kernel/sched/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
  10 /*
  11  * This allows printing both to /sys/kernel/debug/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  */
  25 static 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
  37 static 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
  50 static const char * const sched_feat_names[] = {
  51 #include "features.h"
  52 };
  53
  54 #undef SCHED_FEAT
  55
  56 static 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
  78 struct static_key sched_feat_keys[__SCHED_FEAT_NR] = {
  79 #include "features.h"
  80 };
  81
  82 #undef SCHED_FEAT
  83
  84 static void sched_feat_disable(int i)
  85 {
  86         static_key_disable_cpuslocked(&sched_feat_keys[i]);
  87 }
  88
  89 static void sched_feat_enable(int i)
  90 {
  91         static_key_enable_cpuslocked(&sched_feat_keys[i]);
  92 }
  93 #else
  94 static void sched_feat_disable(int i) { };
  95 static void sched_feat_enable(int i) { };
  96 #endif /* CONFIG_JUMP_LABEL */
  97
  98 static 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
 123 static ssize_t
 124 sched_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
 156 static int sched_feat_open(struct inode *inode, struct file *filp)
 157 {
 158         return single_open(filp, sched_feat_show, NULL);
 159 }
 160
 161 static 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
 171 static 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
 198 static 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
 204 static int sched_scaling_open(struct inode *inode, struct file *filp)
 205 {
 206         return single_open(filp, sched_scaling_show, NULL);
 207 }
 208
 209 static 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
 221 static 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
 245 static 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
 266 static int sched_dynamic_open(struct inode *inode, struct file *filp)
 267 {
 268         return single_open(filp, sched_dynamic_show, NULL);
 269 }
 270
 271 static 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
 283 #ifdef CONFIG_SMP
 284 static struct dentry           *sd_dentry;
 285
 286
 287 static ssize_t sched_verbose_write(struct file *filp, const char __user *ubuf,
 288                                   size_t cnt, loff_t *ppos)
 289 {
 290         ssize_t result;
 291         bool orig;
 292
 293         cpus_read_lock();
 294         mutex_lock(&sched_domains_mutex);
 295
 296         orig = sched_debug_verbose;
 297         result = debugfs_write_file_bool(filp, ubuf, cnt, ppos);
 298
 299         if (sched_debug_verbose && !orig)
 300                 update_sched_domain_debugfs();
 301         else if (!sched_debug_verbose && orig) {
 302                 debugfs_remove(sd_dentry);
 303                 sd_dentry = NULL;
 304         }
 305
 306         mutex_unlock(&sched_domains_mutex);
 307         cpus_read_unlock();
 308
 309         return result;
 310 }
 311 #else
 312 #define sched_verbose_write debugfs_write_file_bool
 313 #endif
 314
 315 static const struct file_operations sched_verbose_fops = {
 316         .read =         debugfs_read_file_bool,
 317         .write =        sched_verbose_write,
 318         .open =         simple_open,
 319         .llseek =       default_llseek,
 320 };
 321
 322 static const struct seq_operations sched_debug_sops;
 323
 324 static int sched_debug_open(struct inode *inode, struct file *filp)
 325 {
 326         return seq_open(filp, &sched_debug_sops);
 327 }
 328
 329 static const struct file_operations sched_debug_fops = {
 330         .open           = sched_debug_open,
 331         .read           = seq_read,
 332         .llseek         = seq_lseek,
 333         .release        = seq_release,
 334 };
 335
 336 static struct dentry *debugfs_sched;
 337
 338 static __init int sched_init_debug(void)
 339 {
 340         struct dentry __maybe_unused *numa;
 341
