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