debug.c

   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 <linux/debugfs.h>
  10#include <linux/nmi.h>
  11#include "sched.h"
  12
  13/*
  14 * This allows printing both to /sys/kernel/debug/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 /* !CONFIG_JUMP_LABEL: */
  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
 172static ssize_t sched_scaling_write(struct file *filp, const char __user *ubuf,
 173				   size_t cnt, loff_t *ppos)
 174{
 175	char buf[16];
 176	unsigned int scaling;
 177
 178	if (cnt > 15)
 179		cnt = 15;
 180
 181	if (copy_from_user(&buf, ubuf, cnt))
 182		return -EFAULT;
 183	buf[cnt] = '\0';
 184
 185	if (kstrtouint(buf, 10, &scaling))
 186		return -EINVAL;
 187
 188	if (scaling >= SCHED_TUNABLESCALING_END)
 189		return -EINVAL;
 190
 191	sysctl_sched_tunable_scaling = scaling;
 192	if (sched_update_scaling())
 193		return -EINVAL;
 194
 195	*ppos += cnt;
 196	return cnt;
 197}
 198
 199static int sched_scaling_show(struct seq_file *m, void *v)
 200{
 201	seq_printf(m, "%d\n", sysctl_sched_tunable_scaling);
 202	return 0;
 203}
 204
 205static int sched_scaling_open(struct inode *inode, struct file *filp)
 206{
 207	return single_open(filp, sched_scaling_show, NULL);
 208}
 209
 210static const struct file_operations sched_scaling_fops = {
 211	.open		= sched_scaling_open,
 212	.write		= sched_scaling_write,
 213	.read		= seq_read,
 214	.llseek		= seq_lseek,
 215	.release	= single_release,
 216};
 217
 218#ifdef CONFIG_PREEMPT_DYNAMIC
 219
 220static ssize_t sched_dynamic_write(struct file *filp, const char __user *ubuf,
 221				   size_t cnt, loff_t *ppos)
 222{
 223	char buf[16];
 224	int mode;
 225
 226	if (cnt > 15)
 227		cnt = 15;
 228
 229	if (copy_from_user(&buf, ubuf, cnt))
 230		return -EFAULT;
 231
 232	buf[cnt] = 0;
 233	mode = sched_dynamic_mode(strstrip(buf));
 234	if (mode < 0)
 235		return mode;
 236
 237	sched_dynamic_update(mode);
 238
 239	*ppos += cnt;
 240
 241	return cnt;
 242}
 243
 244static int sched_dynamic_show(struct seq_file *m, void *v)
 245{
 246	int i = IS_ENABLED(CONFIG_PREEMPT_RT) * 2;
 247	int j;
 248
 249	/* Count entries in NULL terminated preempt_modes */
 250	for (j = 0; preempt_modes[j]; j++)
 251		;
 252	j -= !IS_ENABLED(CONFIG_ARCH_HAS_PREEMPT_LAZY);
 253
 254	for (; i < j; i++) {
 255		if (preempt_dynamic_mode == i)
 256			seq_puts(m, "(");
 257		seq_puts(m, preempt_modes[i]);
 258		if (preempt_dynamic_mode == i)
 259			seq_puts(m, ")");
 260
 261		seq_puts(m, " ");
 262	}
 263
 264	seq_puts(m, "\n");
 265	return 0;
 266}
 267
 268static int sched_dynamic_open(struct inode *inode, struct file *filp)
 269{
 270	return single_open(filp, sched_dynamic_show, NULL);
 271}
 272
 273static const struct file_operations sched_dynamic_fops = {
 274	.open		= sched_dynamic_open,
 275	.write		= sched_dynamic_write,
 276	.read		= seq_read,
 277	.llseek		= seq_lseek,
 278	.release	= single_release,
 279};
 280
 281#endif /* CONFIG_PREEMPT_DYNAMIC */
 282
 283__read_mostly bool sched_debug_verbose;
 284
 285static struct dentry           *sd_dentry;
 286
 287
 288static ssize_t sched_verbose_write(struct file *filp, const char __user *ubuf,
 289				  size_t cnt, loff_t *ppos)
 290{
 291	ssize_t result;
 292	bool orig;
 293
 294	cpus_read_lock();
 295	sched_domains_mutex_lock();
 296
 297	orig = sched_debug_verbose;
 298	result = debugfs_write_file_bool(filp, ubuf, cnt, ppos);
 299
 300	if (sched_debug_verbose && !orig)
 301		update_sched_domain_debugfs();
 302	else if (!sched_debug_verbose && orig) {
 303		debugfs_remove(sd_dentry);
 304		sd_dentry = NULL;
 305	}
 306
 307	sched_domains_mutex_unlock();
 308	cpus_read_unlock();
 309
 310	return result;
 311}
 312
 313static const struct file_operations sched_verbose_fops = {
 314	.read =         debugfs_read_file_bool,
 315	.write =        sched_verbose_write,
 316	.open =         simple_open,
 317	.llseek =       default_llseek,
 318};
 319
 320static const struct seq_operations sched_debug_sops;
 321
 322static int sched_debug_open(struct inode *inode, struct file *filp)
 323{
 324	return seq_open(filp, &sched_debug_sops);
 325}
 326
 327static const struct file_operations sched_debug_fops = {
