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 /*
  12  * This allows printing both to /proc/sched_debug and
  13  * to the console
  14  */
  15 #define SEQ_printf(m, x...)                     \
  16  do {                                           \
  17         if (m)                                  \
  18                 seq_printf(m, x);               \
  19         else                                    \
  20                 pr_cont(x);                     \
  21  } while (0)
  22
  23 /*
  24  * Ease the printing of nsec fields:
  25  */
  26 static long long nsec_high(unsigned long long nsec)
  27 {
  28         if ((long long)nsec < 0) {
  29                 nsec = -nsec;
  30                 do_div(nsec, 1000000);
  31                 return -nsec;
  32         }
  33         do_div(nsec, 1000000);
  34
  35         return nsec;
  36 }
  37
  38 static unsigned long nsec_low(unsigned long long nsec)
  39 {
  40         if ((long long)nsec < 0)
  41                 nsec = -nsec;
  42
  43         return do_div(nsec, 1000000);
  44 }
  45
  46 #define SPLIT_NS(x) nsec_high(x), nsec_low(x)
  47
  48 #define SCHED_FEAT(name, enabled)       \
  49         #name ,
  50
  51 static const char * const sched_feat_names[] = {
  52 #include "features.h"
  53 };
  54
  55 #undef SCHED_FEAT
  56
  57 static int sched_feat_show(struct seq_file *m, void *v)
  58 {
  59         int i;
  60
  61         for (i = 0; i < __SCHED_FEAT_NR; i++) {
  62                 if (!(sysctl_sched_features & (1UL << i)))
  63                         seq_puts(m, "NO_");
  64                 seq_printf(m, "%s ", sched_feat_names[i]);
  65         }
  66         seq_puts(m, "\n");
  67
  68         return 0;
  69 }
  70
  71 #ifdef CONFIG_JUMP_LABEL
  72
  73 #define jump_label_key__true  STATIC_KEY_INIT_TRUE
  74 #define jump_label_key__false STATIC_KEY_INIT_FALSE
  75
  76 #define SCHED_FEAT(name, enabled)       \
  77         jump_label_key__##enabled ,
  78
  79 struct static_key sched_feat_keys[__SCHED_FEAT_NR] = {
  80 #include "features.h"
  81 };
  82
  83 #undef SCHED_FEAT
  84
  85 static void sched_feat_disable(int i)
  86 {
  87         static_key_disable_cpuslocked(&sched_feat_keys[i]);
  88 }
  89
  90 static void sched_feat_enable(int i)
  91 {
  92         static_key_enable_cpuslocked(&sched_feat_keys[i]);
  93 }
  94 #else
  95 static void sched_feat_disable(int i) { };
  96 static void sched_feat_enable(int i) { };
  97 #endif /* CONFIG_JUMP_LABEL */
  98
  99 static int sched_feat_set(char *cmp)
 100 {
 101         int i;
 102         int neg = 0;
 103
 104         if (strncmp(cmp, "NO_", 3) == 0) {
 105                 neg = 1;
 106                 cmp += 3;
 107         }
 108
 109         i = match_string(sched_feat_names, __SCHED_FEAT_NR, cmp);
 110         if (i < 0)
 111                 return i;
 112
 113         if (neg) {
 114                 sysctl_sched_features &= ~(1UL << i);
 115                 sched_feat_disable(i);
 116         } else {
 117                 sysctl_sched_features |= (1UL << i);
 118                 sched_feat_enable(i);
 119         }
 120
 121         return 0;
 122 }
 123
 124 static ssize_t
 125 sched_feat_write(struct file *filp, const char __user *ubuf,
 126                 size_t cnt, loff_t *ppos)
 127 {
 128         char buf[64];
 129         char *cmp;
 130         int ret;
 131         struct inode *inode;
 132
 133         if (cnt > 63)
 134                 cnt = 63;
 135
 136         if (copy_from_user(&buf, ubuf, cnt))
 137                 return -EFAULT;
 138
 139         buf[cnt] = 0;
 140         cmp = strstrip(buf);
 141
 142         /* Ensure the static_key remains in a consistent state */
 143         inode = file_inode(filp);
