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 CONFIG_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_cpuslocked(&sched_feat_keys[i]);
  93 }
  94
  95 static void sched_feat_enable(int i)
  96 {
  97         static_key_enable_cpuslocked(&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 /* CONFIG_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         i = match_string(sched_feat_names, __SCHED_FEAT_NR, cmp);
 115         if (i < 0)
 116                 return i;
 117
 118         if (neg) {
 119                 sysctl_sched_features &= ~(1UL << i);
 120                 sched_feat_disable(i);
 121         } else {
 122                 sysctl_sched_features |= (1UL << i);
 123                 sched_feat_enable(i);
 124         }
 125
 126         return 0;
 127 }
 128
 129 static ssize_t
 130 sched_feat_write(struct file *filp, const char __user *ubuf,
 131                 size_t cnt, loff_t *ppos)
 132 {
 133         char buf[64];
 134         char *cmp;
 135         int ret;
 136         struct inode *inode;
 137
 138         if (cnt > 63)
 139                 cnt = 63;
 140
 141         if (copy_from_user(&buf, ubuf, cnt))
 142                 return -EFAULT;
 143
 144         buf[cnt] = 0;
 145         cmp = strstrip(buf);
 146
 147         /* Ensure the static_key remains in a consistent state */
 148         inode = file_inode(filp);
 149         cpus_read_lock();
 150         inode_lock(inode);
 151         ret = sched_feat_set(cmp);
 152         inode_unlock(inode);
 153         cpus_read_unlock();
 154         if (ret < 0)
 155                 return ret;
 156
 157         *ppos += cnt;
 158
 159         return cnt;
 160 }
 161
 162 static int sched_feat_open(struct inode *inode, struct file *filp)
 163 {
 164         return single_open(filp, sched_feat_show, NULL);
 165 }
 166
 167 static const struct file_operations sched_feat_fops = {
 168         .open           = sched_feat_open,
 169         .write          = sched_feat_write,
 170         .read           = seq_read,
 171         .llseek         = seq_lseek,
 172         .release        = single_release,
 173 };
 174
 175 __read_mostly bool sched_debug_enabled;
 176
 177 static __init int sched_init_debug(void)
 178 {
 179         debugfs_create_file("sched_features", 0644, NULL, NULL,
 180                         &sched_feat_fops);
 181
 182         debugfs_create_bool("sched_debug", 0644, NULL,
 183                         &sched_debug_enabled);
 184
 185         return 0;
 186 }
 187 late_initcall(sched_init_debug);
 188
 189 #ifdef CONFIG_SMP
 190
 191 #ifdef CONFIG_SYSCTL
 192
 193 static struct ctl_table sd_ctl_dir[] = {
 194         {
 195                 .procname       = "sched_domain",
 196                 .mode           = 0555,
 197         },
 198         {}
 199 };
 200
 201 static struct ctl_table sd_ctl_root[] = {
 202         {
 203                 .procname       = "kernel",
 204                 .mode           = 0555,
 205                 .child          = sd_ctl_dir,
 206         },
 207         {}
 208 };
 209
 210 static struct ctl_table *sd_alloc_ctl_entry(int n)
 211 {
 212         struct ctl_table *entry =
 213                 kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL);
 214
 215         return entry;
 216 }
 217
 218 static void sd_free_ctl_entry(struct ctl_table **tablep)
 219 {
 220         struct ctl_table *entry;
 221
 222         /*
 223          * In the intermediate directories, both the child directory and
 224          * procname are dynamically allocated and could fail but the mode
 225          * will always be set. In the lowest directory the names are
 226          * static strings and all have proc handlers.
