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 #endif
 303
 304 #ifdef CONFIG_NUMA_BALANCING
 305         numa = debugfs_create_dir("numa_balancing", debugfs_sched);
 306
 307         debugfs_create_u32("scan_delay_ms", 0644, numa, &sysctl_numa_balancing_scan_delay);
 308         debugfs_create_u32("scan_period_min_ms", 0644, numa, &sysctl_numa_balancing_scan_period_min);
 309         debugfs_create_u32("scan_period_max_ms", 0644, numa, &sysctl_numa_balancing_scan_period_max);
 310         debugfs_create_u32("scan_size_mb", 0644, numa, &sysctl_numa_balancing_scan_size);
 311 #endif
 312
 313         return 0;
 314 }
 315 late_initcall(sched_init_debug);
 316
 317 #ifdef CONFIG_SMP
 318
 319 #ifdef CONFIG_SYSCTL
 320
 321 static struct ctl_table sd_ctl_dir[] = {
 322         {
 323                 .procname       = "sched_domain",
 324                 .mode           = 0555,
 325         },
 326         {}
 327 };
 328
 329 static struct ctl_table sd_ctl_root[] = {
 330         {
 331                 .procname       = "kernel",
 332                 .mode           = 0555,
 333                 .child          = sd_ctl_dir,
 334         },
 335         {}
 336 };
 337
 338 static struct ctl_table *sd_alloc_ctl_entry(int n)
 339 {
 340         struct ctl_table *entry =
 341                 kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL);
 342
 343         return entry;
 344 }
 345
 346 static void sd_free_ctl_entry(struct ctl_table **tablep)
 347 {
 348         struct ctl_table *entry;
 349
 350         /*
 351          * In the intermediate directories, both the child directory and
 352          * procname are dynamically allocated and could fail but the mode
 353          * will always be set. In the lowest directory the names are
 354          * static strings and all have proc handlers.
 355          */
 356         for (entry = *tablep; entry->mode; entry++) {
 357                 if (entry->child)
 358                         sd_free_ctl_entry(&entry->child);
 359                 if (entry->proc_handler == NULL)
 360                         kfree(entry->procname);
 361         }
 362
 363         kfree(*tablep);
 364         *tablep = NULL;
 365 }
 366
 367 static void
 368 set_table_entry(struct ctl_table *entry,
 369                 const char *procname, void *data, int maxlen,
 370                 umode_t mode, proc_handler *proc_handler)
 371 {
 372         entry->procname = procname;
 373         entry->data = data;
 374         entry->maxlen = maxlen;
 375         entry->mode = mode;
 376         entry->proc_handler = proc_handler;
 377 }
 378
 379 static int sd_ctl_doflags(struct ctl_table *table, int write,
 380                           void *buffer, size_t *lenp, loff_t *ppos)
 381 {
 382         unsigned long flags = *(unsigned long *)table->data;
 383         size_t data_size = 0;
 384         size_t len = 0;
 385         char *tmp, *buf;
 386         int idx;
 387
 388         if (write)
 389                 return 0;
 390
 391         for_each_set_bit(idx, &flags, __SD_FLAG_CNT) {
 392                 char *name = sd_flag_debug[idx].name;
 393
 394                 /* Name plus whitespace */
 395                 data_size += strlen(name) + 1;
 396         }
 397
 398         if (*ppos > data_size) {
 399                 *lenp = 0;
 400                 return 0;
 401         }
 402
 403         buf = kcalloc(data_size + 1, sizeof(*buf), GFP_KERNEL);
 404         if (!buf)
 405                 return -ENOMEM;
 406
 407         for_each_set_bit(idx, &flags, __SD_FLAG_CNT) {
 408                 char *name = sd_flag_debug[idx].name;
 409
 410                 len += snprintf(buf + len, strlen(name) + 2, "%s ", name);
 411         }
 412
 413         tmp = buf + *ppos;
 414         len -= *ppos;
 415
 416         if (len > *lenp)
 417                 len = *lenp;
 418         if (len)
 419                 memcpy(buffer, tmp, len);
 420         if (len < *lenp) {
