kernel/sched/debug.c

   1 /*
   2  * kernel/sched/debug.c
   3  *
   4  * Print the CFS rbtree and other debugging details
   5  *
   6  * Copyright(C) 2007, Red Hat, Inc., Ingo Molnar
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 as
  10  * published by the Free Software Foundation.
  11  */
  12 #include "sched.h"
  13
  14 static DEFINE_SPINLOCK(sched_debug_lock);
  15
  16 /*
  17  * This allows printing both to /proc/sched_debug and
  18  * to the console
  19  */
  20 #define SEQ_printf(m, x...)                     \
  21  do {                                           \
  22         if (m)                                  \
  23                 seq_printf(m, x);               \
  24         else                                    \
  25                 pr_cont(x);                     \
  26  } while (0)
  27
  28 /*
  29  * Ease the printing of nsec fields:
  30  */
  31 static long long nsec_high(unsigned long long nsec)
  32 {
  33         if ((long long)nsec < 0) {
  34                 nsec = -nsec;
  35                 do_div(nsec, 1000000);
  36                 return -nsec;
  37         }
  38         do_div(nsec, 1000000);
  39
  40         return nsec;
  41 }
  42
  43 static unsigned long nsec_low(unsigned long long nsec)
  44 {
  45         if ((long long)nsec < 0)
  46                 nsec = -nsec;
  47
  48         return do_div(nsec, 1000000);
  49 }
  50
  51 #define SPLIT_NS(x) nsec_high(x), nsec_low(x)
  52
  53 #define SCHED_FEAT(name, enabled)       \
  54         #name ,
  55
  56 static const char * const sched_feat_names[] = {
  57 #include "features.h"
  58 };
  59
  60 #undef SCHED_FEAT
  61
  62 static int sched_feat_show(struct seq_file *m, void *v)
  63 {
  64         int i;
  65
  66         for (i = 0; i < __SCHED_FEAT_NR; i++) {
  67                 if (!(sysctl_sched_features & (1UL << i)))
  68                         seq_puts(m, "NO_");
  69                 seq_printf(m, "%s ", sched_feat_names[i]);
  70         }
  71         seq_puts(m, "\n");
  72
  73         return 0;
  74 }
  75
  76 #ifdef HAVE_JUMP_LABEL
  77
  78 #define jump_label_key__true  STATIC_KEY_INIT_TRUE
  79 #define jump_label_key__false STATIC_KEY_INIT_FALSE
  80
  81 #define SCHED_FEAT(name, enabled)       \
  82         jump_label_key__##enabled ,
  83
  84 struct static_key sched_feat_keys[__SCHED_FEAT_NR] = {
  85 #include "features.h"
  86 };
  87
  88 #undef SCHED_FEAT
  89
  90 static void sched_feat_disable(int i)
  91 {
  92         static_key_disable(&sched_feat_keys[i]);
  93 }
  94
  95 static void sched_feat_enable(int i)
  96 {
  97         static_key_enable(&sched_feat_keys[i]);
  98 }
  99 #else
 100 static void sched_feat_disable(int i) { };
 101 static void sched_feat_enable(int i) { };
 102 #endif /* HAVE_JUMP_LABEL */
 103
 104 static int sched_feat_set(char *cmp)
 105 {
 106         int i;
 107         int neg = 0;
 108
 109         if (strncmp(cmp, "NO_", 3) == 0) {
 110                 neg = 1;
 111                 cmp += 3;
 112         }
 113
 114         for (i = 0; i < __SCHED_FEAT_NR; i++) {
 115                 if (strcmp(cmp, sched_feat_names[i]) == 0) {
 116                         if (neg) {
 117                                 sysctl_sched_features &= ~(1UL << i);
 118                                 sched_feat_disable(i);
 119                         } else {
 120                                 sysctl_sched_features |= (1UL << i);
 121                                 sched_feat_enable(i);
 122                         }
 123                         break;
 124                 }
 125         }
 126
 127         return i;
 128 }
 129
 130 static ssize_t
 131 sched_feat_write(struct file *filp, const char __user *ubuf,
 132                 size_t cnt, loff_t *ppos)
 133 {
 134         char buf[64];
 135         char *cmp;
 136         int i;
 137         struct inode *inode;
 138
 139         if (cnt > 63)
 140                 cnt = 63;
 141
 142         if (copy_from_user(&buf, ubuf, cnt))
 143                 return -EFAULT;
 144
 145         buf[cnt] = 0;
 146         cmp = strstrip(buf);
 147
 148         /* Ensure the static_key remains in a consistent state */
 149         inode = file_inode(filp);
 150         inode_lock(inode);
 151         i = sched_feat_set(cmp);
 152         inode_unlock(inode);
 153         if (i == __SCHED_FEAT_NR)
 154                 return -EINVAL;
 155
 156         *ppos += cnt;
 157
 158         return cnt;
 159 }
 160
 161 static int sched_feat_open(struct inode *inode, struct file *filp)
 162 {
 163         return single_open(filp, sched_feat_show, NULL);
