Linux Audio

Check our new training course

Loading...
v6.2
   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
  10/*
  11 * This allows printing both to /proc/sched_debug and
  12 * to the console
  13 */
  14#define SEQ_printf(m, x...)			\
  15 do {						\
  16	if (m)					\
  17		seq_printf(m, x);		\
  18	else					\
  19		pr_cont(x);			\
  20 } while (0)
  21
  22/*
  23 * Ease the printing of nsec fields:
  24 */
  25static long long nsec_high(unsigned long long nsec)
  26{
  27	if ((long long)nsec < 0) {
  28		nsec = -nsec;
  29		do_div(nsec, 1000000);
  30		return -nsec;
  31	}
  32	do_div(nsec, 1000000);
  33
  34	return nsec;
  35}
  36
  37static unsigned long nsec_low(unsigned long long nsec)
  38{
  39	if ((long long)nsec < 0)
  40		nsec = -nsec;
  41
  42	return do_div(nsec, 1000000);
  43}
  44
  45#define SPLIT_NS(x) nsec_high(x), nsec_low(x)
  46
  47#define SCHED_FEAT(name, enabled)	\
  48	#name ,
  49
  50static const char * const sched_feat_names[] = {
  51#include "features.h"
  52};
  53
  54#undef SCHED_FEAT
  55
  56static int sched_feat_show(struct seq_file *m, void *v)
  57{
  58	int i;
  59
  60	for (i = 0; i < __SCHED_FEAT_NR; i++) {
  61		if (!(sysctl_sched_features & (1UL << i)))
  62			seq_puts(m, "NO_");
  63		seq_printf(m, "%s ", sched_feat_names[i]);
  64	}
  65	seq_puts(m, "\n");
  66
  67	return 0;
  68}
  69
  70#ifdef CONFIG_JUMP_LABEL
  71
  72#define jump_label_key__true  STATIC_KEY_INIT_TRUE
  73#define jump_label_key__false STATIC_KEY_INIT_FALSE
  74
  75#define SCHED_FEAT(name, enabled)	\
  76	jump_label_key__##enabled ,
  77
  78struct static_key sched_feat_keys[__SCHED_FEAT_NR] = {
  79#include "features.h"
  80};
  81
  82#undef SCHED_FEAT
  83
  84static void sched_feat_disable(int i)
  85{
  86	static_key_disable_cpuslocked(&sched_feat_keys[i]);
  87}
  88
  89static void sched_feat_enable(int i)
  90{
  91	static_key_enable_cpuslocked(&sched_feat_keys[i]);
  92}
  93#else
  94static void sched_feat_disable(int i) { };
  95static void sched_feat_enable(int i) { };
  96#endif /* CONFIG_JUMP_LABEL */
  97
  98static int sched_feat_set(char *cmp)
  99{
 100	int i;
 101	int neg = 0;
 102
 103	if (strncmp(cmp, "NO_", 3) == 0) {
 104		neg = 1;
 105		cmp += 3;
 106	}
 107
 108	i = match_string(sched_feat_names, __SCHED_FEAT_NR, cmp);
 109	if (i < 0)
 110		return i;
 111
 112	if (neg) {
 113		sysctl_sched_features &= ~(1UL << i);
 114		sched_feat_disable(i);
 115	} else {
 116		sysctl_sched_features |= (1UL << i);
 117		sched_feat_enable(i);
 118	}
 119
 120	return 0;
 121}
 122
 123static ssize_t
 124sched_feat_write(struct file *filp, const char __user *ubuf,
 125		size_t cnt, loff_t *ppos)
 126{
 127	char buf[64];
 128	char *cmp;
 129	int ret;
 130	struct inode *inode;
 131
 132	if (cnt > 63)
 133		cnt = 63;
 134
 135	if (copy_from_user(&buf, ubuf, cnt))
 136		return -EFAULT;
 137
 138	buf[cnt] = 0;
 139	cmp = strstrip(buf);
 140
 141	/* Ensure the static_key remains in a consistent state */
 142	inode = file_inode(filp);
 143	cpus_read_lock();
 144	inode_lock(inode);
 145	ret = sched_feat_set(cmp);
 146	inode_unlock(inode);
 147	cpus_read_unlock();
 148	if (ret < 0)
 149		return ret;
 150
 151	*ppos += cnt;
 152
 153	return cnt;
 154}
 155
 156static int sched_feat_open(struct inode *inode, struct file *filp)
 157{
 158	return single_open(filp, sched_feat_show, NULL);
 159}
 160
 161static const struct file_operations sched_feat_fops = {
 162	.open		= sched_feat_open,
 163	.write		= sched_feat_write,
 164	.read		= seq_read,
 165	.llseek		= seq_lseek,
 166	.release	= single_release,
 167};
 168
 169#ifdef CONFIG_SMP
 170
 171static ssize_t sched_scaling_write(struct file *filp, const char __user *ubuf,
 172				   size_t cnt, loff_t *ppos)
 173{
 174	char buf[16];
 175	unsigned int scaling;
 176
 177	if (cnt > 15)
 178		cnt = 15;
 179
 180	if (copy_from_user(&buf, ubuf, cnt))
 181		return -EFAULT;
 182	buf[cnt] = '\0';
 183
 184	if (kstrtouint(buf, 10, &scaling))
 185		return -EINVAL;
 186
 187	if (scaling >= SCHED_TUNABLESCALING_END)
 188		return -EINVAL;
 189
 190	sysctl_sched_tunable_scaling = scaling;
 191	if (sched_update_scaling())
 192		return -EINVAL;
 193
 194	*ppos += cnt;
 195	return cnt;
 196}
 197
 198static int sched_scaling_show(struct seq_file *m, void *v)
 199{
 200	seq_printf(m, "%d\n", sysctl_sched_tunable_scaling);
 201	return 0;
 202}
 
