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 struct dentry *debugfs_sched;
 281
 282 static __init int sched_init_debug(void)
 283 {
 284         struct dentry __maybe_unused *numa;
 285
 286         debugfs_sched = debugfs_create_dir("sched", NULL);
 287
 288         debugfs_create_file("features", 0644, debugfs_sched, NULL, &sched_feat_fops);
 289         debugfs_create_bool("debug_enabled", 0644, debugfs_sched, &sched_debug_enabled);
 290 #ifdef CONFIG_PREEMPT_DYNAMIC
 291         debugfs_create_file("preempt", 0644, debugfs_sched, NULL, &sched_dynamic_fops);
 292 #endif
 293
 294         debugfs_create_u32("latency_ns", 0644, debugfs_sched, &sysctl_sched_latency);
 295         debugfs_create_u32("min_granularity_ns", 0644, debugfs_sched, &sysctl_sched_min_granularity);
 296         debugfs_create_u32("wakeup_granularity_ns", 0644, debugfs_sched, &sysctl_sched_wakeup_granularity);
 297
 298 #ifdef CONFIG_SMP
 299         debugfs_create_file("tunable_scaling", 0644, debugfs_sched, NULL, &sched_scaling_fops);
 300         debugfs_create_u32("migration_cost_ns", 0644, debugfs_sched, &sysctl_sched_migration_cost);
 301         debugfs_create_u32("nr_migrate", 0644, debugfs_sched, &sysctl_sched_nr_migrate);
 302
 303         mutex_lock(&sched_domains_mutex);
 304         update_sched_domain_debugfs();
 305         mutex_unlock(&sched_domains_mutex);
 306 #endif
 307
 308 #ifdef CONFIG_NUMA_BALANCING
 309         numa = debugfs_create_dir("numa_balancing", debugfs_sched);
 310
 311         debugfs_create_u32("scan_delay_ms", 0644, numa, &sysctl_numa_balancing_scan_delay);
 312         debugfs_create_u32("scan_period_min_ms", 0644, numa, &sysctl_numa_balancing_scan_period_min);
 313         debugfs_create_u32("scan_period_max_ms", 0644, numa, &sysctl_numa_balancing_scan_period_max);
 314         debugfs_create_u32("scan_size_mb", 0644, numa, &sysctl_numa_balancing_scan_size);
 315 #endif
 316
 317         return 0;
 318 }
 319 late_initcall(sched_init_debug);
 320
 321 #ifdef CONFIG_SMP
 322
 323 static cpumask_var_t            sd_sysctl_cpus;
 324 static struct dentry            *sd_dentry;
 325
 326 static int sd_flags_show(struct seq_file *m, void *v)
 327 {
 328         unsigned long flags = *(unsigned int *)m->private;
 329         int idx;
 330
 331         for_each_set_bit(idx, &flags, __SD_FLAG_CNT) {
 332                 seq_puts(m, sd_flag_debug[idx].name);
 333                 seq_puts(m, " ");
 334         }
 335         seq_puts(m, "\n");
 336
 337         return 0;
 338 }
 339
 340 static int sd_flags_open(struct inode *inode, struct file *file)
 341 {
 342         return single_open(file, sd_flags_show, inode->i_private);
 343 }
 344
 345 static const struct file_operations sd_flags_fops = {
 346         .open           = sd_flags_open,
 347         .read           = seq_read,
 348         .llseek         = seq_lseek,
 349         .release        = single_release,
 350 };
 351
 352 static void register_sd(struct sched_domain *sd, struct dentry *parent)
 353 {
 354 #define SDM(type, mode, member) \
 355         debugfs_create_##type(#member, mode, parent, &sd->member)
 356
 357         SDM(ulong, 0644, min_interval);
 358         SDM(ulong, 0644, max_interval);
 359         SDM(u64,   0644, max_newidle_lb_cost);
 360         SDM(u32,   0644, busy_factor);
 361         SDM(u32,   0644, imbalance_pct);
 362         SDM(u32,   0644, cache_nice_tries);
 363         SDM(str,   0444, name);
 364
 365 #undef SDM
 366
 367         debugfs_create_file("flags", 0444, parent, &sd->flags, &sd_flags_fops);
 368 }
 369
 370 void update_sched_domain_debugfs(void)
 371 {
 372         int cpu, i;
 373
 374         if (!cpumask_available(sd_sysctl_cpus)) {
 375                 if (!alloc_cpumask_var(&sd_sysctl_cpus, GFP_KERNEL))
 376                         return;
 377                 cpumask_copy(sd_sysctl_cpus, cpu_possible_mask);
 378         }
 379
 380         if (!sd_dentry)
