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",
  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",
  76                 0LL, 0LL, 0LL, 0L, 0LL, 0L, 0LL, 0L);
  77 #endif
  78
  79 #ifdef CONFIG_CGROUP_SCHED
  80         {
  81                 char path[64];
  82
  83                 cgroup_path(task_group(p)->css.cgroup, path, sizeof(path));
  84                 SEQ_printf(m, " %s", path);
  85         }
  86 #endif
  87         SEQ_printf(m, "\n");
  88 }
  89
  90 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
  91 {
  92         struct task_struct *g, *p;
  93         unsigned long flags;
  94
  95         SEQ_printf(m,
  96         "\nrunnable tasks:\n"
  97         "            task   PID         tree-key  switches  prio"
  98         "     exec-runtime         sum-exec        sum-sleep\n"
  99         "------------------------------------------------------"
 100         "----------------------------------------------------\n");
 101
 102         read_lock_irqsave(&tasklist_lock, flags);
 103
 104         do_each_thread(g, p) {
 105                 if (!p->se.on_rq || task_cpu(p) != rq_cpu)
 106                         continue;
 107
 108                 print_task(m, rq, p);
 109         } while_each_thread(g, p);
 110
 111         read_unlock_irqrestore(&tasklist_lock, flags);
 112 }
 113
 114 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
 115 {
 116         s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
 117                 spread, rq0_min_vruntime, spread0;
 118         struct rq *rq = &per_cpu(runqueues, cpu);
 119         struct sched_entity *last;
 120         unsigned long flags;
 121
 122 #if !defined(CONFIG_CGROUP_SCHED) || !defined(CONFIG_USER_SCHED)
 123         SEQ_printf(m, "\ncfs_rq[%d]:\n", cpu);
 124 #else
 125         char path[128] = "";
 126         struct cgroup *cgroup = NULL;
 127         struct task_group *tg = cfs_rq->tg;
 128
 129         if (tg)
 130                 cgroup = tg->css.cgroup;
 131
 132         if (cgroup)
 133                 cgroup_path(cgroup, path, sizeof(path));
 134
 135         SEQ_printf(m, "\ncfs_rq[%d]:%s\n", cpu, path);
 136 #endif
 137
 138         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "exec_clock",
 139                         SPLIT_NS(cfs_rq->exec_clock));
 140
 141         spin_lock_irqsave(&rq->lock, flags);
 142         if (cfs_rq->rb_leftmost)
 143                 MIN_vruntime = (__pick_next_entity(cfs_rq))->vruntime;
 144         last = __pick_last_entity(cfs_rq);
 145         if (last)
 146                 max_vruntime = last->vruntime;
 147         min_vruntime = rq->cfs.min_vruntime;
 148         rq0_min_vruntime = per_cpu(runqueues, 0).cfs.min_vruntime;
 149         spin_unlock_irqrestore(&rq->lock, flags);
 150         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "MIN_vruntime",
 151                         SPLIT_NS(MIN_vruntime));
 152         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "min_vruntime",
 153                         SPLIT_NS(min_vruntime));
 154         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "max_vruntime",
 155                         SPLIT_NS(max_vruntime));
 156         spread = max_vruntime - MIN_vruntime;
 157         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread",
 158                         SPLIT_NS(spread));
 159         spread0 = min_vruntime - rq0_min_vruntime;
 160         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread0",
 161                         SPLIT_NS(spread0));
 162         SEQ_printf(m, "  .%-30s: %ld\n", "nr_running", cfs_rq->nr_running);
 163         SEQ_printf(m, "  .%-30s: %ld\n", "load", cfs_rq->load.weight);
 164 #ifdef CONFIG_SCHEDSTATS
 165         SEQ_printf(m, "  .%-30s: %d\n", "bkl_count",
 166                         rq->bkl_count);
 167 #endif
 168         SEQ_printf(m, "  .%-30s: %ld\n", "nr_spread_over",
 169                         cfs_rq->nr_spread_over);
 170 }
 171
 172 static void print_cpu(struct seq_file *m, int cpu)
 173 {
 174         struct rq *rq = &per_cpu(runqueues, cpu);
 175
 176 #ifdef CONFIG_X86
 177         {
 178                 unsigned int freq = cpu_khz ? : 1;
 179