 144         cpus_read_lock();
 145         inode_lock(inode);
 146         ret = sched_feat_set(cmp);
 147         inode_unlock(inode);
 148         cpus_read_unlock();
 149         if (ret < 0)
 150                 return ret;
 151
 152         *ppos += cnt;
 153
 154         return cnt;
 155 }
 156
 157 static int sched_feat_open(struct inode *inode, struct file *filp)
 158 {
 159         return single_open(filp, sched_feat_show, NULL);
 160 }
 161
 162 static const struct file_operations sched_feat_fops = {
 163         .open           = sched_feat_open,
 164         .write          = sched_feat_write,
 165         .read           = seq_read,
 166         .llseek         = seq_lseek,
 167         .release        = single_release,
 168 };
 169
 170 #ifdef CONFIG_SMP
 171
 172 static ssize_t sched_scaling_write(struct file *filp, const char __user *ubuf,
 173                                    size_t cnt, loff_t *ppos)
 174 {
 175         char buf[16];
 176         unsigned int scaling;
 177
 178         if (cnt > 15)
 179                 cnt = 15;
 180
 181         if (copy_from_user(&buf, ubuf, cnt))
 182                 return -EFAULT;
 183         buf[cnt] = '\0';
 184
 185         if (kstrtouint(buf, 10, &scaling))
 186                 return -EINVAL;
 187
 188         if (scaling >= SCHED_TUNABLESCALING_END)
 189                 return -EINVAL;
 190
 191         sysctl_sched_tunable_scaling = scaling;
 192         if (sched_update_scaling())
 193                 return -EINVAL;
 194
 195         *ppos += cnt;
 196         return cnt;
 197 }
 198
 199 static int sched_scaling_show(struct seq_file *m, void *v)
 200 {
 201         seq_printf(m, "%d\n", sysctl_sched_tunable_scaling);
 202         return 0;
 203 }
 204
 205 static int sched_scaling_open(struct inode *inode, struct file *filp)
 206 {
 207         return single_open(filp, sched_scaling_show, NULL);
 208 }
 209
 210 static const struct file_operations sched_scaling_fops = {
 211         .open           = sched_scaling_open,
 212         .write          = sched_scaling_write,
 213         .read           = seq_read,
 214         .llseek         = seq_lseek,
 215         .release        = single_release,
 216 };
 217
 218 #endif /* SMP */
 219
 220 #ifdef CONFIG_PREEMPT_DYNAMIC
 221
 222 static ssize_t sched_dynamic_write(struct file *filp, const char __user *ubuf,
 223                                    size_t cnt, loff_t *ppos)
 224 {
 225         char buf[16];
 226         int mode;
 227
 228         if (cnt > 15)
 229                 cnt = 15;
 230
 231         if (copy_from_user(&buf, ubuf, cnt))
 232                 return -EFAULT;
 233
 234         buf[cnt] = 0;
 235         mode = sched_dynamic_mode(strstrip(buf));
 236         if (mode < 0)
 237                 return mode;
 238
 239         sched_dynamic_update(mode);
 240
 241         *ppos += cnt;
 242
 243         return cnt;
 244 }
 245
 246 static int sched_dynamic_show(struct seq_file *m, void *v)
 247 {
 248         static const char * preempt_modes[] = {
 249                 "none", "voluntary", "full"
 250         };
 251         int i;
 252
 253         for (i = 0; i < ARRAY_SIZE(preempt_modes); i++) {
 254                 if (preempt_dynamic_mode == i)
 255                         seq_puts(m, "(");
 256                 seq_puts(m, preempt_modes[i]);
 257                 if (preempt_dynamic_mode == i)
 258                         seq_puts(m, ")");
 259
 260                 seq_puts(m, " ");
 261         }
 262
 263         seq_puts(m, "\n");
 264         return 0;
 265 }
 266
 267 static int sched_dynamic_open(struct inode *inode, struct file *filp)
 268 {
 269         return single_open(filp, sched_dynamic_show, NULL);
 270 }
 271
