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 __read_mostly bool sched_debug_enabled;
 171
 172 static __init int sched_init_debug(void)
 173 {
 174         debugfs_create_file("sched_features", 0644, NULL, NULL,
 175                         &sched_feat_fops);
 176
 177         debugfs_create_bool("sched_debug", 0644, NULL,
 178                         &sched_debug_enabled);
 179
 180         return 0;
 181 }
 182 late_initcall(sched_init_debug);
 183
 184 #ifdef CONFIG_SMP
 185
 186 #ifdef CONFIG_SYSCTL
 187
 188 static struct ctl_table sd_ctl_dir[] = {
 189         {
 190                 .procname       = "sched_domain",
 191                 .mode           = 0555,
 192         },
 193         {}
 194 };
 195
 196 static struct ctl_table sd_ctl_root[] = {
 197         {
 198                 .procname       = "kernel",
 199                 .mode           = 0555,
 200                 .child          = sd_ctl_dir,
 201         },
 202         {}
 203 };
 204
 205 static struct ctl_table *sd_alloc_ctl_entry(int n)
 206 {
 207         struct ctl_table *entry =
 208                 kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL);
 209
 210         return entry;
 211 }
 212
 213 static void sd_free_ctl_entry(struct ctl_table **tablep)
 214 {
 215         struct ctl_table *entry;
 216
 217         /*
 218          * In the intermediate directories, both the child directory and
 219          * procname are dynamically allocated and could fail but the mode
 220          * will always be set. In the lowest directory the names are
 221          * static strings and all have proc handlers.
 222          */
 223         for (entry = *tablep; entry->mode; entry++) {
 224                 if (entry->child)
 225                         sd_free_ctl_entry(&entry->child);
 226                 if (entry->proc_handler == NULL)
 227                         kfree(entry->procname);
 228         }
 229
 230         kfree(*tablep);
 231         *tablep = NULL;
 232 }
 233
 234 static void
 235 set_table_entry(struct ctl_table *entry,
 236                 const char *procname, void *data, int maxlen,
 237                 umode_t mode, proc_handler *proc_handler)
 238 {
 239         entry->procname = procname;
 240         entry->data = data;
 241         entry->maxlen = maxlen;
 242         entry->mode = mode;
 243         entry->proc_handler = proc_handler;
 244 }
 245
 246 static struct ctl_table *
 247 sd_alloc_ctl_domain_table(struct sched_domain *sd)
 248 {
 249         struct ctl_table *table = sd_alloc_ctl_entry(9);
 250
 251         if (table == NULL)
 252                 return NULL;
 253
 254         set_table_entry(&table[0], "min_interval",        &sd->min_interval,        sizeof(long), 0644, proc_doulongvec_minmax);
 255         set_table_entry(&table[1], "max_interval",        &sd->max_interval,        sizeof(long), 0644, proc_doulongvec_minmax);
 256         set_table_entry(&table[2], "busy_factor",         &sd->busy_factor,         sizeof(int),  0644, proc_dointvec_minmax);
 257         set_table_entry(&table[3], "imbalance_pct",       &sd->imbalance_pct,       sizeof(int),  0644, proc_dointvec_minmax);
 258         set_table_entry(&table[4], "cache_nice_tries",    &sd->cache_nice_tries,    sizeof(int),  0644, proc_dointvec_minmax);
 259         set_table_entry(&table[5], "flags",               &sd->flags,               sizeof(int),  0444, proc_dointvec_minmax);
 260         set_table_entry(&table[6], "max_newidle_lb_cost", &sd->max_newidle_lb_cost, sizeof(long), 0644, proc_doulongvec_minmax);
 261         set_table_entry(&table[7], "name",                sd->name,            CORENAME_MAX_SIZE, 0444, proc_dostring);
 262         /* &table[8] is terminator */
 263
 264         return table;
 265 }
 266
 267 static struct ctl_table *sd_alloc_ctl_cpu_table(int cpu)
 268 {
