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/mm.h>
  15 #include <linux/sched/task.h>
  16 #include <linux/seq_file.h>
  17 #include <linux/kallsyms.h>
  18 #include <linux/utsname.h>
  19 #include <linux/mempolicy.h>
  20 #include <linux/debugfs.h>
  21
  22 #include "sched.h"
  23
  24 static DEFINE_SPINLOCK(sched_debug_lock);
  25
  26 /*
  27  * This allows printing both to /proc/sched_debug and
  28  * to the console
  29  */
  30 #define SEQ_printf(m, x...)                     \
  31  do {                                           \
  32         if (m)                                  \
  33                 seq_printf(m, x);               \
  34         else                                    \
  35                 printk(x);                      \
  36  } while (0)
  37
  38 /*
  39  * Ease the printing of nsec fields:
  40  */
  41 static long long nsec_high(unsigned long long nsec)
  42 {
  43         if ((long long)nsec < 0) {
  44                 nsec = -nsec;
  45                 do_div(nsec, 1000000);
  46                 return -nsec;
  47         }
  48         do_div(nsec, 1000000);
  49
  50         return nsec;
  51 }
  52
  53 static unsigned long nsec_low(unsigned long long nsec)
  54 {
  55         if ((long long)nsec < 0)
  56                 nsec = -nsec;
  57
  58         return do_div(nsec, 1000000);
  59 }
  60
  61 #define SPLIT_NS(x) nsec_high(x), nsec_low(x)
  62
  63 #define SCHED_FEAT(name, enabled)       \
  64         #name ,
  65
  66 static const char * const sched_feat_names[] = {
  67 #include "features.h"
  68 };
  69
  70 #undef SCHED_FEAT
  71
  72 static int sched_feat_show(struct seq_file *m, void *v)
  73 {
  74         int i;
  75
  76         for (i = 0; i < __SCHED_FEAT_NR; i++) {
  77                 if (!(sysctl_sched_features & (1UL << i)))
  78                         seq_puts(m, "NO_");
  79                 seq_printf(m, "%s ", sched_feat_names[i]);
  80         }
  81         seq_puts(m, "\n");
  82
  83         return 0;
  84 }
  85
  86 #ifdef HAVE_JUMP_LABEL
  87
  88 #define jump_label_key__true  STATIC_KEY_INIT_TRUE
  89 #define jump_label_key__false STATIC_KEY_INIT_FALSE
  90
  91 #define SCHED_FEAT(name, enabled)       \
  92         jump_label_key__##enabled ,
  93
  94 struct static_key sched_feat_keys[__SCHED_FEAT_NR] = {
  95 #include "features.h"
  96 };
  97
  98 #undef SCHED_FEAT
  99
 100 static void sched_feat_disable(int i)
 101 {
 102         static_key_disable(&sched_feat_keys[i]);
 103 }
 104
 105 static void sched_feat_enable(int i)
 106 {
 107         static_key_enable(&sched_feat_keys[i]);
 108 }
 109 #else
 110 static void sched_feat_disable(int i) { };
 111 static void sched_feat_enable(int i) { };
 112 #endif /* HAVE_JUMP_LABEL */
 113
 114 static int sched_feat_set(char *cmp)
 115 {
 116         int i;
 117         int neg = 0;
 118
 119         if (strncmp(cmp, "NO_", 3) == 0) {
 120                 neg = 1;
 121                 cmp += 3;
 122         }
 123
 124         for (i = 0; i < __SCHED_FEAT_NR; i++) {
 125                 if (strcmp(cmp, sched_feat_names[i]) == 0) {
 126                         if (neg) {
 127                                 sysctl_sched_features &= ~(1UL << i);
 128                                 sched_feat_disable(i);
 129                         } else {
 130                                 sysctl_sched_features |= (1UL << i);
 131                                 sched_feat_enable(i);
 132                         }
 133                         break;
 134                 }
 135         }
 136
 137         return i;
 138 }
 139
 140 static ssize_t
 141 sched_feat_write(struct file *filp, const char __user *ubuf,
 142                 size_t cnt, loff_t *ppos)
 143 {
 144         char buf[64];
 145         char *cmp;
 146         int i;
 147         struct inode *inode;
 148
 149         if (cnt > 63)
 150                 cnt = 63;
 151
 152         if (copy_from_user(&buf, ubuf, cnt))
 153                 return -EFAULT;
 154
 155         buf[cnt] = 0;
 156         cmp = strstrip(buf);
 157
 158         /* Ensure the static_key remains in a consistent state */
 159         inode = file_inode(filp);
 160         inode_lock(inode);
 161         i = sched_feat_set(cmp);
 162         inode_unlock(inode);
 163         if (i == __SCHED_FEAT_NR)
 164                 return -EINVAL;
 165
 166         *ppos += cnt;
 167
 168         return cnt;
 169 }
 170
 171 static int sched_feat_open(struct inode *inode, struct file *filp)
 172 {
