kernel/sched/debug.c

   1 /*
   2  * kernel/sched/debug.c
   3  *
   4  * Print the CFS rbtree and other debugging details
   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 #include "sched.h"
  13
  14 static DEFINE_SPINLOCK(sched_debug_lock);
  15
  16 /*
  17  * This allows printing both to /proc/sched_debug and
  18  * to the console
  19  */
  20 #define SEQ_printf(m, x...)                     \
  21  do {                                           \
  22         if (m)                                  \
  23                 seq_printf(m, x);               \
  24         else                                    \
  25                 pr_cont(x);                     \
  26  } while (0)
  27
  28 /*
  29  * Ease the printing of nsec fields:
  30  */
  31 static long long nsec_high(unsigned long long nsec)
  32 {
  33         if ((long long)nsec < 0) {
  34                 nsec = -nsec;
  35                 do_div(nsec, 1000000);
  36                 return -nsec;
  37         }
  38         do_div(nsec, 1000000);
  39
  40         return nsec;
  41 }
  42
  43 static unsigned long nsec_low(unsigned long long nsec)
  44 {
  45         if ((long long)nsec < 0)
  46                 nsec = -nsec;
  47
  48         return do_div(nsec, 1000000);
  49 }
  50
  51 #define SPLIT_NS(x) nsec_high(x), nsec_low(x)
  52
  53 #define SCHED_FEAT(name, enabled)       \
  54         #name ,
  55
  56 static const char * const sched_feat_names[] = {
  57 #include "features.h"
  58 };
  59
  60 #undef SCHED_FEAT
  61
  62 static int sched_feat_show(struct seq_file *m, void *v)
  63 {
  64         int i;
  65
  66         for (i = 0; i < __SCHED_FEAT_NR; i++) {
  67                 if (!(sysctl_sched_features & (1UL << i)))
  68                         seq_puts(m, "NO_");
  69                 seq_printf(m, "%s ", sched_feat_names[i]);
  70         }
  71         seq_puts(m, "\n");
  72
  73         return 0;
  74 }
  75
  76 #ifdef HAVE_JUMP_LABEL
  77
  78 #define jump_label_key__true  STATIC_KEY_INIT_TRUE
  79 #define jump_label_key__false STATIC_KEY_INIT_FALSE
  80
  81 #define SCHED_FEAT(name, enabled)       \
  82         jump_label_key__##enabled ,
  83
  84 struct static_key sched_feat_keys[__SCHED_FEAT_NR] = {
  85 #include "features.h"
  86 };
  87
  88 #undef SCHED_FEAT
  89
  90 static void sched_feat_disable(int i)
  91 {
  92         static_key_disable(&sched_feat_keys[i]);
  93 }
  94
  95 static void sched_feat_enable(int i)
  96 {
  97         static_key_enable(&sched_feat_keys[i]);
  98 }
  99 #else
 100 static void sched_feat_disable(int i) { };
 101 static void sched_feat_enable(int i) { };
 102 #endif /* HAVE_JUMP_LABEL */
 103
 104 static int sched_feat_set(char *cmp)
 105 {
 106         int i;
 107         int neg = 0;
 108
 109         if (strncmp(cmp, "NO_", 3) == 0) {
 110                 neg = 1;
 111                 cmp += 3;
 112         }
 113
 114         for (i = 0; i < __SCHED_FEAT_NR; i++) {
 115                 if (strcmp(cmp, sched_feat_names[i]) == 0) {
 116                         if (neg) {
 117                                 sysctl_sched_features &= ~(1UL << i);
 118                                 sched_feat_disable(i);
 119                         } else {
 120                                 sysctl_sched_features |= (1UL << i);
 121                                 sched_feat_enable(i);
 122                         }
 123                         break;
 124                 }
 125         }
 126
 127         return i;
 128 }
 129
 130 static ssize_t
 131 sched_feat_write(struct file *filp, const char __user *ubuf,
 132                 size_t cnt, loff_t *ppos)
 133 {
 134         char buf[64];
 135         char *cmp;
 136         int i;
 137         struct inode *inode;
 138
 139         if (cnt > 63)
 140                 cnt = 63;
 141
 142         if (copy_from_user(&buf, ubuf, cnt))
 143                 return -EFAULT;
 144
 145         buf[cnt] = 0;
 146         cmp = strstrip(buf);
 147
 148         /* Ensure the static_key remains in a consistent state */
 149         inode = file_inode(filp);
 150         inode_lock(inode);
 151         i = sched_feat_set(cmp);
 152         inode_unlock(inode);
 153         if (i == __SCHED_FEAT_NR)
 154                 return -EINVAL;
 155
 156         *ppos += cnt;
 157
 158         return cnt;
 159 }
 160
 161 static int sched_feat_open(struct inode *inode, struct file *filp)
 162 {
 163         return single_open(filp, sched_feat_show, NULL);
 164 }
 165
 166 static const struct file_operations sched_feat_fops = {
