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