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,
  58                 struct task_group *tg)
  59 {
  60         struct sched_entity *se = tg->se[cpu];
  61         if (!se)
  62                 return;
  63
  64 #define P(F) \
  65         SEQ_printf(m, "  .%-30s: %lld\n", #F, (long long)F)
  66 #define PN(F) \
  67         SEQ_printf(m, "  .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F))
  68
  69         PN(se->exec_start);
  70         PN(se->vruntime);
  71         PN(se->sum_exec_runtime);
  72 #ifdef CONFIG_SCHEDSTATS
  73         PN(se->wait_start);
  74         PN(se->sleep_start);
  75         PN(se->block_start);
  76         PN(se->sleep_max);
  77         PN(se->block_max);
  78         PN(se->exec_max);
  79         PN(se->slice_max);
  80         PN(se->wait_max);
  81         PN(se->wait_sum);
  82         P(se->wait_count);
  83 #endif
  84         P(se->load.weight);
  85 #undef PN
  86 #undef P
  87 }
  88 #endif
  89
  90 static void
  91 print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
  92 {
  93         if (rq->curr == p)
  94                 SEQ_printf(m, "R");
  95         else
  96                 SEQ_printf(m, " ");
  97
  98         SEQ_printf(m, "%15s %5d %9Ld.%06ld %9Ld %5d ",
  99                 p->comm, p->pid,
 100                 SPLIT_NS(p->se.vruntime),
 101                 (long long)(p->nvcsw + p->nivcsw),
 102                 p->prio);
 103 #ifdef CONFIG_SCHEDSTATS
 104         SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
 105                 SPLIT_NS(p->se.vruntime),
 106                 SPLIT_NS(p->se.sum_exec_runtime),
 107                 SPLIT_NS(p->se.sum_sleep_runtime));
 108 #else
 109         SEQ_printf(m, "%15Ld %15Ld %15Ld.%06ld %15Ld.%06ld %15Ld.%06ld",
 110                 0LL, 0LL, 0LL, 0L, 0LL, 0L, 0LL, 0L);
 111 #endif
 112
 113 #ifdef CONFIG_CGROUP_SCHED
 114         {
 115                 char path[64];
 116
 117                 cgroup_path(task_group(p)->css.cgroup, path, sizeof(path));
 118                 SEQ_printf(m, " %s", path);
 119         }
 120 #endif
 121         SEQ_printf(m, "\n");
 122 }
 123
 124 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
 125 {
 126         struct task_struct *g, *p;
 127         unsigned long flags;
 128
 129         SEQ_printf(m,
 130         "\nrunnable tasks:\n"
 131         "            task   PID         tree-key  switches  prio"
 132         "     exec-runtime         sum-exec        sum-sleep\n"
 133         "------------------------------------------------------"
 134         "----------------------------------------------------\n");
 135
 136         read_lock_irqsave(&tasklist_lock, flags);
 137
 138         do_each_thread(g, p) {
 139                 if (!p->se.on_rq || task_cpu(p) != rq_cpu)
 140                         continue;
 141
 142                 print_task(m, rq, p);
 143         } while_each_thread(g, p);
 144
 145         read_unlock_irqrestore(&tasklist_lock, flags);
 146 }
 147
 148 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
 149 {
 150         s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
 151                 spread, rq0_min_vruntime, spread0;
 152         struct rq *rq = &per_cpu(runqueues, cpu);
 153         struct sched_entity *last;
 154         unsigned long flags;
 155
 156 #if defined(CONFIG_CGROUP_SCHED) && defined(CONFIG_FAIR_GROUP_SCHED)
 157         char path[128] = "";
 158         struct task_group *tg = cfs_rq->tg;
 159
 160         cgroup_path(tg->css.cgroup, path, sizeof(path));
 161
 162         SEQ_printf(m, "\ncfs_rq[%d]:%s\n", cpu, path);
 163 #elif defined(CONFIG_USER_SCHED) && defined(CONFIG_FAIR_GROUP_SCHED)
 164         {
 165                 uid_t uid = cfs_rq->tg->uid;
 166                 SEQ_printf(m, "\ncfs_rq[%d] for UID: %u\n", cpu, uid);
 167         }
 168 #else
 169         SEQ_printf(m, "\ncfs_rq[%d]:\n", cpu);
 170 #endif
