xref: /linux/kernel/sched/debug.c (revision e5c86679d5e864947a52fb31e45a425dea3e7fa9)
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/mm.h>
15 #include <linux/sched/task.h>
16 #include <linux/seq_file.h>
17 #include <linux/kallsyms.h>
18 #include <linux/utsname.h>
19 #include <linux/mempolicy.h>
20 #include <linux/debugfs.h>
21 
22 #include "sched.h"
23 
24 static DEFINE_SPINLOCK(sched_debug_lock);
25 
26 /*
27  * This allows printing both to /proc/sched_debug and
28  * to the console
29  */
30 #define SEQ_printf(m, x...)			\
31  do {						\
32 	if (m)					\
33 		seq_printf(m, x);		\
34 	else					\
35 		printk(x);			\
36  } while (0)
37 
38 /*
39  * Ease the printing of nsec fields:
40  */
41 static long long nsec_high(unsigned long long nsec)
42 {
43 	if ((long long)nsec < 0) {
44 		nsec = -nsec;
45 		do_div(nsec, 1000000);
46 		return -nsec;
47 	}
48 	do_div(nsec, 1000000);
49 
50 	return nsec;
51 }
52 
53 static unsigned long nsec_low(unsigned long long nsec)
54 {
55 	if ((long long)nsec < 0)
56 		nsec = -nsec;
57 
58 	return do_div(nsec, 1000000);
59 }
60 
61 #define SPLIT_NS(x) nsec_high(x), nsec_low(x)
62 
63 #define SCHED_FEAT(name, enabled)	\
64 	#name ,
65 
66 static const char * const sched_feat_names[] = {
67 #include "features.h"
68 };
69 
70 #undef SCHED_FEAT
71 
72 static int sched_feat_show(struct seq_file *m, void *v)
73 {
74 	int i;
75 
76 	for (i = 0; i < __SCHED_FEAT_NR; i++) {
77 		if (!(sysctl_sched_features & (1UL << i)))
78 			seq_puts(m, "NO_");
79 		seq_printf(m, "%s ", sched_feat_names[i]);
80 	}
81 	seq_puts(m, "\n");
82 
83 	return 0;
84 }
85 
86 #ifdef HAVE_JUMP_LABEL
87 
88 #define jump_label_key__true  STATIC_KEY_INIT_TRUE
89 #define jump_label_key__false STATIC_KEY_INIT_FALSE
90 
91 #define SCHED_FEAT(name, enabled)	\
92 	jump_label_key__##enabled ,
93 
94 struct static_key sched_feat_keys[__SCHED_FEAT_NR] = {
95 #include "features.h"
96 };
97 
98 #undef SCHED_FEAT
99 
100 static void sched_feat_disable(int i)
101 {
102 	static_key_disable(&sched_feat_keys[i]);
103 }
104 
105 static void sched_feat_enable(int i)
106 {
107 	static_key_enable(&sched_feat_keys[i]);
108 }
109 #else
110 static void sched_feat_disable(int i) { };
111 static void sched_feat_enable(int i) { };
112 #endif /* HAVE_JUMP_LABEL */
113 
114 static int sched_feat_set(char *cmp)
115 {
116 	int i;
117 	int neg = 0;
118 
119 	if (strncmp(cmp, "NO_", 3) == 0) {
120 		neg = 1;
121 		cmp += 3;
122 	}
123 
124 	for (i = 0; i < __SCHED_FEAT_NR; i++) {
125 		if (strcmp(cmp, sched_feat_names[i]) == 0) {
126 			if (neg) {
127 				sysctl_sched_features &= ~(1UL << i);
128 				sched_feat_disable(i);
129 			} else {
130 				sysctl_sched_features |= (1UL << i);
131 				sched_feat_enable(i);
132 			}
133 			break;
134 		}
135 	}
136 
137 	return i;
138 }
139 
140 static ssize_t
141 sched_feat_write(struct file *filp, const char __user *ubuf,
142 		size_t cnt, loff_t *ppos)
143 {
144 	char buf[64];
145 	char *cmp;
146 	int i;
147 	struct inode *inode;
148 
149 	if (cnt > 63)
150 		cnt = 63;
151 
152 	if (copy_from_user(&buf, ubuf, cnt))
153 		return -EFAULT;
154 
155 	buf[cnt] = 0;
156 	cmp = strstrip(buf);
157 
158 	/* Ensure the static_key remains in a consistent state */
159 	inode = file_inode(filp);
160 	inode_lock(inode);
161 	i = sched_feat_set(cmp);
162 	inode_unlock(inode);
163 	if (i == __SCHED_FEAT_NR)
164 		return -EINVAL;
165 
