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 /*
  20  * This allows printing both to /proc/sched_debug and
  21  * to the console
  22  */
  23 #define SEQ_printf(m, x...)                     \
  24  do {                                           \
  25         if (m)                                  \
  26                 seq_printf(m, x);               \
  27         else                                    \
  28                 printk(x);                      \
  29  } while (0)
  30
  31 /*
  32  * Ease the printing of nsec fields:
  33  */
  34 static long long nsec_high(unsigned long long nsec)
  35 {
  36         if ((long long)nsec < 0) {
  37                 nsec = -nsec;
  38                 do_div(nsec, 1000000);
  39                 return -nsec;
  40         }
  41         do_div(nsec, 1000000);
  42
  43         return nsec;
  44 }
  45
  46 static unsigned long nsec_low(unsigned long long nsec)
  47 {
  48         if ((long long)nsec < 0)
  49                 nsec = -nsec;
  50
  51         return do_div(nsec, 1000000);
  52 }
  53
  54 #define SPLIT_NS(x) nsec_high(x), nsec_low(x)
  55
  56 #ifdef CONFIG_FAIR_GROUP_SCHED
  57 static void print_cfs_group_stats(struct seq_file *m, int cpu, struct task_group *tg)
  58 {
  59         struct sched_entity *se = tg->se[cpu];
  60         if (!se)
  61                 return;
  62
  63 #define P(F) \
  64         SEQ_printf(m, "  .%-30s: %lld\n", #F, (long long)F)
  65 #define PN(F) \
  66         SEQ_printf(m, "  .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F))
  67
  68         PN(se->exec_start);
  69         PN(se->vruntime);
  70         PN(se->sum_exec_runtime);
  71 #ifdef CONFIG_SCHEDSTATS
  72         PN(se->statistics.wait_start);
  73         PN(se->statistics.sleep_start);
  74         PN(se->statistics.block_start);
  75         PN(se->statistics.sleep_max);
  76         PN(se->statistics.block_max);
  77         PN(se->statistics.exec_max);
  78         PN(se->statistics.slice_max);
  79         PN(se->statistics.wait_max);
  80         PN(se->statistics.wait_sum);
  81         P(se->statistics.wait_count);
  82 #endif
  83         P(se->load.weight);
  84 #undef PN
  85 #undef P
  86 }
  87 #endif
  88
  89 static void
  90 print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
  91 {
  92         if (rq->curr == p)
  93                 SEQ_printf(m, "R");
  94         else
  95                 SEQ_printf(m, " ");
  96
  97         SEQ_printf(m, "%15s %5d %9Ld.%06ld %9Ld %5d ",
  98                 p->comm, p->pid,
  99                 SPLIT_NS(p->se.vruntime),
 100                 (long long)(p->nvcsw + p->nivcsw),
 101                 p->prio);
 102 #ifdef CONFIG_SCHEDSTATS
 103         SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
 104                 SPLIT_NS(p->se.vruntime),
 105                 SPLIT_NS(p->se.sum_exec_runtime),
 106                 SPLIT_NS(p->se.statistics.sum_sleep_runtime));
 107 #else
 108         SEQ_printf(m, "%15Ld %15Ld %15Ld.%06ld %15Ld.%06ld %15Ld.%06ld",
 109                 0LL, 0LL, 0LL, 0L, 0LL, 0L, 0LL, 0L);
 110 #endif
 111
 112         SEQ_printf(m, "\n");
 113 }
 114
 115 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
 116 {
 117         struct task_struct *g, *p;
 118         unsigned long flags;
 119
 120         SEQ_printf(m,
 121         "\nrunnable tasks:\n"
 122         "            task   PID         tree-key  switches  prio"
 123         "     exec-runtime         sum-exec        sum-sleep\n"
 124         "------------------------------------------------------"
 125         "----------------------------------------------------\n");
 126
 127         read_lock_irqsave(&tasklist_lock, flags);
 128
 129         do_each_thread(g, p) {
 130                 if (!p->se.on_rq || task_cpu(p) != rq_cpu)
 131                         continue;
 132
 133                 print_task(m, rq, p);
 134         } while_each_thread(g, p);
 135