 171         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "exec_clock",
 172                         SPLIT_NS(cfs_rq->exec_clock));
 173
 174         spin_lock_irqsave(&rq->lock, flags);
 175         if (cfs_rq->rb_leftmost)
 176                 MIN_vruntime = (__pick_next_entity(cfs_rq))->vruntime;
 177         last = __pick_last_entity(cfs_rq);
 178         if (last)
 179                 max_vruntime = last->vruntime;
 180         min_vruntime = cfs_rq->min_vruntime;
 181         rq0_min_vruntime = per_cpu(runqueues, 0).cfs.min_vruntime;
 182         spin_unlock_irqrestore(&rq->lock, flags);
 183         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "MIN_vruntime",
 184                         SPLIT_NS(MIN_vruntime));
 185         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "min_vruntime",
 186                         SPLIT_NS(min_vruntime));
 187         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "max_vruntime",
 188                         SPLIT_NS(max_vruntime));
 189         spread = max_vruntime - MIN_vruntime;
 190         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread",
 191                         SPLIT_NS(spread));
 192         spread0 = min_vruntime - rq0_min_vruntime;
 193         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread0",
 194                         SPLIT_NS(spread0));
 195         SEQ_printf(m, "  .%-30s: %ld\n", "nr_running", cfs_rq->nr_running);
 196         SEQ_printf(m, "  .%-30s: %ld\n", "load", cfs_rq->load.weight);
 197
 198         SEQ_printf(m, "  .%-30s: %d\n", "nr_spread_over",
 199                         cfs_rq->nr_spread_over);
 200 #ifdef CONFIG_FAIR_GROUP_SCHED
 201 #ifdef CONFIG_SMP
 202         SEQ_printf(m, "  .%-30s: %lu\n", "shares", cfs_rq->shares);
 203 #endif
 204         print_cfs_group_stats(m, cpu, cfs_rq->tg);
 205 #endif
 206 }
 207
 208 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
 209 {
 210 #if defined(CONFIG_CGROUP_SCHED) && defined(CONFIG_RT_GROUP_SCHED)
 211         char path[128] = "";
 212         struct task_group *tg = rt_rq->tg;
 213
 214         cgroup_path(tg->css.cgroup, path, sizeof(path));
 215
 216         SEQ_printf(m, "\nrt_rq[%d]:%s\n", cpu, path);
 217 #else
 218         SEQ_printf(m, "\nrt_rq[%d]:\n", cpu);
 219 #endif
 220
 221
 222 #define P(x) \
 223         SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
 224 #define PN(x) \
 225         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
 226
 227         P(rt_nr_running);
 228         P(rt_throttled);
 229         PN(rt_time);
 230         PN(rt_runtime);
 231
 232 #undef PN
 233 #undef P
 234 }
 235
 236 static void print_cpu(struct seq_file *m, int cpu)
 237 {
 238         struct rq *rq = &per_cpu(runqueues, cpu);
 239
 240 #ifdef CONFIG_X86
 241         {
 242                 unsigned int freq = cpu_khz ? : 1;
 243
 244                 SEQ_printf(m, "\ncpu#%d, %u.%03u MHz\n",
 245                            cpu, freq / 1000, (freq % 1000));
 246         }
 247 #else
 248         SEQ_printf(m, "\ncpu#%d\n", cpu);
 249 #endif
 250
 251 #define P(x) \
 252         SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rq->x))
 253 #define PN(x) \
 254         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
 255
 256         P(nr_running);
 257         SEQ_printf(m, "  .%-30s: %lu\n", "load",
 258                    rq->load.weight);
 259         P(nr_switches);
 260         P(nr_load_updates);
 261         P(nr_uninterruptible);
 262         SEQ_printf(m, "  .%-30s: %lu\n", "jiffies", jiffies);
 263         PN(next_balance);
 264         P(curr->pid);
 265         PN(clock);
 266         P(cpu_load[0]);
 267         P(cpu_load[1]);
 268         P(cpu_load[2]);
 269         P(cpu_load[3]);
 270         P(cpu_load[4]);
 271 #undef P
 272 #undef PN
 273
 274 #ifdef CONFIG_SCHEDSTATS
 275 #define P(n) SEQ_printf(m, "  .%-30s: %d\n", #n, rq->n);
 276