 164 }
 165
 166 static const struct file_operations sched_feat_fops = {
 167         .open           = sched_feat_open,
 168         .write          = sched_feat_write,
 169         .read           = seq_read,
 170         .llseek         = seq_lseek,
 171         .release        = single_release,
 172 };
 173
 174 __read_mostly bool sched_debug_enabled;
 175
 176 static __init int sched_init_debug(void)
 177 {
 178         debugfs_create_file("sched_features", 0644, NULL, NULL,
 179                         &sched_feat_fops);
 180
 181         debugfs_create_bool("sched_debug", 0644, NULL,
 182                         &sched_debug_enabled);
 183
 184         return 0;
 185 }
 186 late_initcall(sched_init_debug);
 187
 188 #ifdef CONFIG_SMP
 189
 190 #ifdef CONFIG_SYSCTL
 191
 192 static struct ctl_table sd_ctl_dir[] = {
 193         {
 194                 .procname       = "sched_domain",
 195                 .mode           = 0555,
 196         },
 197         {}
 198 };
 199
 200 static struct ctl_table sd_ctl_root[] = {
 201         {
 202                 .procname       = "kernel",
 203                 .mode           = 0555,
 204                 .child          = sd_ctl_dir,
 205         },
 206         {}
 207 };
 208
 209 static struct ctl_table *sd_alloc_ctl_entry(int n)
 210 {
 211         struct ctl_table *entry =
 212                 kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL);
 213
 214         return entry;
 215 }
 216
 217 static void sd_free_ctl_entry(struct ctl_table **tablep)
 218 {
 219         struct ctl_table *entry;
 220
 221         /*
 222          * In the intermediate directories, both the child directory and
 223          * procname are dynamically allocated and could fail but the mode
 224          * will always be set. In the lowest directory the names are
 225          * static strings and all have proc handlers.
 226          */
 227         for (entry = *tablep; entry->mode; entry++) {
 228                 if (entry->child)
 229                         sd_free_ctl_entry(&entry->child);
 230                 if (entry->proc_handler == NULL)
 231                         kfree(entry->procname);
 232         }
 233
 234         kfree(*tablep);
 235         *tablep = NULL;
 236 }
 237
 238 static int min_load_idx = 0;
 239 static int max_load_idx = CPU_LOAD_IDX_MAX-1;
 240
 241 static void
 242 set_table_entry(struct ctl_table *entry,
 243                 const char *procname, void *data, int maxlen,
 244                 umode_t mode, proc_handler *proc_handler,
 245                 bool load_idx)
 246 {
 247         entry->procname = procname;
 248         entry->data = data;
 249         entry->maxlen = maxlen;
 250         entry->mode = mode;
 251         entry->proc_handler = proc_handler;
 252
 253         if (load_idx) {
 254                 entry->extra1 = &min_load_idx;
 255                 entry->extra2 = &max_load_idx;
 256         }
 257 }
 258
 259 static struct ctl_table *
 260 sd_alloc_ctl_domain_table(struct sched_domain *sd)
 261 {
 262         struct ctl_table *table = sd_alloc_ctl_entry(14);
 263
 264         if (table == NULL)
 265                 return NULL;
 266
 267         set_table_entry(&table[0] , "min_interval",        &sd->min_interval,        sizeof(long), 0644, proc_doulongvec_minmax, false);
 268         set_table_entry(&table[1] , "max_interval",        &sd->max_interval,        sizeof(long), 0644, proc_doulongvec_minmax, false);
 269         set_table_entry(&table[2] , "busy_idx",            &sd->busy_idx,            sizeof(int) , 0644, proc_dointvec_minmax,   true );
 270         set_table_entry(&table[3] , "idle_idx",            &sd->idle_idx,            sizeof(int) , 0644, proc_dointvec_minmax,   true );
 271         set_table_entry(&table[4] , "newidle_idx",         &sd->newidle_idx,         sizeof(int) , 0644, proc_dointvec_minmax,   true );
 272         set_table_entry(&table[5] , "wake_idx",            &sd->wake_idx,            sizeof(int) , 0644, proc_dointvec_minmax,   true );
 273         set_table_entry(&table[6] , "forkexec_idx",        &sd->forkexec_idx,        sizeof(int) , 0644, proc_dointvec_minmax,   true );
 274         set_table_entry(&table[7] , "busy_factor",         &sd->busy_factor,         sizeof(int) , 0644, proc_dointvec_minmax,   false);
 275         set_table_entry(&table[8] , "imbalance_pct",       &sd->imbalance_pct,       sizeof(int) , 0644, proc_dointvec_minmax,   false);
 276         set_table_entry(&table[9] , "cache_nice_tries",    &sd->cache_nice_tries,    sizeof(int) , 0644, proc_dointvec_minmax,   false);