 272 static const struct file_operations sched_dynamic_fops = {
 273         .open           = sched_dynamic_open,
 274         .write          = sched_dynamic_write,
 275         .read           = seq_read,
 276         .llseek         = seq_lseek,
 277         .release        = single_release,
 278 };
 279
 280 #endif /* CONFIG_PREEMPT_DYNAMIC */
 281
 282 __read_mostly bool sched_debug_verbose;
 283
 284 static const struct seq_operations sched_debug_sops;
 285
 286 static int sched_debug_open(struct inode *inode, struct file *filp)
 287 {
 288         return seq_open(filp, &sched_debug_sops);
 289 }
 290
 291 static const struct file_operations sched_debug_fops = {
 292         .open           = sched_debug_open,
 293         .read           = seq_read,
 294         .llseek         = seq_lseek,
 295         .release        = seq_release,
 296 };
 297
 298 static struct dentry *debugfs_sched;
 299
 300 static __init int sched_init_debug(void)
 301 {
 302         struct dentry __maybe_unused *numa;
 303
 304         debugfs_sched = debugfs_create_dir("sched", NULL);
 305
 306         debugfs_create_file("features", 0644, debugfs_sched, NULL, &sched_feat_fops);
 307         debugfs_create_bool("verbose", 0644, debugfs_sched, &sched_debug_verbose);
 308 #ifdef CONFIG_PREEMPT_DYNAMIC
 309         debugfs_create_file("preempt", 0644, debugfs_sched, NULL, &sched_dynamic_fops);
 310 #endif
 311
 312         debugfs_create_u32("latency_ns", 0644, debugfs_sched, &sysctl_sched_latency);
 313         debugfs_create_u32("min_granularity_ns", 0644, debugfs_sched, &sysctl_sched_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",       #F, (long long)schedstat_val(F))
 452 #define PN(F)           SEQ_printf(m, "  .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F))
 453 #define PN_SCHEDSTAT(F) SEQ_printf(m, "  .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)schedstat_val(F)))
 454
 455         if (!se)
 456                 return;
 457
 458         PN(se->exec_start);
 459         PN(se->vruntime);
 460         PN(se->sum_exec_runtime);
 461
 462         if (schedstat_enabled()) {
 463                 PN_SCHEDSTAT(se->statistics.wait_start);
 464                 PN_SCHEDSTAT(se->statistics.sleep_start);
 465                 PN_SCHEDSTAT(se->statistics.block_start);
 466                 PN_SCHEDSTAT(se->statistics.sleep_max);
 467                 PN_SCHEDSTAT(se->statistics.block_max);
 468                 PN_SCHEDSTAT(se->statistics.exec_max);
 469                 PN_SCHEDSTAT(se->statistics.slice_max);
 470                 PN_SCHEDSTAT(se->statistics.wait_max);
 471                 PN_SCHEDSTAT(se->statistics.wait_sum);
 472                 P_SCHEDSTAT(se->statistics.wait_count);
 473         }
 474
 475         P(se->load.weight);
 476 #ifdef CONFIG_SMP
 477         P(se->avg.load_avg);
 478         P(se->avg.util_avg);
 479         P(se->avg.runnable_avg);
 480 #endif
 481
 482 #undef PN_SCHEDSTAT
 483 #undef PN
 484 #undef P_SCHEDSTAT
 485 #undef P
 486 }
 487 #endif
 488
 489 #ifdef CONFIG_CGROUP_SCHED
 490 static DEFINE_SPINLOCK(sched_debug_lock);
 491 static char group_path[PATH_MAX];
 492
 493 static void task_group_path(struct task_group *tg, char *path, int plen)
 494 {
 495         if (autogroup_path(tg, path, plen))
 496                 return;
 497
 498         cgroup_path(tg->css.cgroup, path, plen);
 499 }
 500
 501 /*
 502  * Only 1 SEQ_printf_task_group_path() caller can use the full length
 503  * group_path[] for cgroup path. Other simultaneous callers will have
 504  * to use a shorter stack buffer. A "..." suffix is appended at the end
 505  * of the stack buffer so that it will show up in case the output length
 506  * matches the given buffer size to indicate possible path name truncation.