166 	*ppos += cnt;
167 
168 	return cnt;
169 }
170 
171 static int sched_feat_open(struct inode *inode, struct file *filp)
172 {
173 	return single_open(filp, sched_feat_show, NULL);
174 }
175 
176 static const struct file_operations sched_feat_fops = {
177 	.open		= sched_feat_open,
178 	.write		= sched_feat_write,
179 	.read		= seq_read,
180 	.llseek		= seq_lseek,
181 	.release	= single_release,
182 };
183 
184 static __init int sched_init_debug(void)
185 {
186 	debugfs_create_file("sched_features", 0644, NULL, NULL,
187 			&sched_feat_fops);
188 
189 	return 0;
190 }
191 late_initcall(sched_init_debug);
192 
193 #ifdef CONFIG_SMP
194 
195 #ifdef CONFIG_SYSCTL
196 
197 static struct ctl_table sd_ctl_dir[] = {
198 	{
199 		.procname	= "sched_domain",
200 		.mode		= 0555,
201 	},
202 	{}
203 };
204 
205 static struct ctl_table sd_ctl_root[] = {
206 	{
207 		.procname	= "kernel",
208 		.mode		= 0555,
209 		.child		= sd_ctl_dir,
210 	},
211 	{}
212 };
213 
214 static struct ctl_table *sd_alloc_ctl_entry(int n)
215 {
216 	struct ctl_table *entry =
217 		kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL);
218 
219 	return entry;
220 }
221 
222 static void sd_free_ctl_entry(struct ctl_table **tablep)
223 {
224 	struct ctl_table *entry;
225 
226 	/*
227 	 * In the intermediate directories, both the child directory and
228 	 * procname are dynamically allocated and could fail but the mode
229 	 * will always be set. In the lowest directory the names are
230 	 * static strings and all have proc handlers.
231 	 */
232 	for (entry = *tablep; entry->mode; entry++) {
233 		if (entry->child)
234 			sd_free_ctl_entry(&entry->child);
235 		if (entry->proc_handler == NULL)
236 			kfree(entry->procname);
237 	}
238 
239 	kfree(*tablep);
240 	*tablep = NULL;
241 }
242 
243 static int min_load_idx = 0;
244 static int max_load_idx = CPU_LOAD_IDX_MAX-1;
245 
246 static void
247 set_table_entry(struct ctl_table *entry,
248 		const char *procname, void *data, int maxlen,
249 		umode_t mode, proc_handler *proc_handler,
250 		bool load_idx)
251 {
252 	entry->procname = procname;
253 	entry->data = data;
254 	entry->maxlen = maxlen;
255 	entry->mode = mode;
256 	entry->proc_handler = proc_handler;
257 
258 	if (load_idx) {
259 		entry->extra1 = &min_load_idx;
260 		entry->extra2 = &max_load_idx;
261 	}
262 }
263 
264 static struct ctl_table *
265 sd_alloc_ctl_domain_table(struct sched_domain *sd)
266 {
267 	struct ctl_table *table = sd_alloc_ctl_entry(14);
268 
269 	if (table == NULL)
270 		return NULL;
271 
272 	set_table_entry(&table[0], "min_interval", &sd->min_interval,
273 		sizeof(long), 0644, proc_doulongvec_minmax, false);
274 	set_table_entry(&table[1], "max_interval", &sd->max_interval,
275 		sizeof(long), 0644, proc_doulongvec_minmax, false);
276 	set_table_entry(&table[2], "busy_idx", &sd->busy_idx,
277 		sizeof(int), 0644, proc_dointvec_minmax, true);
278 	set_table_entry(&table[3], "idle_idx", &sd->idle_idx,
279 		sizeof(int), 0644, proc_dointvec_minmax, true);
280 	set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx,
281 		sizeof(int), 0644, proc_dointvec_minmax, true);
282 	set_table_entry(&table[5], "wake_idx", &sd->wake_idx,
283 		sizeof(int), 0644, proc_dointvec_minmax, true);
284 	set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx,
285 		sizeof(int), 0644, proc_dointvec_minmax, true);
286 	set_table_entry(&table[7], "busy_factor", &sd->busy_factor,