 227          */
 228         for (entry = *tablep; entry->mode; entry++) {
 229                 if (entry->child)
 230                         sd_free_ctl_entry(&entry->child);
 231                 if (entry->proc_handler == NULL)
 232                         kfree(entry->procname);
 233         }
 234
 235         kfree(*tablep);
 236         *tablep = NULL;
 237 }
 238
 239 static int min_load_idx = 0;
 240 static int max_load_idx = CPU_LOAD_IDX_MAX-1;
 241
 242 static void
 243 set_table_entry(struct ctl_table *entry,
 244                 const char *procname, void *data, int maxlen,
 245                 umode_t mode, proc_handler *proc_handler,
 246                 bool load_idx)
 247 {
 248         entry->procname = procname;
 249         entry->data = data;
 250         entry->maxlen = maxlen;
 251         entry->mode = mode;
 252         entry->proc_handler = proc_handler;
 253
 254         if (load_idx) {
 255                 entry->extra1 = &min_load_idx;
 256                 entry->extra2 = &max_load_idx;
 257         }
 258 }
 259
 260 static struct ctl_table *
 261 sd_alloc_ctl_domain_table(struct sched_domain *sd)
 262 {
 263         struct ctl_table *table = sd_alloc_ctl_entry(14);
 264
 265         if (table == NULL)
 266                 return NULL;
 267
 268         set_table_entry(&table[0] , "min_interval",        &sd->min_interval,        sizeof(long), 0644, proc_doulongvec_minmax, false);
 269         set_table_entry(&table[1] , "max_interval",        &sd->max_interval,        sizeof(long), 0644, proc_doulongvec_minmax, false);
 270         set_table_entry(&table[2] , "busy_idx",            &sd->busy_idx,            sizeof(int) , 0644, proc_dointvec_minmax,   true );
 271         set_table_entry(&table[3] , "idle_idx",            &sd->idle_idx,            sizeof(int) , 0644, proc_dointvec_minmax,   true );
 272         set_table_entry(&table[4] , "newidle_idx",         &sd->newidle_idx,         sizeof(int) , 0644, proc_dointvec_minmax,   true );
 273         set_table_entry(&table[5] , "wake_idx",            &sd->wake_idx,            sizeof(int) , 0644, proc_dointvec_minmax,   true );
 274         set_table_entry(&table[6] , "forkexec_idx",        &sd->forkexec_idx,        sizeof(int) , 0644, proc_dointvec_minmax,   true );
 275         set_table_entry(&table[7] , "busy_factor",         &sd->busy_factor,         sizeof(int) , 0644, proc_dointvec_minmax,   false);
 276         set_table_entry(&table[8] , "imbalance_pct",       &sd->imbalance_pct,       sizeof(int) , 0644, proc_dointvec_minmax,   false);
 277         set_table_entry(&table[9] , "cache_nice_tries",    &sd->cache_nice_tries,    sizeof(int) , 0644, proc_dointvec_minmax,   false);
 278         set_table_entry(&table[10], "flags",               &sd->flags,               sizeof(int) , 0644, proc_dointvec_minmax,   false);
 279         set_table_entry(&table[11], "max_newidle_lb_cost", &sd->max_newidle_lb_cost, sizeof(long), 0644, proc_doulongvec_minmax, false);
 280         set_table_entry(&table[12], "name",                sd->name,            CORENAME_MAX_SIZE, 0444, proc_dostring,          false);
 281         /* &table[13] is terminator */
 282
 283         return table;
 284 }
 285
 286 static struct ctl_table *sd_alloc_ctl_cpu_table(int cpu)
 287 {
 288         struct ctl_table *entry, *table;
 289         struct sched_domain *sd;
 290         int domain_num = 0, i;
 291         char buf[32];
 292
 293         for_each_domain(cpu, sd)
 294                 domain_num++;
 295         entry = table = sd_alloc_ctl_entry(domain_num + 1);
 296         if (table == NULL)
 297                 return NULL;
 298
 299         i = 0;
 300         for_each_domain(cpu, sd) {
 301                 snprintf(buf, 32, "domain%d", i);
 302                 entry->procname = kstrdup(buf, GFP_KERNEL);
 303                 entry->mode = 0555;
 304                 entry->child = sd_alloc_ctl_domain_table(sd);
 305                 entry++;
 306                 i++;
 307         }
 308         return table;
 309 }
 310
 311 static cpumask_var_t            sd_sysctl_cpus;
 312 static struct ctl_table_header  *sd_sysctl_header;
 313
 314 void register_sched_domain_sysctl(void)
 315 {
 316         static struct ctl_table *cpu_entries;
 317         static struct ctl_table **cpu_idx;