 328	.open		= sched_debug_open,
 329	.read		= seq_read,
 330	.llseek		= seq_lseek,
 331	.release	= seq_release,
 332};
 333
 334enum dl_param {
 335	DL_RUNTIME = 0,
 336	DL_PERIOD,
 337};
 338
 339static unsigned long fair_server_period_max = (1UL << 22) * NSEC_PER_USEC; /* ~4 seconds */
 340static unsigned long fair_server_period_min = (100) * NSEC_PER_USEC;     /* 100 us */
 341
 342static ssize_t sched_fair_server_write(struct file *filp, const char __user *ubuf,
 343				       size_t cnt, loff_t *ppos, enum dl_param param)
 344{
 345	long cpu = (long) ((struct seq_file *) filp->private_data)->private;
 346	struct rq *rq = cpu_rq(cpu);
 347	u64 runtime, period;
 348	int retval = 0;
 349	size_t err;
 350	u64 value;
 351
 352	err = kstrtoull_from_user(ubuf, cnt, 10, &value);
 353	if (err)
 354		return err;
 355
 356	scoped_guard (rq_lock_irqsave, rq) {
 357		runtime  = rq->fair_server.dl_runtime;
 358		period = rq->fair_server.dl_period;
 359
 360		switch (param) {
 361		case DL_RUNTIME:
 362			if (runtime == value)
 363				break;
 364			runtime = value;
 365			break;
 366		case DL_PERIOD:
 367			if (value == period)
 368				break;
 369			period = value;
 370			break;
 371		}
 372
 373		if (runtime > period ||
 374		    period > fair_server_period_max ||
 375		    period < fair_server_period_min) {
 376			return  -EINVAL;
 377		}
 378
 379		update_rq_clock(rq);
 380		dl_server_stop(&rq->fair_server);
 381
 382		retval = dl_server_apply_params(&rq->fair_server, runtime, period, 0);
 383
 384		if (!runtime)
 385			printk_deferred("Fair server disabled in CPU %d, system may crash due to starvation.\n",
 386					cpu_of(rq));
 387
 388		if (rq->cfs.h_nr_queued)
 389			dl_server_start(&rq->fair_server);
 390
 391		if (retval < 0)
 392			return retval;
 393	}
 394
 395	*ppos += cnt;
 396	return cnt;
 397}
 398
 399static size_t sched_fair_server_show(struct seq_file *m, void *v, enum dl_param param)
 400{
 401	unsigned long cpu = (unsigned long) m->private;
 402	struct rq *rq = cpu_rq(cpu);
 403	u64 value;
 404
 405	switch (param) {
 406	case DL_RUNTIME:
 407		value = rq->fair_server.dl_runtime;
 408		break;
 409	case DL_PERIOD:
 410		value = rq->fair_server.dl_period;
 411		break;
 412	}
 413
 414	seq_printf(m, "%llu\n", value);
 415	return 0;
 416
 417}
 418
 419static ssize_t
 420sched_fair_server_runtime_write(struct file *filp, const char __user *ubuf,
 421				size_t cnt, loff_t *ppos)
 422{
 423	return sched_fair_server_write(filp, ubuf, cnt, ppos, DL_RUNTIME);
 424}
 425
 426static int sched_fair_server_runtime_show(struct seq_file *m, void *v)
 427{
 428	return sched_fair_server_show(m, v, DL_RUNTIME);
 429}
 430
 431static int sched_fair_server_runtime_open(struct inode *inode, struct file *filp)
 432{
 433	return single_open(filp, sched_fair_server_runtime_show, inode->i_private);
 434}
 435
 436static const struct file_operations fair_server_runtime_fops = {
 437	.open		= sched_fair_server_runtime_open,
 438	.write		= sched_fair_server_runtime_write,
 439	.read		= seq_read,
 440	.llseek		= seq_lseek,
 441	.release	= single_release,
 442};
 443
 444static ssize_t
 445sched_fair_server_period_write(struct file *filp, const char __user *ubuf,
 446			       size_t cnt, loff_t *ppos)
 447{
 448	return sched_fair_server_write(filp, ubuf, cnt, ppos, DL_PERIOD);
 449}
 450
 451static int sched_fair_server_period_show(struct seq_file *m, void *v)
 452{
 453	return sched_fair_server_show(m, v, DL_PERIOD);
 454}
 455
 456static int sched_fair_server_period_open(struct inode *inode, struct file *filp)
 457{
 458	return single_open(filp, sched_fair_server_period_show, inode->i_private);
 459}
 460
 461static const struct file_operations fair_server_period_fops = {
 462	.open		= sched_fair_server_period_open,
 463	.write		= sched_fair_server_period_write,
 464	.read		= seq_read,
 465	.llseek		= seq_lseek,
 466	.release	= single_release,
 467};
 468
 469static struct dentry *debugfs_sched;
 470
 471static void debugfs_fair_server_init(void)
 472{
 473	struct dentry *d_fair;
 474	unsigned long cpu;
 475
 476	d_fair = debugfs_create_dir("fair_server", debugfs_sched);
 477	if (!d_fair)
 478		return;