 203
 204static int sched_scaling_open(struct inode *inode, struct file *filp)
 205{
 206	return single_open(filp, sched_scaling_show, NULL);
 207}
 208
 209static const struct file_operations sched_scaling_fops = {
 210	.open		= sched_scaling_open,
 211	.write		= sched_scaling_write,
 212	.read		= seq_read,
 213	.llseek		= seq_lseek,
 214	.release	= single_release,
 215};
 216
 217#endif /* SMP */
 218
 219#ifdef CONFIG_PREEMPT_DYNAMIC
 220
 221static ssize_t sched_dynamic_write(struct file *filp, const char __user *ubuf,
 222				   size_t cnt, loff_t *ppos)
 223{
 224	char buf[16];
 225	int mode;
 226
 227	if (cnt > 15)
 228		cnt = 15;
 229
 230	if (copy_from_user(&buf, ubuf, cnt))
 231		return -EFAULT;
 232
 233	buf[cnt] = 0;
 234	mode = sched_dynamic_mode(strstrip(buf));
 235	if (mode < 0)
 236		return mode;
 237
 238	sched_dynamic_update(mode);
 239
 240	*ppos += cnt;
 241
 242	return cnt;
 243}
 244
 245static int sched_dynamic_show(struct seq_file *m, void *v)
 246{
 247	static const char * preempt_modes[] = {
 248		"none", "voluntary", "full"
 249	};
 250	int i;
 251
 252	for (i = 0; i < ARRAY_SIZE(preempt_modes); i++) {
 253		if (preempt_dynamic_mode == i)
 254			seq_puts(m, "(");
 255		seq_puts(m, preempt_modes[i]);
 256		if (preempt_dynamic_mode == i)
 257			seq_puts(m, ")");
 258
 259		seq_puts(m, " ");
 260	}
 261
 262	seq_puts(m, "\n");
 263	return 0;
 264}
 265
 266static int sched_dynamic_open(struct inode *inode, struct file *filp)
 267{
 268	return single_open(filp, sched_dynamic_show, NULL);
 269}
 270
 271static const struct file_operations sched_dynamic_fops = {
 272	.open		= sched_dynamic_open,
 273	.write		= sched_dynamic_write,
 274	.read		= seq_read,
 275	.llseek		= seq_lseek,
 276	.release	= single_release,
 277};
 278
 279#endif /* CONFIG_PREEMPT_DYNAMIC */
 280
 281__read_mostly bool sched_debug_verbose;
 282
 283static const struct seq_operations sched_debug_sops;
 284
 285static int sched_debug_open(struct inode *inode, struct file *filp)
 286{
 287	return seq_open(filp, &sched_debug_sops);
 288}
 289
 290static const struct file_operations sched_debug_fops = {
 291	.open		= sched_debug_open,
 292	.read		= seq_read,
 293	.llseek		= seq_lseek,
 294	.release	= seq_release,
 295};
 296
 297static struct dentry *debugfs_sched;
 298
 299static __init int sched_init_debug(void)
 300{
 301	struct dentry __maybe_unused *numa;
 302
 303	debugfs_sched = debugfs_create_dir("sched", NULL);
 304
 305	debugfs_create_file("features", 0644, debugfs_sched, NULL, &sched_feat_fops);
 306	debugfs_create_bool("verbose", 0644, debugfs_sched, &sched_debug_verbose);
 307#ifdef CONFIG_PREEMPT_DYNAMIC
 308	debugfs_create_file("preempt", 0644, debugfs_sched, NULL, &sched_dynamic_fops);
 309#endif
 310
 311	debugfs_create_u32("latency_ns", 0644, debugfs_sched, &sysctl_sched_latency);
 312	debugfs_create_u32("min_granularity_ns", 0644, debugfs_sched, &sysctl_sched_min_granularity);
 313	debugfs_create_u32("idle_min_granularity_ns", 0644, debugfs_sched, &sysctl_sched_idle_min_granularity);
 314	debugfs_create_u32("wakeup_granularity_ns", 0644, debugfs_sched, &sysctl_sched_wakeup_granularity);
 315
 316	debugfs_create_u32("latency_warn_ms", 0644, debugfs_sched, &sysctl_resched_latency_warn_ms);
 317	debugfs_create_u32("latency_warn_once", 0644, debugfs_sched, &sysctl_resched_latency_warn_once);
 318
 319#ifdef CONFIG_SMP
 320	debugfs_create_file("tunable_scaling", 0644, debugfs_sched, NULL, &sched_scaling_fops);
 321	debugfs_create_u32("migration_cost_ns", 0644, debugfs_sched, &sysctl_sched_migration_cost);
 322	debugfs_create_u32("nr_migrate", 0644, debugfs_sched, &sysctl_sched_nr_migrate);
 323
 324	mutex_lock(&sched_domains_mutex);
 325	update_sched_domain_debugfs();
 326	mutex_unlock(&sched_domains_mutex);
 327#endif
 328
 329#ifdef CONFIG_NUMA_BALANCING
 330	numa = debugfs_create_dir("numa_balancing", debugfs_sched);
 331
 332	debugfs_create_u32("scan_delay_ms", 0644, numa, &sysctl_numa_balancing_scan_delay);
 333	debugfs_create_u32("scan_period_min_ms", 0644, numa, &sysctl_numa_balancing_scan_period_min);
 334	debugfs_create_u32("scan_period_max_ms", 0644, numa, &sysctl_numa_balancing_scan_period_max);
 335	debugfs_create_u32("scan_size_mb", 0644, numa, &sysctl_numa_balancing_scan_size);
 336	debugfs_create_u32("hot_threshold_ms", 0644, numa, &sysctl_numa_balancing_hot_threshold);
 337#endif
 338
 339	debugfs_create_file("debug", 0444, debugfs_sched, NULL, &sched_debug_fops);
 340
 341	return 0;
 342}
 343late_initcall(sched_init_debug);
 344
 345#ifdef CONFIG_SMP
 346
 347static cpumask_var_t		sd_sysctl_cpus;
 348static struct dentry		*sd_dentry;
 349
 350static int sd_flags_show(struct seq_file *m, void *v)
 351{
 352	unsigned long flags = *(unsigned int *)m->private;
 353	int idx;
 354
 355	for_each_set_bit(idx, &flags, __SD_FLAG_CNT) {
 356		seq_puts(m, sd_flag_debug[idx].name);
 357		seq_puts(m, " ");
 
 
 
 
 
 
 
 
 358	}
 359	seq_puts(m, "\n");
 360
 361	return 0;
 
 362}
 363
 364static int sd_flags_open(struct inode *inode, struct file *file)
 365{
 366	return single_open(file, sd_flags_show, inode->i_private);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 367}
 368
 369static const struct file_operations sd_flags_fops = {
 370	.open		= sd_flags_open,
 371	.read		= seq_read,
 372	.llseek		= seq_lseek,
 373	.release	= single_release,
 374};
 375
 376static void register_sd(struct sched_domain *sd, struct dentry *parent)
 377{
 378#define SDM(type, mode, member)	\
 379	debugfs_create_##type(#member, mode, parent, &sd->member)
 
 
 
 380
 381	SDM(ulong, 0644, min_interval);
 382	SDM(ulong, 0644, max_interval);
 383	SDM(u64,   0644, max_newidle_lb_cost);
 384	SDM(u32,   0644, busy_factor);
 385	SDM(u32,   0644, imbalance_pct);
 386	SDM(u32,   0644, cache_nice_tries);
 387	SDM(str,   0444, name);
 388
 389#undef SDM
 
 
 390
 391	debugfs_create_file("flags", 0444, parent, &sd->flags, &sd_flags_fops);
 392}
 393
 394void update_sched_domain_debugfs(void)
 395{
 396	int cpu, i;
 397
 398	/*
 399	 * This can unfortunately be invoked before sched_debug_init() creates
 400	 * the debug directory. Don't touch sd_sysctl_cpus until then.
 401	 */
 402	if (!debugfs_sched)
 403		return;
 404
 405	if (!cpumask_available(sd_sysctl_cpus)) {
 406		if (!alloc_cpumask_var(&sd_sysctl_cpus, GFP_KERNEL))
 407			return;
 408		cpumask_copy(sd_sysctl_cpus, cpu_possible_mask);
 409	}
 410
 411	if (!sd_dentry)
 412		sd_dentry = debugfs_create_dir("domains", debugfs_sched);
 
 
 
 413
 414	for_each_cpu(cpu, sd_sysctl_cpus) {
 415		struct sched_domain *sd;
 416		struct dentry *d_cpu;
 417		char buf[32];
 418
 419		snprintf(buf, sizeof(buf), "cpu%d", cpu);
 420		debugfs_lookup_and_remove(buf, sd_dentry);
 421		d_cpu = debugfs_create_dir(buf, sd_dentry);
 422
 423		i = 0;
 424		for_each_domain(cpu, sd) {
 425			struct dentry *d_sd;
 426
 427			snprintf(buf, sizeof(buf), "domain%d", i);
 428			d_sd = debugfs_create_dir(buf, d_cpu);
 429
 430			register_sd(sd, d_sd);
 431			i++;
 
 432		}
 
 
 433
 434		__cpumask_clear_cpu(cpu, sd_sysctl_cpus);
 435	}
 
 
 
 436}
 437
 438void dirty_sched_domain_sysctl(int cpu)
 439{
 440	if (cpumask_available(sd_sysctl_cpus))
 441		__cpumask_set_cpu(cpu, sd_sysctl_cpus);
 442}
 443
 
 
 
 
 
 
 
