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->statistics.wait_start);
  74         PN(se->statistics.sleep_start);
  75         PN(se->statistics.block_start);
  76         PN(se->statistics.sleep_max);
  77         PN(se->statistics.block_max);
  78         PN(se->statistics.exec_max);
  79         PN(se->statistics.slice_max);
  80         PN(se->statistics.wait_max);
  81         PN(se->statistics.wait_sum);
  82         P(se->statistics.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.statistics.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 #else
 179         SEQ_printf(m, "\ncfs_rq[%d]:\n", cpu);
 180 #endif
 181         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "exec_clock",
 182                         SPLIT_NS(cfs_rq->exec_clock));
 183
 184         raw_spin_lock_irqsave(&rq->lock, flags);
 185         if (cfs_rq->rb_leftmost)
 186                 MIN_vruntime = (__pick_next_entity(cfs_rq))->vruntime;
 187         last = __pick_last_entity(cfs_rq);
 188         if (last)
 189                 max_vruntime = last->vruntime;
 190         min_vruntime = cfs_rq->min_vruntime;
 191         rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
 192         raw_spin_unlock_irqrestore(&rq->lock, flags);
 193         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "MIN_vruntime",
 194                         SPLIT_NS(MIN_vruntime));
 195         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "min_vruntime",
 196                         SPLIT_NS(min_vruntime));
 197         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "max_vruntime",
 198                         SPLIT_NS(max_vruntime));
 199         spread = max_vruntime - MIN_vruntime;
 200         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread",
 201                         SPLIT_NS(spread));
 202         spread0 = min_vruntime - rq0_min_vruntime;
 203         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread0",
 204                         SPLIT_NS(spread0));
 205         SEQ_printf(m, "  .%-30s: %d\n", "nr_spread_over",
 206                         cfs_rq->nr_spread_over);
 207         SEQ_printf(m, "  .%-30s: %ld\n", "nr_running", cfs_rq->nr_running);
 208         SEQ_printf(m, "  .%-30s: %ld\n", "load", cfs_rq->load.weight);
 209 #ifdef CONFIG_FAIR_GROUP_SCHED
 210 #ifdef CONFIG_SMP
 211         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "load_avg",
 212                         SPLIT_NS(cfs_rq->load_avg));
 213         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "load_period",
 214                         SPLIT_NS(cfs_rq->load_period));
 215         SEQ_printf(m, "  .%-30s: %ld\n", "load_contrib",
 216                         cfs_rq->load_contribution);
 217         SEQ_printf(m, "  .%-30s: %d\n", "load_tg",
 218                         atomic_read(&tg->load_weight));
 219 #endif
 220
 221         print_cfs_group_stats(m, cpu, cfs_rq->tg);
 222 #endif
 223 }
 224
 225 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
 226 {
 227 #if defined(CONFIG_CGROUP_SCHED) && defined(CONFIG_RT_GROUP_SCHED)
 228         char path[128];
 229         struct task_group *tg = rt_rq->tg;
 230
 231         task_group_path(tg, path, sizeof(path));
 232
 233         SEQ_printf(m, "\nrt_rq[%d]:%s\n", cpu, path);
 234 #else
 235         SEQ_printf(m, "\nrt_rq[%d]:\n", cpu);
 236 #endif
 237
 238
 239 #define P(x) \
 240         SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
 241 #define PN(x) \
 242         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
 243
 244         P(rt_nr_running);
 245         P(rt_throttled);
 246         PN(rt_time);
 247         PN(rt_runtime);
 248
 249 #undef PN
 250 #undef P
 251 }
 252
 253 extern __read_mostly int sched_clock_running;
 254
 255 static void print_cpu(struct seq_file *m, int cpu)
 256 {
 257         struct rq *rq = cpu_rq(cpu);
 258
 259 #ifdef CONFIG_X86
 260         {
 261                 unsigned int freq = cpu_khz ? : 1;
 262
 263                 SEQ_printf(m, "\ncpu#%d, %u.%03u MHz\n",
 264                            cpu, freq / 1000, (freq % 1000));
