xref: /openbmc/linux/kernel/sched/debug.c (revision 22d55f02)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * kernel/sched/debug.c
4  *
5  * Print the CFS rbtree and other debugging details
6  *
7  * Copyright(C) 2007, Red Hat, Inc., Ingo Molnar
8  */
9 #include "sched.h"
10 
11 static DEFINE_SPINLOCK(sched_debug_lock);
12 
13 /*
14  * This allows printing both to /proc/sched_debug and
15  * to the console
16  */
17 #define SEQ_printf(m, x...)			\
18  do {						\
19 	if (m)					\
20 		seq_printf(m, x);		\
21 	else					\
22 		pr_cont(x);			\
23  } while (0)
24 
25 /*
26  * Ease the printing of nsec fields:
27  */
28 static long long nsec_high(unsigned long long nsec)
29 {
30 	if ((long long)nsec < 0) {
31 		nsec = -nsec;
32 		do_div(nsec, 1000000);
33 		return -nsec;
34 	}
35 	do_div(nsec, 1000000);
36 
37 	return nsec;
38 }
39 
40 static unsigned long nsec_low(unsigned long long nsec)
41 {
42 	if ((long long)nsec < 0)
43 		nsec = -nsec;
44 
45 	return do_div(nsec, 1000000);
46 }
47 
48 #define SPLIT_NS(x) nsec_high(x), nsec_low(x)
49 
50 #define SCHED_FEAT(name, enabled)	\
51 	#name ,
52 
53 static const char * const sched_feat_names[] = {
54 #include "features.h"
55 };
56 
57 #undef SCHED_FEAT
58 
59 static int sched_feat_show(struct seq_file *m, void *v)
60 {
61 	int i;
62 
63 	for (i = 0; i < __SCHED_FEAT_NR; i++) {
64 		if (!(sysctl_sched_features & (1UL << i)))
65 			seq_puts(m, "NO_");
66 		seq_printf(m, "%s ", sched_feat_names[i]);
67 	}
68 	seq_puts(m, "\n");
69 
70 	return 0;
71 }
72 
73 #ifdef CONFIG_JUMP_LABEL
74 
75 #define jump_label_key__true  STATIC_KEY_INIT_TRUE
76 #define jump_label_key__false STATIC_KEY_INIT_FALSE
77 
78 #define SCHED_FEAT(name, enabled)	\
79 	jump_label_key__##enabled ,
80 
81 struct static_key sched_feat_keys[__SCHED_FEAT_NR] = {
82 #include "features.h"
83 };
84 
85 #undef SCHED_FEAT
86 
87 static void sched_feat_disable(int i)
88 {
89 	static_key_disable_cpuslocked(&sched_feat_keys[i]);
90 }
91 
92 static void sched_feat_enable(int i)
93 {
94 	static_key_enable_cpuslocked(&sched_feat_keys[i]);
95 }
96 #else
97 static void sched_feat_disable(int i) { };
98 static void sched_feat_enable(int i) { };
99 #endif /* CONFIG_JUMP_LABEL */
100 
101 static int sched_feat_set(char *cmp)
102 {
103 	int i;
104 	int neg = 0;
105 
106 	if (strncmp(cmp, "NO_", 3) == 0) {
107 		neg = 1;
108 		cmp += 3;
109 	}
110 
111 	i = match_string(sched_feat_names, __SCHED_FEAT_NR, cmp);
112 	if (i < 0)
113 		return i;
114 
115 	if (neg) {
116 		sysctl_sched_features &= ~(1UL << i);
117 		sched_feat_disable(i);
118 	} else {
119 		sysctl_sched_features |= (1UL << i);
120 		sched_feat_enable(i);
121 	}
122 
123 	return 0;
124 }
125 
126 static ssize_t
127 sched_feat_write(struct file *filp, const char __user *ubuf,
128 		size_t cnt, loff_t *ppos)
129 {
130 	char buf[64];
131 	char *cmp;
132 	int ret;
133 	struct inode *inode;
134 
135 	if (cnt > 63)
136 		cnt = 63;
137 
138 	if (copy_from_user(&buf, ubuf, cnt))
139 		return -EFAULT;
140 
141 	buf[cnt] = 0;
142 	cmp = strstrip(buf);
143 
144 	/* Ensure the static_key remains in a consistent state */
145 	inode = file_inode(filp);
146 	cpus_read_lock();
147 	inode_lock(inode);
148 	ret = sched_feat_set(cmp);
149 	inode_unlock(inode);
150 	cpus_read_unlock();
151 	if (ret < 0)
152 		return ret;
153 
154 	*ppos += cnt;
155 
156 	return cnt;
157 }
158 
159 static int sched_feat_open(struct inode *inode, struct file *filp)
160 {
161 	return single_open(filp, sched_feat_show, NULL);
162 }
163 
164 static const struct file_operations sched_feat_fops = {
165 	.open		= sched_feat_open,
166 	.write		= sched_feat_write,
167 	.read		= seq_read,
168 	.llseek		= seq_lseek,