 167         .open           = sched_feat_open,
 168         .write          = sched_feat_write,
 169         .read           = seq_read,
 170         .llseek         = seq_lseek,
 171         .release        = single_release,
 172 };
 173
 174 __read_mostly bool sched_debug_enabled;
 175
 176 static __init int sched_init_debug(void)
 177 {
 178         debugfs_create_file("sched_features", 0644, NULL, NULL,
 179                         &sched_feat_fops);
 180
 181         debugfs_create_bool("sched_debug", 0644, NULL,
 182                         &sched_debug_enabled);
 183
 184         return 0;
 185 }
 186 late_initcall(sched_init_debug);
 187
 188 #ifdef CONFIG_SMP
 189
 190 #ifdef CONFIG_SYSCTL
 191
 192 static struct ctl_table sd_ctl_dir[] = {
 193         {
 194                 .procname       = "sched_domain",
 195                 .mode           = 0555,
 196         },
 197         {}
 198 };
 199
 200 static struct ctl_table sd_ctl_root[] = {
 201         {
 202                 .procname       = "kernel",
 203                 .mode           = 0555,
 204                 .child          = sd_ctl_dir,
 205         },
 206         {}
 207 };
 208
 209 static struct ctl_table *sd_alloc_ctl_entry(int n)
 210 {
 211         struct ctl_table *entry =
 212                 kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL);
 213
 214         return entry;
 215 }
 216
 217 static void sd_free_ctl_entry(struct ctl_table **tablep)
 218 {
 219         struct ctl_table *entry;
 220
 221         /*
 222          * In the intermediate directories, both the child directory and
 223          * procname are dynamically allocated and could fail but the mode
 224          * will always be set. In the lowest directory the names are
 225          * static strings and all have proc handlers.
 226          */
 227         for (entry = *tablep; entry->mode; entry++) {
 228                 if (entry->child)
 229                         sd_free_ctl_entry(&entry->child);
 230                 if (entry->proc_handler == NULL)
 231                         kfree(entry->procname);
 232         }
 233
 234         kfree(*tablep);
 235         *tablep = NULL;
 236 }
 237
 238 static int min_load_idx = 0;
 239 static int max_load_idx = CPU_LOAD_IDX_MAX-1;
 240
 241 static void
 242 set_table_entry(struct ctl_table *entry,
 243                 const char *procname, void *data, int maxlen,
 244                 umode_t mode, proc_handler *proc_handler,
 245                 bool load_idx)
 246 {
 247         entry->procname = procname;
 248         entry->data = data;
 249         entry->maxlen = maxlen;
 250         entry->mode = mode;
 251         entry->proc_handler = proc_handler;
 252
 253         if (load_idx) {
 254                 entry->extra1 = &min_load_idx;
 255                 entry->extra2 = &max_load_idx;
 256         }
 257 }
 258
 259 static struct ctl_table *
 260 sd_alloc_ctl_domain_table(struct sched_domain *sd)
 261 {
 262         struct ctl_table *table = sd_alloc_ctl_entry(14);
 263
 264         if (table == NULL)
 265                 return NULL;
 266
 267         set_table_entry(&table[0] , "min_interval",        &sd->min_interval,        sizeof(long), 0644, proc_doulongvec_minmax, false);
 268         set_table_entry(&table[1] , "max_interval",        &sd->max_interval,        sizeof(long), 0644, proc_doulongvec_minmax, false);
 269         set_table_entry(&table[2] , "busy_idx",            &sd->busy_idx,            sizeof(int) , 0644, proc_dointvec_minmax,   true );
 270         set_table_entry(&table[3] , "idle_idx",            &sd->idle_idx,            sizeof(int) , 0644, proc_dointvec_minmax,   true );
 271         set_table_entry(&table[4] , "newidle_idx",         &sd->newidle_idx,         sizeof(int) , 0644, proc_dointvec_minmax,   true );
 272         set_table_entry(&table[5] , "wake_idx",            &sd->wake_idx,            sizeof(int) , 0644, proc_dointvec_minmax,   true );
 273         set_table_entry(&table[6] , "forkexec_idx",        &sd->forkexec_idx,        sizeof(int) , 0644, proc_dointvec_minmax,   true );
 274         set_table_entry(&table[7] , "busy_factor",         &sd->busy_factor,         sizeof(int) , 0644, proc_dointvec_minmax,   false);
 275         set_table_entry(&table[8] , "imbalance_pct",       &sd->imbalance_pct,       sizeof(int) , 0644, proc_dointvec_minmax,   false);
 276         set_table_entry(&table[9] , "cache_nice_tries",    &sd->cache_nice_tries,    sizeof(int) , 0644, proc_dointvec_minmax,   false);