 421                 ((char *)buffer)[len] = '\n';
 422                 len++;
 423         }
 424
 425         *lenp = len;
 426         *ppos += len;
 427
 428         kfree(buf);
 429
 430         return 0;
 431 }
 432
 433 static struct ctl_table *
 434 sd_alloc_ctl_domain_table(struct sched_domain *sd)
 435 {
 436         struct ctl_table *table = sd_alloc_ctl_entry(9);
 437
 438         if (table == NULL)
 439                 return NULL;
 440
 441         set_table_entry(&table[0], "min_interval",        &sd->min_interval,        sizeof(long), 0644, proc_doulongvec_minmax);
 442         set_table_entry(&table[1], "max_interval",        &sd->max_interval,        sizeof(long), 0644, proc_doulongvec_minmax);
 443         set_table_entry(&table[2], "busy_factor",         &sd->busy_factor,         sizeof(int),  0644, proc_dointvec_minmax);
 444         set_table_entry(&table[3], "imbalance_pct",       &sd->imbalance_pct,       sizeof(int),  0644, proc_dointvec_minmax);
 445         set_table_entry(&table[4], "cache_nice_tries",    &sd->cache_nice_tries,    sizeof(int),  0644, proc_dointvec_minmax);
 446         set_table_entry(&table[5], "flags",               &sd->flags,               sizeof(int),  0444, sd_ctl_doflags);
 447         set_table_entry(&table[6], "max_newidle_lb_cost", &sd->max_newidle_lb_cost, sizeof(long), 0644, proc_doulongvec_minmax);
 448         set_table_entry(&table[7], "name",                sd->name,            CORENAME_MAX_SIZE, 0444, proc_dostring);
 449         /* &table[8] is terminator */
 450
 451         return table;
 452 }
 453
 454 static struct ctl_table *sd_alloc_ctl_cpu_table(int cpu)
 455 {
 456         struct ctl_table *entry, *table;
 457         struct sched_domain *sd;
 458         int domain_num = 0, i;
 459         char buf[32];
 460
 461         for_each_domain(cpu, sd)
 462                 domain_num++;
 463         entry = table = sd_alloc_ctl_entry(domain_num + 1);
 464         if (table == NULL)
 465                 return NULL;
 466
 467         i = 0;
 468         for_each_domain(cpu, sd) {
 469                 snprintf(buf, 32, "domain%d", i);
 470                 entry->procname = kstrdup(buf, GFP_KERNEL);
 471                 entry->mode = 0555;
 472                 entry->child = sd_alloc_ctl_domain_table(sd);
 473                 entry++;
 474                 i++;
 475         }
 476         return table;
 477 }
 478
 479 static cpumask_var_t            sd_sysctl_cpus;
 480 static struct ctl_table_header  *sd_sysctl_header;
 481
 482 void register_sched_domain_sysctl(void)
 483 {
 484         static struct ctl_table *cpu_entries;
 485         static struct ctl_table **cpu_idx;
 486         static bool init_done = false;
 487         char buf[32];
 488         int i;
 489
 490         if (!cpu_entries) {
 491                 cpu_entries = sd_alloc_ctl_entry(num_possible_cpus() + 1);
 492                 if (!cpu_entries)
 493                         return;
 494
 495                 WARN_ON(sd_ctl_dir[0].child);
 496                 sd_ctl_dir[0].child = cpu_entries;
 497         }
 498
 499         if (!cpu_idx) {
 500                 struct ctl_table *e = cpu_entries;
 501
 502                 cpu_idx = kcalloc(nr_cpu_ids, sizeof(struct ctl_table*), GFP_KERNEL);
 503                 if (!cpu_idx)
 504                         return;
 505
 506                 /* deal with sparse possible map */
 507                 for_each_possible_cpu(i) {
 508                         cpu_idx[i] = e;
 509                         e++;
 510                 }
 511         }
 512
 513         if (!cpumask_available(sd_sysctl_cpus)) {
 514                 if (!alloc_cpumask_var(&sd_sysctl_cpus, GFP_KERNEL))
 515                         return;
 516         }
 517
 518         if (!init_done) {
 519                 init_done = true;
 520                 /* init to possible to not have holes in @cpu_entries */
 521                 cpumask_copy(sd_sysctl_cpus, cpu_possible_mask);