 507  */
 508 #define SEQ_printf_task_group_path(m, tg, fmt...)                       \
 509 {                                                                       \
 510         if (spin_trylock(&sched_debug_lock)) {                          \
 511                 task_group_path(tg, group_path, sizeof(group_path));    \
 512                 SEQ_printf(m, fmt, group_path);                         \
 513                 spin_unlock(&sched_debug_lock);                         \
 514         } else {                                                        \
 515                 char buf[128];                                          \
 516                 char *bufend = buf + sizeof(buf) - 3;                   \
 517                 task_group_path(tg, buf, bufend - buf);                 \
 518                 strcpy(bufend - 1, "...");                              \
 519                 SEQ_printf(m, fmt, buf);                                \
 520         }                                                               \
 521 }
 522 #endif
 523
 524 static void
 525 print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
 526 {
 527         if (task_current(rq, p))
 528                 SEQ_printf(m, ">R");
 529         else
 530                 SEQ_printf(m, " %c", task_state_to_char(p));
 531
 532         SEQ_printf(m, " %15s %5d %9Ld.%06ld %9Ld %5d ",
 533                 p->comm, task_pid_nr(p),
 534                 SPLIT_NS(p->se.vruntime),
 535                 (long long)(p->nvcsw + p->nivcsw),
 536                 p->prio);
 537
 538         SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
 539                 SPLIT_NS(schedstat_val_or_zero(p->se.statistics.wait_sum)),
 540                 SPLIT_NS(p->se.sum_exec_runtime),
 541                 SPLIT_NS(schedstat_val_or_zero(p->se.statistics.sum_sleep_runtime)));
 542
 543 #ifdef CONFIG_NUMA_BALANCING
 544         SEQ_printf(m, " %d %d", task_node(p), task_numa_group_id(p));
 545 #endif
 546 #ifdef CONFIG_CGROUP_SCHED
 547         SEQ_printf_task_group_path(m, task_group(p), " %s")
 548 #endif
 549
 550         SEQ_printf(m, "\n");
 551 }
 552
 553 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
 554 {
 555         struct task_struct *g, *p;
 556
 557         SEQ_printf(m, "\n");
 558         SEQ_printf(m, "runnable tasks:\n");
 559         SEQ_printf(m, " S            task   PID         tree-key  switches  prio"
 560                    "     wait-time             sum-exec        sum-sleep\n");
 561         SEQ_printf(m, "-------------------------------------------------------"
 562                    "------------------------------------------------------\n");
 563
 564         rcu_read_lock();
 565         for_each_process_thread(g, p) {
 566                 if (task_cpu(p) != rq_cpu)
 567                         continue;
 568
 569                 print_task(m, rq, p);
 570         }
 571         rcu_read_unlock();
 572 }
 573
 574 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
 575 {
 576         s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
 577                 spread, rq0_min_vruntime, spread0;
 578         struct rq *rq = cpu_rq(cpu);
 579         struct sched_entity *last;
 580         unsigned long flags;
 581
 582 #ifdef CONFIG_FAIR_GROUP_SCHED
 583         SEQ_printf(m, "\n");
 584         SEQ_printf_task_group_path(m, cfs_rq->tg, "cfs_rq[%d]:%s\n", cpu);
 585 #else
 586         SEQ_printf(m, "\n");
 587         SEQ_printf(m, "cfs_rq[%d]:\n", cpu);
 588 #endif
 589         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "exec_clock",
 590                         SPLIT_NS(cfs_rq->exec_clock));
 591
 592         raw_spin_rq_lock_irqsave(rq, flags);
 593         if (rb_first_cached(&cfs_rq->tasks_timeline))
 594                 MIN_vruntime = (__pick_first_entity(cfs_rq))->vruntime;
 595         last = __pick_last_entity(cfs_rq);
 596         if (last)
 597                 max_vruntime = last->vruntime;
 598         min_vruntime = cfs_rq->min_vruntime;
 599         rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
 600         raw_spin_rq_unlock_irqrestore(rq, flags);
 601         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "MIN_vruntime",
 602                         SPLIT_NS(MIN_vruntime));
 603         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "min_vruntime",
 604                         SPLIT_NS(min_vruntime));
 605         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "max_vruntime",
