kernel/sched/debug.c

   1 /*
   2  * kernel/sched/debug.c
   3  *
   4  * Print the CFS rbtree and other debugging details
   5  *
   6  * Copyright(C) 2007, Red Hat, Inc., Ingo Molnar
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 as
  10  * published by the Free Software Foundation.
  11  */
  12 #include "sched.h"
  13
  14 /*
  15  * This allows printing both to /proc/sched_debug and
  16  * to the console
  17  */
  18 #define SEQ_printf(m, x...)                     \
  19  do {                                           \
  20         if (m)                                  \
  21                 seq_printf(m, x);               \
  22         else                                    \
  23                 pr_cont(x);                     \
  24  } while (0)
  25
  26 /*
  27  * Ease the printing of nsec fields:
  28  */
  29 static long long nsec_high(unsigned long long nsec)
  30 {
  31         if ((long long)nsec < 0) {
  32                 nsec = -nsec;
  33                 do_div(nsec, 1000000);
  34                 return -nsec;
  35         }
  36         do_div(nsec, 1000000);
  37
  38         return nsec;
  39 }
  40
  41 static unsigned long nsec_low(unsigned long long nsec)
  42 {
  43         if ((long long)nsec < 0)
  44                 nsec = -nsec;
  45
  46         return do_div(nsec, 1000000);
  47 }
  48
  49 #define SPLIT_NS(x) nsec_high(x), nsec_low(x)
  50
  51 #define SCHED_FEAT(name, enabled)       \
  52         #name ,
  53
  54 static const char * const sched_feat_names[] = {
  55 #include "features.h"
  56 };
  57
  58 #undef SCHED_FEAT
  59
  60 static int sched_feat_show(struct seq_file *m, void *v)
  61 {
  62         int i;
  63
  64         for (i = 0; i < __SCHED_FEAT_NR; i++) {
  65                 if (!(sysctl_sched_features & (1UL << i)))
  66                         seq_puts(m, "NO_");
  67                 seq_printf(m, "%s ", sched_feat_names[i]);
  68         }
  69         seq_puts(m, "\n");
  70
  71         return 0;
  72 }
  73
  74 #ifdef CONFIG_JUMP_LABEL
  75
  76 #define jump_label_key__true  STATIC_KEY_INIT_TRUE
  77 #define jump_label_key__false STATIC_KEY_INIT_FALSE
  78
  79 #define SCHED_FEAT(name, enabled)       \
  80         jump_label_key__##enabled ,
  81
  82 struct static_key sched_feat_keys[__SCHED_FEAT_NR] = {
  83 #include "features.h"
  84 };
  85
  86 #undef SCHED_FEAT
  87
  88 static void sched_feat_disable(int i)
  89 {
  90         static_key_disable_cpuslocked(&sched_feat_keys[i]);
  91 }
  92
  93 static void sched_feat_enable(int i)
  94 {
  95         static_key_enable_cpuslocked(&sched_feat_keys[i]);
  96 }
  97 #else
  98 static void sched_feat_disable(int i) { };
  99 static void sched_feat_enable(int i) { };
 100 #endif /* CONFIG_JUMP_LABEL */
 101
 102 static int sched_feat_set(char *cmp)
 103 {
 104         int i;
 105         int neg = 0;
 106
 107         if (strncmp(cmp, "NO_", 3) == 0) {
 108                 neg = 1;
 109                 cmp += 3;
 110         }
 111
 112         i = match_string(sched_feat_names, __SCHED_FEAT_NR, cmp);
 113         if (i < 0)
 114                 return i;
 115
 116         if (neg) {
 117                 sysctl_sched_features &= ~(1UL << i);
 118                 sched_feat_disable(i);
 119         } else {
 120                 sysctl_sched_features |= (1UL << i);
 121                 sched_feat_enable(i);
 122         }
 123
 124         return 0;
 125 }
 126
 127 static ssize_t
 128 sched_feat_write(struct file *filp, const char __user *ubuf,
 129                 size_t cnt, loff_t *ppos)
 130 {
 131         char buf[64];
 132         char *cmp;
 133         int ret;
 134         struct inode *inode;
 135
 136         if (cnt > 63)
 137                 cnt = 63;
 138
 139         if (copy_from_user(&buf, ubuf, cnt))
 140                 return -EFAULT;
 141
 142         buf[cnt] = 0;
 143         cmp = strstrip(buf);
 144