287 		sizeof(int), 0644, proc_dointvec_minmax, false);
288 	set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct,
289 		sizeof(int), 0644, proc_dointvec_minmax, false);
290 	set_table_entry(&table[9], "cache_nice_tries",
291 		&sd->cache_nice_tries,
292 		sizeof(int), 0644, proc_dointvec_minmax, false);
293 	set_table_entry(&table[10], "flags", &sd->flags,
294 		sizeof(int), 0644, proc_dointvec_minmax, false);
295 	set_table_entry(&table[11], "max_newidle_lb_cost",
296 		&sd->max_newidle_lb_cost,
297 		sizeof(long), 0644, proc_doulongvec_minmax, false);
298 	set_table_entry(&table[12], "name", sd->name,
299 		CORENAME_MAX_SIZE, 0444, proc_dostring, false);
300 	/* &table[13] is terminator */
301 
302 	return table;
303 }
304 
305 static struct ctl_table *sd_alloc_ctl_cpu_table(int cpu)
306 {
307 	struct ctl_table *entry, *table;
308 	struct sched_domain *sd;
309 	int domain_num = 0, i;
310 	char buf[32];
311 
312 	for_each_domain(cpu, sd)
313 		domain_num++;
314 	entry = table = sd_alloc_ctl_entry(domain_num + 1);
315 	if (table == NULL)
316 		return NULL;
317 
318 	i = 0;
319 	for_each_domain(cpu, sd) {
320 		snprintf(buf, 32, "domain%d", i);
321 		entry->procname = kstrdup(buf, GFP_KERNEL);
322 		entry->mode = 0555;
323 		entry->child = sd_alloc_ctl_domain_table(sd);
324 		entry++;
325 		i++;
326 	}
327 	return table;
328 }
329 
330 static struct ctl_table_header *sd_sysctl_header;
331 void register_sched_domain_sysctl(void)
332 {
333 	int i, cpu_num = num_possible_cpus();
334 	struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1);
335 	char buf[32];
336 
337 	WARN_ON(sd_ctl_dir[0].child);
338 	sd_ctl_dir[0].child = entry;
339 
340 	if (entry == NULL)
341 		return;
342 
343 	for_each_possible_cpu(i) {
344 		snprintf(buf, 32, "cpu%d", i);
345 		entry->procname = kstrdup(buf, GFP_KERNEL);
346 		entry->mode = 0555;
347 		entry->child = sd_alloc_ctl_cpu_table(i);
348 		entry++;
349 	}
350 
351 	WARN_ON(sd_sysctl_header);
352 	sd_sysctl_header = register_sysctl_table(sd_ctl_root);
353 }
354 
355 /* may be called multiple times per register */
356 void unregister_sched_domain_sysctl(void)
357 {
358 	unregister_sysctl_table(sd_sysctl_header);
359 	sd_sysctl_header = NULL;
360 	if (sd_ctl_dir[0].child)
361 		sd_free_ctl_entry(&sd_ctl_dir[0].child);
362 }
363 #endif /* CONFIG_SYSCTL */
364 #endif /* CONFIG_SMP */
365 
366 #ifdef CONFIG_FAIR_GROUP_SCHED
367 static void print_cfs_group_stats(struct seq_file *m, int cpu, struct task_group *tg)
368 {
369 	struct sched_entity *se = tg->se[cpu];
370 
371 #define P(F) \
372 	SEQ_printf(m, "  .%-30s: %lld\n", #F, (long long)F)
373 #define P_SCHEDSTAT(F) \
374 	SEQ_printf(m, "  .%-30s: %lld\n", #F, (long long)schedstat_val(F))
375 #define PN(F) \
376 	SEQ_printf(m, "  .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F))
377 #define PN_SCHEDSTAT(F) \
378 	SEQ_printf(m, "  .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)schedstat_val(F)))
379 
380 	if (!se)
381 		return;
382 
383 	PN(se->exec_start);
384 	PN(se->vruntime);
385 	PN(se->sum_exec_runtime);
386 	if (schedstat_enabled()) {
387 		PN_SCHEDSTAT(se->statistics.wait_start);
388 		PN_SCHEDSTAT(se->statistics.sleep_start);
389 		PN_SCHEDSTAT(se->statistics.block_start);
390 		PN_SCHEDSTAT(se->statistics.sleep_max);
391 		PN_SCHEDSTAT(se->statistics.block_max);
392 		PN_SCHEDSTAT(se->statistics.exec_max);