 444#endif /* CONFIG_SMP */
 445
 446#ifdef CONFIG_FAIR_GROUP_SCHED
 447static void print_cfs_group_stats(struct seq_file *m, int cpu, struct task_group *tg)
 448{
 449	struct sched_entity *se = tg->se[cpu];
 450
 451#define P(F)		SEQ_printf(m, "  .%-30s: %lld\n",	#F, (long long)F)
 452#define P_SCHEDSTAT(F)	SEQ_printf(m, "  .%-30s: %lld\n",	\
 453		#F, (long long)schedstat_val(stats->F))
 454#define PN(F)		SEQ_printf(m, "  .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F))
 455#define PN_SCHEDSTAT(F)	SEQ_printf(m, "  .%-30s: %lld.%06ld\n", \
 456		#F, SPLIT_NS((long long)schedstat_val(stats->F)))
 457
 458	if (!se)
 459		return;
 460
 461	PN(se->exec_start);
 462	PN(se->vruntime);
 463	PN(se->sum_exec_runtime);
 464
 465	if (schedstat_enabled()) {
 466		struct sched_statistics *stats;
 467		stats = __schedstats_from_se(se);
 468
 469		PN_SCHEDSTAT(wait_start);
 470		PN_SCHEDSTAT(sleep_start);
 471		PN_SCHEDSTAT(block_start);
 472		PN_SCHEDSTAT(sleep_max);
 473		PN_SCHEDSTAT(block_max);
 474		PN_SCHEDSTAT(exec_max);
 475		PN_SCHEDSTAT(slice_max);
 476		PN_SCHEDSTAT(wait_max);
 477		PN_SCHEDSTAT(wait_sum);
 478		P_SCHEDSTAT(wait_count);
 479	}
 480
 481	P(se->load.weight);
 482#ifdef CONFIG_SMP
 483	P(se->avg.load_avg);
 484	P(se->avg.util_avg);
 485	P(se->avg.runnable_avg);
 486#endif
 487
 488#undef PN_SCHEDSTAT
 489#undef PN
 490#undef P_SCHEDSTAT
 491#undef P
 492}
 493#endif
 494
 495#ifdef CONFIG_CGROUP_SCHED
 496static DEFINE_SPINLOCK(sched_debug_lock);
 497static char group_path[PATH_MAX];
 498
 499static void task_group_path(struct task_group *tg, char *path, int plen)
 500{
 501	if (autogroup_path(tg, path, plen))
 502		return;
 503
 504	cgroup_path(tg->css.cgroup, path, plen);
 505}
 506
 507/*
 508 * Only 1 SEQ_printf_task_group_path() caller can use the full length
 509 * group_path[] for cgroup path. Other simultaneous callers will have
 510 * to use a shorter stack buffer. A "..." suffix is appended at the end
 511 * of the stack buffer so that it will show up in case the output length
 512 * matches the given buffer size to indicate possible path name truncation.
 513 */
 514#define SEQ_printf_task_group_path(m, tg, fmt...)			\
 515{									\
 516	if (spin_trylock(&sched_debug_lock)) {				\
 517		task_group_path(tg, group_path, sizeof(group_path));	\
 518		SEQ_printf(m, fmt, group_path);				\
 519		spin_unlock(&sched_debug_lock);				\
 520	} else {							\
 521		char buf[128];						\
 522		char *bufend = buf + sizeof(buf) - 3;			\
 523		task_group_path(tg, buf, bufend - buf);			\
 524		strcpy(bufend - 1, "...");				\
 525		SEQ_printf(m, fmt, buf);				\
 526	}								\
 527}
 528#endif
 529
 530static void
 531print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
 532{
 533	if (task_current(rq, p))
 534		SEQ_printf(m, ">R");
 535	else
 536		SEQ_printf(m, " %c", task_state_to_char(p));
 537
 538	SEQ_printf(m, " %15s %5d %9Ld.%06ld %9Ld %5d ",
 539		p->comm, task_pid_nr(p),
 540		SPLIT_NS(p->se.vruntime),
 541		(long long)(p->nvcsw + p->nivcsw),
 542		p->prio);
 543
 544	SEQ_printf(m, "%9lld.%06ld %9lld.%06ld %9lld.%06ld %9lld.%06ld",
 545		SPLIT_NS(schedstat_val_or_zero(p->stats.wait_sum)),
 546		SPLIT_NS(p->se.sum_exec_runtime),
 547		SPLIT_NS(schedstat_val_or_zero(p->stats.sum_sleep_runtime)),
 548		SPLIT_NS(schedstat_val_or_zero(p->stats.sum_block_runtime)));
 549
 550#ifdef CONFIG_NUMA_BALANCING
 551	SEQ_printf(m, " %d %d", task_node(p), task_numa_group_id(p));
 552#endif
 553#ifdef CONFIG_CGROUP_SCHED
 554	SEQ_printf_task_group_path(m, task_group(p), " %s")
 555#endif
 556
 557	SEQ_printf(m, "\n");
 558}
 559
 560static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
 561{
 562	struct task_struct *g, *p;
 563
 564	SEQ_printf(m, "\n");
 565	SEQ_printf(m, "runnable tasks:\n");
 566	SEQ_printf(m, " S            task   PID         tree-key  switches  prio"
 567		   "     wait-time             sum-exec        sum-sleep\n");
 568	SEQ_printf(m, "-------------------------------------------------------"
 569		   "------------------------------------------------------\n");
 570
 571	rcu_read_lock();
 572	for_each_process_thread(g, p) {
 573		if (task_cpu(p) != rq_cpu)
 574			continue;
 575
 576		print_task(m, rq, p);
 577	}
 578	rcu_read_unlock();
 579}
 580
 581void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
 582{
 583	s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
 584		spread, rq0_min_vruntime, spread0;
 585	struct rq *rq = cpu_rq(cpu);
 586	struct sched_entity *last;
 587	unsigned long flags;
 588
 589#ifdef CONFIG_FAIR_GROUP_SCHED
 590	SEQ_printf(m, "\n");
 591	SEQ_printf_task_group_path(m, cfs_rq->tg, "cfs_rq[%d]:%s\n", cpu);
 592#else
 593	SEQ_printf(m, "\n");
 594	SEQ_printf(m, "cfs_rq[%d]:\n", cpu);
 595#endif
 596	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "exec_clock",
 597			SPLIT_NS(cfs_rq->exec_clock));
 598
 599	raw_spin_rq_lock_irqsave(rq, flags);
 600	if (rb_first_cached(&cfs_rq->tasks_timeline))
 601		MIN_vruntime = (__pick_first_entity(cfs_rq))->vruntime;
 602	last = __pick_last_entity(cfs_rq);
 603	if (last)
 604		max_vruntime = last->vruntime;
 605	min_vruntime = cfs_rq->min_vruntime;
 606	rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
 607	raw_spin_rq_unlock_irqrestore(rq, flags);
 608	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "MIN_vruntime",
 609			SPLIT_NS(MIN_vruntime));
 610	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "min_vruntime",
 611			SPLIT_NS(min_vruntime));
 612	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "max_vruntime",
 613			SPLIT_NS(max_vruntime));
 614	spread = max_vruntime - MIN_vruntime;
 615	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread",
 616			SPLIT_NS(spread));
 617	spread0 = min_vruntime - rq0_min_vruntime;
 618	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread0",
 619			SPLIT_NS(spread0));
 620	SEQ_printf(m, "  .%-30s: %d\n", "nr_spread_over",
 621			cfs_rq->nr_spread_over);
 622	SEQ_printf(m, "  .%-30s: %d\n", "nr_running", cfs_rq->nr_running);
 623	SEQ_printf(m, "  .%-30s: %d\n", "h_nr_running", cfs_rq->h_nr_running);
 624	SEQ_printf(m, "  .%-30s: %d\n", "idle_nr_running",
 625			cfs_rq->idle_nr_running);
 626	SEQ_printf(m, "  .%-30s: %d\n", "idle_h_nr_running",
 627			cfs_rq->idle_h_nr_running);
 628	SEQ_printf(m, "  .%-30s: %ld\n", "load", cfs_rq->load.weight);
 629#ifdef CONFIG_SMP
 630	SEQ_printf(m, "  .%-30s: %lu\n", "load_avg",
 631			cfs_rq->avg.load_avg);
 632	SEQ_printf(m, "  .%-30s: %lu\n", "runnable_avg",
 633			cfs_rq->avg.runnable_avg);
 634	SEQ_printf(m, "  .%-30s: %lu\n", "util_avg",
 635			cfs_rq->avg.util_avg);
 636	SEQ_printf(m, "  .%-30s: %u\n", "util_est_enqueued",
 637			cfs_rq->avg.util_est.enqueued);
 638	SEQ_printf(m, "  .%-30s: %ld\n", "removed.load_avg",
 639			cfs_rq->removed.load_avg);
 640	SEQ_printf(m, "  .%-30s: %ld\n", "removed.util_avg",
 641			cfs_rq->removed.util_avg);
 642	SEQ_printf(m, "  .%-30s: %ld\n", "removed.runnable_avg",
 643			cfs_rq->removed.runnable_avg);
 644#ifdef CONFIG_FAIR_GROUP_SCHED
 645	SEQ_printf(m, "  .%-30s: %lu\n", "tg_load_avg_contrib",
 646			cfs_rq->tg_load_avg_contrib);
 647	SEQ_printf(m, "  .%-30s: %ld\n", "tg_load_avg",
 648			atomic_long_read(&cfs_rq->tg->load_avg));
 649#endif
 650#endif
 651#ifdef CONFIG_CFS_BANDWIDTH
 652	SEQ_printf(m, "  .%-30s: %d\n", "throttled",
 653			cfs_rq->throttled);
 654	SEQ_printf(m, "  .%-30s: %d\n", "throttle_count",
 655			cfs_rq->throttle_count);
 656#endif
 657
 658#ifdef CONFIG_FAIR_GROUP_SCHED
 659	print_cfs_group_stats(m, cpu, cfs_rq->tg);
 660#endif
 661}
 662
 663void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
 664{
 665#ifdef CONFIG_RT_GROUP_SCHED
 666	SEQ_printf(m, "\n");
 667	SEQ_printf_task_group_path(m, rt_rq->tg, "rt_rq[%d]:%s\n", cpu);
 668#else
 669	SEQ_printf(m, "\n");
 670	SEQ_printf(m, "rt_rq[%d]:\n", cpu);
 671#endif
 672
 673#define P(x) \
 674	SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
 675#define PU(x) \
 676	SEQ_printf(m, "  .%-30s: %lu\n", #x, (unsigned long)(rt_rq->x))
 677#define PN(x) \
 678	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
 679
 680	PU(rt_nr_running);
 681#ifdef CONFIG_SMP
 682	PU(rt_nr_migratory);
 683#endif
 684	P(rt_throttled);
 685	PN(rt_time);
 686	PN(rt_runtime);
 687
 688#undef PN
 689#undef PU
 690#undef P
 691}
 692
 693void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq)
 694{
 695	struct dl_bw *dl_bw;
 696
 697	SEQ_printf(m, "\n");
 698	SEQ_printf(m, "dl_rq[%d]:\n", cpu);
 699
 700#define PU(x) \
 701	SEQ_printf(m, "  .%-30s: %lu\n", #x, (unsigned long)(dl_rq->x))
 702
 703	PU(dl_nr_running);
 704#ifdef CONFIG_SMP
 705	PU(dl_nr_migratory);
 706	dl_bw = &cpu_rq(cpu)->rd->dl_bw;
 707#else
 708	dl_bw = &dl_rq->dl_bw;
 709#endif
 710	SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->bw", dl_bw->bw);
 711	SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->total_bw", dl_bw->total_bw);
 712
 713#undef PU
 714}
 715
 716static void print_cpu(struct seq_file *m, int cpu)
 717{
 718	struct rq *rq = cpu_rq(cpu);
 
