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 static void
  57 print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
  58 {
  59         if (rq->curr == p)
  60                 SEQ_printf(m, "R");
  61         else
  62                 SEQ_printf(m, " ");
  63
  64         SEQ_printf(m, "%15s %5d %9Ld.%06ld %9Ld %5d ",
  65                 p->comm, p->pid,
  66                 SPLIT_NS(p->se.vruntime),
  67                 (long long)(p->nvcsw + p->nivcsw),
  68                 p->prio);
  69 #ifdef CONFIG_SCHEDSTATS
  70         SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld\n",
  71                 SPLIT_NS(p->se.vruntime),
  72                 SPLIT_NS(p->se.sum_exec_runtime),
  73                 SPLIT_NS(p->se.sum_sleep_runtime));
  74 #else
  75         SEQ_printf(m, "%15Ld %15Ld %15Ld.%06ld %15Ld.%06ld %15Ld.%06ld\n",
  76                 0LL, 0LL, 0LL, 0L, 0LL, 0L, 0LL, 0L);
  77 #endif
  78 }
  79
  80 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
  81 {
  82         struct task_struct *g, *p;
  83         unsigned long flags;
  84
  85         SEQ_printf(m,
  86         "\nrunnable tasks:\n"
  87         "            task   PID         tree-key  switches  prio"
  88         "     exec-runtime         sum-exec        sum-sleep\n"
  89         "------------------------------------------------------"
  90         "----------------------------------------------------\n");
  91
  92         read_lock_irqsave(&tasklist_lock, flags);
  93
  94         do_each_thread(g, p) {
  95                 if (!p->se.on_rq || task_cpu(p) != rq_cpu)
  96                         continue;
  97
  98                 print_task(m, rq, p);
  99         } while_each_thread(g, p);
 100
 101         read_unlock_irqrestore(&tasklist_lock, flags);
 102 }
 103
 104 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
 105 {
 106         s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
 107                 spread, rq0_min_vruntime, spread0;
 108         struct rq *rq = &per_cpu(runqueues, cpu);
 109         struct sched_entity *last;
 110         unsigned long flags;
 111
 112         SEQ_printf(m, "\ncfs_rq\n");
 113
 114         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "exec_clock",
 115                         SPLIT_NS(cfs_rq->exec_clock));
 116
 117         spin_lock_irqsave(&rq->lock, flags);
 118         if (cfs_rq->rb_leftmost)
 119                 MIN_vruntime = (__pick_next_entity(cfs_rq))->vruntime;
 120         last = __pick_last_entity(cfs_rq);
 121         if (last)
 122                 max_vruntime = last->vruntime;
 123         min_vruntime = rq->cfs.min_vruntime;
 124         rq0_min_vruntime = per_cpu(runqueues, 0).cfs.min_vruntime;
 125         spin_unlock_irqrestore(&rq->lock, flags);
 126         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "MIN_vruntime",
 127                         SPLIT_NS(MIN_vruntime));
 128         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "min_vruntime",
 129                         SPLIT_NS(min_vruntime));
 130         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "max_vruntime",
 131                         SPLIT_NS(max_vruntime));
 132         spread = max_vruntime - MIN_vruntime;
 133         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread",
 134                         SPLIT_NS(spread));
 135         spread0 = min_vruntime - rq0_min_vruntime;
 136         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread0",
 137                         SPLIT_NS(spread0));
 138         SEQ_printf(m, "  .%-30s: %ld\n", "nr_running", cfs_rq->nr_running);
 139         SEQ_printf(m, "  .%-30s: %ld\n", "load", cfs_rq->load.weight);
 140 #ifdef CONFIG_SCHEDSTATS
 141         SEQ_printf(m, "  .%-30s: %d\n", "bkl_count",
 142                         rq->bkl_count);
 143 #endif
 144         SEQ_printf(m, "  .%-30s: %ld\n", "nr_spread_over",
 145                         cfs_rq->nr_spread_over);
 146 }
 147
 148 static void print_cpu(struct seq_file *m, int cpu)
 149 {
 150         struct rq *rq = &per_cpu(runqueues, cpu);
 151
 152 #ifdef CONFIG_X86
 153         {
 154                 unsigned int freq = cpu_khz ? : 1;
 155
 156                 SEQ_printf(m, "\ncpu#%d, %u.%03u MHz\n",
 157                            cpu, freq / 1000, (freq % 1000));