 342         debugfs_sched = debugfs_create_dir("sched", NULL);
 343
 344         debugfs_create_file("features", 0644, debugfs_sched, NULL, &sched_feat_fops);
 345         debugfs_create_file_unsafe("verbose", 0644, debugfs_sched, &sched_debug_verbose, &sched_verbose_fops);
 346 #ifdef CONFIG_PREEMPT_DYNAMIC
 347         debugfs_create_file("preempt", 0644, debugfs_sched, NULL, &sched_dynamic_fops);
 348 #endif
 349
 350         debugfs_create_u32("base_slice_ns", 0644, debugfs_sched, &sysctl_sched_base_slice);
 351
 352         debugfs_create_u32("latency_warn_ms", 0644, debugfs_sched, &sysctl_resched_latency_warn_ms);
 353         debugfs_create_u32("latency_warn_once", 0644, debugfs_sched, &sysctl_resched_latency_warn_once);
 354
 355 #ifdef CONFIG_SMP
 356         debugfs_create_file("tunable_scaling", 0644, debugfs_sched, NULL, &sched_scaling_fops);
 357         debugfs_create_u32("migration_cost_ns", 0644, debugfs_sched, &sysctl_sched_migration_cost);
 358         debugfs_create_u32("nr_migrate", 0644, debugfs_sched, &sysctl_sched_nr_migrate);
 359
 360         mutex_lock(&sched_domains_mutex);
 361         update_sched_domain_debugfs();
 362         mutex_unlock(&sched_domains_mutex);
 363 #endif
 364
 365 #ifdef CONFIG_NUMA_BALANCING
 366         numa = debugfs_create_dir("numa_balancing", debugfs_sched);
 367
 368         debugfs_create_u32("scan_delay_ms", 0644, numa, &sysctl_numa_balancing_scan_delay);
 369         debugfs_create_u32("scan_period_min_ms", 0644, numa, &sysctl_numa_balancing_scan_period_min);
 370         debugfs_create_u32("scan_period_max_ms", 0644, numa, &sysctl_numa_balancing_scan_period_max);
 371         debugfs_create_u32("scan_size_mb", 0644, numa, &sysctl_numa_balancing_scan_size);
 372         debugfs_create_u32("hot_threshold_ms", 0644, numa, &sysctl_numa_balancing_hot_threshold);
 373 #endif
 374
 375         debugfs_create_file("debug", 0444, debugfs_sched, NULL, &sched_debug_fops);
 376
 377         return 0;
 378 }
 379 late_initcall(sched_init_debug);
 380
 381 #ifdef CONFIG_SMP
 382
 383 static cpumask_var_t            sd_sysctl_cpus;
 384
 385 static int sd_flags_show(struct seq_file *m, void *v)
 386 {
 387         unsigned long flags = *(unsigned int *)m->private;
 388         int idx;
 389
 390         for_each_set_bit(idx, &flags, __SD_FLAG_CNT) {
 391                 seq_puts(m, sd_flag_debug[idx].name);
 392                 seq_puts(m, " ");
 393         }
 394         seq_puts(m, "\n");
 395
 396         return 0;
 397 }
 398
 399 static int sd_flags_open(struct inode *inode, struct file *file)
 400 {
 401         return single_open(file, sd_flags_show, inode->i_private);
 402 }
 403
 404 static const struct file_operations sd_flags_fops = {
 405         .open           = sd_flags_open,
 406         .read           = seq_read,
 407         .llseek         = seq_lseek,
 408         .release        = single_release,
 409 };
 410
 411 static void register_sd(struct sched_domain *sd, struct dentry *parent)
 412 {
 413 #define SDM(type, mode, member) \
 414         debugfs_create_##type(#member, mode, parent, &sd->member)
 415
 416         SDM(ulong, 0644, min_interval);
 417         SDM(ulong, 0644, max_interval);
 418         SDM(u64,   0644, max_newidle_lb_cost);
 419         SDM(u32,   0644, busy_factor);
 420         SDM(u32,   0644, imbalance_pct);
 421         SDM(u32,   0644, cache_nice_tries);
 422         SDM(str,   0444, name);
 423
 424 #undef SDM
 425
 426         debugfs_create_file("flags", 0444, parent, &sd->flags, &sd_flags_fops);
 427         debugfs_create_file("groups_flags", 0444, parent, &sd->groups->flags, &sd_flags_fops);
 428         debugfs_create_u32("level", 0444, parent, (u32 *)&sd->level);
 429 }
 430
 431 void update_sched_domain_debugfs(void)
 432 {
 433         int cpu, i;
 434
 435         /*
 436          * This can unfortunately be invoked before sched_debug_init() creates
 437          * the debug directory. Don't touch sd_sysctl_cpus until then.