 719
 720#ifdef CONFIG_X86
 721	{
 722		unsigned int freq = cpu_khz ? : 1;
 723
 724		SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
 725			   cpu, freq / 1000, (freq % 1000));
 726	}
 727#else
 728	SEQ_printf(m, "cpu#%d\n", cpu);
 729#endif
 730
 731#define P(x)								\
 732do {									\
 733	if (sizeof(rq->x) == 4)						\
 734		SEQ_printf(m, "  .%-30s: %ld\n", #x, (long)(rq->x));	\
 735	else								\
 736		SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rq->x));\
 737} while (0)
 738
 739#define PN(x) \
 740	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
 741
 742	P(nr_running);
 743	P(nr_switches);
 744	P(nr_uninterruptible);
 745	PN(next_balance);
 746	SEQ_printf(m, "  .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr)));
 747	PN(clock);
 748	PN(clock_task);
 749#undef P
 750#undef PN
 751
 752#ifdef CONFIG_SMP
 753#define P64(n) SEQ_printf(m, "  .%-30s: %Ld\n", #n, rq->n);
 754	P64(avg_idle);
 755	P64(max_idle_balance_cost);
 756#undef P64
 757#endif
 758
 759#define P(n) SEQ_printf(m, "  .%-30s: %d\n", #n, schedstat_val(rq->n));
 760	if (schedstat_enabled()) {
 761		P(yld_count);
 762		P(sched_count);
 763		P(sched_goidle);
 764		P(ttwu_count);
 765		P(ttwu_local);
 766	}
 767#undef P
 768
 
 769	print_cfs_stats(m, cpu);
 770	print_rt_stats(m, cpu);
 771	print_dl_stats(m, cpu);
 772
 773	print_rq(m, rq, cpu);
 
 774	SEQ_printf(m, "\n");
 775}
 776
 777static const char *sched_tunable_scaling_names[] = {
 778	"none",
 779	"logarithmic",
 780	"linear"
 781};
 782
 783static void sched_debug_header(struct seq_file *m)
 784{
 785	u64 ktime, sched_clk, cpu_clk;
 786	unsigned long flags;
 787
 788	local_irq_save(flags);
 789	ktime = ktime_to_ns(ktime_get());
 790	sched_clk = sched_clock();
 791	cpu_clk = local_clock();
 792	local_irq_restore(flags);
 793
 794	SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n",
 795		init_utsname()->release,
 796		(int)strcspn(init_utsname()->version, " "),
 797		init_utsname()->version);
 798
 799#define P(x) \
 800	SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
 801#define PN(x) \
 802	SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
 803	PN(ktime);
 804	PN(sched_clk);
 805	PN(cpu_clk);
 806	P(jiffies);
 807#ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
 808	P(sched_clock_stable());
 809#endif
 810#undef PN
 811#undef P
 812
 813	SEQ_printf(m, "\n");
 814	SEQ_printf(m, "sysctl_sched\n");
 815
 816#define P(x) \
 817	SEQ_printf(m, "  .%-40s: %Ld\n", #x, (long long)(x))
 818#define PN(x) \
 819	SEQ_printf(m, "  .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
 820	PN(sysctl_sched_latency);
 821	PN(sysctl_sched_min_granularity);
 822	PN(sysctl_sched_idle_min_granularity);
 823	PN(sysctl_sched_wakeup_granularity);
 824	P(sysctl_sched_child_runs_first);
 825	P(sysctl_sched_features);
 826#undef PN
 827#undef P
 828
 829	SEQ_printf(m, "  .%-40s: %d (%s)\n",
 830		"sysctl_sched_tunable_scaling",
 831		sysctl_sched_tunable_scaling,
 832		sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
 833	SEQ_printf(m, "\n");
 834}
 835
 836static int sched_debug_show(struct seq_file *m, void *v)
 837{
 838	int cpu = (unsigned long)(v - 2);
 839
 840	if (cpu != -1)
 841		print_cpu(m, cpu);
 842	else
 843		sched_debug_header(m);
 844
 845	return 0;
 846}
 847
 848void sysrq_sched_debug_show(void)
 849{
 850	int cpu;
 851
 852	sched_debug_header(NULL);
 853	for_each_online_cpu(cpu) {
 854		/*
 855		 * Need to reset softlockup watchdogs on all CPUs, because
 856		 * another CPU might be blocked waiting for us to process
 857		 * an IPI or stop_machine.
 858		 */
 859		touch_nmi_watchdog();
 860		touch_all_softlockup_watchdogs();
 861		print_cpu(NULL, cpu);
 862	}
 863}
 864
 865/*
 866 * This iterator needs some explanation.
 867 * It returns 1 for the header position.
 868 * This means 2 is CPU 0.
 869 * In a hotplugged system some CPUs, including CPU 0, may be missing so we have
 870 * to use cpumask_* to iterate over the CPUs.
 871 */
 872static void *sched_debug_start(struct seq_file *file, loff_t *offset)
 873{
 874	unsigned long n = *offset;
 875
 876	if (n == 0)
 877		return (void *) 1;
 878
 879	n--;
 880
 881	if (n > 0)
 882		n = cpumask_next(n - 1, cpu_online_mask);
 883	else
 884		n = cpumask_first(cpu_online_mask);
 885
 886	*offset = n + 1;
 887
 888	if (n < nr_cpu_ids)
 889		return (void *)(unsigned long)(n + 2);
 890
 891	return NULL;
 892}
 893
 894static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
 895{
 896	(*offset)++;
 897	return sched_debug_start(file, offset);
 898}
 899
 900static void sched_debug_stop(struct seq_file *file, void *data)
 901{
 902}
 903
 904static const struct seq_operations sched_debug_sops = {
 905	.start		= sched_debug_start,
 906	.next		= sched_debug_next,
 907	.stop		= sched_debug_stop,
 908	.show		= sched_debug_show,
 909};
 910
 
 
 
 
 
 
 
 
 
 911#define __PS(S, F) SEQ_printf(m, "%-45s:%21Ld\n", S, (long long)(F))
 912#define __P(F) __PS(#F, F)
 913#define   P(F) __PS(#F, p->F)
 914#define   PM(F, M) __PS(#F, p->F & (M))
 915#define __PSN(S, F) SEQ_printf(m, "%-45s:%14Ld.%06ld\n", S, SPLIT_NS((long long)(F)))
 916#define __PN(F) __PSN(#F, F)
 917#define   PN(F) __PSN(#F, p->F)
 918
 919
 920#ifdef CONFIG_NUMA_BALANCING
 921void print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
 922		unsigned long tpf, unsigned long gsf, unsigned long gpf)
 923{
 924	SEQ_printf(m, "numa_faults node=%d ", node);
 925	SEQ_printf(m, "task_private=%lu task_shared=%lu ", tpf, tsf);
 926	SEQ_printf(m, "group_private=%lu group_shared=%lu\n", gpf, gsf);
 927}
 928#endif
 929
 930
 931static void sched_show_numa(struct task_struct *p, struct seq_file *m)
 932{
 933#ifdef CONFIG_NUMA_BALANCING
 
 
 934	if (p->mm)
 935		P(mm->numa_scan_seq);
 936
 
 
 
 
 
 
 
