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
 177         if (cnt > 15)
 178                 cnt = 15;
 179
 180         if (copy_from_user(&buf, ubuf, cnt))
 181                 return -EFAULT;
 182
 183         if (kstrtouint(buf, 10, &sysctl_sched_tunable_scaling))
 184                 return -EINVAL;
 185
 186         if (sched_update_scaling())
 187                 return -EINVAL;
 188
 189         *ppos += cnt;
 190         return cnt;
 191 }
 192
 193 static int sched_scaling_show(struct seq_file *m, void *v)
 194 {
 195         seq_printf(m, "%d\n", sysctl_sched_tunable_scaling);
 196         return 0;
 197 }
 198
 199 static int sched_scaling_open(struct inode *inode, struct file *filp)
 200 {
 201         return single_open(filp, sched_scaling_show, NULL);
 202 }
 203
 204 static const struct file_operations sched_scaling_fops = {
 205         .open           = sched_scaling_open,
 206         .write          = sched_scaling_write,
 207         .read           = seq_read,
 208         .llseek         = seq_lseek,
 209         .release        = single_release,
 210 };
 211
 212 #endif /* SMP */
 213
 214 #ifdef CONFIG_PREEMPT_DYNAMIC
 215
 216 static ssize_t sched_dynamic_write(struct file *filp, const char __user *ubuf,
 217                                    size_t cnt, loff_t *ppos)
 218 {
 219         char buf[16];
 220         int mode;
 221
 222         if (cnt > 15)
 223                 cnt = 15;
 224
 225         if (copy_from_user(&buf, ubuf, cnt))
 226                 return -EFAULT;
 227
 228         buf[cnt] = 0;
 229         mode = sched_dynamic_mode(strstrip(buf));
 230         if (mode < 0)
 231                 return mode;
 232
 233         sched_dynamic_update(mode);
 234
 235         *ppos += cnt;
 236
 237         return cnt;
 238 }
 239
 240 static int sched_dynamic_show(struct seq_file *m, void *v)
 241 {
 242         static const char * preempt_modes[] = {
 243                 "none", "voluntary", "full"
 244         };
 245         int i;
 246
 247         for (i = 0; i < ARRAY_SIZE(preempt_modes); i++) {
 248                 if (preempt_dynamic_mode == i)
 249                         seq_puts(m, "(");
 250                 seq_puts(m, preempt_modes[i]);
 251                 if (preempt_dynamic_mode == i)
 252                         seq_puts(m, ")");
 253
 254                 seq_puts(m, " ");
 255         }
 256
 257         seq_puts(m, "\n");
 258         return 0;
 259 }
 260
 261 static int sched_dynamic_open(struct inode *inode, struct file *filp)
 262 {
 263         return single_open(filp, sched_dynamic_show, NULL);
 264 }
 265
 266 static const struct file_operations sched_dynamic_fops = {
 267         .open           = sched_dynamic_open,
 268         .write          = sched_dynamic_write,
 269         .read           = seq_read,
 270         .llseek         = seq_lseek,
 271         .release        = single_release,
 272 };
 273
 274 #endif /* CONFIG_PREEMPT_DYNAMIC */
 275
 276 __read_mostly bool sched_debug_verbose;
 277
 278 static const struct seq_operations sched_debug_sops;
 279
 280 static int sched_debug_open(struct inode *inode, struct file *filp)
 281 {
 282         return seq_open(filp, &sched_debug_sops);
 283 }
 284
 285 static const struct file_operations sched_debug_fops = {
 286         .open           = sched_debug_open,
 287         .read           = seq_read,
 288         .llseek         = seq_lseek,
 289         .release        = seq_release,
 290 };
 291
 292 static struct dentry *debugfs_sched;
 293
 294 static __init int sched_init_debug(void)
 295 {
 296         struct dentry __maybe_unused *numa;
 297
 298         debugfs_sched = debugfs_create_dir("sched", NULL);
 299
 300         debugfs_create_file("features", 0644, debugfs_sched, NULL, &sched_feat_fops);
