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 P_SCHEDSTAT(F) \
 373         SEQ_printf(m, "  .%-30s: %lld\n", #F, (long long)schedstat_val(F))
 374 #define PN(F) \
 375         SEQ_printf(m, "  .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F))
 376 #define PN_SCHEDSTAT(F) \
 377         SEQ_printf(m, "  .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)schedstat_val(F)))
 378
 379         if (!se)
 380                 return;
 381
 382         PN(se->exec_start);
 383         PN(se->vruntime);
 384         PN(se->sum_exec_runtime);
 385         if (schedstat_enabled()) {
 386                 PN_SCHEDSTAT(se->statistics.wait_start);
 387                 PN_SCHEDSTAT(se->statistics.sleep_start);
 388                 PN_SCHEDSTAT(se->statistics.block_start);
 389                 PN_SCHEDSTAT(se->statistics.sleep_max);
 390                 PN_SCHEDSTAT(se->statistics.block_max);
 391                 PN_SCHEDSTAT(se->statistics.exec_max);
 392                 PN_SCHEDSTAT(se->statistics.slice_max);
 393                 PN_SCHEDSTAT(se->statistics.wait_max);
 394                 PN_SCHEDSTAT(se->statistics.wait_sum);
 395                 P_SCHEDSTAT(se->statistics.wait_count);
 396         }
 397         P(se->load.weight);
 398 #ifdef CONFIG_SMP
 399         P(se->avg.load_avg);
 400         P(se->avg.util_avg);
 401 #endif
 402
 403 #undef PN_SCHEDSTAT
 404 #undef PN
 405 #undef P_SCHEDSTAT
 406 #undef P
 407 }
 408 #endif
 409
 410 #ifdef CONFIG_CGROUP_SCHED
 411 static char group_path[PATH_MAX];
 412
 413 static char *task_group_path(struct task_group *tg)
 414 {
 415         if (autogroup_path(tg, group_path, PATH_MAX))
 416                 return group_path;
 417
 418         cgroup_path(tg->css.cgroup, group_path, PATH_MAX);
 419         return group_path;
 420 }
 421 #endif
 422
 423 static void
 424 print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
 425 {
 426         if (rq->curr == p)
 427                 SEQ_printf(m, "R");
 428         else
 429                 SEQ_printf(m, " ");
 430
 431         SEQ_printf(m, "%15s %5d %9Ld.%06ld %9Ld %5d ",
 432                 p->comm, task_pid_nr(p),
 433                 SPLIT_NS(p->se.vruntime),
 434                 (long long)(p->nvcsw + p->nivcsw),
 435                 p->prio);
 436
 437         SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
 438                 SPLIT_NS(schedstat_val_or_zero(p->se.statistics.wait_sum)),
 439                 SPLIT_NS(p->se.sum_exec_runtime),
 440                 SPLIT_NS(schedstat_val_or_zero(p->se.statistics.sum_sleep_runtime)));
 441
 442 #ifdef CONFIG_NUMA_BALANCING
 443         SEQ_printf(m, " %d %d", task_node(p), task_numa_group_id(p));
 444 #endif
 445 #ifdef CONFIG_CGROUP_SCHED
 446         SEQ_printf(m, " %s", task_group_path(task_group(p)));
 447 #endif
 448
 449         SEQ_printf(m, "\n");
 450 }
 451
 452 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
 453 {
 454         struct task_struct *g, *p;
 455
 456         SEQ_printf(m,
 457         "\nrunnable tasks:\n"
 458         "            task   PID         tree-key  switches  prio"
 459         "     wait-time             sum-exec        sum-sleep\n"
 460         "------------------------------------------------------"
 461         "----------------------------------------------------\n");
 462
 463         rcu_read_lock();
