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                 rcu_read_lock();
 118                 cgroup_path(task_group(p)->css.cgroup, path, sizeof(path));
 119                 rcu_read_unlock();
 120                 SEQ_printf(m, " %s", path);
 121         }
 122 #endif
 123         SEQ_printf(m, "\n");
 124 }
 125
 126 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
 127 {
 128         struct task_struct *g, *p;
 129         unsigned long flags;
 130
 131         SEQ_printf(m,
 132         "\nrunnable tasks:\n"
 133         "            task   PID         tree-key  switches  prio"
 134         "     exec-runtime         sum-exec        sum-sleep\n"
 135         "------------------------------------------------------"
 136         "----------------------------------------------------\n");
 137
 138         read_lock_irqsave(&tasklist_lock, flags);
 139
 140         do_each_thread(g, p) {
 141                 if (!p->se.on_rq || task_cpu(p) != rq_cpu)
 142                         continue;
 143
 144                 print_task(m, rq, p);
 145         } while_each_thread(g, p);
 146
 147         read_unlock_irqrestore(&tasklist_lock, flags);
 148 }
 149
 150 #if defined(CONFIG_CGROUP_SCHED) && \
 151         (defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED))
 152 static void task_group_path(struct task_group *tg, char *buf, int buflen)
 153 {
 154         /* may be NULL if the underlying cgroup isn't fully-created yet */
 155         if (!tg->css.cgroup) {
 156                 buf[0] = '\0';
 157                 return;
 158         }
 159         cgroup_path(tg->css.cgroup, buf, buflen);
 160 }
 161 #endif
 162
 163 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
 164 {
 165         s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
 166                 spread, rq0_min_vruntime, spread0;
 167         struct rq *rq = cpu_rq(cpu);
 168         struct sched_entity *last;
 169         unsigned long flags;
 170
 171 #if defined(CONFIG_CGROUP_SCHED) && defined(CONFIG_FAIR_GROUP_SCHED)
 172         char path[128];
 173         struct task_group *tg = cfs_rq->tg;
 174
 175         task_group_path(tg, path, sizeof(path));
 176
 177         SEQ_printf(m, "\ncfs_rq[%d]:%s\n", cpu, path);
 178 #elif defined(CONFIG_USER_SCHED) && defined(CONFIG_FAIR_GROUP_SCHED)
 179         {
 180                 uid_t uid = cfs_rq->tg->uid;
 181                 SEQ_printf(m, "\ncfs_rq[%d] for UID: %u\n", cpu, uid);
 182         }
 183 #else
 184         SEQ_printf(m, "\ncfs_rq[%d]:\n", cpu);
 185 #endif
 186         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "exec_clock",
 187                         SPLIT_NS(cfs_rq->exec_clock));
 188
 189         raw_spin_lock_irqsave(&rq->lock, flags);
 190         if (cfs_rq->rb_leftmost)
 191                 MIN_vruntime = (__pick_next_entity(cfs_rq))->vruntime;
 192         last = __pick_last_entity(cfs_rq);
 193         if (last)
 194                 max_vruntime = last->vruntime;
 195         min_vruntime = cfs_rq->min_vruntime;
 196         rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
 197         raw_spin_unlock_irqrestore(&rq->lock, flags);
 198         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "MIN_vruntime",
 199                         SPLIT_NS(MIN_vruntime));
 200         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "min_vruntime",
 201                         SPLIT_NS(min_vruntime));
 202         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "max_vruntime",
 203                         SPLIT_NS(max_vruntime));
 204         spread = max_vruntime - MIN_vruntime;
 205         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread",
 206                         SPLIT_NS(spread));
 207         spread0 = min_vruntime - rq0_min_vruntime;
 208         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread0",
 209                         SPLIT_NS(spread0));
 210         SEQ_printf(m, "  .%-30s: %ld\n", "nr_running", cfs_rq->nr_running);
 211         SEQ_printf(m, "  .%-30s: %ld\n", "load", cfs_rq->load.weight);
 212
 213         SEQ_printf(m, "  .%-30s: %d\n", "nr_spread_over",
 214                         cfs_rq->nr_spread_over);
 215 #ifdef CONFIG_FAIR_GROUP_SCHED
 216 #ifdef CONFIG_SMP
 217         SEQ_printf(m, "  .%-30s: %lu\n", "shares", cfs_rq->shares);
 218 #endif
 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 #if defined(CONFIG_CGROUP_SCHED) && defined(CONFIG_RT_GROUP_SCHED)
 226         char path[128];
 227         struct task_group *tg = rt_rq->tg;
 228
 229         task_group_path(tg, path, sizeof(path));
 230
 231         SEQ_printf(m, "\nrt_rq[%d]:%s\n", cpu, path);
 232 #else
 233         SEQ_printf(m, "\nrt_rq[%d]:\n", cpu);
 234 #endif
 235