 301         debugfs_create_bool("verbose", 0644, debugfs_sched, &sched_debug_verbose);
 302 #ifdef CONFIG_PREEMPT_DYNAMIC
 303         debugfs_create_file("preempt", 0644, debugfs_sched, NULL, &sched_dynamic_fops);
 304 #endif
 305
 306         debugfs_create_u32("latency_ns", 0644, debugfs_sched, &sysctl_sched_latency);
 307         debugfs_create_u32("min_granularity_ns", 0644, debugfs_sched, &sysctl_sched_min_granularity);
 308         debugfs_create_u32("wakeup_granularity_ns", 0644, debugfs_sched, &sysctl_sched_wakeup_granularity);
 309
 310         debugfs_create_u32("latency_warn_ms", 0644, debugfs_sched, &sysctl_resched_latency_warn_ms);
 311         debugfs_create_u32("latency_warn_once", 0644, debugfs_sched, &sysctl_resched_latency_warn_once);
 312
 313 #ifdef CONFIG_SMP
 314         debugfs_create_file("tunable_scaling", 0644, debugfs_sched, NULL, &sched_scaling_fops);
 315         debugfs_create_u32("migration_cost_ns", 0644, debugfs_sched, &sysctl_sched_migration_cost);
 316         debugfs_create_u32("nr_migrate", 0644, debugfs_sched, &sysctl_sched_nr_migrate);
 317
 318         mutex_lock(&sched_domains_mutex);
 319         update_sched_domain_debugfs();
 320         mutex_unlock(&sched_domains_mutex);
 321 #endif
 322
 323 #ifdef CONFIG_NUMA_BALANCING
 324         numa = debugfs_create_dir("numa_balancing", debugfs_sched);
 325
 326         debugfs_create_u32("scan_delay_ms", 0644, numa, &sysctl_numa_balancing_scan_delay);
 327         debugfs_create_u32("scan_period_min_ms", 0644, numa, &sysctl_numa_balancing_scan_period_min);
 328         debugfs_create_u32("scan_period_max_ms", 0644, numa, &sysctl_numa_balancing_scan_period_max);
 329         debugfs_create_u32("scan_size_mb", 0644, numa, &sysctl_numa_balancing_scan_size);
 330 #endif
 331
 332         debugfs_create_file("debug", 0444, debugfs_sched, NULL, &sched_debug_fops);
 333
 334         return 0;
 335 }
 336 late_initcall(sched_init_debug);
 337
 338 #ifdef CONFIG_SMP
 339
 340 static cpumask_var_t            sd_sysctl_cpus;
 341 static struct dentry            *sd_dentry;
 342
 343 static int sd_flags_show(struct seq_file *m, void *v)
 344 {
 345         unsigned long flags = *(unsigned int *)m->private;
 346         int idx;
 347
 348         for_each_set_bit(idx, &flags, __SD_FLAG_CNT) {
 349                 seq_puts(m, sd_flag_debug[idx].name);
 350                 seq_puts(m, " ");
 351         }
 352         seq_puts(m, "\n");
 353
 354         return 0;
 355 }
 356
 357 static int sd_flags_open(struct inode *inode, struct file *file)
 358 {
 359         return single_open(file, sd_flags_show, inode->i_private);
 360 }
 361
 362 static const struct file_operations sd_flags_fops = {
 363         .open           = sd_flags_open,
 364         .read           = seq_read,
 365         .llseek         = seq_lseek,
 366         .release        = single_release,
 367 };
 368
 369 static void register_sd(struct sched_domain *sd, struct dentry *parent)
 370 {
 371 #define SDM(type, mode, member) \
 372         debugfs_create_##type(#member, mode, parent, &sd->member)
 373
 374         SDM(ulong, 0644, min_interval);
 375         SDM(ulong, 0644, max_interval);
 376         SDM(u64,   0644, max_newidle_lb_cost);
 377         SDM(u32,   0644, busy_factor);
 378         SDM(u32,   0644, imbalance_pct);
 379         SDM(u32,   0644, cache_nice_tries);
 380         SDM(str,   0444, name);
 381
 382 #undef SDM
 383
 384         debugfs_create_file("flags", 0444, parent, &sd->flags, &sd_flags_fops);
 385 }
 386
 387 void update_sched_domain_debugfs(void)
 388 {
 389         int cpu, i;
 390
 391         if (!cpumask_available(sd_sysctl_cpus)) {
 392                 if (!alloc_cpumask_var(&sd_sysctl_cpus, GFP_KERNEL))