 277         set_table_entry(&table[10], "flags",               &sd->flags,               sizeof(int) , 0644, proc_dointvec_minmax,   false);
 278         set_table_entry(&table[11], "max_newidle_lb_cost", &sd->max_newidle_lb_cost, sizeof(long), 0644, proc_doulongvec_minmax, false);
 279         set_table_entry(&table[12], "name",                sd->name,            CORENAME_MAX_SIZE, 0444, proc_dostring,          false);
 280         /* &table[13] is terminator */
 281
 282         return table;
 283 }
 284
 285 static struct ctl_table *sd_alloc_ctl_cpu_table(int cpu)
 286 {
 287         struct ctl_table *entry, *table;
 288         struct sched_domain *sd;
 289         int domain_num = 0, i;
 290         char buf[32];
 291
 292         for_each_domain(cpu, sd)
 293                 domain_num++;
 294         entry = table = sd_alloc_ctl_entry(domain_num + 1);
 295         if (table == NULL)
 296                 return NULL;
 297
 298         i = 0;
 299         for_each_domain(cpu, sd) {
 300                 snprintf(buf, 32, "domain%d", i);
 301                 entry->procname = kstrdup(buf, GFP_KERNEL);
 302                 entry->mode = 0555;
 303                 entry->child = sd_alloc_ctl_domain_table(sd);
 304                 entry++;
 305                 i++;
 306         }
 307         return table;
 308 }
 309
 310 static cpumask_var_t            sd_sysctl_cpus;
 311 static struct ctl_table_header  *sd_sysctl_header;
 312
 313 void register_sched_domain_sysctl(void)
 314 {
 315         static struct ctl_table *cpu_entries;
 316         static struct ctl_table **cpu_idx;
 317         char buf[32];
 318         int i;
 319
 320         if (!cpu_entries) {
 321                 cpu_entries = sd_alloc_ctl_entry(num_possible_cpus() + 1);
 322                 if (!cpu_entries)
 323                         return;
 324
 325                 WARN_ON(sd_ctl_dir[0].child);
 326                 sd_ctl_dir[0].child = cpu_entries;
 327         }
 328
 329         if (!cpu_idx) {
 330                 struct ctl_table *e = cpu_entries;
 331
 332                 cpu_idx = kcalloc(nr_cpu_ids, sizeof(struct ctl_table*), GFP_KERNEL);
 333                 if (!cpu_idx)
 334                         return;
 335
 336                 /* deal with sparse possible map */
 337                 for_each_possible_cpu(i) {
 338                         cpu_idx[i] = e;
 339                         e++;
 340                 }
 341         }
 342
 343         if (!cpumask_available(sd_sysctl_cpus)) {
 344                 if (!alloc_cpumask_var(&sd_sysctl_cpus, GFP_KERNEL))
 345                         return;
 346
 347                 /* init to possible to not have holes in @cpu_entries */
 348                 cpumask_copy(sd_sysctl_cpus, cpu_possible_mask);
 349         }
 350
 351         for_each_cpu(i, sd_sysctl_cpus) {
 352                 struct ctl_table *e = cpu_idx[i];
 353
 354                 if (e->child)
 355                         sd_free_ctl_entry(&e->child);
 356
 357                 if (!e->procname) {
 358                         snprintf(buf, 32, "cpu%d", i);
 359                         e->procname = kstrdup(buf, GFP_KERNEL);
 360                 }
 361                 e->mode = 0555;
 362                 e->child = sd_alloc_ctl_cpu_table(i);
 363
 364                 __cpumask_clear_cpu(i, sd_sysctl_cpus);
 365         }
 366
 367         WARN_ON(sd_sysctl_header);
 368         sd_sysctl_header = register_sysctl_table(sd_ctl_root);
 369 }
 370
 371 void dirty_sched_domain_sysctl(int cpu)
 372 {
 373         if (cpumask_available(sd_sysctl_cpus))
 374                 __cpumask_set_cpu(cpu, sd_sysctl_cpus);
 375 }
 376
 377 /* may be called multiple times per register */
 378 void unregister_sched_domain_sysctl(void)
 379 {
 380         unregister_sysctl_table(sd_sysctl_header);
 381         sd_sysctl_header = NULL;
 382 }
 383 #endif /* CONFIG_SYSCTL */
 384 #endif /* CONFIG_SMP */
 385
 386 #ifdef CONFIG_FAIR_GROUP_SCHED
 387 static void print_cfs_group_stats(struct seq_file *m, int cpu, struct task_group *tg)
 388 {
 389         struct sched_entity *se = tg->se[cpu];
 390
 391 #define P(F)            SEQ_printf(m, "  .%-30s: %lld\n",       #F, (long long)F)
 392 #define P_SCHEDSTAT(F)  SEQ_printf(m, "  .%-30s: %lld\n",       #F, (long long)schedstat_val(F))
 393 #define PN(F)           SEQ_printf(m, "  .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F))
 394 #define PN_SCHEDSTAT(F) SEQ_printf(m, "  .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)schedstat_val(F)))
 395
 396         if (!se)
 397                 return;
 398
 399         PN(se->exec_start);