 269         struct ctl_table *entry, *table;
 270         struct sched_domain *sd;
 271         int domain_num = 0, i;
 272         char buf[32];
 273
 274         for_each_domain(cpu, sd)
 275                 domain_num++;
 276         entry = table = sd_alloc_ctl_entry(domain_num + 1);
 277         if (table == NULL)
 278                 return NULL;
 279
 280         i = 0;
 281         for_each_domain(cpu, sd) {
 282                 snprintf(buf, 32, "domain%d", i);
 283                 entry->procname = kstrdup(buf, GFP_KERNEL);
 284                 entry->mode = 0555;
 285                 entry->child = sd_alloc_ctl_domain_table(sd);
 286                 entry++;
 287                 i++;
 288         }
 289         return table;
 290 }
 291
 292 static cpumask_var_t            sd_sysctl_cpus;
 293 static struct ctl_table_header  *sd_sysctl_header;
 294
 295 void register_sched_domain_sysctl(void)
 296 {
 297         static struct ctl_table *cpu_entries;
 298         static struct ctl_table **cpu_idx;
 299         static bool init_done = false;
 300         char buf[32];
 301         int i;
 302
 303         if (!cpu_entries) {
 304                 cpu_entries = sd_alloc_ctl_entry(num_possible_cpus() + 1);
 305                 if (!cpu_entries)
 306                         return;
 307
 308                 WARN_ON(sd_ctl_dir[0].child);
 309                 sd_ctl_dir[0].child = cpu_entries;
 310         }
 311
 312         if (!cpu_idx) {
 313                 struct ctl_table *e = cpu_entries;
 314
 315                 cpu_idx = kcalloc(nr_cpu_ids, sizeof(struct ctl_table*), GFP_KERNEL);
 316                 if (!cpu_idx)
 317                         return;
 318
 319                 /* deal with sparse possible map */
 320                 for_each_possible_cpu(i) {
 321                         cpu_idx[i] = e;
 322                         e++;
 323                 }
 324         }
 325
 326         if (!cpumask_available(sd_sysctl_cpus)) {
 327                 if (!alloc_cpumask_var(&sd_sysctl_cpus, GFP_KERNEL))
 328                         return;
 329         }
 330
 331         if (!init_done) {
 332                 init_done = true;
 333                 /* init to possible to not have holes in @cpu_entries */
 334                 cpumask_copy(sd_sysctl_cpus, cpu_possible_mask);
 335         }
 336
 337         for_each_cpu(i, sd_sysctl_cpus) {
 338                 struct ctl_table *e = cpu_idx[i];
 339
 340                 if (e->child)
 341                         sd_free_ctl_entry(&e->child);
 342
 343                 if (!e->procname) {
 344                         snprintf(buf, 32, "cpu%d", i);
 345                         e->procname = kstrdup(buf, GFP_KERNEL);
 346                 }
 347                 e->mode = 0555;
 348                 e->child = sd_alloc_ctl_cpu_table(i);
 349
 350                 __cpumask_clear_cpu(i, sd_sysctl_cpus);
 351         }
 352
 353         WARN_ON(sd_sysctl_header);
 354         sd_sysctl_header = register_sysctl_table(sd_ctl_root);
 355 }
 356
 357 void dirty_sched_domain_sysctl(int cpu)
 358 {
 359         if (cpumask_available(sd_sysctl_cpus))
 360                 __cpumask_set_cpu(cpu, sd_sysctl_cpus);
 361 }
 362
 363 /* may be called multiple times per register */
 364 void unregister_sched_domain_sysctl(void)
 365 {
 366         unregister_sysctl_table(sd_sysctl_header);
 367         sd_sysctl_header = NULL;
 368 }
 369 #endif /* CONFIG_SYSCTL */
 370 #endif /* CONFIG_SMP */
 371
 372 #ifdef CONFIG_FAIR_GROUP_SCHED
 373 static void print_cfs_group_stats(struct seq_file *m, int cpu, struct task_group *tg)
 374 {
 375         struct sched_entity *se = tg->se[cpu];
 376
 377 #define P(F)            SEQ_printf(m, "  .%-30s: %lld\n",       #F, (long long)F)
 378 #define P_SCHEDSTAT(F)  SEQ_printf(m, "  .%-30s: %lld\n",       #F, (long long)schedstat_val(F))