 277         P(yld_exp_empty);
 278         P(yld_act_empty);
 279         P(yld_both_empty);
 280         P(yld_count);
 281
 282         P(sched_switch);
 283         P(sched_count);
 284         P(sched_goidle);
 285
 286         P(ttwu_count);
 287         P(ttwu_local);
 288
 289         P(bkl_count);
 290
 291 #undef P
 292 #endif
 293         print_cfs_stats(m, cpu);
 294         print_rt_stats(m, cpu);
 295
 296         print_rq(m, rq, cpu);
 297 }
 298
 299 static int sched_debug_show(struct seq_file *m, void *v)
 300 {
 301         u64 now = ktime_to_ns(ktime_get());
 302         int cpu;
 303
 304         SEQ_printf(m, "Sched Debug Version: v0.08, %s %.*s\n",
 305                 init_utsname()->release,
 306                 (int)strcspn(init_utsname()->version, " "),
 307                 init_utsname()->version);
 308
 309         SEQ_printf(m, "now at %Lu.%06ld msecs\n", SPLIT_NS(now));
 310
 311 #define P(x) \
 312         SEQ_printf(m, "  .%-40s: %Ld\n", #x, (long long)(x))
 313 #define PN(x) \
 314         SEQ_printf(m, "  .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
 315         PN(sysctl_sched_latency);
 316         PN(sysctl_sched_min_granularity);
 317         PN(sysctl_sched_wakeup_granularity);
 318         PN(sysctl_sched_child_runs_first);
 319         P(sysctl_sched_features);
 320 #undef PN
 321 #undef P
 322
 323         for_each_online_cpu(cpu)
 324                 print_cpu(m, cpu);
 325
 326         SEQ_printf(m, "\n");
 327
 328         return 0;
 329 }
 330
 331 static void sysrq_sched_debug_show(void)
 332 {
 333         sched_debug_show(NULL, NULL);
 334 }
 335
 336 static int sched_debug_open(struct inode *inode, struct file *filp)
 337 {
 338         return single_open(filp, sched_debug_show, NULL);
 339 }
 340
 341 static const struct file_operations sched_debug_fops = {
 342         .open           = sched_debug_open,
 343         .read           = seq_read,
 344         .llseek         = seq_lseek,
 345         .release        = single_release,
 346 };
 347
 348 static int __init init_sched_debug_procfs(void)
 349 {
 350         struct proc_dir_entry *pe;
 351
 352         pe = proc_create("sched_debug", 0444, NULL, &sched_debug_fops);
 353         if (!pe)
 354                 return -ENOMEM;
 355         return 0;
 356 }
 357
 358 __initcall(init_sched_debug_procfs);
 359
 360 void proc_sched_show_task(struct task_struct *p, struct seq_file *m)
 361 {
 362         unsigned long nr_switches;
 363         unsigned long flags;
 364         int num_threads = 1;
 365
 366         if (lock_task_sighand(p, &flags)) {
 367                 num_threads = atomic_read(&p->signal->count);
 368                 unlock_task_sighand(p, &flags);
 369         }
 370
 371         SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, p->pid, num_threads);
 372         SEQ_printf(m,
 373                 "---------------------------------------------------------\n");
 374 #define __P(F) \
 375         SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)F)
 376 #define P(F) \
 377         SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)p->F)
 378 #define __PN(F) \
 379         SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
 380 #define PN(F) \
 381         SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
 382
 383         PN(se.exec_start);
 384         PN(se.vruntime);
 385         PN(se.sum_exec_runtime);
 386         PN(se.avg_overlap);
 387
 388         nr_switches = p->nvcsw + p->nivcsw;
 389
 390 #ifdef CONFIG_SCHEDSTATS
 391         PN(se.wait_start);
 392         PN(se.sleep_start);
 393         PN(se.block_start);
 394         PN(se.sleep_max);
 395         PN(se.block_max);
 396         PN(se.exec_max);
 397         PN(se.slice_max);
 398         PN(se.wait_max);
 399         PN(se.wait_sum);
 400         P(se.wait_count);
 401         P(sched_info.bkl_count);
 402         P(se.nr_migrations);