169 	.release	= single_release,
170 };
171 
172 __read_mostly bool sched_debug_enabled;
173 
174 static __init int sched_init_debug(void)
175 {
176 	debugfs_create_file("sched_features", 0644, NULL, NULL,
177 			&sched_feat_fops);
178 
179 	debugfs_create_bool("sched_debug", 0644, NULL,
180 			&sched_debug_enabled);
181 
182 	return 0;
183 }
184 late_initcall(sched_init_debug);
185 
186 #ifdef CONFIG_SMP
187 
188 #ifdef CONFIG_SYSCTL
189 
190 static struct ctl_table sd_ctl_dir[] = {
191 	{
192 		.procname	= "sched_domain",
193 		.mode		= 0555,
194 	},
195 	{}
196 };
197 
198 static struct ctl_table sd_ctl_root[] = {
199 	{
200 		.procname	= "kernel",
201 		.mode		= 0555,
202 		.child		= sd_ctl_dir,
203 	},
204 	{}
205 };
206 
207 static struct ctl_table *sd_alloc_ctl_entry(int n)
208 {
209 	struct ctl_table *entry =
210 		kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL);
211 
212 	return entry;
213 }
214 
215 static void sd_free_ctl_entry(struct ctl_table **tablep)
216 {
217 	struct ctl_table *entry;
218 
219 	/*
220 	 * In the intermediate directories, both the child directory and
221 	 * procname are dynamically allocated and could fail but the mode
222 	 * will always be set. In the lowest directory the names are
223 	 * static strings and all have proc handlers.
224 	 */
225 	for (entry = *tablep; entry->mode; entry++) {
226 		if (entry->child)
227 			sd_free_ctl_entry(&entry->child);
228 		if (entry->proc_handler == NULL)
229 			kfree(entry->procname);
230 	}
231 
232 	kfree(*tablep);
233 	*tablep = NULL;
234 }
235 
236 static int min_load_idx = 0;
237 static int max_load_idx = CPU_LOAD_IDX_MAX-1;
238 
239 static void
240 set_table_entry(struct ctl_table *entry,
241 		const char *procname, void *data, int maxlen,
242 		umode_t mode, proc_handler *proc_handler,
243 		bool load_idx)
244 {
245 	entry->procname = procname;
246 	entry->data = data;
247 	entry->maxlen = maxlen;
248 	entry->mode = mode;
249 	entry->proc_handler = proc_handler;
250 
251 	if (load_idx) {
252 		entry->extra1 = &min_load_idx;
253 		entry->extra2 = &max_load_idx;
254 	}
255 }
256 
257 static struct ctl_table *
258 sd_alloc_ctl_domain_table(struct sched_domain *sd)
259 {
260 	struct ctl_table *table = sd_alloc_ctl_entry(14);
261 
262 	if (table == NULL)
263 		return NULL;
264 
265 	set_table_entry(&table[0] , "min_interval",	   &sd->min_interval,	     sizeof(long), 0644, proc_doulongvec_minmax, false);
266 	set_table_entry(&table[1] , "max_interval",	   &sd->max_interval,	     sizeof(long), 0644, proc_doulongvec_minmax, false);
267 	set_table_entry(&table[2] , "busy_idx",		   &sd->busy_idx,	     sizeof(int) , 0644, proc_dointvec_minmax,   true );
268 	set_table_entry(&table[3] , "idle_idx",		   &sd->idle_idx,	     sizeof(int) , 0644, proc_dointvec_minmax,   true );
269 	set_table_entry(&table[4] , "newidle_idx",	   &sd->newidle_idx,	     sizeof(int) , 0644, proc_dointvec_minmax,   true );
270 	set_table_entry(&table[5] , "wake_idx",		   &sd->wake_idx,	     sizeof(int) , 0644, proc_dointvec_minmax,   true );
271 	set_table_entry(&table[6] , "forkexec_idx",	   &sd->forkexec_idx,	     sizeof(int) , 0644, proc_dointvec_minmax,   true );
272 	set_table_entry(&table[7] , "busy_factor",	   &sd->busy_factor,	     sizeof(int) , 0644, proc_dointvec_minmax,   false);
273 	set_table_entry(&table[8] , "imbalance_pct",	   &sd->imbalance_pct,	     sizeof(int) , 0644, proc_dointvec_minmax,   false);
274 	set_table_entry(&table[9] , "cache_nice_tries",	   &sd->cache_nice_tries,    sizeof(int) , 0644, proc_dointvec_minmax,   false);
275 	set_table_entry(&table[10], "flags",		   &sd->flags,		     sizeof(int) , 0644, proc_dointvec_minmax,   false);