 393                         return;
 394                 cpumask_copy(sd_sysctl_cpus, cpu_possible_mask);
 395         }
 396
 397         if (!sd_dentry)
 398                 sd_dentry = debugfs_create_dir("domains", debugfs_sched);
 399
 400         for_each_cpu(cpu, sd_sysctl_cpus) {
 401                 struct sched_domain *sd;
 402                 struct dentry *d_cpu;
 403                 char buf[32];
 404
 405                 snprintf(buf, sizeof(buf), "cpu%d", cpu);
 406                 debugfs_remove(debugfs_lookup(buf, sd_dentry));
 407                 d_cpu = debugfs_create_dir(buf, sd_dentry);
 408
 409                 i = 0;
 410                 for_each_domain(cpu, sd) {
 411                         struct dentry *d_sd;
 412
 413                         snprintf(buf, sizeof(buf), "domain%d", i);
 414                         d_sd = debugfs_create_dir(buf, d_cpu);
 415
 416                         register_sd(sd, d_sd);
 417                         i++;
 418                 }
 419
 420                 __cpumask_clear_cpu(cpu, sd_sysctl_cpus);
 421         }
 422 }
 423
 424 void dirty_sched_domain_sysctl(int cpu)
 425 {
 426         if (cpumask_available(sd_sysctl_cpus))
 427                 __cpumask_set_cpu(cpu, sd_sysctl_cpus);
 428 }
 429
 430 #endif /* CONFIG_SMP */
 431
 432 #ifdef CONFIG_FAIR_GROUP_SCHED
 433 static void print_cfs_group_stats(struct seq_file *m, int cpu, struct task_group *tg)
 434 {
 435         struct sched_entity *se = tg->se[cpu];
 436
 437 #define P(F)            SEQ_printf(m, "  .%-30s: %lld\n",       #F, (long long)F)
 438 #define P_SCHEDSTAT(F)  SEQ_printf(m, "  .%-30s: %lld\n",       #F, (long long)schedstat_val(F))
 439 #define PN(F)           SEQ_printf(m, "  .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F))
 440 #define PN_SCHEDSTAT(F) SEQ_printf(m, "  .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)schedstat_val(F)))
 441
 442         if (!se)
 443                 return;
 444
 445         PN(se->exec_start);
 446         PN(se->vruntime);
 447         PN(se->sum_exec_runtime);
 448
 449         if (schedstat_enabled()) {
 450                 PN_SCHEDSTAT(se->statistics.wait_start);
 451                 PN_SCHEDSTAT(se->statistics.sleep_start);
 452                 PN_SCHEDSTAT(se->statistics.block_start);
 453                 PN_SCHEDSTAT(se->statistics.sleep_max);
 454                 PN_SCHEDSTAT(se->statistics.block_max);
 455                 PN_SCHEDSTAT(se->statistics.exec_max);
 456                 PN_SCHEDSTAT(se->statistics.slice_max);
 457                 PN_SCHEDSTAT(se->statistics.wait_max);
 458                 PN_SCHEDSTAT(se->statistics.wait_sum);
 459                 P_SCHEDSTAT(se->statistics.wait_count);
 460         }
 461
 462         P(se->load.weight);
 463 #ifdef CONFIG_SMP
 464         P(se->avg.load_avg);
 465         P(se->avg.util_avg);
 466         P(se->avg.runnable_avg);
 467 #endif
 468
 469 #undef PN_SCHEDSTAT
 470 #undef PN
 471 #undef P_SCHEDSTAT
 472 #undef P
 473 }
 474 #endif
 475
 476 #ifdef CONFIG_CGROUP_SCHED
 477 static DEFINE_SPINLOCK(sched_debug_lock);
 478 static char group_path[PATH_MAX];
 479
 480 static void task_group_path(struct task_group *tg, char *path, int plen)
 481 {
 482         if (autogroup_path(tg, path, plen))
 483                 return;
 484
 485         cgroup_path(tg->css.cgroup, path, plen);
 486 }
 487
 488 /*
 489  * Only 1 SEQ_printf_task_group_path() caller can use the full length
 490  * group_path[] for cgroup path. Other simultaneous callers will have
 491  * to use a shorter stack buffer. A "..." suffix is appended at the end
 492  * of the stack buffer so that it will show up in case the output length
 493  * matches the given buffer size to indicate possible path name truncation.
