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, task_pid_nr(p),
 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_SMP
 213         SEQ_printf(m, "  .%-30s: %ld\n", "runnable_load_avg",
 214                         cfs_rq->runnable_load_avg);
 215         SEQ_printf(m, "  .%-30s: %ld\n", "blocked_load_avg",
 216                         cfs_rq->blocked_load_avg);
 217 #ifdef CONFIG_FAIR_GROUP_SCHED
 218         SEQ_printf(m, "  .%-30s: %ld\n", "tg_load_contrib",
 219                         cfs_rq->tg_load_contrib);
 220         SEQ_printf(m, "  .%-30s: %d\n", "tg_runnable_contrib",
 221                         cfs_rq->tg_runnable_contrib);
 222         SEQ_printf(m, "  .%-30s: %ld\n", "tg_load_avg",
 223                         atomic_long_read(&cfs_rq->tg->load_avg));
 224         SEQ_printf(m, "  .%-30s: %d\n", "tg->runnable_avg",
 225                         atomic_read(&cfs_rq->tg->runnable_avg));
 226 #endif
 227 #endif
 228
 229 #ifdef CONFIG_FAIR_GROUP_SCHED
 230         print_cfs_group_stats(m, cpu, cfs_rq->tg);
 231 #endif
 232 }
 233
 234 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
 235 {
 236 #ifdef CONFIG_RT_GROUP_SCHED
 237         SEQ_printf(m, "\nrt_rq[%d]:%s\n", cpu, task_group_path(rt_rq->tg));
 238 #else
 239         SEQ_printf(m, "\nrt_rq[%d]:\n", cpu);
 240 #endif
 241
 242 #define P(x) \
 243         SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
 244 #define PN(x) \
 245         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
 246
 247         P(rt_nr_running);
 248         P(rt_throttled);
 249         PN(rt_time);
 250         PN(rt_runtime);
 251
 252 #undef PN
 253 #undef P
 254 }
 255
 256 extern __read_mostly int sched_clock_running;
 257
 258 static void print_cpu(struct seq_file *m, int cpu)
 259 {
 260         struct rq *rq = cpu_rq(cpu);
 261         unsigned long flags;
 262
 263 #ifdef CONFIG_X86
 264         {
 265                 unsigned int freq = cpu_khz ? : 1;
 266
 267                 SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
 268                            cpu, freq / 1000, (freq % 1000));
 269         }
 270 #else
 271         SEQ_printf(m, "cpu#%d\n", cpu);
 272 #endif
 273
 274 #define P(x)                                                            \
 275 do {                                                                    \
 276         if (sizeof(rq->x) == 4)                                         \
 277                 SEQ_printf(m, "  .%-30s: %ld\n", #x, (long)(rq->x));    \
 278         else                                                            \
 279                 SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rq->x));\
 280 } while (0)
 281
 282 #define PN(x) \
 283         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
 284
 285         P(nr_running);
 286         SEQ_printf(m, "  .%-30s: %lu\n", "load",
 287                    rq->load.weight);
 288         P(nr_switches);
 289         P(nr_load_updates);
 290         P(nr_uninterruptible);
 291         PN(next_balance);
 292         SEQ_printf(m, "  .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr)));
 293         PN(clock);
 294         P(cpu_load[0]);
 295         P(cpu_load[1]);
 296         P(cpu_load[2]);
 297         P(cpu_load[3]);
 298         P(cpu_load[4]);
 299 #undef P
 300 #undef PN
 301
 302 #ifdef CONFIG_SCHEDSTATS
 303 #define P(n) SEQ_printf(m, "  .%-30s: %d\n", #n, rq->n);
 304 #define P64(n) SEQ_printf(m, "  .%-30s: %Ld\n", #n, rq->n);
 305
 306         P(yld_count);
 307
 308         P(sched_count);
 309         P(sched_goidle);
 310 #ifdef CONFIG_SMP
 311         P64(avg_idle);
 312 #endif
 313
 314         P(ttwu_count);
 315         P(ttwu_local);
 316
 317 #undef P
 318 #undef P64
 319 #endif
 320         spin_lock_irqsave(&sched_debug_lock, flags);
 321         print_cfs_stats(m, cpu);
 322         print_rt_stats(m, cpu);
 323
 324         rcu_read_lock();
 325         print_rq(m, rq, cpu);
 326         rcu_read_unlock();
 327         spin_unlock_irqrestore(&sched_debug_lock, flags);
 328         SEQ_printf(m, "\n");
 329 }
 330
 331 static const char *sched_tunable_scaling_names[] = {
 332         "none",
 333         "logaritmic",
 334         "linear"
 335 };
 336
 337 static void sched_debug_header(struct seq_file *m)
 338 {