 318         static bool init_done = false;
 319         char buf[32];
 320         int i;
 321
 322         if (!cpu_entries) {
 323                 cpu_entries = sd_alloc_ctl_entry(num_possible_cpus() + 1);
 324                 if (!cpu_entries)
 325                         return;
 326
 327                 WARN_ON(sd_ctl_dir[0].child);
 328                 sd_ctl_dir[0].child = cpu_entries;
 329         }
 330
 331         if (!cpu_idx) {
 332                 struct ctl_table *e = cpu_entries;
 333
 334                 cpu_idx = kcalloc(nr_cpu_ids, sizeof(struct ctl_table*), GFP_KERNEL);
 335                 if (!cpu_idx)
 336                         return;
 337
 338                 /* deal with sparse possible map */
 339                 for_each_possible_cpu(i) {
 340                         cpu_idx[i] = e;
 341                         e++;
 342                 }
 343         }
 344
 345         if (!cpumask_available(sd_sysctl_cpus)) {
 346                 if (!alloc_cpumask_var(&sd_sysctl_cpus, GFP_KERNEL))
 347                         return;
 348         }
 349
 350         if (!init_done) {
 351                 init_done = true;
 352                 /* init to possible to not have holes in @cpu_entries */
 353                 cpumask_copy(sd_sysctl_cpus, cpu_possible_mask);
 354         }
 355
 356         for_each_cpu(i, sd_sysctl_cpus) {
 357                 struct ctl_table *e = cpu_idx[i];
 358
 359                 if (e->child)
 360                         sd_free_ctl_entry(&e->child);
 361
 362                 if (!e->procname) {
 363                         snprintf(buf, 32, "cpu%d", i);
 364                         e->procname = kstrdup(buf, GFP_KERNEL);
 365                 }
 366                 e->mode = 0555;
 367                 e->child = sd_alloc_ctl_cpu_table(i);
 368
 369                 __cpumask_clear_cpu(i, sd_sysctl_cpus);
 370         }
 371
 372         WARN_ON(sd_sysctl_header);
 373         sd_sysctl_header = register_sysctl_table(sd_ctl_root);
 374 }
 375
 376 void dirty_sched_domain_sysctl(int cpu)
 377 {
 378         if (cpumask_available(sd_sysctl_cpus))
 379                 __cpumask_set_cpu(cpu, sd_sysctl_cpus);
 380 }
 381
 382 /* may be called multiple times per register */
 383 void unregister_sched_domain_sysctl(void)
 384 {
 385         unregister_sysctl_table(sd_sysctl_header);
 386         sd_sysctl_header = NULL;
 387 }
 388 #endif /* CONFIG_SYSCTL */
 389 #endif /* CONFIG_SMP */
 390
 391 #ifdef CONFIG_FAIR_GROUP_SCHED
 392 static void print_cfs_group_stats(struct seq_file *m, int cpu, struct task_group *tg)
 393 {
 394         struct sched_entity *se = tg->se[cpu];
 395
 396 #define P(F)            SEQ_printf(m, "  .%-30s: %lld\n",       #F, (long long)F)
 397 #define P_SCHEDSTAT(F)  SEQ_printf(m, "  .%-30s: %lld\n",       #F, (long long)schedstat_val(F))
 398 #define PN(F)           SEQ_printf(m, "  .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F))
 399 #define PN_SCHEDSTAT(F) SEQ_printf(m, "  .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)schedstat_val(F)))
 400
 401         if (!se)
 402                 return;
 403
 404         PN(se->exec_start);
 405         PN(se->vruntime);
 406         PN(se->sum_exec_runtime);
 407
 408         if (schedstat_enabled()) {
 409                 PN_SCHEDSTAT(se->statistics.wait_start);
 410                 PN_SCHEDSTAT(se->statistics.sleep_start);
 411                 PN_SCHEDSTAT(se->statistics.block_start);
 412                 PN_SCHEDSTAT(se->statistics.sleep_max);
 413                 PN_SCHEDSTAT(se->statistics.block_max);
 414                 PN_SCHEDSTAT(se->statistics.exec_max);
 415                 PN_SCHEDSTAT(se->statistics.slice_max);
 416                 PN_SCHEDSTAT(se->statistics.wait_max);
 417                 PN_SCHEDSTAT(se->statistics.wait_sum);
 418                 P_SCHEDSTAT(se->statistics.wait_count);
 419         }
 420
 421         P(se->load.weight);
 422         P(se->runnable_weight);
 423 #ifdef CONFIG_SMP
 424         P(se->avg.load_avg);
 425         P(se->avg.util_avg);
 426         P(se->avg.runnable_load_avg);
 427 #endif
 428
 429 #undef PN_SCHEDSTAT
 430 #undef PN
 431 #undef P_SCHEDSTAT
 432 #undef P
 433 }
 434 #endif
 435
 436 #ifdef CONFIG_CGROUP_SCHED