 379 #define PN(F)           SEQ_printf(m, "  .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F))
 380 #define PN_SCHEDSTAT(F) SEQ_printf(m, "  .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)schedstat_val(F)))
 381
 382         if (!se)
 383                 return;
 384
 385         PN(se->exec_start);
 386         PN(se->vruntime);
 387         PN(se->sum_exec_runtime);
 388
 389         if (schedstat_enabled()) {
 390                 PN_SCHEDSTAT(se->statistics.wait_start);
 391                 PN_SCHEDSTAT(se->statistics.sleep_start);
 392                 PN_SCHEDSTAT(se->statistics.block_start);
 393                 PN_SCHEDSTAT(se->statistics.sleep_max);
 394                 PN_SCHEDSTAT(se->statistics.block_max);
 395                 PN_SCHEDSTAT(se->statistics.exec_max);
 396                 PN_SCHEDSTAT(se->statistics.slice_max);
 397                 PN_SCHEDSTAT(se->statistics.wait_max);
 398                 PN_SCHEDSTAT(se->statistics.wait_sum);
 399                 P_SCHEDSTAT(se->statistics.wait_count);
 400         }
 401
 402         P(se->load.weight);
 403         P(se->runnable_weight);
 404 #ifdef CONFIG_SMP
 405         P(se->avg.load_avg);
 406         P(se->avg.util_avg);
 407         P(se->avg.runnable_load_avg);
 408 #endif
 409
 410 #undef PN_SCHEDSTAT
 411 #undef PN
 412 #undef P_SCHEDSTAT
 413 #undef P
 414 }
 415 #endif
 416
 417 #ifdef CONFIG_CGROUP_SCHED
 418 static DEFINE_SPINLOCK(sched_debug_lock);
 419 static char group_path[PATH_MAX];
 420
 421 static void task_group_path(struct task_group *tg, char *path, int plen)
 422 {
 423         if (autogroup_path(tg, path, plen))
 424                 return;
 425
 426         cgroup_path(tg->css.cgroup, path, plen);
 427 }
 428
 429 /*
 430  * Only 1 SEQ_printf_task_group_path() caller can use the full length
 431  * group_path[] for cgroup path. Other simultaneous callers will have
 432  * to use a shorter stack buffer. A "..." suffix is appended at the end
 433  * of the stack buffer so that it will show up in case the output length
 434  * matches the given buffer size to indicate possible path name truncation.
 435  */
 436 #define SEQ_printf_task_group_path(m, tg, fmt...)                       \
 437 {                                                                       \
 438         if (spin_trylock(&sched_debug_lock)) {                          \
 439                 task_group_path(tg, group_path, sizeof(group_path));    \
 440                 SEQ_printf(m, fmt, group_path);                         \
 441                 spin_unlock(&sched_debug_lock);                         \
 442         } else {                                                        \
 443                 char buf[128];                                          \
 444                 char *bufend = buf + sizeof(buf) - 3;                   \
 445                 task_group_path(tg, buf, bufend - buf);                 \
 446                 strcpy(bufend - 1, "...");                              \
 447                 SEQ_printf(m, fmt, buf);                                \
 448         }                                                               \
 449 }
 450 #endif
 451
 452 static void
 453 print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
 454 {
 455         if (rq->curr == p)
 456                 SEQ_printf(m, ">R");
 457         else
 458                 SEQ_printf(m, " %c", task_state_to_char(p));
 459
 460         SEQ_printf(m, "%15s %5d %9Ld.%06ld %9Ld %5d ",
 461                 p->comm, task_pid_nr(p),
 462                 SPLIT_NS(p->se.vruntime),
 463                 (long long)(p->nvcsw + p->nivcsw),
 464                 p->prio);
 465
 466         SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
 467                 SPLIT_NS(schedstat_val_or_zero(p->se.statistics.wait_sum)),