 265         }
 266 #else
 267         SEQ_printf(m, "\ncpu#%d\n", cpu);
 268 #endif
 269
 270 #define P(x) \
 271         SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rq->x))
 272 #define PN(x) \
 273         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
 274
 275         P(nr_running);
 276         SEQ_printf(m, "  .%-30s: %lu\n", "load",
 277                    rq->load.weight);
 278         P(nr_switches);
 279         P(nr_load_updates);
 280         P(nr_uninterruptible);
 281         PN(next_balance);
 282         P(curr->pid);
 283         PN(clock);
 284         P(cpu_load[0]);
 285         P(cpu_load[1]);
 286         P(cpu_load[2]);
 287         P(cpu_load[3]);
 288         P(cpu_load[4]);
 289 #undef P
 290 #undef PN
 291
 292 #ifdef CONFIG_SCHEDSTATS
 293 #define P(n) SEQ_printf(m, "  .%-30s: %d\n", #n, rq->n);
 294 #define P64(n) SEQ_printf(m, "  .%-30s: %Ld\n", #n, rq->n);
 295
 296         P(yld_count);
 297
 298         P(sched_switch);
 299         P(sched_count);
 300         P(sched_goidle);
 301 #ifdef CONFIG_SMP
 302         P64(avg_idle);
 303 #endif
 304
 305         P(ttwu_count);
 306         P(ttwu_local);
 307
 308         P(bkl_count);
 309
 310 #undef P
 311 #endif
 312         print_cfs_stats(m, cpu);
 313         print_rt_stats(m, cpu);
 314
 315         print_rq(m, rq, cpu);
 316 }
 317
 318 static const char *sched_tunable_scaling_names[] = {
 319         "none",
 320         "logaritmic",
 321         "linear"
 322 };
 323
 324 static int sched_debug_show(struct seq_file *m, void *v)
 325 {
 326         u64 ktime, sched_clk, cpu_clk;
 327         unsigned long flags;
 328         int cpu;
 329
 330         local_irq_save(flags);
 331         ktime = ktime_to_ns(ktime_get());
 332         sched_clk = sched_clock();
 333         cpu_clk = local_clock();
 334         local_irq_restore(flags);
 335
 336         SEQ_printf(m, "Sched Debug Version: v0.10, %s %.*s\n",
 337                 init_utsname()->release,
 338                 (int)strcspn(init_utsname()->version, " "),
 339                 init_utsname()->version);
 340
 341 #define P(x) \
 342         SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
 343 #define PN(x) \
 344         SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
 345         PN(ktime);
 346         PN(sched_clk);
 347         PN(cpu_clk);
 348         P(jiffies);
 349 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
 350         P(sched_clock_stable);
 351 #endif
 352 #undef PN
 353 #undef P
 354
 355         SEQ_printf(m, "\n");
 356         SEQ_printf(m, "sysctl_sched\n");
 357
 358 #define P(x) \
 359         SEQ_printf(m, "  .%-40s: %Ld\n", #x, (long long)(x))
 360 #define PN(x) \
 361         SEQ_printf(m, "  .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
 362         PN(sysctl_sched_latency);
 363         PN(sysctl_sched_min_granularity);
 364         PN(sysctl_sched_wakeup_granularity);
 365         P(sysctl_sched_child_runs_first);
 366         P(sysctl_sched_features);
 367 #undef PN
 368 #undef P
 369
 370         SEQ_printf(m, "  .%-40s: %d (%s)\n", "sysctl_sched_tunable_scaling",
 371                 sysctl_sched_tunable_scaling,
 372                 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
 373
 374         for_each_online_cpu(cpu)
 375                 print_cpu(m, cpu);
 376
 377         SEQ_printf(m, "\n");
 378
 379         return 0;
 380 }
 381
 382 static void sysrq_sched_debug_show(void)
 383 {
 384         sched_debug_show(NULL, NULL);
 385 }
 386
 387 static int sched_debug_open(struct inode *inode, struct file *filp)
 388 {
 389         return single_open(filp, sched_debug_show, NULL);
 390 }
 391
 392 static const struct file_operations sched_debug_fops = {
 393         .open           = sched_debug_open,
 394         .read           = seq_read,
 395         .llseek         = seq_lseek,
 396         .release        = single_release,
 397 };
 398
 399 static int __init init_sched_debug_procfs(void)
 400 {
 401         struct proc_dir_entry *pe;
 402
 403         pe = proc_create("sched_debug", 0444, NULL, &sched_debug_fops);