 145         /* Ensure the static_key remains in a consistent state */
 146         inode = file_inode(filp);
 147         cpus_read_lock();
 148         inode_lock(inode);
 149         ret = sched_feat_set(cmp);
 150         inode_unlock(inode);
 151         cpus_read_unlock();
 152         if (ret < 0)
 153                 return ret;
 154
 155         *ppos += cnt;
 156
 157         return cnt;
 158 }
 159
 160 static int sched_feat_open(struct inode *inode, struct file *filp)
 161 {
 162         return single_open(filp, sched_feat_show, NULL);
 163 }
 164
 165 static const struct file_operations sched_feat_fops = {
 166         .open           = sched_feat_open,
 167         .write          = sched_feat_write,
 168         .read           = seq_read,
 169         .llseek         = seq_lseek,
 170         .release        = single_release,
 171 };
 172
 173 __read_mostly bool sched_debug_enabled;
 174
 175 static __init int sched_init_debug(void)
 176 {
 177         debugfs_create_file("sched_features", 0644, NULL, NULL,
 178                         &sched_feat_fops);
 179
 180         debugfs_create_bool("sched_debug", 0644, NULL,
 181                         &sched_debug_enabled);
 182
 183         return 0;
 184 }
 185 late_initcall(sched_init_debug);
 186
 187 #ifdef CONFIG_SMP
 188
 189 #ifdef CONFIG_SYSCTL
 190
 191 static struct ctl_table sd_ctl_dir[] = {
 192         {
 193                 .procname       = "sched_domain",
 194                 .mode           = 0555,
 195         },
 196         {}
 197 };
 198
 199 static struct ctl_table sd_ctl_root[] = {
 200         {
 201                 .procname       = "kernel",
 202                 .mode           = 0555,
 203                 .child          = sd_ctl_dir,
 204         },
 205         {}
 206 };
 207
 208 static struct ctl_table *sd_alloc_ctl_entry(int n)
 209 {
 210         struct ctl_table *entry =
 211                 kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL);
 212
 213         return entry;
 214 }
 215
 216 static void sd_free_ctl_entry(struct ctl_table **tablep)
 217 {
 218         struct ctl_table *entry;
 219
 220         /*
 221          * In the intermediate directories, both the child directory and
 222          * procname are dynamically allocated and could fail but the mode
 223          * will always be set. In the lowest directory the names are
 224          * static strings and all have proc handlers.
 225          */
 226         for (entry = *tablep; entry->mode; entry++) {
 227                 if (entry->child)
 228                         sd_free_ctl_entry(&entry->child);
 229                 if (entry->proc_handler == NULL)
 230                         kfree(entry->procname);
 231         }
 232
 233         kfree(*tablep);
 234         *tablep = NULL;
 235 }
 236
 237 static int min_load_idx = 0;
 238 static int max_load_idx = CPU_LOAD_IDX_MAX-1;
 239
 240 static void
 241 set_table_entry(struct ctl_table *entry,
 242                 const char *procname, void *data, int maxlen,
 243                 umode_t mode, proc_handler *proc_handler,
 244                 bool load_idx)
 245 {
 246         entry->procname = procname;
 247         entry->data = data;
 248         entry->maxlen = maxlen;
 249         entry->mode = mode;
 250         entry->proc_handler = proc_handler;
 251
 252         if (load_idx) {
 253                 entry->extra1 = &min_load_idx;
 254                 entry->extra2 = &max_load_idx;
 255         }
 256 }
 257
 258 static struct ctl_table *
 259 sd_alloc_ctl_domain_table(struct sched_domain *sd)
 260 {
 261         struct ctl_table *table = sd_alloc_ctl_entry(14);
 262
 263         if (table == NULL)
 264                 return NULL;
 265
 266         set_table_entry(&table[0] , "min_interval",        &sd->min_interval,        sizeof(long), 0644, proc_doulongvec_minmax, false);