 468                 SPLIT_NS(p->se.sum_exec_runtime),
 469                 SPLIT_NS(schedstat_val_or_zero(p->se.statistics.sum_sleep_runtime)));
 470
 471 #ifdef CONFIG_NUMA_BALANCING
 472         SEQ_printf(m, " %d %d", task_node(p), task_numa_group_id(p));
 473 #endif
 474 #ifdef CONFIG_CGROUP_SCHED
 475         SEQ_printf_task_group_path(m, task_group(p), " %s")
 476 #endif
 477
 478         SEQ_printf(m, "\n");
 479 }
 480
 481 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
 482 {
 483         struct task_struct *g, *p;
 484
 485         SEQ_printf(m, "\n");
 486         SEQ_printf(m, "runnable tasks:\n");
 487         SEQ_printf(m, " S           task   PID         tree-key  switches  prio"
 488                    "     wait-time             sum-exec        sum-sleep\n");
 489         SEQ_printf(m, "-------------------------------------------------------"
 490                    "----------------------------------------------------\n");
 491
 492         rcu_read_lock();
 493         for_each_process_thread(g, p) {
 494                 if (task_cpu(p) != rq_cpu)
 495                         continue;
 496
 497                 print_task(m, rq, p);
 498         }
 499         rcu_read_unlock();
 500 }
 501
 502 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
 503 {
 504         s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
 505                 spread, rq0_min_vruntime, spread0;
 506         struct rq *rq = cpu_rq(cpu);
 507         struct sched_entity *last;
 508         unsigned long flags;
 509
 510 #ifdef CONFIG_FAIR_GROUP_SCHED
 511         SEQ_printf(m, "\n");
 512         SEQ_printf_task_group_path(m, cfs_rq->tg, "cfs_rq[%d]:%s\n", cpu);
 513 #else
 514         SEQ_printf(m, "\n");
 515         SEQ_printf(m, "cfs_rq[%d]:\n", cpu);
 516 #endif
 517         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "exec_clock",
 518                         SPLIT_NS(cfs_rq->exec_clock));
 519
 520         raw_spin_lock_irqsave(&rq->lock, flags);
 521         if (rb_first_cached(&cfs_rq->tasks_timeline))
 522                 MIN_vruntime = (__pick_first_entity(cfs_rq))->vruntime;
 523         last = __pick_last_entity(cfs_rq);
 524         if (last)
 525                 max_vruntime = last->vruntime;
 526         min_vruntime = cfs_rq->min_vruntime;
 527         rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
 528         raw_spin_unlock_irqrestore(&rq->lock, flags);
 529         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "MIN_vruntime",
 530                         SPLIT_NS(MIN_vruntime));
 531         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "min_vruntime",
 532                         SPLIT_NS(min_vruntime));
 533         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "max_vruntime",
 534                         SPLIT_NS(max_vruntime));
 535         spread = max_vruntime - MIN_vruntime;
 536         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread",
 537                         SPLIT_NS(spread));
 538         spread0 = min_vruntime - rq0_min_vruntime;
 539         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread0",
 540                         SPLIT_NS(spread0));
 541         SEQ_printf(m, "  .%-30s: %d\n", "nr_spread_over",
 542                         cfs_rq->nr_spread_over);
 543         SEQ_printf(m, "  .%-30s: %d\n", "nr_running", cfs_rq->nr_running);
 544         SEQ_printf(m, "  .%-30s: %ld\n", "load", cfs_rq->load.weight);
 545 #ifdef CONFIG_SMP
 546         SEQ_printf(m, "  .%-30s: %ld\n", "runnable_weight", cfs_rq->runnable_weight);
 547         SEQ_printf(m, "  .%-30s: %lu\n", "load_avg",
 548                         cfs_rq->avg.load_avg);
 549         SEQ_printf(m, "  .%-30s: %lu\n", "runnable_load_avg",
 550                         cfs_rq->avg.runnable_load_avg);