 173         return single_open(filp, sched_feat_show, NULL);
 174 }
 175
 176 static const struct file_operations sched_feat_fops = {
 177         .open           = sched_feat_open,
 178         .write          = sched_feat_write,
 179         .read           = seq_read,
 180         .llseek         = seq_lseek,
 181         .release        = single_release,
 182 };
 183
 184 __read_mostly bool sched_debug_enabled;
 185
 186 static __init int sched_init_debug(void)
 187 {
 188         debugfs_create_file("sched_features", 0644, NULL, NULL,
 189                         &sched_feat_fops);
 190
 191         debugfs_create_bool("sched_debug", 0644, NULL,
 192                         &sched_debug_enabled);
 193
 194         return 0;
 195 }
 196 late_initcall(sched_init_debug);
 197
 198 #ifdef CONFIG_SMP
 199
 200 #ifdef CONFIG_SYSCTL
 201
 202 static struct ctl_table sd_ctl_dir[] = {
 203         {
 204                 .procname       = "sched_domain",
 205                 .mode           = 0555,
 206         },
 207         {}
 208 };
 209
 210 static struct ctl_table sd_ctl_root[] = {
 211         {
 212                 .procname       = "kernel",
 213                 .mode           = 0555,
 214                 .child          = sd_ctl_dir,
 215         },
 216         {}
 217 };
 218
 219 static struct ctl_table *sd_alloc_ctl_entry(int n)
 220 {
 221         struct ctl_table *entry =
 222                 kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL);
 223
 224         return entry;
 225 }
 226
 227 static void sd_free_ctl_entry(struct ctl_table **tablep)
 228 {
 229         struct ctl_table *entry;
 230
 231         /*
 232          * In the intermediate directories, both the child directory and
 233          * procname are dynamically allocated and could fail but the mode
 234          * will always be set. In the lowest directory the names are
 235          * static strings and all have proc handlers.
 236          */
 237         for (entry = *tablep; entry->mode; entry++) {
 238                 if (entry->child)
 239                         sd_free_ctl_entry(&entry->child);
 240                 if (entry->proc_handler == NULL)
 241                         kfree(entry->procname);
 242         }
 243
 244         kfree(*tablep);
 245         *tablep = NULL;
 246 }
 247
 248 static int min_load_idx = 0;
 249 static int max_load_idx = CPU_LOAD_IDX_MAX-1;
 250
 251 static void
 252 set_table_entry(struct ctl_table *entry,
 253                 const char *procname, void *data, int maxlen,
 254                 umode_t mode, proc_handler *proc_handler,
 255                 bool load_idx)
 256 {
 257         entry->procname = procname;
 258         entry->data = data;
 259         entry->maxlen = maxlen;
 260         entry->mode = mode;
 261         entry->proc_handler = proc_handler;
 262
 263         if (load_idx) {
 264                 entry->extra1 = &min_load_idx;
 265                 entry->extra2 = &max_load_idx;
 266         }
 267 }
 268
 269 static struct ctl_table *
 270 sd_alloc_ctl_domain_table(struct sched_domain *sd)
 271 {
 272         struct ctl_table *table = sd_alloc_ctl_entry(14);
 273
 274         if (table == NULL)
 275                 return NULL;
 276
 277         set_table_entry(&table[0], "min_interval", &sd->min_interval,
 278                 sizeof(long), 0644, proc_doulongvec_minmax, false);
 279         set_table_entry(&table[1], "max_interval", &sd->max_interval,
 280                 sizeof(long), 0644, proc_doulongvec_minmax, false);
 281         set_table_entry(&table[2], "busy_idx", &sd->busy_idx,
 282                 sizeof(int), 0644, proc_dointvec_minmax, true);
 283         set_table_entry(&table[3], "idle_idx", &sd->idle_idx,
 284                 sizeof(int), 0644, proc_dointvec_minmax, true);
 285         set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx,
 286                 sizeof(int), 0644, proc_dointvec_minmax, true);
 287         set_table_entry(&table[5], "wake_idx", &sd->wake_idx,
 288                 sizeof(int), 0644, proc_dointvec_minmax, true);
 289         set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx,
 290                 sizeof(int), 0644, proc_dointvec_minmax, true);
 291         set_table_entry(&table[7], "busy_factor", &sd->busy_factor,
 292                 sizeof(int), 0644, proc_dointvec_minmax, false);
 293         set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct,
 294                 sizeof(int), 0644, proc_dointvec_minmax, false);
 295         set_table_entry(&table[9], "cache_nice_tries",
