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 #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         if (!se)
  65                 return;
  66
  67 #define P(F) \
  68         SEQ_printf(m, "  .%-30s: %lld\n", #F, (long long)F)
  69 #define PN(F) \
  70         SEQ_printf(m, "  .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F))
  71
  72         PN(se->exec_start);
  73         PN(se->vruntime);
  74         PN(se->sum_exec_runtime);
  75 #ifdef CONFIG_SCHEDSTATS
  76         PN(se->statistics.wait_start);
  77         PN(se->statistics.sleep_start);
  78         PN(se->statistics.block_start);
  79         PN(se->statistics.sleep_max);
  80         PN(se->statistics.block_max);
  81         PN(se->statistics.exec_max);
  82         PN(se->statistics.slice_max);
  83         PN(se->statistics.wait_max);
  84         PN(se->statistics.wait_sum);
  85         P(se->statistics.wait_count);
  86 #endif
  87         P(se->load.weight);
  88 #undef PN
  89 #undef P
  90 }
  91 #endif
  92
  93 #ifdef CONFIG_CGROUP_SCHED
  94 static char group_path[PATH_MAX];
  95
  96 static char *task_group_path(struct task_group *tg)
  97 {
  98         if (autogroup_path(tg, group_path, PATH_MAX))
  99                 return group_path;
 100
 101         /*
 102          * May be NULL if the underlying cgroup isn't fully-created yet
 103          */
 104         if (!tg->css.cgroup) {
 105                 group_path[0] = '\0';
 106                 return group_path;
 107         }
 108         cgroup_path(tg->css.cgroup, group_path, PATH_MAX);
 109         return group_path;
 110 }
 111 #endif
 112
 113 static void
 114 print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
 115 {
 116         if (rq->curr == p)
 117                 SEQ_printf(m, "R");
 118         else
 119                 SEQ_printf(m, " ");
 120
 121         SEQ_printf(m, "%15s %5d %9Ld.%06ld %9Ld %5d ",
 122                 p->comm, p->pid,
 123                 SPLIT_NS(p->se.vruntime),
 124                 (long long)(p->nvcsw + p->nivcsw),
 125                 p->prio);
 126 #ifdef CONFIG_SCHEDSTATS
 127         SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
 128                 SPLIT_NS(p->se.vruntime),
 129                 SPLIT_NS(p->se.sum_exec_runtime),
 130                 SPLIT_NS(p->se.statistics.sum_sleep_runtime));
 131 #else
 132         SEQ_printf(m, "%15Ld %15Ld %15Ld.%06ld %15Ld.%06ld %15Ld.%06ld",
 133                 0LL, 0LL, 0LL, 0L, 0LL, 0L, 0LL, 0L);
 134 #endif
 135 #ifdef CONFIG_CGROUP_SCHED
 136         SEQ_printf(m, " %s", task_group_path(task_group(p)));
 137 #endif
 138
 139         SEQ_printf(m, "\n");
 140 }
 141
 142 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
 143 {
 144         struct task_struct *g, *p;
 145         unsigned long flags;
 146
 147         SEQ_printf(m,
 148         "\nrunnable tasks:\n"
 149         "            task   PID         tree-key  switches  prio"
 150         "     exec-runtime         sum-exec        sum-sleep\n"
 151         "------------------------------------------------------"
 152         "----------------------------------------------------\n");
 153
 154         read_lock_irqsave(&tasklist_lock, flags);
 155
 156         do_each_thread(g, p) {
 157                 if (!p->on_rq || task_cpu(p) != rq_cpu)
 158                         continue;
 159
 160                 print_task(m, rq, p);
 161         } while_each_thread(g, p);
 162
 163         read_unlock_irqrestore(&tasklist_lock, flags);
 164 }
 165
 166 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
 167 {
 168         s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
 169                 spread, rq0_min_vruntime, spread0;
 170         struct rq *rq = cpu_rq(cpu);
 171         struct sched_entity *last;
 172         unsigned long flags;
 173
 174 #ifdef CONFIG_FAIR_GROUP_SCHED
 175         SEQ_printf(m, "\ncfs_rq[%d]:%s\n", cpu, task_group_path(cfs_rq->tg));