 522         }
 523
 524         for_each_cpu(i, sd_sysctl_cpus) {
 525                 struct ctl_table *e = cpu_idx[i];
 526
 527                 if (e->child)
 528                         sd_free_ctl_entry(&e->child);
 529
 530                 if (!e->procname) {
 531                         snprintf(buf, 32, "cpu%d", i);
 532                         e->procname = kstrdup(buf, GFP_KERNEL);
 533                 }
 534                 e->mode = 0555;
 535                 e->child = sd_alloc_ctl_cpu_table(i);
 536
 537                 __cpumask_clear_cpu(i, sd_sysctl_cpus);
 538         }
 539
 540         WARN_ON(sd_sysctl_header);
 541         sd_sysctl_header = register_sysctl_table(sd_ctl_root);
 542 }
 543
 544 void dirty_sched_domain_sysctl(int cpu)
 545 {
 546         if (cpumask_available(sd_sysctl_cpus))
 547                 __cpumask_set_cpu(cpu, sd_sysctl_cpus);
 548 }
 549
 550 /* may be called multiple times per register */
 551 void unregister_sched_domain_sysctl(void)
 552 {
 553         unregister_sysctl_table(sd_sysctl_header);
 554         sd_sysctl_header = NULL;
 555 }
 556 #endif /* CONFIG_SYSCTL */
 557 #endif /* CONFIG_SMP */
 558
 559 #ifdef CONFIG_FAIR_GROUP_SCHED
 560 static void print_cfs_group_stats(struct seq_file *m, int cpu, struct task_group *tg)
 561 {
 562         struct sched_entity *se = tg->se[cpu];
 563
 564 #define P(F)            SEQ_printf(m, "  .%-30s: %lld\n",       #F, (long long)F)
 565 #define P_SCHEDSTAT(F)  SEQ_printf(m, "  .%-30s: %lld\n",       #F, (long long)schedstat_val(F))
 566 #define PN(F)           SEQ_printf(m, "  .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F))
 567 #define PN_SCHEDSTAT(F) SEQ_printf(m, "  .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)schedstat_val(F)))
 568
 569         if (!se)
 570                 return;
 571
 572         PN(se->exec_start);
 573         PN(se->vruntime);
 574         PN(se->sum_exec_runtime);
 575
 576         if (schedstat_enabled()) {
 577                 PN_SCHEDSTAT(se->statistics.wait_start);
 578                 PN_SCHEDSTAT(se->statistics.sleep_start);
 579                 PN_SCHEDSTAT(se->statistics.block_start);
 580                 PN_SCHEDSTAT(se->statistics.sleep_max);
 581                 PN_SCHEDSTAT(se->statistics.block_max);
 582                 PN_SCHEDSTAT(se->statistics.exec_max);
 583                 PN_SCHEDSTAT(se->statistics.slice_max);
 584                 PN_SCHEDSTAT(se->statistics.wait_max);
 585                 PN_SCHEDSTAT(se->statistics.wait_sum);
 586                 P_SCHEDSTAT(se->statistics.wait_count);
 587         }
 588
 589         P(se->load.weight);
 590 #ifdef CONFIG_SMP
 591         P(se->avg.load_avg);
 592         P(se->avg.util_avg);
 593         P(se->avg.runnable_avg);
 594 #endif
 595
 596 #undef PN_SCHEDSTAT
 597 #undef PN
 598 #undef P_SCHEDSTAT
 599 #undef P
 600 }
 601 #endif
 602
 603 #ifdef CONFIG_CGROUP_SCHED
 604 static char group_path[PATH_MAX];
 605
 606 static char *task_group_path(struct task_group *tg)
 607 {
 608         if (autogroup_path(tg, group_path, PATH_MAX))
 609                 return group_path;
 610
 611         cgroup_path(tg->css.cgroup, group_path, PATH_MAX);
 612
 613         return group_path;
 614 }
 615 #endif
 616
 617 static void
 618 print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
 619 {
 620         if (task_current(rq, p))
 621                 SEQ_printf(m, ">R");
 622         else
 623                 SEQ_printf(m, " %c", task_state_to_char(p));
 624
 625         SEQ_printf(m, " %15s %5d %9Ld.%06ld %9Ld %5d ",
 626                 p->comm, task_pid_nr(p),
 627                 SPLIT_NS(p->se.vruntime),
 628                 (long long)(p->nvcsw + p->nivcsw),
 629                 p->prio);
 630
 631         SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
 632                 SPLIT_NS(schedstat_val_or_zero(p->se.statistics.wait_sum)),