 937	P(numa_pages_migrated);
 938	P(numa_preferred_nid);
 939	P(total_numa_faults);
 940	SEQ_printf(m, "current_node=%d, numa_group_id=%d\n",
 941			task_node(p), task_numa_group_id(p));
 942	show_numa_stats(p, m);
 
 943#endif
 944}
 945
 946void proc_sched_show_task(struct task_struct *p, struct pid_namespace *ns,
 947						  struct seq_file *m)
 948{
 949	unsigned long nr_switches;
 950
 951	SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr_ns(p, ns),
 952						get_nr_threads(p));
 953	SEQ_printf(m,
 954		"---------------------------------------------------------"
 955		"----------\n");
 956
 957#define P_SCHEDSTAT(F)  __PS(#F, schedstat_val(p->stats.F))
 958#define PN_SCHEDSTAT(F) __PSN(#F, schedstat_val(p->stats.F))
 959
 960	PN(se.exec_start);
 961	PN(se.vruntime);
 962	PN(se.sum_exec_runtime);
 963
 964	nr_switches = p->nvcsw + p->nivcsw;
 965
 966	P(se.nr_migrations);
 967
 968	if (schedstat_enabled()) {
 969		u64 avg_atom, avg_per_cpu;
 970
 971		PN_SCHEDSTAT(sum_sleep_runtime);
 972		PN_SCHEDSTAT(sum_block_runtime);
 973		PN_SCHEDSTAT(wait_start);
 974		PN_SCHEDSTAT(sleep_start);
 975		PN_SCHEDSTAT(block_start);
 976		PN_SCHEDSTAT(sleep_max);
 977		PN_SCHEDSTAT(block_max);
 978		PN_SCHEDSTAT(exec_max);
 979		PN_SCHEDSTAT(slice_max);
 980		PN_SCHEDSTAT(wait_max);
 981		PN_SCHEDSTAT(wait_sum);
 982		P_SCHEDSTAT(wait_count);
 983		PN_SCHEDSTAT(iowait_sum);
 984		P_SCHEDSTAT(iowait_count);
 985		P_SCHEDSTAT(nr_migrations_cold);
 986		P_SCHEDSTAT(nr_failed_migrations_affine);
 987		P_SCHEDSTAT(nr_failed_migrations_running);
 988		P_SCHEDSTAT(nr_failed_migrations_hot);
 989		P_SCHEDSTAT(nr_forced_migrations);
 990		P_SCHEDSTAT(nr_wakeups);
 991		P_SCHEDSTAT(nr_wakeups_sync);
 992		P_SCHEDSTAT(nr_wakeups_migrate);
 993		P_SCHEDSTAT(nr_wakeups_local);
 994		P_SCHEDSTAT(nr_wakeups_remote);
 995		P_SCHEDSTAT(nr_wakeups_affine);
 996		P_SCHEDSTAT(nr_wakeups_affine_attempts);
 997		P_SCHEDSTAT(nr_wakeups_passive);
 998		P_SCHEDSTAT(nr_wakeups_idle);
 999
1000		avg_atom = p->se.sum_exec_runtime;
1001		if (nr_switches)
1002			avg_atom = div64_ul(avg_atom, nr_switches);
1003		else
1004			avg_atom = -1LL;
1005
1006		avg_per_cpu = p->se.sum_exec_runtime;
1007		if (p->se.nr_migrations) {
1008			avg_per_cpu = div64_u64(avg_per_cpu,
1009						p->se.nr_migrations);
1010		} else {
1011			avg_per_cpu = -1LL;
1012		}
1013
1014		__PN(avg_atom);
1015		__PN(avg_per_cpu);
1016
1017#ifdef CONFIG_SCHED_CORE
1018		PN_SCHEDSTAT(core_forceidle_sum);
1019#endif
1020	}
1021
1022	__P(nr_switches);
1023	__PS("nr_voluntary_switches", p->nvcsw);
1024	__PS("nr_involuntary_switches", p->nivcsw);
1025
1026	P(se.load.weight);
1027#ifdef CONFIG_SMP
1028	P(se.avg.load_sum);
1029	P(se.avg.runnable_sum);
1030	P(se.avg.util_sum);
1031	P(se.avg.load_avg);
1032	P(se.avg.runnable_avg);
1033	P(se.avg.util_avg);
1034	P(se.avg.last_update_time);
1035	P(se.avg.util_est.ewma);
1036	PM(se.avg.util_est.enqueued, ~UTIL_AVG_UNCHANGED);
1037#endif
1038#ifdef CONFIG_UCLAMP_TASK
1039	__PS("uclamp.min", p->uclamp_req[UCLAMP_MIN].value);
1040	__PS("uclamp.max", p->uclamp_req[UCLAMP_MAX].value);
1041	__PS("effective uclamp.min", uclamp_eff_value(p, UCLAMP_MIN));
1042	__PS("effective uclamp.max", uclamp_eff_value(p, UCLAMP_MAX));
1043#endif
1044	P(policy);
1045	P(prio);
1046	if (task_has_dl_policy(p)) {
1047		P(dl.runtime);
1048		P(dl.deadline);
1049	}
1050#undef PN_SCHEDSTAT
1051#undef P_SCHEDSTAT
1052
1053	{
1054		unsigned int this_cpu = raw_smp_processor_id();
1055		u64 t0, t1;
1056
1057		t0 = cpu_clock(this_cpu);
1058		t1 = cpu_clock(this_cpu);
1059		__PS("clock-delta", t1-t0);
1060	}
1061
1062	sched_show_numa(p, m);
1063}
1064
1065void proc_sched_set_task(struct task_struct *p)
1066{
1067#ifdef CONFIG_SCHEDSTATS
1068	memset(&p->stats, 0, sizeof(p->stats));
1069#endif
1070}
1071
1072void resched_latency_warn(int cpu, u64 latency)
1073{
1074	static DEFINE_RATELIMIT_STATE(latency_check_ratelimit, 60 * 60 * HZ, 1);
1075
1076	WARN(__ratelimit(&latency_check_ratelimit),
1077	     "sched: CPU %d need_resched set for > %llu ns (%d ticks) "
1078	     "without schedule\n",
1079	     cpu, latency, cpu_rq(cpu)->ticks_without_resched);
1080}
v5.9
  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
 11static 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 */
 28static 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
 40static 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
 53static const char * const sched_feat_names[] = {
 54#include "features.h"
 55};
 56
 57#undef SCHED_FEAT
 58
 59static 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
 81struct static_key sched_feat_keys[__SCHED_FEAT_NR] = {
 82#include "features.h"
 83};
 84
 85#undef SCHED_FEAT
 86
 87static void sched_feat_disable(int i)
 88{
 89	static_key_disable_cpuslocked(&sched_feat_keys[i]);
 90}
 91
 92static void sched_feat_enable(int i)
 93{
 94	static_key_enable_cpuslocked(&sched_feat_keys[i]);
 95}
 96#else
 97static void sched_feat_disable(int i) { };
 98static void sched_feat_enable(int i) { };
 99#endif /* CONFIG_JUMP_LABEL */
100
101static 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
126static ssize_t
127sched_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
159static int sched_feat_open(struct inode *inode, struct file *filp)
160{
161	return single_open(filp, sched_feat_show, NULL);
162}
163
164static 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
174static __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}
184late_initcall(sched_init_debug);
185
186#ifdef CONFIG_SMP
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
187
188#ifdef CONFIG_SYSCTL
 
 
 
189
190static struct ctl_table sd_ctl_dir[] = {
191	{
192		.procname	= "sched_domain",
193		.mode		= 0555,
194	},
195	{}
196};
197
198static struct ctl_table sd_ctl_root[] = {
199	{
200		.procname	= "kernel",
201		.mode		= 0555,
202		.child		= sd_ctl_dir,
203	},
204	{}
 
 
 
 
 
 
 
 
 