 438          */
 439         if (!debugfs_sched)
 440                 return;
 441
 442         if (!sched_debug_verbose)
 443                 return;
 444
 445         if (!cpumask_available(sd_sysctl_cpus)) {
 446                 if (!alloc_cpumask_var(&sd_sysctl_cpus, GFP_KERNEL))
 447                         return;
 448                 cpumask_copy(sd_sysctl_cpus, cpu_possible_mask);
 449         }
 450
 451         if (!sd_dentry) {
 452                 sd_dentry = debugfs_create_dir("domains", debugfs_sched);
 453
 454                 /* rebuild sd_sysctl_cpus if empty since it gets cleared below */
 455                 if (cpumask_empty(sd_sysctl_cpus))
 456                         cpumask_copy(sd_sysctl_cpus, cpu_online_mask);
 457         }
 458
 459         for_each_cpu(cpu, sd_sysctl_cpus) {
 460                 struct sched_domain *sd;
 461                 struct dentry *d_cpu;
 462                 char buf[32];
 463
 464                 snprintf(buf, sizeof(buf), "cpu%d", cpu);
 465                 debugfs_lookup_and_remove(buf, sd_dentry);
 466                 d_cpu = debugfs_create_dir(buf, sd_dentry);
 467
 468                 i = 0;
 469                 for_each_domain(cpu, sd) {
 470                         struct dentry *d_sd;
 471
 472                         snprintf(buf, sizeof(buf), "domain%d", i);
 473                         d_sd = debugfs_create_dir(buf, d_cpu);
 474
 475                         register_sd(sd, d_sd);
 476                         i++;
 477                 }
 478
 479                 __cpumask_clear_cpu(cpu, sd_sysctl_cpus);
 480         }
 481 }
 482
 483 void dirty_sched_domain_sysctl(int cpu)
 484 {
 485         if (cpumask_available(sd_sysctl_cpus))
 486                 __cpumask_set_cpu(cpu, sd_sysctl_cpus);
 487 }
 488
 489 #endif /* CONFIG_SMP */
 490
 491 #ifdef CONFIG_FAIR_GROUP_SCHED
 492 static void print_cfs_group_stats(struct seq_file *m, int cpu, struct task_group *tg)
 493 {
 494         struct sched_entity *se = tg->se[cpu];
 495
 496 #define P(F)            SEQ_printf(m, "  .%-30s: %lld\n",       #F, (long long)F)
 497 #define P_SCHEDSTAT(F)  SEQ_printf(m, "  .%-30s: %lld\n",       \
 498                 #F, (long long)schedstat_val(stats->F))
 499 #define PN(F)           SEQ_printf(m, "  .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F))
 500 #define PN_SCHEDSTAT(F) SEQ_printf(m, "  .%-30s: %lld.%06ld\n", \
 501                 #F, SPLIT_NS((long long)schedstat_val(stats->F)))
 502
 503         if (!se)
 504                 return;
 505
 506         PN(se->exec_start);
 507         PN(se->vruntime);
 508         PN(se->sum_exec_runtime);
 509
 510         if (schedstat_enabled()) {
 511                 struct sched_statistics *stats;
 512                 stats = __schedstats_from_se(se);
 513
 514                 PN_SCHEDSTAT(wait_start);
 515                 PN_SCHEDSTAT(sleep_start);
 516                 PN_SCHEDSTAT(block_start);
 517                 PN_SCHEDSTAT(sleep_max);
 518                 PN_SCHEDSTAT(block_max);
 519                 PN_SCHEDSTAT(exec_max);
 520                 PN_SCHEDSTAT(slice_max);
 521                 PN_SCHEDSTAT(wait_max);
 522                 PN_SCHEDSTAT(wait_sum);
 523                 P_SCHEDSTAT(wait_count);
 524         }
 525
 526         P(se->load.weight);
 527 #ifdef CONFIG_SMP
 528         P(se->avg.load_avg);
 529         P(se->avg.util_avg);
 530         P(se->avg.runnable_avg);
 531 #endif
 532
 533 #undef PN_SCHEDSTAT
 534 #undef PN
 535 #undef P_SCHEDSTAT
 536 #undef P
 537 }
 538 #endif
 539
 540 #ifdef CONFIG_CGROUP_SCHED
 541 static DEFINE_SPINLOCK(sched_debug_lock);
 542 static char group_path[PATH_MAX];
 543
 544 static void task_group_path(struct task_group *tg, char *path, int plen)
 545 {
 546         if (autogroup_path(tg, path, plen))
 547                 return;
 548
 549         cgroup_path(tg->css.cgroup, path, plen);
 550 }
 551
 552 /*
 553  * Only 1 SEQ_printf_task_group_path() caller can use the full length
 554  * group_path[] for cgroup path. Other simultaneous callers will have
 555  * to use a shorter stack buffer. A "..." suffix is appended at the end
 556  * of the stack buffer so that it will show up in case the output length
 557  * matches the given buffer size to indicate possible path name truncation.