 606                         SPLIT_NS(max_vruntime));
 607         spread = max_vruntime - MIN_vruntime;
 608         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread",
 609                         SPLIT_NS(spread));
 610         spread0 = min_vruntime - rq0_min_vruntime;
 611         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread0",
 612                         SPLIT_NS(spread0));
 613         SEQ_printf(m, "  .%-30s: %d\n", "nr_spread_over",
 614                         cfs_rq->nr_spread_over);
 615         SEQ_printf(m, "  .%-30s: %d\n", "nr_running", cfs_rq->nr_running);
 616         SEQ_printf(m, "  .%-30s: %d\n", "h_nr_running", cfs_rq->h_nr_running);
 617         SEQ_printf(m, "  .%-30s: %d\n", "idle_h_nr_running",
 618                         cfs_rq->idle_h_nr_running);
 619         SEQ_printf(m, "  .%-30s: %ld\n", "load", cfs_rq->load.weight);
 620 #ifdef CONFIG_SMP
 621         SEQ_printf(m, "  .%-30s: %lu\n", "load_avg",
 622                         cfs_rq->avg.load_avg);
 623         SEQ_printf(m, "  .%-30s: %lu\n", "runnable_avg",
 624                         cfs_rq->avg.runnable_avg);
 625         SEQ_printf(m, "  .%-30s: %lu\n", "util_avg",
 626                         cfs_rq->avg.util_avg);
 627         SEQ_printf(m, "  .%-30s: %u\n", "util_est_enqueued",
 628                         cfs_rq->avg.util_est.enqueued);
 629         SEQ_printf(m, "  .%-30s: %ld\n", "removed.load_avg",
 630                         cfs_rq->removed.load_avg);
 631         SEQ_printf(m, "  .%-30s: %ld\n", "removed.util_avg",
 632                         cfs_rq->removed.util_avg);
 633         SEQ_printf(m, "  .%-30s: %ld\n", "removed.runnable_avg",
 634                         cfs_rq->removed.runnable_avg);
 635 #ifdef CONFIG_FAIR_GROUP_SCHED
 636         SEQ_printf(m, "  .%-30s: %lu\n", "tg_load_avg_contrib",
 637                         cfs_rq->tg_load_avg_contrib);
 638         SEQ_printf(m, "  .%-30s: %ld\n", "tg_load_avg",
 639                         atomic_long_read(&cfs_rq->tg->load_avg));
 640 #endif
 641 #endif
 642 #ifdef CONFIG_CFS_BANDWIDTH
 643         SEQ_printf(m, "  .%-30s: %d\n", "throttled",
 644                         cfs_rq->throttled);
 645         SEQ_printf(m, "  .%-30s: %d\n", "throttle_count",
 646                         cfs_rq->throttle_count);
 647 #endif
 648
 649 #ifdef CONFIG_FAIR_GROUP_SCHED
 650         print_cfs_group_stats(m, cpu, cfs_rq->tg);
 651 #endif
 652 }
 653
 654 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
 655 {
 656 #ifdef CONFIG_RT_GROUP_SCHED
 657         SEQ_printf(m, "\n");
 658         SEQ_printf_task_group_path(m, rt_rq->tg, "rt_rq[%d]:%s\n", cpu);
 659 #else
 660         SEQ_printf(m, "\n");
 661         SEQ_printf(m, "rt_rq[%d]:\n", cpu);
 662 #endif
 663
 664 #define P(x) \
 665         SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
 666 #define PU(x) \
 667         SEQ_printf(m, "  .%-30s: %lu\n", #x, (unsigned long)(rt_rq->x))
 668 #define PN(x) \
 669         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
 670
 671         PU(rt_nr_running);
 672 #ifdef CONFIG_SMP
 673         PU(rt_nr_migratory);
 674 #endif
 675         P(rt_throttled);
 676         PN(rt_time);
 677         PN(rt_runtime);
 678
 679 #undef PN
 680 #undef PU
 681 #undef P
 682 }
 683
 684 void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq)
 685 {
 686         struct dl_bw *dl_bw;
 687
 688         SEQ_printf(m, "\n");
 689         SEQ_printf(m, "dl_rq[%d]:\n", cpu);
 690
 691 #define PU(x) \
 692         SEQ_printf(m, "  .%-30s: %lu\n", #x, (unsigned long)(dl_rq->x))
 693
 694         PU(dl_nr_running);
 695 #ifdef CONFIG_SMP
 696         PU(dl_nr_migratory);
 697         dl_bw = &cpu_rq(cpu)->rd->dl_bw;
 698 #else
 699         dl_bw = &dl_rq->dl_bw;
 700 #endif
 701         SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->bw", dl_bw->bw);
 702         SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->total_bw", dl_bw->total_bw);
 703
 704 #undef PU
 705 }
 706
 707 static void print_cpu(struct seq_file *m, int cpu)
 708 {
 709         struct rq *rq = cpu_rq(cpu);
 710
 711 #ifdef CONFIG_X86
 712         {
 713                 unsigned int freq = cpu_khz ? : 1;
 714
 715                 SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
 716                            cpu, freq / 1000, (freq % 1000));
 717         }
 718 #else
 719         SEQ_printf(m, "cpu#%d\n", cpu);
 720 #endif
 721