276 	set_table_entry(&table[11], "max_newidle_lb_cost", &sd->max_newidle_lb_cost, sizeof(long), 0644, proc_doulongvec_minmax, false);
277 	set_table_entry(&table[12], "name",		   sd->name,		CORENAME_MAX_SIZE, 0444, proc_dostring,		 false);
278 	/* &table[13] is terminator */
279 
280 	return table;
281 }
282 
283 static struct ctl_table *sd_alloc_ctl_cpu_table(int cpu)
284 {
285 	struct ctl_table *entry, *table;
286 	struct sched_domain *sd;
287 	int domain_num = 0, i;
288 	char buf[32];
289 
290 	for_each_domain(cpu, sd)
291 		domain_num++;
292 	entry = table = sd_alloc_ctl_entry(domain_num + 1);
293 	if (table == NULL)
294 		return NULL;
295 
296 	i = 0;
297 	for_each_domain(cpu, sd) {
298 		snprintf(buf, 32, "domain%d", i);
299 		entry->procname = kstrdup(buf, GFP_KERNEL);
300 		entry->mode = 0555;
301 		entry->child = sd_alloc_ctl_domain_table(sd);
302 		entry++;
303 		i++;
304 	}
305 	return table;
306 }
307 
308 static cpumask_var_t		sd_sysctl_cpus;
309 static struct ctl_table_header	*sd_sysctl_header;
310 
311 void register_sched_domain_sysctl(void)
312 {
313 	static struct ctl_table *cpu_entries;
314 	static struct ctl_table **cpu_idx;
315 	static bool init_done = false;
316 	char buf[32];
317 	int i;
318 
319 	if (!cpu_entries) {
320 		cpu_entries = sd_alloc_ctl_entry(num_possible_cpus() + 1);
321 		if (!cpu_entries)
322 			return;
323 
324 		WARN_ON(sd_ctl_dir[0].child);
325 		sd_ctl_dir[0].child = cpu_entries;
326 	}
327 
328 	if (!cpu_idx) {
329 		struct ctl_table *e = cpu_entries;
330 
331 		cpu_idx = kcalloc(nr_cpu_ids, sizeof(struct ctl_table*), GFP_KERNEL);
332 		if (!cpu_idx)
333 			return;
334 
335 		/* deal with sparse possible map */
336 		for_each_possible_cpu(i) {
337 			cpu_idx[i] = e;
338 			e++;
339 		}
340 	}
341 
342 	if (!cpumask_available(sd_sysctl_cpus)) {
343 		if (!alloc_cpumask_var(&sd_sysctl_cpus, GFP_KERNEL))
344 			return;
345 	}
346 
347 	if (!init_done) {
348 		init_done = true;
349 		/* init to possible to not have holes in @cpu_entries */
350 		cpumask_copy(sd_sysctl_cpus, cpu_possible_mask);
351 	}
352 
353 	for_each_cpu(i, sd_sysctl_cpus) {
354 		struct ctl_table *e = cpu_idx[i];
355 
356 		if (e->child)
357 			sd_free_ctl_entry(&e->child);
358 
359 		if (!e->procname) {
360 			snprintf(buf, 32, "cpu%d", i);
361 			e->procname = kstrdup(buf, GFP_KERNEL);
362 		}
363 		e->mode = 0555;
364 		e->child = sd_alloc_ctl_cpu_table(i);
365 
366 		__cpumask_clear_cpu(i, sd_sysctl_cpus);
367 	}
368 
369 	WARN_ON(sd_sysctl_header);
370 	sd_sysctl_header = register_sysctl_table(sd_ctl_root);
371 }
372 
373 void dirty_sched_domain_sysctl(int cpu)
374 {
375 	if (cpumask_available(sd_sysctl_cpus))
376 		__cpumask_set_cpu(cpu, sd_sysctl_cpus);
377 }
378 
379 /* may be called multiple times per register */
380 void unregister_sched_domain_sysctl(void)
381 {
382 	unregister_sysctl_table(sd_sysctl_header);
383 	sd_sysctl_header = NULL;
384 }
385 #endif /* CONFIG_SYSCTL */
386 #endif /* CONFIG_SMP */
387 
388 #ifdef CONFIG_FAIR_GROUP_SCHED
389 static void print_cfs_group_stats(struct seq_file *m, int cpu, struct task_group *tg)
390 {
391 	struct sched_entity *se = tg->se[cpu];
392 
393 #define P(F)		SEQ_printf(m, "  .%-30s: %lld\n",	#F, (long long)F)
394 #define P_SCHEDSTAT(F)	SEQ_printf(m, "  .%-30s: %lld\n",	#F, (long long)schedstat_val(F))
395 #define PN(F)		SEQ_printf(m, "  .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F))
396 #define PN_SCHEDSTAT(F)	SEQ_printf(m, "  .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)schedstat_val(F)))
397 
398 	if (!se)
399 		return;
400 
401 	PN(se->exec_start);
402 	PN(se->vruntime);
403 	PN(se->sum_exec_runtime);
404 