 558  */
 559 #define SEQ_printf_task_group_path(m, tg, fmt...)                       \
 560 {                                                                       \
 561         if (spin_trylock(&sched_debug_lock)) {                          \
 562                 task_group_path(tg, group_path, sizeof(group_path));    \
 563                 SEQ_printf(m, fmt, group_path);                         \
 564                 spin_unlock(&sched_debug_lock);                         \
 565         } else {                                                        \
 566                 char buf[128];                                          \
 567                 char *bufend = buf + sizeof(buf) - 3;                   \
 568                 task_group_path(tg, buf, bufend - buf);                 \
 569                 strcpy(bufend - 1, "...");                              \
 570                 SEQ_printf(m, fmt, buf);                                \
 571         }                                                               \
 572 }
 573 #endif
 574
 575 static void
 576 print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
 577 {
 578         if (task_current(rq, p))
 579                 SEQ_printf(m, ">R");
 580         else
 581                 SEQ_printf(m, " %c", task_state_to_char(p));
 582
 583         SEQ_printf(m, "%15s %5d %9Ld.%06ld %c %9Ld.%06ld %9Ld.%06ld %9Ld.%06ld %9Ld %5d ",
 584                 p->comm, task_pid_nr(p),
 585                 SPLIT_NS(p->se.vruntime),
 586                 entity_eligible(cfs_rq_of(&p->se), &p->se) ? 'E' : 'N',
 587                 SPLIT_NS(p->se.deadline),
 588                 SPLIT_NS(p->se.slice),
 589                 SPLIT_NS(p->se.sum_exec_runtime),
 590                 (long long)(p->nvcsw + p->nivcsw),
 591                 p->prio);
 592
 593         SEQ_printf(m, "%9lld.%06ld %9lld.%06ld %9lld.%06ld %9lld.%06ld",
 594                 SPLIT_NS(schedstat_val_or_zero(p->stats.wait_sum)),
 595                 SPLIT_NS(p->se.sum_exec_runtime),
 596                 SPLIT_NS(schedstat_val_or_zero(p->stats.sum_sleep_runtime)),
 597                 SPLIT_NS(schedstat_val_or_zero(p->stats.sum_block_runtime)));
 598
 599 #ifdef CONFIG_NUMA_BALANCING
 600         SEQ_printf(m, " %d %d", task_node(p), task_numa_group_id(p));
 601 #endif
 602 #ifdef CONFIG_CGROUP_SCHED
 603         SEQ_printf_task_group_path(m, task_group(p), " %s")
 604 #endif
 605
 606         SEQ_printf(m, "\n");
 607 }
 608
 609 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
 610 {
 611         struct task_struct *g, *p;
 612
 613         SEQ_printf(m, "\n");
 614         SEQ_printf(m, "runnable tasks:\n");
 615         SEQ_printf(m, " S            task   PID         tree-key  switches  prio"
 616                    "     wait-time             sum-exec        sum-sleep\n");
 617         SEQ_printf(m, "-------------------------------------------------------"
 618                    "------------------------------------------------------\n");
 619
 620         rcu_read_lock();
 621         for_each_process_thread(g, p) {
 622                 if (task_cpu(p) != rq_cpu)
 623                         continue;
 624
 625                 print_task(m, rq, p);
 626         }
 627         rcu_read_unlock();
 628 }
 629
 630 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
 631 {
 632         s64 left_vruntime = -1, min_vruntime, right_vruntime = -1, left_deadline = -1, spread;
 633         struct sched_entity *last, *first, *root;
 634         struct rq *rq = cpu_rq(cpu);
 635         unsigned long flags;
 636
 637 #ifdef CONFIG_FAIR_GROUP_SCHED
 638         SEQ_printf(m, "\n");
 639         SEQ_printf_task_group_path(m, cfs_rq->tg, "cfs_rq[%d]:%s\n", cpu);
 640 #else
 641         SEQ_printf(m, "\n");
 642         SEQ_printf(m, "cfs_rq[%d]:\n", cpu);
 643 #endif
 644         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "exec_clock",
 645                         SPLIT_NS(cfs_rq->exec_clock));
 646
 647         raw_spin_rq_lock_irqsave(rq, flags);
 648         root = __pick_root_entity(cfs_rq);
 649         if (root)
 650                 left_vruntime = root->min_vruntime;
 651         first = __pick_first_entity(cfs_rq);
 652         if (first)
 653                 left_deadline = first->deadline;
 654         last = __pick_last_entity(cfs_rq);
 655         if (last)
 656                 right_vruntime = last->vruntime;
 657         min_vruntime = cfs_rq->min_vruntime;
 658         raw_spin_rq_unlock_irqrestore(rq, flags);