 479
 480	for_each_possible_cpu(cpu) {
 481		struct dentry *d_cpu;
 482		char buf[32];
 483
 484		snprintf(buf, sizeof(buf), "cpu%lu", cpu);
 485		d_cpu = debugfs_create_dir(buf, d_fair);
 486
 487		debugfs_create_file("runtime", 0644, d_cpu, (void *) cpu, &fair_server_runtime_fops);
 488		debugfs_create_file("period", 0644, d_cpu, (void *) cpu, &fair_server_period_fops);
 489	}
 490}
 491
 492static __init int sched_init_debug(void)
 493{
 494	struct dentry __maybe_unused *numa;
 495
 496	debugfs_sched = debugfs_create_dir("sched", NULL);
 497
 498	debugfs_create_file("features", 0644, debugfs_sched, NULL, &sched_feat_fops);
 499	debugfs_create_file_unsafe("verbose", 0644, debugfs_sched, &sched_debug_verbose, &sched_verbose_fops);
 500#ifdef CONFIG_PREEMPT_DYNAMIC
 501	debugfs_create_file("preempt", 0644, debugfs_sched, NULL, &sched_dynamic_fops);
 502#endif
 503
 504	debugfs_create_u32("base_slice_ns", 0644, debugfs_sched, &sysctl_sched_base_slice);
 505
 506	debugfs_create_u32("latency_warn_ms", 0644, debugfs_sched, &sysctl_resched_latency_warn_ms);
 507	debugfs_create_u32("latency_warn_once", 0644, debugfs_sched, &sysctl_resched_latency_warn_once);
 508
 509	debugfs_create_file("tunable_scaling", 0644, debugfs_sched, NULL, &sched_scaling_fops);
 510	debugfs_create_u32("migration_cost_ns", 0644, debugfs_sched, &sysctl_sched_migration_cost);
 511	debugfs_create_u32("nr_migrate", 0644, debugfs_sched, &sysctl_sched_nr_migrate);
 512
 513	sched_domains_mutex_lock();
 514	update_sched_domain_debugfs();
 515	sched_domains_mutex_unlock();
 516
 517#ifdef CONFIG_NUMA_BALANCING
 518	numa = debugfs_create_dir("numa_balancing", debugfs_sched);
 519
 520	debugfs_create_u32("scan_delay_ms", 0644, numa, &sysctl_numa_balancing_scan_delay);
 521	debugfs_create_u32("scan_period_min_ms", 0644, numa, &sysctl_numa_balancing_scan_period_min);
 522	debugfs_create_u32("scan_period_max_ms", 0644, numa, &sysctl_numa_balancing_scan_period_max);
 523	debugfs_create_u32("scan_size_mb", 0644, numa, &sysctl_numa_balancing_scan_size);
 524	debugfs_create_u32("hot_threshold_ms", 0644, numa, &sysctl_numa_balancing_hot_threshold);
 525#endif /* CONFIG_NUMA_BALANCING */
 526
 527	debugfs_create_file("debug", 0444, debugfs_sched, NULL, &sched_debug_fops);
 528
 529	debugfs_fair_server_init();
 530
 531	return 0;
 532}
 533late_initcall(sched_init_debug);
 534
 535static cpumask_var_t		sd_sysctl_cpus;
 536
 537static int sd_flags_show(struct seq_file *m, void *v)
 538{
 539	unsigned long flags = *(unsigned int *)m->private;
 540	int idx;
 541
 542	for_each_set_bit(idx, &flags, __SD_FLAG_CNT) {
 543		seq_puts(m, sd_flag_debug[idx].name);
 544		seq_puts(m, " ");
 545	}
 546	seq_puts(m, "\n");
 547
 548	return 0;
 549}
 550
 551static int sd_flags_open(struct inode *inode, struct file *file)
 552{
 553	return single_open(file, sd_flags_show, inode->i_private);
 554}
 555
 556static const struct file_operations sd_flags_fops = {
 557	.open		= sd_flags_open,
 558	.read		= seq_read,
 559	.llseek		= seq_lseek,
 560	.release	= single_release,
 561};
 562
 563static void register_sd(struct sched_domain *sd, struct dentry *parent)
 564{
 565#define SDM(type, mode, member)	\
 566	debugfs_create_##type(#member, mode, parent, &sd->member)
 567
 568	SDM(ulong, 0644, min_interval);
 569	SDM(ulong, 0644, max_interval);
 570	SDM(u64,   0644, max_newidle_lb_cost);
 571	SDM(u32,   0644, busy_factor);
 572	SDM(u32,   0644, imbalance_pct);
 573	SDM(u32,   0644, cache_nice_tries);
 574	SDM(str,   0444, name);
 575
 576#undef SDM
 577
 578	debugfs_create_file("flags", 0444, parent, &sd->flags, &sd_flags_fops);
 579	debugfs_create_file("groups_flags", 0444, parent, &sd->groups->flags, &sd_flags_fops);
 580	debugfs_create_u32("level", 0444, parent, (u32 *)&sd->level);
 581
 582	if (sd->flags & SD_ASYM_PACKING)
 583		debugfs_create_u32("group_asym_prefer_cpu", 0444, parent,
 584				   (u32 *)&sd->groups->asym_prefer_cpu);
 585}
 586
 587void update_sched_domain_debugfs(void)
 588{
 589	int cpu, i;
 590
 591	/*
 592	 * This can unfortunately be invoked before sched_debug_init() creates
 593	 * the debug directory. Don't touch sd_sysctl_cpus until then.
