	if (n < nr_cpu_ids)
758 		return (void *)(unsigned long)(n + 2);
759 	return NULL;
760 }
761 
762 static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
763 {
764 	(*offset)++;
765 	return sched_debug_start(file, offset);
766 }
767 
768 static void sched_debug_stop(struct seq_file *file, void *data)
769 {
770 }
771 
772 static const struct seq_operations sched_debug_sops = {
773 	.start = sched_debug_start,
774 	.next = sched_debug_next,
775 	.stop = sched_debug_stop,
776 	.show = sched_debug_show,
777 };
778 
779 static int sched_debug_release(struct inode *inode, struct file *file)
780 {
781 	seq_release(inode, file);
782 
783 	return 0;
784 }
785 
786 static int sched_debug_open(struct inode *inode, struct file *filp)
787 {
788 	int ret = 0;
789 
790 	ret = seq_open(filp, &sched_debug_sops);
791 
792 	return ret;
793 }
794 
795 static const struct file_operations sched_debug_fops = {
796 	.open		= sched_debug_open,
797 	.read		= seq_read,
798 	.llseek		= seq_lseek,
799 	.release	= sched_debug_release,
800 };
801 
802 static int __init init_sched_debug_procfs(void)
803 {
804 	struct proc_dir_entry *pe;
805 
806 	pe = proc_create("sched_debug", 0444, NULL, &sched_debug_fops);
807 	if (!pe)
808 		return -ENOMEM;
809 	return 0;
810 }
811 
812 __initcall(init_sched_debug_procfs);
813 
814 #define __P(F) \
815 	SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)F)
816 #define P(F) \
817 	SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)p->F)
818 #define __PN(F) \
819 	SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
820 #define PN(F) \
821 	SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
822 
823 
824 #ifdef CONFIG_NUMA_BALANCING
825 void print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
826 		unsigned long tpf, unsigned long gsf, unsigned long gpf)
827 {
828 	SEQ_printf(m, "numa_faults node=%d ", node);
829 	SEQ_printf(m, "task_private=%lu task_shared=%lu ", tsf, tpf);
830 	SEQ_printf(m, "group_private=%lu group_shared=%lu\n", gsf, gpf);
831 }
832 #endif
833 
834 
835 static void sched_show_numa(struct task_struct *p, struct seq_file *m)
836 {
837 #ifdef CONFIG_NUMA_BALANCING
838 	struct mempolicy *pol;
839 
840 	if (p->mm)
841 		P(mm->numa_scan_seq);
842 
843 	task_lock(p);
844 	pol = p->mempolicy;
845 	if (pol && !(pol->flags & MPOL_F_MORON))
846 		pol = NULL;
847 	mpol_get(pol);
848 	task_unlock(p);
849 
850 	P(numa_pages_migrated);
851 	P(numa_preferred_nid);
852 	P(total_numa_faults);
853 	SEQ_printf(m, "current_node=%d, numa_group_id=%d\n",
854 			task_node(p), task_numa_group_id(p));
855 	show_numa_stats(p, m);
856 	mpol_put(pol);
857 #endif
858 }
859 
860 void proc_sched_show_task(struct task_struct *p, struct seq_file *m)
861 {
862 	unsigned long nr_switches;
863 
864 	SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr(p),
865 						get_nr_threads(p));
866 	SEQ_printf(m,
867 		"---------------------------------------------------------"
868 		"----------\n");
869 #define __P(F) \
870 	SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)F)
871 #define P(F) \
872 	SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)p->F)
873 #define P_SCHEDSTAT(F) \
874 	SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)schedstat_val(p->F))
875 #define __PN(F) \