 136         read_unlock_irqrestore(&tasklist_lock, flags);
 137 }
 138
 139 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
 140 {
 141         s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
 142                 spread, rq0_min_vruntime, spread0;
 143         struct rq *rq = cpu_rq(cpu);
 144         struct sched_entity *last;
 145         unsigned long flags;
 146
 147         SEQ_printf(m, "\ncfs_rq[%d]:\n", cpu);
 148         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "exec_clock",
 149                         SPLIT_NS(cfs_rq->exec_clock));
 150
 151         raw_spin_lock_irqsave(&rq->lock, flags);
 152         if (cfs_rq->rb_leftmost)
 153                 MIN_vruntime = (__pick_next_entity(cfs_rq))->vruntime;
 154         last = __pick_last_entity(cfs_rq);
 155         if (last)
 156                 max_vruntime = last->vruntime;
 157         min_vruntime = cfs_rq->min_vruntime;
 158         rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
 159         raw_spin_unlock_irqrestore(&rq->lock, flags);
 160         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "MIN_vruntime",
 161                         SPLIT_NS(MIN_vruntime));
 162         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "min_vruntime",
 163                         SPLIT_NS(min_vruntime));
 164         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "max_vruntime",
 165                         SPLIT_NS(max_vruntime));
 166         spread = max_vruntime - MIN_vruntime;
 167         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread",
 168                         SPLIT_NS(spread));
 169         spread0 = min_vruntime - rq0_min_vruntime;
 170         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread0",
 171                         SPLIT_NS(spread0));
 172         SEQ_printf(m, "  .%-30s: %d\n", "nr_spread_over",
 173                         cfs_rq->nr_spread_over);
 174         SEQ_printf(m, "  .%-30s: %ld\n", "nr_running", cfs_rq->nr_running);
 175         SEQ_printf(m, "  .%-30s: %ld\n", "load", cfs_rq->load.weight);
 176 #ifdef CONFIG_FAIR_GROUP_SCHED
 177 #ifdef CONFIG_SMP
 178         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "load_avg",
 179                         SPLIT_NS(cfs_rq->load_avg));
 180         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "load_period",
 181                         SPLIT_NS(cfs_rq->load_period));
 182         SEQ_printf(m, "  .%-30s: %ld\n", "load_contrib",
 183                         cfs_rq->load_contribution);
 184         SEQ_printf(m, "  .%-30s: %d\n", "load_tg",
 185                         atomic_read(&cfs_rq->tg->load_weight));
 186 #endif
 187
 188         print_cfs_group_stats(m, cpu, cfs_rq->tg);
 189 #endif
 190 }
 191
 192 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
 193 {
 194         SEQ_printf(m, "\nrt_rq[%d]:\n", cpu);
 195
 196 #define P(x) \
 197         SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
 198 #define PN(x) \
 199         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
 200
 201         P(rt_nr_running);
 202         P(rt_throttled);
 203         PN(rt_time);
 204         PN(rt_runtime);
 205
 206 #undef PN
 207 #undef P
 208 }
 209
 210 extern __read_mostly int sched_clock_running;
 211
 212 static void print_cpu(struct seq_file *m, int cpu)
 213 {
 214         struct rq *rq = cpu_rq(cpu);
 215
 216 #ifdef CONFIG_X86
 217         {
 218                 unsigned int freq = cpu_khz ? : 1;
 219
 220                 SEQ_printf(m, "\ncpu#%d, %u.%03u MHz\n",
 221                            cpu, freq / 1000, (freq % 1000));
 222         }
 223 #else
 224         SEQ_printf(m, "\ncpu#%d\n", cpu);
 225 #endif
 226
 227 #define P(x) \
 228         SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rq->x))
 229 #define PN(x) \
 230         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
 231
 232         P(nr_running);