 594	 */
 595	if (!debugfs_sched)
 596		return;
 597
 598	if (!sched_debug_verbose)
 599		return;
 600
 601	if (!cpumask_available(sd_sysctl_cpus)) {
 602		if (!alloc_cpumask_var(&sd_sysctl_cpus, GFP_KERNEL))
 603			return;
 604		cpumask_copy(sd_sysctl_cpus, cpu_possible_mask);
 605	}
 606
 607	if (!sd_dentry) {
 608		sd_dentry = debugfs_create_dir("domains", debugfs_sched);
 609
 610		/* rebuild sd_sysctl_cpus if empty since it gets cleared below */
 611		if (cpumask_empty(sd_sysctl_cpus))
 612			cpumask_copy(sd_sysctl_cpus, cpu_online_mask);
 613	}
 614
 615	for_each_cpu(cpu, sd_sysctl_cpus) {
 616		struct sched_domain *sd;
 617		struct dentry *d_cpu;
 618		char buf[32];
 619
 620		snprintf(buf, sizeof(buf), "cpu%d", cpu);
 621		debugfs_lookup_and_remove(buf, sd_dentry);
 622		d_cpu = debugfs_create_dir(buf, sd_dentry);
 623
 624		i = 0;
 625		for_each_domain(cpu, sd) {
 626			struct dentry *d_sd;
 627
 628			snprintf(buf, sizeof(buf), "domain%d", i);
 629			d_sd = debugfs_create_dir(buf, d_cpu);
 630
 631			register_sd(sd, d_sd);
 632			i++;
 633		}
 634
 635		__cpumask_clear_cpu(cpu, sd_sysctl_cpus);
 636	}
 637}
 638
 639void dirty_sched_domain_sysctl(int cpu)
 640{
 641	if (cpumask_available(sd_sysctl_cpus))
 642		__cpumask_set_cpu(cpu, sd_sysctl_cpus);
 643}
 644
 645#ifdef CONFIG_FAIR_GROUP_SCHED
 646static void print_cfs_group_stats(struct seq_file *m, int cpu, struct task_group *tg)
 647{
 648	struct sched_entity *se = tg->se[cpu];
 649
 650#define P(F)		SEQ_printf(m, "  .%-30s: %lld\n",	#F, (long long)F)
 651#define P_SCHEDSTAT(F)	SEQ_printf(m, "  .%-30s: %lld\n",	\
 652		#F, (long long)schedstat_val(stats->F))
 653#define PN(F)		SEQ_printf(m, "  .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F))
 654#define PN_SCHEDSTAT(F)	SEQ_printf(m, "  .%-30s: %lld.%06ld\n", \
 655		#F, SPLIT_NS((long long)schedstat_val(stats->F)))
 656
 657	if (!se)
 658		return;
 659
 660	PN(se->exec_start);
 661	PN(se->vruntime);
 662	PN(se->sum_exec_runtime);
 663
 664	if (schedstat_enabled()) {
 665		struct sched_statistics *stats;
 666		stats = __schedstats_from_se(se);
 667
 668		PN_SCHEDSTAT(wait_start);
 669		PN_SCHEDSTAT(sleep_start);
 670		PN_SCHEDSTAT(block_start);
 671		PN_SCHEDSTAT(sleep_max);
 672		PN_SCHEDSTAT(block_max);
 673		PN_SCHEDSTAT(exec_max);
 674		PN_SCHEDSTAT(slice_max);
 675		PN_SCHEDSTAT(wait_max);
 676		PN_SCHEDSTAT(wait_sum);
 677		P_SCHEDSTAT(wait_count);
 678	}
 679
 680	P(se->load.weight);
 681	P(se->avg.load_avg);
 682	P(se->avg.util_avg);
 683	P(se->avg.runnable_avg);
 684
 685#undef PN_SCHEDSTAT
 686#undef PN
 687#undef P_SCHEDSTAT
 688#undef P
 689}
 690#endif /* CONFIG_FAIR_GROUP_SCHED */
 691
 692#ifdef CONFIG_CGROUP_SCHED
 693static DEFINE_SPINLOCK(sched_debug_lock);
 694static char group_path[PATH_MAX];
 695
 696static void task_group_path(struct task_group *tg, char *path, int plen)
 697{
 698	if (autogroup_path(tg, path, plen))
 699		return;
 700
 701	cgroup_path(tg->css.cgroup, path, plen);
 702}
 703
 704/*
 705 * Only 1 SEQ_printf_task_group_path() caller can use the full length
 706 * group_path[] for cgroup path. Other simultaneous callers will have
 707 * to use a shorter stack buffer. A "..." suffix is appended at the end
 708 * of the stack buffer so that it will show up in case the output length
 709 * matches the given buffer size to indicate possible path name truncation.