 722 #define P(x)                                                            \
 723 do {                                                                    \
 724         if (sizeof(rq->x) == 4)                                         \
 725                 SEQ_printf(m, "  .%-30s: %ld\n", #x, (long)(rq->x));    \
 726         else                                                            \
 727                 SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rq->x));\
 728 } while (0)
 729
 730 #define PN(x) \
 731         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
 732
 733         P(nr_running);
 734         P(nr_switches);
 735         P(nr_uninterruptible);
 736         PN(next_balance);
 737         SEQ_printf(m, "  .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr)));
 738         PN(clock);
 739         PN(clock_task);
 740 #undef P
 741 #undef PN
 742
 743 #ifdef CONFIG_SMP
 744 #define P64(n) SEQ_printf(m, "  .%-30s: %Ld\n", #n, rq->n);
 745         P64(avg_idle);
 746         P64(max_idle_balance_cost);
 747 #undef P64
 748 #endif
 749
 750 #define P(n) SEQ_printf(m, "  .%-30s: %d\n", #n, schedstat_val(rq->n));
 751         if (schedstat_enabled()) {
 752                 P(yld_count);
 753                 P(sched_count);
 754                 P(sched_goidle);
 755                 P(ttwu_count);
 756                 P(ttwu_local);
 757         }
 758 #undef P
 759
 760         print_cfs_stats(m, cpu);
 761         print_rt_stats(m, cpu);
 762         print_dl_stats(m, cpu);
 763
 764         print_rq(m, rq, cpu);
 765         SEQ_printf(m, "\n");
 766 }
 767
 768 static const char *sched_tunable_scaling_names[] = {
 769         "none",
 770         "logarithmic",
 771         "linear"
 772 };
 773
 774 static void sched_debug_header(struct seq_file *m)
 775 {
 776         u64 ktime, sched_clk, cpu_clk;
 777         unsigned long flags;
 778
 779         local_irq_save(flags);
 780         ktime = ktime_to_ns(ktime_get());
 781         sched_clk = sched_clock();
 782         cpu_clk = local_clock();
 783         local_irq_restore(flags);
 784
 785         SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n",
 786                 init_utsname()->release,
 787                 (int)strcspn(init_utsname()->version, " "),
 788                 init_utsname()->version);
 789
 790 #define P(x) \
 791         SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
 792 #define PN(x) \
 793         SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
 794         PN(ktime);
 795         PN(sched_clk);
 796         PN(cpu_clk);
 797         P(jiffies);
 798 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
 799         P(sched_clock_stable());
 800 #endif
 801 #undef PN
 802 #undef P
 803
 804         SEQ_printf(m, "\n");
 805         SEQ_printf(m, "sysctl_sched\n");
 806
 807 #define P(x) \
 808         SEQ_printf(m, "  .%-40s: %Ld\n", #x, (long long)(x))
 809 #define PN(x) \
 810         SEQ_printf(m, "  .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
 811         PN(sysctl_sched_latency);
 812         PN(sysctl_sched_min_granularity);
 813         PN(sysctl_sched_wakeup_granularity);
 814         P(sysctl_sched_child_runs_first);
 815         P(sysctl_sched_features);
 816 #undef PN
 817 #undef P
 818
 819         SEQ_printf(m, "  .%-40s: %d (%s)\n",
 820                 "sysctl_sched_tunable_scaling",
 821                 sysctl_sched_tunable_scaling,
 822                 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
 823         SEQ_printf(m, "\n");
 824 }
 825
 826 static int sched_debug_show(struct seq_file *m, void *v)
 827 {
 828         int cpu = (unsigned long)(v - 2);
 829
 830         if (cpu != -1)
 831                 print_cpu(m, cpu);
 832         else
 833                 sched_debug_header(m);
 834
 835         return 0;
 836 }
 837
 838 void sysrq_sched_debug_show(void)
 839 {
 840         int cpu;
 841
 842         sched_debug_header(NULL);
 843         for_each_online_cpu(cpu) {
 844                 /*
 845                  * Need to reset softlockup watchdogs on all CPUs, because
 846                  * another CPU might be blocked waiting for us to process
 847                  * an IPI or stop_machine.
 848                  */
 849                 touch_nmi_watchdog();
 850                 touch_all_softlockup_watchdogs();
 851                 print_cpu(NULL, cpu);
 852         }
 853 }
 854
 855 /*
 856  * This iterator needs some explanation.