 296                 &sd->cache_nice_tries,
 297                 sizeof(int), 0644, proc_dointvec_minmax, false);
 298         set_table_entry(&table[10], "flags", &sd->flags,
 299                 sizeof(int), 0644, proc_dointvec_minmax, false);
 300         set_table_entry(&table[11], "max_newidle_lb_cost",
 301                 &sd->max_newidle_lb_cost,
 302                 sizeof(long), 0644, proc_doulongvec_minmax, false);
 303         set_table_entry(&table[12], "name", sd->name,
 304                 CORENAME_MAX_SIZE, 0444, proc_dostring, false);
 305         /* &table[13] is terminator */
 306
 307         return table;
 308 }
 309
 310 static struct ctl_table *sd_alloc_ctl_cpu_table(int cpu)
 311 {
 312         struct ctl_table *entry, *table;
 313         struct sched_domain *sd;
 314         int domain_num = 0, i;
 315         char buf[32];
 316
 317         for_each_domain(cpu, sd)
 318                 domain_num++;
 319         entry = table = sd_alloc_ctl_entry(domain_num + 1);
 320         if (table == NULL)
 321                 return NULL;
 322
 323         i = 0;
 324         for_each_domain(cpu, sd) {
 325                 snprintf(buf, 32, "domain%d", i);
 326                 entry->procname = kstrdup(buf, GFP_KERNEL);
 327                 entry->mode = 0555;
 328                 entry->child = sd_alloc_ctl_domain_table(sd);
 329                 entry++;
 330                 i++;
 331         }
 332         return table;
 333 }
 334
 335 static cpumask_var_t sd_sysctl_cpus;
 336 static struct ctl_table_header *sd_sysctl_header;
 337
 338 void register_sched_domain_sysctl(void)
 339 {
 340         static struct ctl_table *cpu_entries;
 341         static struct ctl_table **cpu_idx;
 342         static bool init_done = false;
 343         char buf[32];
 344         int i;
 345
 346         if (!cpu_entries) {
 347                 cpu_entries = sd_alloc_ctl_entry(num_possible_cpus() + 1);
 348                 if (!cpu_entries)
 349                         return;
 350
 351                 WARN_ON(sd_ctl_dir[0].child);
 352                 sd_ctl_dir[0].child = cpu_entries;
 353         }
 354
 355         if (!cpu_idx) {
 356                 struct ctl_table *e = cpu_entries;
 357
 358                 cpu_idx = kcalloc(nr_cpu_ids, sizeof(struct ctl_table*), GFP_KERNEL);
 359                 if (!cpu_idx)
 360                         return;
 361
 362                 /* deal with sparse possible map */
 363                 for_each_possible_cpu(i) {
 364                         cpu_idx[i] = e;
 365                         e++;
 366                 }
 367         }
 368
 369         if (!cpumask_available(sd_sysctl_cpus)) {
 370                 if (!alloc_cpumask_var(&sd_sysctl_cpus, GFP_KERNEL))
 371                         return;
 372         }
 373
 374         if (!init_done) {
 375                 init_done = true;
 376                 /* init to possible to not have holes in @cpu_entries */
 377                 cpumask_copy(sd_sysctl_cpus, cpu_possible_mask);
 378         }
 379
 380         for_each_cpu(i, sd_sysctl_cpus) {
 381                 struct ctl_table *e = cpu_idx[i];
 382
 383                 if (e->child)
 384                         sd_free_ctl_entry(&e->child);
 385
 386                 if (!e->procname) {
 387                         snprintf(buf, 32, "cpu%d", i);
 388                         e->procname = kstrdup(buf, GFP_KERNEL);
 389                 }
 390                 e->mode = 0555;
 391                 e->child = sd_alloc_ctl_cpu_table(i);
 392
 393                 __cpumask_clear_cpu(i, sd_sysctl_cpus);
 394         }
 395
 396         WARN_ON(sd_sysctl_header);
 397         sd_sysctl_header = register_sysctl_table(sd_ctl_root);
 398 }
 399
 400 void dirty_sched_domain_sysctl(int cpu)
 401 {
 402         if (cpumask_available(sd_sysctl_cpus))
 403                 __cpumask_set_cpu(cpu, sd_sysctl_cpus);
 404 }
 405
 406 /* may be called multiple times per register */
 407 void unregister_sched_domain_sysctl(void)
 408 {
 409         unregister_sysctl_table(sd_sysctl_header);
 410         sd_sysctl_header = NULL;
 411 }
 412 #endif /* CONFIG_SYSCTL */
 413 #endif /* CONFIG_SMP */
 414
 415 #ifdef CONFIG_FAIR_GROUP_SCHED
 416 static void print_cfs_group_stats(struct seq_file *m, int cpu, struct task_group *tg)
 417 {
 418         struct sched_entity *se = tg->se[cpu];
 419
 420 #define P(F) \
 421         SEQ_printf(m, "  .%-30s: %lld\n", #F, (long long)F)
 422 #define P_SCHEDSTAT(F) \