 710 */
 711#define SEQ_printf_task_group_path(m, tg, fmt...)			\
 712{									\
 713	if (spin_trylock(&sched_debug_lock)) {				\
 714		task_group_path(tg, group_path, sizeof(group_path));	\
 715		SEQ_printf(m, fmt, group_path);				\
 716		spin_unlock(&sched_debug_lock);				\
 717	} else {							\
 718		char buf[128];						\
 719		char *bufend = buf + sizeof(buf) - 3;			\
 720		task_group_path(tg, buf, bufend - buf);			\
 721		strcpy(bufend - 1, "...");				\
 722		SEQ_printf(m, fmt, buf);				\
 723	}								\
 724}
 725#endif
 726
 727static void
 728print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
 729{
 730	if (task_current(rq, p))
 731		SEQ_printf(m, ">R");
 732	else
 733		SEQ_printf(m, " %c", task_state_to_char(p));
 734
 735	SEQ_printf(m, " %15s %5d %9Ld.%06ld   %c   %9Ld.%06ld %c %9Ld.%06ld %9Ld.%06ld %9Ld   %5d ",
 736		p->comm, task_pid_nr(p),
 737		SPLIT_NS(p->se.vruntime),
 738		entity_eligible(cfs_rq_of(&p->se), &p->se) ? 'E' : 'N',
 739		SPLIT_NS(p->se.deadline),
 740		p->se.custom_slice ? 'S' : ' ',
 741		SPLIT_NS(p->se.slice),
 742		SPLIT_NS(p->se.sum_exec_runtime),
 743		(long long)(p->nvcsw + p->nivcsw),
 744		p->prio);
 745
 746	SEQ_printf(m, "%9lld.%06ld %9lld.%06ld %9lld.%06ld",
 747		SPLIT_NS(schedstat_val_or_zero(p->stats.wait_sum)),
 748		SPLIT_NS(schedstat_val_or_zero(p->stats.sum_sleep_runtime)),
 749		SPLIT_NS(schedstat_val_or_zero(p->stats.sum_block_runtime)));
 750
 751#ifdef CONFIG_NUMA_BALANCING
 752	SEQ_printf(m, "   %d      %d", task_node(p), task_numa_group_id(p));
 753#endif
 754#ifdef CONFIG_CGROUP_SCHED
 755	SEQ_printf_task_group_path(m, task_group(p), "        %s")
 756#endif
 757
 758	SEQ_printf(m, "\n");
 759}
 760
 761static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
 762{
 763	struct task_struct *g, *p;
 764
 765	SEQ_printf(m, "\n");
 766	SEQ_printf(m, "runnable tasks:\n");
 767	SEQ_printf(m, " S            task   PID       vruntime   eligible    "
 768		   "deadline             slice          sum-exec      switches  "
 769		   "prio         wait-time        sum-sleep       sum-block"
 770#ifdef CONFIG_NUMA_BALANCING
 771		   "  node   group-id"
 772#endif
 773#ifdef CONFIG_CGROUP_SCHED
 774		   "  group-path"
 775#endif
 776		   "\n");
 777	SEQ_printf(m, "-------------------------------------------------------"
 778		   "------------------------------------------------------"
 779		   "------------------------------------------------------"
 780#ifdef CONFIG_NUMA_BALANCING
 781		   "--------------"
 782#endif
 783#ifdef CONFIG_CGROUP_SCHED
 784		   "--------------"
 785#endif
 786		   "\n");
 787
 788	rcu_read_lock();
 789	for_each_process_thread(g, p) {
 790		if (task_cpu(p) != rq_cpu)
 791			continue;
 792
 793		print_task(m, rq, p);
 794	}
 795	rcu_read_unlock();
 796}
 797
 798void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
 799{
 800	s64 left_vruntime = -1, zero_vruntime, right_vruntime = -1, left_deadline = -1, spread;
 801	u64 avruntime;
 802	struct sched_entity *last, *first, *root;
 803	struct rq *rq = cpu_rq(cpu);
 804	unsigned long flags;
 805
 806#ifdef CONFIG_FAIR_GROUP_SCHED
 807	SEQ_printf(m, "\n");
 808	SEQ_printf_task_group_path(m, cfs_rq->tg, "cfs_rq[%d]:%s\n", cpu);
 809#else
 810	SEQ_printf(m, "\n");
 811	SEQ_printf(m, "cfs_rq[%d]:\n", cpu);
 812#endif
 813
 814	raw_spin_rq_lock_irqsave(rq, flags);
 815	root = __pick_root_entity(cfs_rq);
 816	if (root)
 817		left_vruntime = root->min_vruntime;
 818	first = __pick_first_entity(cfs_rq);
 819	if (first)
 820		left_deadline = first->deadline;
 821	last = __pick_last_entity(cfs_rq);
 822	if (last)
 823		right_vruntime = last->vruntime;
 824	zero_vruntime = cfs_rq->zero_vruntime;
 825	avruntime = avg_vruntime(cfs_rq);
 826	raw_spin_rq_unlock_irqrestore(rq, flags);
 827
 828	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "left_deadline",
 829			SPLIT_NS(left_deadline));
 830	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "left_vruntime",
 831			SPLIT_NS(left_vruntime));
 832	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "zero_vruntime",
 833			SPLIT_NS(zero_vruntime));
 834	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "avg_vruntime",
 835			SPLIT_NS(avruntime));
 836	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "right_vruntime",
 837			SPLIT_NS(right_vruntime));