 464         for_each_process_thread(g, p) {
 465                 if (task_cpu(p) != rq_cpu)
 466                         continue;
 467
 468                 print_task(m, rq, p);
 469         }
 470         rcu_read_unlock();
 471 }
 472
 473 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
 474 {
 475         s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
 476                 spread, rq0_min_vruntime, spread0;
 477         struct rq *rq = cpu_rq(cpu);
 478         struct sched_entity *last;
 479         unsigned long flags;
 480
 481 #ifdef CONFIG_FAIR_GROUP_SCHED
 482         SEQ_printf(m, "\ncfs_rq[%d]:%s\n", cpu, task_group_path(cfs_rq->tg));
 483 #else
 484         SEQ_printf(m, "\ncfs_rq[%d]:\n", cpu);
 485 #endif
 486         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "exec_clock",
 487                         SPLIT_NS(cfs_rq->exec_clock));
 488
 489         raw_spin_lock_irqsave(&rq->lock, flags);
 490         if (cfs_rq->rb_leftmost)
 491                 MIN_vruntime = (__pick_first_entity(cfs_rq))->vruntime;
 492         last = __pick_last_entity(cfs_rq);
 493         if (last)
 494                 max_vruntime = last->vruntime;
 495         min_vruntime = cfs_rq->min_vruntime;
 496         rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
 497         raw_spin_unlock_irqrestore(&rq->lock, flags);
 498         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "MIN_vruntime",
 499                         SPLIT_NS(MIN_vruntime));
 500         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "min_vruntime",
 501                         SPLIT_NS(min_vruntime));
 502         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "max_vruntime",
 503                         SPLIT_NS(max_vruntime));
 504         spread = max_vruntime - MIN_vruntime;
 505         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread",
 506                         SPLIT_NS(spread));
 507         spread0 = min_vruntime - rq0_min_vruntime;
 508         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread0",
 509                         SPLIT_NS(spread0));
 510         SEQ_printf(m, "  .%-30s: %d\n", "nr_spread_over",
 511                         cfs_rq->nr_spread_over);
 512         SEQ_printf(m, "  .%-30s: %d\n", "nr_running", cfs_rq->nr_running);
 513         SEQ_printf(m, "  .%-30s: %ld\n", "load", cfs_rq->load.weight);
 514 #ifdef CONFIG_SMP
 515         SEQ_printf(m, "  .%-30s: %lu\n", "load_avg",
 516                         cfs_rq->avg.load_avg);
 517         SEQ_printf(m, "  .%-30s: %lu\n", "runnable_load_avg",
 518                         cfs_rq->runnable_load_avg);
 519         SEQ_printf(m, "  .%-30s: %lu\n", "util_avg",
 520                         cfs_rq->avg.util_avg);
 521         SEQ_printf(m, "  .%-30s: %ld\n", "removed_load_avg",
 522                         atomic_long_read(&cfs_rq->removed_load_avg));
 523         SEQ_printf(m, "  .%-30s: %ld\n", "removed_util_avg",
 524                         atomic_long_read(&cfs_rq->removed_util_avg));
 525 #ifdef CONFIG_FAIR_GROUP_SCHED
 526         SEQ_printf(m, "  .%-30s: %lu\n", "tg_load_avg_contrib",
 527                         cfs_rq->tg_load_avg_contrib);
 528         SEQ_printf(m, "  .%-30s: %ld\n", "tg_load_avg",
 529                         atomic_long_read(&cfs_rq->tg->load_avg));
 530 #endif
 531 #endif
 532 #ifdef CONFIG_CFS_BANDWIDTH