 400         PN(se->vruntime);
 401         PN(se->sum_exec_runtime);
 402
 403         if (schedstat_enabled()) {
 404                 PN_SCHEDSTAT(se->statistics.wait_start);
 405                 PN_SCHEDSTAT(se->statistics.sleep_start);
 406                 PN_SCHEDSTAT(se->statistics.block_start);
 407                 PN_SCHEDSTAT(se->statistics.sleep_max);
 408                 PN_SCHEDSTAT(se->statistics.block_max);
 409                 PN_SCHEDSTAT(se->statistics.exec_max);
 410                 PN_SCHEDSTAT(se->statistics.slice_max);
 411                 PN_SCHEDSTAT(se->statistics.wait_max);
 412                 PN_SCHEDSTAT(se->statistics.wait_sum);
 413                 P_SCHEDSTAT(se->statistics.wait_count);
 414         }
 415
 416         P(se->load.weight);
 417         P(se->runnable_weight);
 418 #ifdef CONFIG_SMP
 419         P(se->avg.load_avg);
 420         P(se->avg.util_avg);
 421         P(se->avg.runnable_load_avg);
 422 #endif
 423
 424 #undef PN_SCHEDSTAT
 425 #undef PN
 426 #undef P_SCHEDSTAT
 427 #undef P
 428 }
 429 #endif
 430
 431 #ifdef CONFIG_CGROUP_SCHED
 432 static char group_path[PATH_MAX];
 433
 434 static char *task_group_path(struct task_group *tg)
 435 {
 436         if (autogroup_path(tg, group_path, PATH_MAX))
 437                 return group_path;
 438
 439         cgroup_path(tg->css.cgroup, group_path, PATH_MAX);
 440
 441         return group_path;
 442 }
 443 #endif
 444
 445 static void
 446 print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
 447 {
 448         if (rq->curr == p)
 449                 SEQ_printf(m, ">R");
 450         else
 451                 SEQ_printf(m, " %c", task_state_to_char(p));
 452
 453         SEQ_printf(m, "%15s %5d %9Ld.%06ld %9Ld %5d ",
 454                 p->comm, task_pid_nr(p),
 455                 SPLIT_NS(p->se.vruntime),
 456                 (long long)(p->nvcsw + p->nivcsw),
 457                 p->prio);
 458
 459         SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
 460                 SPLIT_NS(schedstat_val_or_zero(p->se.statistics.wait_sum)),
 461                 SPLIT_NS(p->se.sum_exec_runtime),
 462                 SPLIT_NS(schedstat_val_or_zero(p->se.statistics.sum_sleep_runtime)));
 463
 464 #ifdef CONFIG_NUMA_BALANCING
 465         SEQ_printf(m, " %d %d", task_node(p), task_numa_group_id(p));
 466 #endif
 467 #ifdef CONFIG_CGROUP_SCHED
 468         SEQ_printf(m, " %s", task_group_path(task_group(p)));
 469 #endif
 470
 471         SEQ_printf(m, "\n");
 472 }
 473
 474 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
 475 {
 476         struct task_struct *g, *p;
 477
 478         SEQ_printf(m, "\n");
 479         SEQ_printf(m, "runnable tasks:\n");
 480         SEQ_printf(m, " S           task   PID         tree-key  switches  prio"
 481                    "     wait-time             sum-exec        sum-sleep\n");
 482         SEQ_printf(m, "-------------------------------------------------------"
 483                    "----------------------------------------------------\n");
 484
 485         rcu_read_lock();
 486         for_each_process_thread(g, p) {
 487                 if (task_cpu(p) != rq_cpu)
 488                         continue;
 489
 490                 print_task(m, rq, p);
 491         }
 492         rcu_read_unlock();
 493 }
 494
 495 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
 496 {
 497         s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
 498                 spread, rq0_min_vruntime, spread0;
 499         struct rq *rq = cpu_rq(cpu);
 500         struct sched_entity *last;
 501         unsigned long flags;
 502
 503 #ifdef CONFIG_FAIR_GROUP_SCHED
 504         SEQ_printf(m, "\n");
 505         SEQ_printf(m, "cfs_rq[%d]:%s\n", cpu, task_group_path(cfs_rq->tg));
 506 #else
 507         SEQ_printf(m, "\n");
 508         SEQ_printf(m, "cfs_rq[%d]:\n", cpu);
 509 #endif
 510         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "exec_clock",
 511                         SPLIT_NS(cfs_rq->exec_clock));
 512
 513         raw_spin_lock_irqsave(&rq->lock, flags);
 514         if (rb_first_cached(&cfs_rq->tasks_timeline))
 515                 MIN_vruntime = (__pick_first_entity(cfs_rq))->vruntime;
 516         last = __pick_last_entity(cfs_rq);
 517         if (last)
 518                 max_vruntime = last->vruntime;
 519         min_vruntime = cfs_rq->min_vruntime;
 520         rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
 521         raw_spin_unlock_irqrestore(&rq->lock, flags);
 522         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "MIN_vruntime",