 180                 SEQ_printf(m, "\ncpu#%d, %u.%03u MHz\n",
 181                            cpu, freq / 1000, (freq % 1000));
 182         }
 183 #else
 184         SEQ_printf(m, "\ncpu#%d\n", cpu);
 185 #endif
 186
 187 #define P(x) \
 188         SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rq->x))
 189 #define PN(x) \
 190         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
 191
 192         P(nr_running);
 193         SEQ_printf(m, "  .%-30s: %lu\n", "load",
 194                    rq->load.weight);
 195         P(nr_switches);
 196         P(nr_load_updates);
 197         P(nr_uninterruptible);
 198         SEQ_printf(m, "  .%-30s: %lu\n", "jiffies", jiffies);
 199         PN(next_balance);
 200         P(curr->pid);
 201         PN(clock);
 202         P(cpu_load[0]);
 203         P(cpu_load[1]);
 204         P(cpu_load[2]);
 205         P(cpu_load[3]);
 206         P(cpu_load[4]);
 207 #undef P
 208 #undef PN
 209
 210         print_cfs_stats(m, cpu);
 211
 212         print_rq(m, rq, cpu);
 213 }
 214
 215 static int sched_debug_show(struct seq_file *m, void *v)
 216 {
 217         u64 now = ktime_to_ns(ktime_get());
 218         int cpu;
 219
 220         SEQ_printf(m, "Sched Debug Version: v0.07, %s %.*s\n",
 221                 init_utsname()->release,
 222                 (int)strcspn(init_utsname()->version, " "),
 223                 init_utsname()->version);
 224
 225         SEQ_printf(m, "now at %Lu.%06ld msecs\n", SPLIT_NS(now));
 226
 227 #define P(x) \
 228         SEQ_printf(m, "  .%-40s: %Ld\n", #x, (long long)(x))
 229 #define PN(x) \
 230         SEQ_printf(m, "  .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
 231         PN(sysctl_sched_latency);
 232         PN(sysctl_sched_min_granularity);
 233         PN(sysctl_sched_wakeup_granularity);
 234         PN(sysctl_sched_child_runs_first);
 235         P(sysctl_sched_features);
 236 #undef PN
 237 #undef P
 238
 239         for_each_online_cpu(cpu)
 240                 print_cpu(m, cpu);
 241
 242         SEQ_printf(m, "\n");
 243
 244         return 0;
 245 }
 246
 247 static void sysrq_sched_debug_show(void)
 248 {
 249         sched_debug_show(NULL, NULL);
 250 }
 251
 252 static int sched_debug_open(struct inode *inode, struct file *filp)
 253 {
 254         return single_open(filp, sched_debug_show, NULL);
 255 }
 256
 257 static const struct file_operations sched_debug_fops = {
 258         .open           = sched_debug_open,
 259         .read           = seq_read,
 260         .llseek         = seq_lseek,
 261         .release        = single_release,
 262 };
 263
 264 static int __init init_sched_debug_procfs(void)
 265 {
 266         struct proc_dir_entry *pe;
 267
 268         pe = proc_create("sched_debug", 0644, NULL, &sched_debug_fops);
 269         if (!pe)
 270                 return -ENOMEM;
 271         return 0;
 272 }
 273
 274 __initcall(init_sched_debug_procfs);
 275
 276 void proc_sched_show_task(struct task_struct *p, struct seq_file *m)
 277 {
 278         unsigned long nr_switches;
 279         unsigned long flags;
 280         int num_threads = 1;
 281
 282         rcu_read_lock();
 283         if (lock_task_sighand(p, &flags)) {
 284                 num_threads = atomic_read(&p->signal->count);
 285                 unlock_task_sighand(p, &flags);
 286         }
 287         rcu_read_unlock();
 288
 289         SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, p->pid, num_threads);
 290         SEQ_printf(m,
 291                 "---------------------------------------------------------\n");
 292 #define __P(F) \
 293         SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)F)
 294 #define P(F) \
 295         SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)p->F)
 296 #define __PN(F) \
 297         SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
 298 #define PN(F) \
 299         SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
 300
 301         PN(se.exec_start);
 302         PN(se.vruntime);
 303         PN(se.sum_exec_runtime);
 304         PN(se.avg_overlap);
 305
 306         nr_switches = p->nvcsw + p->nivcsw;
 307
 308 #ifdef CONFIG_SCHEDSTATS
 309         PN(se.wait_start);