 423         SEQ_printf(m, "  .%-30s: %lld\n", #F, (long long)schedstat_val(F))
 424 #define PN(F) \
 425         SEQ_printf(m, "  .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F))
 426 #define PN_SCHEDSTAT(F) \
 427         SEQ_printf(m, "  .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)schedstat_val(F)))
 428
 429         if (!se)
 430                 return;
 431
 432         PN(se->exec_start);
 433         PN(se->vruntime);
 434         PN(se->sum_exec_runtime);
 435         if (schedstat_enabled()) {
 436                 PN_SCHEDSTAT(se->statistics.wait_start);
 437                 PN_SCHEDSTAT(se->statistics.sleep_start);
 438                 PN_SCHEDSTAT(se->statistics.block_start);
 439                 PN_SCHEDSTAT(se->statistics.sleep_max);
 440                 PN_SCHEDSTAT(se->statistics.block_max);
 441                 PN_SCHEDSTAT(se->statistics.exec_max);
 442                 PN_SCHEDSTAT(se->statistics.slice_max);
 443                 PN_SCHEDSTAT(se->statistics.wait_max);
 444                 PN_SCHEDSTAT(se->statistics.wait_sum);
 445                 P_SCHEDSTAT(se->statistics.wait_count);
 446         }
 447         P(se->load.weight);
 448 #ifdef CONFIG_SMP
 449         P(se->avg.load_avg);
 450         P(se->avg.util_avg);
 451 #endif
 452
 453 #undef PN_SCHEDSTAT
 454 #undef PN
 455 #undef P_SCHEDSTAT
 456 #undef P
 457 }
 458 #endif
 459
 460 #ifdef CONFIG_CGROUP_SCHED
 461 static char group_path[PATH_MAX];
 462
 463 static char *task_group_path(struct task_group *tg)
 464 {
 465         if (autogroup_path(tg, group_path, PATH_MAX))
 466                 return group_path;
 467
 468         cgroup_path(tg->css.cgroup, group_path, PATH_MAX);
 469         return group_path;
 470 }
 471 #endif
 472
 473 static void
 474 print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
 475 {
 476         if (rq->curr == p)
 477                 SEQ_printf(m, ">R");
 478         else
 479                 SEQ_printf(m, " %c", task_state_to_char(p));
 480
 481         SEQ_printf(m, "%15s %5d %9Ld.%06ld %9Ld %5d ",
 482                 p->comm, task_pid_nr(p),
 483                 SPLIT_NS(p->se.vruntime),
 484                 (long long)(p->nvcsw + p->nivcsw),
 485                 p->prio);
 486
 487         SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
 488                 SPLIT_NS(schedstat_val_or_zero(p->se.statistics.wait_sum)),
 489                 SPLIT_NS(p->se.sum_exec_runtime),
 490                 SPLIT_NS(schedstat_val_or_zero(p->se.statistics.sum_sleep_runtime)));
 491
 492 #ifdef CONFIG_NUMA_BALANCING
 493         SEQ_printf(m, " %d %d", task_node(p), task_numa_group_id(p));
 494 #endif
 495 #ifdef CONFIG_CGROUP_SCHED
 496         SEQ_printf(m, " %s", task_group_path(task_group(p)));
 497 #endif
 498
 499         SEQ_printf(m, "\n");
 500 }
 501
 502 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
 503 {
 504         struct task_struct *g, *p;
 505
 506         SEQ_printf(m,
 507         "\nrunnable tasks:\n"
 508         " S           task   PID         tree-key  switches  prio"
 509         "     wait-time             sum-exec        sum-sleep\n"
 510         "-------------------------------------------------------"
 511         "----------------------------------------------------\n");
 512
 513         rcu_read_lock();
 514         for_each_process_thread(g, p) {
 515                 if (task_cpu(p) != rq_cpu)
 516                         continue;
 517
 518                 print_task(m, rq, p);
 519         }
 520         rcu_read_unlock();
 521 }
 522
 523 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
 524 {
 525         s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
 526                 spread, rq0_min_vruntime, spread0;
 527         struct rq *rq = cpu_rq(cpu);
 528         struct sched_entity *last;
 529         unsigned long flags;
 530
 531 #ifdef CONFIG_FAIR_GROUP_SCHED
 532         SEQ_printf(m, "\ncfs_rq[%d]:%s\n", cpu, task_group_path(cfs_rq->tg));
 533 #else
 534         SEQ_printf(m, "\ncfs_rq[%d]:\n", cpu);
 535 #endif
 536         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "exec_clock",
 537                         SPLIT_NS(cfs_rq->exec_clock));
 538
 539         raw_spin_lock_irqsave(&rq->lock, flags);
 540         if (rb_first_cached(&cfs_rq->tasks_timeline))
 541                 MIN_vruntime = (__pick_first_entity(cfs_rq))->vruntime;
 542         last = __pick_last_entity(cfs_rq);
 543         if (last)
 544                 max_vruntime = last->vruntime;