 267         set_table_entry(&table[1] , "max_interval",        &sd->max_interval,        sizeof(long), 0644, proc_doulongvec_minmax, false);
 268         set_table_entry(&table[2] , "busy_idx",            &sd->busy_idx,            sizeof(int) , 0644, proc_dointvec_minmax,   true );
 269         set_table_entry(&table[3] , "idle_idx",            &sd->idle_idx,            sizeof(int) , 0644, proc_dointvec_minmax,   true );
 270         set_table_entry(&table[4] , "newidle_idx",         &sd->newidle_idx,         sizeof(int) , 0644, proc_dointvec_minmax,   true );
 271         set_table_entry(&table[5] , "wake_idx",            &sd->wake_idx,            sizeof(int) , 0644, proc_dointvec_minmax,   true );
 272         set_table_entry(&table[6] , "forkexec_idx",        &sd->forkexec_idx,        sizeof(int) , 0644, proc_dointvec_minmax,   true );
 273         set_table_entry(&table[7] , "busy_factor",         &sd->busy_factor,         sizeof(int) , 0644, proc_dointvec_minmax,   false);
 274         set_table_entry(&table[8] , "imbalance_pct",       &sd->imbalance_pct,       sizeof(int) , 0644, proc_dointvec_minmax,   false);
 275         set_table_entry(&table[9] , "cache_nice_tries",    &sd->cache_nice_tries,    sizeof(int) , 0644, proc_dointvec_minmax,   false);
 276         set_table_entry(&table[10], "flags",               &sd->flags,               sizeof(int) , 0644, proc_dointvec_minmax,   false);
 277         set_table_entry(&table[11], "max_newidle_lb_cost", &sd->max_newidle_lb_cost, sizeof(long), 0644, proc_doulongvec_minmax, false);
 278         set_table_entry(&table[12], "name",                sd->name,            CORENAME_MAX_SIZE, 0444, proc_dostring,          false);
 279         /* &table[13] is terminator */
 280
 281         return table;
 282 }
 283
 284 static struct ctl_table *sd_alloc_ctl_cpu_table(int cpu)
 285 {
 286         struct ctl_table *entry, *table;
 287         struct sched_domain *sd;
 288         int domain_num = 0, i;
 289         char buf[32];
 290
 291         for_each_domain(cpu, sd)
 292                 domain_num++;
 293         entry = table = sd_alloc_ctl_entry(domain_num + 1);
 294         if (table == NULL)
 295                 return NULL;
 296
 297         i = 0;
 298         for_each_domain(cpu, sd) {
 299                 snprintf(buf, 32, "domain%d", i);
 300                 entry->procname = kstrdup(buf, GFP_KERNEL);
 301                 entry->mode = 0555;
 302                 entry->child = sd_alloc_ctl_domain_table(sd);
 303                 entry++;
 304                 i++;
 305         }
 306         return table;
 307 }
 308
 309 static cpumask_var_t            sd_sysctl_cpus;
 310 static struct ctl_table_header  *sd_sysctl_header;
 311
 312 void register_sched_domain_sysctl(void)
 313 {
 314         static struct ctl_table *cpu_entries;
 315         static struct ctl_table **cpu_idx;
 316         static bool init_done = false;
 317         char buf[32];
 318         int i;
 319
 320         if (!cpu_entries) {
 321                 cpu_entries = sd_alloc_ctl_entry(num_possible_cpus() + 1);
 322                 if (!cpu_entries)
 323                         return;
 324
 325                 WARN_ON(sd_ctl_dir[0].child);
 326                 sd_ctl_dir[0].child = cpu_entries;
 327         }
 328
 329         if (!cpu_idx) {
 330                 struct ctl_table *e = cpu_entries;
 331
 332                 cpu_idx = kcalloc(nr_cpu_ids, sizeof(struct ctl_table*), GFP_KERNEL);
 333                 if (!cpu_idx)
 334                         return;
 335
 336                 /* deal with sparse possible map */
 337                 for_each_possible_cpu(i) {
 338                         cpu_idx[i] = e;
 339                         e++;
 340                 }
 341         }
 342
 343         if (!cpumask_available(sd_sysctl_cpus)) {
 344                 if (!alloc_cpumask_var(&sd_sysctl_cpus, GFP_KERNEL))
 345                         return;
 346         }