 381                 sd_dentry = debugfs_create_dir("domains", debugfs_sched);
 382
 383         for_each_cpu(cpu, sd_sysctl_cpus) {
 384                 struct sched_domain *sd;
 385                 struct dentry *d_cpu;
 386                 char buf[32];
 387
 388                 snprintf(buf, sizeof(buf), "cpu%d", cpu);
 389                 debugfs_remove(debugfs_lookup(buf, sd_dentry));
 390                 d_cpu = debugfs_create_dir(buf, sd_dentry);
 391
 392                 i = 0;
 393                 for_each_domain(cpu, sd) {
 394                         struct dentry *d_sd;
 395
 396                         snprintf(buf, sizeof(buf), "domain%d", i);
 397                         d_sd = debugfs_create_dir(buf, d_cpu);
 398
 399                         register_sd(sd, d_sd);
 400                         i++;
 401                 }
 402
 403                 __cpumask_clear_cpu(cpu, sd_sysctl_cpus);
 404         }
 405 }
 406
 407 void dirty_sched_domain_sysctl(int cpu)
 408 {
 409         if (cpumask_available(sd_sysctl_cpus))
 410                 __cpumask_set_cpu(cpu, sd_sysctl_cpus);
 411 }
 412
 413 #endif /* CONFIG_SMP */
 414
 415 #ifdef CONFIG_FAIR_GROUP_SCHED
 416 static void print_cfs_group_stats(struct seq_file *m, int cpu, struct task_group *tg)
 417 {
 418         struct sched_entity *se = tg->se[cpu];
 419
 420 #define P(F)            SEQ_printf(m, "  .%-30s: %lld\n",       #F, (long long)F)
 421 #define P_SCHEDSTAT(F)  SEQ_printf(m, "  .%-30s: %lld\n",       #F, (long long)schedstat_val(F))
 422 #define PN(F)           SEQ_printf(m, "  .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F))
 423 #define PN_SCHEDSTAT(F) SEQ_printf(m, "  .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)schedstat_val(F)))
 424
 425         if (!se)
 426                 return;
 427
 428         PN(se->exec_start);
 429         PN(se->vruntime);
 430         PN(se->sum_exec_runtime);
 431
 432         if (schedstat_enabled()) {
 433                 PN_SCHEDSTAT(se->statistics.wait_start);
 434                 PN_SCHEDSTAT(se->statistics.sleep_start);
 435                 PN_SCHEDSTAT(se->statistics.block_start);
 436                 PN_SCHEDSTAT(se->statistics.sleep_max);
 437                 PN_SCHEDSTAT(se->statistics.block_max);
 438                 PN_SCHEDSTAT(se->statistics.exec_max);
 439                 PN_SCHEDSTAT(se->statistics.slice_max);
 440                 PN_SCHEDSTAT(se->statistics.wait_max);
 441                 PN_SCHEDSTAT(se->statistics.wait_sum);
 442                 P_SCHEDSTAT(se->statistics.wait_count);
 443         }
 444
 445         P(se->load.weight);
 446 #ifdef CONFIG_SMP
 447         P(se->avg.load_avg);
 448         P(se->avg.util_avg);
 449         P(se->avg.runnable_avg);
 450 #endif
 451
 452 #undef PN_SCHEDSTAT
 453 #undef PN
 454 #undef P_SCHEDSTAT
 455 #undef P
 456 }
 457 #endif
 458
 459 #ifdef CONFIG_CGROUP_SCHED
 460 static char group_path[PATH_MAX];
 461
 462 static char *task_group_path(struct task_group *tg)
 463 {
 464         if (autogroup_path(tg, group_path, PATH_MAX))
 465                 return group_path;
 466
 467         cgroup_path(tg->css.cgroup, group_path, PATH_MAX);
 468
 469         return group_path;
 470 }
 471 #endif
 472
 473 static void
 474 print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
 475 {
 476         if (task_current(rq, p))
 477                 SEQ_printf(m, ">R");
 478         else
 479                 SEQ_printf(m, " %c", task_state_to_char(p));
 480
 481         SEQ_printf(m, " %15s %5d %9Ld.%06ld %9Ld %5d ",
 482                 p->comm, task_pid_nr(p),
 483                 SPLIT_NS(p->se.vruntime),
 484                 (long long)(p->nvcsw + p->nivcsw),
 485                 p->prio);
 486
 487         SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
 488                 SPLIT_NS(schedstat_val_or_zero(p->se.statistics.wait_sum)),
 489                 SPLIT_NS(p->se.sum_exec_runtime),
