kernel/sched_debug.c

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