393 		PN_SCHEDSTAT(se->statistics.slice_max);
394 		PN_SCHEDSTAT(se->statistics.wait_max);
395 		PN_SCHEDSTAT(se->statistics.wait_sum);
396 		P_SCHEDSTAT(se->statistics.wait_count);
397 	}
398 	P(se->load.weight);
399 #ifdef CONFIG_SMP
400 	P(se->avg.load_avg);
401 	P(se->avg.util_avg);
402 #endif
403 
404 #undef PN_SCHEDSTAT
405 #undef PN
406 #undef P_SCHEDSTAT
407 #undef P
408 }
409 #endif
410 
411 #ifdef CONFIG_CGROUP_SCHED
412 static char group_path[PATH_MAX];
413 
414 static char *task_group_path(struct task_group *tg)
415 {
416 	if (autogroup_path(tg, group_path, PATH_MAX))
417 		return group_path;
418 
419 	cgroup_path(tg->css.cgroup, group_path, PATH_MAX);
420 	return group_path;
421 }
422 #endif
423 
424 static void
425 print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
426 {
427 	if (rq->curr == p)
428 		SEQ_printf(m, "R");
429 	else
430 		SEQ_printf(m, " ");
431 
432 	SEQ_printf(m, "%15s %5d %9Ld.%06ld %9Ld %5d ",
433 		p->comm, task_pid_nr(p),
434 		SPLIT_NS(p->se.vruntime),
435 		(long long)(p->nvcsw + p->nivcsw),
436 		p->prio);
437 
438 	SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
439 		SPLIT_NS(schedstat_val_or_zero(p->se.statistics.wait_sum)),
440 		SPLIT_NS(p->se.sum_exec_runtime),
441 		SPLIT_NS(schedstat_val_or_zero(p->se.statistics.sum_sleep_runtime)));
442 
443 #ifdef CONFIG_NUMA_BALANCING
444 	SEQ_printf(m, " %d %d", task_node(p), task_numa_group_id(p));
445 #endif
446 #ifdef CONFIG_CGROUP_SCHED
447 	SEQ_printf(m, " %s", task_group_path(task_group(p)));
448 #endif
449 
450 	SEQ_printf(m, "\n");
451 }
452 
453 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
454 {
455 	struct task_struct *g, *p;
456 
457 	SEQ_printf(m,
458 	"\nrunnable tasks:\n"
459 	"            task   PID         tree-key  switches  prio"
460 	"     wait-time             sum-exec        sum-sleep\n"
461 	"------------------------------------------------------"
462 	"----------------------------------------------------\n");
463 
464 	rcu_read_lock();
465 	for_each_process_thread(g, p) {
466 		if (task_cpu(p) != rq_cpu)
467 			continue;
468 
469 		print_task(m, rq, p);
470 	}
471 	rcu_read_unlock();
472 }
473 
474 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
475 {
476 	s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
477 		spread, rq0_min_vruntime, spread0;
478 	struct rq *rq = cpu_rq(cpu);
479 	struct sched_entity *last;
480 	unsigned long flags;
481 
482 #ifdef CONFIG_FAIR_GROUP_SCHED
483 	SEQ_printf(m, "\ncfs_rq[%d]:%s\n", cpu, task_group_path(cfs_rq->tg));
484 #else
485 	SEQ_printf(m, "\ncfs_rq[%d]:\n", cpu);
486 #endif
487 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "exec_clock",
488 			SPLIT_NS(cfs_rq->exec_clock));
489 
490 	raw_spin_lock_irqsave(&rq->lock, flags);
491 	if (cfs_rq->rb_leftmost)
492 		MIN_vruntime = (__pick_first_entity(cfs_rq))->vruntime;
493 	last = __pick_last_entity(cfs_rq);
494 	if (last)
495 		max_vruntime = last->vruntime;
496 	min_vruntime = cfs_rq->min_vruntime;
497 	rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
498 	raw_spin_unlock_irqrestore(&rq->lock, flags);
499 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "MIN_vruntime",
500 			SPLIT_NS(MIN_vruntime));
501 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "min_vruntime",
502 			SPLIT_NS(min_vruntime));
503 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "max_vruntime",