 857  * It returns 1 for the header position.
 858  * This means 2 is CPU 0.
 859  * In a hotplugged system some CPUs, including CPU 0, may be missing so we have
 860  * to use cpumask_* to iterate over the CPUs.
 861  */
 862 static void *sched_debug_start(struct seq_file *file, loff_t *offset)
 863 {
 864         unsigned long n = *offset;
 865
 866         if (n == 0)
 867                 return (void *) 1;
 868
 869         n--;
 870
 871         if (n > 0)
 872                 n = cpumask_next(n - 1, cpu_online_mask);
 873         else
 874                 n = cpumask_first(cpu_online_mask);
 875
 876         *offset = n + 1;
 877
 878         if (n < nr_cpu_ids)
 879                 return (void *)(unsigned long)(n + 2);
 880
 881         return NULL;
 882 }
 883
 884 static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
 885 {
 886         (*offset)++;
 887         return sched_debug_start(file, offset);
 888 }
 889
 890 static void sched_debug_stop(struct seq_file *file, void *data)
 891 {
 892 }
 893
 894 static const struct seq_operations sched_debug_sops = {
 895         .start          = sched_debug_start,
 896         .next           = sched_debug_next,
 897         .stop           = sched_debug_stop,
 898         .show           = sched_debug_show,
 899 };
 900
 901 #define __PS(S, F) SEQ_printf(m, "%-45s:%21Ld\n", S, (long long)(F))
 902 #define __P(F) __PS(#F, F)
 903 #define   P(F) __PS(#F, p->F)
 904 #define   PM(F, M) __PS(#F, p->F & (M))
 905 #define __PSN(S, F) SEQ_printf(m, "%-45s:%14Ld.%06ld\n", S, SPLIT_NS((long long)(F)))
 906 #define __PN(F) __PSN(#F, F)
 907 #define   PN(F) __PSN(#F, p->F)
 908
 909
 910 #ifdef CONFIG_NUMA_BALANCING
 911 void print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
 912                 unsigned long tpf, unsigned long gsf, unsigned long gpf)
 913 {
 914         SEQ_printf(m, "numa_faults node=%d ", node);
 915         SEQ_printf(m, "task_private=%lu task_shared=%lu ", tpf, tsf);
 916         SEQ_printf(m, "group_private=%lu group_shared=%lu\n", gpf, gsf);
 917 }
 918 #endif
 919
 920
 921 static void sched_show_numa(struct task_struct *p, struct seq_file *m)
 922 {
 923 #ifdef CONFIG_NUMA_BALANCING
 924         struct mempolicy *pol;
 925
 926         if (p->mm)
 927                 P(mm->numa_scan_seq);
 928
 929         task_lock(p);
 930         pol = p->mempolicy;
 931         if (pol && !(pol->flags & MPOL_F_MORON))
 932                 pol = NULL;
 933         mpol_get(pol);
 934         task_unlock(p);
 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         mpol_put(pol);
 943 #endif
 944 }
 945
 946 void proc_sched_show_task(struct task_struct *p, struct pid_namespace *ns,
 947                                                   struct seq_file *m)
 948 {
 949         unsigned long nr_switches;
 950
 951         SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr_ns(p, ns),
 952                                                 get_nr_threads(p));
 953         SEQ_printf(m,
 954                 "---------------------------------------------------------"
 955                 "----------\n");
 956
 957 #define P_SCHEDSTAT(F)  __PS(#F, schedstat_val(p->F))
 958 #define PN_SCHEDSTAT(F) __PSN(#F, schedstat_val(p->F))
 959
 960         PN(se.exec_start);
 961         PN(se.vruntime);
 962         PN(se.sum_exec_runtime);
 963
 964         nr_switches = p->nvcsw + p->nivcsw;
 965
 966         P(se.nr_migrations);
 967
 968         if (schedstat_enabled()) {
 969                 u64 avg_atom, avg_per_cpu;
 970
 971                 PN_SCHEDSTAT(se.statistics.sum_sleep_runtime);
 972                 PN_SCHEDSTAT(se.statistics.wait_start);