 310         PN(se.sleep_start);
 311         PN(se.block_start);
 312         PN(se.sleep_max);
 313         PN(se.block_max);
 314         PN(se.exec_max);
 315         PN(se.slice_max);
 316         PN(se.wait_max);
 317         PN(se.wait_sum);
 318         P(se.wait_count);
 319         P(sched_info.bkl_count);
 320         P(se.nr_migrations);
 321         P(se.nr_migrations_cold);
 322         P(se.nr_failed_migrations_affine);
 323         P(se.nr_failed_migrations_running);
 324         P(se.nr_failed_migrations_hot);
 325         P(se.nr_forced_migrations);
 326         P(se.nr_forced2_migrations);
 327         P(se.nr_wakeups);
 328         P(se.nr_wakeups_sync);
 329         P(se.nr_wakeups_migrate);
 330         P(se.nr_wakeups_local);
 331         P(se.nr_wakeups_remote);
 332         P(se.nr_wakeups_affine);
 333         P(se.nr_wakeups_affine_attempts);
 334         P(se.nr_wakeups_passive);
 335         P(se.nr_wakeups_idle);
 336
 337         {
 338                 u64 avg_atom, avg_per_cpu;
 339
 340                 avg_atom = p->se.sum_exec_runtime;
 341                 if (nr_switches)
 342                         do_div(avg_atom, nr_switches);
 343                 else
 344                         avg_atom = -1LL;
 345
 346                 avg_per_cpu = p->se.sum_exec_runtime;
 347                 if (p->se.nr_migrations) {
 348                         avg_per_cpu = div64_u64(avg_per_cpu,
 349                                                 p->se.nr_migrations);
 350                 } else {
 351                         avg_per_cpu = -1LL;
 352                 }
 353
 354                 __PN(avg_atom);
 355                 __PN(avg_per_cpu);
 356         }
 357 #endif
 358         __P(nr_switches);
 359         SEQ_printf(m, "%-35s:%21Ld\n",
 360                    "nr_voluntary_switches", (long long)p->nvcsw);
 361         SEQ_printf(m, "%-35s:%21Ld\n",
 362                    "nr_involuntary_switches", (long long)p->nivcsw);
 363
 364         P(se.load.weight);
 365         P(policy);
 366         P(prio);
 367 #undef PN
 368 #undef __PN
 369 #undef P
 370 #undef __P
 371
 372         {
 373                 u64 t0, t1;
 374
 375                 t0 = sched_clock();
 376                 t1 = sched_clock();
 377                 SEQ_printf(m, "%-35s:%21Ld\n",
 378                            "clock-delta", (long long)(t1-t0));
 379         }
 380 }
 381
 382 void proc_sched_set_task(struct task_struct *p)
 383 {
 384 #ifdef CONFIG_SCHEDSTATS
 385         p->se.wait_max                          = 0;
 386         p->se.wait_sum                          = 0;
 387         p->se.wait_count                        = 0;
 388         p->se.sleep_max                         = 0;
 389         p->se.sum_sleep_runtime                 = 0;
 390         p->se.block_max                         = 0;
 391         p->se.exec_max                          = 0;
 392         p->se.slice_max                         = 0;
 393         p->se.nr_migrations                     = 0;
 394         p->se.nr_migrations_cold                = 0;
 395         p->se.nr_failed_migrations_affine       = 0;
 396         p->se.nr_failed_migrations_running      = 0;
 397         p->se.nr_failed_migrations_hot          = 0;
 398         p->se.nr_forced_migrations              = 0;
 399         p->se.nr_forced2_migrations             = 0;
 400         p->se.nr_wakeups                        = 0;
 401         p->se.nr_wakeups_sync                   = 0;
 402         p->se.nr_wakeups_migrate                = 0;
 403         p->se.nr_wakeups_local                  = 0;
 404         p->se.nr_wakeups_remote                 = 0;
 405         p->se.nr_wakeups_affine                 = 0;
 406         p->se.nr_wakeups_affine_attempts        = 0;
 407         p->se.nr_wakeups_passive                = 0;
 408         p->se.nr_wakeups_idle                   = 0;
 409         p->sched_info.bkl_count                 = 0;
 410 #endif
 411         p->se.sum_exec_runtime                  = 0;
 412         p->se.prev_sum_exec_runtime             = 0;
 413         p->nvcsw                                = 0;
 414         p->nivcsw                               = 0;
 415 }