405 	if (schedstat_enabled()) {
406 		PN_SCHEDSTAT(se->statistics.wait_start);
407 		PN_SCHEDSTAT(se->statistics.sleep_start);
408 		PN_SCHEDSTAT(se->statistics.block_start);
409 		PN_SCHEDSTAT(se->statistics.sleep_max);
410 		PN_SCHEDSTAT(se->statistics.block_max);
411 		PN_SCHEDSTAT(se->statistics.exec_max);
412 		PN_SCHEDSTAT(se->statistics.slice_max);
413 		PN_SCHEDSTAT(se->statistics.wait_max);
414 		PN_SCHEDSTAT(se->statistics.wait_sum);
415 		P_SCHEDSTAT(se->statistics.wait_count);
416 	}
417 
418 	P(se->load.weight);
419 	P(se->runnable_weight);
420 #ifdef CONFIG_SMP
421 	P(se->avg.load_avg);
422 	P(se->avg.util_avg);
423 	P(se->avg.runnable_load_avg);
424 #endif
425 
426 #undef PN_SCHEDSTAT
427 #undef PN
428 #undef P_SCHEDSTAT
429 #undef P
430 }
431 #endif
432 
433 #ifdef CONFIG_CGROUP_SCHED
434 static char group_path[PATH_MAX];
435 
436 static char *task_group_path(struct task_group *tg)
437 {
438 	if (autogroup_path(tg, group_path, PATH_MAX))
439 		return group_path;
440 
441 	cgroup_path(tg->css.cgroup, group_path, PATH_MAX);
442 
443 	return group_path;
444 }
445 #endif
446 
447 static void
448 print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
449 {
450 	if (rq->curr == p)
451 		SEQ_printf(m, ">R");
452 	else
453 		SEQ_printf(m, " %c", task_state_to_char(p));
454 
455 	SEQ_printf(m, "%15s %5d %9Ld.%06ld %9Ld %5d ",
456 		p->comm, task_pid_nr(p),
457 		SPLIT_NS(p->se.vruntime),
458 		(long long)(p->nvcsw + p->nivcsw),
459 		p->prio);
460 
461 	SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
462 		SPLIT_NS(schedstat_val_or_zero(p->se.statistics.wait_sum)),
463 		SPLIT_NS(p->se.sum_exec_runtime),
464 		SPLIT_NS(schedstat_val_or_zero(p->se.statistics.sum_sleep_runtime)));
465 
466 #ifdef CONFIG_NUMA_BALANCING
467 	SEQ_printf(m, " %d %d", task_node(p), task_numa_group_id(p));
468 #endif
469 #ifdef CONFIG_CGROUP_SCHED
470 	SEQ_printf(m, " %s", task_group_path(task_group(p)));
471 #endif
472 
473 	SEQ_printf(m, "\n");
474 }
475 
476 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
477 {
478 	struct task_struct *g, *p;
479 
480 	SEQ_printf(m, "\n");
481 	SEQ_printf(m, "runnable tasks:\n");
482 	SEQ_printf(m, " S           task   PID         tree-key  switches  prio"
483 		   "     wait-time             sum-exec        sum-sleep\n");
484 	SEQ_printf(m, "-------------------------------------------------------"
485 		   "----------------------------------------------------\n");
486 
487 	rcu_read_lock();
488 	for_each_process_thread(g, p) {
489 		if (task_cpu(p) != rq_cpu)
490 			continue;
491 
492 		print_task(m, rq, p);
493 	}
494 	rcu_read_unlock();
495 }
496 
497 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
498 {
499 	s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
500 		spread, rq0_min_vruntime, spread0;
501 	struct rq *rq = cpu_rq(cpu);
502 	struct sched_entity *last;
503 	unsigned long flags;
504 
505 #ifdef CONFIG_FAIR_GROUP_SCHED
506 	SEQ_printf(m, "\n");
507 	SEQ_printf(m, "cfs_rq[%d]:%s\n", cpu, task_group_path(cfs_rq->tg));
508 #else
509 	SEQ_printf(m, "\n");
510 	SEQ_printf(m, "cfs_rq[%d]:\n", cpu);
511 #endif
512 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "exec_clock",
513 			SPLIT_NS(cfs_rq->exec_clock));
514 
515 	raw_spin_lock_irqsave(&rq->lock, flags);
516 	if (rb_first_cached(&cfs_rq->tasks_timeline))
517 		MIN_vruntime = (__pick_first_entity(cfs_rq))->vruntime;
518 	last = __pick_last_entity(cfs_rq);
519 	if (last)
520 		max_vruntime = last->vruntime;
521 	min_vruntime = cfs_rq->min_vruntime;
522 	rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
523 	raw_spin_unlock_irqrestore(&rq->lock, flags);
524 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "MIN_vruntime",