 633                 SPLIT_NS(p->se.sum_exec_runtime),
 634                 SPLIT_NS(schedstat_val_or_zero(p->se.statistics.sum_sleep_runtime)));
 635
 636 #ifdef CONFIG_NUMA_BALANCING
 637         SEQ_printf(m, " %d %d", task_node(p), task_numa_group_id(p));
 638 #endif
 639 #ifdef CONFIG_CGROUP_SCHED
 640         SEQ_printf(m, " %s", task_group_path(task_group(p)));
 641 #endif
 642
 643         SEQ_printf(m, "\n");
 644 }
 645
 646 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
 647 {
 648         struct task_struct *g, *p;
 649
 650         SEQ_printf(m, "\n");
 651         SEQ_printf(m, "runnable tasks:\n");
 652         SEQ_printf(m, " S            task   PID         tree-key  switches  prio"
 653                    "     wait-time             sum-exec        sum-sleep\n");
 654         SEQ_printf(m, "-------------------------------------------------------"
 655                    "------------------------------------------------------\n");
 656
 657         rcu_read_lock();
 658         for_each_process_thread(g, p) {
 659                 if (task_cpu(p) != rq_cpu)
 660                         continue;
 661
 662                 print_task(m, rq, p);
 663         }
 664         rcu_read_unlock();
 665 }
 666
 667 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
 668 {
 669         s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
 670                 spread, rq0_min_vruntime, spread0;
 671         struct rq *rq = cpu_rq(cpu);
 672         struct sched_entity *last;
 673         unsigned long flags;
 674
 675 #ifdef CONFIG_FAIR_GROUP_SCHED
 676         SEQ_printf(m, "\n");
 677         SEQ_printf(m, "cfs_rq[%d]:%s\n", cpu, task_group_path(cfs_rq->tg));
 678 #else
 679         SEQ_printf(m, "\n");
 680         SEQ_printf(m, "cfs_rq[%d]:\n", cpu);
 681 #endif
 682         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "exec_clock",
 683                         SPLIT_NS(cfs_rq->exec_clock));
 684
 685         raw_spin_lock_irqsave(&rq->lock, flags);
 686         if (rb_first_cached(&cfs_rq->tasks_timeline))
 687                 MIN_vruntime = (__pick_first_entity(cfs_rq))->vruntime;
 688         last = __pick_last_entity(cfs_rq);
 689         if (last)
 690                 max_vruntime = last->vruntime;
 691         min_vruntime = cfs_rq->min_vruntime;
 692         rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
 693         raw_spin_unlock_irqrestore(&rq->lock, flags);
 694         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "MIN_vruntime",
 695                         SPLIT_NS(MIN_vruntime));
 696         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "min_vruntime",
 697                         SPLIT_NS(min_vruntime));
 698         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "max_vruntime",
 699                         SPLIT_NS(max_vruntime));
 700         spread = max_vruntime - MIN_vruntime;
 701         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread",
 702                         SPLIT_NS(spread));
 703         spread0 = min_vruntime - rq0_min_vruntime;
 704         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread0",
 705                         SPLIT_NS(spread0));
 706         SEQ_printf(m, "  .%-30s: %d\n", "nr_spread_over",
 707                         cfs_rq->nr_spread_over);
 708         SEQ_printf(m, "  .%-30s: %d\n", "nr_running", cfs_rq->nr_running);
 709         SEQ_printf(m, "  .%-30s: %ld\n", "load", cfs_rq->load.weight);
 710 #ifdef CONFIG_SMP
 711         SEQ_printf(m, "  .%-30s: %lu\n", "load_avg",
 712                         cfs_rq->avg.load_avg);
 713         SEQ_printf(m, "  .%-30s: %lu\n", "runnable_avg",
 714                         cfs_rq->avg.runnable_avg);
 715         SEQ_printf(m, "  .%-30s: %lu\n", "util_avg",
 716                         cfs_rq->avg.util_avg);
 717         SEQ_printf(m, "  .%-30s: %u\n", "util_est_enqueued",
 718                         cfs_rq->avg.util_est.enqueued);