 533         SEQ_printf(m, "  .%-30s: %d\n", "throttled",
 534                         cfs_rq->throttled);
 535         SEQ_printf(m, "  .%-30s: %d\n", "throttle_count",
 536                         cfs_rq->throttle_count);
 537 #endif
 538
 539 #ifdef CONFIG_FAIR_GROUP_SCHED
 540         print_cfs_group_stats(m, cpu, cfs_rq->tg);
 541 #endif
 542 }
 543
 544 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
 545 {
 546 #ifdef CONFIG_RT_GROUP_SCHED
 547         SEQ_printf(m, "\nrt_rq[%d]:%s\n", cpu, task_group_path(rt_rq->tg));
 548 #else
 549         SEQ_printf(m, "\nrt_rq[%d]:\n", cpu);
 550 #endif
 551
 552 #define P(x) \
 553         SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
 554 #define PN(x) \
 555         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
 556
 557         P(rt_nr_running);
 558         P(rt_throttled);
 559         PN(rt_time);
 560         PN(rt_runtime);
 561
 562 #undef PN
 563 #undef P
 564 }
 565
 566 void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq)
 567 {
 568         struct dl_bw *dl_bw;
 569
 570         SEQ_printf(m, "\ndl_rq[%d]:\n", cpu);
 571         SEQ_printf(m, "  .%-30s: %ld\n", "dl_nr_running", dl_rq->dl_nr_running);
 572 #ifdef CONFIG_SMP
 573         dl_bw = &cpu_rq(cpu)->rd->dl_bw;
 574 #else
 575         dl_bw = &dl_rq->dl_bw;
 576 #endif
 577         SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->bw", dl_bw->bw);
 578         SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->total_bw", dl_bw->total_bw);
 579 }
 580
 581 extern __read_mostly int sched_clock_running;
 582
 583 static void print_cpu(struct seq_file *m, int cpu)
 584 {
 585         struct rq *rq = cpu_rq(cpu);
 586         unsigned long flags;
 587
 588 #ifdef CONFIG_X86
 589         {
 590                 unsigned int freq = cpu_khz ? : 1;
 591
 592                 SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
 593                            cpu, freq / 1000, (freq % 1000));
 594         }
 595 #else
 596         SEQ_printf(m, "cpu#%d\n", cpu);
 597 #endif
 598
 599 #define P(x)                                                            \
 600 do {                                                                    \
 601         if (sizeof(rq->x) == 4)                                         \
 602                 SEQ_printf(m, "  .%-30s: %ld\n", #x, (long)(rq->x));    \
 603         else                                                            \
 604                 SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rq->x));\
 605 } while (0)
 606
 607 #define PN(x) \
 608         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
 609
 610         P(nr_running);
 611         SEQ_printf(m, "  .%-30s: %lu\n", "load",
 612                    rq->load.weight);
 613         P(nr_switches);
 614         P(nr_load_updates);
 615         P(nr_uninterruptible);
 616         PN(next_balance);
 617         SEQ_printf(m, "  .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr)));
 618         PN(clock);
 619         PN(clock_task);
 620         P(cpu_load[0]);
 621         P(cpu_load[1]);
 622         P(cpu_load[2]);
 623         P(cpu_load[3]);
 624         P(cpu_load[4]);
 625 #undef P
 626 #undef PN
 627
 628 #ifdef CONFIG_SMP
 629 #define P64(n) SEQ_printf(m, "  .%-30s: %Ld\n", #n, rq->n);
 630         P64(avg_idle);
 631         P64(max_idle_balance_cost);
 632 #undef P64