 347
 348         if (!init_done) {
 349                 init_done = true;
 350                 /* init to possible to not have holes in @cpu_entries */
 351                 cpumask_copy(sd_sysctl_cpus, cpu_possible_mask);
 352         }
 353
 354         for_each_cpu(i, sd_sysctl_cpus) {
 355                 struct ctl_table *e = cpu_idx[i];
 356
 357                 if (e->child)
 358                         sd_free_ctl_entry(&e->child);
 359
 360                 if (!e->procname) {
 361                         snprintf(buf, 32, "cpu%d", i);
 362                         e->procname = kstrdup(buf, GFP_KERNEL);
 363                 }
 364                 e->mode = 0555;
 365                 e->child = sd_alloc_ctl_cpu_table(i);
 366
 367                 __cpumask_clear_cpu(i, sd_sysctl_cpus);
 368         }
 369
 370         WARN_ON(sd_sysctl_header);
 371         sd_sysctl_header = register_sysctl_table(sd_ctl_root);
 372 }
 373
 374 void dirty_sched_domain_sysctl(int cpu)
 375 {
 376         if (cpumask_available(sd_sysctl_cpus))
 377                 __cpumask_set_cpu(cpu, sd_sysctl_cpus);
 378 }
 379
 380 /* may be called multiple times per register */
 381 void unregister_sched_domain_sysctl(void)
 382 {
 383         unregister_sysctl_table(sd_sysctl_header);
 384         sd_sysctl_header = NULL;
 385 }
 386 #endif /* CONFIG_SYSCTL */
 387 #endif /* CONFIG_SMP */
 388
 389 #ifdef CONFIG_FAIR_GROUP_SCHED
 390 static void print_cfs_group_stats(struct seq_file *m, int cpu, struct task_group *tg)
 391 {
 392         struct sched_entity *se = tg->se[cpu];
 393
 394 #define P(F)            SEQ_printf(m, "  .%-30s: %lld\n",       #F, (long long)F)
 395 #define P_SCHEDSTAT(F)  SEQ_printf(m, "  .%-30s: %lld\n",       #F, (long long)schedstat_val(F))
 396 #define PN(F)           SEQ_printf(m, "  .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F))
 397 #define PN_SCHEDSTAT(F) SEQ_printf(m, "  .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)schedstat_val(F)))
 398
 399         if (!se)
 400                 return;
 401
 402         PN(se->exec_start);
 403         PN(se->vruntime);
 404         PN(se->sum_exec_runtime);
 405
 406         if (schedstat_enabled()) {
 407                 PN_SCHEDSTAT(se->statistics.wait_start);
 408                 PN_SCHEDSTAT(se->statistics.sleep_start);
 409                 PN_SCHEDSTAT(se->statistics.block_start);
 410                 PN_SCHEDSTAT(se->statistics.sleep_max);
 411                 PN_SCHEDSTAT(se->statistics.block_max);
 412                 PN_SCHEDSTAT(se->statistics.exec_max);
 413                 PN_SCHEDSTAT(se->statistics.slice_max);
 414                 PN_SCHEDSTAT(se->statistics.wait_max);
 415                 PN_SCHEDSTAT(se->statistics.wait_sum);
 416                 P_SCHEDSTAT(se->statistics.wait_count);
 417         }
 418
 419         P(se->load.weight);
 420         P(se->runnable_weight);
 421 #ifdef CONFIG_SMP
 422         P(se->avg.load_avg);
 423         P(se->avg.util_avg);
 424         P(se->avg.runnable_load_avg);
 425 #endif
 426
 427 #undef PN_SCHEDSTAT
 428 #undef PN
 429 #undef P_SCHEDSTAT
 430 #undef P
 431 }
 432 #endif
 433
 434 #ifdef CONFIG_CGROUP_SCHED
 435 static DEFINE_SPINLOCK(sched_debug_lock);
 436 static char group_path[PATH_MAX];
 437
 438 static void task_group_path(struct task_group *tg, char *path, int plen)
 439 {
 440         if (autogroup_path(tg, path, plen))
 441                 return;
 442
 443         cgroup_path(tg->css.cgroup, path, plen);
 444 }
 445
 446 /*
 447  * Only 1 SEQ_printf_task_group_path() caller can use the full length
 448  * group_path[] for cgroup path. Other simultaneous callers will have
 449  * to use a shorter stack buffer. A "..." suffix is appended at the end
 450  * of the stack buffer so that it will show up in case the output length
 451  * matches the given buffer size to indicate possible path name truncation.
