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