 277         set_table_entry(&table[10], "flags",               &sd->flags,               sizeof(int) , 0644, proc_dointvec_minmax,   false);
 278         set_table_entry(&table[11], "max_newidle_lb_cost", &sd->max_newidle_lb_cost, sizeof(long), 0644, proc_doulongvec_minmax, false);
 279         set_table_entry(&table[12], "name",                sd->name,            CORENAME_MAX_SIZE, 0444, proc_dostring,          false);
 280         /* &table[13] is terminator */
 281
 282         return table;
 283 }
 284
 285 static struct ctl_table *sd_alloc_ctl_cpu_table(int cpu)
 286 {
 287         struct ctl_table *entry, *table;
 288         struct sched_domain *sd;
 289         int domain_num = 0, i;
 290         char buf[32];
 291
 292         for_each_domain(cpu, sd)
 293                 domain_num++;
 294         entry = table = sd_alloc_ctl_entry(domain_num + 1);
 295         if (table == NULL)
 296                 return NULL;
 297
 298         i = 0;
 299         for_each_domain(cpu, sd) {
 300                 snprintf(buf, 32, "domain%d", i);
 301                 entry->procname = kstrdup(buf, GFP_KERNEL);
 302                 entry->mode = 0555;
 303                 entry->child = sd_alloc_ctl_domain_table(sd);
 304                 entry++;
 305                 i++;
 306         }
 307         return table;
 308 }
 309
 310 static cpumask_var_t            sd_sysctl_cpus;
 311 static struct ctl_table_header  *sd_sysctl_header;
 312
 313 void register_sched_domain_sysctl(void)
 314 {
 315         static struct ctl_table *cpu_entries;
 316         static struct ctl_table **cpu_idx;
 317         char buf[32];
 318         int i;
 319
 320         if (!cpu_entries) {
 321                 cpu_entries = sd_alloc_ctl_entry(num_possible_cpus() + 1);
 322                 if (!cpu_entries)
 323                         return;
 324
 325                 WARN_ON(sd_ctl_dir[0].child);
 326                 sd_ctl_dir[0].child = cpu_entries;
 327         }
 328
 329         if (!cpu_idx) {
 330                 struct ctl_table *e = cpu_entries;
 331
 332                 cpu_idx = kcalloc(nr_cpu_ids, sizeof(struct ctl_table*), GFP_KERNEL);
 333                 if (!cpu_idx)
 334                         return;
 335
 336                 /* deal with sparse possible map */
 337                 for_each_possible_cpu(i) {
 338                         cpu_idx[i] = e;
 339                         e++;
 340                 }
 341         }
 342
 343         if (!cpumask_available(sd_sysctl_cpus)) {
 344                 if (!alloc_cpumask_var(&sd_sysctl_cpus, GFP_KERNEL))
 345                         return;
 346
 347                 /* init to possible to not have holes in @cpu_entries */
 348                 cpumask_copy(sd_sysctl_cpus, cpu_possible_mask);
 349         }
 350
 351         for_each_cpu(i, sd_sysctl_cpus) {
 352                 struct ctl_table *e = cpu_idx[i];
 353
 354                 if (e->child)
 355                         sd_free_ctl_entry(&e->child);
 356
 357                 if (!e->procname) {
 358                         snprintf(buf, 32, "cpu%d", i);
 359                         e->procname = kstrdup(buf, GFP_KERNEL);
 360                 }
 361                 e->mode = 0555;
 362                 e->child = sd_alloc_ctl_cpu_table(i);
 363
 364                 __cpumask_clear_cpu(i, sd_sysctl_cpus);
 365         }
 366
 367         WARN_ON(sd_sysctl_header);
 368         sd_sysctl_header = register_sysctl_table(sd_ctl_root);
 369 }
 370
 371 void dirty_sched_domain_sysctl(int cpu)
 372 {
 373         if (cpumask_available(sd_sysctl_cpus))
 374                 __cpumask_set_cpu(cpu, sd_sysctl_cpus);
 375 }
 376
 377 /* may be called multiple times per register */
 378 void unregister_sched_domain_sysctl(void)
 379 {
 380         unregister_sysctl_table(sd_sysctl_header);
 381         sd_sysctl_header = NULL;
 382 }
 383 #endif /* CONFIG_SYSCTL */
 384 #endif /* CONFIG_SMP */
 385
 386 #ifdef CONFIG_FAIR_GROUP_SCHED
 387 static void print_cfs_group_stats(struct seq_file *m, int cpu, struct task_group *tg)
 388 {
 389         struct sched_entity *se = tg->se[cpu];
 390
 391 #define P(F)            SEQ_printf(m, "  .%-30s: %lld\n",       #F, (long long)F)
 392 #define P_SCHEDSTAT(F)  SEQ_printf(m, "  .%-30s: %lld\n",       #F, (long long)schedstat_val(F))
 393 #define PN(F)           SEQ_printf(m, "  .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F))
 394 #define PN_SCHEDSTAT(F) SEQ_printf(m, "  .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)schedstat_val(F)))
 395
 396         if (!se)
 397                 return;
 398
 399         PN(se->exec_start);