 452  */
 453 #define SEQ_printf_task_group_path(m, tg, fmt...)                       \
 454 {                                                                       \
 455         if (spin_trylock(&sched_debug_lock)) {                          \
 456                 task_group_path(tg, group_path, sizeof(group_path));    \
 457                 SEQ_printf(m, fmt, group_path);                         \
 458                 spin_unlock(&sched_debug_lock);                         \
 459         } else {                                                        \
 460                 char buf[128];                                          \
 461                 char *bufend = buf + sizeof(buf) - 3;                   \
 462                 task_group_path(tg, buf, bufend - buf);                 \
 463                 strcpy(bufend - 1, "...");                              \
 464                 SEQ_printf(m, fmt, buf);                                \
 465         }                                                               \
 466 }
 467 #endif
 468
 469 static void
 470 print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
 471 {
 472         if (rq->curr == p)
 473                 SEQ_printf(m, ">R");
 474         else
 475                 SEQ_printf(m, " %c", task_state_to_char(p));
 476
 477         SEQ_printf(m, "%15s %5d %9Ld.%06ld %9Ld %5d ",
 478                 p->comm, task_pid_nr(p),
 479                 SPLIT_NS(p->se.vruntime),
 480                 (long long)(p->nvcsw + p->nivcsw),
 481                 p->prio);
 482
 483         SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
 484                 SPLIT_NS(schedstat_val_or_zero(p->se.statistics.wait_sum)),
 485                 SPLIT_NS(p->se.sum_exec_runtime),
 486                 SPLIT_NS(schedstat_val_or_zero(p->se.statistics.sum_sleep_runtime)));
 487
 488 #ifdef CONFIG_NUMA_BALANCING
 489         SEQ_printf(m, " %d %d", task_node(p), task_numa_group_id(p));
 490 #endif
 491 #ifdef CONFIG_CGROUP_SCHED
 492         SEQ_printf_task_group_path(m, task_group(p), " %s")
 493 #endif
 494
 495         SEQ_printf(m, "\n");
 496 }
 497
 498 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
 499 {
 500         struct task_struct *g, *p;
 501
 502         SEQ_printf(m, "\n");
 503         SEQ_printf(m, "runnable tasks:\n");
 504         SEQ_printf(m, " S           task   PID         tree-key  switches  prio"
 505                    "     wait-time             sum-exec        sum-sleep\n");
 506         SEQ_printf(m, "-------------------------------------------------------"
 507                    "----------------------------------------------------\n");
 508
 509         rcu_read_lock();
 510         for_each_process_thread(g, p) {
 511                 if (task_cpu(p) != rq_cpu)
 512                         continue;
 513
 514                 print_task(m, rq, p);
 515         }
 516         rcu_read_unlock();
 517 }
 518
 519 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
 520 {
 521         s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
 522                 spread, rq0_min_vruntime, spread0;
 523         struct rq *rq = cpu_rq(cpu);
 524         struct sched_entity *last;
 525         unsigned long flags;
 526
 527 #ifdef CONFIG_FAIR_GROUP_SCHED
 528         SEQ_printf(m, "\n");
 529         SEQ_printf_task_group_path(m, cfs_rq->tg, "cfs_rq[%d]:%s\n", cpu);
 530 #else
 531         SEQ_printf(m, "\n");
 532         SEQ_printf(m, "cfs_rq[%d]:\n", cpu);
 533 #endif
 534         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "exec_clock",
 535                         SPLIT_NS(cfs_rq->exec_clock));
 536
 537         raw_spin_lock_irqsave(&rq->lock, flags);
 538         if (rb_first_cached(&cfs_rq->tasks_timeline))