 551         SEQ_printf(m, "  .%-30s: %lu\n", "util_avg",
 552                         cfs_rq->avg.util_avg);
 553         SEQ_printf(m, "  .%-30s: %u\n", "util_est_enqueued",
 554                         cfs_rq->avg.util_est.enqueued);
 555         SEQ_printf(m, "  .%-30s: %ld\n", "removed.load_avg",
 556                         cfs_rq->removed.load_avg);
 557         SEQ_printf(m, "  .%-30s: %ld\n", "removed.util_avg",
 558                         cfs_rq->removed.util_avg);
 559         SEQ_printf(m, "  .%-30s: %ld\n", "removed.runnable_sum",
 560                         cfs_rq->removed.runnable_sum);
 561 #ifdef CONFIG_FAIR_GROUP_SCHED
 562         SEQ_printf(m, "  .%-30s: %lu\n", "tg_load_avg_contrib",
 563                         cfs_rq->tg_load_avg_contrib);
 564         SEQ_printf(m, "  .%-30s: %ld\n", "tg_load_avg",
 565                         atomic_long_read(&cfs_rq->tg->load_avg));
 566 #endif
 567 #endif
 568 #ifdef CONFIG_CFS_BANDWIDTH
 569         SEQ_printf(m, "  .%-30s: %d\n", "throttled",
 570                         cfs_rq->throttled);
 571         SEQ_printf(m, "  .%-30s: %d\n", "throttle_count",
 572                         cfs_rq->throttle_count);
 573 #endif
 574
 575 #ifdef CONFIG_FAIR_GROUP_SCHED
 576         print_cfs_group_stats(m, cpu, cfs_rq->tg);
 577 #endif
 578 }
 579
 580 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
 581 {
 582 #ifdef CONFIG_RT_GROUP_SCHED
 583         SEQ_printf(m, "\n");
 584         SEQ_printf_task_group_path(m, rt_rq->tg, "rt_rq[%d]:%s\n", cpu);
 585 #else
 586         SEQ_printf(m, "\n");
 587         SEQ_printf(m, "rt_rq[%d]:\n", cpu);
 588 #endif
 589
 590 #define P(x) \
 591         SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
 592 #define PU(x) \
 593         SEQ_printf(m, "  .%-30s: %lu\n", #x, (unsigned long)(rt_rq->x))
 594 #define PN(x) \
 595         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
 596
 597         PU(rt_nr_running);
 598 #ifdef CONFIG_SMP
 599         PU(rt_nr_migratory);
 600 #endif
 601         P(rt_throttled);
 602         PN(rt_time);
 603         PN(rt_runtime);
 604
 605 #undef PN
 606 #undef PU
 607 #undef P
 608 }
 609
 610 void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq)
 611 {
 612         struct dl_bw *dl_bw;
 613
 614         SEQ_printf(m, "\n");
 615         SEQ_printf(m, "dl_rq[%d]:\n", cpu);
 616
 617 #define PU(x) \
 618         SEQ_printf(m, "  .%-30s: %lu\n", #x, (unsigned long)(dl_rq->x))
 619
 620         PU(dl_nr_running);
 621 #ifdef CONFIG_SMP
 622         PU(dl_nr_migratory);
 623         dl_bw = &cpu_rq(cpu)->rd->dl_bw;
 624 #else
 625         dl_bw = &dl_rq->dl_bw;
 626 #endif
 627         SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->bw", dl_bw->bw);
 628         SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->total_bw", dl_bw->total_bw);
 629
 630 #undef PU
 631 }
 632
 633 static void print_cpu(struct seq_file *m, int cpu)
 634 {
 635         struct rq *rq = cpu_rq(cpu);
 636
 637 #ifdef CONFIG_X86
 638         {
 639                 unsigned int freq = cpu_khz ? : 1;
 640
 641                 SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
 642                            cpu, freq / 1000, (freq % 1000));
 643         }
 644 #else
 645         SEQ_printf(m, "cpu#%d\n", cpu);
 646 #endif
 647
 648 #define P(x)                                                            \
 649 do {                                                                    \
 650         if (sizeof(rq->x) == 4)                                         \
 651                 SEQ_printf(m, "  .%-30s: %ld\n", #x, (long)(rq->x));    \
 652         else                                                            \
 653                 SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rq->x));\
 654 } while (0)
 655
 656 #define PN(x) \