 400         PN(se->vruntime);
 401         PN(se->sum_exec_runtime);
 402
 403         if (schedstat_enabled()) {
 404                 PN_SCHEDSTAT(se->statistics.wait_start);
 405                 PN_SCHEDSTAT(se->statistics.sleep_start);
 406                 PN_SCHEDSTAT(se->statistics.block_start);
 407                 PN_SCHEDSTAT(se->statistics.sleep_max);
 408                 PN_SCHEDSTAT(se->statistics.block_max);
 409                 PN_SCHEDSTAT(se->statistics.exec_max);
 410                 PN_SCHEDSTAT(se->statistics.slice_max);
 411                 PN_SCHEDSTAT(se->statistics.wait_max);
 412                 PN_SCHEDSTAT(se->statistics.wait_sum);
 413                 P_SCHEDSTAT(se->statistics.wait_count);
 414         }
 415
 416         P(se->load.weight);
 417         P(se->runnable_weight);
 418 #ifdef CONFIG_SMP
 419         P(se->avg.load_avg);
 420         P(se->avg.util_avg);
 421         P(se->avg.runnable_load_avg);
 422 #endif
 423
 424 #undef PN_SCHEDSTAT
 425 #undef PN
 426 #undef P_SCHEDSTAT
 427 #undef P
 428 }
 429 #endif
 430
 431 #ifdef CONFIG_CGROUP_SCHED
 432 static char group_path[PATH_MAX];
 433
 434 static char *task_group_path(struct task_group *tg)
 435 {
 436         if (autogroup_path(tg, group_path, PATH_MAX))
 437                 return group_path;
 438
 439         cgroup_path(tg->css.cgroup, group_path, PATH_MAX);
 440
 441         return group_path;
 442 }
 443 #endif
 444
 445 static void
 446 print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
 447 {
 448         if (rq->curr == p)
 449                 SEQ_printf(m, ">R");
 450         else
 451                 SEQ_printf(m, " %c", task_state_to_char(p));
 452
 453         SEQ_printf(m, "%15s %5d %9Ld.%06ld %9Ld %5d ",
 454                 p->comm, task_pid_nr(p),
 455                 SPLIT_NS(p->se.vruntime),
 456                 (long long)(p->nvcsw + p->nivcsw),
 457                 p->prio);
 458
 459         SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
 460                 SPLIT_NS(schedstat_val_or_zero(p->se.statistics.wait_sum)),
 461                 SPLIT_NS(p->se.sum_exec_runtime),
 462                 SPLIT_NS(schedstat_val_or_zero(p->se.statistics.sum_sleep_runtime)));
 463
 464 #ifdef CONFIG_NUMA_BALANCING
 465         SEQ_printf(m, " %d %d", task_node(p), task_numa_group_id(p));
 466 #endif
 467 #ifdef CONFIG_CGROUP_SCHED
 468         SEQ_printf(m, " %s", task_group_path(task_group(p)));
 469 #endif
 470
 471         SEQ_printf(m, "\n");
 472 }
 473
 474 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
 475 {
 476         struct task_struct *g, *p;
 477
 478         SEQ_printf(m, "\n");
 479         SEQ_printf(m, "runnable tasks:\n");
 480         SEQ_printf(m, " S           task   PID         tree-key  switches  prio"
 481                    "     wait-time             sum-exec        sum-sleep\n");
 482         SEQ_printf(m, "-------------------------------------------------------"
 483                    "----------------------------------------------------\n");
 484
 485         rcu_read_lock();
 486         for_each_process_thread(g, p) {
 487                 if (task_cpu(p) != rq_cpu)
 488                         continue;
 489
 490                 print_task(m, rq, p);
 491         }
 492         rcu_read_unlock();
 493 }
 494
 495 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
 496 {
 497         s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
 498                 spread, rq0_min_vruntime, spread0;
 499         struct rq *rq = cpu_rq(cpu);
 500         struct sched_entity *last;
 501         unsigned long flags;
 502
 503 #ifdef CONFIG_FAIR_GROUP_SCHED
 504         SEQ_printf(m, "\n");
 505         SEQ_printf(m, "cfs_rq[%d]:%s\n", cpu, task_group_path(cfs_rq->tg));
 506 #else
 507         SEQ_printf(m, "\n");
 508         SEQ_printf(m, "cfs_rq[%d]:\n", cpu);
 509 #endif
 510         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "exec_clock",
 511                         SPLIT_NS(cfs_rq->exec_clock));
 512
 513         raw_spin_lock_irqsave(&rq->lock, flags);
 514         if (rb_first_cached(&cfs_rq->tasks_timeline))
 515                 MIN_vruntime = (__pick_first_entity(cfs_rq))->vruntime;
 516         last = __pick_last_entity(cfs_rq);
 517         if (last)
 518                 max_vruntime = last->vruntime;
 519         min_vruntime = cfs_rq->min_vruntime;
 520         rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
 521         raw_spin_unlock_irqrestore(&rq->lock, flags);