 838	spread = right_vruntime - left_vruntime;
 839	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread", SPLIT_NS(spread));
 840	SEQ_printf(m, "  .%-30s: %d\n", "nr_queued", cfs_rq->nr_queued);
 841	SEQ_printf(m, "  .%-30s: %d\n", "h_nr_runnable", cfs_rq->h_nr_runnable);
 842	SEQ_printf(m, "  .%-30s: %d\n", "h_nr_queued", cfs_rq->h_nr_queued);
 843	SEQ_printf(m, "  .%-30s: %d\n", "h_nr_idle", cfs_rq->h_nr_idle);
 844	SEQ_printf(m, "  .%-30s: %ld\n", "load", cfs_rq->load.weight);
 845	SEQ_printf(m, "  .%-30s: %lu\n", "load_avg",
 846			cfs_rq->avg.load_avg);
 847	SEQ_printf(m, "  .%-30s: %lu\n", "runnable_avg",
 848			cfs_rq->avg.runnable_avg);
 849	SEQ_printf(m, "  .%-30s: %lu\n", "util_avg",
 850			cfs_rq->avg.util_avg);
 851	SEQ_printf(m, "  .%-30s: %u\n", "util_est",
 852			cfs_rq->avg.util_est);
 853	SEQ_printf(m, "  .%-30s: %ld\n", "removed.load_avg",
 854			cfs_rq->removed.load_avg);
 855	SEQ_printf(m, "  .%-30s: %ld\n", "removed.util_avg",
 856			cfs_rq->removed.util_avg);
 857	SEQ_printf(m, "  .%-30s: %ld\n", "removed.runnable_avg",
 858			cfs_rq->removed.runnable_avg);
 859#ifdef CONFIG_FAIR_GROUP_SCHED
 860	SEQ_printf(m, "  .%-30s: %lu\n", "tg_load_avg_contrib",
 861			cfs_rq->tg_load_avg_contrib);
 862	SEQ_printf(m, "  .%-30s: %ld\n", "tg_load_avg",
 863			atomic_long_read(&cfs_rq->tg->load_avg));
 864#endif /* CONFIG_FAIR_GROUP_SCHED */
 865#ifdef CONFIG_CFS_BANDWIDTH
 866	SEQ_printf(m, "  .%-30s: %d\n", "throttled",
 867			cfs_rq->throttled);
 868	SEQ_printf(m, "  .%-30s: %d\n", "throttle_count",
 869			cfs_rq->throttle_count);
 870#endif
 871
 872#ifdef CONFIG_FAIR_GROUP_SCHED
 873	print_cfs_group_stats(m, cpu, cfs_rq->tg);
 874#endif
 875}
 876
 877void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
 878{
 879#ifdef CONFIG_RT_GROUP_SCHED
 880	SEQ_printf(m, "\n");
 881	SEQ_printf_task_group_path(m, rt_rq->tg, "rt_rq[%d]:%s\n", cpu);
 882#else
 883	SEQ_printf(m, "\n");
 884	SEQ_printf(m, "rt_rq[%d]:\n", cpu);
 885#endif
 886
 887#define P(x) \
 888	SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
 889#define PU(x) \
 890	SEQ_printf(m, "  .%-30s: %lu\n", #x, (unsigned long)(rt_rq->x))
 891#define PN(x) \
 892	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
 893
 894	PU(rt_nr_running);
 895
 896#ifdef CONFIG_RT_GROUP_SCHED
 897	P(rt_throttled);
 898	PN(rt_time);
 899	PN(rt_runtime);
 900#endif
 901
 902#undef PN
 903#undef PU
 904#undef P
 905}
 906
 907void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq)
 908{
 909	struct dl_bw *dl_bw;
 910
 911	SEQ_printf(m, "\n");
 912	SEQ_printf(m, "dl_rq[%d]:\n", cpu);
 913
 914#define PU(x) \
 915	SEQ_printf(m, "  .%-30s: %lu\n", #x, (unsigned long)(dl_rq->x))
 916
 917	PU(dl_nr_running);
 918	dl_bw = &cpu_rq(cpu)->rd->dl_bw;
 919	SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->bw", dl_bw->bw);
 920	SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->total_bw", dl_bw->total_bw);
 921
 922#undef PU
 923}
 924
 925static void print_cpu(struct seq_file *m, int cpu)
 926{
 927	struct rq *rq = cpu_rq(cpu);
 928
 929#ifdef CONFIG_X86
 930	{
 931		unsigned int freq = cpu_khz ? : 1;
 932
 933		SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
 934			   cpu, freq / 1000, (freq % 1000));
 935	}
 936#else /* !CONFIG_X86: */
 937	SEQ_printf(m, "cpu#%d\n", cpu);
 938#endif /* !CONFIG_X86 */
 939
 940#define P(x)								\
 941do {									\
 942	if (sizeof(rq->x) == 4)						\
 943		SEQ_printf(m, "  .%-30s: %d\n", #x, (int)(rq->x));	\
 944	else								\
 945		SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rq->x));\
 946} while (0)
 947
 948#define PN(x) \
 949	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
 950
 951	P(nr_running);
 952	P(nr_switches);
 953	P(nr_uninterruptible);
 954	PN(next_balance);
 955	SEQ_printf(m, "  .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr)));
 956	PN(clock);
 957	PN(clock_task);
 958#undef P
 959#undef PN
 960
 961#define P64(n) SEQ_printf(m, "  .%-30s: %Ld\n", #n, rq->n);
 962	P64(avg_idle);
 963	P64(max_idle_balance_cost);
 964#undef P64
 965
 966#define P(n) SEQ_printf(m, "  .%-30s: %d\n", #n, schedstat_val(rq->n));
 967	if (schedstat_enabled()) {
 968		P(yld_count);
 969		P(sched_count);
 970		P(sched_goidle);
 971		P(ttwu_count);
 972		P(ttwu_local);
 973	}
 974#undef P
 975
 976	print_cfs_stats(m, cpu);