504 			SPLIT_NS(max_vruntime));
505 	spread = max_vruntime - MIN_vruntime;
506 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread",
507 			SPLIT_NS(spread));
508 	spread0 = min_vruntime - rq0_min_vruntime;
509 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread0",
510 			SPLIT_NS(spread0));
511 	SEQ_printf(m, "  .%-30s: %d\n", "nr_spread_over",
512 			cfs_rq->nr_spread_over);
513 	SEQ_printf(m, "  .%-30s: %d\n", "nr_running", cfs_rq->nr_running);
514 	SEQ_printf(m, "  .%-30s: %ld\n", "load", cfs_rq->load.weight);
515 #ifdef CONFIG_SMP
516 	SEQ_printf(m, "  .%-30s: %lu\n", "load_avg",
517 			cfs_rq->avg.load_avg);
518 	SEQ_printf(m, "  .%-30s: %lu\n", "runnable_load_avg",
519 			cfs_rq->runnable_load_avg);
520 	SEQ_printf(m, "  .%-30s: %lu\n", "util_avg",
521 			cfs_rq->avg.util_avg);
522 	SEQ_printf(m, "  .%-30s: %ld\n", "removed_load_avg",
523 			atomic_long_read(&cfs_rq->removed_load_avg));
524 	SEQ_printf(m, "  .%-30s: %ld\n", "removed_util_avg",
525 			atomic_long_read(&cfs_rq->removed_util_avg));
526 #ifdef CONFIG_FAIR_GROUP_SCHED
527 	SEQ_printf(m, "  .%-30s: %lu\n", "tg_load_avg_contrib",
528 			cfs_rq->tg_load_avg_contrib);
529 	SEQ_printf(m, "  .%-30s: %ld\n", "tg_load_avg",
530 			atomic_long_read(&cfs_rq->tg->load_avg));
531 #endif
532 #endif
533 #ifdef CONFIG_CFS_BANDWIDTH
534 	SEQ_printf(m, "  .%-30s: %d\n", "throttled",
535 			cfs_rq->throttled);
536 	SEQ_printf(m, "  .%-30s: %d\n", "throttle_count",
537 			cfs_rq->throttle_count);
538 #endif
539 
540 #ifdef CONFIG_FAIR_GROUP_SCHED
541 	print_cfs_group_stats(m, cpu, cfs_rq->tg);
542 #endif
543 }
544 
545 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
546 {
547 #ifdef CONFIG_RT_GROUP_SCHED
548 	SEQ_printf(m, "\nrt_rq[%d]:%s\n", cpu, task_group_path(rt_rq->tg));
549 #else
550 	SEQ_printf(m, "\nrt_rq[%d]:\n", cpu);
551 #endif
552 
553 #define P(x) \
554 	SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
555 #define PN(x) \
556 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
557 
558 	P(rt_nr_running);
559 	P(rt_throttled);
560 	PN(rt_time);
561 	PN(rt_runtime);
562 
563 #undef PN
564 #undef P
565 }
566 
567 void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq)
568 {
569 	struct dl_bw *dl_bw;
570 
571 	SEQ_printf(m, "\ndl_rq[%d]:\n", cpu);
572 	SEQ_printf(m, "  .%-30s: %ld\n", "dl_nr_running", dl_rq->dl_nr_running);
573 #ifdef CONFIG_SMP
574 	dl_bw = &cpu_rq(cpu)->rd->dl_bw;
575 #else
576 	dl_bw = &dl_rq->dl_bw;
577 #endif
578 	SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->bw", dl_bw->bw);
579 	SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->total_bw", dl_bw->total_bw);
580 }
581 
582 extern __read_mostly int sched_clock_running;
583 
584 static void print_cpu(struct seq_file *m, int cpu)
585 {
586 	struct rq *rq = cpu_rq(cpu);
587 	unsigned long flags;
588 
589 #ifdef CONFIG_X86
590 	{
591 		unsigned int freq = cpu_khz ? : 1;
592 
593 		SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
594 			   cpu, freq / 1000, (freq % 1000));
595 	}
596 #else
597 	SEQ_printf(m, "cpu#%d\n", cpu);
598 #endif
599 
600 #define P(x)								\
601 do {									\
602 	if (sizeof(rq->x) == 4)						\
603 		SEQ_printf(m, "  .%-30s: %ld\n", #x, (long)(rq->x));	\
604 	else								\
605 		SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rq->x));\
606 } while (0)
607 
608 #define PN(x) \
609 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
610 
