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 /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  */
  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 static const struct seq_operations sched_debug_sops;
 284
 285 static int sched_debug_open(struct inode *inode, struct file *filp)
 286 {
 287         return seq_open(filp, &sched_debug_sops);
 288 }
 289
 290 static 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
 297 static struct dentry *debugfs_sched;
 298
 299 static __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 #endif
 337
 338         debugfs_create_file("debug", 0444, debugfs_sched, NULL, &sched_debug_fops);
 339
 340         return 0;
 341 }
 342 late_initcall(sched_init_debug);
 343
 344 #ifdef CONFIG_SMP
 345
 346 static cpumask_var_t            sd_sysctl_cpus;
 347 static struct dentry            *sd_dentry;
 348
 349 static int sd_flags_show(struct seq_file *m, void *v)
 350 {
 351         unsigned long flags = *(unsigned int *)m->private;
 352         int idx;
 353
 354         for_each_set_bit(idx, &flags, __SD_FLAG_CNT) {
 355                 seq_puts(m, sd_flag_debug[idx].name);
 356                 seq_puts(m, " ");
 357         }
 358         seq_puts(m, "\n");
 359
 360         return 0;
 361 }
 362
 363 static int sd_flags_open(struct inode *inode, struct file *file)
 364 {
 365         return single_open(file, sd_flags_show, inode->i_private);
 366 }
 367
 368 static const struct file_operations sd_flags_fops = {
 369         .open           = sd_flags_open,
 370         .read           = seq_read,
 371         .llseek         = seq_lseek,
 372         .release        = single_release,
 373 };
 374
 375 static void register_sd(struct sched_domain *sd, struct dentry *parent)
 376 {
 377 #define SDM(type, mode, member) \
 378         debugfs_create_##type(#member, mode, parent, &sd->member)
 379
 380         SDM(ulong, 0644, min_interval);
 381         SDM(ulong, 0644, max_interval);
 382         SDM(u64,   0644, max_newidle_lb_cost);
 383         SDM(u32,   0644, busy_factor);
 384         SDM(u32,   0644, imbalance_pct);
 385         SDM(u32,   0644, cache_nice_tries);
 386         SDM(str,   0444, name);
 387
 388 #undef SDM
 389
 390         debugfs_create_file("flags", 0444, parent, &sd->flags, &sd_flags_fops);
 391 }
 392
 393 void update_sched_domain_debugfs(void)
 394 {
 395         int cpu, i;
 396
 397         /*
 398          * This can unfortunately be invoked before sched_debug_init() creates
 399          * the debug directory. Don't touch sd_sysctl_cpus until then.
 400          */
 401         if (!debugfs_sched)
 402                 return;
 403
 404         if (!cpumask_available(sd_sysctl_cpus)) {
 405                 if (!alloc_cpumask_var(&sd_sysctl_cpus, GFP_KERNEL))
 406                         return;
 407                 cpumask_copy(sd_sysctl_cpus, cpu_possible_mask);
 408         }
 409
 410         if (!sd_dentry)
 411                 sd_dentry = debugfs_create_dir("domains", debugfs_sched);
 412
 413         for_each_cpu(cpu, sd_sysctl_cpus) {
 414                 struct sched_domain *sd;
 415                 struct dentry *d_cpu;
 416                 char buf[32];
 417
 418                 snprintf(buf, sizeof(buf), "cpu%d", cpu);
 419                 debugfs_remove(debugfs_lookup(buf, sd_dentry));
 420                 d_cpu = debugfs_create_dir(buf, sd_dentry);
 421
 422                 i = 0;
 423                 for_each_domain(cpu, sd) {
 424                         struct dentry *d_sd;
 425
 426                         snprintf(buf, sizeof(buf), "domain%d", i);
 427                         d_sd = debugfs_create_dir(buf, d_cpu);
 428
 429                         register_sd(sd, d_sd);
 430                         i++;
 431                 }
 432
 433                 __cpumask_clear_cpu(cpu, sd_sysctl_cpus);
 434         }
 435 }
 436
 437 void dirty_sched_domain_sysctl(int cpu)
 438 {
 439         if (cpumask_available(sd_sysctl_cpus))
 440                 __cpumask_set_cpu(cpu, sd_sysctl_cpus);
 441 }
 442
 443 #endif /* CONFIG_SMP */
 444