 657         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
 658
 659         P(nr_running);
 660         P(nr_switches);
 661         P(nr_load_updates);
 662         P(nr_uninterruptible);
 663         PN(next_balance);
 664         SEQ_printf(m, "  .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr)));
 665         PN(clock);
 666         PN(clock_task);
 667 #undef P
 668 #undef PN
 669
 670 #ifdef CONFIG_SMP
 671 #define P64(n) SEQ_printf(m, "  .%-30s: %Ld\n", #n, rq->n);
 672         P64(avg_idle);
 673         P64(max_idle_balance_cost);
 674 #undef P64
 675 #endif
 676
 677 #define P(n) SEQ_printf(m, "  .%-30s: %d\n", #n, schedstat_val(rq->n));
 678         if (schedstat_enabled()) {
 679                 P(yld_count);
 680                 P(sched_count);
 681                 P(sched_goidle);
 682                 P(ttwu_count);
 683                 P(ttwu_local);
 684         }
 685 #undef P
 686
 687         print_cfs_stats(m, cpu);
 688         print_rt_stats(m, cpu);
 689         print_dl_stats(m, cpu);
 690
 691         print_rq(m, rq, cpu);
 692         SEQ_printf(m, "\n");
 693 }
 694
 695 static const char *sched_tunable_scaling_names[] = {
 696         "none",
 697         "logarithmic",
 698         "linear"
 699 };
 700
 701 static void sched_debug_header(struct seq_file *m)
 702 {
 703         u64 ktime, sched_clk, cpu_clk;
 704         unsigned long flags;
 705
 706         local_irq_save(flags);
 707         ktime = ktime_to_ns(ktime_get());
 708         sched_clk = sched_clock();
 709         cpu_clk = local_clock();
 710         local_irq_restore(flags);
 711
 712         SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n",
 713                 init_utsname()->release,
 714                 (int)strcspn(init_utsname()->version, " "),
 715                 init_utsname()->version);
 716
 717 #define P(x) \
 718         SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
 719 #define PN(x) \
 720         SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
 721         PN(ktime);
 722         PN(sched_clk);
 723         PN(cpu_clk);
 724         P(jiffies);
 725 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
 726         P(sched_clock_stable());
 727 #endif
 728 #undef PN
 729 #undef P
 730
 731         SEQ_printf(m, "\n");
 732         SEQ_printf(m, "sysctl_sched\n");
 733
 734 #define P(x) \
 735         SEQ_printf(m, "  .%-40s: %Ld\n", #x, (long long)(x))
 736 #define PN(x) \
 737         SEQ_printf(m, "  .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
 738         PN(sysctl_sched_latency);
 739         PN(sysctl_sched_min_granularity);
 740         PN(sysctl_sched_wakeup_granularity);
 741         P(sysctl_sched_child_runs_first);
 742         P(sysctl_sched_features);
 743 #undef PN
 744 #undef P
 745
 746         SEQ_printf(m, "  .%-40s: %d (%s)\n",
 747                 "sysctl_sched_tunable_scaling",
 748                 sysctl_sched_tunable_scaling,
 749                 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
 750         SEQ_printf(m, "\n");
 751 }
 752
 753 static int sched_debug_show(struct seq_file *m, void *v)
 754 {
 755         int cpu = (unsigned long)(v - 2);
 756
 757         if (cpu != -1)
 758                 print_cpu(m, cpu);
 759         else
 760                 sched_debug_header(m);
 761
 762         return 0;
 763 }
 764
 765 void sysrq_sched_debug_show(void)
 766 {
 767         int cpu;
 768
 769         sched_debug_header(NULL);
 770         for_each_online_cpu(cpu)
 771                 print_cpu(NULL, cpu);
 772
 773 }
 774
 775 /*
 776  * This itererator needs some explanation.
 777  * It returns 1 for the header position.
 778  * This means 2 is CPU 0.
 779  * In a hotplugged system some CPUs, including CPU 0, may be missing so we have
 780  * to use cpumask_* to iterate over the CPUs.