 176 #else
 177         SEQ_printf(m, "\ncfs_rq[%d]:\n", cpu);
 178 #endif
 179         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "exec_clock",
 180                         SPLIT_NS(cfs_rq->exec_clock));
 181
 182         raw_spin_lock_irqsave(&rq->lock, flags);
 183         if (cfs_rq->rb_leftmost)
 184                 MIN_vruntime = (__pick_first_entity(cfs_rq))->vruntime;
 185         last = __pick_last_entity(cfs_rq);
 186         if (last)
 187                 max_vruntime = last->vruntime;
 188         min_vruntime = cfs_rq->min_vruntime;
 189         rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
 190         raw_spin_unlock_irqrestore(&rq->lock, flags);
 191         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "MIN_vruntime",
 192                         SPLIT_NS(MIN_vruntime));
 193         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "min_vruntime",
 194                         SPLIT_NS(min_vruntime));
 195         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "max_vruntime",
 196                         SPLIT_NS(max_vruntime));
 197         spread = max_vruntime - MIN_vruntime;
 198         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread",
 199                         SPLIT_NS(spread));
 200         spread0 = min_vruntime - rq0_min_vruntime;
 201         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread0",
 202                         SPLIT_NS(spread0));
 203         SEQ_printf(m, "  .%-30s: %d\n", "nr_spread_over",
 204                         cfs_rq->nr_spread_over);
 205         SEQ_printf(m, "  .%-30s: %ld\n", "nr_running", cfs_rq->nr_running);
 206         SEQ_printf(m, "  .%-30s: %ld\n", "load", cfs_rq->load.weight);
 207 #ifdef CONFIG_FAIR_GROUP_SCHED
 208 #ifdef CONFIG_SMP
 209         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "load_avg",
 210                         SPLIT_NS(cfs_rq->load_avg));
 211         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "load_period",
 212                         SPLIT_NS(cfs_rq->load_period));
 213         SEQ_printf(m, "  .%-30s: %ld\n", "load_contrib",
 214                         cfs_rq->load_contribution);
 215         SEQ_printf(m, "  .%-30s: %d\n", "load_tg",
 216                         atomic_read(&cfs_rq->tg->load_weight));
 217 #endif
 218
 219         print_cfs_group_stats(m, cpu, cfs_rq->tg);
 220 #endif
 221 }
 222
 223 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
 224 {
 225 #ifdef CONFIG_RT_GROUP_SCHED
 226         SEQ_printf(m, "\nrt_rq[%d]:%s\n", cpu, task_group_path(rt_rq->tg));
 227 #else
 228         SEQ_printf(m, "\nrt_rq[%d]:\n", cpu);
 229 #endif
 230
 231 #define P(x) \
 232         SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
 233 #define PN(x) \
 234         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
 235
 236         P(rt_nr_running);
 237         P(rt_throttled);
 238         PN(rt_time);
 239         PN(rt_runtime);
 240
 241 #undef PN
 242 #undef P
 243 }
 244
 245 extern __read_mostly int sched_clock_running;
 246
 247 static void print_cpu(struct seq_file *m, int cpu)
 248 {
 249         struct rq *rq = cpu_rq(cpu);
 250         unsigned long flags;
 251
 252 #ifdef CONFIG_X86
 253         {
 254                 unsigned int freq = cpu_khz ? : 1;
 255
 256                 SEQ_printf(m, "\ncpu#%d, %u.%03u MHz\n",
 257                            cpu, freq / 1000, (freq % 1000));
 258         }
 259 #else
 260         SEQ_printf(m, "\ncpu#%d\n", cpu);
 261 #endif
 262
 263 #define P(x) \
 264         SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rq->x))
 265 #define PN(x) \
 266         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
 267
 268         P(nr_running);
 269         SEQ_printf(m, "  .%-30s: %lu\n", "load",
 270                    rq->load.weight);
 271         P(nr_switches);
 272         P(nr_load_updates);
 273         P(nr_uninterruptible);