 490                 SPLIT_NS(schedstat_val_or_zero(p->se.statistics.sum_sleep_runtime)));
 491
 492 #ifdef CONFIG_NUMA_BALANCING
 493         SEQ_printf(m, " %d %d", task_node(p), task_numa_group_id(p));
 494 #endif
 495 #ifdef CONFIG_CGROUP_SCHED
 496         SEQ_printf(m, " %s", task_group_path(task_group(p)));
 497 #endif
 498
 499         SEQ_printf(m, "\n");
 500 }
 501
 502 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
 503 {
 504         struct task_struct *g, *p;
 505
 506         SEQ_printf(m, "\n");
 507         SEQ_printf(m, "runnable tasks:\n");
 508         SEQ_printf(m, " S            task   PID         tree-key  switches  prio"
 509                    "     wait-time             sum-exec        sum-sleep\n");
 510         SEQ_printf(m, "-------------------------------------------------------"
 511                    "------------------------------------------------------\n");
 512
 513         rcu_read_lock();
 514         for_each_process_thread(g, p) {
 515                 if (task_cpu(p) != rq_cpu)
 516                         continue;
 517
 518                 print_task(m, rq, p);
 519         }
 520         rcu_read_unlock();
 521 }
 522
 523 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
 524 {
 525         s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
 526                 spread, rq0_min_vruntime, spread0;
 527         struct rq *rq = cpu_rq(cpu);
 528         struct sched_entity *last;
 529         unsigned long flags;
 530
 531 #ifdef CONFIG_FAIR_GROUP_SCHED
 532         SEQ_printf(m, "\n");
 533         SEQ_printf(m, "cfs_rq[%d]:%s\n", cpu, task_group_path(cfs_rq->tg));
 534 #else
 535         SEQ_printf(m, "\n");
 536         SEQ_printf(m, "cfs_rq[%d]:\n", cpu);
 537 #endif
 538         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "exec_clock",
 539                         SPLIT_NS(cfs_rq->exec_clock));
 540
 541         raw_spin_lock_irqsave(&rq->lock, flags);
 542         if (rb_first_cached(&cfs_rq->tasks_timeline))
 543                 MIN_vruntime = (__pick_first_entity(cfs_rq))->vruntime;
 544         last = __pick_last_entity(cfs_rq);
 545         if (last)
 546                 max_vruntime = last->vruntime;
 547         min_vruntime = cfs_rq->min_vruntime;
 548         rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
 549         raw_spin_unlock_irqrestore(&rq->lock, flags);
 550         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "MIN_vruntime",
 551                         SPLIT_NS(MIN_vruntime));
 552         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "min_vruntime",
 553                         SPLIT_NS(min_vruntime));
 554         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "max_vruntime",
 555                         SPLIT_NS(max_vruntime));
 556         spread = max_vruntime - MIN_vruntime;
 557         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread",
 558                         SPLIT_NS(spread));
 559         spread0 = min_vruntime - rq0_min_vruntime;
 560         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread0",
 561                         SPLIT_NS(spread0));
 562         SEQ_printf(m, "  .%-30s: %d\n", "nr_spread_over",
 563                         cfs_rq->nr_spread_over);
 564         SEQ_printf(m, "  .%-30s: %d\n", "nr_running", cfs_rq->nr_running);
 565         SEQ_printf(m, "  .%-30s: %ld\n", "load", cfs_rq->load.weight);
 566 #ifdef CONFIG_SMP
 567         SEQ_printf(m, "  .%-30s: %lu\n", "load_avg",
 568                         cfs_rq->avg.load_avg);
 569         SEQ_printf(m, "  .%-30s: %lu\n", "runnable_avg",
 570                         cfs_rq->avg.runnable_avg);
 571         SEQ_printf(m, "  .%-30s: %lu\n", "util_avg",
 572                         cfs_rq->avg.util_avg);
 573         SEQ_printf(m, "  .%-30s: %u\n", "util_est_enqueued",
 574                         cfs_rq->avg.util_est.enqueued);
 575         SEQ_printf(m, "  .%-30s: %ld\n", "removed.load_avg",