525 			SPLIT_NS(MIN_vruntime));
526 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "min_vruntime",
527 			SPLIT_NS(min_vruntime));
528 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "max_vruntime",
529 			SPLIT_NS(max_vruntime));
530 	spread = max_vruntime - MIN_vruntime;
531 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread",
532 			SPLIT_NS(spread));
533 	spread0 = min_vruntime - rq0_min_vruntime;
534 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread0",
535 			SPLIT_NS(spread0));
536 	SEQ_printf(m, "  .%-30s: %d\n", "nr_spread_over",
537 			cfs_rq->nr_spread_over);
538 	SEQ_printf(m, "  .%-30s: %d\n", "nr_running", cfs_rq->nr_running);
539 	SEQ_printf(m, "  .%-30s: %ld\n", "load", cfs_rq->load.weight);
540 #ifdef CONFIG_SMP
541 	SEQ_printf(m, "  .%-30s: %ld\n", "runnable_weight", cfs_rq->runnable_weight);
542 	SEQ_printf(m, "  .%-30s: %lu\n", "load_avg",
543 			cfs_rq->avg.load_avg);
544 	SEQ_printf(m, "  .%-30s: %lu\n", "runnable_load_avg",
545 			cfs_rq->avg.runnable_load_avg);
546 	SEQ_printf(m, "  .%-30s: %lu\n", "util_avg",
547 			cfs_rq->avg.util_avg);
548 	SEQ_printf(m, "  .%-30s: %u\n", "util_est_enqueued",
549 			cfs_rq->avg.util_est.enqueued);
550 	SEQ_printf(m, "  .%-30s: %ld\n", "removed.load_avg",
551 			cfs_rq->removed.load_avg);
552 	SEQ_printf(m, "  .%-30s: %ld\n", "removed.util_avg",
553 			cfs_rq->removed.util_avg);
554 	SEQ_printf(m, "  .%-30s: %ld\n", "removed.runnable_sum",
555 			cfs_rq->removed.runnable_sum);
556 #ifdef CONFIG_FAIR_GROUP_SCHED
557 	SEQ_printf(m, "  .%-30s: %lu\n", "tg_load_avg_contrib",
558 			cfs_rq->tg_load_avg_contrib);
559 	SEQ_printf(m, "  .%-30s: %ld\n", "tg_load_avg",
560 			atomic_long_read(&cfs_rq->tg->load_avg));
561 #endif
562 #endif
563 #ifdef CONFIG_CFS_BANDWIDTH
564 	SEQ_printf(m, "  .%-30s: %d\n", "throttled",
565 			cfs_rq->throttled);
566 	SEQ_printf(m, "  .%-30s: %d\n", "throttle_count",
567 			cfs_rq->throttle_count);
568 #endif
569 
570 #ifdef CONFIG_FAIR_GROUP_SCHED
571 	print_cfs_group_stats(m, cpu, cfs_rq->tg);
572 #endif
573 }
574 
575 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
576 {
577 #ifdef CONFIG_RT_GROUP_SCHED
578 	SEQ_printf(m, "\n");
579 	SEQ_printf(m, "rt_rq[%d]:%s\n", cpu, task_group_path(rt_rq->tg));
580 #else
581 	SEQ_printf(m, "\n");
582 	SEQ_printf(m, "rt_rq[%d]:\n", cpu);
583 #endif
584 
585 #define P(x) \
586 	SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
587 #define PU(x) \
588 	SEQ_printf(m, "  .%-30s: %lu\n", #x, (unsigned long)(rt_rq->x))
589 #define PN(x) \
590 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
591 
592 	PU(rt_nr_running);
593 #ifdef CONFIG_SMP
594 	PU(rt_nr_migratory);
595 #endif
596 	P(rt_throttled);
597 	PN(rt_time);
598 	PN(rt_runtime);
599 
600 #undef PN
601 #undef PU
602 #undef P
603 }
604 
605 void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq)
606 {
607 	struct dl_bw *dl_bw;
608 
609 	SEQ_printf(m, "\n");
610 	SEQ_printf(m, "dl_rq[%d]:\n", cpu);
611 
612 #define PU(x) \
613 	SEQ_printf(m, "  .%-30s: %lu\n", #x, (unsigned long)(dl_rq->x))
614 
615 	PU(dl_nr_running);
616 #ifdef CONFIG_SMP
617 	PU(dl_nr_migratory);
618 	dl_bw = &cpu_rq(cpu)->rd->dl_bw;
619 #else
620 	dl_bw = &dl_rq->dl_bw;
621 #endif
622 	SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->bw", dl_bw->bw);
623 	SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->total_bw", dl_bw->total_bw);
624 
625 #undef PU
626 }
627 
628 static void print_cpu(struct seq_file *m, int cpu)
629 {
630 	struct rq *rq = cpu_rq(cpu);
631 	unsigned long flags;
632 
633 #ifdef CONFIG_X86
634 	{
635 		unsigned int freq = cpu_khz ? : 1;
636 