 659
 660         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "left_deadline",
 661                         SPLIT_NS(left_deadline));
 662         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "left_vruntime",
 663                         SPLIT_NS(left_vruntime));
 664         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "min_vruntime",
 665                         SPLIT_NS(min_vruntime));
 666         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "avg_vruntime",
 667                         SPLIT_NS(avg_vruntime(cfs_rq)));
 668         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "right_vruntime",
 669                         SPLIT_NS(right_vruntime));
 670         spread = right_vruntime - left_vruntime;
 671         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread", SPLIT_NS(spread));
 672         SEQ_printf(m, "  .%-30s: %d\n", "nr_spread_over",
 673                         cfs_rq->nr_spread_over);
 674         SEQ_printf(m, "  .%-30s: %d\n", "nr_running", cfs_rq->nr_running);
 675         SEQ_printf(m, "  .%-30s: %d\n", "h_nr_running", cfs_rq->h_nr_running);
 676         SEQ_printf(m, "  .%-30s: %d\n", "idle_nr_running",
 677                         cfs_rq->idle_nr_running);
 678         SEQ_printf(m, "  .%-30s: %d\n", "idle_h_nr_running",
 679                         cfs_rq->idle_h_nr_running);
 680         SEQ_printf(m, "  .%-30s: %ld\n", "load", cfs_rq->load.weight);
 681 #ifdef CONFIG_SMP
 682         SEQ_printf(m, "  .%-30s: %lu\n", "load_avg",
 683                         cfs_rq->avg.load_avg);
 684         SEQ_printf(m, "  .%-30s: %lu\n", "runnable_avg",
 685                         cfs_rq->avg.runnable_avg);
 686         SEQ_printf(m, "  .%-30s: %lu\n", "util_avg",
 687                         cfs_rq->avg.util_avg);
 688         SEQ_printf(m, "  .%-30s: %u\n", "util_est",
 689                         cfs_rq->avg.util_est);
 690         SEQ_printf(m, "  .%-30s: %ld\n", "removed.load_avg",
 691                         cfs_rq->removed.load_avg);
 692         SEQ_printf(m, "  .%-30s: %ld\n", "removed.util_avg",
 693                         cfs_rq->removed.util_avg);
 694         SEQ_printf(m, "  .%-30s: %ld\n", "removed.runnable_avg",
 695                         cfs_rq->removed.runnable_avg);
 696 #ifdef CONFIG_FAIR_GROUP_SCHED
 697         SEQ_printf(m, "  .%-30s: %lu\n", "tg_load_avg_contrib",
 698                         cfs_rq->tg_load_avg_contrib);
 699         SEQ_printf(m, "  .%-30s: %ld\n", "tg_load_avg",
 700                         atomic_long_read(&cfs_rq->tg->load_avg));
 701 #endif
 702 #endif
 703 #ifdef CONFIG_CFS_BANDWIDTH
 704         SEQ_printf(m, "  .%-30s: %d\n", "throttled",
 705                         cfs_rq->throttled);
 706         SEQ_printf(m, "  .%-30s: %d\n", "throttle_count",
 707                         cfs_rq->throttle_count);
 708 #endif
 709
 710 #ifdef CONFIG_FAIR_GROUP_SCHED
 711         print_cfs_group_stats(m, cpu, cfs_rq->tg);
 712 #endif
 713 }
 714
 715 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
 716 {
 717 #ifdef CONFIG_RT_GROUP_SCHED
 718         SEQ_printf(m, "\n");
 719         SEQ_printf_task_group_path(m, rt_rq->tg, "rt_rq[%d]:%s\n", cpu);
 720 #else
 721         SEQ_printf(m, "\n");
 722         SEQ_printf(m, "rt_rq[%d]:\n", cpu);
 723 #endif
 724
 725 #define P(x) \
 726         SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
 727 #define PU(x) \
 728         SEQ_printf(m, "  .%-30s: %lu\n", #x, (unsigned long)(rt_rq->x))
 729 #define PN(x) \
 730         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
 731
 732         PU(rt_nr_running);
 733         P(rt_throttled);
 734         PN(rt_time);
 735         PN(rt_runtime);
 736
 737 #undef PN
 738 #undef PU
 739 #undef P
 740 }
 741
 742 void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq)
 743 {
 744         struct dl_bw *dl_bw;
 745
 746         SEQ_printf(m, "\n");
 747         SEQ_printf(m, "dl_rq[%d]:\n", cpu);
 748
 749 #define PU(x) \
 750         SEQ_printf(m, "  .%-30s: %lu\n", #x, (unsigned long)(dl_rq->x))
 751
 752         PU(dl_nr_running);
 753 #ifdef CONFIG_SMP
 754         dl_bw = &cpu_rq(cpu)->rd->dl_bw;
 755 #else
 756         dl_bw = &dl_rq->dl_bw;
 757 #endif
 758         SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->bw", dl_bw->bw);
 759         SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->total_bw", dl_bw->total_bw);
 760
 761 #undef PU
 762 }
 763
 764 static void print_cpu(struct seq_file *m, int cpu)