205};
206
207static 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
215static 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
236static void
237set_table_entry(struct ctl_table *entry,
238		const char *procname, void *data, int maxlen,
239		umode_t mode, proc_handler *proc_handler)
240{
241	entry->procname = procname;
242	entry->data = data;
243	entry->maxlen = maxlen;
244	entry->mode = mode;
245	entry->proc_handler = proc_handler;
246}
247
248static struct ctl_table *
249sd_alloc_ctl_domain_table(struct sched_domain *sd)
250{
251	struct ctl_table *table = sd_alloc_ctl_entry(9);
252
253	if (table == NULL)
254		return NULL;
255
256	set_table_entry(&table[0], "min_interval",	  &sd->min_interval,	    sizeof(long), 0644, proc_doulongvec_minmax);
257	set_table_entry(&table[1], "max_interval",	  &sd->max_interval,	    sizeof(long), 0644, proc_doulongvec_minmax);
258	set_table_entry(&table[2], "busy_factor",	  &sd->busy_factor,	    sizeof(int),  0644, proc_dointvec_minmax);
259	set_table_entry(&table[3], "imbalance_pct",	  &sd->imbalance_pct,	    sizeof(int),  0644, proc_dointvec_minmax);
260	set_table_entry(&table[4], "cache_nice_tries",	  &sd->cache_nice_tries,    sizeof(int),  0644, proc_dointvec_minmax);
261	set_table_entry(&table[5], "flags",		  &sd->flags,		    sizeof(int),  0444, proc_dointvec_minmax);
262	set_table_entry(&table[6], "max_newidle_lb_cost", &sd->max_newidle_lb_cost, sizeof(long), 0644, proc_doulongvec_minmax);
263	set_table_entry(&table[7], "name",		  sd->name,	       CORENAME_MAX_SIZE, 0444, proc_dostring);
264	/* &table[8] is terminator */
265
266	return table;
267}
268
269static struct ctl_table *sd_alloc_ctl_cpu_table(int cpu)
270{
271	struct ctl_table *entry, *table;
272	struct sched_domain *sd;
273	int domain_num = 0, i;
274	char buf[32];
275
276	for_each_domain(cpu, sd)
277		domain_num++;
278	entry = table = sd_alloc_ctl_entry(domain_num + 1);
279	if (table == NULL)
280		return NULL;
281
282	i = 0;
283	for_each_domain(cpu, sd) {
284		snprintf(buf, 32, "domain%d", i);
285		entry->procname = kstrdup(buf, GFP_KERNEL);
286		entry->mode = 0555;
287		entry->child = sd_alloc_ctl_domain_table(sd);
288		entry++;
289		i++;
290	}
291	return table;
292}
293
294static cpumask_var_t		sd_sysctl_cpus;
295static struct ctl_table_header	*sd_sysctl_header;
 
 
 
 
296
297void register_sched_domain_sysctl(void)
298{
299	static struct ctl_table *cpu_entries;
300	static struct ctl_table **cpu_idx;
301	static bool init_done = false;
302	char buf[32];
303	int i;
304
305	if (!cpu_entries) {
306		cpu_entries = sd_alloc_ctl_entry(num_possible_cpus() + 1);
307		if (!cpu_entries)
308			return;
 
 
 
309
310		WARN_ON(sd_ctl_dir[0].child);
311		sd_ctl_dir[0].child = cpu_entries;
312	}
313
314	if (!cpu_idx) {
315		struct ctl_table *e = cpu_entries;
316
317		cpu_idx = kcalloc(nr_cpu_ids, sizeof(struct ctl_table*), GFP_KERNEL);
318		if (!cpu_idx)
319			return;
320
321		/* deal with sparse possible map */
322		for_each_possible_cpu(i) {
323			cpu_idx[i] = e;
324			e++;
325		}
326	}
327
328	if (!cpumask_available(sd_sysctl_cpus)) {
329		if (!alloc_cpumask_var(&sd_sysctl_cpus, GFP_KERNEL))
330			return;
 
331	}
332
333	if (!init_done) {
334		init_done = true;
335		/* init to possible to not have holes in @cpu_entries */
336		cpumask_copy(sd_sysctl_cpus, cpu_possible_mask);
337	}
338
339	for_each_cpu(i, sd_sysctl_cpus) {
340		struct ctl_table *e = cpu_idx[i];
 
 
 
 
 
 
 
 
 
 
341
342		if (e->child)
343			sd_free_ctl_entry(&e->child);
344
345		if (!e->procname) {
346			snprintf(buf, 32, "cpu%d", i);
347			e->procname = kstrdup(buf, GFP_KERNEL);
348		}
349		e->mode = 0555;
350		e->child = sd_alloc_ctl_cpu_table(i);
351
352		__cpumask_clear_cpu(i, sd_sysctl_cpus);
353	}
354
355	WARN_ON(sd_sysctl_header);
356	sd_sysctl_header = register_sysctl_table(sd_ctl_root);
357}
358
359void dirty_sched_domain_sysctl(int cpu)
360{
361	if (cpumask_available(sd_sysctl_cpus))
362		__cpumask_set_cpu(cpu, sd_sysctl_cpus);
363}
364
365/* may be called multiple times per register */
366void unregister_sched_domain_sysctl(void)
367{
368	unregister_sysctl_table(sd_sysctl_header);
369	sd_sysctl_header = NULL;
370}
371#endif /* CONFIG_SYSCTL */
372#endif /* CONFIG_SMP */
373
374#ifdef CONFIG_FAIR_GROUP_SCHED
375static void print_cfs_group_stats(struct seq_file *m, int cpu, struct task_group *tg)
376{
377	struct sched_entity *se = tg->se[cpu];
378
379#define P(F)		SEQ_printf(m, "  .%-30s: %lld\n",	#F, (long long)F)
380#define P_SCHEDSTAT(F)	SEQ_printf(m, "  .%-30s: %lld\n",	#F, (long long)schedstat_val(F))
 
381#define PN(F)		SEQ_printf(m, "  .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F))
382#define PN_SCHEDSTAT(F)	SEQ_printf(m, "  .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)schedstat_val(F)))
 
383
384	if (!se)
385		return;
386
387	PN(se->exec_start);
388	PN(se->vruntime);
389	PN(se->sum_exec_runtime);
390
391	if (schedstat_enabled()) {
392		PN_SCHEDSTAT(se->statistics.wait_start);
393		PN_SCHEDSTAT(se->statistics.sleep_start);
394		PN_SCHEDSTAT(se->statistics.block_start);
395		PN_SCHEDSTAT(se->statistics.sleep_max);
396		PN_SCHEDSTAT(se->statistics.block_max);
397		PN_SCHEDSTAT(se->statistics.exec_max);
398		PN_SCHEDSTAT(se->statistics.slice_max);
399		PN_SCHEDSTAT(se->statistics.wait_max);
400		PN_SCHEDSTAT(se->statistics.wait_sum);
401		P_SCHEDSTAT(se->statistics.wait_count);
 
 
 
402	}
403
404	P(se->load.weight);
405#ifdef CONFIG_SMP
406	P(se->avg.load_avg);
407	P(se->avg.util_avg);
408	P(se->avg.runnable_avg);
409#endif
410
411#undef PN_SCHEDSTAT
412#undef PN
413#undef P_SCHEDSTAT
414#undef P
415}
416#endif
417
418#ifdef CONFIG_CGROUP_SCHED
 
419static char group_path[PATH_MAX];
420
421static char *task_group_path(struct task_group *tg)
422{
423	if (autogroup_path(tg, group_path, PATH_MAX))
424		return group_path;
425
426	cgroup_path(tg->css.cgroup, group_path, PATH_MAX);
 
427
428	return group_path;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
429}
430#endif
431
432static void
433print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
434{
435	if (rq->curr == p)
436		SEQ_printf(m, ">R");
437	else
438		SEQ_printf(m, " %c", task_state_to_char(p));
439
440	SEQ_printf(m, " %15s %5d %9Ld.%06ld %9Ld %5d ",
441		p->comm, task_pid_nr(p),
442		SPLIT_NS(p->se.vruntime),
443		(long long)(p->nvcsw + p->nivcsw),
444		p->prio);
445
446	SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
447		SPLIT_NS(schedstat_val_or_zero(p->se.statistics.wait_sum)),
448		SPLIT_NS(p->se.sum_exec_runtime),
449		SPLIT_NS(schedstat_val_or_zero(p->se.statistics.sum_sleep_runtime)));
 