 445 #ifdef CONFIG_FAIR_GROUP_SCHED
 446 static void print_cfs_group_stats(struct seq_file *m, int cpu, struct task_group *tg)
 447 {
 448         struct sched_entity *se = tg->se[cpu];
 449
 450 #define P(F)            SEQ_printf(m, "  .%-30s: %lld\n",       #F, (long long)F)
 451 #define P_SCHEDSTAT(F)  SEQ_printf(m, "  .%-30s: %lld\n",       \
 452                 #F, (long long)schedstat_val(stats->F))
 453 #define PN(F)           SEQ_printf(m, "  .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F))
 454 #define PN_SCHEDSTAT(F) SEQ_printf(m, "  .%-30s: %lld.%06ld\n", \
 455                 #F, SPLIT_NS((long long)schedstat_val(stats->F)))
 456
 457         if (!se)
 458                 return;
 459
 460         PN(se->exec_start);
 461         PN(se->vruntime);
 462         PN(se->sum_exec_runtime);
 463
 464         if (schedstat_enabled()) {
 465                 struct sched_statistics *stats;
 466                 stats = __schedstats_from_se(se);
 467
 468                 PN_SCHEDSTAT(wait_start);
 469                 PN_SCHEDSTAT(sleep_start);
 470                 PN_SCHEDSTAT(block_start);
 471                 PN_SCHEDSTAT(sleep_max);
 472                 PN_SCHEDSTAT(block_max);
 473                 PN_SCHEDSTAT(exec_max);
 474                 PN_SCHEDSTAT(slice_max);
 475                 PN_SCHEDSTAT(wait_max);
 476                 PN_SCHEDSTAT(wait_sum);
 477                 P_SCHEDSTAT(wait_count);
 478         }
 479
 480         P(se->load.weight);
 481 #ifdef CONFIG_SMP
 482         P(se->avg.load_avg);
 483         P(se->avg.util_avg);
 484         P(se->avg.runnable_avg);
 485 #endif
 486
 487 #undef PN_SCHEDSTAT
 488 #undef PN
 489 #undef P_SCHEDSTAT
 490 #undef P
 491 }
 492 #endif
 493
 494 #ifdef CONFIG_CGROUP_SCHED
 495 static DEFINE_SPINLOCK(sched_debug_lock);
 496 static char group_path[PATH_MAX];
 497
 498 static void task_group_path(struct task_group *tg, char *path, int plen)
 499 {
 500         if (autogroup_path(tg, path, plen))
 501                 return;
 502
 503         cgroup_path(tg->css.cgroup, path, plen);
 504 }
 505
 506 /*
 507  * Only 1 SEQ_printf_task_group_path() caller can use the full length
 508  * group_path[] for cgroup path. Other simultaneous callers will have
 509  * to use a shorter stack buffer. A "..." suffix is appended at the end
 510  * of the stack buffer so that it will show up in case the output length
 511  * matches the given buffer size to indicate possible path name truncation.
 512  */
 513 #define SEQ_printf_task_group_path(m, tg, fmt...)                       \
 514 {                                                                       \
 515         if (spin_trylock(&sched_debug_lock)) {                          \
 516                 task_group_path(tg, group_path, sizeof(group_path));    \
 517                 SEQ_printf(m, fmt, group_path);                         \
 518                 spin_unlock(&sched_debug_lock);                         \
 519         } else {                                                        \
 520                 char buf[128];                                          \
 521                 char *bufend = buf + sizeof(buf) - 3;                   \
 522                 task_group_path(tg, buf, bufend - buf);                 \
 523                 strcpy(bufend - 1, "...");                              \
 524                 SEQ_printf(m, fmt, buf);                                \
 525         }                                                               \
 526 }
 527 #endif
 528
 529 static void
 530 print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
 531 {
 532         if (task_current(rq, p))
 533                 SEQ_printf(m, ">R");
 534         else
 535                 SEQ_printf(m, " %c", task_state_to_char(p));
 536
 537         SEQ_printf(m, " %15s %5d %9Ld.%06ld %9Ld %5d ",
 538                 p->comm, task_pid_nr(p),
 539                 SPLIT_NS(p->se.vruntime),
 540                 (long long)(p->nvcsw + p->nivcsw),
 541                 p->prio);
 542
 543         SEQ_printf(m, "%9lld.%06ld %9lld.%06ld %9lld.%06ld %9lld.%06ld",