 781  */
 782 static void *sched_debug_start(struct seq_file *file, loff_t *offset)
 783 {
 784         unsigned long n = *offset;
 785
 786         if (n == 0)
 787                 return (void *) 1;
 788
 789         n--;
 790
 791         if (n > 0)
 792                 n = cpumask_next(n - 1, cpu_online_mask);
 793         else
 794                 n = cpumask_first(cpu_online_mask);
 795
 796         *offset = n + 1;
 797
 798         if (n < nr_cpu_ids)
 799                 return (void *)(unsigned long)(n + 2);
 800
 801         return NULL;
 802 }
 803
 804 static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
 805 {
 806         (*offset)++;
 807         return sched_debug_start(file, offset);
 808 }
 809
 810 static void sched_debug_stop(struct seq_file *file, void *data)
 811 {
 812 }
 813
 814 static const struct seq_operations sched_debug_sops = {
 815         .start          = sched_debug_start,
 816         .next           = sched_debug_next,
 817         .stop           = sched_debug_stop,
 818         .show           = sched_debug_show,
 819 };
 820
 821 static int __init init_sched_debug_procfs(void)
 822 {
 823         if (!proc_create_seq("sched_debug", 0444, NULL, &sched_debug_sops))
 824                 return -ENOMEM;
 825         return 0;
 826 }
 827
 828 __initcall(init_sched_debug_procfs);
 829
 830 #define __P(F)  SEQ_printf(m, "%-45s:%21Ld\n",       #F, (long long)F)
 831 #define   P(F)  SEQ_printf(m, "%-45s:%21Ld\n",       #F, (long long)p->F)
 832 #define __PN(F) SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
 833 #define   PN(F) SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
 834
 835
 836 #ifdef CONFIG_NUMA_BALANCING
 837 void print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
 838                 unsigned long tpf, unsigned long gsf, unsigned long gpf)
 839 {
 840         SEQ_printf(m, "numa_faults node=%d ", node);
 841         SEQ_printf(m, "task_private=%lu task_shared=%lu ", tpf, tsf);
 842         SEQ_printf(m, "group_private=%lu group_shared=%lu\n", gpf, gsf);
 843 }
 844 #endif
 845
 846
 847 static void sched_show_numa(struct task_struct *p, struct seq_file *m)
 848 {
 849 #ifdef CONFIG_NUMA_BALANCING
 850         if (p->mm)
 851                 P(mm->numa_scan_seq);
 852
 853         P(numa_pages_migrated);
 854         P(numa_preferred_nid);
 855         P(total_numa_faults);
 856         SEQ_printf(m, "current_node=%d, numa_group_id=%d\n",
 857                         task_node(p), task_numa_group_id(p));
 858         show_numa_stats(p, m);
 859 #endif
 860 }
 861
 862 void proc_sched_show_task(struct task_struct *p, struct pid_namespace *ns,
 863                                                   struct seq_file *m)
 864 {
 865         unsigned long nr_switches;
 866
 867         SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr_ns(p, ns),
 868                                                 get_nr_threads(p));
 869         SEQ_printf(m,
 870                 "---------------------------------------------------------"
 871                 "----------\n");
 872 #define __P(F) \
 873         SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)F)
 874 #define P(F) \
 875         SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)p->F)
 876 #define P_SCHEDSTAT(F) \
 877         SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)schedstat_val(p->F))
 878 #define __PN(F) \
 879         SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
 880 #define PN(F) \
 881         SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
 882 #define PN_SCHEDSTAT(F) \
 883         SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)schedstat_val(p->F)))
 884
 885         PN(se.exec_start);
 886         PN(se.vruntime);
 887         PN(se.sum_exec_runtime);
 888
 889         nr_switches = p->nvcsw + p->nivcsw;
 890
 891         P(se.nr_migrations);
 892
 893         if (schedstat_enabled()) {
 894                 u64 avg_atom, avg_per_cpu;
 895