 523                         SPLIT_NS(MIN_vruntime));
 524         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "min_vruntime",
 525                         SPLIT_NS(min_vruntime));
 526         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "max_vruntime",
 527                         SPLIT_NS(max_vruntime));
 528         spread = max_vruntime - MIN_vruntime;
 529         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread",
 530                         SPLIT_NS(spread));
 531         spread0 = min_vruntime - rq0_min_vruntime;
 532         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread0",
 533                         SPLIT_NS(spread0));
 534         SEQ_printf(m, "  .%-30s: %d\n", "nr_spread_over",
 535                         cfs_rq->nr_spread_over);
 536         SEQ_printf(m, "  .%-30s: %d\n", "nr_running", cfs_rq->nr_running);
 537         SEQ_printf(m, "  .%-30s: %ld\n", "load", cfs_rq->load.weight);
 538 #ifdef CONFIG_SMP
 539         SEQ_printf(m, "  .%-30s: %ld\n", "runnable_weight", cfs_rq->runnable_weight);
 540         SEQ_printf(m, "  .%-30s: %lu\n", "load_avg",
 541                         cfs_rq->avg.load_avg);
 542         SEQ_printf(m, "  .%-30s: %lu\n", "runnable_load_avg",
 543                         cfs_rq->avg.runnable_load_avg);
 544         SEQ_printf(m, "  .%-30s: %lu\n", "util_avg",
 545                         cfs_rq->avg.util_avg);
 546         SEQ_printf(m, "  .%-30s: %u\n", "util_est_enqueued",
 547                         cfs_rq->avg.util_est.enqueued);
 548         SEQ_printf(m, "  .%-30s: %ld\n", "removed.load_avg",
 549                         cfs_rq->removed.load_avg);
 550         SEQ_printf(m, "  .%-30s: %ld\n", "removed.util_avg",
 551                         cfs_rq->removed.util_avg);
 552         SEQ_printf(m, "  .%-30s: %ld\n", "removed.runnable_sum",
 553                         cfs_rq->removed.runnable_sum);
 554 #ifdef CONFIG_FAIR_GROUP_SCHED
 555         SEQ_printf(m, "  .%-30s: %lu\n", "tg_load_avg_contrib",
 556                         cfs_rq->tg_load_avg_contrib);
 557         SEQ_printf(m, "  .%-30s: %ld\n", "tg_load_avg",
 558                         atomic_long_read(&cfs_rq->tg->load_avg));
 559 #endif
 560 #endif
 561 #ifdef CONFIG_CFS_BANDWIDTH
 562         SEQ_printf(m, "  .%-30s: %d\n", "throttled",
 563                         cfs_rq->throttled);
 564         SEQ_printf(m, "  .%-30s: %d\n", "throttle_count",
 565                         cfs_rq->throttle_count);
 566 #endif
 567
 568 #ifdef CONFIG_FAIR_GROUP_SCHED
 569         print_cfs_group_stats(m, cpu, cfs_rq->tg);
 570 #endif
 571 }
 572
 573 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
 574 {
 575 #ifdef CONFIG_RT_GROUP_SCHED
 576         SEQ_printf(m, "\n");
 577         SEQ_printf(m, "rt_rq[%d]:%s\n", cpu, task_group_path(rt_rq->tg));
 578 #else
 579         SEQ_printf(m, "\n");
 580         SEQ_printf(m, "rt_rq[%d]:\n", cpu);
 581 #endif
 582
 583 #define P(x) \
 584         SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
 585 #define PU(x) \
 586         SEQ_printf(m, "  .%-30s: %lu\n", #x, (unsigned long)(rt_rq->x))
 587 #define PN(x) \
 588         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
 589
 590         PU(rt_nr_running);
 591 #ifdef CONFIG_SMP
 592         PU(rt_nr_migratory);
 593 #endif
 594         P(rt_throttled);
 595         PN(rt_time);
 596         PN(rt_runtime);
 597
 598 #undef PN
 599 #undef PU
 600 #undef P
 601 }
 602
 603 void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq)
 604 {
 605         struct dl_bw *dl_bw;
 606
 607         SEQ_printf(m, "\n");
 608         SEQ_printf(m, "dl_rq[%d]:\n", cpu);
 609
 610 #define PU(x) \
 611         SEQ_printf(m, "  .%-30s: %lu\n", #x, (unsigned long)(dl_rq->x))
 612
 613         PU(dl_nr_running);
 614 #ifdef CONFIG_SMP
 615         PU(dl_nr_migratory);
 616         dl_bw = &cpu_rq(cpu)->rd->dl_bw;
 617 #else
 618         dl_bw = &dl_rq->dl_bw;
 619 #endif
 620         SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->bw", dl_bw->bw);
 621         SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->total_bw", dl_bw->total_bw);
 622
 623 #undef PU
 624 }
 625
 626 extern __read_mostly int sched_clock_running;
 627
 628 static void print_cpu(struct seq_file *m, int cpu)
 629 {
 630         struct rq *rq = cpu_rq(cpu);
 631         unsigned long flags;
 632
 633 #ifdef CONFIG_X86
 634         {
 635                 unsigned int freq = cpu_khz ? : 1;
 636
 637                 SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
 638                            cpu, freq / 1000, (freq % 1000));