 437 static char group_path[PATH_MAX];
 438
 439 static char *task_group_path(struct task_group *tg)
 440 {
 441         if (autogroup_path(tg, group_path, PATH_MAX))
 442                 return group_path;
 443
 444         cgroup_path(tg->css.cgroup, group_path, PATH_MAX);
 445
 446         return group_path;
 447 }
 448 #endif
 449
 450 static void
 451 print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
 452 {
 453         if (rq->curr == p)
 454                 SEQ_printf(m, ">R");
 455         else
 456                 SEQ_printf(m, " %c", task_state_to_char(p));
 457
 458         SEQ_printf(m, "%15s %5d %9Ld.%06ld %9Ld %5d ",
 459                 p->comm, task_pid_nr(p),
 460                 SPLIT_NS(p->se.vruntime),
 461                 (long long)(p->nvcsw + p->nivcsw),
 462                 p->prio);
 463
 464         SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
 465                 SPLIT_NS(schedstat_val_or_zero(p->se.statistics.wait_sum)),
 466                 SPLIT_NS(p->se.sum_exec_runtime),
 467                 SPLIT_NS(schedstat_val_or_zero(p->se.statistics.sum_sleep_runtime)));
 468
 469 #ifdef CONFIG_NUMA_BALANCING
 470         SEQ_printf(m, " %d %d", task_node(p), task_numa_group_id(p));
 471 #endif
 472 #ifdef CONFIG_CGROUP_SCHED
 473         SEQ_printf(m, " %s", task_group_path(task_group(p)));
 474 #endif
 475
 476         SEQ_printf(m, "\n");
 477 }
 478
 479 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
 480 {
 481         struct task_struct *g, *p;
 482
 483         SEQ_printf(m, "\n");
 484         SEQ_printf(m, "runnable tasks:\n");
 485         SEQ_printf(m, " S           task   PID         tree-key  switches  prio"
 486                    "     wait-time             sum-exec        sum-sleep\n");
 487         SEQ_printf(m, "-------------------------------------------------------"
 488                    "----------------------------------------------------\n");
 489
 490         rcu_read_lock();
 491         for_each_process_thread(g, p) {
 492                 if (task_cpu(p) != rq_cpu)
 493                         continue;
 494
 495                 print_task(m, rq, p);
 496         }
 497         rcu_read_unlock();
 498 }
 499
 500 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
 501 {
 502         s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
 503                 spread, rq0_min_vruntime, spread0;
 504         struct rq *rq = cpu_rq(cpu);
 505         struct sched_entity *last;
 506         unsigned long flags;
 507
 508 #ifdef CONFIG_FAIR_GROUP_SCHED
 509         SEQ_printf(m, "\n");
 510         SEQ_printf(m, "cfs_rq[%d]:%s\n", cpu, task_group_path(cfs_rq->tg));
 511 #else
 512         SEQ_printf(m, "\n");
 513         SEQ_printf(m, "cfs_rq[%d]:\n", cpu);
 514 #endif
 515         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "exec_clock",
 516                         SPLIT_NS(cfs_rq->exec_clock));
 517
 518         raw_spin_lock_irqsave(&rq->lock, flags);
 519         if (rb_first_cached(&cfs_rq->tasks_timeline))
 520                 MIN_vruntime = (__pick_first_entity(cfs_rq))->vruntime;
 521         last = __pick_last_entity(cfs_rq);
 522         if (last)
 523                 max_vruntime = last->vruntime;
 524         min_vruntime = cfs_rq->min_vruntime;
 525         rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
 526         raw_spin_unlock_irqrestore(&rq->lock, flags);
 527         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "MIN_vruntime",
 528                         SPLIT_NS(MIN_vruntime));
 529         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "min_vruntime",
 530                         SPLIT_NS(min_vruntime));
 531         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "max_vruntime",
 532                         SPLIT_NS(max_vruntime));
 533         spread = max_vruntime - MIN_vruntime;
 534         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread",
 535                         SPLIT_NS(spread));
 536         spread0 = min_vruntime - rq0_min_vruntime;
 537         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread0",
 538                         SPLIT_NS(spread0));
 539         SEQ_printf(m, "  .%-30s: %d\n", "nr_spread_over",
 540                         cfs_rq->nr_spread_over);