 274         PN(next_balance);
 275         P(curr->pid);
 276         PN(clock);
 277         P(cpu_load[0]);
 278         P(cpu_load[1]);
 279         P(cpu_load[2]);
 280         P(cpu_load[3]);
 281         P(cpu_load[4]);
 282 #undef P
 283 #undef PN
 284
 285 #ifdef CONFIG_SCHEDSTATS
 286 #define P(n) SEQ_printf(m, "  .%-30s: %d\n", #n, rq->n);
 287 #define P64(n) SEQ_printf(m, "  .%-30s: %Ld\n", #n, rq->n);
 288
 289         P(yld_count);
 290
 291         P(sched_switch);
 292         P(sched_count);
 293         P(sched_goidle);
 294 #ifdef CONFIG_SMP
 295         P64(avg_idle);
 296 #endif
 297
 298         P(ttwu_count);
 299         P(ttwu_local);
 300
 301 #undef P
 302 #undef P64
 303 #endif
 304         spin_lock_irqsave(&sched_debug_lock, flags);
 305         print_cfs_stats(m, cpu);
 306         print_rt_stats(m, cpu);
 307
 308         rcu_read_lock();
 309         print_rq(m, rq, cpu);
 310         rcu_read_unlock();
 311         spin_unlock_irqrestore(&sched_debug_lock, flags);
 312 }
 313
 314 static const char *sched_tunable_scaling_names[] = {
 315         "none",
 316         "logaritmic",
 317         "linear"
 318 };
 319
 320 static int sched_debug_show(struct seq_file *m, void *v)
 321 {
 322         u64 ktime, sched_clk, cpu_clk;
 323         unsigned long flags;
 324         int cpu;
 325
 326         local_irq_save(flags);
 327         ktime = ktime_to_ns(ktime_get());
 328         sched_clk = sched_clock();
 329         cpu_clk = local_clock();
 330         local_irq_restore(flags);
 331
 332         SEQ_printf(m, "Sched Debug Version: v0.10, %s %.*s\n",
 333                 init_utsname()->release,
 334                 (int)strcspn(init_utsname()->version, " "),
 335                 init_utsname()->version);
 336
 337 #define P(x) \
 338         SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
 339 #define PN(x) \
 340         SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
 341         PN(ktime);
 342         PN(sched_clk);
 343         PN(cpu_clk);
 344         P(jiffies);
 345 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
 346         P(sched_clock_stable);
 347 #endif
 348 #undef PN
 349 #undef P
 350
 351         SEQ_printf(m, "\n");
 352         SEQ_printf(m, "sysctl_sched\n");
 353
 354 #define P(x) \
 355         SEQ_printf(m, "  .%-40s: %Ld\n", #x, (long long)(x))
 356 #define PN(x) \
 357         SEQ_printf(m, "  .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
 358         PN(sysctl_sched_latency);
 359         PN(sysctl_sched_min_granularity);
 360         PN(sysctl_sched_wakeup_granularity);
 361         P(sysctl_sched_child_runs_first);
 362         P(sysctl_sched_features);
 363 #undef PN
 364 #undef P
 365
 366         SEQ_printf(m, "  .%-40s: %d (%s)\n", "sysctl_sched_tunable_scaling",
 367                 sysctl_sched_tunable_scaling,
 368                 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
 369
 370         for_each_online_cpu(cpu)
 371                 print_cpu(m, cpu);
 372
 373         SEQ_printf(m, "\n");
 374
 375         return 0;
 376 }
 377
 378 void sysrq_sched_debug_show(void)
 379 {
 380         sched_debug_show(NULL, NULL);
 381 }
 382
 383 static int sched_debug_open(struct inode *inode, struct file *filp)
 384 {
 385         return single_open(filp, sched_debug_show, NULL);
 386 }
 387
 388 static const struct file_operations sched_debug_fops = {
 389         .open           = sched_debug_open,
 390         .read           = seq_read,
 391         .llseek         = seq_lseek,
 392         .release        = single_release,
 393 };
 394
 395 static int __init init_sched_debug_procfs(void)
 396 {
 397         struct proc_dir_entry *pe;
 398
 399         pe = proc_create("sched_debug", 0444, NULL, &sched_debug_fops);
 400         if (!pe)
 401                 return -ENOMEM;
 402         return 0;
 403 }
 404
 405 __initcall(init_sched_debug_procfs);