 545         min_vruntime = cfs_rq->min_vruntime;
 546         rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
 547         raw_spin_unlock_irqrestore(&rq->lock, flags);
 548         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "MIN_vruntime",
 549                         SPLIT_NS(MIN_vruntime));
 550         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "min_vruntime",
 551                         SPLIT_NS(min_vruntime));
 552         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "max_vruntime",
 553                         SPLIT_NS(max_vruntime));
 554         spread = max_vruntime - MIN_vruntime;
 555         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread",
 556                         SPLIT_NS(spread));
 557         spread0 = min_vruntime - rq0_min_vruntime;
 558         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread0",
 559                         SPLIT_NS(spread0));
 560         SEQ_printf(m, "  .%-30s: %d\n", "nr_spread_over",
 561                         cfs_rq->nr_spread_over);
 562         SEQ_printf(m, "  .%-30s: %d\n", "nr_running", cfs_rq->nr_running);
 563         SEQ_printf(m, "  .%-30s: %ld\n", "load", cfs_rq->load.weight);
 564 #ifdef CONFIG_SMP
 565         SEQ_printf(m, "  .%-30s: %lu\n", "load_avg",
 566                         cfs_rq->avg.load_avg);
 567         SEQ_printf(m, "  .%-30s: %lu\n", "runnable_load_avg",
 568                         cfs_rq->runnable_load_avg);
 569         SEQ_printf(m, "  .%-30s: %lu\n", "util_avg",
 570                         cfs_rq->avg.util_avg);
 571         SEQ_printf(m, "  .%-30s: %ld\n", "removed_load_avg",
 572                         atomic_long_read(&cfs_rq->removed_load_avg));
 573         SEQ_printf(m, "  .%-30s: %ld\n", "removed_util_avg",
 574                         atomic_long_read(&cfs_rq->removed_util_avg));
 575 #ifdef CONFIG_FAIR_GROUP_SCHED
 576         SEQ_printf(m, "  .%-30s: %lu\n", "tg_load_avg_contrib",
 577                         cfs_rq->tg_load_avg_contrib);
 578         SEQ_printf(m, "  .%-30s: %ld\n", "tg_load_avg",
 579                         atomic_long_read(&cfs_rq->tg->load_avg));
 580 #endif
 581 #endif
 582 #ifdef CONFIG_CFS_BANDWIDTH
 583         SEQ_printf(m, "  .%-30s: %d\n", "throttled",
 584                         cfs_rq->throttled);
 585         SEQ_printf(m, "  .%-30s: %d\n", "throttle_count",
 586                         cfs_rq->throttle_count);
 587 #endif
 588
 589 #ifdef CONFIG_FAIR_GROUP_SCHED
 590         print_cfs_group_stats(m, cpu, cfs_rq->tg);
 591 #endif
 592 }
 593
 594 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
 595 {
 596 #ifdef CONFIG_RT_GROUP_SCHED
 597         SEQ_printf(m, "\nrt_rq[%d]:%s\n", cpu, task_group_path(rt_rq->tg));
 598 #else
 599         SEQ_printf(m, "\nrt_rq[%d]:\n", cpu);
 600 #endif
 601
 602 #define P(x) \
 603         SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
 604 #define PU(x) \
 605         SEQ_printf(m, "  .%-30s: %lu\n", #x, (unsigned long)(rt_rq->x))
 606 #define PN(x) \
 607         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
 608
 609         PU(rt_nr_running);
 610 #ifdef CONFIG_SMP
 611         PU(rt_nr_migratory);
 612 #endif
 613         P(rt_throttled);
 614         PN(rt_time);
 615         PN(rt_runtime);
 616
 617 #undef PN
 618 #undef PU
 619 #undef P
 620 }
 621
 622 void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq)
 623 {
 624         struct dl_bw *dl_bw;
 625
 626         SEQ_printf(m, "\ndl_rq[%d]:\n", cpu);
 627
 628 #define PU(x) \
 629         SEQ_printf(m, "  .%-30s: %lu\n", #x, (unsigned long)(dl_rq->x))
 630
 631         PU(dl_nr_running);
 632 #ifdef CONFIG_SMP
 633         PU(dl_nr_migratory);
 634         dl_bw = &cpu_rq(cpu)->rd->dl_bw;
 635 #else
 636         dl_bw = &dl_rq->dl_bw;
 637 #endif
 638         SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->bw", dl_bw->bw);
 639         SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->total_bw", dl_bw->total_bw);
 640
 641 #undef PU
 642 }
 643
 644 extern __read_mostly int sched_clock_running;
 645
 646 static void print_cpu(struct seq_file *m, int cpu)
 647 {
 648         struct rq *rq = cpu_rq(cpu);
 649         unsigned long flags;
 650
 651 #ifdef CONFIG_X86
 652         {
 653                 unsigned int freq = cpu_khz ? : 1;
 654
 655                 SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
 656                            cpu, freq / 1000, (freq % 1000));
 657         }
 658 #else
 659         SEQ_printf(m, "cpu#%d\n", cpu);