 576                         cfs_rq->removed.load_avg);
 577         SEQ_printf(m, "  .%-30s: %ld\n", "removed.util_avg",
 578                         cfs_rq->removed.util_avg);
 579         SEQ_printf(m, "  .%-30s: %ld\n", "removed.runnable_avg",
 580                         cfs_rq->removed.runnable_avg);
 581 #ifdef CONFIG_FAIR_GROUP_SCHED
 582         SEQ_printf(m, "  .%-30s: %lu\n", "tg_load_avg_contrib",
 583                         cfs_rq->tg_load_avg_contrib);
 584         SEQ_printf(m, "  .%-30s: %ld\n", "tg_load_avg",
 585                         atomic_long_read(&cfs_rq->tg->load_avg));
 586 #endif
 587 #endif
 588 #ifdef CONFIG_CFS_BANDWIDTH
 589         SEQ_printf(m, "  .%-30s: %d\n", "throttled",
 590                         cfs_rq->throttled);
 591         SEQ_printf(m, "  .%-30s: %d\n", "throttle_count",
 592                         cfs_rq->throttle_count);
 593 #endif
 594
 595 #ifdef CONFIG_FAIR_GROUP_SCHED
 596         print_cfs_group_stats(m, cpu, cfs_rq->tg);
 597 #endif
 598 }
 599
 600 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
 601 {
 602 #ifdef CONFIG_RT_GROUP_SCHED
 603         SEQ_printf(m, "\n");
 604         SEQ_printf(m, "rt_rq[%d]:%s\n", cpu, task_group_path(rt_rq->tg));
 605 #else
 606         SEQ_printf(m, "\n");
 607         SEQ_printf(m, "rt_rq[%d]:\n", cpu);
 608 #endif
 609
 610 #define P(x) \
 611         SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
 612 #define PU(x) \
 613         SEQ_printf(m, "  .%-30s: %lu\n", #x, (unsigned long)(rt_rq->x))
 614 #define PN(x) \
 615         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
 616
 617         PU(rt_nr_running);
 618 #ifdef CONFIG_SMP
 619         PU(rt_nr_migratory);
 620 #endif
 621         P(rt_throttled);
 622         PN(rt_time);
 623         PN(rt_runtime);
 624
 625 #undef PN
 626 #undef PU
 627 #undef P
 628 }
 629
 630 void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq)
 631 {
 632         struct dl_bw *dl_bw;
 633
 634         SEQ_printf(m, "\n");
 635         SEQ_printf(m, "dl_rq[%d]:\n", cpu);
 636
 637 #define PU(x) \
 638         SEQ_printf(m, "  .%-30s: %lu\n", #x, (unsigned long)(dl_rq->x))
 639
 640         PU(dl_nr_running);
 641 #ifdef CONFIG_SMP
 642         PU(dl_nr_migratory);
 643         dl_bw = &cpu_rq(cpu)->rd->dl_bw;
 644 #else
 645         dl_bw = &dl_rq->dl_bw;
 646 #endif
 647         SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->bw", dl_bw->bw);
 648         SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->total_bw", dl_bw->total_bw);
 649
 650 #undef PU
 651 }
 652
 653 static void print_cpu(struct seq_file *m, int cpu)
 654 {
 655         struct rq *rq = cpu_rq(cpu);
 656         unsigned long flags;
 657
 658 #ifdef CONFIG_X86
 659         {
 660                 unsigned int freq = cpu_khz ? : 1;
 661
 662                 SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
 663                            cpu, freq / 1000, (freq % 1000));
 664         }
 665 #else
 666         SEQ_printf(m, "cpu#%d\n", cpu);
 667 #endif
 668
 669 #define P(x)                                                            \
 670 do {                                                                    \
 671         if (sizeof(rq->x) == 4)                                         \
 672                 SEQ_printf(m, "  .%-30s: %ld\n", #x, (long)(rq->x));    \
 673         else                                                            \
 674                 SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rq->x));\
 675 } while (0)
 676
 677 #define PN(x) \
 678         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
 679
 680         P(nr_running);
 681         P(nr_switches);
 682         P(nr_uninterruptible);
 683         PN(next_balance);
 684         SEQ_printf(m, "  .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr)));
 685         PN(clock);
 686         PN(clock_task);
 687 #undef P