 541         SEQ_printf(m, "  .%-30s: %d\n", "nr_running", cfs_rq->nr_running);
 542         SEQ_printf(m, "  .%-30s: %ld\n", "load", cfs_rq->load.weight);
 543 #ifdef CONFIG_SMP
 544         SEQ_printf(m, "  .%-30s: %ld\n", "runnable_weight", cfs_rq->runnable_weight);
 545         SEQ_printf(m, "  .%-30s: %lu\n", "load_avg",
 546                         cfs_rq->avg.load_avg);
 547         SEQ_printf(m, "  .%-30s: %lu\n", "runnable_load_avg",
 548                         cfs_rq->avg.runnable_load_avg);
 549         SEQ_printf(m, "  .%-30s: %lu\n", "util_avg",
 550                         cfs_rq->avg.util_avg);
 551         SEQ_printf(m, "  .%-30s: %u\n", "util_est_enqueued",
 552                         cfs_rq->avg.util_est.enqueued);
 553         SEQ_printf(m, "  .%-30s: %ld\n", "removed.load_avg",
 554                         cfs_rq->removed.load_avg);
 555         SEQ_printf(m, "  .%-30s: %ld\n", "removed.util_avg",
 556                         cfs_rq->removed.util_avg);
 557         SEQ_printf(m, "  .%-30s: %ld\n", "removed.runnable_sum",
 558                         cfs_rq->removed.runnable_sum);
 559 #ifdef CONFIG_FAIR_GROUP_SCHED
 560         SEQ_printf(m, "  .%-30s: %lu\n", "tg_load_avg_contrib",
 561                         cfs_rq->tg_load_avg_contrib);
 562         SEQ_printf(m, "  .%-30s: %ld\n", "tg_load_avg",
 563                         atomic_long_read(&cfs_rq->tg->load_avg));
 564 #endif
 565 #endif
 566 #ifdef CONFIG_CFS_BANDWIDTH
 567         SEQ_printf(m, "  .%-30s: %d\n", "throttled",
 568                         cfs_rq->throttled);
 569         SEQ_printf(m, "  .%-30s: %d\n", "throttle_count",
 570                         cfs_rq->throttle_count);
 571 #endif
 572
 573 #ifdef CONFIG_FAIR_GROUP_SCHED
 574         print_cfs_group_stats(m, cpu, cfs_rq->tg);
 575 #endif
 576 }
 577
 578 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
 579 {
 580 #ifdef CONFIG_RT_GROUP_SCHED
 581         SEQ_printf(m, "\n");
 582         SEQ_printf(m, "rt_rq[%d]:%s\n", cpu, task_group_path(rt_rq->tg));
 583 #else
 584         SEQ_printf(m, "\n");
 585         SEQ_printf(m, "rt_rq[%d]:\n", cpu);
 586 #endif
 587
 588 #define P(x) \
 589         SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
 590 #define PU(x) \
 591         SEQ_printf(m, "  .%-30s: %lu\n", #x, (unsigned long)(rt_rq->x))
 592 #define PN(x) \
 593         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
 594
 595         PU(rt_nr_running);
 596 #ifdef CONFIG_SMP
 597         PU(rt_nr_migratory);
 598 #endif
 599         P(rt_throttled);
 600         PN(rt_time);
 601         PN(rt_runtime);
 602
 603 #undef PN
 604 #undef PU
 605 #undef P
 606 }
 607
 608 void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq)
 609 {
 610         struct dl_bw *dl_bw;
 611
 612         SEQ_printf(m, "\n");
 613         SEQ_printf(m, "dl_rq[%d]:\n", cpu);
 614
 615 #define PU(x) \
 616         SEQ_printf(m, "  .%-30s: %lu\n", #x, (unsigned long)(dl_rq->x))
 617
 618         PU(dl_nr_running);
 619 #ifdef CONFIG_SMP
 620         PU(dl_nr_migratory);
 621         dl_bw = &cpu_rq(cpu)->rd->dl_bw;
 622 #else
 623         dl_bw = &dl_rq->dl_bw;
 624 #endif
 625         SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->bw", dl_bw->bw);
 626         SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->total_bw", dl_bw->total_bw);
 627
 628 #undef PU
 629 }
 630
 631 static void print_cpu(struct seq_file *m, int cpu)
 632 {
 633         struct rq *rq = cpu_rq(cpu);
 634         unsigned long flags;
 635
 636 #ifdef CONFIG_X86
 637         {
 638                 unsigned int freq = cpu_khz ? : 1;
 639
 640                 SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
 641                            cpu, freq / 1000, (freq % 1000));
 642         }
 643 #else
 644         SEQ_printf(m, "cpu#%d\n", cpu);
 645 #endif
 646
 647 #define P(x)                                                            \
 648 do {                                                                    \
 649         if (sizeof(rq->x) == 4)                                         \
 650                 SEQ_printf(m, "  .%-30s: %ld\n", #x, (long)(rq->x));    \
 651         else                                                            \
 652                 SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rq->x));\