 719         SEQ_printf(m, "  .%-30s: %ld\n", "removed.load_avg",
 720                         cfs_rq->removed.load_avg);
 721         SEQ_printf(m, "  .%-30s: %ld\n", "removed.util_avg",
 722                         cfs_rq->removed.util_avg);
 723         SEQ_printf(m, "  .%-30s: %ld\n", "removed.runnable_avg",
 724                         cfs_rq->removed.runnable_avg);
 725 #ifdef CONFIG_FAIR_GROUP_SCHED
 726         SEQ_printf(m, "  .%-30s: %lu\n", "tg_load_avg_contrib",
 727                         cfs_rq->tg_load_avg_contrib);
 728         SEQ_printf(m, "  .%-30s: %ld\n", "tg_load_avg",
 729                         atomic_long_read(&cfs_rq->tg->load_avg));
 730 #endif
 731 #endif
 732 #ifdef CONFIG_CFS_BANDWIDTH
 733         SEQ_printf(m, "  .%-30s: %d\n", "throttled",
 734                         cfs_rq->throttled);
 735         SEQ_printf(m, "  .%-30s: %d\n", "throttle_count",
 736                         cfs_rq->throttle_count);
 737 #endif
 738
 739 #ifdef CONFIG_FAIR_GROUP_SCHED
 740         print_cfs_group_stats(m, cpu, cfs_rq->tg);
 741 #endif
 742 }
 743
 744 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
 745 {
 746 #ifdef CONFIG_RT_GROUP_SCHED
 747         SEQ_printf(m, "\n");
 748         SEQ_printf(m, "rt_rq[%d]:%s\n", cpu, task_group_path(rt_rq->tg));
 749 #else
 750         SEQ_printf(m, "\n");
 751         SEQ_printf(m, "rt_rq[%d]:\n", cpu);
 752 #endif
 753
 754 #define P(x) \
 755         SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
 756 #define PU(x) \
 757         SEQ_printf(m, "  .%-30s: %lu\n", #x, (unsigned long)(rt_rq->x))
 758 #define PN(x) \
 759         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
 760
 761         PU(rt_nr_running);
 762 #ifdef CONFIG_SMP
 763         PU(rt_nr_migratory);
 764 #endif
 765         P(rt_throttled);
 766         PN(rt_time);
 767         PN(rt_runtime);
 768
 769 #undef PN
 770 #undef PU
 771 #undef P
 772 }
 773
 774 void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq)
 775 {
 776         struct dl_bw *dl_bw;
 777
 778         SEQ_printf(m, "\n");
 779         SEQ_printf(m, "dl_rq[%d]:\n", cpu);
 780
 781 #define PU(x) \
 782         SEQ_printf(m, "  .%-30s: %lu\n", #x, (unsigned long)(dl_rq->x))
 783
 784         PU(dl_nr_running);
 785 #ifdef CONFIG_SMP
 786         PU(dl_nr_migratory);
 787         dl_bw = &cpu_rq(cpu)->rd->dl_bw;
 788 #else
 789         dl_bw = &dl_rq->dl_bw;
 790 #endif
 791         SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->bw", dl_bw->bw);
 792         SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->total_bw", dl_bw->total_bw);
 793
 794 #undef PU
 795 }
 796
 797 static void print_cpu(struct seq_file *m, int cpu)
 798 {
 799         struct rq *rq = cpu_rq(cpu);
 800         unsigned long flags;
 801
 802 #ifdef CONFIG_X86
 803         {
 804                 unsigned int freq = cpu_khz ? : 1;
 805
 806                 SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
 807                            cpu, freq / 1000, (freq % 1000));
 808         }
 809 #else
 810         SEQ_printf(m, "cpu#%d\n", cpu);
 811 #endif
 812
 813 #define P(x)                                                            \
 814 do {                                                                    \
 815         if (sizeof(rq->x) == 4)                                         \
 816                 SEQ_printf(m, "  .%-30s: %ld\n", #x, (long)(rq->x));    \
 817         else                                                            \
 818                 SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rq->x));\
 819 } while (0)
 820
 821 #define PN(x) \
 822         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
 823
 824         P(nr_running);
 825         P(nr_switches);
 826         P(nr_uninterruptible);
 827         PN(next_balance);
 828         SEQ_printf(m, "  .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr)));
 829         PN(clock);
 830         PN(clock_task);