876 	SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
877 #define PN(F) \
878 	SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
879 #define PN_SCHEDSTAT(F) \
880 	SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)schedstat_val(p->F)))
881 
882 	PN(se.exec_start);
883 	PN(se.vruntime);
884 	PN(se.sum_exec_runtime);
885 
886 	nr_switches = p->nvcsw + p->nivcsw;
887 
888 	P(se.nr_migrations);
889 
890 	if (schedstat_enabled()) {
891 		u64 avg_atom, avg_per_cpu;
892 
893 		PN_SCHEDSTAT(se.statistics.sum_sleep_runtime);
894 		PN_SCHEDSTAT(se.statistics.wait_start);
895 		PN_SCHEDSTAT(se.statistics.sleep_start);
896 		PN_SCHEDSTAT(se.statistics.block_start);
897 		PN_SCHEDSTAT(se.statistics.sleep_max);
898 		PN_SCHEDSTAT(se.statistics.block_max);
899 		PN_SCHEDSTAT(se.statistics.exec_max);
900 		PN_SCHEDSTAT(se.statistics.slice_max);
901 		PN_SCHEDSTAT(se.statistics.wait_max);
902 		PN_SCHEDSTAT(se.statistics.wait_sum);
903 		P_SCHEDSTAT(se.statistics.wait_count);
904 		PN_SCHEDSTAT(se.statistics.iowait_sum);
905 		P_SCHEDSTAT(se.statistics.iowait_count);
906 		P_SCHEDSTAT(se.statistics.nr_migrations_cold);
907 		P_SCHEDSTAT(se.statistics.nr_failed_migrations_affine);
908 		P_SCHEDSTAT(se.statistics.nr_failed_migrations_running);
909 		P_SCHEDSTAT(se.statistics.nr_failed_migrations_hot);
910 		P_SCHEDSTAT(se.statistics.nr_forced_migrations);
911 		P_SCHEDSTAT(se.statistics.nr_wakeups);
912 		P_SCHEDSTAT(se.statistics.nr_wakeups_sync);
913 		P_SCHEDSTAT(se.statistics.nr_wakeups_migrate);
914 		P_SCHEDSTAT(se.statistics.nr_wakeups_local);
915 		P_SCHEDSTAT(se.statistics.nr_wakeups_remote);
916 		P_SCHEDSTAT(se.statistics.nr_wakeups_affine);
917 		P_SCHEDSTAT(se.statistics.nr_wakeups_affine_attempts);
918 		P_SCHEDSTAT(se.statistics.nr_wakeups_passive);
919 		P_SCHEDSTAT(se.statistics.nr_wakeups_idle);
920 
921 		avg_atom = p->se.sum_exec_runtime;
922 		if (nr_switches)
923 			avg_atom = div64_ul(avg_atom, nr_switches);
924 		else
925 			avg_atom = -1LL;
926 
927 		avg_per_cpu = p->se.sum_exec_runtime;
928 		if (p->se.nr_migrations) {
929 			avg_per_cpu = div64_u64(avg_per_cpu,
930 						p->se.nr_migrations);
931 		} else {
932 			avg_per_cpu = -1LL;
933 		}
934 
935 		__PN(avg_atom);
936 		__PN(avg_per_cpu);
937 	}
938 
939 	__P(nr_switches);
940 	SEQ_printf(m, "%-45s:%21Ld\n",
941 		   "nr_voluntary_switches", (long long)p->nvcsw);
942 	SEQ_printf(m, "%-45s:%21Ld\n",
943 		   "nr_involuntary_switches", (long long)p->nivcsw);
944 
945 	P(se.load.weight);
946 #ifdef CONFIG_SMP
947 	P(se.avg.load_sum);
948 	P(se.avg.util_sum);
949 	P(se.avg.load_avg);
950 	P(se.avg.util_avg);
951 	P(se.avg.last_update_time);
952 #endif
953 	P(policy);
954 	P(prio);
955 #undef PN_SCHEDSTAT
956 #undef PN
957 #undef __PN
958 #undef P_SCHEDSTAT
959 #undef P
960 #undef __P
961 
962 	{
963 		unsigned int this_cpu = raw_smp_processor_id();
964 		u64 t0, t1;
965 
966 		t0 = cpu_clock(this_cpu);
967 		t1 = cpu_clock(this_cpu);
968 		SEQ_printf(m, "%-45s:%21Ld\n",
969 			   "clock-delta", (long long)(t1-t0));
970 	}
971 
972 	sched_show_numa(p, m);
973 }
974 
975 void proc_sched_set_task(struct task_struct *p)
976 {
977 #ifdef CONFIG_SCHEDSTATS
978 	memset(&p->se.statistics, 0, sizeof(p->se.statistics));
979 #endif
980 }
981