kernel/sched/debug.c

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