 765 {
 766         struct rq *rq = cpu_rq(cpu);
 767
 768 #ifdef CONFIG_X86
 769         {
 770                 unsigned int freq = cpu_khz ? : 1;
 771
 772                 SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
 773                            cpu, freq / 1000, (freq % 1000));
 774         }
 775 #else
 776         SEQ_printf(m, "cpu#%d\n", cpu);
 777 #endif
 778
 779 #define P(x)                                                            \
 780 do {                                                                    \
 781         if (sizeof(rq->x) == 4)                                         \
 782                 SEQ_printf(m, "  .%-30s: %d\n", #x, (int)(rq->x));      \
 783         else                                                            \
 784                 SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rq->x));\
 785 } while (0)
 786
 787 #define PN(x) \
 788         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
 789
 790         P(nr_running);
 791         P(nr_switches);
 792         P(nr_uninterruptible);
 793         PN(next_balance);
 794         SEQ_printf(m, "  .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr)));
 795         PN(clock);
 796         PN(clock_task);
 797 #undef P
 798 #undef PN
 799
 800 #ifdef CONFIG_SMP
 801 #define P64(n) SEQ_printf(m, "  .%-30s: %Ld\n", #n, rq->n);
 802         P64(avg_idle);
 803         P64(max_idle_balance_cost);
 804 #undef P64
 805 #endif
 806
 807 #define P(n) SEQ_printf(m, "  .%-30s: %d\n", #n, schedstat_val(rq->n));
 808         if (schedstat_enabled()) {
 809                 P(yld_count);
 810                 P(sched_count);
 811                 P(sched_goidle);
 812                 P(ttwu_count);
 813                 P(ttwu_local);
 814         }
 815 #undef P
 816
 817         print_cfs_stats(m, cpu);
 818         print_rt_stats(m, cpu);
 819         print_dl_stats(m, cpu);
 820
 821         print_rq(m, rq, cpu);
 822         SEQ_printf(m, "\n");
 823 }
 824
 825 static const char *sched_tunable_scaling_names[] = {
 826         "none",
 827         "logarithmic",
 828         "linear"
 829 };
 830
 831 static void sched_debug_header(struct seq_file *m)
 832 {
 833         u64 ktime, sched_clk, cpu_clk;
 834         unsigned long flags;
 835
 836         local_irq_save(flags);
 837         ktime = ktime_to_ns(ktime_get());
 838         sched_clk = sched_clock();
 839         cpu_clk = local_clock();
 840         local_irq_restore(flags);
 841
 842         SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n",
 843                 init_utsname()->release,
 844                 (int)strcspn(init_utsname()->version, " "),
 845                 init_utsname()->version);
 846
 847 #define P(x) \
 848         SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
 849 #define PN(x) \
 850         SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
 851         PN(ktime);
 852         PN(sched_clk);
 853         PN(cpu_clk);
 854         P(jiffies);
 855 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
 856         P(sched_clock_stable());
 857 #endif
 858 #undef PN
 859 #undef P
 860
 861         SEQ_printf(m, "\n");
 862         SEQ_printf(m, "sysctl_sched\n");
 863
 864 #define P(x) \
 865         SEQ_printf(m, "  .%-40s: %Ld\n", #x, (long long)(x))
 866 #define PN(x) \
 867         SEQ_printf(m, "  .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
 868         PN(sysctl_sched_base_slice);
 869         P(sysctl_sched_features);
 870 #undef PN
 871 #undef P
 872
 873         SEQ_printf(m, "  .%-40s: %d (%s)\n",
 874                 "sysctl_sched_tunable_scaling",
 875                 sysctl_sched_tunable_scaling,
 876                 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
 877         SEQ_printf(m, "\n");
 878 }
 879
 880 static int sched_debug_show(struct seq_file *m, void *v)
 881 {
 882         int cpu = (unsigned long)(v - 2);
 883
 884         if (cpu != -1)
 885                 print_cpu(m, cpu);
 886         else
 887                 sched_debug_header(m);
 888
 889         return 0;
 890 }
 891
 892 void sysrq_sched_debug_show(void)
 893 {
 894         int cpu;
 895
 896         sched_debug_header(NULL);
 897         for_each_online_cpu(cpu) {
 898                 /*
 899                  * Need to reset softlockup watchdogs on all CPUs, because
 900                  * another CPU might be blocked waiting for us to process
 901                  * an IPI or stop_machine.