 973                 PN_SCHEDSTAT(se.statistics.sleep_start);
 974                 PN_SCHEDSTAT(se.statistics.block_start);
 975                 PN_SCHEDSTAT(se.statistics.sleep_max);
 976                 PN_SCHEDSTAT(se.statistics.block_max);
 977                 PN_SCHEDSTAT(se.statistics.exec_max);
 978                 PN_SCHEDSTAT(se.statistics.slice_max);
 979                 PN_SCHEDSTAT(se.statistics.wait_max);
 980                 PN_SCHEDSTAT(se.statistics.wait_sum);
 981                 P_SCHEDSTAT(se.statistics.wait_count);
 982                 PN_SCHEDSTAT(se.statistics.iowait_sum);
 983                 P_SCHEDSTAT(se.statistics.iowait_count);
 984                 P_SCHEDSTAT(se.statistics.nr_migrations_cold);
 985                 P_SCHEDSTAT(se.statistics.nr_failed_migrations_affine);
 986                 P_SCHEDSTAT(se.statistics.nr_failed_migrations_running);
 987                 P_SCHEDSTAT(se.statistics.nr_failed_migrations_hot);
 988                 P_SCHEDSTAT(se.statistics.nr_forced_migrations);
 989                 P_SCHEDSTAT(se.statistics.nr_wakeups);
 990                 P_SCHEDSTAT(se.statistics.nr_wakeups_sync);
 991                 P_SCHEDSTAT(se.statistics.nr_wakeups_migrate);
 992                 P_SCHEDSTAT(se.statistics.nr_wakeups_local);
 993                 P_SCHEDSTAT(se.statistics.nr_wakeups_remote);
 994                 P_SCHEDSTAT(se.statistics.nr_wakeups_affine);
 995                 P_SCHEDSTAT(se.statistics.nr_wakeups_affine_attempts);
 996                 P_SCHEDSTAT(se.statistics.nr_wakeups_passive);
 997                 P_SCHEDSTAT(se.statistics.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
1017         __P(nr_switches);
1018         __PS("nr_voluntary_switches", p->nvcsw);
1019         __PS("nr_involuntary_switches", p->nivcsw);
1020
1021         P(se.load.weight);
1022 #ifdef CONFIG_SMP
1023         P(se.avg.load_sum);
1024         P(se.avg.runnable_sum);
1025         P(se.avg.util_sum);
1026         P(se.avg.load_avg);
1027         P(se.avg.runnable_avg);
1028         P(se.avg.util_avg);
1029         P(se.avg.last_update_time);
1030         P(se.avg.util_est.ewma);
1031         PM(se.avg.util_est.enqueued, ~UTIL_AVG_UNCHANGED);
1032 #endif
1033 #ifdef CONFIG_UCLAMP_TASK
1034         __PS("uclamp.min", p->uclamp_req[UCLAMP_MIN].value);
1035         __PS("uclamp.max", p->uclamp_req[UCLAMP_MAX].value);
1036         __PS("effective uclamp.min", uclamp_eff_value(p, UCLAMP_MIN));
1037         __PS("effective uclamp.max", uclamp_eff_value(p, UCLAMP_MAX));
1038 #endif
1039         P(policy);
1040         P(prio);
1041         if (task_has_dl_policy(p)) {
1042                 P(dl.runtime);
1043                 P(dl.deadline);
1044         }
1045 #undef PN_SCHEDSTAT
1046 #undef P_SCHEDSTAT
1047
1048         {
1049                 unsigned int this_cpu = raw_smp_processor_id();
1050                 u64 t0, t1;
1051
1052                 t0 = cpu_clock(this_cpu);
1053                 t1 = cpu_clock(this_cpu);
1054                 __PS("clock-delta", t1-t0);
1055         }
1056
1057         sched_show_numa(p, m);
1058 }
1059
1060 void proc_sched_set_task(struct task_struct *p)
1061 {
1062 #ifdef CONFIG_SCHEDSTATS
1063         memset(&p->se.statistics, 0, sizeof(p->se.statistics));
1064 #endif
1065 }
1066
1067 void resched_latency_warn(int cpu, u64 latency)
1068 {
1069         static DEFINE_RATELIMIT_STATE(latency_check_ratelimit, 60 * 60 * HZ, 1);
1070
1071         WARN(__ratelimit(&latency_check_ratelimit),
1072              "sched: CPU %d need_resched set for > %llu ns (%d ticks) "
1073              "without schedule\n",
1074              cpu, latency, cpu_rq(cpu)->ticks_without_resched);
1075 }