 406
 407 void proc_sched_show_task(struct task_struct *p, struct seq_file *m)
 408 {
 409         unsigned long nr_switches;
 410
 411         SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, p->pid,
 412                                                 get_nr_threads(p));
 413         SEQ_printf(m,
 414                 "---------------------------------------------------------\n");
 415 #define __P(F) \
 416         SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)F)
 417 #define P(F) \
 418         SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)p->F)
 419 #define __PN(F) \
 420         SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
 421 #define PN(F) \
 422         SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
 423
 424         PN(se.exec_start);
 425         PN(se.vruntime);
 426         PN(se.sum_exec_runtime);
 427
 428         nr_switches = p->nvcsw + p->nivcsw;
 429
 430 #ifdef CONFIG_SCHEDSTATS
 431         PN(se.statistics.wait_start);
 432         PN(se.statistics.sleep_start);
 433         PN(se.statistics.block_start);
 434         PN(se.statistics.sleep_max);
 435         PN(se.statistics.block_max);
 436         PN(se.statistics.exec_max);
 437         PN(se.statistics.slice_max);
 438         PN(se.statistics.wait_max);
 439         PN(se.statistics.wait_sum);
 440         P(se.statistics.wait_count);
 441         PN(se.statistics.iowait_sum);
 442         P(se.statistics.iowait_count);
 443         P(se.nr_migrations);
 444         P(se.statistics.nr_migrations_cold);
 445         P(se.statistics.nr_failed_migrations_affine);
 446         P(se.statistics.nr_failed_migrations_running);
 447         P(se.statistics.nr_failed_migrations_hot);
 448         P(se.statistics.nr_forced_migrations);
 449         P(se.statistics.nr_wakeups);
 450         P(se.statistics.nr_wakeups_sync);
 451         P(se.statistics.nr_wakeups_migrate);
 452         P(se.statistics.nr_wakeups_local);
 453         P(se.statistics.nr_wakeups_remote);
 454         P(se.statistics.nr_wakeups_affine);
 455         P(se.statistics.nr_wakeups_affine_attempts);
 456         P(se.statistics.nr_wakeups_passive);
 457         P(se.statistics.nr_wakeups_idle);
 458
 459         {
 460                 u64 avg_atom, avg_per_cpu;
 461
 462                 avg_atom = p->se.sum_exec_runtime;
 463                 if (nr_switches)
 464                         do_div(avg_atom, nr_switches);
 465                 else
 466                         avg_atom = -1LL;
 467
 468                 avg_per_cpu = p->se.sum_exec_runtime;
 469                 if (p->se.nr_migrations) {
 470                         avg_per_cpu = div64_u64(avg_per_cpu,
 471                                                 p->se.nr_migrations);
 472                 } else {
 473                         avg_per_cpu = -1LL;
 474                 }
 475
 476                 __PN(avg_atom);
 477                 __PN(avg_per_cpu);
 478         }
 479 #endif
 480         __P(nr_switches);
 481         SEQ_printf(m, "%-35s:%21Ld\n",
 482                    "nr_voluntary_switches", (long long)p->nvcsw);
 483         SEQ_printf(m, "%-35s:%21Ld\n",
 484                    "nr_involuntary_switches", (long long)p->nivcsw);
 485
 486         P(se.load.weight);
 487         P(policy);
 488         P(prio);
 489 #undef PN
 490 #undef __PN
 491 #undef P
 492 #undef __P
 493
 494         {
 495                 unsigned int this_cpu = raw_smp_processor_id();
 496                 u64 t0, t1;
 497
 498                 t0 = cpu_clock(this_cpu);
 499                 t1 = cpu_clock(this_cpu);
 500                 SEQ_printf(m, "%-35s:%21Ld\n",
 501                            "clock-delta", (long long)(t1-t0));
 502         }
 503 }
 504
 505 void proc_sched_set_task(struct task_struct *p)
 506 {
 507 #ifdef CONFIG_SCHEDSTATS
 508         memset(&p->se.statistics, 0, sizeof(p->se.statistics));
 509 #endif
 510 }