637 		SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
638 			   cpu, freq / 1000, (freq % 1000));
639 	}
640 #else
641 	SEQ_printf(m, "cpu#%d\n", cpu);
642 #endif
643 
644 #define P(x)								\
645 do {									\
646 	if (sizeof(rq->x) == 4)						\
647 		SEQ_printf(m, "  .%-30s: %ld\n", #x, (long)(rq->x));	\
648 	else								\
649 		SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rq->x));\
650 } while (0)
651 
652 #define PN(x) \
653 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
654 
655 	P(nr_running);
656 	SEQ_printf(m, "  .%-30s: %lu\n", "load",
657 		   rq->load.weight);
658 	P(nr_switches);
659 	P(nr_load_updates);
660 	P(nr_uninterruptible);
661 	PN(next_balance);
662 	SEQ_printf(m, "  .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr)));
663 	PN(clock);
664 	PN(clock_task);
665 	P(cpu_load[0]);
666 	P(cpu_load[1]);
667 	P(cpu_load[2]);
668 	P(cpu_load[3]);
669 	P(cpu_load[4]);
670 #undef P
671 #undef PN
672 
673 #ifdef CONFIG_SMP
674 #define P64(n) SEQ_printf(m, "  .%-30s: %Ld\n", #n, rq->n);
675 	P64(avg_idle);
676 	P64(max_idle_balance_cost);
677 #undef P64
678 #endif
679 
680 #define P(n) SEQ_printf(m, "  .%-30s: %d\n", #n, schedstat_val(rq->n));
681 	if (schedstat_enabled()) {
682 		P(yld_count);
683 		P(sched_count);
684 		P(sched_goidle);
685 		P(ttwu_count);
686 		P(ttwu_local);
687 	}
688 #undef P
689 
690 	spin_lock_irqsave(&sched_debug_lock, flags);
691 	print_cfs_stats(m, cpu);
692 	print_rt_stats(m, cpu);
693 	print_dl_stats(m, cpu);
694 
695 	print_rq(m, rq, cpu);
696 	spin_unlock_irqrestore(&sched_debug_lock, flags);
697 	SEQ_printf(m, "\n");
698 }
699 
700 static const char *sched_tunable_scaling_names[] = {
701 	"none",
702 	"logarithmic",
703 	"linear"
704 };
705 
706 static void sched_debug_header(struct seq_file *m)
707 {
708 	u64 ktime, sched_clk, cpu_clk;
709 	unsigned long flags;
710 
711 	local_irq_save(flags);
712 	ktime = ktime_to_ns(ktime_get());
713 	sched_clk = sched_clock();
714 	cpu_clk = local_clock();
715 	local_irq_restore(flags);
716 
717 	SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n",
718 		init_utsname()->release,
719 		(int)strcspn(init_utsname()->version, " "),
720 		init_utsname()->version);
721 
722 #define P(x) \
723 	SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
724 #define PN(x) \
725 	SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
726 	PN(ktime);
727 	PN(sched_clk);
728 	PN(cpu_clk);
729 	P(jiffies);
730 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
731 	P(sched_clock_stable());
732 #endif
733 #undef PN
734 #undef P
735 
736 	SEQ_printf(m, "\n");
737 	SEQ_printf(m, "sysctl_sched\n");
738 
739 #define P(x) \
740 	SEQ_printf(m, "  .%-40s: %Ld\n", #x, (long long)(x))
741 #define PN(x) \
742 	SEQ_printf(m, "  .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
743 	PN(sysctl_sched_latency);
744 	PN(sysctl_sched_min_granularity);
745 	PN(sysctl_sched_wakeup_granularity);
746 	P(sysctl_sched_child_runs_first);
747 	P(sysctl_sched_features);
748 #undef PN
749 #undef P
750 
751 	SEQ_printf(m, "  .%-40s: %d (%s)\n",
752 		"sysctl_sched_tunable_scaling",
753 		sysctl_sched_tunable_scaling,
754 		sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
755 	SEQ_printf(m, "\n");
756 }
757 
758 static int sched_debug_show(struct seq_file *m, void *v)
759 {
760 	int cpu = (unsigned long)(v - 2);
761 
762 	if (cpu != -1)
763 		print_cpu(m, cpu);
764 	else
765 		sched_debug_header(m);
766 
767 	return 0;
768 }
769 
770 void sysrq_sched_debug_show(void)
771 {
772 	int cpu;
773 
774 	sched_debug_header(NULL);
775 	for_each_online_cpu(cpu)
776 		print_cpu(NULL, cpu);
777 
778 }
779 
780 /*
781  * This itererator needs some explanation.