611 	P(nr_running);
612 	SEQ_printf(m, "  .%-30s: %lu\n", "load",
613 		   rq->load.weight);
614 	P(nr_switches);
615 	P(nr_load_updates);
616 	P(nr_uninterruptible);
617 	PN(next_balance);
618 	SEQ_printf(m, "  .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr)));
619 	PN(clock);
620 	PN(clock_task);
621 	P(cpu_load[0]);
622 	P(cpu_load[1]);
623 	P(cpu_load[2]);
624 	P(cpu_load[3]);
625 	P(cpu_load[4]);
626 #undef P
627 #undef PN
628 
629 #ifdef CONFIG_SMP
630 #define P64(n) SEQ_printf(m, "  .%-30s: %Ld\n", #n, rq->n);
631 	P64(avg_idle);
632 	P64(max_idle_balance_cost);
633 #undef P64
634 #endif
635 
636 #define P(n) SEQ_printf(m, "  .%-30s: %d\n", #n, schedstat_val(rq->n));
637 	if (schedstat_enabled()) {
638 		P(yld_count);
639 		P(sched_count);
640 		P(sched_goidle);
641 		P(ttwu_count);
642 		P(ttwu_local);
643 	}
644 #undef P
645 
646 	spin_lock_irqsave(&sched_debug_lock, flags);
647 	print_cfs_stats(m, cpu);
648 	print_rt_stats(m, cpu);
649 	print_dl_stats(m, cpu);
650 
651 	print_rq(m, rq, cpu);
652 	spin_unlock_irqrestore(&sched_debug_lock, flags);
653 	SEQ_printf(m, "\n");
654 }
655 
656 static const char *sched_tunable_scaling_names[] = {
657 	"none",
658 	"logaritmic",
659 	"linear"
660 };
661 
662 static void sched_debug_header(struct seq_file *m)
663 {
664 	u64 ktime, sched_clk, cpu_clk;
665 	unsigned long flags;
666 
667 	local_irq_save(flags);
668 	ktime = ktime_to_ns(ktime_get());
669 	sched_clk = sched_clock();
670 	cpu_clk = local_clock();
671 	local_irq_restore(flags);
672 
673 	SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n",
674 		init_utsname()->release,
675 		(int)strcspn(init_utsname()->version, " "),
676 		init_utsname()->version);
677 
678 #define P(x) \
679 	SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
680 #define PN(x) \
681 	SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
682 	PN(ktime);
683 	PN(sched_clk);
684 	PN(cpu_clk);
685 	P(jiffies);
686 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
687 	P(sched_clock_stable());
688 #endif
689 #undef PN
690 #undef P
691 
692 	SEQ_printf(m, "\n");
693 	SEQ_printf(m, "sysctl_sched\n");
694 
695 #define P(x) \
696 	SEQ_printf(m, "  .%-40s: %Ld\n", #x, (long long)(x))
697 #define PN(x) \
698 	SEQ_printf(m, "  .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
699 	PN(sysctl_sched_latency);
700 	PN(sysctl_sched_min_granularity);
701 	PN(sysctl_sched_wakeup_granularity);
702 	P(sysctl_sched_child_runs_first);
703 	P(sysctl_sched_features);
704 #undef PN
705 #undef P
706 
707 	SEQ_printf(m, "  .%-40s: %d (%s)\n",
708 		"sysctl_sched_tunable_scaling",
709 		sysctl_sched_tunable_scaling,
710 		sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
711 	SEQ_printf(m, "\n");
712 }
713 
714 static int sched_debug_show(struct seq_file *m, void *v)
715 {
716 	int cpu = (unsigned long)(v - 2);
717 
718 	if (cpu != -1)
719 		print_cpu(m, cpu);
720 	else
721 		sched_debug_header(m);
722 
723 	return 0;
724 }
725 
726 void sysrq_sched_debug_show(void)
727 {
728 	int cpu;
729 
730 	sched_debug_header(NULL);
731 	for_each_online_cpu(cpu)
732 		print_cpu(NULL, cpu);
733 
734 }
735 
736 /*
737  * This itererator needs some explanation.
738  * It returns 1 for the header position.
739  * This means 2 is cpu 0.
740  * In a hotplugged system some cpus, including cpu 0, may be missing so we have
741  * to use cpumask_* to iterate over the cpus.