 633 #endif
 634
 635 #define P(n) SEQ_printf(m, "  .%-30s: %d\n", #n, schedstat_val(rq->n));
 636         if (schedstat_enabled()) {
 637                 P(yld_count);
 638                 P(sched_count);
 639                 P(sched_goidle);
 640                 P(ttwu_count);
 641                 P(ttwu_local);
 642         }
 643 #undef P
 644
 645         spin_lock_irqsave(&sched_debug_lock, flags);
 646         print_cfs_stats(m, cpu);
 647         print_rt_stats(m, cpu);
 648         print_dl_stats(m, cpu);
 649
 650         print_rq(m, rq, cpu);
 651         spin_unlock_irqrestore(&sched_debug_lock, flags);
 652         SEQ_printf(m, "\n");
 653 }
 654
 655 static const char *sched_tunable_scaling_names[] = {
 656         "none",
 657         "logaritmic",
 658         "linear"
 659 };
 660
 661 static void sched_debug_header(struct seq_file *m)
 662 {
 663         u64 ktime, sched_clk, cpu_clk;
 664         unsigned long flags;
 665
 666         local_irq_save(flags);
 667         ktime = ktime_to_ns(ktime_get());
 668         sched_clk = sched_clock();
 669         cpu_clk = local_clock();
 670         local_irq_restore(flags);
 671
 672         SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n",
 673                 init_utsname()->release,
 674                 (int)strcspn(init_utsname()->version, " "),
 675                 init_utsname()->version);
 676
 677 #define P(x) \
 678         SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
 679 #define PN(x) \
 680         SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
 681         PN(ktime);
 682         PN(sched_clk);
 683         PN(cpu_clk);
 684         P(jiffies);
 685 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
 686         P(sched_clock_stable());
 687 #endif
 688 #undef PN
 689 #undef P
 690
 691         SEQ_printf(m, "\n");
 692         SEQ_printf(m, "sysctl_sched\n");
 693
 694 #define P(x) \
 695         SEQ_printf(m, "  .%-40s: %Ld\n", #x, (long long)(x))
 696 #define PN(x) \
 697         SEQ_printf(m, "  .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
 698         PN(sysctl_sched_latency);
 699         PN(sysctl_sched_min_granularity);
 700         PN(sysctl_sched_wakeup_granularity);
 701         P(sysctl_sched_child_runs_first);
 702         P(sysctl_sched_features);
 703 #undef PN
 704 #undef P
 705
 706         SEQ_printf(m, "  .%-40s: %d (%s)\n",
 707                 "sysctl_sched_tunable_scaling",
 708                 sysctl_sched_tunable_scaling,
 709                 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
 710         SEQ_printf(m, "\n");
 711 }
 712
 713 static int sched_debug_show(struct seq_file *m, void *v)
 714 {
 715         int cpu = (unsigned long)(v - 2);
 716
 717         if (cpu != -1)
 718                 print_cpu(m, cpu);
 719         else
 720                 sched_debug_header(m);
 721
 722         return 0;
 723 }
 724
 725 void sysrq_sched_debug_show(void)
 726 {
 727         int cpu;
 728
 729         sched_debug_header(NULL);
 730         for_each_online_cpu(cpu)
 731                 print_cpu(NULL, cpu);
 732
 733 }
 734
 735 /*
 736  * This itererator needs some explanation.
 737  * It returns 1 for the header position.
 738  * This means 2 is cpu 0.
 739  * In a hotplugged system some cpus, including cpu 0, may be missing so we have
 740  * to use cpumask_* to iterate over the cpus.