 653 } while (0)
 654
 655 #define PN(x) \
 656         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
 657
 658         P(nr_running);
 659         SEQ_printf(m, "  .%-30s: %lu\n", "load",
 660                    rq->load.weight);
 661         P(nr_switches);
 662         P(nr_load_updates);
 663         P(nr_uninterruptible);
 664         PN(next_balance);
 665         SEQ_printf(m, "  .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr)));
 666         PN(clock);
 667         PN(clock_task);
 668         P(cpu_load[0]);
 669         P(cpu_load[1]);
 670         P(cpu_load[2]);
 671         P(cpu_load[3]);
 672         P(cpu_load[4]);
 673 #undef P
 674 #undef PN
 675
 676 #ifdef CONFIG_SMP
 677 #define P64(n) SEQ_printf(m, "  .%-30s: %Ld\n", #n, rq->n);
 678         P64(avg_idle);
 679         P64(max_idle_balance_cost);
 680 #undef P64
 681 #endif
 682
 683 #define P(n) SEQ_printf(m, "  .%-30s: %d\n", #n, schedstat_val(rq->n));
 684         if (schedstat_enabled()) {
 685                 P(yld_count);
 686                 P(sched_count);
 687                 P(sched_goidle);
 688                 P(ttwu_count);
 689                 P(ttwu_local);
 690         }
 691 #undef P
 692
 693         spin_lock_irqsave(&sched_debug_lock, flags);
 694         print_cfs_stats(m, cpu);
 695         print_rt_stats(m, cpu);
 696         print_dl_stats(m, cpu);
 697
 698         print_rq(m, rq, cpu);
 699         spin_unlock_irqrestore(&sched_debug_lock, flags);
 700         SEQ_printf(m, "\n");
 701 }
 702
 703 static const char *sched_tunable_scaling_names[] = {
 704         "none",
 705         "logarithmic",
 706         "linear"
 707 };
 708
 709 static void sched_debug_header(struct seq_file *m)
 710 {
 711         u64 ktime, sched_clk, cpu_clk;
 712         unsigned long flags;
 713
 714         local_irq_save(flags);
 715         ktime = ktime_to_ns(ktime_get());
 716         sched_clk = sched_clock();
 717         cpu_clk = local_clock();
 718         local_irq_restore(flags);
 719
 720         SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n",
 721                 init_utsname()->release,
 722                 (int)strcspn(init_utsname()->version, " "),
 723                 init_utsname()->version);
 724
 725 #define P(x) \
 726         SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
 727 #define PN(x) \
 728         SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
 729         PN(ktime);
 730         PN(sched_clk);
 731         PN(cpu_clk);
 732         P(jiffies);
 733 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
 734         P(sched_clock_stable());
 735 #endif
 736 #undef PN
 737 #undef P
 738
 739         SEQ_printf(m, "\n");
 740         SEQ_printf(m, "sysctl_sched\n");
 741
 742 #define P(x) \
 743         SEQ_printf(m, "  .%-40s: %Ld\n", #x, (long long)(x))
 744 #define PN(x) \
 745         SEQ_printf(m, "  .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
 746         PN(sysctl_sched_latency);
 747         PN(sysctl_sched_min_granularity);
 748         PN(sysctl_sched_wakeup_granularity);
 749         P(sysctl_sched_child_runs_first);
 750         P(sysctl_sched_features);
 751 #undef PN
 752 #undef P
 753
 754         SEQ_printf(m, "  .%-40s: %d (%s)\n",
 755                 "sysctl_sched_tunable_scaling",
 756                 sysctl_sched_tunable_scaling,
 757                 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
 758         SEQ_printf(m, "\n");
 759 }
 760
 761 static int sched_debug_show(struct seq_file *m, void *v)
 762 {
 763         int cpu = (unsigned long)(v - 2);
 764
 765         if (cpu != -1)
 766                 print_cpu(m, cpu);
 767         else
 768                 sched_debug_header(m);
 769
 770         return 0;
 771 }
 772
 773 void sysrq_sched_debug_show(void)
 774 {
 775         int cpu;
 776
 777         sched_debug_header(NULL);
 778         for_each_online_cpu(cpu)
 779                 print_cpu(NULL, cpu);
 780
 781 }
 782
 783 /*
 784  * This itererator needs some explanation.
 785  * It returns 1 for the header position.
 786  * This means 2 is CPU 0.
 787  * In a hotplugged system some CPUs, including CPU 0, may be missing so we have
 788  * to use cpumask_* to iterate over the CPUs.