 404         if (!pe)
 405                 return -ENOMEM;
 406         return 0;
 407 }
 408
 409 __initcall(init_sched_debug_procfs);
 410
 411 void proc_sched_show_task(struct task_struct *p, struct seq_file *m)
 412 {
 413         unsigned long nr_switches;
 414
 415         SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, p->pid,
 416                                                 get_nr_threads(p));
 417         SEQ_printf(m,
 418                 "---------------------------------------------------------\n");
 419 #define __P(F) \
 420         SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)F)
 421 #define P(F) \
 422         SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)p->F)
 423 #define __PN(F) \
 424         SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
 425 #define PN(F) \
 426         SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
 427
 428         PN(se.exec_start);
 429         PN(se.vruntime);
 430         PN(se.sum_exec_runtime);
 431
 432         nr_switches = p->nvcsw + p->nivcsw;
 433
 434 #ifdef CONFIG_SCHEDSTATS
 435         PN(se.statistics.wait_start);
 436         PN(se.statistics.sleep_start);
 437         PN(se.statistics.block_start);
 438         PN(se.statistics.sleep_max);
 439         PN(se.statistics.block_max);
 440         PN(se.statistics.exec_max);
 441         PN(se.statistics.slice_max);
 442         PN(se.statistics.wait_max);
 443         PN(se.statistics.wait_sum);
 444         P(se.statistics.wait_count);
 445         PN(se.statistics.iowait_sum);
 446         P(se.statistics.iowait_count);
 447         P(sched_info.bkl_count);
 448         P(se.nr_migrations);
 449         P(se.statistics.nr_migrations_cold);
 450         P(se.statistics.nr_failed_migrations_affine);
 451         P(se.statistics.nr_failed_migrations_running);
 452         P(se.statistics.nr_failed_migrations_hot);
 453         P(se.statistics.nr_forced_migrations);
 454         P(se.statistics.nr_wakeups);
 455         P(se.statistics.nr_wakeups_sync);
 456         P(se.statistics.nr_wakeups_migrate);
 457         P(se.statistics.nr_wakeups_local);
 458         P(se.statistics.nr_wakeups_remote);
 459         P(se.statistics.nr_wakeups_affine);
 460         P(se.statistics.nr_wakeups_affine_attempts);
 461         P(se.statistics.nr_wakeups_passive);
 462         P(se.statistics.nr_wakeups_idle);
 463
 464         {
 465                 u64 avg_atom, avg_per_cpu;
 466
 467                 avg_atom = p->se.sum_exec_runtime;
 468                 if (nr_switches)
 469                         do_div(avg_atom, nr_switches);
 470                 else
 471                         avg_atom = -1LL;
 472
 473                 avg_per_cpu = p->se.sum_exec_runtime;
 474                 if (p->se.nr_migrations) {
 475                         avg_per_cpu = div64_u64(avg_per_cpu,
 476                                                 p->se.nr_migrations);
 477                 } else {
 478                         avg_per_cpu = -1LL;
 479                 }
 480
 481                 __PN(avg_atom);
 482                 __PN(avg_per_cpu);
 483         }
 484 #endif
 485         __P(nr_switches);
 486         SEQ_printf(m, "%-35s:%21Ld\n",
 487                    "nr_voluntary_switches", (long long)p->nvcsw);
 488         SEQ_printf(m, "%-35s:%21Ld\n",
 489                    "nr_involuntary_switches", (long long)p->nivcsw);
 490
 491         P(se.load.weight);
 492         P(policy);
 493         P(prio);
 494 #undef PN
 495 #undef __PN
 496 #undef P
 497 #undef __P
 498
 499         {
 500                 unsigned int this_cpu = raw_smp_processor_id();
 501                 u64 t0, t1;
 502
 503                 t0 = cpu_clock(this_cpu);
 504                 t1 = cpu_clock(this_cpu);
 505                 SEQ_printf(m, "%-35s:%21Ld\n",
 506                            "clock-delta", (long long)(t1-t0));
 507         }
 508 }
 509
 510 void proc_sched_set_task(struct task_struct *p)
 511 {
 512 #ifdef CONFIG_SCHEDSTATS
 513         memset(&p->se.statistics, 0, sizeof(p->se.statistics));
 514 #endif
 515 }