 831 #undef P
 832 #undef PN
 833
 834 #ifdef CONFIG_SMP
 835 #define P64(n) SEQ_printf(m, "  .%-30s: %Ld\n", #n, rq->n);
 836         P64(avg_idle);
 837         P64(max_idle_balance_cost);
 838 #undef P64
 839 #endif
 840
 841 #define P(n) SEQ_printf(m, "  .%-30s: %d\n", #n, schedstat_val(rq->n));
 842         if (schedstat_enabled()) {
 843                 P(yld_count);
 844                 P(sched_count);
 845                 P(sched_goidle);
 846                 P(ttwu_count);
 847                 P(ttwu_local);
 848         }
 849 #undef P
 850
 851         spin_lock_irqsave(&sched_debug_lock, flags);
 852         print_cfs_stats(m, cpu);
 853         print_rt_stats(m, cpu);
 854         print_dl_stats(m, cpu);
 855
 856         print_rq(m, rq, cpu);
 857         spin_unlock_irqrestore(&sched_debug_lock, flags);
 858         SEQ_printf(m, "\n");
 859 }
 860
 861 static const char *sched_tunable_scaling_names[] = {
 862         "none",
 863         "logarithmic",
 864         "linear"
 865 };
 866
 867 static void sched_debug_header(struct seq_file *m)
 868 {
 869         u64 ktime, sched_clk, cpu_clk;
 870         unsigned long flags;
 871
 872         local_irq_save(flags);
 873         ktime = ktime_to_ns(ktime_get());
 874         sched_clk = sched_clock();
 875         cpu_clk = local_clock();
 876         local_irq_restore(flags);
 877
 878         SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n",
 879                 init_utsname()->release,
 880                 (int)strcspn(init_utsname()->version, " "),
 881                 init_utsname()->version);
 882
 883 #define P(x) \
 884         SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
 885 #define PN(x) \
 886         SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
 887         PN(ktime);
 888         PN(sched_clk);
 889         PN(cpu_clk);
 890         P(jiffies);
 891 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
 892         P(sched_clock_stable());
 893 #endif
 894 #undef PN
 895 #undef P
 896
 897         SEQ_printf(m, "\n");
 898         SEQ_printf(m, "sysctl_sched\n");
 899
 900 #define P(x) \
 901         SEQ_printf(m, "  .%-40s: %Ld\n", #x, (long long)(x))
 902 #define PN(x) \
 903         SEQ_printf(m, "  .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
 904         PN(sysctl_sched_latency);
 905         PN(sysctl_sched_min_granularity);
 906         PN(sysctl_sched_wakeup_granularity);
 907         P(sysctl_sched_child_runs_first);
 908         P(sysctl_sched_features);
 909 #undef PN
 910 #undef P
 911
 912         SEQ_printf(m, "  .%-40s: %d (%s)\n",
 913                 "sysctl_sched_tunable_scaling",
 914                 sysctl_sched_tunable_scaling,
 915                 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
 916         SEQ_printf(m, "\n");
 917 }
 918
 919 static int sched_debug_show(struct seq_file *m, void *v)
 920 {
 921         int cpu = (unsigned long)(v - 2);
 922
 923         if (cpu != -1)
 924                 print_cpu(m, cpu);
 925         else
 926                 sched_debug_header(m);
 927
 928         return 0;
 929 }
 930
 931 void sysrq_sched_debug_show(void)
 932 {
 933         int cpu;
 934
 935         sched_debug_header(NULL);
 936         for_each_online_cpu(cpu) {
 937                 /*
 938                  * Need to reset softlockup watchdogs on all CPUs, because
 939                  * another CPU might be blocked waiting for us to process
 940                  * an IPI or stop_machine.
 941                  */
 942                 touch_nmi_watchdog();
 943                 touch_all_softlockup_watchdogs();
 944                 print_cpu(NULL, cpu);
 945         }
 946 }
 947
 948 /*
 949  * This iterator needs some explanation.
 950  * It returns 1 for the header position.
 951  * This means 2 is CPU 0.
 952  * In a hotplugged system some CPUs, including CPU 0, may be missing so we have
 953  * to use cpumask_* to iterate over the CPUs.