782  * It returns 1 for the header position.
783  * This means 2 is CPU 0.
784  * In a hotplugged system some CPUs, including CPU 0, may be missing so we have
785  * to use cpumask_* to iterate over the CPUs.
786  */
787 static void *sched_debug_start(struct seq_file *file, loff_t *offset)
788 {
789 	unsigned long n = *offset;
790 
791 	if (n == 0)
792 		return (void *) 1;
793 
794 	n--;
795 
796 	if (n > 0)
797 		n = cpumask_next(n - 1, cpu_online_mask);
798 	else
799 		n = cpumask_first(cpu_online_mask);
800 
801 	*offset = n + 1;
802 
803 	if (n < nr_cpu_ids)
804 		return (void *)(unsigned long)(n + 2);
805 
806 	return NULL;
807 }
808 
809 static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
810 {
811 	(*offset)++;
812 	return sched_debug_start(file, offset);
813 }
814 
815 static void sched_debug_stop(struct seq_file *file, void *data)
816 {
817 }
818 
819 static const struct seq_operations sched_debug_sops = {
820 	.start		= sched_debug_start,
821 	.next		= sched_debug_next,
822 	.stop		= sched_debug_stop,
823 	.show		= sched_debug_show,
824 };
825 
826 static int __init init_sched_debug_procfs(void)
827 {
828 	if (!proc_create_seq("sched_debug", 0444, NULL, &sched_debug_sops))
829 		return -ENOMEM;
830 	return 0;
831 }
832 
833 __initcall(init_sched_debug_procfs);
834 
835 #define __P(F)	SEQ_printf(m, "%-45s:%21Ld\n",	     #F, (long long)F)
836 #define   P(F)	SEQ_printf(m, "%-45s:%21Ld\n",	     #F, (long long)p->F)
837 #define __PN(F)	SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
838 #define   PN(F)	SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
839 
840 
841 #ifdef CONFIG_NUMA_BALANCING
842 void print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
843 		unsigned long tpf, unsigned long gsf, unsigned long gpf)
844 {
845 	SEQ_printf(m, "numa_faults node=%d ", node);
846 	SEQ_printf(m, "task_private=%lu task_shared=%lu ", tpf, tsf);
847 	SEQ_printf(m, "group_private=%lu group_shared=%lu\n", gpf, gsf);
848 }
849 #endif
850 
851 
852 static void sched_show_numa(struct task_struct *p, struct seq_file *m)
853 {
854 #ifdef CONFIG_NUMA_BALANCING
855 	struct mempolicy *pol;
856 
857 	if (p->mm)
858 		P(mm->numa_scan_seq);
859 
860 	task_lock(p);
861 	pol = p->mempolicy;
862 	if (pol && !(pol->flags & MPOL_F_MORON))
863 		pol = NULL;
864 	mpol_get(pol);
865 	task_unlock(p);
866 
867 	P(numa_pages_migrated);
868 	P(numa_preferred_nid);
869 	P(total_numa_faults);
870 	SEQ_printf(m, "current_node=%d, numa_group_id=%d\n",
871 			task_node(p), task_numa_group_id(p));
872 	show_numa_stats(p, m);
873 	mpol_put(pol);
874 #endif
875 }
876 
877 void proc_sched_show_task(struct task_struct *p, struct pid_namespace *ns,
878 						  struct seq_file *m)
879 {
880 	unsigned long nr_switches;
881 
882 	SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr_ns(p, ns),
883 						get_nr_threads(p));
884 	SEQ_printf(m,
885 		"---------------------------------------------------------"
886 		"----------\n");
887 #define __P(F) \
888 	SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)F)
889 #define P(F) \
890 	SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)p->F)
891 #define P_SCHEDSTAT(F) \
892 	SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)schedstat_val(p->F))
893 #define __PN(F) \
894 	SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
895 #define PN(F) \
896 	SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