 522         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "MIN_vruntime",
 523                         SPLIT_NS(MIN_vruntime));
 524         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "min_vruntime",
 525                         SPLIT_NS(min_vruntime));
 526         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "max_vruntime",
 527                         SPLIT_NS(max_vruntime));
 528         spread = max_vruntime - MIN_vruntime;
 529         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread",
 530                         SPLIT_NS(spread));
 531         spread0 = min_vruntime - rq0_min_vruntime;
 532         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread0",
 533                         SPLIT_NS(spread0));
 534         SEQ_printf(m, "  .%-30s: %d\n", "nr_spread_over",
 535                         cfs_rq->nr_spread_over);
 536         SEQ_printf(m, "  .%-30s: %d\n", "nr_running", cfs_rq->nr_running);
 537         SEQ_printf(m, "  .%-30s: %ld\n", "load", cfs_rq->load.weight);
 538 #ifdef CONFIG_SMP
 539         SEQ_printf(m, "  .%-30s: %ld\n", "runnable_weight", cfs_rq->runnable_weight);
 540         SEQ_printf(m, "  .%-30s: %lu\n", "load_avg",
 541                         cfs_rq->avg.load_avg);
 542         SEQ_printf(m, "  .%-30s: %lu\n", "runnable_load_avg",
 543                         cfs_rq->avg.runnable_load_avg);
 544         SEQ_printf(m, "  .%-30s: %lu\n", "util_avg",
 545                         cfs_rq->avg.util_avg);
 546         SEQ_printf(m, "  .%-30s: %ld\n", "removed.load_avg",
 547                         cfs_rq->removed.load_avg);
 548         SEQ_printf(m, "  .%-30s: %ld\n", "removed.util_avg",
 549                         cfs_rq->removed.util_avg);
 550         SEQ_printf(m, "  .%-30s: %ld\n", "removed.runnable_sum",
 551                         cfs_rq->removed.runnable_sum);
 552 #ifdef CONFIG_FAIR_GROUP_SCHED
 553         SEQ_printf(m, "  .%-30s: %lu\n", "tg_load_avg_contrib",
 554                         cfs_rq->tg_load_avg_contrib);
 555         SEQ_printf(m, "  .%-30s: %ld\n", "tg_load_avg",
 556                         atomic_long_read(&cfs_rq->tg->load_avg));
 557 #endif
 558 #endif
 559 #ifdef CONFIG_CFS_BANDWIDTH
 560         SEQ_printf(m, "  .%-30s: %d\n", "throttled",
 561                         cfs_rq->throttled);
 562         SEQ_printf(m, "  .%-30s: %d\n", "throttle_count",
 563                         cfs_rq->throttle_count);
 564 #endif
 565
 566 #ifdef CONFIG_FAIR_GROUP_SCHED
 567         print_cfs_group_stats(m, cpu, cfs_rq->tg);
 568 #endif
 569 }
 570
 571 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
 572 {
 573 #ifdef CONFIG_RT_GROUP_SCHED
 574         SEQ_printf(m, "\n");
 575         SEQ_printf(m, "rt_rq[%d]:%s\n", cpu, task_group_path(rt_rq->tg));
 576 #else
 577         SEQ_printf(m, "\n");
 578         SEQ_printf(m, "rt_rq[%d]:\n", cpu);
 579 #endif
 580
 581 #define P(x) \
 582         SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
 583 #define PU(x) \
 584         SEQ_printf(m, "  .%-30s: %lu\n", #x, (unsigned long)(rt_rq->x))
 585 #define PN(x) \
 586         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
 587
 588         PU(rt_nr_running);
 589 #ifdef CONFIG_SMP
 590         PU(rt_nr_migratory);
 591 #endif
 592         P(rt_throttled);
 593         PN(rt_time);
 594         PN(rt_runtime);
 595
 596 #undef PN
 597 #undef PU
 598 #undef P
 599 }
 600
 601 void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq)
 602 {
 603         struct dl_bw *dl_bw;
 604
 605         SEQ_printf(m, "\n");
 606         SEQ_printf(m, "dl_rq[%d]:\n", cpu);
 607
 608 #define PU(x) \
 609         SEQ_printf(m, "  .%-30s: %lu\n", #x, (unsigned long)(dl_rq->x))
 610
 611         PU(dl_nr_running);
 612 #ifdef CONFIG_SMP
 613         PU(dl_nr_migratory);
 614         dl_bw = &cpu_rq(cpu)->rd->dl_bw;
 615 #else
 616         dl_bw = &dl_rq->dl_bw;
 617 #endif
 618         SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->bw", dl_bw->bw);
 619         SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->total_bw", dl_bw->total_bw);
 620
 621 #undef PU
 622 }
 623
 624 extern __read_mostly int sched_clock_running;
 625
 626 static void print_cpu(struct seq_file *m, int cpu)
 627 {
 628         struct rq *rq = cpu_rq(cpu);
 629         unsigned long flags;
 630
 631 #ifdef CONFIG_X86
 632         {
 633                 unsigned int freq = cpu_khz ? : 1;
 634
 635                 SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
 636                            cpu, freq / 1000, (freq % 1000));
 637         }
 638 #else