 339         u64 ktime, sched_clk, cpu_clk;
 340         unsigned long flags;
 341
 342         local_irq_save(flags);
 343         ktime = ktime_to_ns(ktime_get());
 344         sched_clk = sched_clock();
 345         cpu_clk = local_clock();
 346         local_irq_restore(flags);
 347
 348         SEQ_printf(m, "Sched Debug Version: v0.10, %s %.*s\n",
 349                 init_utsname()->release,
 350                 (int)strcspn(init_utsname()->version, " "),
 351                 init_utsname()->version);
 352
 353 #define P(x) \
 354         SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
 355 #define PN(x) \
 356         SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
 357         PN(ktime);
 358         PN(sched_clk);
 359         PN(cpu_clk);
 360         P(jiffies);
 361 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
 362         P(sched_clock_stable);
 363 #endif
 364 #undef PN
 365 #undef P
 366
 367         SEQ_printf(m, "\n");
 368         SEQ_printf(m, "sysctl_sched\n");
 369
 370 #define P(x) \
 371         SEQ_printf(m, "  .%-40s: %Ld\n", #x, (long long)(x))
 372 #define PN(x) \
 373         SEQ_printf(m, "  .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
 374         PN(sysctl_sched_latency);
 375         PN(sysctl_sched_min_granularity);
 376         PN(sysctl_sched_wakeup_granularity);
 377         P(sysctl_sched_child_runs_first);
 378         P(sysctl_sched_features);
 379 #undef PN
 380 #undef P
 381
 382         SEQ_printf(m, "  .%-40s: %d (%s)\n",
 383                 "sysctl_sched_tunable_scaling",
 384                 sysctl_sched_tunable_scaling,
 385                 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
 386         SEQ_printf(m, "\n");
 387 }
 388
 389 static int sched_debug_show(struct seq_file *m, void *v)
 390 {
 391         int cpu = (unsigned long)(v - 2);
 392
 393         if (cpu != -1)
 394                 print_cpu(m, cpu);
 395         else
 396                 sched_debug_header(m);
 397
 398         return 0;
 399 }
 400
 401 void sysrq_sched_debug_show(void)
 402 {
 403         int cpu;
 404
 405         sched_debug_header(NULL);
 406         for_each_online_cpu(cpu)
 407                 print_cpu(NULL, cpu);
 408
 409 }
 410
 411 /*
 412  * This itererator needs some explanation.
 413  * It returns 1 for the header position.
 414  * This means 2 is cpu 0.
 415  * In a hotplugged system some cpus, including cpu 0, may be missing so we have
 416  * to use cpumask_* to iterate over the cpus.
 417  */
 418 static void *sched_debug_start(struct seq_file *file, loff_t *offset)
 419 {
 420         unsigned long n = *offset;
 421
 422         if (n == 0)
 423                 return (void *) 1;
 424
 425         n--;
 426
 427         if (n > 0)
 428                 n = cpumask_next(n - 1, cpu_online_mask);
 429         else
 430                 n = cpumask_first(cpu_online_mask);
 431
 432         *offset = n + 1;
 433
 434         if (n < nr_cpu_ids)
 435                 return (void *)(unsigned long)(n + 2);
 436         return NULL;
 437 }
 438
 439 static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
 440 {
 441         (*offset)++;
 442         return sched_debug_start(file, offset);
 443 }
 444
 445 static void sched_debug_stop(struct seq_file *file, void *data)
 446 {
 447 }
 448
 449 static const struct seq_operations sched_debug_sops = {
 450         .start = sched_debug_start,
 451         .next = sched_debug_next,
 452         .stop = sched_debug_stop,
 453         .show = sched_debug_show,
 454 };
 455
 456 static int sched_debug_release(struct inode *inode, struct file *file)
 457 {
 458         seq_release(inode, file);
 459
 460         return 0;
 461 }
 462
 463 static int sched_debug_open(struct inode *inode, struct file *filp)
 464 {
 465         int ret = 0;
 466
 467         ret = seq_open(filp, &sched_debug_sops);
 468
 469         return ret;
 470 }
 471
 472 static const struct file_operations sched_debug_fops = {
 473         .open           = sched_debug_open,
 474         .read           = seq_read,
 475         .llseek         = seq_lseek,
 476         .release        = sched_debug_release,
 477 };
 478
 479 static int __init init_sched_debug_procfs(void)
 480 {
 481         struct proc_dir_entry *pe;
 482
 483         pe = proc_create("sched_debug", 0444, NULL, &sched_debug_fops);