 954  */
 955 static void *sched_debug_start(struct seq_file *file, loff_t *offset)
 956 {
 957         unsigned long n = *offset;
 958
 959         if (n == 0)
 960                 return (void *) 1;
 961
 962         n--;
 963
 964         if (n > 0)
 965                 n = cpumask_next(n - 1, cpu_online_mask);
 966         else
 967                 n = cpumask_first(cpu_online_mask);
 968
 969         *offset = n + 1;
 970
 971         if (n < nr_cpu_ids)
 972                 return (void *)(unsigned long)(n + 2);
 973
 974         return NULL;
 975 }
 976
 977 static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
 978 {
 979         (*offset)++;
 980         return sched_debug_start(file, offset);
 981 }
 982
 983 static void sched_debug_stop(struct seq_file *file, void *data)
 984 {
 985 }
 986
 987 static const struct seq_operations sched_debug_sops = {
 988         .start          = sched_debug_start,
 989         .next           = sched_debug_next,
 990         .stop           = sched_debug_stop,
 991         .show           = sched_debug_show,
 992 };
 993
 994 static int __init init_sched_debug_procfs(void)
 995 {
 996         if (!proc_create_seq("sched_debug", 0444, NULL, &sched_debug_sops))
 997                 return -ENOMEM;
 998         return 0;
 999 }
1000
1001 __initcall(init_sched_debug_procfs);
1002
1003 #define __PS(S, F) SEQ_printf(m, "%-45s:%21Ld\n", S, (long long)(F))
1004 #define __P(F) __PS(#F, F)
1005 #define   P(F) __PS(#F, p->F)
1006 #define __PSN(S, F) SEQ_printf(m, "%-45s:%14Ld.%06ld\n", S, SPLIT_NS((long long)(F)))
1007 #define __PN(F) __PSN(#F, F)
1008 #define   PN(F) __PSN(#F, p->F)
1009
1010
1011 #ifdef CONFIG_NUMA_BALANCING
1012 void print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
1013                 unsigned long tpf, unsigned long gsf, unsigned long gpf)
1014 {
1015         SEQ_printf(m, "numa_faults node=%d ", node);
1016         SEQ_printf(m, "task_private=%lu task_shared=%lu ", tpf, tsf);
1017         SEQ_printf(m, "group_private=%lu group_shared=%lu\n", gpf, gsf);
1018 }
1019 #endif
1020
1021
1022 static void sched_show_numa(struct task_struct *p, struct seq_file *m)
1023 {
1024 #ifdef CONFIG_NUMA_BALANCING
1025         struct mempolicy *pol;
1026
1027         if (p->mm)
1028                 P(mm->numa_scan_seq);
1029
1030         task_lock(p);
1031         pol = p->mempolicy;
1032         if (pol && !(pol->flags & MPOL_F_MORON))
1033                 pol = NULL;
1034         mpol_get(pol);
1035         task_unlock(p);
1036
1037         P(numa_pages_migrated);
1038         P(numa_preferred_nid);
1039         P(total_numa_faults);
1040         SEQ_printf(m, "current_node=%d, numa_group_id=%d\n",
1041                         task_node(p), task_numa_group_id(p));
1042         show_numa_stats(p, m);
1043         mpol_put(pol);
1044 #endif
1045 }
1046
1047 void proc_sched_show_task(struct task_struct *p, struct pid_namespace *ns,
1048                                                   struct seq_file *m)
1049 {
1050         unsigned long nr_switches;
1051
1052         SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr_ns(p, ns),
1053                                                 get_nr_threads(p));
1054         SEQ_printf(m,
1055                 "---------------------------------------------------------"
1056                 "----------\n");
1057
1058 #define P_SCHEDSTAT(F)  __PS(#F, schedstat_val(p->F))
1059 #define PN_SCHEDSTAT(F) __PSN(#F, schedstat_val(p->F))
1060
1061         PN(se.exec_start);
1062         PN(se.vruntime);
1063         PN(se.sum_exec_runtime);
1064
1065         nr_switches = p->nvcsw + p->nivcsw;
1066
1067         P(se.nr_migrations);
1068
1069         if (schedstat_enabled()) {
1070                 u64 avg_atom, avg_per_cpu;
1071
1072                 PN_SCHEDSTAT(se.statistics.sum_sleep_runtime);
1073                 PN_SCHEDSTAT(se.statistics.wait_start);
