kernel/sched/debug.c

   1 /*
   2  * kernel/sched/debug.c
   3  *
   4  * Print the CFS rbtree
   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
  13 #include <linux/proc_fs.h>
  14 #include <linux/sched.h>
  15 #include <linux/seq_file.h>
  16 #include <linux/kallsyms.h>
  17 #include <linux/utsname.h>
  18
  19 #include "sched.h"
  20
  21 static DEFINE_SPINLOCK(sched_debug_lock);
  22
  23 /*
  24  * This allows printing both to /proc/sched_debug and
  25  * to the console
  26  */
  27 #define SEQ_printf(m, x...)                     \
  28  do {                                           \
  29         if (m)                                  \
  30                 seq_printf(m, x);               \
  31         else                                    \
  32                 printk(x);                      \
  33  } while (0)
  34
  35 /*
  36  * Ease the printing of nsec fields:
  37  */
  38 static long long nsec_high(unsigned long long nsec)
  39 {
  40         if ((long long)nsec < 0) {
  41                 nsec = -nsec;
  42                 do_div(nsec, 1000000);
  43                 return -nsec;
  44         }
  45         do_div(nsec, 1000000);
  46
  47         return nsec;
  48 }
  49
  50 static unsigned long nsec_low(unsigned long long nsec)
  51 {
  52         if ((long long)nsec < 0)
  53                 nsec = -nsec;
  54
  55         return do_div(nsec, 1000000);
  56 }
  57
  58 #define SPLIT_NS(x) nsec_high(x), nsec_low(x)
  59
  60 #ifdef CONFIG_FAIR_GROUP_SCHED
  61 static void print_cfs_group_stats(struct seq_file *m, int cpu, struct task_group *tg)
  62 {
  63         struct sched_entity *se = tg->se[cpu];
  64
  65 #define P(F) \
  66         SEQ_printf(m, "  .%-30s: %lld\n", #F, (long long)F)
  67 #define PN(F) \
  68         SEQ_printf(m, "  .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F))
  69
  70         if (!se) {
  71                 struct sched_avg *avg = &cpu_rq(cpu)->avg;
  72                 P(avg->runnable_avg_sum);
  73                 P(avg->runnable_avg_period);
  74                 return;
  75         }
  76
  77
  78         PN(se->exec_start);
  79         PN(se->vruntime);
  80         PN(se->sum_exec_runtime);
  81 #ifdef CONFIG_SCHEDSTATS
  82         PN(se->statistics.wait_start);
  83         PN(se->statistics.sleep_start);
  84         PN(se->statistics.block_start);
  85         PN(se->statistics.sleep_max);
  86         PN(se->statistics.block_max);
  87         PN(se->statistics.exec_max);
  88         PN(se->statistics.slice_max);
  89         PN(se->statistics.wait_max);
  90         PN(se->statistics.wait_sum);
  91         P(se->statistics.wait_count);
  92 #endif
  93         P(se->load.weight);
  94 #ifdef CONFIG_SMP
  95         P(se->avg.runnable_avg_sum);
  96         P(se->avg.runnable_avg_period);
  97         P(se->avg.usage_avg_sum);
  98         P(se->avg.load_avg_contrib);
  99         P(se->avg.decay_count);
 100 #endif
 101 #undef PN
 102 #undef P
 103 }
 104 #endif
 105
 106 #ifdef CONFIG_CGROUP_SCHED
 107 static char group_path[PATH_MAX];
 108
 109 static char *task_group_path(struct task_group *tg)
 110 {
 111         if (autogroup_path(tg, group_path, PATH_MAX))
 112                 return group_path;
 113
 114         cgroup_path(tg->css.cgroup, group_path, PATH_MAX);
 115         return group_path;
 116 }
 117 #endif
 118
 119 static void
 120 print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
 121 {
 122         if (rq->curr == p)
 123                 SEQ_printf(m, "R");
 124         else
 125                 SEQ_printf(m, " ");
 126
 127         SEQ_printf(m, "%15s %5d %9Ld.%06ld %9Ld %5d ",
 128                 p->comm, p->pid,
 129                 SPLIT_NS(p->se.vruntime),
 130                 (long long)(p->nvcsw + p->nivcsw),
 131                 p->prio);
 132 #ifdef CONFIG_SCHEDSTATS
 133         SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
 134                 SPLIT_NS(p->se.vruntime),
 135                 SPLIT_NS(p->se.sum_exec_runtime),
 136                 SPLIT_NS(p->se.statistics.sum_sleep_runtime));
 137 #else
 138         SEQ_printf(m, "%15Ld %15Ld %15Ld.%06ld %15Ld.%06ld %15Ld.%06ld",