 236
 237 #define P(x) \
 238         SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
 239 #define PN(x) \
 240         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
 241
 242         P(rt_nr_running);
 243         P(rt_throttled);
 244         PN(rt_time);
 245         PN(rt_runtime);
 246
 247 #undef PN
 248 #undef P
 249 }
 250
 251 static void print_cpu(struct seq_file *m, int cpu)
 252 {
 253         struct rq *rq = cpu_rq(cpu);
 254
 255 #ifdef CONFIG_X86
 256         {
 257                 unsigned int freq = cpu_khz ? : 1;
 258
 259                 SEQ_printf(m, "\ncpu#%d, %u.%03u MHz\n",
 260                            cpu, freq / 1000, (freq % 1000));
 261         }
 262 #else
 263         SEQ_printf(m, "\ncpu#%d\n", cpu);
 264 #endif
 265
 266 #define P(x) \
 267         SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rq->x))
 268 #define PN(x) \
 269         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
 270
 271         P(nr_running);
 272         SEQ_printf(m, "  .%-30s: %lu\n", "load",
 273                    rq->load.weight);
 274         P(nr_switches);
 275         P(nr_load_updates);
 276         P(nr_uninterruptible);
 277         PN(next_balance);
 278         P(curr->pid);
 279         PN(clock);
 280         P(cpu_load[0]);
 281         P(cpu_load[1]);
 282         P(cpu_load[2]);
 283         P(cpu_load[3]);
 284         P(cpu_load[4]);
 285 #undef P
 286 #undef PN
 287
 288 #ifdef CONFIG_SCHEDSTATS
 289 #define P(n) SEQ_printf(m, "  .%-30s: %d\n", #n, rq->n);
 290 #define P64(n) SEQ_printf(m, "  .%-30s: %Ld\n", #n, rq->n);
 291
 292         P(yld_count);
 293
 294         P(sched_switch);
 295         P(sched_count);
 296         P(sched_goidle);
 297 #ifdef CONFIG_SMP
 298         P64(avg_idle);
 299 #endif
 300
 301         P(ttwu_count);
 302         P(ttwu_local);
 303
 304         P(bkl_count);
 305
 306 #undef P
 307 #endif
 308         print_cfs_stats(m, cpu);
 309         print_rt_stats(m, cpu);
 310
 311         print_rq(m, rq, cpu);
 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 now = ktime_to_ns(ktime_get());
 323         int cpu;
 324
 325         SEQ_printf(m, "Sched Debug Version: v0.09, %s %.*s\n",
 326                 init_utsname()->release,
 327                 (int)strcspn(init_utsname()->version, " "),
 328                 init_utsname()->version);
 329
 330         SEQ_printf(m, "now at %Lu.%06ld msecs\n", SPLIT_NS(now));
 331
 332 #define P(x) \
 333         SEQ_printf(m, "  .%-40s: %Ld\n", #x, (long long)(x))
 334 #define PN(x) \
 335         SEQ_printf(m, "  .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
 336         P(jiffies);
 337         PN(sysctl_sched_latency);
 338         PN(sysctl_sched_min_granularity);
 339         PN(sysctl_sched_wakeup_granularity);
 340         PN(sysctl_sched_child_runs_first);
 341         P(sysctl_sched_features);
 342 #undef PN
 343 #undef P
 344
 345         SEQ_printf(m, "  .%-40s: %d (%s)\n", "sysctl_sched_tunable_scaling",
 346                 sysctl_sched_tunable_scaling,
 347                 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
 348
 349         for_each_online_cpu(cpu)
 350                 print_cpu(m, cpu);
 351
 352         SEQ_printf(m, "\n");
 353
 354         return 0;
 355 }
 356
 357 static void sysrq_sched_debug_show(void)
 358 {
 359         sched_debug_show(NULL, NULL);
 360 }
 361
 362 static int sched_debug_open(struct inode *inode, struct file *filp)
 363 {
 364         return single_open(filp, sched_debug_show, NULL);
 365 }
 366
 367 static const struct file_operations sched_debug_fops = {
 368         .open           = sched_debug_open,
 369         .read           = seq_read,
 370         .llseek         = seq_lseek,
 371         .release        = single_release,
 372 };
 373
 374 static int __init init_sched_debug_procfs(void)
 375 {
 376         struct proc_dir_entry *pe;
 377
 378         pe = proc_create("sched_debug", 0444, NULL, &sched_debug_fops);
 379         if (!pe)
 380                 return -ENOMEM;
 381         return 0;
 382 }
 383
 384 __initcall(init_sched_debug_procfs);
 385
 386 void proc_sched_show_task(struct task_struct *p, struct seq_file *m)
 387 {
 388         unsigned long nr_switches;
 389         unsigned long flags;
 390         int num_threads = 1;
 391
 392         if (lock_task_sighand(p, &flags)) {
 393                 num_threads = atomic_read(&p->signal->count);
 394                 unlock_task_sighand(p, &flags);
 395         }
 396
 397         SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, p->pid, num_threads);
 398         SEQ_printf(m,
 399                 "---------------------------------------------------------\n");
 400 #define __P(F) \