 158         }
 159 #else
 160         SEQ_printf(m, "\ncpu#%d\n", cpu);
 161 #endif
 162
 163 #define P(x) \
 164         SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rq->x))
 165 #define PN(x) \
 166         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
 167
 168         P(nr_running);
 169         SEQ_printf(m, "  .%-30s: %lu\n", "load",
 170                    rq->load.weight);
 171         P(nr_switches);
 172         P(nr_load_updates);
 173         P(nr_uninterruptible);
 174         SEQ_printf(m, "  .%-30s: %lu\n", "jiffies", jiffies);
 175         PN(next_balance);
 176         P(curr->pid);
 177         PN(clock);
 178         PN(idle_clock);
 179         PN(prev_clock_raw);
 180         P(clock_warps);
 181         P(clock_overflows);
 182         P(clock_underflows);
 183         P(clock_deep_idle_events);
 184         PN(clock_max_delta);
 185         P(cpu_load[0]);
 186         P(cpu_load[1]);
 187         P(cpu_load[2]);
 188         P(cpu_load[3]);
 189         P(cpu_load[4]);
 190 #undef P
 191 #undef PN
 192
 193         print_cfs_stats(m, cpu);
 194
 195         print_rq(m, rq, cpu);
 196 }
 197
 198 static int sched_debug_show(struct seq_file *m, void *v)
 199 {
 200         u64 now = ktime_to_ns(ktime_get());
 201         int cpu;
 202
 203         SEQ_printf(m, "Sched Debug Version: v0.07, %s %.*s\n",
 204                 init_utsname()->release,
 205                 (int)strcspn(init_utsname()->version, " "),
 206                 init_utsname()->version);
 207
 208         SEQ_printf(m, "now at %Lu.%06ld msecs\n", SPLIT_NS(now));
 209
 210 #define P(x) \
 211         SEQ_printf(m, "  .%-40s: %Ld\n", #x, (long long)(x))
 212 #define PN(x) \
 213         SEQ_printf(m, "  .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
 214         PN(sysctl_sched_latency);
 215         PN(sysctl_sched_min_granularity);
 216         PN(sysctl_sched_wakeup_granularity);
 217         PN(sysctl_sched_batch_wakeup_granularity);
 218         PN(sysctl_sched_child_runs_first);
 219         P(sysctl_sched_features);
 220 #undef PN
 221 #undef P
 222
 223         for_each_online_cpu(cpu)
 224                 print_cpu(m, cpu);
 225
 226         SEQ_printf(m, "\n");
 227
 228         return 0;
 229 }
 230
 231 static void sysrq_sched_debug_show(void)
 232 {
 233         sched_debug_show(NULL, NULL);
 234 }
 235
 236 static int sched_debug_open(struct inode *inode, struct file *filp)
 237 {
 238         return single_open(filp, sched_debug_show, NULL);
 239 }
 240
 241 static const struct file_operations sched_debug_fops = {
 242         .open           = sched_debug_open,
 243         .read           = seq_read,
 244         .llseek         = seq_lseek,
 245         .release        = single_release,
 246 };
 247
 248 static int __init init_sched_debug_procfs(void)
 249 {
 250         struct proc_dir_entry *pe;
 251
 252         pe = create_proc_entry("sched_debug", 0644, NULL);
 253         if (!pe)
 254                 return -ENOMEM;
 255
 256         pe->proc_fops = &sched_debug_fops;
 257
 258         return 0;
 259 }
 260
 261 __initcall(init_sched_debug_procfs);
 262
 263 void proc_sched_show_task(struct task_struct *p, struct seq_file *m)
 264 {
 265         unsigned long nr_switches;
 266         unsigned long flags;
 267         int num_threads = 1;
 268
 269         rcu_read_lock();
 270         if (lock_task_sighand(p, &flags)) {
 271                 num_threads = atomic_read(&p->signal->count);
 272                 unlock_task_sighand(p, &flags);
 273         }
 274         rcu_read_unlock();
 275
 276         SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, p->pid, num_threads);
 277         SEQ_printf(m,
 278                 "---------------------------------------------------------\n");
 279 #define __P(F) \
 280         SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)F)
 281 #define P(F) \
 282         SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)p->F)
 283 #define __PN(F) \
 284         SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
 285 #define PN(F) \
 286         SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
 287
 288         PN(se.exec_start);
 289         PN(se.vruntime);
 290         PN(se.sum_exec_runtime);
 291