 789  */
 790 static void *sched_debug_start(struct seq_file *file, loff_t *offset)
 791 {
 792         unsigned long n = *offset;
 793
 794         if (n == 0)
 795                 return (void *) 1;
 796
 797         n--;
 798
 799         if (n > 0)
 800                 n = cpumask_next(n - 1, cpu_online_mask);
 801         else
 802                 n = cpumask_first(cpu_online_mask);
 803
 804         *offset = n + 1;
 805
 806         if (n < nr_cpu_ids)
 807                 return (void *)(unsigned long)(n + 2);
 808
 809         return NULL;
 810 }
 811
 812 static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
 813 {
 814         (*offset)++;
 815         return sched_debug_start(file, offset);
 816 }
 817
 818 static void sched_debug_stop(struct seq_file *file, void *data)
 819 {
 820 }
 821
 822 static const struct seq_operations sched_debug_sops = {
 823         .start          = sched_debug_start,
 824         .next           = sched_debug_next,
 825         .stop           = sched_debug_stop,
 826         .show           = sched_debug_show,
 827 };
 828
 829 static int __init init_sched_debug_procfs(void)
 830 {
 831         if (!proc_create_seq("sched_debug", 0444, NULL, &sched_debug_sops))
 832                 return -ENOMEM;
 833         return 0;
 834 }
 835
 836 __initcall(init_sched_debug_procfs);
 837
 838 #define __P(F)  SEQ_printf(m, "%-45s:%21Ld\n",       #F, (long long)F)
 839 #define   P(F)  SEQ_printf(m, "%-45s:%21Ld\n",       #F, (long long)p->F)
 840 #define __PN(F) SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
 841 #define   PN(F) SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
 842
 843
 844 #ifdef CONFIG_NUMA_BALANCING
 845 void print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
 846                 unsigned long tpf, unsigned long gsf, unsigned long gpf)
 847 {
 848         SEQ_printf(m, "numa_faults node=%d ", node);
 849         SEQ_printf(m, "task_private=%lu task_shared=%lu ", tpf, tsf);
 850         SEQ_printf(m, "group_private=%lu group_shared=%lu\n", gpf, gsf);
 851 }
 852 #endif
 853
 854
 855 static void sched_show_numa(struct task_struct *p, struct seq_file *m)
 856 {
 857 #ifdef CONFIG_NUMA_BALANCING
 858         struct mempolicy *pol;
 859
 860         if (p->mm)
 861                 P(mm->numa_scan_seq);
 862
 863         task_lock(p);
 864         pol = p->mempolicy;
 865         if (pol && !(pol->flags & MPOL_F_MORON))
 866                 pol = NULL;
 867         mpol_get(pol);
 868         task_unlock(p);
 869
 870         P(numa_pages_migrated);
 871         P(numa_preferred_nid);
 872         P(total_numa_faults);
 873         SEQ_printf(m, "current_node=%d, numa_group_id=%d\n",
 874                         task_node(p), task_numa_group_id(p));
 875         show_numa_stats(p, m);
 876         mpol_put(pol);
 877 #endif
 878 }
 879
 880 void proc_sched_show_task(struct task_struct *p, struct pid_namespace *ns,
 881                                                   struct seq_file *m)
 882 {
 883         unsigned long nr_switches;
 884
 885         SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr_ns(p, ns),
 886                                                 get_nr_threads(p));
 887         SEQ_printf(m,
 888                 "---------------------------------------------------------"
 889                 "----------\n");
 890 #define __P(F) \
 891         SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)F)
 892 #define P(F) \
 893         SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)p->F)
 894 #define P_SCHEDSTAT(F) \
 895         SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)schedstat_val(p->F))
 896 #define __PN(F) \
 897         SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
 898 #define PN(F) \
 899         SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
 900 #define PN_SCHEDSTAT(F) \
 901         SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)schedstat_val(p->F)))
 902
 903         PN(se.exec_start);
 904         PN(se.vruntime);
 905         PN(se.sum_exec_runtime);
 906
 907         nr_switches = p->nvcsw + p->nivcsw;
 908
 909         P(se.nr_migrations);