 741  */
 742 static void *sched_debug_start(struct seq_file *file, loff_t *offset)
 743 {
 744         unsigned long n = *offset;
 745
 746         if (n == 0)
 747                 return (void *) 1;
 748
 749         n--;
 750
 751         if (n > 0)
 752                 n = cpumask_next(n - 1, cpu_online_mask);
 753         else
 754                 n = cpumask_first(cpu_online_mask);
 755
 756         *offset = n + 1;
 757
 758         if (n < nr_cpu_ids)
 759                 return (void *)(unsigned long)(n + 2);
 760         return NULL;
 761 }
 762
 763 static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
 764 {
 765         (*offset)++;
 766         return sched_debug_start(file, offset);
 767 }
 768
 769 static void sched_debug_stop(struct seq_file *file, void *data)
 770 {
 771 }
 772
 773 static const struct seq_operations sched_debug_sops = {
 774         .start = sched_debug_start,
 775         .next = sched_debug_next,
 776         .stop = sched_debug_stop,
 777         .show = sched_debug_show,
 778 };
 779
 780 static int sched_debug_release(struct inode *inode, struct file *file)
 781 {
 782         seq_release(inode, file);
 783
 784         return 0;
 785 }
 786
 787 static int sched_debug_open(struct inode *inode, struct file *filp)
 788 {
 789         int ret = 0;
 790
 791         ret = seq_open(filp, &sched_debug_sops);
 792
 793         return ret;
 794 }
 795
 796 static const struct file_operations sched_debug_fops = {
 797         .open           = sched_debug_open,
 798         .read           = seq_read,
 799         .llseek         = seq_lseek,
 800         .release        = sched_debug_release,
 801 };
 802
 803 static int __init init_sched_debug_procfs(void)
 804 {
 805         struct proc_dir_entry *pe;
 806
 807         pe = proc_create("sched_debug", 0444, NULL, &sched_debug_fops);
 808         if (!pe)
 809                 return -ENOMEM;
 810         return 0;
 811 }
 812
 813 __initcall(init_sched_debug_procfs);
 814
 815 #define __P(F) \
 816         SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)F)
 817 #define P(F) \
 818         SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)p->F)
 819 #define __PN(F) \
 820         SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
 821 #define PN(F) \
 822         SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
 823
 824
 825 #ifdef CONFIG_NUMA_BALANCING
 826 void print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
 827                 unsigned long tpf, unsigned long gsf, unsigned long gpf)
 828 {
 829         SEQ_printf(m, "numa_faults node=%d ", node);
 830         SEQ_printf(m, "task_private=%lu task_shared=%lu ", tsf, tpf);
 831         SEQ_printf(m, "group_private=%lu group_shared=%lu\n", gsf, gpf);
 832 }
 833 #endif
 834
 835
 836 static void sched_show_numa(struct task_struct *p, struct seq_file *m)
 837 {
 838 #ifdef CONFIG_NUMA_BALANCING
 839         struct mempolicy *pol;
 840
 841         if (p->mm)
 842                 P(mm->numa_scan_seq);
 843
 844         task_lock(p);
 845         pol = p->mempolicy;
 846         if (pol && !(pol->flags & MPOL_F_MORON))
 847                 pol = NULL;
 848         mpol_get(pol);
 849         task_unlock(p);
 850
 851         P(numa_pages_migrated);
 852         P(numa_preferred_nid);
 853         P(total_numa_faults);
 854         SEQ_printf(m, "current_node=%d, numa_group_id=%d\n",
 855                         task_node(p), task_numa_group_id(p));
 856         show_numa_stats(p, m);
 857         mpol_put(pol);
 858 #endif
 859 }
 860
 861 void proc_sched_show_task(struct task_struct *p, struct seq_file *m)
 862 {
 863         unsigned long nr_switches;
 864
 865         SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr(p),
 866                                                 get_nr_threads(p));
 867         SEQ_printf(m,
 868                 "---------------------------------------------------------"
 869                 "----------\n");
 870 #define __P(F) \
 871         SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)F)
 872 #define P(F) \
 873         SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)p->F)
 874 #define P_SCHEDSTAT(F) \