 544                 SPLIT_NS(schedstat_val_or_zero(p->stats.wait_sum)),
 545                 SPLIT_NS(p->se.sum_exec_runtime),
 546                 SPLIT_NS(schedstat_val_or_zero(p->stats.sum_sleep_runtime)),
 547                 SPLIT_NS(schedstat_val_or_zero(p->stats.sum_block_runtime)));
 548
 549 #ifdef CONFIG_NUMA_BALANCING
 550         SEQ_printf(m, " %d %d", task_node(p), task_numa_group_id(p));
 551 #endif
 552 #ifdef CONFIG_CGROUP_SCHED
 553         SEQ_printf_task_group_path(m, task_group(p), " %s")
 554 #endif
 555
 556         SEQ_printf(m, "\n");
 557 }
 558
 559 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
 560 {
 561         struct task_struct *g, *p;
 562
 563         SEQ_printf(m, "\n");
 564         SEQ_printf(m, "runnable tasks:\n");
 565         SEQ_printf(m, " S            task   PID         tree-key  switches  prio"
 566                    "     wait-time             sum-exec        sum-sleep\n");
 567         SEQ_printf(m, "-------------------------------------------------------"
 568                    "------------------------------------------------------\n");
 569
 570         rcu_read_lock();
 571         for_each_process_thread(g, p) {
 572                 if (task_cpu(p) != rq_cpu)
 573                         continue;
 574
 575                 print_task(m, rq, p);
 576         }
 577         rcu_read_unlock();
 578 }
 579
 580 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
 581 {
 582         s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
 583                 spread, rq0_min_vruntime, spread0;
 584         struct rq *rq = cpu_rq(cpu);
 585         struct sched_entity *last;
 586         unsigned long flags;
 587
 588 #ifdef CONFIG_FAIR_GROUP_SCHED
 589         SEQ_printf(m, "\n");
 590         SEQ_printf_task_group_path(m, cfs_rq->tg, "cfs_rq[%d]:%s\n", cpu);
 591 #else
 592         SEQ_printf(m, "\n");
 593         SEQ_printf(m, "cfs_rq[%d]:\n", cpu);
 594 #endif
 595         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "exec_clock",
 596                         SPLIT_NS(cfs_rq->exec_clock));
 597
 598         raw_spin_rq_lock_irqsave(rq, flags);
 599         if (rb_first_cached(&cfs_rq->tasks_timeline))
 600                 MIN_vruntime = (__pick_first_entity(cfs_rq))->vruntime;
 601         last = __pick_last_entity(cfs_rq);
 602         if (last)
 603                 max_vruntime = last->vruntime;
 604         min_vruntime = cfs_rq->min_vruntime;
 605         rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
 606         raw_spin_rq_unlock_irqrestore(rq, flags);
 607         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "MIN_vruntime",
 608                         SPLIT_NS(MIN_vruntime));
 609         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "min_vruntime",
 610                         SPLIT_NS(min_vruntime));
 611         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "max_vruntime",
 612                         SPLIT_NS(max_vruntime));
 613         spread = max_vruntime - MIN_vruntime;
 614         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread",
 615                         SPLIT_NS(spread));
 616         spread0 = min_vruntime - rq0_min_vruntime;
 617         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread0",
 618                         SPLIT_NS(spread0));
 619         SEQ_printf(m, "  .%-30s: %d\n", "nr_spread_over",
 620                         cfs_rq->nr_spread_over);
 621         SEQ_printf(m, "  .%-30s: %d\n", "nr_running", cfs_rq->nr_running);
 622         SEQ_printf(m, "  .%-30s: %d\n", "h_nr_running", cfs_rq->h_nr_running);
 623         SEQ_printf(m, "  .%-30s: %d\n", "idle_nr_running",
 624                         cfs_rq->idle_nr_running);
 625         SEQ_printf(m, "  .%-30s: %d\n", "idle_h_nr_running",
 626                         cfs_rq->idle_h_nr_running);
 627         SEQ_printf(m, "  .%-30s: %ld\n", "load", cfs_rq->load.weight);
 628 #ifdef CONFIG_SMP
 629         SEQ_printf(m, "  .%-30s: %lu\n", "load_avg",
 630                         cfs_rq->avg.load_avg);