 977	print_rt_stats(m, cpu);
 978	print_dl_stats(m, cpu);
 979
 980	print_rq(m, rq, cpu);
 981	SEQ_printf(m, "\n");
 982}
 983
 984static const char *sched_tunable_scaling_names[] = {
 985	"none",
 986	"logarithmic",
 987	"linear"
 988};
 989
 990static void sched_debug_header(struct seq_file *m)
 991{
 992	u64 ktime, sched_clk, cpu_clk;
 993	unsigned long flags;
 994
 995	local_irq_save(flags);
 996	ktime = ktime_to_ns(ktime_get());
 997	sched_clk = sched_clock();
 998	cpu_clk = local_clock();
 999	local_irq_restore(flags);
1000
1001	SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n",
1002		init_utsname()->release,
1003		(int)strcspn(init_utsname()->version, " "),
1004		init_utsname()->version);
1005
1006#define P(x) \
1007	SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
1008#define PN(x) \
1009	SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
1010	PN(ktime);
1011	PN(sched_clk);
1012	PN(cpu_clk);
1013	P(jiffies);
1014#ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
1015	P(sched_clock_stable());
1016#endif
1017#undef PN
1018#undef P
1019
1020	SEQ_printf(m, "\n");
1021	SEQ_printf(m, "sysctl_sched\n");
1022
1023#define P(x) \
1024	SEQ_printf(m, "  .%-40s: %Ld\n", #x, (long long)(x))
1025#define PN(x) \
1026	SEQ_printf(m, "  .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
1027	PN(sysctl_sched_base_slice);
1028	P(sysctl_sched_features);
1029#undef PN
1030#undef P
1031
1032	SEQ_printf(m, "  .%-40s: %d (%s)\n",
1033		"sysctl_sched_tunable_scaling",
1034		sysctl_sched_tunable_scaling,
1035		sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
1036	SEQ_printf(m, "\n");
1037}
1038
1039static int sched_debug_show(struct seq_file *m, void *v)
1040{
1041	int cpu = (unsigned long)(v - 2);
1042
1043	if (cpu != -1)
1044		print_cpu(m, cpu);
1045	else
1046		sched_debug_header(m);
1047
1048	return 0;
1049}
1050
1051void sysrq_sched_debug_show(void)
1052{
1053	int cpu;
1054
1055	sched_debug_header(NULL);
1056	for_each_online_cpu(cpu) {
1057		/*
1058		 * Need to reset softlockup watchdogs on all CPUs, because
1059		 * another CPU might be blocked waiting for us to process
1060		 * an IPI or stop_machine.
1061		 */
1062		touch_nmi_watchdog();
1063		touch_all_softlockup_watchdogs();
1064		print_cpu(NULL, cpu);
1065	}
1066}
1067
1068/*
1069 * This iterator needs some explanation.
1070 * It returns 1 for the header position.
1071 * This means 2 is CPU 0.
1072 * In a hotplugged system some CPUs, including CPU 0, may be missing so we have
1073 * to use cpumask_* to iterate over the CPUs.
1074 */
1075static void *sched_debug_start(struct seq_file *file, loff_t *offset)
1076{
1077	unsigned long n = *offset;
1078
1079	if (n == 0)
1080		return (void *) 1;
1081
1082	n--;
1083
1084	if (n > 0)
1085		n = cpumask_next(n - 1, cpu_online_mask);
1086	else
1087		n = cpumask_first(cpu_online_mask);
1088
1089	*offset = n + 1;
1090
1091	if (n < nr_cpu_ids)
1092		return (void *)(unsigned long)(n + 2);
1093
1094	return NULL;
1095}
1096
1097static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
1098{
1099	(*offset)++;
1100	return sched_debug_start(file, offset);
1101}
1102
1103static void sched_debug_stop(struct seq_file *file, void *data)
1104{
1105}
1106
1107static const struct seq_operations sched_debug_sops = {
1108	.start		= sched_debug_start,
1109	.next		= sched_debug_next,
1110	.stop		= sched_debug_stop,
1111	.show		= sched_debug_show,
1112};
1113
1114#define __PS(S, F) SEQ_printf(m, "%-45s:%21Ld\n", S, (long long)(F))
1115#define __P(F) __PS(#F, F)
1116#define   P(F) __PS(#F, p->F)
1117#define   PM(F, M) __PS(#F, p->F & (M))
1118#define __PSN(S, F) SEQ_printf(m, "%-45s:%14Ld.%06ld\n", S, SPLIT_NS((long long)(F)))
1119#define __PN(F) __PSN(#F, F)
1120#define   PN(F) __PSN(#F, p->F)
1121
1122
1123#ifdef CONFIG_NUMA_BALANCING
1124void print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
1125		unsigned long tpf, unsigned long gsf, unsigned long gpf)
1126{
1127	SEQ_printf(m, "numa_faults node=%d ", node);
1128	SEQ_printf(m, "task_private=%lu task_shared=%lu ", tpf, tsf);
1129	SEQ_printf(m, "group_private=%lu group_shared=%lu\n", gpf, gsf);
1130}
1131#endif
1132
1133
1134static void sched_show_numa(struct task_struct *p, struct seq_file *m)
1135{
1136#ifdef CONFIG_NUMA_BALANCING
1137	if (p->mm)
1138		P(mm->numa_scan_seq);
1139
