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         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 #ifdef CONFIG_SMP
  89         P(se->avg.runnable_avg_sum);
  90         P(se->avg.runnable_avg_period);
  91 #endif
  92 #undef PN
  93 #undef P
  94 }
  95 #endif
  96
  97 #ifdef CONFIG_CGROUP_SCHED
  98 static char group_path[PATH_MAX];
  99
 100 static char *task_group_path(struct task_group *tg)
 101 {
 102         if (autogroup_path(tg, group_path, PATH_MAX))
 103                 return group_path;
 104
 105         /*
 106          * May be NULL if the underlying cgroup isn't fully-created yet
 107          */
 108         if (!tg->css.cgroup) {
 109                 group_path[0] = '\0';
 110                 return group_path;
 111         }
 112         cgroup_path(tg->css.cgroup, group_path, PATH_MAX);
 113         return group_path;
 114 }
 115 #endif
 116
 117 static void
 118 print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
 119 {
 120         if (rq->curr == p)
 121                 SEQ_printf(m, "R");
 122         else
 123                 SEQ_printf(m, " ");
 124
 125         SEQ_printf(m, "%15s %5d %9Ld.%06ld %9Ld %5d ",
 126                 p->comm, p->pid,
 127                 SPLIT_NS(p->se.vruntime),
 128                 (long long)(p->nvcsw + p->nivcsw),
 129                 p->prio);
 130 #ifdef CONFIG_SCHEDSTATS
 131         SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
 132                 SPLIT_NS(p->se.vruntime),
 133                 SPLIT_NS(p->se.sum_exec_runtime),
 134                 SPLIT_NS(p->se.statistics.sum_sleep_runtime));
 135 #else
 136         SEQ_printf(m, "%15Ld %15Ld %15Ld.%06ld %15Ld.%06ld %15Ld.%06ld",
 137                 0LL, 0LL, 0LL, 0L, 0LL, 0L, 0LL, 0L);
 138 #endif
 139 #ifdef CONFIG_CGROUP_SCHED
 140         SEQ_printf(m, " %s", task_group_path(task_group(p)));
 141 #endif
 142
 143         SEQ_printf(m, "\n");
 144 }
 145
 146 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
 147 {
 148         struct task_struct *g, *p;
 149         unsigned long flags;
 150
 151         SEQ_printf(m,
 152         "\nrunnable tasks:\n"
 153         "            task   PID         tree-key  switches  prio"
 154         "     exec-runtime         sum-exec        sum-sleep\n"
 155         "------------------------------------------------------"
 156         "----------------------------------------------------\n");
 157
 158         read_lock_irqsave(&tasklist_lock, flags);
 159
 160         do_each_thread(g, p) {
 161                 if (!p->on_rq || task_cpu(p) != rq_cpu)
 162                         continue;
 163
 164                 print_task(m, rq, p);
 165         } while_each_thread(g, p);
 166
 167         read_unlock_irqrestore(&tasklist_lock, flags);
 168 }
 169
 170 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
 171 {
 172         s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
 173                 spread, rq0_min_vruntime, spread0;
 174         struct rq *rq = cpu_rq(cpu);
 175         struct sched_entity *last;
 176         unsigned long flags;
 177
 178 #ifdef CONFIG_FAIR_GROUP_SCHED
 179         SEQ_printf(m, "\ncfs_rq[%d]:%s\n", cpu, task_group_path(cfs_rq->tg));
 180 #else
 181         SEQ_printf(m, "\ncfs_rq[%d]:\n", cpu);
 182 #endif
 183         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "exec_clock",
 184                         SPLIT_NS(cfs_rq->exec_clock));
 185
 186         raw_spin_lock_irqsave(&rq->lock, flags);
 187         if (cfs_rq->rb_leftmost)
 188                 MIN_vruntime = (__pick_first_entity(cfs_rq))->vruntime;
 189         last = __pick_last_entity(cfs_rq);
 190         if (last)
 191                 max_vruntime = last->vruntime;
 192         min_vruntime = cfs_rq->min_vruntime;
 193         rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