 688 #undef PN
 689
 690 #ifdef CONFIG_SMP
 691 #define P64(n) SEQ_printf(m, "  .%-30s: %Ld\n", #n, rq->n);
 692         P64(avg_idle);
 693         P64(max_idle_balance_cost);
 694 #undef P64
 695 #endif
 696
 697 #define P(n) SEQ_printf(m, "  .%-30s: %d\n", #n, schedstat_val(rq->n));
 698         if (schedstat_enabled()) {
 699                 P(yld_count);
 700                 P(sched_count);
 701                 P(sched_goidle);
 702                 P(ttwu_count);
 703                 P(ttwu_local);
 704         }
 705 #undef P
 706
 707         spin_lock_irqsave(&sched_debug_lock, flags);
 708         print_cfs_stats(m, cpu);
 709         print_rt_stats(m, cpu);
 710         print_dl_stats(m, cpu);
 711
 712         print_rq(m, rq, cpu);
 713         spin_unlock_irqrestore(&sched_debug_lock, flags);
 714         SEQ_printf(m, "\n");
 715 }
 716
 717 static const char *sched_tunable_scaling_names[] = {
 718         "none",
 719         "logarithmic",
 720         "linear"
 721 };
 722
 723 static void sched_debug_header(struct seq_file *m)
 724 {
 725         u64 ktime, sched_clk, cpu_clk;
 726         unsigned long flags;
 727
 728         local_irq_save(flags);
 729         ktime = ktime_to_ns(ktime_get());
 730         sched_clk = sched_clock();
 731         cpu_clk = local_clock();
 732         local_irq_restore(flags);
 733
 734         SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n",
 735                 init_utsname()->release,
 736                 (int)strcspn(init_utsname()->version, " "),
 737                 init_utsname()->version);
 738
 739 #define P(x) \
 740         SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
 741 #define PN(x) \
 742         SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
 743         PN(ktime);
 744         PN(sched_clk);
 745         PN(cpu_clk);
 746         P(jiffies);
 747 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
 748         P(sched_clock_stable());
 749 #endif
 750 #undef PN
 751 #undef P
 752
 753         SEQ_printf(m, "\n");
 754         SEQ_printf(m, "sysctl_sched\n");
 755
 756 #define P(x) \
 757         SEQ_printf(m, "  .%-40s: %Ld\n", #x, (long long)(x))
 758 #define PN(x) \
 759         SEQ_printf(m, "  .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
 760         PN(sysctl_sched_latency);
 761         PN(sysctl_sched_min_granularity);
 762         PN(sysctl_sched_wakeup_granularity);
 763         P(sysctl_sched_child_runs_first);
 764         P(sysctl_sched_features);
 765 #undef PN
 766 #undef P
 767
 768         SEQ_printf(m, "  .%-40s: %d (%s)\n",
 769                 "sysctl_sched_tunable_scaling",
 770                 sysctl_sched_tunable_scaling,
 771                 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
 772         SEQ_printf(m, "\n");
 773 }
 774
 775 static int sched_debug_show(struct seq_file *m, void *v)
 776 {
 777         int cpu = (unsigned long)(v - 2);
 778
 779         if (cpu != -1)
 780                 print_cpu(m, cpu);
 781         else
 782                 sched_debug_header(m);
 783
 784         return 0;
 785 }
 786
 787 void sysrq_sched_debug_show(void)
 788 {
 789         int cpu;
 790
 791         sched_debug_header(NULL);
 792         for_each_online_cpu(cpu) {
 793                 /*
 794                  * Need to reset softlockup watchdogs on all CPUs, because
 795                  * another CPU might be blocked waiting for us to process
 796                  * an IPI or stop_machine.
 797                  */
 798                 touch_nmi_watchdog();
 799                 touch_all_softlockup_watchdogs();
 800                 print_cpu(NULL, cpu);
 801         }
 802 }
 803
 804 /*
 805  * This iterator needs some explanation.
 806  * It returns 1 for the header position.
 807  * This means 2 is CPU 0.
 808  * In a hotplugged system some CPUs, including CPU 0, may be missing so we have
 809  * to use cpumask_* to iterate over the CPUs.