 139                 0LL, 0LL, 0LL, 0L, 0LL, 0L, 0LL, 0L);
 140 #endif
 141 #ifdef CONFIG_CGROUP_SCHED
 142         SEQ_printf(m, " %s", task_group_path(task_group(p)));
 143 #endif
 144
 145         SEQ_printf(m, "\n");
 146 }
 147
 148 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
 149 {
 150         struct task_struct *g, *p;
 151         unsigned long flags;
 152
 153         SEQ_printf(m,
 154         "\nrunnable tasks:\n"
 155         "            task   PID         tree-key  switches  prio"
 156         "     exec-runtime         sum-exec        sum-sleep\n"
 157         "------------------------------------------------------"
 158         "----------------------------------------------------\n");
 159
 160         read_lock_irqsave(&tasklist_lock, flags);
 161
 162         do_each_thread(g, p) {
 163                 if (!p->on_rq || task_cpu(p) != rq_cpu)
 164                         continue;
 165
 166                 print_task(m, rq, p);
 167         } while_each_thread(g, p);
 168
 169         read_unlock_irqrestore(&tasklist_lock, flags);
 170 }
 171
 172 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
 173 {
 174         s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
 175                 spread, rq0_min_vruntime, spread0;
 176         struct rq *rq = cpu_rq(cpu);
 177         struct sched_entity *last;
 178         unsigned long flags;
 179
 180 #ifdef CONFIG_FAIR_GROUP_SCHED
 181         SEQ_printf(m, "\ncfs_rq[%d]:%s\n", cpu, task_group_path(cfs_rq->tg));
 182 #else
 183         SEQ_printf(m, "\ncfs_rq[%d]:\n", cpu);
 184 #endif
 185         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "exec_clock",
 186                         SPLIT_NS(cfs_rq->exec_clock));
 187
 188         raw_spin_lock_irqsave(&rq->lock, flags);
 189         if (cfs_rq->rb_leftmost)
 190                 MIN_vruntime = (__pick_first_entity(cfs_rq))->vruntime;
 191         last = __pick_last_entity(cfs_rq);
 192         if (last)
 193                 max_vruntime = last->vruntime;
 194         min_vruntime = cfs_rq->min_vruntime;
 195         rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
 196         raw_spin_unlock_irqrestore(&rq->lock, flags);
 197         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "MIN_vruntime",
 198                         SPLIT_NS(MIN_vruntime));
 199         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "min_vruntime",
 200                         SPLIT_NS(min_vruntime));
 201         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "max_vruntime",
 202                         SPLIT_NS(max_vruntime));
 203         spread = max_vruntime - MIN_vruntime;
 204         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread",
 205                         SPLIT_NS(spread));
 206         spread0 = min_vruntime - rq0_min_vruntime;
 207         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread0",
 208                         SPLIT_NS(spread0));
 209         SEQ_printf(m, "  .%-30s: %d\n", "nr_spread_over",
 210                         cfs_rq->nr_spread_over);
 211         SEQ_printf(m, "  .%-30s: %d\n", "nr_running", cfs_rq->nr_running);
 212         SEQ_printf(m, "  .%-30s: %ld\n", "load", cfs_rq->load.weight);
 213 #ifdef CONFIG_FAIR_GROUP_SCHED
 214 #ifdef CONFIG_SMP
 215         SEQ_printf(m, "  .%-30s: %lld\n", "runnable_load_avg",
 216                         cfs_rq->runnable_load_avg);
 217         SEQ_printf(m, "  .%-30s: %lld\n", "blocked_load_avg",
 218                         cfs_rq->blocked_load_avg);
 219         SEQ_printf(m, "  .%-30s: %lld\n", "tg_load_avg",
 220                         (unsigned long long)atomic64_read(&cfs_rq->tg->load_avg));
 221         SEQ_printf(m, "  .%-30s: %lld\n", "tg_load_contrib",
 222                         cfs_rq->tg_load_contrib);
 223         SEQ_printf(m, "  .%-30s: %d\n", "tg_runnable_contrib",
 224                         cfs_rq->tg_runnable_contrib);
 225         SEQ_printf(m, "  .%-30s: %d\n", "tg->runnable_avg",
 226                         atomic_read(&cfs_rq->tg->runnable_avg));
 227         SEQ_printf(m, "  .%-30s: %d\n", "tg->usage_avg",