 194         raw_spin_unlock_irqrestore(&rq->lock, flags);
 195         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "MIN_vruntime",
 196                         SPLIT_NS(MIN_vruntime));
 197         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "min_vruntime",
 198                         SPLIT_NS(min_vruntime));
 199         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "max_vruntime",
 200                         SPLIT_NS(max_vruntime));
 201         spread = max_vruntime - MIN_vruntime;
 202         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread",
 203                         SPLIT_NS(spread));
 204         spread0 = min_vruntime - rq0_min_vruntime;
 205         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread0",
 206                         SPLIT_NS(spread0));
 207         SEQ_printf(m, "  .%-30s: %d\n", "nr_spread_over",
 208                         cfs_rq->nr_spread_over);
 209         SEQ_printf(m, "  .%-30s: %d\n", "nr_running", cfs_rq->nr_running);
 210         SEQ_printf(m, "  .%-30s: %ld\n", "load", cfs_rq->load.weight);
 211 #ifdef CONFIG_FAIR_GROUP_SCHED
 212 #ifdef CONFIG_SMP
 213         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "load_avg",
 214                         SPLIT_NS(cfs_rq->load_avg));
 215         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "load_period",
 216                         SPLIT_NS(cfs_rq->load_period));
 217         SEQ_printf(m, "  .%-30s: %ld\n", "load_contrib",
 218                         cfs_rq->load_contribution);
 219         SEQ_printf(m, "  .%-30s: %d\n", "load_tg",
 220                         atomic_read(&cfs_rq->tg->load_weight));
 221 #endif
 222
 223         print_cfs_group_stats(m, cpu, cfs_rq->tg);
 224 #endif
 225 }
 226
 227 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
 228 {
 229 #ifdef CONFIG_RT_GROUP_SCHED
 230         SEQ_printf(m, "\nrt_rq[%d]:%s\n", cpu, task_group_path(rt_rq->tg));
 231 #else
 232         SEQ_printf(m, "\nrt_rq[%d]:\n", cpu);
 233 #endif
 234
 235 #define P(x) \
 236         SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
 237 #define PN(x) \
 238         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
 239
 240         P(rt_nr_running);
 241         P(rt_throttled);
 242         PN(rt_time);
 243         PN(rt_runtime);
 244
 245 #undef PN
 246 #undef P
 247 }
 248
 249 extern __read_mostly int sched_clock_running;
 250
 251 static void print_cpu(struct seq_file *m, int cpu)
 252 {
 253         struct rq *rq = cpu_rq(cpu);
 254         unsigned long flags;
 255
 256 #ifdef CONFIG_X86
 257         {
 258                 unsigned int freq = cpu_khz ? : 1;
 259
 260                 SEQ_printf(m, "\ncpu#%d, %u.%03u MHz\n",
 261                            cpu, freq / 1000, (freq % 1000));
 262         }
 263 #else
 264         SEQ_printf(m, "\ncpu#%d\n", cpu);
 265 #endif
 266
 267 #define P(x)                                                            \
 268 do {                                                                    \
 269         if (sizeof(rq->x) == 4)                                         \
 270                 SEQ_printf(m, "  .%-30s: %ld\n", #x, (long)(rq->x));    \
 271         else                                                            \
 272                 SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rq->x));\
 273 } while (0)
 274
 275 #define PN(x) \
 276         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
 277
 278         P(nr_running);
 279         SEQ_printf(m, "  .%-30s: %lu\n", "load",
 280                    rq->load.weight);
 281         P(nr_switches);
 282         P(nr_load_updates);
 283         P(nr_uninterruptible);
 284         PN(next_balance);
 285         P(curr->pid);
 286         PN(clock);
 287         P(cpu_load[0]);
 288         P(cpu_load[1]);