 639         SEQ_printf(m, "cpu#%d\n", cpu);
 640 #endif
 641
 642 #define P(x)                                                            \
 643 do {                                                                    \
 644         if (sizeof(rq->x) == 4)                                         \
 645                 SEQ_printf(m, "  .%-30s: %ld\n", #x, (long)(rq->x));    \
 646         else                                                            \
 647                 SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rq->x));\
 648 } while (0)
 649
 650 #define PN(x) \
 651         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
 652
 653         P(nr_running);
 654         SEQ_printf(m, "  .%-30s: %lu\n", "load",
 655                    rq->load.weight);
 656         P(nr_switches);
 657         P(nr_load_updates);
 658         P(nr_uninterruptible);
 659         PN(next_balance);
 660         SEQ_printf(m, "  .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr)));
 661         PN(clock);
 662         PN(clock_task);
 663         P(cpu_load[0]);
 664         P(cpu_load[1]);
 665         P(cpu_load[2]);
 666         P(cpu_load[3]);
 667         P(cpu_load[4]);
 668 #undef P
 669 #undef PN
 670
 671 #ifdef CONFIG_SMP
 672 #define P64(n) SEQ_printf(m, "  .%-30s: %Ld\n", #n, rq->n);
 673         P64(avg_idle);
 674         P64(max_idle_balance_cost);
 675 #undef P64
 676 #endif
 677
 678 #define P(n) SEQ_printf(m, "  .%-30s: %d\n", #n, schedstat_val(rq->n));
 679         if (schedstat_enabled()) {
 680                 P(yld_count);
 681                 P(sched_count);
 682                 P(sched_goidle);
 683                 P(ttwu_count);
 684                 P(ttwu_local);
 685         }
 686 #undef P
 687
 688         spin_lock_irqsave(&sched_debug_lock, flags);
 689         print_cfs_stats(m, cpu);
 690         print_rt_stats(m, cpu);
 691         print_dl_stats(m, cpu);
 692
 693         print_rq(m, rq, cpu);
 694         spin_unlock_irqrestore(&sched_debug_lock, flags);
 695         SEQ_printf(m, "\n");
 696 }
 697
 698 static const char *sched_tunable_scaling_names[] = {
 699         "none",
 700         "logaritmic",
 701         "linear"
 702 };
 703
 704 static void sched_debug_header(struct seq_file *m)
 705 {
 706         u64 ktime, sched_clk, cpu_clk;
 707         unsigned long flags;
 708
 709         local_irq_save(flags);
 710         ktime = ktime_to_ns(ktime_get());
 711         sched_clk = sched_clock();
 712         cpu_clk = local_clock();
 713         local_irq_restore(flags);
 714
 715         SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n",
 716                 init_utsname()->release,
 717                 (int)strcspn(init_utsname()->version, " "),
 718                 init_utsname()->version);
 719
 720 #define P(x) \
 721         SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
 722 #define PN(x) \
 723         SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
 724         PN(ktime);
 725         PN(sched_clk);
 726         PN(cpu_clk);
 727         P(jiffies);
 728 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
 729         P(sched_clock_stable());
 730 #endif
 731 #undef PN
 732 #undef P
 733
 734         SEQ_printf(m, "\n");
 735         SEQ_printf(m, "sysctl_sched\n");
 736
 737 #define P(x) \
 738         SEQ_printf(m, "  .%-40s: %Ld\n", #x, (long long)(x))
 739 #define PN(x) \
 740         SEQ_printf(m, "  .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
 741         PN(sysctl_sched_latency);
 742         PN(sysctl_sched_min_granularity);
 743         PN(sysctl_sched_wakeup_granularity);
 744         P(sysctl_sched_child_runs_first);
 745         P(sysctl_sched_features);
 746 #undef PN
 747 #undef P
 748
 749         SEQ_printf(m, "  .%-40s: %d (%s)\n",
 750                 "sysctl_sched_tunable_scaling",
 751                 sysctl_sched_tunable_scaling,
 752                 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
 753         SEQ_printf(m, "\n");
 754 }
 755
 756 static int sched_debug_show(struct seq_file *m, void *v)
 757 {
 758         int cpu = (unsigned long)(v - 2);
 759
 760         if (cpu != -1)
 761                 print_cpu(m, cpu);
 762         else
 763                 sched_debug_header(m);
 764
 765         return 0;
 766 }
 767
 768 void sysrq_sched_debug_show(void)
 769 {
 770         int cpu;
 771
 772         sched_debug_header(NULL);
 773         for_each_online_cpu(cpu)
 774                 print_cpu(NULL, cpu);
 775
 776 }
 777
 778 /*
 779  * This itererator needs some explanation.