450
451#ifdef CONFIG_NUMA_BALANCING
452	SEQ_printf(m, " %d %d", task_node(p), task_numa_group_id(p));
453#endif
454#ifdef CONFIG_CGROUP_SCHED
455	SEQ_printf(m, " %s", task_group_path(task_group(p)));
456#endif
457
458	SEQ_printf(m, "\n");
459}
460
461static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
462{
463	struct task_struct *g, *p;
464
465	SEQ_printf(m, "\n");
466	SEQ_printf(m, "runnable tasks:\n");
467	SEQ_printf(m, " S            task   PID         tree-key  switches  prio"
468		   "     wait-time             sum-exec        sum-sleep\n");
469	SEQ_printf(m, "-------------------------------------------------------"
470		   "------------------------------------------------------\n");
471
472	rcu_read_lock();
473	for_each_process_thread(g, p) {
474		if (task_cpu(p) != rq_cpu)
475			continue;
476
477		print_task(m, rq, p);
478	}
479	rcu_read_unlock();
480}
481
482void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
483{
484	s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
485		spread, rq0_min_vruntime, spread0;
486	struct rq *rq = cpu_rq(cpu);
487	struct sched_entity *last;
488	unsigned long flags;
489
490#ifdef CONFIG_FAIR_GROUP_SCHED
491	SEQ_printf(m, "\n");
492	SEQ_printf(m, "cfs_rq[%d]:%s\n", cpu, task_group_path(cfs_rq->tg));
493#else
494	SEQ_printf(m, "\n");
495	SEQ_printf(m, "cfs_rq[%d]:\n", cpu);
496#endif
497	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "exec_clock",
498			SPLIT_NS(cfs_rq->exec_clock));
499
500	raw_spin_lock_irqsave(&rq->lock, flags);
501	if (rb_first_cached(&cfs_rq->tasks_timeline))
502		MIN_vruntime = (__pick_first_entity(cfs_rq))->vruntime;
503	last = __pick_last_entity(cfs_rq);
504	if (last)
505		max_vruntime = last->vruntime;
506	min_vruntime = cfs_rq->min_vruntime;
507	rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
508	raw_spin_unlock_irqrestore(&rq->lock, flags);
509	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "MIN_vruntime",
510			SPLIT_NS(MIN_vruntime));
511	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "min_vruntime",
512			SPLIT_NS(min_vruntime));
513	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "max_vruntime",
514			SPLIT_NS(max_vruntime));
515	spread = max_vruntime - MIN_vruntime;
516	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread",
517			SPLIT_NS(spread));
518	spread0 = min_vruntime - rq0_min_vruntime;
519	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread0",
520			SPLIT_NS(spread0));
521	SEQ_printf(m, "  .%-30s: %d\n", "nr_spread_over",
522			cfs_rq->nr_spread_over);
523	SEQ_printf(m, "  .%-30s: %d\n", "nr_running", cfs_rq->nr_running);
 
 
 
 
 
524	SEQ_printf(m, "  .%-30s: %ld\n", "load", cfs_rq->load.weight);
525#ifdef CONFIG_SMP
526	SEQ_printf(m, "  .%-30s: %lu\n", "load_avg",
527			cfs_rq->avg.load_avg);
528	SEQ_printf(m, "  .%-30s: %lu\n", "runnable_avg",
529			cfs_rq->avg.runnable_avg);
530	SEQ_printf(m, "  .%-30s: %lu\n", "util_avg",
531			cfs_rq->avg.util_avg);
532	SEQ_printf(m, "  .%-30s: %u\n", "util_est_enqueued",
533			cfs_rq->avg.util_est.enqueued);
534	SEQ_printf(m, "  .%-30s: %ld\n", "removed.load_avg",
535			cfs_rq->removed.load_avg);
536	SEQ_printf(m, "  .%-30s: %ld\n", "removed.util_avg",
537			cfs_rq->removed.util_avg);
538	SEQ_printf(m, "  .%-30s: %ld\n", "removed.runnable_avg",
539			cfs_rq->removed.runnable_avg);
540#ifdef CONFIG_FAIR_GROUP_SCHED
541	SEQ_printf(m, "  .%-30s: %lu\n", "tg_load_avg_contrib",
542			cfs_rq->tg_load_avg_contrib);
543	SEQ_printf(m, "  .%-30s: %ld\n", "tg_load_avg",
544			atomic_long_read(&cfs_rq->tg->load_avg));
545#endif
546#endif
547#ifdef CONFIG_CFS_BANDWIDTH
548	SEQ_printf(m, "  .%-30s: %d\n", "throttled",
549			cfs_rq->throttled);
550	SEQ_printf(m, "  .%-30s: %d\n", "throttle_count",
551			cfs_rq->throttle_count);
552#endif
553
554#ifdef CONFIG_FAIR_GROUP_SCHED
555	print_cfs_group_stats(m, cpu, cfs_rq->tg);
556#endif
557}
558
559void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
560{
561#ifdef CONFIG_RT_GROUP_SCHED
562	SEQ_printf(m, "\n");
563	SEQ_printf(m, "rt_rq[%d]:%s\n", cpu, task_group_path(rt_rq->tg));
564#else
565	SEQ_printf(m, "\n");
566	SEQ_printf(m, "rt_rq[%d]:\n", cpu);
567#endif
568
569#define P(x) \
570	SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
571#define PU(x) \
572	SEQ_printf(m, "  .%-30s: %lu\n", #x, (unsigned long)(rt_rq->x))
573#define PN(x) \
574	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
575
576	PU(rt_nr_running);
577#ifdef CONFIG_SMP
578	PU(rt_nr_migratory);
579#endif
580	P(rt_throttled);
581	PN(rt_time);
582	PN(rt_runtime);
583
584#undef PN
585#undef PU
586#undef P
587}
588
589void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq)
590{
591	struct dl_bw *dl_bw;
592
593	SEQ_printf(m, "\n");
594	SEQ_printf(m, "dl_rq[%d]:\n", cpu);
595
596#define PU(x) \
597	SEQ_printf(m, "  .%-30s: %lu\n", #x, (unsigned long)(dl_rq->x))
598
599	PU(dl_nr_running);
600#ifdef CONFIG_SMP
601	PU(dl_nr_migratory);
602	dl_bw = &cpu_rq(cpu)->rd->dl_bw;
603#else
604	dl_bw = &dl_rq->dl_bw;
605#endif
606	SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->bw", dl_bw->bw);
607	SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->total_bw", dl_bw->total_bw);
608
609#undef PU
610}
611
612static void print_cpu(struct seq_file *m, int cpu)
613{
614	struct rq *rq = cpu_rq(cpu);
615	unsigned long flags;
616
617#ifdef CONFIG_X86
618	{
619		unsigned int freq = cpu_khz ? : 1;
620
621		SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
622			   cpu, freq / 1000, (freq % 1000));
623	}
624#else
625	SEQ_printf(m, "cpu#%d\n", cpu);
626#endif
627
628#define P(x)								\
629do {									\
630	if (sizeof(rq->x) == 4)						\
631		SEQ_printf(m, "  .%-30s: %ld\n", #x, (long)(rq->x));	\
632	else								\
633		SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rq->x));\
634} while (0)
635
636#define PN(x) \
637	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
638
639	P(nr_running);
640	P(nr_switches);
641	P(nr_uninterruptible);
642	PN(next_balance);
643	SEQ_printf(m, "  .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr)));
644	PN(clock);
645	PN(clock_task);
646#undef P
647#undef PN
648
649#ifdef CONFIG_SMP
650#define P64(n) SEQ_printf(m, "  .%-30s: %Ld\n", #n, rq->n);
651	P64(avg_idle);
652	P64(max_idle_balance_cost);
653#undef P64
654#endif
655
656#define P(n) SEQ_printf(m, "  .%-30s: %d\n", #n, schedstat_val(rq->n));
657	if (schedstat_enabled()) {
658		P(yld_count);
659		P(sched_count);
660		P(sched_goidle);
661		P(ttwu_count);
662		P(ttwu_local);
663	}
664#undef P
665
666	spin_lock_irqsave(&sched_debug_lock, flags);
667	print_cfs_stats(m, cpu);
668	print_rt_stats(m, cpu);
669	print_dl_stats(m, cpu);
670
671	print_rq(m, rq, cpu);
672	spin_unlock_irqrestore(&sched_debug_lock, flags);
673	SEQ_printf(m, "\n");
674}
675
676static const char *sched_tunable_scaling_names[] = {
677	"none",
678	"logarithmic",
679	"linear"
680};
681
682static void sched_debug_header(struct seq_file *m)
683{
684	u64 ktime, sched_clk, cpu_clk;
685	unsigned long flags;
686
687	local_irq_save(flags);
688	ktime = ktime_to_ns(ktime_get());
689	sched_clk = sched_clock();
690	cpu_clk = local_clock();
691	local_irq_restore(flags);
692
693	SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n",
694		init_utsname()->release,
695		(int)strcspn(init_utsname()->version, " "),
696		init_utsname()->version);
697
698#define P(x) \
699	SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
700#define PN(x) \
701	SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
702	PN(ktime);
703	PN(sched_clk);
704	PN(cpu_clk);
705	P(jiffies);
706#ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
707	P(sched_clock_stable());
708#endif
709#undef PN
710#undef P
711
712	SEQ_printf(m, "\n");
713	SEQ_printf(m, "sysctl_sched\n");
714
715#define P(x) \
716	SEQ_printf(m, "  .%-40s: %Ld\n", #x, (long long)(x))
717#define PN(x) \
718	SEQ_printf(m, "  .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
719	PN(sysctl_sched_latency);
720	PN(sysctl_sched_min_granularity);
 