 875         SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)schedstat_val(p->F))
 876 #define __PN(F) \
 877         SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
 878 #define PN(F) \
 879         SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
 880 #define PN_SCHEDSTAT(F) \
 881         SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)schedstat_val(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         P(se.nr_migrations);
 890
 891         if (schedstat_enabled()) {
 892                 u64 avg_atom, avg_per_cpu;
 893
 894                 PN_SCHEDSTAT(se.statistics.sum_sleep_runtime);
 895                 PN_SCHEDSTAT(se.statistics.wait_start);
 896                 PN_SCHEDSTAT(se.statistics.sleep_start);
 897                 PN_SCHEDSTAT(se.statistics.block_start);
 898                 PN_SCHEDSTAT(se.statistics.sleep_max);
 899                 PN_SCHEDSTAT(se.statistics.block_max);
 900                 PN_SCHEDSTAT(se.statistics.exec_max);
 901                 PN_SCHEDSTAT(se.statistics.slice_max);
 902                 PN_SCHEDSTAT(se.statistics.wait_max);
 903                 PN_SCHEDSTAT(se.statistics.wait_sum);
 904                 P_SCHEDSTAT(se.statistics.wait_count);
 905                 PN_SCHEDSTAT(se.statistics.iowait_sum);
 906                 P_SCHEDSTAT(se.statistics.iowait_count);
 907                 P_SCHEDSTAT(se.statistics.nr_migrations_cold);
 908                 P_SCHEDSTAT(se.statistics.nr_failed_migrations_affine);
 909                 P_SCHEDSTAT(se.statistics.nr_failed_migrations_running);
 910                 P_SCHEDSTAT(se.statistics.nr_failed_migrations_hot);
 911                 P_SCHEDSTAT(se.statistics.nr_forced_migrations);
 912                 P_SCHEDSTAT(se.statistics.nr_wakeups);
 913                 P_SCHEDSTAT(se.statistics.nr_wakeups_sync);
 914                 P_SCHEDSTAT(se.statistics.nr_wakeups_migrate);
 915                 P_SCHEDSTAT(se.statistics.nr_wakeups_local);
 916                 P_SCHEDSTAT(se.statistics.nr_wakeups_remote);
 917                 P_SCHEDSTAT(se.statistics.nr_wakeups_affine);
 918                 P_SCHEDSTAT(se.statistics.nr_wakeups_affine_attempts);
 919                 P_SCHEDSTAT(se.statistics.nr_wakeups_passive);
 920                 P_SCHEDSTAT(se.statistics.nr_wakeups_idle);
 921
 922                 avg_atom = p->se.sum_exec_runtime;
 923                 if (nr_switches)
 924                         avg_atom = div64_ul(avg_atom, nr_switches);
 925                 else
 926                         avg_atom = -1LL;
 927
 928                 avg_per_cpu = p->se.sum_exec_runtime;
 929                 if (p->se.nr_migrations) {
 930                         avg_per_cpu = div64_u64(avg_per_cpu,
 931                                                 p->se.nr_migrations);
 932                 } else {
 933                         avg_per_cpu = -1LL;
 934                 }
 935
 936                 __PN(avg_atom);
 937                 __PN(avg_per_cpu);
 938         }
 939
 940         __P(nr_switches);
 941         SEQ_printf(m, "%-45s:%21Ld\n",
 942                    "nr_voluntary_switches", (long long)p->nvcsw);
 943         SEQ_printf(m, "%-45s:%21Ld\n",
 944                    "nr_involuntary_switches", (long long)p->nivcsw);
 945
 946         P(se.load.weight);
 947 #ifdef CONFIG_SMP
 948         P(se.avg.load_sum);
 949         P(se.avg.util_sum);
 950         P(se.avg.load_avg);
 951         P(se.avg.util_avg);
 952         P(se.avg.last_update_time);
 953 #endif
 954         P(policy);
 955         P(prio);
 956 #undef PN_SCHEDSTAT
 957 #undef PN
 958 #undef __PN
 959 #undef P_SCHEDSTAT
 960 #undef P
 961 #undef __P
 962
 963         {
 964                 unsigned int this_cpu = raw_smp_processor_id();
 965                 u64 t0, t1;
 966
 967                 t0 = cpu_clock(this_cpu);
 968                 t1 = cpu_clock(this_cpu);
 969                 SEQ_printf(m, "%-45s:%21Ld\n",
 970                            "clock-delta", (long long)(t1-t0));
 971         }
 972
 973         sched_show_numa(p, m);
 974 }
 975
 976 void proc_sched_set_task(struct task_struct *p)
 977 {
 978 #ifdef CONFIG_SCHEDSTATS
 979         memset(&p->se.statistics, 0, sizeof(p->se.statistics));
 980 #endif
 981 }