 639         }
 640 #else
 641         SEQ_printf(m, "cpu#%d\n", cpu);
 642 #endif
 643
 644 #define P(x)                                                            \
 645 do {                                                                    \
 646         if (sizeof(rq->x) == 4)                                         \
 647                 SEQ_printf(m, "  .%-30s: %ld\n", #x, (long)(rq->x));    \
 648         else                                                            \
 649                 SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rq->x));\
 650 } while (0)
 651
 652 #define PN(x) \
 653         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
 654
 655         P(nr_running);
 656         SEQ_printf(m, "  .%-30s: %lu\n", "load",
 657                    rq->load.weight);
 658         P(nr_switches);
 659         P(nr_load_updates);
 660         P(nr_uninterruptible);
 661         PN(next_balance);
 662         SEQ_printf(m, "  .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr)));
 663         PN(clock);
 664         PN(clock_task);
 665         P(cpu_load[0]);
 666         P(cpu_load[1]);
 667         P(cpu_load[2]);
 668         P(cpu_load[3]);
 669         P(cpu_load[4]);
 670 #undef P
 671 #undef PN
 672
 673 #ifdef CONFIG_SMP
 674 #define P64(n) SEQ_printf(m, "  .%-30s: %Ld\n", #n, rq->n);
 675         P64(avg_idle);
 676         P64(max_idle_balance_cost);
 677 #undef P64
 678 #endif
 679
 680 #define P(n) SEQ_printf(m, "  .%-30s: %d\n", #n, schedstat_val(rq->n));
 681         if (schedstat_enabled()) {
 682                 P(yld_count);
 683                 P(sched_count);
 684                 P(sched_goidle);
 685                 P(ttwu_count);
 686                 P(ttwu_local);
 687         }
 688 #undef P
 689
 690         spin_lock_irqsave(&sched_debug_lock, flags);
 691         print_cfs_stats(m, cpu);
 692         print_rt_stats(m, cpu);
 693         print_dl_stats(m, cpu);
 694
 695         print_rq(m, rq, cpu);
 696         spin_unlock_irqrestore(&sched_debug_lock, flags);
 697         SEQ_printf(m, "\n");
 698 }
 699
 700 static const char *sched_tunable_scaling_names[] = {
 701         "none",
 702         "logaritmic",
 703         "linear"
 704 };
 705
 706 static void sched_debug_header(struct seq_file *m)
 707 {
 708         u64 ktime, sched_clk, cpu_clk;
 709         unsigned long flags;
 710
 711         local_irq_save(flags);
 712         ktime = ktime_to_ns(ktime_get());
 713         sched_clk = sched_clock();
 714         cpu_clk = local_clock();
 715         local_irq_restore(flags);
 716
 717         SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n",
 718                 init_utsname()->release,
 719                 (int)strcspn(init_utsname()->version, " "),
 720                 init_utsname()->version);
 721
 722 #define P(x) \
 723         SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
 724 #define PN(x) \
 725         SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
 726         PN(ktime);
 727         PN(sched_clk);
 728         PN(cpu_clk);
 729         P(jiffies);
 730 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
 731         P(sched_clock_stable());
 732 #endif
 733 #undef PN
 734 #undef P
 735
 736         SEQ_printf(m, "\n");
 737         SEQ_printf(m, "sysctl_sched\n");
 738
 739 #define P(x) \
 740         SEQ_printf(m, "  .%-40s: %Ld\n", #x, (long long)(x))
 741 #define PN(x) \
 742         SEQ_printf(m, "  .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
 743         PN(sysctl_sched_latency);
 744         PN(sysctl_sched_min_granularity);
 745         PN(sysctl_sched_wakeup_granularity);
 746         P(sysctl_sched_child_runs_first);
 747         P(sysctl_sched_features);
 748 #undef PN
 749 #undef P
 750
 751         SEQ_printf(m, "  .%-40s: %d (%s)\n",
 752                 "sysctl_sched_tunable_scaling",
 753                 sysctl_sched_tunable_scaling,
 754                 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
 755         SEQ_printf(m, "\n");
 756 }
 757
 758 static int sched_debug_show(struct seq_file *m, void *v)
 759 {
 760         int cpu = (unsigned long)(v - 2);
 761
 762         if (cpu != -1)
 763                 print_cpu(m, cpu);
 764         else
 765                 sched_debug_header(m);
 766
 767         return 0;
 768 }
 769
 770 void sysrq_sched_debug_show(void)
 771 {
 772         int cpu;
 773
 774         sched_debug_header(NULL);
 775         for_each_online_cpu(cpu)
 776                 print_cpu(NULL, cpu);
 777
 778 }
 779
 780 /*
 781  * This itererator needs some explanation.