 810  */
 811 static void *sched_debug_start(struct seq_file *file, loff_t *offset)
 812 {
 813         unsigned long n = *offset;
 814
 815         if (n == 0)
 816                 return (void *) 1;
 817
 818         n--;
 819
 820         if (n > 0)
 821                 n = cpumask_next(n - 1, cpu_online_mask);
 822         else
 823                 n = cpumask_first(cpu_online_mask);
 824
 825         *offset = n + 1;
 826
 827         if (n < nr_cpu_ids)
 828                 return (void *)(unsigned long)(n + 2);
 829
 830         return NULL;
 831 }
 832
 833 static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
 834 {
 835         (*offset)++;
 836         return sched_debug_start(file, offset);
 837 }
 838
 839 static void sched_debug_stop(struct seq_file *file, void *data)
 840 {
 841 }
 842
 843 static const struct seq_operations sched_debug_sops = {
 844         .start          = sched_debug_start,
 845         .next           = sched_debug_next,
 846         .stop           = sched_debug_stop,
 847         .show           = sched_debug_show,
 848 };
 849
 850 static int __init init_sched_debug_procfs(void)
 851 {
 852         if (!proc_create_seq("sched_debug", 0444, NULL, &sched_debug_sops))
 853                 return -ENOMEM;
 854         return 0;
 855 }
 856
 857 __initcall(init_sched_debug_procfs);
 858
 859 #define __PS(S, F) SEQ_printf(m, "%-45s:%21Ld\n", S, (long long)(F))
 860 #define __P(F) __PS(#F, F)
 861 #define   P(F) __PS(#F, p->F)
 862 #define __PSN(S, F) SEQ_printf(m, "%-45s:%14Ld.%06ld\n", S, SPLIT_NS((long long)(F)))
 863 #define __PN(F) __PSN(#F, F)
 864 #define   PN(F) __PSN(#F, p->F)
 865
 866
 867 #ifdef CONFIG_NUMA_BALANCING
 868 void print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
 869                 unsigned long tpf, unsigned long gsf, unsigned long gpf)
 870 {
 871         SEQ_printf(m, "numa_faults node=%d ", node);
 872         SEQ_printf(m, "task_private=%lu task_shared=%lu ", tpf, tsf);
 873         SEQ_printf(m, "group_private=%lu group_shared=%lu\n", gpf, gsf);
 874 }
 875 #endif
 876
 877
 878 static void sched_show_numa(struct task_struct *p, struct seq_file *m)
 879 {
 880 #ifdef CONFIG_NUMA_BALANCING
 881         struct mempolicy *pol;
 882
 883         if (p->mm)
 884                 P(mm->numa_scan_seq);
 885
 886         task_lock(p);
 887         pol = p->mempolicy;
 888         if (pol && !(pol->flags & MPOL_F_MORON))
 889                 pol = NULL;
 890         mpol_get(pol);
 891         task_unlock(p);
 892
 893         P(numa_pages_migrated);
 894         P(numa_preferred_nid);
 895         P(total_numa_faults);
 896         SEQ_printf(m, "current_node=%d, numa_group_id=%d\n",
 897                         task_node(p), task_numa_group_id(p));
 898         show_numa_stats(p, m);
 899         mpol_put(pol);
 900 #endif
 901 }
 902
 903 void proc_sched_show_task(struct task_struct *p, struct pid_namespace *ns,
 904                                                   struct seq_file *m)
 905 {
 906         unsigned long nr_switches;
 907
 908         SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr_ns(p, ns),
 909                                                 get_nr_threads(p));
 910         SEQ_printf(m,
 911                 "---------------------------------------------------------"
 912                 "----------\n");
 913
 914 #define P_SCHEDSTAT(F)  __PS(#F, schedstat_val(p->F))
 915 #define PN_SCHEDSTAT(F) __PSN(#F, schedstat_val(p->F))
 916
 917         PN(se.exec_start);
 918         PN(se.vruntime);
 919         PN(se.sum_exec_runtime);
 920
 921         nr_switches = p->nvcsw + p->nivcsw;
 922
 923         P(se.nr_migrations);
 924
 925         if (schedstat_enabled()) {
 926                 u64 avg_atom, avg_per_cpu;
 927
 928                 PN_SCHEDSTAT(se.statistics.sum_sleep_runtime);
 929                 PN_SCHEDSTAT(se.statistics.wait_start);