721	PN(sysctl_sched_wakeup_granularity);
722	P(sysctl_sched_child_runs_first);
723	P(sysctl_sched_features);
724#undef PN
725#undef P
726
727	SEQ_printf(m, "  .%-40s: %d (%s)\n",
728		"sysctl_sched_tunable_scaling",
729		sysctl_sched_tunable_scaling,
730		sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
731	SEQ_printf(m, "\n");
732}
733
734static int sched_debug_show(struct seq_file *m, void *v)
735{
736	int cpu = (unsigned long)(v - 2);
737
738	if (cpu != -1)
739		print_cpu(m, cpu);
740	else
741		sched_debug_header(m);
742
743	return 0;
744}
745
746void sysrq_sched_debug_show(void)
747{
748	int cpu;
749
750	sched_debug_header(NULL);
751	for_each_online_cpu(cpu) {
752		/*
753		 * Need to reset softlockup watchdogs on all CPUs, because
754		 * another CPU might be blocked waiting for us to process
755		 * an IPI or stop_machine.
756		 */
757		touch_nmi_watchdog();
758		touch_all_softlockup_watchdogs();
759		print_cpu(NULL, cpu);
760	}
761}
762
763/*
764 * This itererator needs some explanation.
765 * It returns 1 for the header position.
766 * This means 2 is CPU 0.
767 * In a hotplugged system some CPUs, including CPU 0, may be missing so we have
768 * to use cpumask_* to iterate over the CPUs.
769 */
770static void *sched_debug_start(struct seq_file *file, loff_t *offset)
771{
772	unsigned long n = *offset;
773
774	if (n == 0)
775		return (void *) 1;
776
777	n--;
778
779	if (n > 0)
780		n = cpumask_next(n - 1, cpu_online_mask);
781	else
782		n = cpumask_first(cpu_online_mask);
783
784	*offset = n + 1;
785
786	if (n < nr_cpu_ids)
787		return (void *)(unsigned long)(n + 2);
788
789	return NULL;
790}
791
792static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
793{
794	(*offset)++;
795	return sched_debug_start(file, offset);
796}
797
798static void sched_debug_stop(struct seq_file *file, void *data)
799{
800}
801
802static const struct seq_operations sched_debug_sops = {
803	.start		= sched_debug_start,
804	.next		= sched_debug_next,
805	.stop		= sched_debug_stop,
806	.show		= sched_debug_show,
807};
808
809static int __init init_sched_debug_procfs(void)
810{
811	if (!proc_create_seq("sched_debug", 0444, NULL, &sched_debug_sops))
812		return -ENOMEM;
813	return 0;
814}
815
816__initcall(init_sched_debug_procfs);
817
818#define __PS(S, F) SEQ_printf(m, "%-45s:%21Ld\n", S, (long long)(F))
819#define __P(F) __PS(#F, F)
820#define   P(F) __PS(#F, p->F)
 
821#define __PSN(S, F) SEQ_printf(m, "%-45s:%14Ld.%06ld\n", S, SPLIT_NS((long long)(F)))
822#define __PN(F) __PSN(#F, F)
823#define   PN(F) __PSN(#F, p->F)
824
825
826#ifdef CONFIG_NUMA_BALANCING
827void 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 ", tpf, tsf);
832	SEQ_printf(m, "group_private=%lu group_shared=%lu\n", gpf, gsf);
833}
834#endif
835
836
837static 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
862void proc_sched_show_task(struct task_struct *p, struct pid_namespace *ns,
863						  struct seq_file *m)
864{
865	unsigned long nr_switches;
866
867	SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr_ns(p, ns),
868						get_nr_threads(p));
869	SEQ_printf(m,
870		"---------------------------------------------------------"
871		"----------\n");
872
873#define P_SCHEDSTAT(F)  __PS(#F, schedstat_val(p->F))
874#define PN_SCHEDSTAT(F) __PSN(#F, schedstat_val(p->F))
875
876	PN(se.exec_start);
877	PN(se.vruntime);
878	PN(se.sum_exec_runtime);
879
880	nr_switches = p->nvcsw + p->nivcsw;
881
882	P(se.nr_migrations);
883
884	if (schedstat_enabled()) {
885		u64 avg_atom, avg_per_cpu;
886
887		PN_SCHEDSTAT(se.statistics.sum_sleep_runtime);
888		PN_SCHEDSTAT(se.statistics.wait_start);
889		PN_SCHEDSTAT(se.statistics.sleep_start);
890		PN_SCHEDSTAT(se.statistics.block_start);
891		PN_SCHEDSTAT(se.statistics.sleep_max);
892		PN_SCHEDSTAT(se.statistics.block_max);
893		PN_SCHEDSTAT(se.statistics.exec_max);
894		PN_SCHEDSTAT(se.statistics.slice_max);
895		PN_SCHEDSTAT(se.statistics.wait_max);
896		PN_SCHEDSTAT(se.statistics.wait_sum);
897		P_SCHEDSTAT(se.statistics.wait_count);
898		PN_SCHEDSTAT(se.statistics.iowait_sum);
899		P_SCHEDSTAT(se.statistics.iowait_count);
900		P_SCHEDSTAT(se.statistics.nr_migrations_cold);
901		P_SCHEDSTAT(se.statistics.nr_failed_migrations_affine);
902		P_SCHEDSTAT(se.statistics.nr_failed_migrations_running);
903		P_SCHEDSTAT(se.statistics.nr_failed_migrations_hot);
904		P_SCHEDSTAT(se.statistics.nr_forced_migrations);
905		P_SCHEDSTAT(se.statistics.nr_wakeups);
906		P_SCHEDSTAT(se.statistics.nr_wakeups_sync);
907		P_SCHEDSTAT(se.statistics.nr_wakeups_migrate);
908		P_SCHEDSTAT(se.statistics.nr_wakeups_local);
909		P_SCHEDSTAT(se.statistics.nr_wakeups_remote);
910		P_SCHEDSTAT(se.statistics.nr_wakeups_affine);
911		P_SCHEDSTAT(se.statistics.nr_wakeups_affine_attempts);
912		P_SCHEDSTAT(se.statistics.nr_wakeups_passive);
913		P_SCHEDSTAT(se.statistics.nr_wakeups_idle);
 
914
915		avg_atom = p->se.sum_exec_runtime;
916		if (nr_switches)
917			avg_atom = div64_ul(avg_atom, nr_switches);
918		else
919			avg_atom = -1LL;
920
921		avg_per_cpu = p->se.sum_exec_runtime;
922		if (p->se.nr_migrations) {
923			avg_per_cpu = div64_u64(avg_per_cpu,
924						p->se.nr_migrations);
925		} else {
926			avg_per_cpu = -1LL;
927		}
928
929		__PN(avg_atom);
930		__PN(avg_per_cpu);
 
 
 
 
931	}
932
933	__P(nr_switches);
934	__PS("nr_voluntary_switches", p->nvcsw);
935	__PS("nr_involuntary_switches", p->nivcsw);
936
937	P(se.load.weight);
938#ifdef CONFIG_SMP
939	P(se.avg.load_sum);
940	P(se.avg.runnable_sum);
941	P(se.avg.util_sum);
942	P(se.avg.load_avg);
943	P(se.avg.runnable_avg);
944	P(se.avg.util_avg);
945	P(se.avg.last_update_time);
946	P(se.avg.util_est.ewma);
947	P(se.avg.util_est.enqueued);
948#endif
949#ifdef CONFIG_UCLAMP_TASK
950	__PS("uclamp.min", p->uclamp_req[UCLAMP_MIN].value);
951	__PS("uclamp.max", p->uclamp_req[UCLAMP_MAX].value);
952	__PS("effective uclamp.min", uclamp_eff_value(p, UCLAMP_MIN));
953	__PS("effective uclamp.max", uclamp_eff_value(p, UCLAMP_MAX));
954#endif
955	P(policy);
956	P(prio);
957	if (task_has_dl_policy(p)) {
958		P(dl.runtime);
959		P(dl.deadline);
960	}
961#undef PN_SCHEDSTAT
962#undef P_SCHEDSTAT
963
964	{
965		unsigned int this_cpu = raw_smp_processor_id();
966		u64 t0, t1;
967
968		t0 = cpu_clock(this_cpu);
969		t1 = cpu_clock(this_cpu);
970		__PS("clock-delta", t1-t0);
971	}
972
973	sched_show_numa(p, m);
974}
975
976void proc_sched_set_task(struct task_struct *p)
977{
978#ifdef CONFIG_SCHEDSTATS
979	memset(&p->se.statistics, 0, sizeof(p->se.statistics));
980#endif
 
 
 
 
 
 
 
 
 
 
981}