 494  */
 495 #define SEQ_printf_task_group_path(m, tg, fmt...)                       \
 496 {                                                                       \
 497         if (spin_trylock(&sched_debug_lock)) {                          \
 498                 task_group_path(tg, group_path, sizeof(group_path));    \
 499                 SEQ_printf(m, fmt, group_path);                         \
 500                 spin_unlock(&sched_debug_lock);                         \
 501         } else {                                                        \
 502                 char buf[128];                                          \
 503                 char *bufend = buf + sizeof(buf) - 3;                   \
 504                 task_group_path(tg, buf, bufend - buf);                 \
 505                 strcpy(bufend - 1, "...");                              \
 506                 SEQ_printf(m, fmt, buf);                                \
 507         }                                                               \
 508 }
 509 #endif
 510
 511 static void
 512 print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
 513 {
 514         if (task_current(rq, p))
 515                 SEQ_printf(m, ">R");
 516         else
 517                 SEQ_printf(m, " %c", task_state_to_char(p));
 518
 519         SEQ_printf(m, " %15s %5d %9Ld.%06ld %9Ld %5d ",
 520                 p->comm, task_pid_nr(p),
 521                 SPLIT_NS(p->se.vruntime),
 522                 (long long)(p->nvcsw + p->nivcsw),
 523                 p->prio);
 524
 525         SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
 526                 SPLIT_NS(schedstat_val_or_zero(p->se.statistics.wait_sum)),
 527                 SPLIT_NS(p->se.sum_exec_runtime),
 528                 SPLIT_NS(schedstat_val_or_zero(p->se.statistics.sum_sleep_runtime)));
 529
 530 #ifdef CONFIG_NUMA_BALANCING
 531         SEQ_printf(m, " %d %d", task_node(p), task_numa_group_id(p));
 532 #endif
 533 #ifdef CONFIG_CGROUP_SCHED
 534         SEQ_printf_task_group_path(m, task_group(p), " %s")
 535 #endif
 536
 537         SEQ_printf(m, "\n");
 538 }
 539
 540 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
 541 {
 542         struct task_struct *g, *p;
 543
 544         SEQ_printf(m, "\n");
 545         SEQ_printf(m, "runnable tasks:\n");
 546         SEQ_printf(m, " S            task   PID         tree-key  switches  prio"
 547                    "     wait-time             sum-exec        sum-sleep\n");
 548         SEQ_printf(m, "-------------------------------------------------------"
 549                    "------------------------------------------------------\n");
 550
 551         rcu_read_lock();
 552         for_each_process_thread(g, p) {
 553                 if (task_cpu(p) != rq_cpu)
 554                         continue;
 555
 556                 print_task(m, rq, p);
 557         }
 558         rcu_read_unlock();
 559 }
 560
 561 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
 562 {
 563         s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
 564                 spread, rq0_min_vruntime, spread0;
 565         struct rq *rq = cpu_rq(cpu);
 566         struct sched_entity *last;
 567         unsigned long flags;
 568
 569 #ifdef CONFIG_FAIR_GROUP_SCHED
 570         SEQ_printf(m, "\n");
 571         SEQ_printf_task_group_path(m, cfs_rq->tg, "cfs_rq[%d]:%s\n", cpu);
 572 #else
 573         SEQ_printf(m, "\n");
 574         SEQ_printf(m, "cfs_rq[%d]:\n", cpu);
 575 #endif
 576         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "exec_clock",
 577                         SPLIT_NS(cfs_rq->exec_clock));
 578
 579         raw_spin_lock_irqsave(&rq->lock, flags);
 580         if (rb_first_cached(&cfs_rq->tasks_timeline))
 581                 MIN_vruntime = (__pick_first_entity(cfs_rq))->vruntime;
 582         last = __pick_last_entity(cfs_rq);
 583         if (last)
 584                 max_vruntime = last->vruntime;
 585         min_vruntime = cfs_rq->min_vruntime;
 586         rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
 587         raw_spin_unlock_irqrestore(&rq->lock, flags);
 588         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "MIN_vruntime",
 589                         SPLIT_NS(MIN_vruntime));
 590         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "min_vruntime",
 591                         SPLIT_NS(min_vruntime));
 592         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "max_vruntime",