 289         P(cpu_load[2]);
 290         P(cpu_load[3]);
 291         P(cpu_load[4]);
 292 #undef P
 293 #undef PN
 294
 295 #ifdef CONFIG_SCHEDSTATS
 296 #define P(n) SEQ_printf(m, "  .%-30s: %d\n", #n, rq->n);
 297 #define P64(n) SEQ_printf(m, "  .%-30s: %Ld\n", #n, rq->n);
 298
 299         P(yld_count);
 300
 301         P(sched_count);
 302         P(sched_goidle);
 303 #ifdef CONFIG_SMP
 304         P64(avg_idle);
 305 #endif
 306
 307         P(ttwu_count);
 308         P(ttwu_local);
 309
 310 #undef P
 311 #undef P64
 312 #endif
 313         spin_lock_irqsave(&sched_debug_lock, flags);
 314         print_cfs_stats(m, cpu);
 315         print_rt_stats(m, cpu);
 316
 317         rcu_read_lock();
 318         print_rq(m, rq, cpu);
 319         rcu_read_unlock();
 320         spin_unlock_irqrestore(&sched_debug_lock, flags);
 321 }
 322
 323 static const char *sched_tunable_scaling_names[] = {
 324         "none",
 325         "logaritmic",
 326         "linear"
 327 };
 328
 329 static int sched_debug_show(struct seq_file *m, void *v)
 330 {
 331         u64 ktime, sched_clk, cpu_clk;
 332         unsigned long flags;
 333         int cpu;
 334
 335         local_irq_save(flags);
 336         ktime = ktime_to_ns(ktime_get());
 337         sched_clk = sched_clock();
 338         cpu_clk = local_clock();
 339         local_irq_restore(flags);
 340
 341         SEQ_printf(m, "Sched Debug Version: v0.10, %s %.*s\n",
 342                 init_utsname()->release,
 343                 (int)strcspn(init_utsname()->version, " "),
 344                 init_utsname()->version);
 345
 346 #define P(x) \
 347         SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
 348 #define PN(x) \
 349         SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
 350         PN(ktime);
 351         PN(sched_clk);
 352         PN(cpu_clk);
 353         P(jiffies);
 354 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
 355         P(sched_clock_stable);
 356 #endif
 357 #undef PN
 358 #undef P
 359
 360         SEQ_printf(m, "\n");
 361         SEQ_printf(m, "sysctl_sched\n");
 362
 363 #define P(x) \
 364         SEQ_printf(m, "  .%-40s: %Ld\n", #x, (long long)(x))
 365 #define PN(x) \
 366         SEQ_printf(m, "  .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
 367         PN(sysctl_sched_latency);
 368         PN(sysctl_sched_min_granularity);
 369         PN(sysctl_sched_wakeup_granularity);
 370         P(sysctl_sched_child_runs_first);
 371         P(sysctl_sched_features);
 372 #undef PN
 373 #undef P
 374
 375         SEQ_printf(m, "  .%-40s: %d (%s)\n", "sysctl_sched_tunable_scaling",
 376                 sysctl_sched_tunable_scaling,
 377                 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
 378
 379         for_each_online_cpu(cpu)
 380                 print_cpu(m, cpu);
 381
 382         SEQ_printf(m, "\n");
 383
 384         return 0;
 385 }
 386
 387 void sysrq_sched_debug_show(void)
 388 {
 389         sched_debug_show(NULL, NULL);
 390 }
 391
 392 static int sched_debug_open(struct inode *inode, struct file *filp)
 393 {
 394         return single_open(filp, sched_debug_show, NULL);
 395 }
 396
 397 static const struct file_operations sched_debug_fops = {
 398         .open           = sched_debug_open,
 399         .read           = seq_read,
 400         .llseek         = seq_lseek,
 401         .release        = single_release,
 402 };
 403
 404 static int __init init_sched_debug_procfs(void)
 405 {
 406         struct proc_dir_entry *pe;
 407
 408         pe = proc_create("sched_debug", 0444, NULL, &sched_debug_fops);
 409         if (!pe)
 410                 return -ENOMEM;
 411         return 0;
 412 }
 413
 414 __initcall(init_sched_debug_procfs);
 415
 416 void proc_sched_show_task(struct task_struct *p, struct seq_file *m)
 417 {