 902                  */
 903                 touch_nmi_watchdog();
 904                 touch_all_softlockup_watchdogs();
 905                 print_cpu(NULL, cpu);
 906         }
 907 }
 908
 909 /*
 910  * This iterator needs some explanation.
 911  * It returns 1 for the header position.
 912  * This means 2 is CPU 0.
 913  * In a hotplugged system some CPUs, including CPU 0, may be missing so we have
 914  * to use cpumask_* to iterate over the CPUs.
 915  */
 916 static void *sched_debug_start(struct seq_file *file, loff_t *offset)
 917 {
 918         unsigned long n = *offset;
 919
 920         if (n == 0)
 921                 return (void *) 1;
 922
 923         n--;
 924
 925         if (n > 0)
 926                 n = cpumask_next(n - 1, cpu_online_mask);
 927         else
 928                 n = cpumask_first(cpu_online_mask);
 929
 930         *offset = n + 1;
 931
 932         if (n < nr_cpu_ids)
 933                 return (void *)(unsigned long)(n + 2);
 934
 935         return NULL;
 936 }
 937
 938 static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
 939 {
 940         (*offset)++;
 941         return sched_debug_start(file, offset);
 942 }
 943
 944 static void sched_debug_stop(struct seq_file *file, void *data)
 945 {
 946 }
 947
 948 static const struct seq_operations sched_debug_sops = {
 949         .start          = sched_debug_start,
 950         .next           = sched_debug_next,
 951         .stop           = sched_debug_stop,
 952         .show           = sched_debug_show,
 953 };
 954
 955 #define __PS(S, F) SEQ_printf(m, "%-45s:%21Ld\n", S, (long long)(F))
 956 #define __P(F) __PS(#F, F)
 957 #define   P(F) __PS(#F, p->F)
 958 #define   PM(F, M) __PS(#F, p->F & (M))
 959 #define __PSN(S, F) SEQ_printf(m, "%-45s:%14Ld.%06ld\n", S, SPLIT_NS((long long)(F)))
 960 #define __PN(F) __PSN(#F, F)
 961 #define   PN(F) __PSN(#F, p->F)
 962
 963
 964 #ifdef CONFIG_NUMA_BALANCING
 965 void print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
 966                 unsigned long tpf, unsigned long gsf, unsigned long gpf)
 967 {
 968         SEQ_printf(m, "numa_faults node=%d ", node);
 969         SEQ_printf(m, "task_private=%lu task_shared=%lu ", tpf, tsf);
 970         SEQ_printf(m, "group_private=%lu group_shared=%lu\n", gpf, gsf);
 971 }
 972 #endif
 973
 974
 975 static void sched_show_numa(struct task_struct *p, struct seq_file *m)
 976 {
 977 #ifdef CONFIG_NUMA_BALANCING
 978         if (p->mm)
 979                 P(mm->numa_scan_seq);
 980
 981         P(numa_pages_migrated);
 982         P(numa_preferred_nid);
 983         P(total_numa_faults);
 984         SEQ_printf(m, "current_node=%d, numa_group_id=%d\n",
 985                         task_node(p), task_numa_group_id(p));
 986         show_numa_stats(p, m);
 987 #endif
 988 }
 989
 990 void proc_sched_show_task(struct task_struct *p, struct pid_namespace *ns,
 991                                                   struct seq_file *m)
 992 {
 993         unsigned long nr_switches;
 994
 995         SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr_ns(p, ns),
 996                                                 get_nr_threads(p));
 997         SEQ_printf(m,
 998                 "---------------------------------------------------------"
 999                 "----------\n");
1000
1001 #define P_SCHEDSTAT(F)  __PS(#F, schedstat_val(p->stats.F))
1002 #define PN_SCHEDSTAT(F) __PSN(#F, schedstat_val(p->stats.F))
1003
1004         PN(se.exec_start);
1005         PN(se.vruntime);
1006         PN(se.sum_exec_runtime);
1007
1008         nr_switches = p->nvcsw + p->nivcsw;
1009
1010         P(se.nr_migrations);
1011
1012         if (schedstat_enabled()) {
1013                 u64 avg_atom, avg_per_cpu;
1014
1015                 PN_SCHEDSTAT(sum_sleep_runtime);
1016                 PN_SCHEDSTAT(sum_block_runtime);
1017                 PN_SCHEDSTAT(wait_start);
1018                 PN_SCHEDSTAT(sleep_start);