 593                         SPLIT_NS(max_vruntime));
 594         spread = max_vruntime - MIN_vruntime;
 595         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread",
 596                         SPLIT_NS(spread));
 597         spread0 = min_vruntime - rq0_min_vruntime;
 598         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread0",
 599                         SPLIT_NS(spread0));
 600         SEQ_printf(m, "  .%-30s: %d\n", "nr_spread_over",
 601                         cfs_rq->nr_spread_over);
 602         SEQ_printf(m, "  .%-30s: %d\n", "nr_running", cfs_rq->nr_running);
 603         SEQ_printf(m, "  .%-30s: %ld\n", "load", cfs_rq->load.weight);
 604 #ifdef CONFIG_SMP
 605         SEQ_printf(m, "  .%-30s: %lu\n", "load_avg",
 606                         cfs_rq->avg.load_avg);
 607         SEQ_printf(m, "  .%-30s: %lu\n", "runnable_avg",
 608                         cfs_rq->avg.runnable_avg);
 609         SEQ_printf(m, "  .%-30s: %lu\n", "util_avg",
 610                         cfs_rq->avg.util_avg);
 611         SEQ_printf(m, "  .%-30s: %u\n", "util_est_enqueued",
 612                         cfs_rq->avg.util_est.enqueued);
 613         SEQ_printf(m, "  .%-30s: %ld\n", "removed.load_avg",
 614                         cfs_rq->removed.load_avg);
 615         SEQ_printf(m, "  .%-30s: %ld\n", "removed.util_avg",
 616                         cfs_rq->removed.util_avg);
 617         SEQ_printf(m, "  .%-30s: %ld\n", "removed.runnable_avg",
 618                         cfs_rq->removed.runnable_avg);
 619 #ifdef CONFIG_FAIR_GROUP_SCHED
 620         SEQ_printf(m, "  .%-30s: %lu\n", "tg_load_avg_contrib",
 621                         cfs_rq->tg_load_avg_contrib);
 622         SEQ_printf(m, "  .%-30s: %ld\n", "tg_load_avg",
 623                         atomic_long_read(&cfs_rq->tg->load_avg));
 624 #endif
 625 #endif
 626 #ifdef CONFIG_CFS_BANDWIDTH
 627         SEQ_printf(m, "  .%-30s: %d\n", "throttled",
 628                         cfs_rq->throttled);
 629         SEQ_printf(m, "  .%-30s: %d\n", "throttle_count",
 630                         cfs_rq->throttle_count);
 631 #endif
 632
 633 #ifdef CONFIG_FAIR_GROUP_SCHED
 634         print_cfs_group_stats(m, cpu, cfs_rq->tg);
 635 #endif
 636 }
 637
 638 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
 639 {
 640 #ifdef CONFIG_RT_GROUP_SCHED
 641         SEQ_printf(m, "\n");
 642         SEQ_printf_task_group_path(m, rt_rq->tg, "rt_rq[%d]:%s\n", cpu);
 643 #else
 644         SEQ_printf(m, "\n");
 645         SEQ_printf(m, "rt_rq[%d]:\n", cpu);
 646 #endif
 647
 648 #define P(x) \
 649         SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
 650 #define PU(x) \
 651         SEQ_printf(m, "  .%-30s: %lu\n", #x, (unsigned long)(rt_rq->x))
 652 #define PN(x) \
 653         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
 654
 655         PU(rt_nr_running);
 656 #ifdef CONFIG_SMP
 657         PU(rt_nr_migratory);
 658 #endif
 659         P(rt_throttled);
 660         PN(rt_time);
 661         PN(rt_runtime);
 662
 663 #undef PN
 664 #undef PU
 665 #undef P
 666 }
 667
 668 void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq)
 669 {
 670         struct dl_bw *dl_bw;
 671
 672         SEQ_printf(m, "\n");
 673         SEQ_printf(m, "dl_rq[%d]:\n", cpu);
 674
 675 #define PU(x) \
 676         SEQ_printf(m, "  .%-30s: %lu\n", #x, (unsigned long)(dl_rq->x))
 677
 678         PU(dl_nr_running);
 679 #ifdef CONFIG_SMP
 680         PU(dl_nr_migratory);
 681         dl_bw = &cpu_rq(cpu)->rd->dl_bw;
 682 #else
 683         dl_bw = &dl_rq->dl_bw;
 684 #endif
 685         SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->bw", dl_bw->bw);
 686         SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->total_bw", dl_bw->total_bw);
 687
 688 #undef PU
 689 }
 690
 691 static void print_cpu(struct seq_file *m, int cpu)
 692 {
 693         struct rq *rq = cpu_rq(cpu);
 694
 695 #ifdef CONFIG_X86
 696         {
 697                 unsigned int freq = cpu_khz ? : 1;
 698
 699                 SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
 700                            cpu, freq / 1000, (freq % 1000));
 701         }
 702 #else
 703         SEQ_printf(m, "cpu#%d\n", cpu);
 704 #endif
 705
 706 #define P(x)                                                            \
 707 do {                                                                    \
 708         if (sizeof(rq->x) == 4)                                         \