 930                 PN_SCHEDSTAT(se.statistics.sleep_start);
 931                 PN_SCHEDSTAT(se.statistics.block_start);
 932                 PN_SCHEDSTAT(se.statistics.sleep_max);
 933                 PN_SCHEDSTAT(se.statistics.block_max);
 934                 PN_SCHEDSTAT(se.statistics.exec_max);
 935                 PN_SCHEDSTAT(se.statistics.slice_max);
 936                 PN_SCHEDSTAT(se.statistics.wait_max);
 937                 PN_SCHEDSTAT(se.statistics.wait_sum);
 938                 P_SCHEDSTAT(se.statistics.wait_count);
 939                 PN_SCHEDSTAT(se.statistics.iowait_sum);
 940                 P_SCHEDSTAT(se.statistics.iowait_count);
 941                 P_SCHEDSTAT(se.statistics.nr_migrations_cold);
 942                 P_SCHEDSTAT(se.statistics.nr_failed_migrations_affine);
 943                 P_SCHEDSTAT(se.statistics.nr_failed_migrations_running);
 944                 P_SCHEDSTAT(se.statistics.nr_failed_migrations_hot);
 945                 P_SCHEDSTAT(se.statistics.nr_forced_migrations);
 946                 P_SCHEDSTAT(se.statistics.nr_wakeups);
 947                 P_SCHEDSTAT(se.statistics.nr_wakeups_sync);
 948                 P_SCHEDSTAT(se.statistics.nr_wakeups_migrate);
 949                 P_SCHEDSTAT(se.statistics.nr_wakeups_local);
 950                 P_SCHEDSTAT(se.statistics.nr_wakeups_remote);
 951                 P_SCHEDSTAT(se.statistics.nr_wakeups_affine);
 952                 P_SCHEDSTAT(se.statistics.nr_wakeups_affine_attempts);
 953                 P_SCHEDSTAT(se.statistics.nr_wakeups_passive);
 954                 P_SCHEDSTAT(se.statistics.nr_wakeups_idle);
 955
 956                 avg_atom = p->se.sum_exec_runtime;
 957                 if (nr_switches)
 958                         avg_atom = div64_ul(avg_atom, nr_switches);
 959                 else
 960                         avg_atom = -1LL;
 961
 962                 avg_per_cpu = p->se.sum_exec_runtime;
 963                 if (p->se.nr_migrations) {
 964                         avg_per_cpu = div64_u64(avg_per_cpu,
 965                                                 p->se.nr_migrations);
 966                 } else {
 967                         avg_per_cpu = -1LL;
 968                 }
 969
 970                 __PN(avg_atom);
 971                 __PN(avg_per_cpu);
 972         }
 973
 974         __P(nr_switches);
 975         __PS("nr_voluntary_switches", p->nvcsw);
 976         __PS("nr_involuntary_switches", p->nivcsw);
 977
 978         P(se.load.weight);
 979 #ifdef CONFIG_SMP
 980         P(se.avg.load_sum);
 981         P(se.avg.runnable_sum);
 982         P(se.avg.util_sum);
 983         P(se.avg.load_avg);
 984         P(se.avg.runnable_avg);
 985         P(se.avg.util_avg);
 986         P(se.avg.last_update_time);
 987         P(se.avg.util_est.ewma);
 988         P(se.avg.util_est.enqueued);
 989 #endif
 990 #ifdef CONFIG_UCLAMP_TASK
 991         __PS("uclamp.min", p->uclamp_req[UCLAMP_MIN].value);
 992         __PS("uclamp.max", p->uclamp_req[UCLAMP_MAX].value);
 993         __PS("effective uclamp.min", uclamp_eff_value(p, UCLAMP_MIN));
 994         __PS("effective uclamp.max", uclamp_eff_value(p, UCLAMP_MAX));
 995 #endif
 996         P(policy);
 997         P(prio);
 998         if (task_has_dl_policy(p)) {
 999                 P(dl.runtime);
1000                 P(dl.deadline);
1001         }
1002 #undef PN_SCHEDSTAT
1003 #undef P_SCHEDSTAT
1004
1005         {
1006                 unsigned int this_cpu = raw_smp_processor_id();
1007                 u64 t0, t1;
1008
1009                 t0 = cpu_clock(this_cpu);
1010                 t1 = cpu_clock(this_cpu);
1011                 __PS("clock-delta", t1-t0);
1012         }
1013
1014         sched_show_numa(p, m);
1015 }
1016
1017 void proc_sched_set_task(struct task_struct *p)
1018 {
1019 #ifdef CONFIG_SCHEDSTATS
1020         memset(&p->se.statistics, 0, sizeof(p->se.statistics));
1021 #endif
1022 }