 539                 MIN_vruntime = (__pick_first_entity(cfs_rq))->vruntime;
 540         last = __pick_last_entity(cfs_rq);
 541         if (last)
 542                 max_vruntime = last->vruntime;
 543         min_vruntime = cfs_rq->min_vruntime;
 544         rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
 545         raw_spin_unlock_irqrestore(&rq->lock, flags);
 546         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "MIN_vruntime",
 547                         SPLIT_NS(MIN_vruntime));
 548         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "min_vruntime",
 549                         SPLIT_NS(min_vruntime));
 550         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "max_vruntime",
 551                         SPLIT_NS(max_vruntime));
 552         spread = max_vruntime - MIN_vruntime;
 553         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread",
 554                         SPLIT_NS(spread));
 555         spread0 = min_vruntime - rq0_min_vruntime;
 556         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread0",
 557                         SPLIT_NS(spread0));
 558         SEQ_printf(m, "  .%-30s: %d\n", "nr_spread_over",
 559                         cfs_rq->nr_spread_over);
 560         SEQ_printf(m, "  .%-30s: %d\n", "nr_running", cfs_rq->nr_running);
 561         SEQ_printf(m, "  .%-30s: %ld\n", "load", cfs_rq->load.weight);
 562 #ifdef CONFIG_SMP
 563         SEQ_printf(m, "  .%-30s: %ld\n", "runnable_weight", cfs_rq->runnable_weight);
 564         SEQ_printf(m, "  .%-30s: %lu\n", "load_avg",
 565                         cfs_rq->avg.load_avg);
 566         SEQ_printf(m, "  .%-30s: %lu\n", "runnable_load_avg",
 567                         cfs_rq->avg.runnable_load_avg);
 568         SEQ_printf(m, "  .%-30s: %lu\n", "util_avg",
 569                         cfs_rq->avg.util_avg);
 570         SEQ_printf(m, "  .%-30s: %u\n", "util_est_enqueued",
 571                         cfs_rq->avg.util_est.enqueued);
 572         SEQ_printf(m, "  .%-30s: %ld\n", "removed.load_avg",
 573                         cfs_rq->removed.load_avg);
 574         SEQ_printf(m, "  .%-30s: %ld\n", "removed.util_avg",
 575                         cfs_rq->removed.util_avg);
 576         SEQ_printf(m, "  .%-30s: %ld\n", "removed.runnable_sum",
 577                         cfs_rq->removed.runnable_sum);
 578 #ifdef CONFIG_FAIR_GROUP_SCHED
 579         SEQ_printf(m, "  .%-30s: %lu\n", "tg_load_avg_contrib",
 580                         cfs_rq->tg_load_avg_contrib);
 581         SEQ_printf(m, "  .%-30s: %ld\n", "tg_load_avg",
 582                         atomic_long_read(&cfs_rq->tg->load_avg));
 583 #endif
 584 #endif
 585 #ifdef CONFIG_CFS_BANDWIDTH
 586         SEQ_printf(m, "  .%-30s: %d\n", "throttled",
 587                         cfs_rq->throttled);
 588         SEQ_printf(m, "  .%-30s: %d\n", "throttle_count",
 589                         cfs_rq->throttle_count);
 590 #endif
 591
 592 #ifdef CONFIG_FAIR_GROUP_SCHED
 593         print_cfs_group_stats(m, cpu, cfs_rq->tg);
 594 #endif
 595 }
 596
 597 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
 598 {
 599 #ifdef CONFIG_RT_GROUP_SCHED
 600         SEQ_printf(m, "\n");
 601         SEQ_printf_task_group_path(m, rt_rq->tg, "rt_rq[%d]:%s\n", cpu);
 602 #else
 603         SEQ_printf(m, "\n");
 604         SEQ_printf(m, "rt_rq[%d]:\n", cpu);
 605 #endif
 606
 607 #define P(x) \
 608         SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
 609 #define PU(x) \
 610         SEQ_printf(m, "  .%-30s: %lu\n", #x, (unsigned long)(rt_rq->x))
 611 #define PN(x) \
 612         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
 613
 614         PU(rt_nr_running);
 615 #ifdef CONFIG_SMP