742  */
743 static void *sched_debug_start(struct seq_file *file, loff_t *offset)
744 {
745 	unsigned long n = *offset;
746 
747 	if (n == 0)
748 		return (void *) 1;
749 
750 	n--;
751 
752 	if (n > 0)
753 		n = cpumask_next(n - 1, cpu_online_mask);
754 	else
755 		n = cpumask_first(cpu_online_mask);
756 
757 	*offset = n + 1;
758 
759 	if (n < nr_cpu_ids)
760 		return (void *)(unsigned long)(n + 2);
761 	return NULL;
762 }
763 
764 static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
765 {
766 	(*offset)++;
767 	return sched_debug_start(file, offset);
768 }
769 
770 static void sched_debug_stop(struct seq_file *file, void *data)
771 {
772 }
773 
774 static const struct seq_operations sched_debug_sops = {
775 	.start = sched_debug_start,
776 	.next = sched_debug_next,
777 	.stop = sched_debug_stop,
778 	.show = sched_debug_show,
779 };
780 
781 static int sched_debug_release(struct inode *inode, struct file *file)
782 {
783 	seq_release(inode, file);
784 
785 	return 0;
786 }
787 
788 static int sched_debug_open(struct inode *inode, struct file *filp)
789 {
790 	int ret = 0;
791 
792 	ret = seq_open(filp, &sched_debug_sops);
793 
794 	return ret;
795 }
796 
797 static const struct file_operations sched_debug_fops = {
798 	.open		= sched_debug_open,
799 	.read		= seq_read,
800 	.llseek		= seq_lseek,
801 	.release	= sched_debug_release,
802 };
803 
804 static int __init init_sched_debug_procfs(void)
805 {
806 	struct proc_dir_entry *pe;
807 
808 	pe = proc_create("sched_debug", 0444, NULL, &sched_debug_fops);
809 	if (!pe)
810 		return -ENOMEM;
811 	return 0;
812 }
813 
814 __initcall(init_sched_debug_procfs);
815 
816 #define __P(F) \
817 	SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)F)
818 #define P(F) \
819 	SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)p->F)
820 #define __PN(F) \
821 	SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
822 #define PN(F) \
823 	SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
824 
825 
826 #ifdef CONFIG_NUMA_BALANCING
827 void print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
828 		unsigned long tpf, unsigned long gsf, unsigned long gpf)
829 {
830 	SEQ_printf(m, "numa_faults node=%d ", node);
831 	SEQ_printf(m, "task_private=%lu task_shared=%lu ", tsf, tpf);
832 	SEQ_printf(m, "group_private=%lu group_shared=%lu\n", gsf, gpf);
833 }
834 #endif
835 
836 
837 static void sched_show_numa(struct task_struct *p, struct seq_file *m)
838 {
839 #ifdef CONFIG_NUMA_BALANCING
840 	struct mempolicy *pol;
841 
842 	if (p->mm)
843 		P(mm->numa_scan_seq);
844 
845 	task_lock(p);
846 	pol = p->mempolicy;
847 	if (pol && !(pol->flags & MPOL_F_MORON))
848 		pol = NULL;
849 	mpol_get(pol);
850 	task_unlock(p);
851 
852 	P(numa_pages_migrated);
853 	P(numa_preferred_nid);
854 	P(total_numa_faults);
855 	SEQ_printf(m, "current_node=%d, numa_group_id=%d\n",
856 			task_node(p), task_numa_group_id(p));
857 	show_numa_stats(p, m);
858 	mpol_put(pol);
859 #endif
860 }
861 
862 void proc_sched_show_task(struct task_struct *p, struct seq_file *m)
863 {
864 	unsigned long nr_switches;
865 
866 	SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr(p),
867 						get_nr_threads(p));
868 	SEQ_printf(m,
869 		"---------------------------------------------------------"
870 		"----------\n");
871 #define __P(F) \
872 	SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)F)
873 #define P(F) \
874 	SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)p->F)