1074                 PN_SCHEDSTAT(se.statistics.sleep_start);
1075                 PN_SCHEDSTAT(se.statistics.block_start);
1076                 PN_SCHEDSTAT(se.statistics.sleep_max);
1077                 PN_SCHEDSTAT(se.statistics.block_max);
1078                 PN_SCHEDSTAT(se.statistics.exec_max);
1079                 PN_SCHEDSTAT(se.statistics.slice_max);
1080                 PN_SCHEDSTAT(se.statistics.wait_max);
1081                 PN_SCHEDSTAT(se.statistics.wait_sum);
1082                 P_SCHEDSTAT(se.statistics.wait_count);
1083                 PN_SCHEDSTAT(se.statistics.iowait_sum);
1084                 P_SCHEDSTAT(se.statistics.iowait_count);
1085                 P_SCHEDSTAT(se.statistics.nr_migrations_cold);
1086                 P_SCHEDSTAT(se.statistics.nr_failed_migrations_affine);
1087                 P_SCHEDSTAT(se.statistics.nr_failed_migrations_running);
1088                 P_SCHEDSTAT(se.statistics.nr_failed_migrations_hot);
1089                 P_SCHEDSTAT(se.statistics.nr_forced_migrations);
1090                 P_SCHEDSTAT(se.statistics.nr_wakeups);
1091                 P_SCHEDSTAT(se.statistics.nr_wakeups_sync);
1092                 P_SCHEDSTAT(se.statistics.nr_wakeups_migrate);
1093                 P_SCHEDSTAT(se.statistics.nr_wakeups_local);
1094                 P_SCHEDSTAT(se.statistics.nr_wakeups_remote);
1095                 P_SCHEDSTAT(se.statistics.nr_wakeups_affine);
1096                 P_SCHEDSTAT(se.statistics.nr_wakeups_affine_attempts);
1097                 P_SCHEDSTAT(se.statistics.nr_wakeups_passive);
1098                 P_SCHEDSTAT(se.statistics.nr_wakeups_idle);
1099
1100                 avg_atom = p->se.sum_exec_runtime;
1101                 if (nr_switches)
1102                         avg_atom = div64_ul(avg_atom, nr_switches);
1103                 else
1104                         avg_atom = -1LL;
1105
1106                 avg_per_cpu = p->se.sum_exec_runtime;
1107                 if (p->se.nr_migrations) {
1108                         avg_per_cpu = div64_u64(avg_per_cpu,
1109                                                 p->se.nr_migrations);
1110                 } else {
1111                         avg_per_cpu = -1LL;
1112                 }
1113
1114                 __PN(avg_atom);
1115                 __PN(avg_per_cpu);
1116         }
1117
1118         __P(nr_switches);
1119         __PS("nr_voluntary_switches", p->nvcsw);
1120         __PS("nr_involuntary_switches", p->nivcsw);
1121
1122         P(se.load.weight);
1123 #ifdef CONFIG_SMP
1124         P(se.avg.load_sum);
1125         P(se.avg.runnable_sum);
1126         P(se.avg.util_sum);
1127         P(se.avg.load_avg);
1128         P(se.avg.runnable_avg);
1129         P(se.avg.util_avg);
1130         P(se.avg.last_update_time);
1131         P(se.avg.util_est.ewma);
1132         P(se.avg.util_est.enqueued);
1133 #endif
1134 #ifdef CONFIG_UCLAMP_TASK
1135         __PS("uclamp.min", p->uclamp_req[UCLAMP_MIN].value);
1136         __PS("uclamp.max", p->uclamp_req[UCLAMP_MAX].value);
1137         __PS("effective uclamp.min", uclamp_eff_value(p, UCLAMP_MIN));
1138         __PS("effective uclamp.max", uclamp_eff_value(p, UCLAMP_MAX));
1139 #endif
1140         P(policy);
1141         P(prio);
1142         if (task_has_dl_policy(p)) {
1143                 P(dl.runtime);
1144                 P(dl.deadline);
1145         }
1146 #undef PN_SCHEDSTAT
1147 #undef P_SCHEDSTAT
1148
1149         {
1150                 unsigned int this_cpu = raw_smp_processor_id();
1151                 u64 t0, t1;
1152
1153                 t0 = cpu_clock(this_cpu);
1154                 t1 = cpu_clock(this_cpu);
1155                 __PS("clock-delta", t1-t0);
1156         }
1157
1158         sched_show_numa(p, m);
1159 }
1160
1161 void proc_sched_set_task(struct task_struct *p)
1162 {
1163 #ifdef CONFIG_SCHEDSTATS
1164         memset(&p->se.statistics, 0, sizeof(p->se.statistics));
1165 #endif
1166 }