 228                         atomic_read(&cfs_rq->tg->usage_avg));
 229 #endif
 230 #ifdef CONFIG_CFS_BANDWIDTH
 231         SEQ_printf(m, "  .%-30s: %d\n", "tg->cfs_bandwidth.timer_active",
 232                         cfs_rq->tg->cfs_bandwidth.timer_active);
 233         SEQ_printf(m, "  .%-30s: %d\n", "throttled",
 234                         cfs_rq->throttled);
 235         SEQ_printf(m, "  .%-30s: %d\n", "throttle_count",
 236                         cfs_rq->throttle_count);
 237 #endif
 238
 239         print_cfs_group_stats(m, cpu, cfs_rq->tg);
 240 #endif
 241 }
 242
 243 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
 244 {
 245 #ifdef CONFIG_RT_GROUP_SCHED
 246         SEQ_printf(m, "\nrt_rq[%d]:%s\n", cpu, task_group_path(rt_rq->tg));
 247 #else
 248         SEQ_printf(m, "\nrt_rq[%d]:\n", cpu);
 249 #endif
 250
 251 #define P(x) \
 252         SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
 253 #define PN(x) \
 254         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
 255
 256         P(rt_nr_running);
 257         P(rt_throttled);
 258         PN(rt_time);
 259         PN(rt_runtime);
 260
 261 #undef PN
 262 #undef P
 263 }
 264
 265 extern __read_mostly int sched_clock_running;
 266
 267 static void print_cpu(struct seq_file *m, int cpu)
 268 {
 269         struct rq *rq = cpu_rq(cpu);
 270         unsigned long flags;
 271
 272 #ifdef CONFIG_X86
 273         {
 274                 unsigned int freq = cpu_khz ? : 1;
 275
 276                 SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
 277                            cpu, freq / 1000, (freq % 1000));
 278         }
 279 #else
 280         SEQ_printf(m, "cpu#%d\n", cpu);
 281 #endif
 282
 283 #define P(x)                                                            \
 284 do {                                                                    \
 285         if (sizeof(rq->x) == 4)                                         \
 286                 SEQ_printf(m, "  .%-30s: %ld\n", #x, (long)(rq->x));    \
 287         else                                                            \
 288                 SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rq->x));\
 289 } while (0)
 290
 291 #define PN(x) \
 292         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
 293
 294         P(nr_running);
 295         SEQ_printf(m, "  .%-30s: %lu\n", "load",
 296                    rq->load.weight);
 297         P(nr_switches);
 298         P(nr_load_updates);
 299         P(nr_uninterruptible);
 300         PN(next_balance);
 301         P(curr->pid);
 302         PN(clock);
 303         P(cpu_load[0]);
 304         P(cpu_load[1]);
 305         P(cpu_load[2]);
 306         P(cpu_load[3]);
 307         P(cpu_load[4]);
 308 #undef P
 309 #undef PN
 310
 311 #ifdef CONFIG_SCHEDSTATS
 312 #define P(n) SEQ_printf(m, "  .%-30s: %d\n", #n, rq->n);
 313 #define P64(n) SEQ_printf(m, "  .%-30s: %Ld\n", #n, rq->n);
 314
 315         P(yld_count);
 316
 317         P(sched_count);
 318         P(sched_goidle);
 319 #ifdef CONFIG_SMP
 320         P64(avg_idle);
 321 #endif
 322
 323         P(ttwu_count);
 324         P(ttwu_local);
 325
 326 #undef P
 327 #undef P64
 328 #endif
 329         spin_lock_irqsave(&sched_debug_lock, flags);
 330         print_cfs_stats(m, cpu);
 331         print_rt_stats(m, cpu);
 332
 333         rcu_read_lock();
 334         print_rq(m, rq, cpu);
 335         rcu_read_unlock();
 336         spin_unlock_irqrestore(&sched_debug_lock, flags);
 337         SEQ_printf(m, "\n");
 338 }
 339
 340 static const char *sched_tunable_scaling_names[] = {
 341         "none",
 342         "logaritmic",
 343         "linear"
 344 };
 345
 346 static void sched_debug_header(struct seq_file *m)
 347 {
 348         u64 ktime, sched_clk, cpu_clk;
 349         unsigned long flags;
 350
 351         local_irq_save(flags);
 352         ktime = ktime_to_ns(ktime_get());
 353         sched_clk = sched_clock();
 354         cpu_clk = local_clock();
 355         local_irq_restore(flags);
 356
 357         SEQ_printf(m, "Sched Debug Version: v0.10, %s %.*s\n",
 358                 init_utsname()->release,