 910
 911         if (schedstat_enabled()) {
 912                 u64 avg_atom, avg_per_cpu;
 913
 914                 PN_SCHEDSTAT(se.statistics.sum_sleep_runtime);
 915                 PN_SCHEDSTAT(se.statistics.wait_start);
 916                 PN_SCHEDSTAT(se.statistics.sleep_start);
 917                 PN_SCHEDSTAT(se.statistics.block_start);
 918                 PN_SCHEDSTAT(se.statistics.sleep_max);
 919                 PN_SCHEDSTAT(se.statistics.block_max);
 920                 PN_SCHEDSTAT(se.statistics.exec_max);
 921                 PN_SCHEDSTAT(se.statistics.slice_max);
 922                 PN_SCHEDSTAT(se.statistics.wait_max);
 923                 PN_SCHEDSTAT(se.statistics.wait_sum);
 924                 P_SCHEDSTAT(se.statistics.wait_count);
 925                 PN_SCHEDSTAT(se.statistics.iowait_sum);
 926                 P_SCHEDSTAT(se.statistics.iowait_count);
 927                 P_SCHEDSTAT(se.statistics.nr_migrations_cold);
 928                 P_SCHEDSTAT(se.statistics.nr_failed_migrations_affine);
 929                 P_SCHEDSTAT(se.statistics.nr_failed_migrations_running);
 930                 P_SCHEDSTAT(se.statistics.nr_failed_migrations_hot);
 931                 P_SCHEDSTAT(se.statistics.nr_forced_migrations);
 932                 P_SCHEDSTAT(se.statistics.nr_wakeups);
 933                 P_SCHEDSTAT(se.statistics.nr_wakeups_sync);
 934                 P_SCHEDSTAT(se.statistics.nr_wakeups_migrate);
 935                 P_SCHEDSTAT(se.statistics.nr_wakeups_local);
 936                 P_SCHEDSTAT(se.statistics.nr_wakeups_remote);
 937                 P_SCHEDSTAT(se.statistics.nr_wakeups_affine);
 938                 P_SCHEDSTAT(se.statistics.nr_wakeups_affine_attempts);
 939                 P_SCHEDSTAT(se.statistics.nr_wakeups_passive);
 940                 P_SCHEDSTAT(se.statistics.nr_wakeups_idle);
 941
 942                 avg_atom = p->se.sum_exec_runtime;
 943                 if (nr_switches)
 944                         avg_atom = div64_ul(avg_atom, nr_switches);
 945                 else
 946                         avg_atom = -1LL;
 947
 948                 avg_per_cpu = p->se.sum_exec_runtime;
 949                 if (p->se.nr_migrations) {
 950                         avg_per_cpu = div64_u64(avg_per_cpu,
 951                                                 p->se.nr_migrations);
 952                 } else {
 953                         avg_per_cpu = -1LL;
 954                 }
 955
 956                 __PN(avg_atom);
 957                 __PN(avg_per_cpu);
 958         }
 959
 960         __P(nr_switches);
 961         SEQ_printf(m, "%-45s:%21Ld\n",
 962                    "nr_voluntary_switches", (long long)p->nvcsw);
 963         SEQ_printf(m, "%-45s:%21Ld\n",
 964                    "nr_involuntary_switches", (long long)p->nivcsw);
 965
 966         P(se.load.weight);
 967         P(se.runnable_weight);
 968 #ifdef CONFIG_SMP
 969         P(se.avg.load_sum);
 970         P(se.avg.runnable_load_sum);
 971         P(se.avg.util_sum);
 972         P(se.avg.load_avg);
 973         P(se.avg.runnable_load_avg);
 974         P(se.avg.util_avg);
 975         P(se.avg.last_update_time);
 976         P(se.avg.util_est.ewma);
 977         P(se.avg.util_est.enqueued);
 978 #endif
 979         P(policy);
 980         P(prio);
 981         if (task_has_dl_policy(p)) {
 982                 P(dl.runtime);
 983                 P(dl.deadline);
 984         }
 985 #undef PN_SCHEDSTAT
 986 #undef PN
 987 #undef __PN
 988 #undef P_SCHEDSTAT
 989 #undef P
 990 #undef __P
 991
 992         {
 993                 unsigned int this_cpu = raw_smp_processor_id();
 994                 u64 t0, t1;
 995
 996                 t0 = cpu_clock(this_cpu);
 997                 t1 = cpu_clock(this_cpu);
 998                 SEQ_printf(m, "%-45s:%21Ld\n",
 999                            "clock-delta", (long long)(t1-t0));
1000         }
1001
1002         sched_show_numa(p, m);
1003 }
1004
1005 void proc_sched_set_task(struct task_struct *p)
1006 {
1007 #ifdef CONFIG_SCHEDSTATS
1008         memset(&p->se.statistics, 0, sizeof(p->se.statistics));
1009 #endif
1010 }