 631         SEQ_printf(m, "  .%-30s: %lu\n", "runnable_avg",
 632                         cfs_rq->avg.runnable_avg);
 633         SEQ_printf(m, "  .%-30s: %lu\n", "util_avg",
 634                         cfs_rq->avg.util_avg);
 635         SEQ_printf(m, "  .%-30s: %u\n", "util_est_enqueued",
 636                         cfs_rq->avg.util_est.enqueued);
 637         SEQ_printf(m, "  .%-30s: %ld\n", "removed.load_avg",
 638                         cfs_rq->removed.load_avg);
 639         SEQ_printf(m, "  .%-30s: %ld\n", "removed.util_avg",
 640                         cfs_rq->removed.util_avg);
 641         SEQ_printf(m, "  .%-30s: %ld\n", "removed.runnable_avg",
 642                         cfs_rq->removed.runnable_avg);
 643 #ifdef CONFIG_FAIR_GROUP_SCHED
 644         SEQ_printf(m, "  .%-30s: %lu\n", "tg_load_avg_contrib",
 645                         cfs_rq->tg_load_avg_contrib);
 646         SEQ_printf(m, "  .%-30s: %ld\n", "tg_load_avg",
 647                         atomic_long_read(&cfs_rq->tg->load_avg));
 648 #endif
 649 #endif
 650 #ifdef CONFIG_CFS_BANDWIDTH
 651         SEQ_printf(m, "  .%-30s: %d\n", "throttled",
 652                         cfs_rq->throttled);
 653         SEQ_printf(m, "  .%-30s: %d\n", "throttle_count",
 654                         cfs_rq->throttle_count);
 655 #endif
 656
 657 #ifdef CONFIG_FAIR_GROUP_SCHED
 658         print_cfs_group_stats(m, cpu, cfs_rq->tg);
 659 #endif
 660 }
 661
 662 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
 663 {
 664 #ifdef CONFIG_RT_GROUP_SCHED
 665         SEQ_printf(m, "\n");
 666         SEQ_printf_task_group_path(m, rt_rq->tg, "rt_rq[%d]:%s\n", cpu);
 667 #else
 668         SEQ_printf(m, "\n");
 669         SEQ_printf(m, "rt_rq[%d]:\n", cpu);
 670 #endif
 671
 672 #define P(x) \
 673         SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
 674 #define PU(x) \
 675         SEQ_printf(m, "  .%-30s: %lu\n", #x, (unsigned long)(rt_rq->x))
 676 #define PN(x) \
 677         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
 678
 679         PU(rt_nr_running);
 680 #ifdef CONFIG_SMP
 681         PU(rt_nr_migratory);
 682 #endif
 683         P(rt_throttled);
 684         PN(rt_time);
 685         PN(rt_runtime);
 686
 687 #undef PN
 688 #undef PU
 689 #undef P
 690 }
 691
 692 void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq)
 693 {
 694         struct dl_bw *dl_bw;
 695
 696         SEQ_printf(m, "\n");
 697         SEQ_printf(m, "dl_rq[%d]:\n", cpu);
 698
 699 #define PU(x) \
 700         SEQ_printf(m, "  .%-30s: %lu\n", #x, (unsigned long)(dl_rq->x))
 701
 702         PU(dl_nr_running);
 703 #ifdef CONFIG_SMP
 704         PU(dl_nr_migratory);
 705         dl_bw = &cpu_rq(cpu)->rd->dl_bw;
 706 #else
 707         dl_bw = &dl_rq->dl_bw;
 708 #endif
 709         SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->bw", dl_bw->bw);
 710         SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->total_bw", dl_bw->total_bw);
 711
 712 #undef PU
 713 }
 714
 715 static void print_cpu(struct seq_file *m, int cpu)
 716 {
 717         struct rq *rq = cpu_rq(cpu);
 718
 719 #ifdef CONFIG_X86
 720         {
 721                 unsigned int freq = cpu_khz ? : 1;
 722
 723                 SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
 724                            cpu, freq / 1000, (freq % 1000));
 725         }
 726 #else
 727         SEQ_printf(m, "cpu#%d\n", cpu);
 728 #endif
 729
 730 #define P(x)                                                            \
 731 do {                                                                    \
 732         if (sizeof(rq->x) == 4)                                         \
 733                 SEQ_printf(m, "  .%-30s: %ld\n", #x, (long)(rq->x));    \
 734         else                                                            \
 735                 SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rq->x));\
 736 } while (0)
 737
 738 #define PN(x) \
 739         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
 740
 741         P(nr_running);
 742         P(nr_switches);