 616         PU(rt_nr_migratory);
 617 #endif
 618         P(rt_throttled);
 619         PN(rt_time);
 620         PN(rt_runtime);
 621
 622 #undef PN
 623 #undef PU
 624 #undef P
 625 }
 626
 627 void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq)
 628 {
 629         struct dl_bw *dl_bw;
 630
 631         SEQ_printf(m, "\n");
 632         SEQ_printf(m, "dl_rq[%d]:\n", cpu);
 633
 634 #define PU(x) \
 635         SEQ_printf(m, "  .%-30s: %lu\n", #x, (unsigned long)(dl_rq->x))
 636
 637         PU(dl_nr_running);
 638 #ifdef CONFIG_SMP
 639         PU(dl_nr_migratory);
 640         dl_bw = &cpu_rq(cpu)->rd->dl_bw;
 641 #else
 642         dl_bw = &dl_rq->dl_bw;
 643 #endif
 644         SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->bw", dl_bw->bw);
 645         SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->total_bw", dl_bw->total_bw);
 646
 647 #undef PU
 648 }
 649
 650 static void print_cpu(struct seq_file *m, int cpu)
 651 {
 652         struct rq *rq = cpu_rq(cpu);
 653
 654 #ifdef CONFIG_X86
 655         {
 656                 unsigned int freq = cpu_khz ? : 1;
 657
 658                 SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
 659                            cpu, freq / 1000, (freq % 1000));
 660         }
 661 #else
 662         SEQ_printf(m, "cpu#%d\n", cpu);
 663 #endif
 664
 665 #define P(x)                                                            \
 666 do {                                                                    \
 667         if (sizeof(rq->x) == 4)                                         \
 668                 SEQ_printf(m, "  .%-30s: %ld\n", #x, (long)(rq->x));    \
 669         else                                                            \
 670                 SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rq->x));\
 671 } while (0)
 672
 673 #define PN(x) \
 674         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
 675
 676         P(nr_running);
 677         SEQ_printf(m, "  .%-30s: %lu\n", "load",
 678                    rq->load.weight);
 679         P(nr_switches);
 680         P(nr_load_updates);
 681         P(nr_uninterruptible);
 682         PN(next_balance);
 683         SEQ_printf(m, "  .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr)));
 684         PN(clock);
 685         PN(clock_task);
 686         P(cpu_load[0]);
 687         P(cpu_load[1]);
 688         P(cpu_load[2]);
 689         P(cpu_load[3]);
 690         P(cpu_load[4]);
 691 #undef P
 692 #undef PN
 693
 694 #ifdef CONFIG_SMP
 695 #define P64(n) SEQ_printf(m, "  .%-30s: %Ld\n", #n, rq->n);
 696         P64(avg_idle);
 697         P64(max_idle_balance_cost);
 698 #undef P64
 699 #endif
 700
 701 #define P(n) SEQ_printf(m, "  .%-30s: %d\n", #n, schedstat_val(rq->n));
 702         if (schedstat_enabled()) {
 703                 P(yld_count);
 704                 P(sched_count);
 705                 P(sched_goidle);
 706                 P(ttwu_count);
 707                 P(ttwu_local);
 708         }
 709 #undef P
 710
 711         print_cfs_stats(m, cpu);
 712         print_rt_stats(m, cpu);
 713         print_dl_stats(m, cpu);
 714
 715         print_rq(m, rq, cpu);
 716         SEQ_printf(m, "\n");
 717 }
 718
 719 static const char *sched_tunable_scaling_names[] = {
 720         "none",
 721         "logaritmic",
 722         "linear"
 723 };
 724
 725 static void sched_debug_header(struct seq_file *m)
 726 {
 727         u64 ktime, sched_clk, cpu_clk;
 728         unsigned long flags;
 729
 730         local_irq_save(flags);
 731         ktime = ktime_to_ns(ktime_get());
 732         sched_clk = sched_clock();
 733         cpu_clk = local_clock();
 734         local_irq_restore(flags);
 735