 403         P(se.nr_migrations_cold);
 404         P(se.nr_failed_migrations_affine);
 405         P(se.nr_failed_migrations_running);
 406         P(se.nr_failed_migrations_hot);
 407         P(se.nr_forced_migrations);
 408         P(se.nr_forced2_migrations);
 409         P(se.nr_wakeups);
 410         P(se.nr_wakeups_sync);
 411         P(se.nr_wakeups_migrate);
 412         P(se.nr_wakeups_local);
 413         P(se.nr_wakeups_remote);
 414         P(se.nr_wakeups_affine);
 415         P(se.nr_wakeups_affine_attempts);
 416         P(se.nr_wakeups_passive);
 417         P(se.nr_wakeups_idle);
 418
 419         {
 420                 u64 avg_atom, avg_per_cpu;
 421
 422                 avg_atom = p->se.sum_exec_runtime;
 423                 if (nr_switches)
 424                         do_div(avg_atom, nr_switches);
 425                 else
 426                         avg_atom = -1LL;
 427
 428                 avg_per_cpu = p->se.sum_exec_runtime;
 429                 if (p->se.nr_migrations) {
 430                         avg_per_cpu = div64_u64(avg_per_cpu,
 431                                                 p->se.nr_migrations);
 432                 } else {
 433                         avg_per_cpu = -1LL;
 434                 }
 435
 436                 __PN(avg_atom);
 437                 __PN(avg_per_cpu);
 438         }
 439 #endif
 440         __P(nr_switches);
 441         SEQ_printf(m, "%-35s:%21Ld\n",
 442                    "nr_voluntary_switches", (long long)p->nvcsw);
 443         SEQ_printf(m, "%-35s:%21Ld\n",
 444                    "nr_involuntary_switches", (long long)p->nivcsw);
 445
 446         P(se.load.weight);
 447         P(policy);
 448         P(prio);
 449 #undef PN
 450 #undef __PN
 451 #undef P
 452 #undef __P
 453
 454         {
 455                 unsigned int this_cpu = raw_smp_processor_id();
 456                 u64 t0, t1;
 457
 458                 t0 = cpu_clock(this_cpu);
 459                 t1 = cpu_clock(this_cpu);
 460                 SEQ_printf(m, "%-35s:%21Ld\n",
 461                            "clock-delta", (long long)(t1-t0));
 462         }
 463 }
 464
 465 void proc_sched_set_task(struct task_struct *p)
 466 {
 467 #ifdef CONFIG_SCHEDSTATS
 468         p->se.wait_max                          = 0;
 469         p->se.wait_sum                          = 0;
 470         p->se.wait_count                        = 0;
 471         p->se.sleep_max                         = 0;
 472         p->se.sum_sleep_runtime                 = 0;
 473         p->se.block_max                         = 0;
 474         p->se.exec_max                          = 0;
 475         p->se.slice_max                         = 0;
 476         p->se.nr_migrations                     = 0;
 477         p->se.nr_migrations_cold                = 0;
 478         p->se.nr_failed_migrations_affine       = 0;
 479         p->se.nr_failed_migrations_running      = 0;
 480         p->se.nr_failed_migrations_hot          = 0;
 481         p->se.nr_forced_migrations              = 0;
 482         p->se.nr_forced2_migrations             = 0;
 483         p->se.nr_wakeups                        = 0;
 484         p->se.nr_wakeups_sync                   = 0;
 485         p->se.nr_wakeups_migrate                = 0;
 486         p->se.nr_wakeups_local                  = 0;
 487         p->se.nr_wakeups_remote                 = 0;
 488         p->se.nr_wakeups_affine                 = 0;
 489         p->se.nr_wakeups_affine_attempts        = 0;
 490         p->se.nr_wakeups_passive                = 0;
 491         p->se.nr_wakeups_idle                   = 0;
 492         p->sched_info.bkl_count                 = 0;
 493 #endif
 494         p->se.sum_exec_runtime                  = 0;
 495         p->se.prev_sum_exec_runtime             = 0;
 496         p->nvcsw                                = 0;
 497         p->nivcsw                               = 0;
 498 }