875 #define P_SCHEDSTAT(F) \
876 	SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)schedstat_val(p->F))
877 #define __PN(F) \
878 	SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
879 #define PN(F) \
880 	SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
881 #define PN_SCHEDSTAT(F) \
882 	SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)schedstat_val(p->F)))
883 
884 	PN(se.exec_start);
885 	PN(se.vruntime);
886 	PN(se.sum_exec_runtime);
887 
888 	nr_switches = p->nvcsw + p->nivcsw;
889 
890 	P(se.nr_migrations);
891 
892 	if (schedstat_enabled()) {
893 		u64 avg_atom, avg_per_cpu;
894 
895 		PN_SCHEDSTAT(se.statistics.sum_sleep_runtime);
896 		PN_SCHEDSTAT(se.statistics.wait_start);
897 		PN_SCHEDSTAT(se.statistics.sleep_start);
898 		PN_SCHEDSTAT(se.statistics.block_start);
899 		PN_SCHEDSTAT(se.statistics.sleep_max);
900 		PN_SCHEDSTAT(se.statistics.block_max);
901 		PN_SCHEDSTAT(se.statistics.exec_max);
902 		PN_SCHEDSTAT(se.statistics.slice_max);
903 		PN_SCHEDSTAT(se.statistics.wait_max);
904 		PN_SCHEDSTAT(se.statistics.wait_sum);
905 		P_SCHEDSTAT(se.statistics.wait_count);
906 		PN_SCHEDSTAT(se.statistics.iowait_sum);
907 		P_SCHEDSTAT(se.statistics.iowait_count);
908 		P_SCHEDSTAT(se.statistics.nr_migrations_cold);
909 		P_SCHEDSTAT(se.statistics.nr_failed_migrations_affine);
910 		P_SCHEDSTAT(se.statistics.nr_failed_migrations_running);
911 		P_SCHEDSTAT(se.statistics.nr_failed_migrations_hot);
912 		P_SCHEDSTAT(se.statistics.nr_forced_migrations);
913 		P_SCHEDSTAT(se.statistics.nr_wakeups);
914 		P_SCHEDSTAT(se.statistics.nr_wakeups_sync);
915 		P_SCHEDSTAT(se.statistics.nr_wakeups_migrate);
916 		P_SCHEDSTAT(se.statistics.nr_wakeups_local);
917 		P_SCHEDSTAT(se.statistics.nr_wakeups_remote);
918 		P_SCHEDSTAT(se.statistics.nr_wakeups_affine);
919 		P_SCHEDSTAT(se.statistics.nr_wakeups_affine_attempts);
920 		P_SCHEDSTAT(se.statistics.nr_wakeups_passive);
921 		P_SCHEDSTAT(se.statistics.nr_wakeups_idle);
922 
923 		avg_atom = p->se.sum_exec_runtime;
924 		if (nr_switches)
925 			avg_atom = div64_ul(avg_atom, nr_switches);
926 		else
927 			avg_atom = -1LL;
928 
929 		avg_per_cpu = p->se.sum_exec_runtime;
930 		if (p->se.nr_migrations) {
931 			avg_per_cpu = div64_u64(avg_per_cpu,
932 						p->se.nr_migrations);
933 		} else {
934 			avg_per_cpu = -1LL;
935 		}
936 
937 		__PN(avg_atom);
938 		__PN(avg_per_cpu);
939 	}
940 
941 	__P(nr_switches);
942 	SEQ_printf(m, "%-45s:%21Ld\n",
943 		   "nr_voluntary_switches", (long long)p->nvcsw);
944 	SEQ_printf(m, "%-45s:%21Ld\n",
945 		   "nr_involuntary_switches", (long long)p->nivcsw);
946 
947 	P(se.load.weight);
948 #ifdef CONFIG_SMP
949 	P(se.avg.load_sum);
950 	P(se.avg.util_sum);
951 	P(se.avg.load_avg);
952 	P(se.avg.util_avg);
953 	P(se.avg.last_update_time);
954 #endif
955 	P(policy);
956 	P(prio);
957 	if (p->policy == SCHED_DEADLINE) {
958 		P(dl.runtime);
959 		P(dl.deadline);
960 	}
961 #undef PN_SCHEDSTAT
962 #undef PN
963 #undef __PN
964 #undef P_SCHEDSTAT
965 #undef P
966 #undef __P
967 
968 	{
969 		unsigned int this_cpu = raw_smp_processor_id();
970 		u64 t0, t1;
971 
972 		t0 = cpu_clock(this_cpu);
973 		t1 = cpu_clock(this_cpu);
974 		SEQ_printf(m, "%-45s:%21Ld\n",
975 			   "clock-delta", (long long)(t1-t0));
976 	}
977 
978 	sched_show_numa(p, m);
979 }
980 
981 void proc_sched_set_task(struct task_struct *p)
982 {
983 #ifdef CONFIG_SCHEDSTATS
984 	memset(&p->se.statistics, 0, sizeof(p->se.statistics));
985 #endif
986 }
987