 743         P(nr_uninterruptible);
 744         PN(next_balance);
 745         SEQ_printf(m, "  .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr)));
 746         PN(clock);
 747         PN(clock_task);
 748 #undef P
 749 #undef PN
 750
 751 #ifdef CONFIG_SMP
 752 #define P64(n) SEQ_printf(m, "  .%-30s: %Ld\n", #n, rq->n);
 753         P64(avg_idle);
 754         P64(max_idle_balance_cost);
 755 #undef P64
 756 #endif
 757
 758 #define P(n) SEQ_printf(m, "  .%-30s: %d\n", #n, schedstat_val(rq->n));
 759         if (schedstat_enabled()) {
 760                 P(yld_count);
 761                 P(sched_count);
 762                 P(sched_goidle);
 763                 P(ttwu_count);
 764                 P(ttwu_local);
 765         }
 766 #undef P
 767
 768         print_cfs_stats(m, cpu);
 769         print_rt_stats(m, cpu);
 770         print_dl_stats(m, cpu);
 771
 772         print_rq(m, rq, cpu);
 773         SEQ_printf(m, "\n");
 774 }
 775
 776 static const char *sched_tunable_scaling_names[] = {
 777         "none",
 778         "logarithmic",
 779         "linear"
 780 };
 781
 782 static void sched_debug_header(struct seq_file *m)
 783 {
 784         u64 ktime, sched_clk, cpu_clk;
 785         unsigned long flags;
 786
 787         local_irq_save(flags);
 788         ktime = ktime_to_ns(ktime_get());
 789         sched_clk = sched_clock();
 790         cpu_clk = local_clock();
 791         local_irq_restore(flags);
 792
 793         SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n",
 794                 init_utsname()->release,
 795                 (int)strcspn(init_utsname()->version, " "),
 796                 init_utsname()->version);
 797
 798 #define P(x) \
 799         SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
 800 #define PN(x) \
 801         SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
 802         PN(ktime);
 803         PN(sched_clk);
 804         PN(cpu_clk);
 805         P(jiffies);
 806 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
 807         P(sched_clock_stable());
 808 #endif
 809 #undef PN
 810 #undef P
 811
 812         SEQ_printf(m, "\n");
 813         SEQ_printf(m, "sysctl_sched\n");
 814
 815 #define P(x) \
 816         SEQ_printf(m, "  .%-40s: %Ld\n", #x, (long long)(x))
 817 #define PN(x) \
 818         SEQ_printf(m, "  .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
 819         PN(sysctl_sched_latency);
 820         PN(sysctl_sched_min_granularity);
 821         PN(sysctl_sched_idle_min_granularity);
 822         PN(sysctl_sched_wakeup_granularity);
 823         P(sysctl_sched_child_runs_first);
 824         P(sysctl_sched_features);
 825 #undef PN
 826 #undef P
 827
 828         SEQ_printf(m, "  .%-40s: %d (%s)\n",
 829                 "sysctl_sched_tunable_scaling",
 830                 sysctl_sched_tunable_scaling,
 831                 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
 832         SEQ_printf(m, "\n");
 833 }
 834
 835 static int sched_debug_show(struct seq_file *m, void *v)
 836 {
 837         int cpu = (unsigned long)(v - 2);
 838
 839         if (cpu != -1)
 840                 print_cpu(m, cpu);
 841         else
 842                 sched_debug_header(m);
 843
 844         return 0;
 845 }
 846
 847 void sysrq_sched_debug_show(void)
 848 {
 849         int cpu;
 850
 851         sched_debug_header(NULL);
 852         for_each_online_cpu(cpu) {
 853                 /*
 854                  * Need to reset softlockup watchdogs on all CPUs, because
 855                  * another CPU might be blocked waiting for us to process
 856                  * an IPI or stop_machine.
 857                  */
 858                 touch_nmi_watchdog();
 859                 touch_all_softlockup_watchdogs();
 860                 print_cpu(NULL, cpu);
 861         }
 862 }
 863
 864 /*
 865  * This iterator needs some explanation.
 866  * It returns 1 for the header position.
 867  * This means 2 is CPU 0.
 868  * In a hotplugged system some CPUs, including CPU 0, may be missing so we have
 869  * to use cpumask_* to iterate over the CPUs.