897 #define PN_SCHEDSTAT(F) \
898 	SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)schedstat_val(p->F)))
899 
900 	PN(se.exec_start);
901 	PN(se.vruntime);
902 	PN(se.sum_exec_runtime);
903 
904 	nr_switches = p->nvcsw + p->nivcsw;
905 
906 	P(se.nr_migrations);
907 
908 	if (schedstat_enabled()) {
909 		u64 avg_atom, avg_per_cpu;
910 
911 		PN_SCHEDSTAT(se.statistics.sum_sleep_runtime);
912 		PN_SCHEDSTAT(se.statistics.wait_start);
913 		PN_SCHEDSTAT(se.statistics.sleep_start);
914 		PN_SCHEDSTAT(se.statistics.block_start);
915 		PN_SCHEDSTAT(se.statistics.sleep_max);
916 		PN_SCHEDSTAT(se.statistics.block_max);
917 		PN_SCHEDSTAT(se.statistics.exec_max);
918 		PN_SCHEDSTAT(se.statistics.slice_max);
919 		PN_SCHEDSTAT(se.statistics.wait_max);
920 		PN_SCHEDSTAT(se.statistics.wait_sum);
921 		P_SCHEDSTAT(se.statistics.wait_count);
922 		PN_SCHEDSTAT(se.statistics.iowait_sum);
923 		P_SCHEDSTAT(se.statistics.iowait_count);
924 		P_SCHEDSTAT(se.statistics.nr_migrations_cold);
925 		P_SCHEDSTAT(se.statistics.nr_failed_migrations_affine);
926 		P_SCHEDSTAT(se.statistics.nr_failed_migrations_running);
927 		P_SCHEDSTAT(se.statistics.nr_failed_migrations_hot);
928 		P_SCHEDSTAT(se.statistics.nr_forced_migrations);
929 		P_SCHEDSTAT(se.statistics.nr_wakeups);
930 		P_SCHEDSTAT(se.statistics.nr_wakeups_sync);
931 		P_SCHEDSTAT(se.statistics.nr_wakeups_migrate);
932 		P_SCHEDSTAT(se.statistics.nr_wakeups_local);
933 		P_SCHEDSTAT(se.statistics.nr_wakeups_remote);
934 		P_SCHEDSTAT(se.statistics.nr_wakeups_affine);
935 		P_SCHEDSTAT(se.statistics.nr_wakeups_affine_attempts);
936 		P_SCHEDSTAT(se.statistics.nr_wakeups_passive);
937 		P_SCHEDSTAT(se.statistics.nr_wakeups_idle);
938 
939 		avg_atom = p->se.sum_exec_runtime;
940 		if (nr_switches)
941 			avg_atom = div64_ul(avg_atom, nr_switches);
942 		else
943 			avg_atom = -1LL;
944 
945 		avg_per_cpu = p->se.sum_exec_runtime;
946 		if (p->se.nr_migrations) {
947 			avg_per_cpu = div64_u64(avg_per_cpu,
948 						p->se.nr_migrations);
949 		} else {
950 			avg_per_cpu = -1LL;
951 		}
952 
953 		__PN(avg_atom);
954 		__PN(avg_per_cpu);
955 	}
956 
957 	__P(nr_switches);
958 	SEQ_printf(m, "%-45s:%21Ld\n",
959 		   "nr_voluntary_switches", (long long)p->nvcsw);
960 	SEQ_printf(m, "%-45s:%21Ld\n",
961 		   "nr_involuntary_switches", (long long)p->nivcsw);
962 
963 	P(se.load.weight);
964 	P(se.runnable_weight);
965 #ifdef CONFIG_SMP
966 	P(se.avg.load_sum);
967 	P(se.avg.runnable_load_sum);
968 	P(se.avg.util_sum);
969 	P(se.avg.load_avg);
970 	P(se.avg.runnable_load_avg);
971 	P(se.avg.util_avg);
972 	P(se.avg.last_update_time);
973 	P(se.avg.util_est.ewma);
974 	P(se.avg.util_est.enqueued);
975 #endif
976 	P(policy);
977 	P(prio);
978 	if (task_has_dl_policy(p)) {
979 		P(dl.runtime);
980 		P(dl.deadline);
981 	}
982 #undef PN_SCHEDSTAT
983 #undef PN
984 #undef __PN
985 #undef P_SCHEDSTAT
986 #undef P
987 #undef __P
988 
989 	{
990 		unsigned int this_cpu = raw_smp_processor_id();
991 		u64 t0, t1;
992 
993 		t0 = cpu_clock(this_cpu);
994 		t1 = cpu_clock(this_cpu);
995 		SEQ_printf(m, "%-45s:%21Ld\n",
996 			   "clock-delta", (long long)(t1-t0));
997 	}
998 
999 	sched_show_numa(p, m);
1000 }
1001 
1002 void proc_sched_set_task(struct task_struct *p)
1003 {
1004 #ifdef CONFIG_SCHEDSTATS
1005 	memset(&p->se.statistics, 0, sizeof(p->se.statistics));
1006 #endif
1007 }
1008