 418         unsigned long nr_switches;
 419
 420         SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, p->pid,
 421                                                 get_nr_threads(p));
 422         SEQ_printf(m,
 423                 "---------------------------------------------------------\n");
 424 #define __P(F) \
 425         SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)F)
 426 #define P(F) \
 427         SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)p->F)
 428 #define __PN(F) \
 429         SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
 430 #define PN(F) \
 431         SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
 432
 433         PN(se.exec_start);
 434         PN(se.vruntime);
 435         PN(se.sum_exec_runtime);
 436
 437         nr_switches = p->nvcsw + p->nivcsw;
 438
 439 #ifdef CONFIG_SCHEDSTATS
 440         PN(se.statistics.wait_start);
 441         PN(se.statistics.sleep_start);
 442         PN(se.statistics.block_start);
 443         PN(se.statistics.sleep_max);
 444         PN(se.statistics.block_max);
 445         PN(se.statistics.exec_max);
 446         PN(se.statistics.slice_max);
 447         PN(se.statistics.wait_max);
 448         PN(se.statistics.wait_sum);
 449         P(se.statistics.wait_count);
 450         PN(se.statistics.iowait_sum);
 451         P(se.statistics.iowait_count);
 452         P(se.nr_migrations);
 453         P(se.statistics.nr_migrations_cold);
 454         P(se.statistics.nr_failed_migrations_affine);
 455         P(se.statistics.nr_failed_migrations_running);
 456         P(se.statistics.nr_failed_migrations_hot);
 457         P(se.statistics.nr_forced_migrations);
 458         P(se.statistics.nr_wakeups);
 459         P(se.statistics.nr_wakeups_sync);
 460         P(se.statistics.nr_wakeups_migrate);
 461         P(se.statistics.nr_wakeups_local);
 462         P(se.statistics.nr_wakeups_remote);
 463         P(se.statistics.nr_wakeups_affine);
 464         P(se.statistics.nr_wakeups_affine_attempts);
 465         P(se.statistics.nr_wakeups_passive);
 466         P(se.statistics.nr_wakeups_idle);
 467
 468         {
 469                 u64 avg_atom, avg_per_cpu;
 470
 471                 avg_atom = p->se.sum_exec_runtime;
 472                 if (nr_switches)
 473                         do_div(avg_atom, nr_switches);
 474                 else
 475                         avg_atom = -1LL;
 476
 477                 avg_per_cpu = p->se.sum_exec_runtime;
 478                 if (p->se.nr_migrations) {
 479                         avg_per_cpu = div64_u64(avg_per_cpu,
 480                                                 p->se.nr_migrations);
 481                 } else {
 482                         avg_per_cpu = -1LL;
 483                 }
 484
 485                 __PN(avg_atom);
 486                 __PN(avg_per_cpu);
 487         }
 488 #endif
 489         __P(nr_switches);
 490         SEQ_printf(m, "%-35s:%21Ld\n",
 491                    "nr_voluntary_switches", (long long)p->nvcsw);
 492         SEQ_printf(m, "%-35s:%21Ld\n",
 493                    "nr_involuntary_switches", (long long)p->nivcsw);
 494
 495         P(se.load.weight);
 496         P(policy);
 497         P(prio);
 498 #undef PN
 499 #undef __PN
 500 #undef P
 501 #undef __P
 502
 503         {
 504                 unsigned int this_cpu = raw_smp_processor_id();
 505                 u64 t0, t1;
 506
 507                 t0 = cpu_clock(this_cpu);
 508                 t1 = cpu_clock(this_cpu);
 509                 SEQ_printf(m, "%-35s:%21Ld\n",
 510                            "clock-delta", (long long)(t1-t0));
 511         }
 512 }
 513
 514 void proc_sched_set_task(struct task_struct *p)
 515 {
 516 #ifdef CONFIG_SCHEDSTATS
 517         memset(&p->se.statistics, 0, sizeof(p->se.statistics));
 518 #endif
 519 }