 870  */
 871 static void *sched_debug_start(struct seq_file *file, loff_t *offset)
 872 {
 873         unsigned long n = *offset;
 874
 875         if (n == 0)
 876                 return (void *) 1;
 877
 878         n--;
 879
 880         if (n > 0)
 881                 n = cpumask_next(n - 1, cpu_online_mask);
 882         else
 883                 n = cpumask_first(cpu_online_mask);
 884
 885         *offset = n + 1;
 886
 887         if (n < nr_cpu_ids)
 888                 return (void *)(unsigned long)(n + 2);
 889
 890         return NULL;
 891 }
 892
 893 static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
 894 {
 895         (*offset)++;
 896         return sched_debug_start(file, offset);
 897 }
 898
 899 static void sched_debug_stop(struct seq_file *file, void *data)
 900 {
 901 }
 902
 903 static const struct seq_operations sched_debug_sops = {
 904         .start          = sched_debug_start,
 905         .next           = sched_debug_next,
 906         .stop           = sched_debug_stop,
 907         .show           = sched_debug_show,
 908 };
 909
 910 #define __PS(S, F) SEQ_printf(m, "%-45s:%21Ld\n", S, (long long)(F))
 911 #define __P(F) __PS(#F, F)
 912 #define   P(F) __PS(#F, p->F)
 913 #define   PM(F, M) __PS(#F, p->F & (M))
 914 #define __PSN(S, F) SEQ_printf(m, "%-45s:%14Ld.%06ld\n", S, SPLIT_NS((long long)(F)))
 915 #define __PN(F) __PSN(#F, F)
 916 #define   PN(F) __PSN(#F, p->F)
 917
 918
 919 #ifdef CONFIG_NUMA_BALANCING
 920 void print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
 921                 unsigned long tpf, unsigned long gsf, unsigned long gpf)
 922 {
 923         SEQ_printf(m, "numa_faults node=%d ", node);
 924         SEQ_printf(m, "task_private=%lu task_shared=%lu ", tpf, tsf);
 925         SEQ_printf(m, "group_private=%lu group_shared=%lu\n", gpf, gsf);
 926 }
 927 #endif
 928
 929
 930 static void sched_show_numa(struct task_struct *p, struct seq_file *m)
 931 {
 932 #ifdef CONFIG_NUMA_BALANCING
 933         if (p->mm)
 934                 P(mm->numa_scan_seq);
 935
 936         P(numa_pages_migrated);
 937         P(numa_preferred_nid);
 938         P(total_numa_faults);
 939         SEQ_printf(m, "current_node=%d, numa_group_id=%d\n",
 940                         task_node(p), task_numa_group_id(p));
 941         show_numa_stats(p, m);
 942 #endif
 943 }
 944
 945 void proc_sched_show_task(struct task_struct *p, struct pid_namespace *ns,
 946                                                   struct seq_file *m)
 947 {
 948         unsigned long nr_switches;
 949
 950         SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr_ns(p, ns),
 951                                                 get_nr_threads(p));
 952         SEQ_printf(m,
 953                 "---------------------------------------------------------"
 954                 "----------\n");
 955
 956 #define P_SCHEDSTAT(F)  __PS(#F, schedstat_val(p->stats.F))
 957 #define PN_SCHEDSTAT(F) __PSN(#F, schedstat_val(p->stats.F))
 958
 959         PN(se.exec_start);
 960         PN(se.vruntime);
 961         PN(se.sum_exec_runtime);
 962
 963         nr_switches = p->nvcsw + p->nivcsw;
 964
 965         P(se.nr_migrations);
 966
 967         if (schedstat_enabled()) {
 968                 u64 avg_atom, avg_per_cpu;
 969
 970                 PN_SCHEDSTAT(sum_sleep_runtime);
 971                 PN_SCHEDSTAT(sum_block_runtime);
 972                 PN_SCHEDSTAT(wait_start);
 973                 PN_SCHEDSTAT(sleep_start);
 974                 PN_SCHEDSTAT(block_start);
 975                 PN_SCHEDSTAT(sleep_max);
 976                 PN_SCHEDSTAT(block_max);
 977                 PN_SCHEDSTAT(exec_max);
 978                 PN_SCHEDSTAT(slice_max);
 979                 PN_SCHEDSTAT(wait_max);