 782  * It returns 1 for the header position.
 783  * This means 2 is CPU 0.
 784  * In a hotplugged system some CPUs, including CPU 0, may be missing so we have
 785  * to use cpumask_* to iterate over the CPUs.
 786  */
 787 static void *sched_debug_start(struct seq_file *file, loff_t *offset)
 788 {
 789         unsigned long n = *offset;
 790
 791         if (n == 0)
 792                 return (void *) 1;
 793
 794         n--;
 795
 796         if (n > 0)
 797                 n = cpumask_next(n - 1, cpu_online_mask);
 798         else
 799                 n = cpumask_first(cpu_online_mask);
 800
 801         *offset = n + 1;
 802
 803         if (n < nr_cpu_ids)
 804                 return (void *)(unsigned long)(n + 2);
 805
 806         return NULL;
 807 }
 808
 809 static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
 810 {
 811         (*offset)++;
 812         return sched_debug_start(file, offset);
 813 }
 814
 815 static void sched_debug_stop(struct seq_file *file, void *data)
 816 {
 817 }
 818
 819 static const struct seq_operations sched_debug_sops = {
 820         .start          = sched_debug_start,
 821         .next           = sched_debug_next,
 822         .stop           = sched_debug_stop,
 823         .show           = sched_debug_show,
 824 };
 825
 826 static int __init init_sched_debug_procfs(void)
 827 {
 828         if (!proc_create_seq("sched_debug", 0444, NULL, &sched_debug_sops))
 829                 return -ENOMEM;
 830         return 0;
 831 }
 832
 833 __initcall(init_sched_debug_procfs);
 834
 835 #define __P(F)  SEQ_printf(m, "%-45s:%21Ld\n",       #F, (long long)F)
 836 #define   P(F)  SEQ_printf(m, "%-45s:%21Ld\n",       #F, (long long)p->F)
 837 #define __PN(F) SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
 838 #define   PN(F) SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
 839
 840
 841 #ifdef CONFIG_NUMA_BALANCING
 842 void print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
 843                 unsigned long tpf, unsigned long gsf, unsigned long gpf)
 844 {
 845         SEQ_printf(m, "numa_faults node=%d ", node);
 846         SEQ_printf(m, "task_private=%lu task_shared=%lu ", tsf, tpf);
 847         SEQ_printf(m, "group_private=%lu group_shared=%lu\n", gsf, gpf);
 848 }
 849 #endif
 850
 851
 852 static void sched_show_numa(struct task_struct *p, struct seq_file *m)
 853 {
 854 #ifdef CONFIG_NUMA_BALANCING
 855         struct mempolicy *pol;
 856
 857         if (p->mm)
 858                 P(mm->numa_scan_seq);
 859
 860         task_lock(p);
 861         pol = p->mempolicy;
 862         if (pol && !(pol->flags & MPOL_F_MORON))
 863                 pol = NULL;
 864         mpol_get(pol);
 865         task_unlock(p);
 866
 867         P(numa_pages_migrated);
 868         P(numa_preferred_nid);
 869         P(total_numa_faults);
 870         SEQ_printf(m, "current_node=%d, numa_group_id=%d\n",
 871                         task_node(p), task_numa_group_id(p));
 872         show_numa_stats(p, m);
 873         mpol_put(pol);
 874 #endif
 875 }
 876
 877 void proc_sched_show_task(struct task_struct *p, struct pid_namespace *ns,
 878                                                   struct seq_file *m)
 879 {
 880         unsigned long nr_switches;
 881
 882         SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr_ns(p, ns),
 883                                                 get_nr_threads(p));
 884         SEQ_printf(m,
 885                 "---------------------------------------------------------"
 886                 "----------\n");
 887 #define __P(F) \
 888         SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)F)
 889 #define P(F) \
 890         SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)p->F)
 891 #define P_SCHEDSTAT(F) \
 892         SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)schedstat_val(p->F))
 893 #define __PN(F) \
 894         SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
 895 #define PN(F) \
 896         SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
 897 #define PN_SCHEDSTAT(F) \
 898         SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)schedstat_val(p->F)))
 899
 900         PN(se.exec_start);
 901         PN(se.vruntime);