 660 #endif
 661
 662 #define P(x)                                                            \
 663 do {                                                                    \
 664         if (sizeof(rq->x) == 4)                                         \
 665                 SEQ_printf(m, "  .%-30s: %ld\n", #x, (long)(rq->x));    \
 666         else                                                            \
 667                 SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rq->x));\
 668 } while (0)
 669
 670 #define PN(x) \
 671         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
 672
 673         P(nr_running);
 674         SEQ_printf(m, "  .%-30s: %lu\n", "load",
 675                    rq->load.weight);
 676         P(nr_switches);
 677         P(nr_load_updates);
 678         P(nr_uninterruptible);
 679         PN(next_balance);
 680         SEQ_printf(m, "  .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr)));
 681         PN(clock);
 682         PN(clock_task);
 683         P(cpu_load[0]);
 684         P(cpu_load[1]);
 685         P(cpu_load[2]);
 686         P(cpu_load[3]);
 687         P(cpu_load[4]);
 688 #undef P
 689 #undef PN
 690
 691 #ifdef CONFIG_SMP
 692 #define P64(n) SEQ_printf(m, "  .%-30s: %Ld\n", #n, rq->n);
 693         P64(avg_idle);
 694         P64(max_idle_balance_cost);
 695 #undef P64
 696 #endif
 697
 698 #define P(n) SEQ_printf(m, "  .%-30s: %d\n", #n, schedstat_val(rq->n));
 699         if (schedstat_enabled()) {
 700                 P(yld_count);
 701                 P(sched_count);
 702                 P(sched_goidle);
 703                 P(ttwu_count);
 704                 P(ttwu_local);
 705         }
 706 #undef P
 707
 708         spin_lock_irqsave(&sched_debug_lock, flags);
 709         print_cfs_stats(m, cpu);
 710         print_rt_stats(m, cpu);
 711         print_dl_stats(m, cpu);
 712
 713         print_rq(m, rq, cpu);
 714         spin_unlock_irqrestore(&sched_debug_lock, flags);
 715         SEQ_printf(m, "\n");
 716 }
 717
 718 static const char *sched_tunable_scaling_names[] = {
 719         "none",
 720         "logaritmic",
 721         "linear"
 722 };
 723
 724 static void sched_debug_header(struct seq_file *m)
 725 {
 726         u64 ktime, sched_clk, cpu_clk;
 727         unsigned long flags;
 728
 729         local_irq_save(flags);
 730         ktime = ktime_to_ns(ktime_get());
 731         sched_clk = sched_clock();
 732         cpu_clk = local_clock();
 733         local_irq_restore(flags);
 734
 735         SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n",
 736                 init_utsname()->release,
 737                 (int)strcspn(init_utsname()->version, " "),
 738                 init_utsname()->version);
 739
 740 #define P(x) \
 741         SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
 742 #define PN(x) \
 743         SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
 744         PN(ktime);
 745         PN(sched_clk);
 746         PN(cpu_clk);
 747         P(jiffies);
 748 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
 749         P(sched_clock_stable());
 750 #endif
 751 #undef PN
 752 #undef P
 753
 754         SEQ_printf(m, "\n");
 755         SEQ_printf(m, "sysctl_sched\n");
 756
 757 #define P(x) \
 758         SEQ_printf(m, "  .%-40s: %Ld\n", #x, (long long)(x))
 759 #define PN(x) \
 760         SEQ_printf(m, "  .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
 761         PN(sysctl_sched_latency);
 762         PN(sysctl_sched_min_granularity);
 763         PN(sysctl_sched_wakeup_granularity);
 764         P(sysctl_sched_child_runs_first);
 765         P(sysctl_sched_features);
 766 #undef PN
 767 #undef P
 768
 769         SEQ_printf(m, "  .%-40s: %d (%s)\n",
 770                 "sysctl_sched_tunable_scaling",
 771                 sysctl_sched_tunable_scaling,
 772                 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
 773         SEQ_printf(m, "\n");
 774 }
 775
 776 static int sched_debug_show(struct seq_file *m, void *v)
 777 {
 778         int cpu = (unsigned long)(v - 2);
 779
 780         if (cpu != -1)
 781                 print_cpu(m, cpu);
 782         else
 783                 sched_debug_header(m);
 784
 785         return 0;
 786 }
 787
 788 void sysrq_sched_debug_show(void)
 789 {
 790         int cpu;
 791
 792         sched_debug_header(NULL);
 793         for_each_online_cpu(cpu)
 794                 print_cpu(NULL, cpu);
 795
 796 }
 797
 798 /*
 799  * This itererator needs some explanation.