 292         nr_switches = p->nvcsw + p->nivcsw;
 293
 294 #ifdef CONFIG_SCHEDSTATS
 295         PN(se.wait_start);
 296         PN(se.sleep_start);
 297         PN(se.block_start);
 298         PN(se.sleep_max);
 299         PN(se.block_max);
 300         PN(se.exec_max);
 301         PN(se.slice_max);
 302         PN(se.wait_max);
 303         PN(se.wait_sum);
 304         P(se.wait_count);
 305         P(sched_info.bkl_count);
 306         P(se.nr_migrations);
 307         P(se.nr_migrations_cold);
 308         P(se.nr_failed_migrations_affine);
 309         P(se.nr_failed_migrations_running);
 310         P(se.nr_failed_migrations_hot);
 311         P(se.nr_forced_migrations);
 312         P(se.nr_forced2_migrations);
 313         P(se.nr_wakeups);
 314         P(se.nr_wakeups_sync);
 315         P(se.nr_wakeups_migrate);
 316         P(se.nr_wakeups_local);
 317         P(se.nr_wakeups_remote);
 318         P(se.nr_wakeups_affine);
 319         P(se.nr_wakeups_affine_attempts);
 320         P(se.nr_wakeups_passive);
 321         P(se.nr_wakeups_idle);
 322
 323         {
 324                 u64 avg_atom, avg_per_cpu;
 325
 326                 avg_atom = p->se.sum_exec_runtime;
 327                 if (nr_switches)
 328                         do_div(avg_atom, nr_switches);
 329                 else
 330                         avg_atom = -1LL;
 331
 332                 avg_per_cpu = p->se.sum_exec_runtime;
 333                 if (p->se.nr_migrations) {
 334                         avg_per_cpu = div64_64(avg_per_cpu,
 335                                                p->se.nr_migrations);
 336                 } else {
 337                         avg_per_cpu = -1LL;
 338                 }
 339
 340                 __PN(avg_atom);
 341                 __PN(avg_per_cpu);
 342         }
 343 #endif
 344         __P(nr_switches);
 345         SEQ_printf(m, "%-35s:%21Ld\n",
 346                    "nr_voluntary_switches", (long long)p->nvcsw);
 347         SEQ_printf(m, "%-35s:%21Ld\n",
 348                    "nr_involuntary_switches", (long long)p->nivcsw);
 349
 350         P(se.load.weight);
 351         P(policy);
 352         P(prio);
 353 #undef PN
 354 #undef __PN
 355 #undef P
 356 #undef __P
 357
 358         {
 359                 u64 t0, t1;
 360
 361                 t0 = sched_clock();
 362                 t1 = sched_clock();
 363                 SEQ_printf(m, "%-35s:%21Ld\n",
 364                            "clock-delta", (long long)(t1-t0));
 365         }
 366 }
 367
 368 void proc_sched_set_task(struct task_struct *p)
 369 {
 370 #ifdef CONFIG_SCHEDSTATS
 371         p->se.wait_max                          = 0;
 372         p->se.wait_sum                          = 0;
 373         p->se.wait_count                        = 0;
 374         p->se.sleep_max                         = 0;
 375         p->se.sum_sleep_runtime                 = 0;
 376         p->se.block_max                         = 0;
 377         p->se.exec_max                          = 0;
 378         p->se.slice_max                         = 0;
 379         p->se.nr_migrations                     = 0;
 380         p->se.nr_migrations_cold                = 0;
 381         p->se.nr_failed_migrations_affine       = 0;
 382         p->se.nr_failed_migrations_running      = 0;
 383         p->se.nr_failed_migrations_hot          = 0;
 384         p->se.nr_forced_migrations              = 0;
 385         p->se.nr_forced2_migrations             = 0;
 386         p->se.nr_wakeups                        = 0;
 387         p->se.nr_wakeups_sync                   = 0;
 388         p->se.nr_wakeups_migrate                = 0;
 389         p->se.nr_wakeups_local                  = 0;
 390         p->se.nr_wakeups_remote                 = 0;
 391         p->se.nr_wakeups_affine                 = 0;
 392         p->se.nr_wakeups_affine_attempts        = 0;
 393         p->se.nr_wakeups_passive                = 0;
 394         p->se.nr_wakeups_idle                   = 0;
 395         p->sched_info.bkl_count                 = 0;
 396 #endif
 397         p->se.sum_exec_runtime                  = 0;
 398         p->se.prev_sum_exec_runtime             = 0;
 399         p->nvcsw                                = 0;
 400         p->nivcsw                               = 0;
 401 }