 780  * It returns 1 for the header position.
 781  * This means 2 is CPU 0.
 782  * In a hotplugged system some CPUs, including CPU 0, may be missing so we have
 783  * to use cpumask_* to iterate over the CPUs.
 784  */
 785 static void *sched_debug_start(struct seq_file *file, loff_t *offset)
 786 {
 787         unsigned long n = *offset;
 788
 789         if (n == 0)
 790                 return (void *) 1;
 791
 792         n--;
 793
 794         if (n > 0)
 795                 n = cpumask_next(n - 1, cpu_online_mask);
 796         else
 797                 n = cpumask_first(cpu_online_mask);
 798
 799         *offset = n + 1;
 800
 801         if (n < nr_cpu_ids)
 802                 return (void *)(unsigned long)(n + 2);
 803
 804         return NULL;
 805 }
 806
 807 static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
 808 {
 809         (*offset)++;
 810         return sched_debug_start(file, offset);
 811 }
 812
 813 static void sched_debug_stop(struct seq_file *file, void *data)
 814 {
 815 }
 816
 817 static const struct seq_operations sched_debug_sops = {
 818         .start          = sched_debug_start,
 819         .next           = sched_debug_next,
 820         .stop           = sched_debug_stop,
 821         .show           = sched_debug_show,
 822 };
 823
 824 static int sched_debug_release(struct inode *inode, struct file *file)
 825 {
 826         seq_release(inode, file);
 827
 828         return 0;
 829 }
 830
 831 static int sched_debug_open(struct inode *inode, struct file *filp)
 832 {
 833         int ret = 0;
 834
 835         ret = seq_open(filp, &sched_debug_sops);
 836
 837         return ret;
 838 }
 839
 840 static const struct file_operations sched_debug_fops = {
 841         .open           = sched_debug_open,
 842         .read           = seq_read,
 843         .llseek         = seq_lseek,
 844         .release        = sched_debug_release,
 845 };
 846
 847 static int __init init_sched_debug_procfs(void)
 848 {
 849         struct proc_dir_entry *pe;
 850
 851         pe = proc_create("sched_debug", 0444, NULL, &sched_debug_fops);
 852         if (!pe)
 853                 return -ENOMEM;
 854         return 0;
 855 }
 856
 857 __initcall(init_sched_debug_procfs);
 858
 859 #define __P(F)  SEQ_printf(m, "%-45s:%21Ld\n",       #F, (long long)F)
 860 #define   P(F)  SEQ_printf(m, "%-45s:%21Ld\n",       #F, (long long)p->F)
 861 #define __PN(F) SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
 862 #define   PN(F) SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
 863
 864
 865 #ifdef CONFIG_NUMA_BALANCING
 866 void print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
 867                 unsigned long tpf, unsigned long gsf, unsigned long gpf)
 868 {
 869         SEQ_printf(m, "numa_faults node=%d ", node);
 870         SEQ_printf(m, "task_private=%lu task_shared=%lu ", tsf, tpf);
 871         SEQ_printf(m, "group_private=%lu group_shared=%lu\n", gsf, gpf);
 872 }
 873 #endif
 874
 875
 876 static void sched_show_numa(struct task_struct *p, struct seq_file *m)
 877 {
 878 #ifdef CONFIG_NUMA_BALANCING
 879         struct mempolicy *pol;
 880
 881         if (p->mm)
 882                 P(mm->numa_scan_seq);
 883
 884         task_lock(p);
 885         pol = p->mempolicy;
 886         if (pol && !(pol->flags & MPOL_F_MORON))
 887                 pol = NULL;
 888         mpol_get(pol);
 889         task_unlock(p);
 890
 891         P(numa_pages_migrated);
 892         P(numa_preferred_nid);
 893         P(total_numa_faults);
 894         SEQ_printf(m, "current_node=%d, numa_group_id=%d\n",
 895                         task_node(p), task_numa_group_id(p));
 896         show_numa_stats(p, m);
 897         mpol_put(pol);
 898 #endif
 899 }
 900
 901 void proc_sched_show_task(struct task_struct *p, struct pid_namespace *ns,
 902                                                   struct seq_file *m)
 903 {
 904         unsigned long nr_switches;
 905
 906         SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr_ns(p, ns),
 907                                                 get_nr_threads(p));
 908         SEQ_printf(m,
 909                 "---------------------------------------------------------"
 910                 "----------\n");
 911 #define __P(F) \
 912         SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)F)
 913 #define P(F) \
 914         SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)p->F)
 915 #define P_SCHEDSTAT(F) \
 916         SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)schedstat_val(p->F))