 401         SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)F)
 402 #define P(F) \
 403         SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)p->F)
 404 #define __PN(F) \
 405         SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
 406 #define PN(F) \
 407         SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
 408
 409         PN(se.exec_start);
 410         PN(se.vruntime);
 411         PN(se.sum_exec_runtime);
 412         PN(se.avg_overlap);
 413         PN(se.avg_wakeup);
 414
 415         nr_switches = p->nvcsw + p->nivcsw;
 416
 417 #ifdef CONFIG_SCHEDSTATS
 418         PN(se.wait_start);
 419         PN(se.sleep_start);
 420         PN(se.block_start);
 421         PN(se.sleep_max);
 422         PN(se.block_max);
 423         PN(se.exec_max);
 424         PN(se.slice_max);
 425         PN(se.wait_max);
 426         PN(se.wait_sum);
 427         P(se.wait_count);
 428         PN(se.iowait_sum);
 429         P(se.iowait_count);
 430         P(sched_info.bkl_count);
 431         P(se.nr_migrations);
 432         P(se.nr_migrations_cold);
 433         P(se.nr_failed_migrations_affine);
 434         P(se.nr_failed_migrations_running);
 435         P(se.nr_failed_migrations_hot);
 436         P(se.nr_forced_migrations);
 437         P(se.nr_wakeups);
 438         P(se.nr_wakeups_sync);
 439         P(se.nr_wakeups_migrate);
 440         P(se.nr_wakeups_local);
 441         P(se.nr_wakeups_remote);
 442         P(se.nr_wakeups_affine);
 443         P(se.nr_wakeups_affine_attempts);
 444         P(se.nr_wakeups_passive);
 445         P(se.nr_wakeups_idle);
 446
 447         {
 448                 u64 avg_atom, avg_per_cpu;
 449
 450                 avg_atom = p->se.sum_exec_runtime;
 451                 if (nr_switches)
 452                         do_div(avg_atom, nr_switches);
 453                 else
 454                         avg_atom = -1LL;
 455
 456                 avg_per_cpu = p->se.sum_exec_runtime;
 457                 if (p->se.nr_migrations) {
 458                         avg_per_cpu = div64_u64(avg_per_cpu,
 459                                                 p->se.nr_migrations);
 460                 } else {
 461                         avg_per_cpu = -1LL;
 462                 }
 463
 464                 __PN(avg_atom);
 465                 __PN(avg_per_cpu);
 466         }
 467 #endif
 468         __P(nr_switches);
 469         SEQ_printf(m, "%-35s:%21Ld\n",
 470                    "nr_voluntary_switches", (long long)p->nvcsw);
 471         SEQ_printf(m, "%-35s:%21Ld\n",
 472                    "nr_involuntary_switches", (long long)p->nivcsw);
 473
 474         P(se.load.weight);
 475         P(policy);
 476         P(prio);
 477 #undef PN
 478 #undef __PN
 479 #undef P
 480 #undef __P
 481
 482         {
 483                 unsigned int this_cpu = raw_smp_processor_id();
 484                 u64 t0, t1;
 485
 486                 t0 = cpu_clock(this_cpu);
 487                 t1 = cpu_clock(this_cpu);
 488                 SEQ_printf(m, "%-35s:%21Ld\n",
 489                            "clock-delta", (long long)(t1-t0));
 490         }
 491 }
 492
 493 void proc_sched_set_task(struct task_struct *p)
 494 {
 495 #ifdef CONFIG_SCHEDSTATS
 496         p->se.wait_max                          = 0;
 497         p->se.wait_sum                          = 0;
 498         p->se.wait_count                        = 0;
 499         p->se.iowait_sum                        = 0;
 500         p->se.iowait_count                      = 0;
 501         p->se.sleep_max                         = 0;
 502         p->se.sum_sleep_runtime                 = 0;
 503         p->se.block_max                         = 0;
 504         p->se.exec_max                          = 0;
 505         p->se.slice_max                         = 0;
 506         p->se.nr_migrations                     = 0;
 507         p->se.nr_migrations_cold                = 0;
 508         p->se.nr_failed_migrations_affine       = 0;
 509         p->se.nr_failed_migrations_running      = 0;
 510         p->se.nr_failed_migrations_hot          = 0;
 511         p->se.nr_forced_migrations              = 0;
 512         p->se.nr_wakeups                        = 0;
 513         p->se.nr_wakeups_sync                   = 0;
 514         p->se.nr_wakeups_migrate                = 0;
 515         p->se.nr_wakeups_local                  = 0;
 516         p->se.nr_wakeups_remote                 = 0;
 517         p->se.nr_wakeups_affine                 = 0;
 518         p->se.nr_wakeups_affine_attempts        = 0;
 519         p->se.nr_wakeups_passive                = 0;
 520         p->se.nr_wakeups_idle                   = 0;
 521         p->sched_info.bkl_count                 = 0;
 522 #endif
 523 }