 800  * It returns 1 for the header position.
 801  * This means 2 is cpu 0.
 802  * In a hotplugged system some cpus, including cpu 0, may be missing so we have
 803  * to use cpumask_* to iterate over the cpus.
 804  */
 805 static void *sched_debug_start(struct seq_file *file, loff_t *offset)
 806 {
 807         unsigned long n = *offset;
 808
 809         if (n == 0)
 810                 return (void *) 1;
 811
 812         n--;
 813
 814         if (n > 0)
 815                 n = cpumask_next(n - 1, cpu_online_mask);
 816         else
 817                 n = cpumask_first(cpu_online_mask);
 818
 819         *offset = n + 1;
 820
 821         if (n < nr_cpu_ids)
 822                 return (void *)(unsigned long)(n + 2);
 823         return NULL;
 824 }
 825
 826 static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
 827 {
 828         (*offset)++;
 829         return sched_debug_start(file, offset);
 830 }
 831
 832 static void sched_debug_stop(struct seq_file *file, void *data)
 833 {
 834 }
 835
 836 static const struct seq_operations sched_debug_sops = {
 837         .start = sched_debug_start,
 838         .next = sched_debug_next,
 839         .stop = sched_debug_stop,
 840         .show = sched_debug_show,
 841 };
 842
 843 static int sched_debug_release(struct inode *inode, struct file *file)
 844 {
 845         seq_release(inode, file);
 846
 847         return 0;
 848 }
 849
 850 static int sched_debug_open(struct inode *inode, struct file *filp)
 851 {
 852         int ret = 0;
 853
 854         ret = seq_open(filp, &sched_debug_sops);
 855
 856         return ret;
 857 }
 858
 859 static const struct file_operations sched_debug_fops = {
 860         .open           = sched_debug_open,
 861         .read           = seq_read,
 862         .llseek         = seq_lseek,
 863         .release        = sched_debug_release,
 864 };
 865
 866 static int __init init_sched_debug_procfs(void)
 867 {
 868         struct proc_dir_entry *pe;
 869
 870         pe = proc_create("sched_debug", 0444, NULL, &sched_debug_fops);
 871         if (!pe)
 872                 return -ENOMEM;
 873         return 0;
 874 }
 875
 876 __initcall(init_sched_debug_procfs);
 877
 878 #define __P(F) \
 879         SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)F)
 880 #define P(F) \
 881         SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)p->F)
 882 #define __PN(F) \
 883         SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
 884 #define PN(F) \
 885         SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
 886
 887
 888 #ifdef CONFIG_NUMA_BALANCING
 889 void print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
 890                 unsigned long tpf, unsigned long gsf, unsigned long gpf)
 891 {
 892         SEQ_printf(m, "numa_faults node=%d ", node);
 893         SEQ_printf(m, "task_private=%lu task_shared=%lu ", tsf, tpf);
 894         SEQ_printf(m, "group_private=%lu group_shared=%lu\n", gsf, gpf);
 895 }
 896 #endif
 897
 898
 899 static void sched_show_numa(struct task_struct *p, struct seq_file *m)
 900 {
 901 #ifdef CONFIG_NUMA_BALANCING
 902         if (p->mm)
 903                 P(mm->numa_scan_seq);
 904
 905         P(numa_pages_migrated);
 906         P(numa_preferred_nid);
 907         P(total_numa_faults);
 908         SEQ_printf(m, "current_node=%d, numa_group_id=%d\n",
 909                         task_node(p), task_numa_group_id(p));
 910         show_numa_stats(p, m);
 911 #endif
 912 }
 913
 914 void proc_sched_show_task(struct task_struct *p, struct pid_namespace *ns,
 915                                                   struct seq_file *m)
 916 {
 917         unsigned long nr_switches;
 918
 919         SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr_ns(p, ns),
 920                                                 get_nr_threads(p));
 921         SEQ_printf(m,
 922                 "---------------------------------------------------------"
 923                 "----------\n");
 924 #define __P(F) \
 925         SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)F)
 926 #define P(F) \
 927         SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)p->F)
 928 #define P_SCHEDSTAT(F) \
 929         SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)schedstat_val(p->F))
 930 #define __PN(F) \
 931         SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