 233         SEQ_printf(m, "  .%-30s: %lu\n", "load",
 234                    rq->load.weight);
 235         P(nr_switches);
 236         P(nr_load_updates);
 237         P(nr_uninterruptible);
 238         PN(next_balance);
 239         P(curr->pid);
 240         PN(clock);
 241         P(cpu_load[0]);
 242         P(cpu_load[1]);
 243         P(cpu_load[2]);
 244         P(cpu_load[3]);
 245         P(cpu_load[4]);
 246 #undef P
 247 #undef PN
 248
 249 #ifdef CONFIG_SCHEDSTATS
 250 #define P(n) SEQ_printf(m, "  .%-30s: %d\n", #n, rq->n);
 251 #define P64(n) SEQ_printf(m, "  .%-30s: %Ld\n", #n, rq->n);
 252
 253         P(yld_count);
 254
 255         P(sched_switch);
 256         P(sched_count);
 257         P(sched_goidle);
 258 #ifdef CONFIG_SMP
 259         P64(avg_idle);
 260 #endif
 261
 262         P(ttwu_count);
 263         P(ttwu_local);
 264
 265         P(bkl_count);
 266
 267 #undef P
 268 #endif
 269         print_cfs_stats(m, cpu);
 270         print_rt_stats(m, cpu);
 271
 272         print_rq(m, rq, cpu);
 273 }
 274
 275 static const char *sched_tunable_scaling_names[] = {
 276         "none",
 277         "logaritmic",
 278         "linear"
 279 };
 280
 281 static int sched_debug_show(struct seq_file *m, void *v)
 282 {
 283         u64 ktime, sched_clk, cpu_clk;
 284         unsigned long flags;
 285         int cpu;
 286
 287         local_irq_save(flags);
 288         ktime = ktime_to_ns(ktime_get());
 289         sched_clk = sched_clock();
 290         cpu_clk = local_clock();
 291         local_irq_restore(flags);
 292
 293         SEQ_printf(m, "Sched Debug Version: v0.10, %s %.*s\n",
 294                 init_utsname()->release,
 295                 (int)strcspn(init_utsname()->version, " "),
 296                 init_utsname()->version);
 297
 298 #define P(x) \
 299         SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
 300 #define PN(x) \
 301         SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
 302         PN(ktime);
 303         PN(sched_clk);
 304         PN(cpu_clk);
 305         P(jiffies);
 306 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
 307         P(sched_clock_stable);
 308 #endif
 309 #undef PN
 310 #undef P
 311
 312         SEQ_printf(m, "\n");
 313         SEQ_printf(m, "sysctl_sched\n");
 314
 315 #define P(x) \
 316         SEQ_printf(m, "  .%-40s: %Ld\n", #x, (long long)(x))
 317 #define PN(x) \
 318         SEQ_printf(m, "  .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
 319         PN(sysctl_sched_latency);
 320         PN(sysctl_sched_min_granularity);
 321         PN(sysctl_sched_wakeup_granularity);
 322         P(sysctl_sched_child_runs_first);
 323         P(sysctl_sched_features);
 324 #undef PN
 325 #undef P
 326
 327         SEQ_printf(m, "  .%-40s: %d (%s)\n", "sysctl_sched_tunable_scaling",
 328                 sysctl_sched_tunable_scaling,
 329                 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
 330
 331         for_each_online_cpu(cpu)
 332                 print_cpu(m, cpu);
 333
 334         SEQ_printf(m, "\n");
 335
 336         return 0;
 337 }
 338
 339 static void sysrq_sched_debug_show(void)
 340 {
 341         sched_debug_show(NULL, NULL);
 342 }
 343
 344 static int sched_debug_open(struct inode *inode, struct file *filp)
 345 {
 346         return single_open(filp, sched_debug_show, NULL);
 347 }
 348
 349 static const struct file_operations sched_debug_fops = {
 350         .open           = sched_debug_open,
 351         .read           = seq_read,
 352         .llseek         = seq_lseek,
 353         .release        = single_release,
 354 };
 355
 356 static int __init init_sched_debug_procfs(void)
 357 {
 358         struct proc_dir_entry *pe;
 359
 360         pe = proc_create("sched_debug", 0444, NULL, &sched_debug_fops);
 361         if (!pe)
 362                 return -ENOMEM;
 363         return 0;
 364 }
 365
 366 __initcall(init_sched_debug_procfs);
 367