1019                 PN_SCHEDSTAT(block_start);
1020                 PN_SCHEDSTAT(sleep_max);
1021                 PN_SCHEDSTAT(block_max);
1022                 PN_SCHEDSTAT(exec_max);
1023                 PN_SCHEDSTAT(slice_max);
1024                 PN_SCHEDSTAT(wait_max);
1025                 PN_SCHEDSTAT(wait_sum);
1026                 P_SCHEDSTAT(wait_count);
1027                 PN_SCHEDSTAT(iowait_sum);
1028                 P_SCHEDSTAT(iowait_count);
1029                 P_SCHEDSTAT(nr_migrations_cold);
1030                 P_SCHEDSTAT(nr_failed_migrations_affine);
1031                 P_SCHEDSTAT(nr_failed_migrations_running);
1032                 P_SCHEDSTAT(nr_failed_migrations_hot);
1033                 P_SCHEDSTAT(nr_forced_migrations);
1034                 P_SCHEDSTAT(nr_wakeups);
1035                 P_SCHEDSTAT(nr_wakeups_sync);
1036                 P_SCHEDSTAT(nr_wakeups_migrate);
1037                 P_SCHEDSTAT(nr_wakeups_local);
1038                 P_SCHEDSTAT(nr_wakeups_remote);
1039                 P_SCHEDSTAT(nr_wakeups_affine);
1040                 P_SCHEDSTAT(nr_wakeups_affine_attempts);
1041                 P_SCHEDSTAT(nr_wakeups_passive);
1042                 P_SCHEDSTAT(nr_wakeups_idle);
1043
1044                 avg_atom = p->se.sum_exec_runtime;
1045                 if (nr_switches)
1046                         avg_atom = div64_ul(avg_atom, nr_switches);
1047                 else
1048                         avg_atom = -1LL;
1049
1050                 avg_per_cpu = p->se.sum_exec_runtime;
1051                 if (p->se.nr_migrations) {
1052                         avg_per_cpu = div64_u64(avg_per_cpu,
1053                                                 p->se.nr_migrations);
1054                 } else {
1055                         avg_per_cpu = -1LL;
1056                 }
1057
1058                 __PN(avg_atom);
1059                 __PN(avg_per_cpu);
1060
1061 #ifdef CONFIG_SCHED_CORE
1062                 PN_SCHEDSTAT(core_forceidle_sum);
1063 #endif
1064         }
1065
1066         __P(nr_switches);
1067         __PS("nr_voluntary_switches", p->nvcsw);
1068         __PS("nr_involuntary_switches", p->nivcsw);
1069
1070         P(se.load.weight);
1071 #ifdef CONFIG_SMP
1072         P(se.avg.load_sum);
1073         P(se.avg.runnable_sum);
1074         P(se.avg.util_sum);
1075         P(se.avg.load_avg);
1076         P(se.avg.runnable_avg);
1077         P(se.avg.util_avg);
1078         P(se.avg.last_update_time);
1079         PM(se.avg.util_est, ~UTIL_AVG_UNCHANGED);
1080 #endif
1081 #ifdef CONFIG_UCLAMP_TASK
1082         __PS("uclamp.min", p->uclamp_req[UCLAMP_MIN].value);
1083         __PS("uclamp.max", p->uclamp_req[UCLAMP_MAX].value);
1084         __PS("effective uclamp.min", uclamp_eff_value(p, UCLAMP_MIN));
1085         __PS("effective uclamp.max", uclamp_eff_value(p, UCLAMP_MAX));
1086 #endif
1087         P(policy);
1088         P(prio);
1089         if (task_has_dl_policy(p)) {
1090                 P(dl.runtime);
1091                 P(dl.deadline);
1092         }
1093 #undef PN_SCHEDSTAT
1094 #undef P_SCHEDSTAT
1095
1096         {
1097                 unsigned int this_cpu = raw_smp_processor_id();
1098                 u64 t0, t1;
1099
1100                 t0 = cpu_clock(this_cpu);
1101                 t1 = cpu_clock(this_cpu);
1102                 __PS("clock-delta", t1-t0);
1103         }
1104
1105         sched_show_numa(p, m);
1106 }
1107
1108 void proc_sched_set_task(struct task_struct *p)
1109 {
1110 #ifdef CONFIG_SCHEDSTATS
1111         memset(&p->stats, 0, sizeof(p->stats));
1112 #endif
1113 }
1114
1115 void resched_latency_warn(int cpu, u64 latency)
1116 {
1117         static DEFINE_RATELIMIT_STATE(latency_check_ratelimit, 60 * 60 * HZ, 1);
1118
1119         WARN(__ratelimit(&latency_check_ratelimit),
1120              "sched: CPU %d need_resched set for > %llu ns (%d ticks) "
1121              "without schedule\n",
1122              cpu, latency, cpu_rq(cpu)->ticks_without_resched);
1123 }