 932 #define PN(F) \
 933         SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
 934 #define PN_SCHEDSTAT(F) \
 935         SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)schedstat_val(p->F)))
 936
 937         PN(se.exec_start);
 938         PN(se.vruntime);
 939         PN(se.sum_exec_runtime);
 940
 941         nr_switches = p->nvcsw + p->nivcsw;
 942
 943         P(se.nr_migrations);
 944
 945         if (schedstat_enabled()) {
 946                 u64 avg_atom, avg_per_cpu;
 947
 948                 PN_SCHEDSTAT(se.statistics.sum_sleep_runtime);
 949                 PN_SCHEDSTAT(se.statistics.wait_start);
 950                 PN_SCHEDSTAT(se.statistics.sleep_start);
 951                 PN_SCHEDSTAT(se.statistics.block_start);
 952                 PN_SCHEDSTAT(se.statistics.sleep_max);
 953                 PN_SCHEDSTAT(se.statistics.block_max);
 954                 PN_SCHEDSTAT(se.statistics.exec_max);
 955                 PN_SCHEDSTAT(se.statistics.slice_max);
 956                 PN_SCHEDSTAT(se.statistics.wait_max);
 957                 PN_SCHEDSTAT(se.statistics.wait_sum);
 958                 P_SCHEDSTAT(se.statistics.wait_count);
 959                 PN_SCHEDSTAT(se.statistics.iowait_sum);
 960                 P_SCHEDSTAT(se.statistics.iowait_count);
 961                 P_SCHEDSTAT(se.statistics.nr_migrations_cold);
 962                 P_SCHEDSTAT(se.statistics.nr_failed_migrations_affine);
 963                 P_SCHEDSTAT(se.statistics.nr_failed_migrations_running);
 964                 P_SCHEDSTAT(se.statistics.nr_failed_migrations_hot);
 965                 P_SCHEDSTAT(se.statistics.nr_forced_migrations);
 966                 P_SCHEDSTAT(se.statistics.nr_wakeups);
 967                 P_SCHEDSTAT(se.statistics.nr_wakeups_sync);
 968                 P_SCHEDSTAT(se.statistics.nr_wakeups_migrate);
 969                 P_SCHEDSTAT(se.statistics.nr_wakeups_local);
 970                 P_SCHEDSTAT(se.statistics.nr_wakeups_remote);
 971                 P_SCHEDSTAT(se.statistics.nr_wakeups_affine);
 972                 P_SCHEDSTAT(se.statistics.nr_wakeups_affine_attempts);
 973                 P_SCHEDSTAT(se.statistics.nr_wakeups_passive);
 974                 P_SCHEDSTAT(se.statistics.nr_wakeups_idle);
 975
 976                 avg_atom = p->se.sum_exec_runtime;
 977                 if (nr_switches)
 978                         avg_atom = div64_ul(avg_atom, nr_switches);
 979                 else
 980                         avg_atom = -1LL;
 981
 982                 avg_per_cpu = p->se.sum_exec_runtime;
 983                 if (p->se.nr_migrations) {
 984                         avg_per_cpu = div64_u64(avg_per_cpu,
 985                                                 p->se.nr_migrations);
 986                 } else {
 987                         avg_per_cpu = -1LL;
 988                 }
 989
 990                 __PN(avg_atom);
 991                 __PN(avg_per_cpu);
 992         }
 993
 994         __P(nr_switches);
 995         SEQ_printf(m, "%-45s:%21Ld\n",
 996                    "nr_voluntary_switches", (long long)p->nvcsw);
 997         SEQ_printf(m, "%-45s:%21Ld\n",
 998                    "nr_involuntary_switches", (long long)p->nivcsw);
 999
1000         P(se.load.weight);
1001 #ifdef CONFIG_SMP
1002         P(se.avg.load_sum);
1003         P(se.avg.util_sum);
1004         P(se.avg.load_avg);
1005         P(se.avg.util_avg);
1006         P(se.avg.last_update_time);
1007 #endif
1008         P(policy);
1009         P(prio);
1010         if (p->policy == SCHED_DEADLINE) {
1011                 P(dl.runtime);
1012                 P(dl.deadline);
1013         }
1014 #undef PN_SCHEDSTAT
1015 #undef PN
1016 #undef __PN
1017 #undef P_SCHEDSTAT
1018 #undef P
1019 #undef __P
1020
1021         {
1022                 unsigned int this_cpu = raw_smp_processor_id();
1023                 u64 t0, t1;
1024
1025                 t0 = cpu_clock(this_cpu);
1026                 t1 = cpu_clock(this_cpu);
1027                 SEQ_printf(m, "%-45s:%21Ld\n",
1028                            "clock-delta", (long long)(t1-t0));
1029         }
1030
1031         sched_show_numa(p, m);
1032 }
1033
1034 void proc_sched_set_task(struct task_struct *p)
1035 {
1036 #ifdef CONFIG_SCHEDSTATS
1037         memset(&p->se.statistics, 0, sizeof(p->se.statistics));
1038 #endif
1039 }