1140	P(numa_pages_migrated);
1141	P(numa_preferred_nid);
1142	P(total_numa_faults);
1143	SEQ_printf(m, "current_node=%d, numa_group_id=%d\n",
1144			task_node(p), task_numa_group_id(p));
1145	show_numa_stats(p, m);
1146#endif /* CONFIG_NUMA_BALANCING */
1147}
1148
1149void proc_sched_show_task(struct task_struct *p, struct pid_namespace *ns,
1150						  struct seq_file *m)
1151{
1152	unsigned long nr_switches;
1153
1154	SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr_ns(p, ns),
1155						get_nr_threads(p));
1156	SEQ_printf(m,
1157		"---------------------------------------------------------"
1158		"----------\n");
1159
1160#define P_SCHEDSTAT(F)  __PS(#F, schedstat_val(p->stats.F))
1161#define PN_SCHEDSTAT(F) __PSN(#F, schedstat_val(p->stats.F))
1162
1163	PN(se.exec_start);
1164	PN(se.vruntime);
1165	PN(se.sum_exec_runtime);
1166
1167	nr_switches = p->nvcsw + p->nivcsw;
1168
1169	P(se.nr_migrations);
1170
1171	if (schedstat_enabled()) {
1172		u64 avg_atom, avg_per_cpu;
1173
1174		PN_SCHEDSTAT(sum_sleep_runtime);
1175		PN_SCHEDSTAT(sum_block_runtime);
1176		PN_SCHEDSTAT(wait_start);
1177		PN_SCHEDSTAT(sleep_start);
1178		PN_SCHEDSTAT(block_start);
1179		PN_SCHEDSTAT(sleep_max);
1180		PN_SCHEDSTAT(block_max);
1181		PN_SCHEDSTAT(exec_max);
1182		PN_SCHEDSTAT(slice_max);
1183		PN_SCHEDSTAT(wait_max);
1184		PN_SCHEDSTAT(wait_sum);
1185		P_SCHEDSTAT(wait_count);
1186		PN_SCHEDSTAT(iowait_sum);
1187		P_SCHEDSTAT(iowait_count);
1188		P_SCHEDSTAT(nr_migrations_cold);
1189		P_SCHEDSTAT(nr_failed_migrations_affine);
1190		P_SCHEDSTAT(nr_failed_migrations_running);
1191		P_SCHEDSTAT(nr_failed_migrations_hot);
1192		P_SCHEDSTAT(nr_forced_migrations);
1193		P_SCHEDSTAT(nr_wakeups);
1194		P_SCHEDSTAT(nr_wakeups_sync);
1195		P_SCHEDSTAT(nr_wakeups_migrate);
1196		P_SCHEDSTAT(nr_wakeups_local);
1197		P_SCHEDSTAT(nr_wakeups_remote);
1198		P_SCHEDSTAT(nr_wakeups_affine);
1199		P_SCHEDSTAT(nr_wakeups_affine_attempts);
1200		P_SCHEDSTAT(nr_wakeups_passive);
1201		P_SCHEDSTAT(nr_wakeups_idle);
1202
1203		avg_atom = p->se.sum_exec_runtime;
1204		if (nr_switches)
1205			avg_atom = div64_ul(avg_atom, nr_switches);
1206		else
1207			avg_atom = -1LL;
1208
1209		avg_per_cpu = p->se.sum_exec_runtime;
1210		if (p->se.nr_migrations) {
1211			avg_per_cpu = div64_u64(avg_per_cpu,
1212						p->se.nr_migrations);
1213		} else {
1214			avg_per_cpu = -1LL;
1215		}
1216
1217		__PN(avg_atom);
1218		__PN(avg_per_cpu);
1219
1220#ifdef CONFIG_SCHED_CORE
1221		PN_SCHEDSTAT(core_forceidle_sum);
1222#endif
1223	}
1224
1225	__P(nr_switches);
1226	__PS("nr_voluntary_switches", p->nvcsw);
1227	__PS("nr_involuntary_switches", p->nivcsw);
1228
1229	P(se.load.weight);
1230	P(se.avg.load_sum);
1231	P(se.avg.runnable_sum);
1232	P(se.avg.util_sum);
1233	P(se.avg.load_avg);
1234	P(se.avg.runnable_avg);
1235	P(se.avg.util_avg);
1236	P(se.avg.last_update_time);
1237	PM(se.avg.util_est, ~UTIL_AVG_UNCHANGED);
1238#ifdef CONFIG_UCLAMP_TASK
1239	__PS("uclamp.min", p->uclamp_req[UCLAMP_MIN].value);
1240	__PS("uclamp.max", p->uclamp_req[UCLAMP_MAX].value);
1241	__PS("effective uclamp.min", uclamp_eff_value(p, UCLAMP_MIN));
1242	__PS("effective uclamp.max", uclamp_eff_value(p, UCLAMP_MAX));
1243#endif /* CONFIG_UCLAMP_TASK */
1244	P(policy);
1245	P(prio);
1246	if (task_has_dl_policy(p)) {
1247		P(dl.runtime);
1248		P(dl.deadline);
1249	} else if (fair_policy(p->policy)) {
1250		P(se.slice);
1251	}
1252#ifdef CONFIG_SCHED_CLASS_EXT
1253	__PS("ext.enabled", task_on_scx(p));
1254#endif
1255#undef PN_SCHEDSTAT
1256#undef P_SCHEDSTAT
1257
1258	{
1259		unsigned int this_cpu = raw_smp_processor_id();
1260		u64 t0, t1;
1261
1262		t0 = cpu_clock(this_cpu);
1263		t1 = cpu_clock(this_cpu);
1264		__PS("clock-delta", t1-t0);
1265	}
1266
1267	sched_show_numa(p, m);
1268}
1269
1270void proc_sched_set_task(struct task_struct *p)
1271{
1272#ifdef CONFIG_SCHEDSTATS
1273	memset(&p->stats, 0, sizeof(p->stats));
1274#endif
1275}
1276
1277void resched_latency_warn(int cpu, u64 latency)
1278{
1279	static DEFINE_RATELIMIT_STATE(latency_check_ratelimit, 60 * 60 * HZ, 1);
1280
1281	if (likely(!__ratelimit(&latency_check_ratelimit)))
1282		return;
1283
1284	pr_err("sched: CPU %d need_resched set for > %llu ns (%d ticks) without schedule\n",
1285	       cpu, latency, cpu_rq(cpu)->ticks_without_resched);
1286	dump_stack();
1287}