 896                 PN_SCHEDSTAT(se.statistics.sum_sleep_runtime);
 897                 PN_SCHEDSTAT(se.statistics.wait_start);
 898                 PN_SCHEDSTAT(se.statistics.sleep_start);
 899                 PN_SCHEDSTAT(se.statistics.block_start);
 900                 PN_SCHEDSTAT(se.statistics.sleep_max);
 901                 PN_SCHEDSTAT(se.statistics.block_max);
 902                 PN_SCHEDSTAT(se.statistics.exec_max);
 903                 PN_SCHEDSTAT(se.statistics.slice_max);
 904                 PN_SCHEDSTAT(se.statistics.wait_max);
 905                 PN_SCHEDSTAT(se.statistics.wait_sum);
 906                 P_SCHEDSTAT(se.statistics.wait_count);
 907                 PN_SCHEDSTAT(se.statistics.iowait_sum);
 908                 P_SCHEDSTAT(se.statistics.iowait_count);
 909                 P_SCHEDSTAT(se.statistics.nr_migrations_cold);
 910                 P_SCHEDSTAT(se.statistics.nr_failed_migrations_affine);
 911                 P_SCHEDSTAT(se.statistics.nr_failed_migrations_running);
 912                 P_SCHEDSTAT(se.statistics.nr_failed_migrations_hot);
 913                 P_SCHEDSTAT(se.statistics.nr_forced_migrations);
 914                 P_SCHEDSTAT(se.statistics.nr_wakeups);
 915                 P_SCHEDSTAT(se.statistics.nr_wakeups_sync);
 916                 P_SCHEDSTAT(se.statistics.nr_wakeups_migrate);
 917                 P_SCHEDSTAT(se.statistics.nr_wakeups_local);
 918                 P_SCHEDSTAT(se.statistics.nr_wakeups_remote);
 919                 P_SCHEDSTAT(se.statistics.nr_wakeups_affine);
 920                 P_SCHEDSTAT(se.statistics.nr_wakeups_affine_attempts);
 921                 P_SCHEDSTAT(se.statistics.nr_wakeups_passive);
 922                 P_SCHEDSTAT(se.statistics.nr_wakeups_idle);
 923
 924                 avg_atom = p->se.sum_exec_runtime;
 925                 if (nr_switches)
 926                         avg_atom = div64_ul(avg_atom, nr_switches);
 927                 else
 928                         avg_atom = -1LL;
 929
 930                 avg_per_cpu = p->se.sum_exec_runtime;
 931                 if (p->se.nr_migrations) {
 932                         avg_per_cpu = div64_u64(avg_per_cpu,
 933                                                 p->se.nr_migrations);
 934                 } else {
 935                         avg_per_cpu = -1LL;
 936                 }
 937
 938                 __PN(avg_atom);
 939                 __PN(avg_per_cpu);
 940         }
 941
 942         __P(nr_switches);
 943         SEQ_printf(m, "%-45s:%21Ld\n",
 944                    "nr_voluntary_switches", (long long)p->nvcsw);
 945         SEQ_printf(m, "%-45s:%21Ld\n",
 946                    "nr_involuntary_switches", (long long)p->nivcsw);
 947
 948         P(se.load.weight);
 949         P(se.runnable_weight);
 950 #ifdef CONFIG_SMP
 951         P(se.avg.load_sum);
 952         P(se.avg.runnable_load_sum);
 953         P(se.avg.util_sum);
 954         P(se.avg.load_avg);
 955         P(se.avg.runnable_load_avg);
 956         P(se.avg.util_avg);
 957         P(se.avg.last_update_time);
 958         P(se.avg.util_est.ewma);
 959         P(se.avg.util_est.enqueued);
 960 #endif
 961         P(policy);
 962         P(prio);
 963         if (task_has_dl_policy(p)) {
 964                 P(dl.runtime);
 965                 P(dl.deadline);
 966         }
 967 #undef PN_SCHEDSTAT
 968 #undef PN
 969 #undef __PN
 970 #undef P_SCHEDSTAT
 971 #undef P
 972 #undef __P
 973
 974         {
 975                 unsigned int this_cpu = raw_smp_processor_id();
 976                 u64 t0, t1;
 977
 978                 t0 = cpu_clock(this_cpu);
 979                 t1 = cpu_clock(this_cpu);
 980                 SEQ_printf(m, "%-45s:%21Ld\n",
 981                            "clock-delta", (long long)(t1-t0));
 982         }
 983
 984         sched_show_numa(p, m);
 985 }
 986
 987 void proc_sched_set_task(struct task_struct *p)
 988 {
 989 #ifdef CONFIG_SCHEDSTATS
 990         memset(&p->se.statistics, 0, sizeof(p->se.statistics));
 991 #endif
 992 }