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