 902         PN(se.sum_exec_runtime);
 903
 904         nr_switches = p->nvcsw + p->nivcsw;
 905
 906         P(se.nr_migrations);
 907
 908         if (schedstat_enabled()) {
 909                 u64 avg_atom, avg_per_cpu;
 910
 911                 PN_SCHEDSTAT(se.statistics.sum_sleep_runtime);
 912                 PN_SCHEDSTAT(se.statistics.wait_start);
 913                 PN_SCHEDSTAT(se.statistics.sleep_start);
 914                 PN_SCHEDSTAT(se.statistics.block_start);
 915                 PN_SCHEDSTAT(se.statistics.sleep_max);
 916                 PN_SCHEDSTAT(se.statistics.block_max);
 917                 PN_SCHEDSTAT(se.statistics.exec_max);
 918                 PN_SCHEDSTAT(se.statistics.slice_max);
 919                 PN_SCHEDSTAT(se.statistics.wait_max);
 920                 PN_SCHEDSTAT(se.statistics.wait_sum);
 921                 P_SCHEDSTAT(se.statistics.wait_count);
 922                 PN_SCHEDSTAT(se.statistics.iowait_sum);
 923                 P_SCHEDSTAT(se.statistics.iowait_count);
 924                 P_SCHEDSTAT(se.statistics.nr_migrations_cold);
 925                 P_SCHEDSTAT(se.statistics.nr_failed_migrations_affine);
 926                 P_SCHEDSTAT(se.statistics.nr_failed_migrations_running);
 927                 P_SCHEDSTAT(se.statistics.nr_failed_migrations_hot);
 928                 P_SCHEDSTAT(se.statistics.nr_forced_migrations);
 929                 P_SCHEDSTAT(se.statistics.nr_wakeups);
 930                 P_SCHEDSTAT(se.statistics.nr_wakeups_sync);
 931                 P_SCHEDSTAT(se.statistics.nr_wakeups_migrate);
 932                 P_SCHEDSTAT(se.statistics.nr_wakeups_local);
 933                 P_SCHEDSTAT(se.statistics.nr_wakeups_remote);
 934                 P_SCHEDSTAT(se.statistics.nr_wakeups_affine);
 935                 P_SCHEDSTAT(se.statistics.nr_wakeups_affine_attempts);
 936                 P_SCHEDSTAT(se.statistics.nr_wakeups_passive);
 937                 P_SCHEDSTAT(se.statistics.nr_wakeups_idle);
 938
 939                 avg_atom = p->se.sum_exec_runtime;
 940                 if (nr_switches)
 941                         avg_atom = div64_ul(avg_atom, nr_switches);
 942                 else
 943                         avg_atom = -1LL;
 944
 945                 avg_per_cpu = p->se.sum_exec_runtime;
 946                 if (p->se.nr_migrations) {
 947                         avg_per_cpu = div64_u64(avg_per_cpu,
 948                                                 p->se.nr_migrations);
 949                 } else {
 950                         avg_per_cpu = -1LL;
 951                 }
 952
 953                 __PN(avg_atom);
 954                 __PN(avg_per_cpu);
 955         }
 956
 957         __P(nr_switches);
 958         SEQ_printf(m, "%-45s:%21Ld\n",
 959                    "nr_voluntary_switches", (long long)p->nvcsw);
 960         SEQ_printf(m, "%-45s:%21Ld\n",
 961                    "nr_involuntary_switches", (long long)p->nivcsw);
 962
 963         P(se.load.weight);
 964         P(se.runnable_weight);
 965 #ifdef CONFIG_SMP
 966         P(se.avg.load_sum);
 967         P(se.avg.runnable_load_sum);
 968         P(se.avg.util_sum);
 969         P(se.avg.load_avg);
 970         P(se.avg.runnable_load_avg);
 971         P(se.avg.util_avg);
 972         P(se.avg.last_update_time);
 973         P(se.avg.util_est.ewma);
 974         P(se.avg.util_est.enqueued);
 975 #endif
 976         P(policy);
 977         P(prio);
 978         if (p->policy == SCHED_DEADLINE) {
 979                 P(dl.runtime);
 980                 P(dl.deadline);
 981         }
 982 #undef PN_SCHEDSTAT
 983 #undef PN
 984 #undef __PN
 985 #undef P_SCHEDSTAT
 986 #undef P
 987 #undef __P
 988
 989         {
 990                 unsigned int this_cpu = raw_smp_processor_id();
 991                 u64 t0, t1;
 992
 993                 t0 = cpu_clock(this_cpu);
 994                 t1 = cpu_clock(this_cpu);
 995                 SEQ_printf(m, "%-45s:%21Ld\n",
 996                            "clock-delta", (long long)(t1-t0));
 997         }
 998
 999         sched_show_numa(p, m);
1000 }
1001
1002 void proc_sched_set_task(struct task_struct *p)
1003 {
1004 #ifdef CONFIG_SCHEDSTATS
1005         memset(&p->se.statistics, 0, sizeof(p->se.statistics));
1006 #endif
1007 }