 368 void proc_sched_show_task(struct task_struct *p, struct seq_file *m)
 369 {
 370         unsigned long nr_switches;
 371
 372         SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, p->pid,
 373                                                 get_nr_threads(p));
 374         SEQ_printf(m,
 375                 "---------------------------------------------------------\n");
 376 #define __P(F) \
 377         SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)F)
 378 #define P(F) \
 379         SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)p->F)
 380 #define __PN(F) \
 381         SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
 382 #define PN(F) \
 383         SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
 384
 385         PN(se.exec_start);
 386         PN(se.vruntime);
 387         PN(se.sum_exec_runtime);
 388
 389         nr_switches = p->nvcsw + p->nivcsw;
 390
 391 #ifdef CONFIG_SCHEDSTATS
 392         PN(se.statistics.wait_start);
 393         PN(se.statistics.sleep_start);
 394         PN(se.statistics.block_start);
 395         PN(se.statistics.sleep_max);
 396         PN(se.statistics.block_max);
 397         PN(se.statistics.exec_max);
 398         PN(se.statistics.slice_max);
 399         PN(se.statistics.wait_max);
 400         PN(se.statistics.wait_sum);
 401         P(se.statistics.wait_count);
 402         PN(se.statistics.iowait_sum);
 403         P(se.statistics.iowait_count);
 404         P(sched_info.bkl_count);
 405         P(se.nr_migrations);
 406         P(se.statistics.nr_migrations_cold);
 407         P(se.statistics.nr_failed_migrations_affine);
 408         P(se.statistics.nr_failed_migrations_running);
 409         P(se.statistics.nr_failed_migrations_hot);
 410         P(se.statistics.nr_forced_migrations);
 411         P(se.statistics.nr_wakeups);
 412         P(se.statistics.nr_wakeups_sync);
 413         P(se.statistics.nr_wakeups_migrate);
 414         P(se.statistics.nr_wakeups_local);
 415         P(se.statistics.nr_wakeups_remote);
 416         P(se.statistics.nr_wakeups_affine);
 417         P(se.statistics.nr_wakeups_affine_attempts);
 418         P(se.statistics.nr_wakeups_passive);
 419         P(se.statistics.nr_wakeups_idle);
 420
 421         {
 422                 u64 avg_atom, avg_per_cpu;
 423
 424                 avg_atom = p->se.sum_exec_runtime;
 425                 if (nr_switches)
 426                         do_div(avg_atom, nr_switches);
 427                 else
 428                         avg_atom = -1LL;
 429
 430                 avg_per_cpu = p->se.sum_exec_runtime;
 431                 if (p->se.nr_migrations) {
 432                         avg_per_cpu = div64_u64(avg_per_cpu,
 433                                                 p->se.nr_migrations);
 434                 } else {
 435                         avg_per_cpu = -1LL;
 436                 }
 437
 438                 __PN(avg_atom);
 439                 __PN(avg_per_cpu);
 440         }
 441 #endif
 442         __P(nr_switches);
 443         SEQ_printf(m, "%-35s:%21Ld\n",
 444                    "nr_voluntary_switches", (long long)p->nvcsw);
 445         SEQ_printf(m, "%-35s:%21Ld\n",
 446                    "nr_involuntary_switches", (long long)p->nivcsw);
 447
 448         P(se.load.weight);
 449         P(policy);
 450         P(prio);
 451 #undef PN
 452 #undef __PN
 453 #undef P
 454 #undef __P
 455
 456         {
 457                 unsigned int this_cpu = raw_smp_processor_id();
 458                 u64 t0, t1;
 459
 460                 t0 = cpu_clock(this_cpu);
 461                 t1 = cpu_clock(this_cpu);
 462                 SEQ_printf(m, "%-35s:%21Ld\n",
 463                            "clock-delta", (long long)(t1-t0));
 464         }
 465 }
 466
 467 void proc_sched_set_task(struct task_struct *p)
 468 {
 469 #ifdef CONFIG_SCHEDSTATS
 470         memset(&p->se.statistics, 0, sizeof(p->se.statistics));
 471 #endif
 472 }