 709                 SEQ_printf(m, "  .%-30s: %ld\n", #x, (long)(rq->x));    \
 710         else                                                            \
 711                 SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rq->x));\
 712 } while (0)
 713
 714 #define PN(x) \
 715         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
 716
 717         P(nr_running);
 718         P(nr_switches);
 719         P(nr_uninterruptible);
 720         PN(next_balance);
 721         SEQ_printf(m, "  .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr)));
 722         PN(clock);
 723         PN(clock_task);
 724 #undef P
 725 #undef PN
 726
 727 #ifdef CONFIG_SMP
 728 #define P64(n) SEQ_printf(m, "  .%-30s: %Ld\n", #n, rq->n);
 729         P64(avg_idle);
 730         P64(max_idle_balance_cost);
 731 #undef P64
 732 #endif
 733
 734 #define P(n) SEQ_printf(m, "  .%-30s: %d\n", #n, schedstat_val(rq->n));
 735         if (schedstat_enabled()) {
 736                 P(yld_count);
 737                 P(sched_count);
 738                 P(sched_goidle);
 739                 P(ttwu_count);
 740                 P(ttwu_local);
 741         }
 742 #undef P
 743
 744         print_cfs_stats(m, cpu);
 745         print_rt_stats(m, cpu);
 746         print_dl_stats(m, cpu);
 747
 748         print_rq(m, rq, cpu);
 749         SEQ_printf(m, "\n");
 750 }
 751
 752 static const char *sched_tunable_scaling_names[] = {
 753         "none",
 754         "logarithmic",
 755         "linear"
 756 };
 757
 758 static void sched_debug_header(struct seq_file *m)
 759 {
 760         u64 ktime, sched_clk, cpu_clk;
 761         unsigned long flags;
 762
 763         local_irq_save(flags);
 764         ktime = ktime_to_ns(ktime_get());
 765         sched_clk = sched_clock();
 766         cpu_clk = local_clock();
 767         local_irq_restore(flags);
 768
 769         SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n",
 770                 init_utsname()->release,
 771                 (int)strcspn(init_utsname()->version, " "),
 772                 init_utsname()->version);
 773
 774 #define P(x) \
 775         SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
 776 #define PN(x) \
 777         SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
 778         PN(ktime);
 779         PN(sched_clk);
 780         PN(cpu_clk);
 781         P(jiffies);
 782 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
 783         P(sched_clock_stable());
 784 #endif
 785 #undef PN
 786 #undef P
 787
 788         SEQ_printf(m, "\n");
 789         SEQ_printf(m, "sysctl_sched\n");
 790
 791 #define P(x) \
 792         SEQ_printf(m, "  .%-40s: %Ld\n", #x, (long long)(x))
 793 #define PN(x) \
 794         SEQ_printf(m, "  .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
 795         PN(sysctl_sched_latency);
 796         PN(sysctl_sched_min_granularity);
 797         PN(sysctl_sched_wakeup_granularity);
 798         P(sysctl_sched_child_runs_first);
 799         P(sysctl_sched_features);
 800 #undef PN
 801 #undef P
 802
 803         SEQ_printf(m, "  .%-40s: %d (%s)\n",
 804                 "sysctl_sched_tunable_scaling",
 805                 sysctl_sched_tunable_scaling,
 806                 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
 807         SEQ_printf(m, "\n");
 808 }
 809
 810 static int sched_debug_show(struct seq_file *m, void *v)
 811 {
 812         int cpu = (unsigned long)(v - 2);
 813
 814         if (cpu != -1)
 815                 print_cpu(m, cpu);
 816         else
 817                 sched_debug_header(m);
 818
 819         return 0;
 820 }
 821
 822 void sysrq_sched_debug_show(void)
 823 {
 824         int cpu;
 825
 826         sched_debug_header(NULL);
 827         for_each_online_cpu(cpu) {
 828                 /*
 829                  * Need to reset softlockup watchdogs on all CPUs, because
 830                  * another CPU might be blocked waiting for us to process
 831                  * an IPI or stop_machine.
 832                  */
 833                 touch_nmi_watchdog();
 834                 touch_all_softlockup_watchdogs();
 835                 print_cpu(NULL, cpu);
 836         }
 837 }
 838
 839 /*
 840  * This iterator needs some explanation.
 841  * It returns 1 for the header position.
 842  * This means 2 is CPU 0.
 843  * In a hotplugged system some CPUs, including CPU 0, may be missing so we have
 844  * to use cpumask_* to iterate over the CPUs.