 980                 PN_SCHEDSTAT(wait_sum);
 981                 P_SCHEDSTAT(wait_count);
 982                 PN_SCHEDSTAT(iowait_sum);
 983                 P_SCHEDSTAT(iowait_count);
 984                 P_SCHEDSTAT(nr_migrations_cold);
 985                 P_SCHEDSTAT(nr_failed_migrations_affine);
 986                 P_SCHEDSTAT(nr_failed_migrations_running);
 987                 P_SCHEDSTAT(nr_failed_migrations_hot);
 988                 P_SCHEDSTAT(nr_forced_migrations);
 989                 P_SCHEDSTAT(nr_wakeups);
 990                 P_SCHEDSTAT(nr_wakeups_sync);
 991                 P_SCHEDSTAT(nr_wakeups_migrate);
 992                 P_SCHEDSTAT(nr_wakeups_local);
 993                 P_SCHEDSTAT(nr_wakeups_remote);
 994                 P_SCHEDSTAT(nr_wakeups_affine);
 995                 P_SCHEDSTAT(nr_wakeups_affine_attempts);
 996                 P_SCHEDSTAT(nr_wakeups_passive);
 997                 P_SCHEDSTAT(nr_wakeups_idle);
 998
 999                 avg_atom = p->se.sum_exec_runtime;
1000                 if (nr_switches)
1001                         avg_atom = div64_ul(avg_atom, nr_switches);
1002                 else
1003                         avg_atom = -1LL;
1004
1005                 avg_per_cpu = p->se.sum_exec_runtime;
1006                 if (p->se.nr_migrations) {
1007                         avg_per_cpu = div64_u64(avg_per_cpu,
1008                                                 p->se.nr_migrations);
1009                 } else {
1010                         avg_per_cpu = -1LL;
1011                 }
1012
1013                 __PN(avg_atom);
1014                 __PN(avg_per_cpu);
1015
1016 #ifdef CONFIG_SCHED_CORE
1017                 PN_SCHEDSTAT(core_forceidle_sum);
1018 #endif
1019         }
1020
1021         __P(nr_switches);
1022         __PS("nr_voluntary_switches", p->nvcsw);
1023         __PS("nr_involuntary_switches", p->nivcsw);
1024
1025         P(se.load.weight);
1026 #ifdef CONFIG_SMP
1027         P(se.avg.load_sum);
1028         P(se.avg.runnable_sum);
1029         P(se.avg.util_sum);
1030         P(se.avg.load_avg);
1031         P(se.avg.runnable_avg);
1032         P(se.avg.util_avg);
1033         P(se.avg.last_update_time);
1034         P(se.avg.util_est.ewma);
1035         PM(se.avg.util_est.enqueued, ~UTIL_AVG_UNCHANGED);
1036 #endif
1037 #ifdef CONFIG_UCLAMP_TASK
1038         __PS("uclamp.min", p->uclamp_req[UCLAMP_MIN].value);
1039         __PS("uclamp.max", p->uclamp_req[UCLAMP_MAX].value);
1040         __PS("effective uclamp.min", uclamp_eff_value(p, UCLAMP_MIN));
1041         __PS("effective uclamp.max", uclamp_eff_value(p, UCLAMP_MAX));
1042 #endif
1043         P(policy);
1044         P(prio);
1045         if (task_has_dl_policy(p)) {
1046                 P(dl.runtime);
1047                 P(dl.deadline);
1048         }
1049 #undef PN_SCHEDSTAT
1050 #undef P_SCHEDSTAT
1051
1052         {
1053                 unsigned int this_cpu = raw_smp_processor_id();
1054                 u64 t0, t1;
1055
1056                 t0 = cpu_clock(this_cpu);
1057                 t1 = cpu_clock(this_cpu);
1058                 __PS("clock-delta", t1-t0);
1059         }
1060
1061         sched_show_numa(p, m);
1062 }
1063
1064 void proc_sched_set_task(struct task_struct *p)
1065 {
1066 #ifdef CONFIG_SCHEDSTATS
1067         memset(&p->stats, 0, sizeof(p->stats));
1068 #endif
1069 }
1070
1071 void resched_latency_warn(int cpu, u64 latency)
1072 {
1073         static DEFINE_RATELIMIT_STATE(latency_check_ratelimit, 60 * 60 * HZ, 1);
1074
1075         WARN(__ratelimit(&latency_check_ratelimit),
1076              "sched: CPU %d need_resched set for > %llu ns (%d ticks) "
1077              "without schedule\n",
1078              cpu, latency, cpu_rq(cpu)->ticks_without_resched);
1079 }