 484         if (!pe)
 485                 return -ENOMEM;
 486         return 0;
 487 }
 488
 489 __initcall(init_sched_debug_procfs);
 490
 491 void proc_sched_show_task(struct task_struct *p, struct seq_file *m)
 492 {
 493         unsigned long nr_switches;
 494
 495         SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr(p),
 496                                                 get_nr_threads(p));
 497         SEQ_printf(m,
 498                 "---------------------------------------------------------"
 499                 "----------\n");
 500 #define __P(F) \
 501         SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)F)
 502 #define P(F) \
 503         SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)p->F)
 504 #define __PN(F) \
 505         SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
 506 #define PN(F) \
 507         SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
 508
 509         PN(se.exec_start);
 510         PN(se.vruntime);
 511         PN(se.sum_exec_runtime);
 512
 513         nr_switches = p->nvcsw + p->nivcsw;
 514
 515 #ifdef CONFIG_SCHEDSTATS
 516         PN(se.statistics.wait_start);
 517         PN(se.statistics.sleep_start);
 518         PN(se.statistics.block_start);
 519         PN(se.statistics.sleep_max);
 520         PN(se.statistics.block_max);
 521         PN(se.statistics.exec_max);
 522         PN(se.statistics.slice_max);
 523         PN(se.statistics.wait_max);
 524         PN(se.statistics.wait_sum);
 525         P(se.statistics.wait_count);
 526         PN(se.statistics.iowait_sum);
 527         P(se.statistics.iowait_count);
 528         P(se.nr_migrations);
 529         P(se.statistics.nr_migrations_cold);
 530         P(se.statistics.nr_failed_migrations_affine);
 531         P(se.statistics.nr_failed_migrations_running);
 532         P(se.statistics.nr_failed_migrations_hot);
 533         P(se.statistics.nr_forced_migrations);
 534         P(se.statistics.nr_wakeups);
 535         P(se.statistics.nr_wakeups_sync);
 536         P(se.statistics.nr_wakeups_migrate);
 537         P(se.statistics.nr_wakeups_local);
 538         P(se.statistics.nr_wakeups_remote);
 539         P(se.statistics.nr_wakeups_affine);
 540         P(se.statistics.nr_wakeups_affine_attempts);
 541         P(se.statistics.nr_wakeups_passive);
 542         P(se.statistics.nr_wakeups_idle);
 543
 544         {
 545                 u64 avg_atom, avg_per_cpu;
 546
 547                 avg_atom = p->se.sum_exec_runtime;
 548                 if (nr_switches)
 549                         do_div(avg_atom, nr_switches);
 550                 else
 551                         avg_atom = -1LL;
 552
 553                 avg_per_cpu = p->se.sum_exec_runtime;
 554                 if (p->se.nr_migrations) {
 555                         avg_per_cpu = div64_u64(avg_per_cpu,
 556                                                 p->se.nr_migrations);
 557                 } else {
 558                         avg_per_cpu = -1LL;
 559                 }
 560
 561                 __PN(avg_atom);
 562                 __PN(avg_per_cpu);
 563         }
 564 #endif
 565         __P(nr_switches);
 566         SEQ_printf(m, "%-45s:%21Ld\n",
 567                    "nr_voluntary_switches", (long long)p->nvcsw);
 568         SEQ_printf(m, "%-45s:%21Ld\n",
 569                    "nr_involuntary_switches", (long long)p->nivcsw);
 570
 571         P(se.load.weight);
 572 #ifdef CONFIG_SMP
 573         P(se.avg.runnable_avg_sum);
 574         P(se.avg.runnable_avg_period);
 575         P(se.avg.load_avg_contrib);
 576         P(se.avg.decay_count);
 577 #endif
 578         P(policy);
 579         P(prio);
 580 #undef PN
 581 #undef __PN
 582 #undef P
 583 #undef __P
 584
 585         {
 586                 unsigned int this_cpu = raw_smp_processor_id();
 587                 u64 t0, t1;
 588
 589                 t0 = cpu_clock(this_cpu);
 590                 t1 = cpu_clock(this_cpu);
 591                 SEQ_printf(m, "%-45s:%21Ld\n",
 592                            "clock-delta", (long long)(t1-t0));
 593         }
 594 }
 595
 596 void proc_sched_set_task(struct task_struct *p)
 597 {
 598 #ifdef CONFIG_SCHEDSTATS
 599         memset(&p->se.statistics, 0, sizeof(p->se.statistics));
 600 #endif
 601 }