 845  */
 846 static void *sched_debug_start(struct seq_file *file, loff_t *offset)
 847 {
 848         unsigned long n = *offset;
 849
 850         if (n == 0)
 851                 return (void *) 1;
 852
 853         n--;
 854
 855         if (n > 0)
 856                 n = cpumask_next(n - 1, cpu_online_mask);
 857         else
 858                 n = cpumask_first(cpu_online_mask);
 859
 860         *offset = n + 1;
 861
 862         if (n < nr_cpu_ids)
 863                 return (void *)(unsigned long)(n + 2);
 864
 865         return NULL;
 866 }
 867
 868 static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
 869 {
 870         (*offset)++;
 871         return sched_debug_start(file, offset);
 872 }
 873
 874 static void sched_debug_stop(struct seq_file *file, void *data)
 875 {
 876 }
 877
 878 static const struct seq_operations sched_debug_sops = {
 879         .start          = sched_debug_start,
 880         .next           = sched_debug_next,
 881         .stop           = sched_debug_stop,
 882         .show           = sched_debug_show,
 883 };
 884
 885 #define __PS(S, F) SEQ_printf(m, "%-45s:%21Ld\n", S, (long long)(F))
 886 #define __P(F) __PS(#F, F)
 887 #define   P(F) __PS(#F, p->F)
 888 #define   PM(F, M) __PS(#F, p->F & (M))
 889 #define __PSN(S, F) SEQ_printf(m, "%-45s:%14Ld.%06ld\n", S, SPLIT_NS((long long)(F)))
 890 #define __PN(F) __PSN(#F, F)
 891 #define   PN(F) __PSN(#F, p->F)
 892
 893
 894 #ifdef CONFIG_NUMA_BALANCING
 895 void print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
 896                 unsigned long tpf, unsigned long gsf, unsigned long gpf)
 897 {
 898         SEQ_printf(m, "numa_faults node=%d ", node);
 899         SEQ_printf(m, "task_private=%lu task_shared=%lu ", tpf, tsf);
 900         SEQ_printf(m, "group_private=%lu group_shared=%lu\n", gpf, gsf);
 901 }
 902 #endif
 903
 904
 905 static void sched_show_numa(struct task_struct *p, struct seq_file *m)
 906 {
 907 #ifdef CONFIG_NUMA_BALANCING
 908         struct mempolicy *pol;
 909
 910         if (p->mm)
 911                 P(mm->numa_scan_seq);
 912
 913         task_lock(p);
 914         pol = p->mempolicy;
 915         if (pol && !(pol->flags & MPOL_F_MORON))
 916                 pol = NULL;
 917         mpol_get(pol);
 918         task_unlock(p);
 919
 920         P(numa_pages_migrated);
 921         P(numa_preferred_nid);
 922         P(total_numa_faults);
 923         SEQ_printf(m, "current_node=%d, numa_group_id=%d\n",
 924                         task_node(p), task_numa_group_id(p));
 925         show_numa_stats(p, m);
 926         mpol_put(pol);
 927 #endif
 928 }
 929
 930 void proc_sched_show_task(struct task_struct *p, struct pid_namespace *ns,
 931                                                   struct seq_file *m)
 932 {
 933         unsigned long nr_switches;
 934
 935         SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr_ns(p, ns),
 936                                                 get_nr_threads(p));
 937         SEQ_printf(m,
 938                 "---------------------------------------------------------"
 939                 "----------\n");
 940
 941 #define P_SCHEDSTAT(F)  __PS(#F, schedstat_val(p->F))
 942 #define PN_SCHEDSTAT(F) __PSN(#F, schedstat_val(p->F))
 943
 944         PN(se.exec_start);
 945         PN(se.vruntime);
 946         PN(se.sum_exec_runtime);
 947
 948         nr_switches = p->nvcsw + p->nivcsw;
 949
 950         P(se.nr_migrations);
 951
 952         if (schedstat_enabled()) {
 953                 u64 avg_atom, avg_per_cpu;
 954
 955                 PN_SCHEDSTAT(se.statistics.sum_sleep_runtime);
 956                 PN_SCHEDSTAT(se.statistics.wait_start);
 957                 PN_SCHEDSTAT(se.statistics.sleep_start);
 958                 PN_SCHEDSTAT(se.statistics.block_start);
 959                 PN_SCHEDSTAT(se.statistics.sleep_max);
 960                 PN_SCHEDSTAT(se.statistics.block_max);
 961                 PN_SCHEDSTAT(se.statistics.exec_max);
 962                 PN_SCHEDSTAT(se.statistics.slice_max);