 359                 (int)strcspn(init_utsname()->version, " "),
 360                 init_utsname()->version);
 361
 362 #define P(x) \
 363         SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
 364 #define PN(x) \
 365         SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
 366         PN(ktime);
 367         PN(sched_clk);
 368         PN(cpu_clk);
 369         P(jiffies);
 370 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
 371         P(sched_clock_stable);
 372 #endif
 373 #undef PN
 374 #undef P
 375
 376         SEQ_printf(m, "\n");
 377         SEQ_printf(m, "sysctl_sched\n");
 378
 379 #define P(x) \
 380         SEQ_printf(m, "  .%-40s: %Ld\n", #x, (long long)(x))
 381 #define PN(x) \
 382         SEQ_printf(m, "  .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
 383         PN(sysctl_sched_latency);
 384         PN(sysctl_sched_min_granularity);
 385         PN(sysctl_sched_wakeup_granularity);
 386         P(sysctl_sched_child_runs_first);
 387         P(sysctl_sched_features);
 388 #undef PN
 389 #undef P
 390
 391         SEQ_printf(m, "  .%-40s: %d (%s)\n",
 392                 "sysctl_sched_tunable_scaling",
 393                 sysctl_sched_tunable_scaling,
 394                 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
 395         SEQ_printf(m, "\n");
 396 }
 397
 398 static int sched_debug_show(struct seq_file *m, void *v)
 399 {
 400         int cpu = (unsigned long)(v - 2);
 401
 402         if (cpu != -1)
 403                 print_cpu(m, cpu);
 404         else
 405                 sched_debug_header(m);
 406
 407         return 0;
 408 }
 409
 410 void sysrq_sched_debug_show(void)
 411 {
 412         int cpu;
 413
 414         sched_debug_header(NULL);
 415         for_each_online_cpu(cpu)
 416                 print_cpu(NULL, cpu);
 417
 418 }
 419
 420 /*
 421  * This itererator needs some explanation.
 422  * It returns 1 for the header position.
 423  * This means 2 is cpu 0.
 424  * In a hotplugged system some cpus, including cpu 0, may be missing so we have
 425  * to use cpumask_* to iterate over the cpus.
 426  */
 427 static void *sched_debug_start(struct seq_file *file, loff_t *offset)
 428 {
 429         unsigned long n = *offset;
 430
 431         if (n == 0)
 432                 return (void *) 1;
 433
 434         n--;
 435
 436         if (n > 0)
 437                 n = cpumask_next(n - 1, cpu_online_mask);
 438         else
 439                 n = cpumask_first(cpu_online_mask);
 440
 441         *offset = n + 1;
 442
 443         if (n < nr_cpu_ids)
 444                 return (void *)(unsigned long)(n + 2);
 445         return NULL;
 446 }
 447
 448 static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
 449 {
 450         (*offset)++;
 451         return sched_debug_start(file, offset);
 452 }
 453
 454 static void sched_debug_stop(struct seq_file *file, void *data)
 455 {
 456 }
 457
 458 static const struct seq_operations sched_debug_sops = {
 459         .start = sched_debug_start,
 460         .next = sched_debug_next,
 461         .stop = sched_debug_stop,
 462         .show = sched_debug_show,
 463 };
 464
 465 static int sched_debug_release(struct inode *inode, struct file *file)
 466 {
 467         seq_release(inode, file);
 468
 469         return 0;
 470 }
 471
 472 static int sched_debug_open(struct inode *inode, struct file *filp)
 473 {
 474         int ret = 0;
 475
 476         ret = seq_open(filp, &sched_debug_sops);
 477
 478         return ret;
 479 }
 480
 481 static const struct file_operations sched_debug_fops = {
 482         .open           = sched_debug_open,
 483         .read           = seq_read,
 484         .llseek         = seq_lseek,
 485         .release        = sched_debug_release,
 486 };
 487
 488 static int __init init_sched_debug_procfs(void)
 489 {
 490         struct proc_dir_entry *pe;
 491
 492         pe = proc_create("sched_debug", 0444, NULL, &sched_debug_fops);
 493         if (!pe)
 494                 return -ENOMEM;
 495         return 0;
 496 }
 497
 498 __initcall(init_sched_debug_procfs);
 499