 917 #define __PN(F) \
 918         SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
 919 #define PN(F) \
 920         SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
 921 #define PN_SCHEDSTAT(F) \
 922         SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)schedstat_val(p->F)))
 923
 924         PN(se.exec_start);
 925         PN(se.vruntime);
 926         PN(se.sum_exec_runtime);
 927
 928         nr_switches = p->nvcsw + p->nivcsw;
 929
 930         P(se.nr_migrations);
 931
 932         if (schedstat_enabled()) {
 933                 u64 avg_atom, avg_per_cpu;
 934
 935                 PN_SCHEDSTAT(se.statistics.sum_sleep_runtime);
 936                 PN_SCHEDSTAT(se.statistics.wait_start);
 937                 PN_SCHEDSTAT(se.statistics.sleep_start);
 938                 PN_SCHEDSTAT(se.statistics.block_start);
 939                 PN_SCHEDSTAT(se.statistics.sleep_max);
 940                 PN_SCHEDSTAT(se.statistics.block_max);
 941                 PN_SCHEDSTAT(se.statistics.exec_max);
 942                 PN_SCHEDSTAT(se.statistics.slice_max);
 943                 PN_SCHEDSTAT(se.statistics.wait_max);
 944                 PN_SCHEDSTAT(se.statistics.wait_sum);
 945                 P_SCHEDSTAT(se.statistics.wait_count);
 946                 PN_SCHEDSTAT(se.statistics.iowait_sum);
 947                 P_SCHEDSTAT(se.statistics.iowait_count);
 948                 P_SCHEDSTAT(se.statistics.nr_migrations_cold);
 949                 P_SCHEDSTAT(se.statistics.nr_failed_migrations_affine);
 950                 P_SCHEDSTAT(se.statistics.nr_failed_migrations_running);
 951                 P_SCHEDSTAT(se.statistics.nr_failed_migrations_hot);
 952                 P_SCHEDSTAT(se.statistics.nr_forced_migrations);
 953                 P_SCHEDSTAT(se.statistics.nr_wakeups);
 954                 P_SCHEDSTAT(se.statistics.nr_wakeups_sync);
 955                 P_SCHEDSTAT(se.statistics.nr_wakeups_migrate);
 956                 P_SCHEDSTAT(se.statistics.nr_wakeups_local);
 957                 P_SCHEDSTAT(se.statistics.nr_wakeups_remote);
 958                 P_SCHEDSTAT(se.statistics.nr_wakeups_affine);
 959                 P_SCHEDSTAT(se.statistics.nr_wakeups_affine_attempts);
 960                 P_SCHEDSTAT(se.statistics.nr_wakeups_passive);
 961                 P_SCHEDSTAT(se.statistics.nr_wakeups_idle);
 962
 963                 avg_atom = p->se.sum_exec_runtime;
 964                 if (nr_switches)
 965                         avg_atom = div64_ul(avg_atom, nr_switches);
 966                 else
 967                         avg_atom = -1LL;
 968
 969                 avg_per_cpu = p->se.sum_exec_runtime;
 970                 if (p->se.nr_migrations) {
 971                         avg_per_cpu = div64_u64(avg_per_cpu,
 972                                                 p->se.nr_migrations);
 973                 } else {
 974                         avg_per_cpu = -1LL;
 975                 }
 976
 977                 __PN(avg_atom);
 978                 __PN(avg_per_cpu);
 979         }
 980
 981         __P(nr_switches);
 982         SEQ_printf(m, "%-45s:%21Ld\n",
 983                    "nr_voluntary_switches", (long long)p->nvcsw);
 984         SEQ_printf(m, "%-45s:%21Ld\n",
 985                    "nr_involuntary_switches", (long long)p->nivcsw);
 986
 987         P(se.load.weight);
 988         P(se.runnable_weight);
 989 #ifdef CONFIG_SMP
 990         P(se.avg.load_sum);
 991         P(se.avg.runnable_load_sum);
 992         P(se.avg.util_sum);
 993         P(se.avg.load_avg);
 994         P(se.avg.runnable_load_avg);
 995         P(se.avg.util_avg);
 996         P(se.avg.last_update_time);
 997 #endif
 998         P(policy);
 999         P(prio);
1000         if (p->policy == SCHED_DEADLINE) {
1001                 P(dl.runtime);
1002                 P(dl.deadline);
1003         }
1004 #undef PN_SCHEDSTAT
1005 #undef PN
1006 #undef __PN
1007 #undef P_SCHEDSTAT
1008 #undef P
1009 #undef __P
1010
1011         {
1012                 unsigned int this_cpu = raw_smp_processor_id();
1013                 u64 t0, t1;
1014
1015                 t0 = cpu_clock(this_cpu);
1016                 t1 = cpu_clock(this_cpu);
1017                 SEQ_printf(m, "%-45s:%21Ld\n",
1018                            "clock-delta", (long long)(t1-t0));
1019         }
1020
1021         sched_show_numa(p, m);
1022 }
1023
1024 void proc_sched_set_task(struct task_struct *p)
1025 {
1026 #ifdef CONFIG_SCHEDSTATS
1027         memset(&p->se.statistics, 0, sizeof(p->se.statistics));
1028 #endif
1029 }