 963                 PN_SCHEDSTAT(se.statistics.wait_max);
 964                 PN_SCHEDSTAT(se.statistics.wait_sum);
 965                 P_SCHEDSTAT(se.statistics.wait_count);
 966                 PN_SCHEDSTAT(se.statistics.iowait_sum);
 967                 P_SCHEDSTAT(se.statistics.iowait_count);
 968                 P_SCHEDSTAT(se.statistics.nr_migrations_cold);
 969                 P_SCHEDSTAT(se.statistics.nr_failed_migrations_affine);
 970                 P_SCHEDSTAT(se.statistics.nr_failed_migrations_running);
 971                 P_SCHEDSTAT(se.statistics.nr_failed_migrations_hot);
 972                 P_SCHEDSTAT(se.statistics.nr_forced_migrations);
 973                 P_SCHEDSTAT(se.statistics.nr_wakeups);
 974                 P_SCHEDSTAT(se.statistics.nr_wakeups_sync);
 975                 P_SCHEDSTAT(se.statistics.nr_wakeups_migrate);
 976                 P_SCHEDSTAT(se.statistics.nr_wakeups_local);
 977                 P_SCHEDSTAT(se.statistics.nr_wakeups_remote);
 978                 P_SCHEDSTAT(se.statistics.nr_wakeups_affine);
 979                 P_SCHEDSTAT(se.statistics.nr_wakeups_affine_attempts);
 980                 P_SCHEDSTAT(se.statistics.nr_wakeups_passive);
 981                 P_SCHEDSTAT(se.statistics.nr_wakeups_idle);
 982
 983                 avg_atom = p->se.sum_exec_runtime;
 984                 if (nr_switches)
 985                         avg_atom = div64_ul(avg_atom, nr_switches);
 986                 else
 987                         avg_atom = -1LL;
 988
 989                 avg_per_cpu = p->se.sum_exec_runtime;
 990                 if (p->se.nr_migrations) {
 991                         avg_per_cpu = div64_u64(avg_per_cpu,
 992                                                 p->se.nr_migrations);
 993                 } else {
 994                         avg_per_cpu = -1LL;
 995                 }
 996
 997                 __PN(avg_atom);
 998                 __PN(avg_per_cpu);
 999         }
1000
1001         __P(nr_switches);
1002         __PS("nr_voluntary_switches", p->nvcsw);
1003         __PS("nr_involuntary_switches", p->nivcsw);
1004
1005         P(se.load.weight);
1006 #ifdef CONFIG_SMP
1007         P(se.avg.load_sum);
1008         P(se.avg.runnable_sum);
1009         P(se.avg.util_sum);
1010         P(se.avg.load_avg);
1011         P(se.avg.runnable_avg);
1012         P(se.avg.util_avg);
1013         P(se.avg.last_update_time);
1014         P(se.avg.util_est.ewma);
1015         PM(se.avg.util_est.enqueued, ~UTIL_AVG_UNCHANGED);
1016 #endif
1017 #ifdef CONFIG_UCLAMP_TASK
1018         __PS("uclamp.min", p->uclamp_req[UCLAMP_MIN].value);
1019         __PS("uclamp.max", p->uclamp_req[UCLAMP_MAX].value);
1020         __PS("effective uclamp.min", uclamp_eff_value(p, UCLAMP_MIN));
1021         __PS("effective uclamp.max", uclamp_eff_value(p, UCLAMP_MAX));
1022 #endif
1023         P(policy);
1024         P(prio);
1025         if (task_has_dl_policy(p)) {
1026                 P(dl.runtime);
1027                 P(dl.deadline);
1028         }
1029 #undef PN_SCHEDSTAT
1030 #undef P_SCHEDSTAT
1031
1032         {
1033                 unsigned int this_cpu = raw_smp_processor_id();
1034                 u64 t0, t1;
1035
1036                 t0 = cpu_clock(this_cpu);
1037                 t1 = cpu_clock(this_cpu);
1038                 __PS("clock-delta", t1-t0);
1039         }
1040
1041         sched_show_numa(p, m);
1042 }
1043
1044 void proc_sched_set_task(struct task_struct *p)
1045 {
1046 #ifdef CONFIG_SCHEDSTATS
1047         memset(&p->se.statistics, 0, sizeof(p->se.statistics));
1048 #endif
1049 }
1050
1051 void resched_latency_warn(int cpu, u64 latency)
1052 {
1053         static DEFINE_RATELIMIT_STATE(latency_check_ratelimit, 60 * 60 * HZ, 1);
1054
1055         WARN(__ratelimit(&latency_check_ratelimit),
1056              "sched: CPU %d need_resched set for > %llu ns (%d ticks) "
1057              "without schedule\n",
1058              cpu, latency, cpu_rq(cpu)->ticks_without_resched);
1059 }