 500 void proc_sched_show_task(struct task_struct *p, struct seq_file *m)
 501 {
 502         unsigned long nr_switches;
 503
 504         SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, p->pid,
 505                                                 get_nr_threads(p));
 506         SEQ_printf(m,
 507                 "---------------------------------------------------------\n");
 508 #define __P(F) \
 509         SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)F)
 510 #define P(F) \
 511         SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)p->F)
 512 #define __PN(F) \
 513         SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
 514 #define PN(F) \
 515         SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
 516
 517         PN(se.exec_start);
 518         PN(se.vruntime);
 519         PN(se.sum_exec_runtime);
 520
 521         nr_switches = p->nvcsw + p->nivcsw;
 522
 523 #ifdef CONFIG_SCHEDSTATS
 524         PN(se.statistics.wait_start);
 525         PN(se.statistics.sleep_start);
 526         PN(se.statistics.block_start);
 527         PN(se.statistics.sleep_max);
 528         PN(se.statistics.block_max);
 529         PN(se.statistics.exec_max);
 530         PN(se.statistics.slice_max);
 531         PN(se.statistics.wait_max);
 532         PN(se.statistics.wait_sum);
 533         P(se.statistics.wait_count);
 534         PN(se.statistics.iowait_sum);
 535         P(se.statistics.iowait_count);
 536         P(se.nr_migrations);
 537         P(se.statistics.nr_migrations_cold);
 538         P(se.statistics.nr_failed_migrations_affine);
 539         P(se.statistics.nr_failed_migrations_running);
 540         P(se.statistics.nr_failed_migrations_hot);
 541         P(se.statistics.nr_forced_migrations);
 542         P(se.statistics.nr_wakeups);
 543         P(se.statistics.nr_wakeups_sync);
 544         P(se.statistics.nr_wakeups_migrate);
 545         P(se.statistics.nr_wakeups_local);
 546         P(se.statistics.nr_wakeups_remote);
 547         P(se.statistics.nr_wakeups_affine);
 548         P(se.statistics.nr_wakeups_affine_attempts);
 549         P(se.statistics.nr_wakeups_passive);
 550         P(se.statistics.nr_wakeups_idle);
 551
 552         {
 553                 u64 avg_atom, avg_per_cpu;
 554
 555                 avg_atom = p->se.sum_exec_runtime;
 556                 if (nr_switches)
 557                         avg_atom = div64_ul(avg_atom, nr_switches);
 558                 else
 559                         avg_atom = -1LL;
 560
 561                 avg_per_cpu = p->se.sum_exec_runtime;
 562                 if (p->se.nr_migrations) {
 563                         avg_per_cpu = div64_u64(avg_per_cpu,
 564                                                 p->se.nr_migrations);
 565                 } else {
 566                         avg_per_cpu = -1LL;
 567                 }
 568
 569                 __PN(avg_atom);
 570                 __PN(avg_per_cpu);
 571         }
 572 #endif
 573         __P(nr_switches);
 574         SEQ_printf(m, "%-35s:%21Ld\n",
 575                    "nr_voluntary_switches", (long long)p->nvcsw);
 576         SEQ_printf(m, "%-35s:%21Ld\n",
 577                    "nr_involuntary_switches", (long long)p->nivcsw);
 578
 579         P(se.load.weight);
 580 #if defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)
 581         P(se.avg.runnable_avg_sum);
 582         P(se.avg.runnable_avg_period);
 583         P(se.avg.load_avg_contrib);
 584         P(se.avg.decay_count);
 585 #endif
 586         P(policy);
 587         P(prio);
 588 #undef PN
 589 #undef __PN
 590 #undef P
 591 #undef __P
 592
 593         {
 594                 unsigned int this_cpu = raw_smp_processor_id();
 595                 u64 t0, t1;
 596
 597                 t0 = cpu_clock(this_cpu);
 598                 t1 = cpu_clock(this_cpu);
 599                 SEQ_printf(m, "%-35s:%21Ld\n",
 600                            "clock-delta", (long long)(t1-t0));
 601         }
 602 }
 603
 604 void proc_sched_set_task(struct task_struct *p)
 605 {
 606 #ifdef CONFIG_SCHEDSTATS
 607         memset(&p->se.statistics, 0, sizeof(p->se.statistics));
 608 #endif
 609 }