 736         SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n",
 737                 init_utsname()->release,
 738                 (int)strcspn(init_utsname()->version, " "),
 739                 init_utsname()->version);
 740
 741 #define P(x) \
 742         SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
 743 #define PN(x) \
 744         SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
 745         PN(ktime);
 746         PN(sched_clk);
 747         PN(cpu_clk);
 748         P(jiffies);
 749 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
 750         P(sched_clock_stable());
 751 #endif
 752 #undef PN
 753 #undef P
 754
 755         SEQ_printf(m, "\n");
 756         SEQ_printf(m, "sysctl_sched\n");
 757
 758 #define P(x) \
 759         SEQ_printf(m, "  .%-40s: %Ld\n", #x, (long long)(x))
 760 #define PN(x) \
 761         SEQ_printf(m, "  .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
 762         PN(sysctl_sched_latency);
 763         PN(sysctl_sched_min_granularity);
 764         PN(sysctl_sched_wakeup_granularity);
 765         P(sysctl_sched_child_runs_first);
 766         P(sysctl_sched_features);
 767 #undef PN
 768 #undef P
 769
 770         SEQ_printf(m, "  .%-40s: %d (%s)\n",
 771                 "sysctl_sched_tunable_scaling",
 772                 sysctl_sched_tunable_scaling,
 773                 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
 774         SEQ_printf(m, "\n");
 775 }
 776
 777 static int sched_debug_show(struct seq_file *m, void *v)
 778 {
 779         int cpu = (unsigned long)(v - 2);
 780
 781         if (cpu != -1)
 782                 print_cpu(m, cpu);
 783         else
 784                 sched_debug_header(m);
 785
 786         return 0;
 787 }
 788
 789 void sysrq_sched_debug_show(void)
 790 {
 791         int cpu;
 792
 793         sched_debug_header(NULL);
 794         for_each_online_cpu(cpu)
 795                 print_cpu(NULL, cpu);
 796
 797 }
 798
 799 /*
 800  * This itererator needs some explanation.
 801  * It returns 1 for the header position.
 802  * This means 2 is CPU 0.
 803  * In a hotplugged system some CPUs, including CPU 0, may be missing so we have
 804  * to use cpumask_* to iterate over the CPUs.
 805  */
 806 static void *sched_debug_start(struct seq_file *file, loff_t *offset)
 807 {
 808         unsigned long n = *offset;
 809
 810         if (n == 0)
 811                 return (void *) 1;
 812
 813         n--;
 814
 815         if (n > 0)
 816                 n = cpumask_next(n - 1, cpu_online_mask);
 817         else
 818                 n = cpumask_first(cpu_online_mask);
 819
 820         *offset = n + 1;
 821
 822         if (n < nr_cpu_ids)
 823                 return (void *)(unsigned long)(n + 2);
 824
 825         return NULL;
 826 }
 827
 828 static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
 829 {
 830         (*offset)++;
 831         return sched_debug_start(file, offset);
 832 }
 833
 834 static void sched_debug_stop(struct seq_file *file, void *data)
 835 {
 836 }
 837
 838 static const struct seq_operations sched_debug_sops = {
 839         .start          = sched_debug_start,
 840         .next           = sched_debug_next,
 841         .stop           = sched_debug_stop,
 842         .show           = sched_debug_show,
 843 };
 844
 845 static int __init init_sched_debug_procfs(void)
 846 {
 847         if (!proc_create_seq("sched_debug", 0444, NULL, &sched_debug_sops))
 848                 return -ENOMEM;
 849         return 0;
 850 }
 851
 852 __initcall(init_sched_debug_procfs);
 853
 854 #define __P(F)  SEQ_printf(m, "%-45s:%21Ld\n",       #F, (long long)F)
 855 #define   P(F)  SEQ_printf(m, "%-45s:%21Ld\n",       #F, (long long)p->F)
 856 #define __PN(F) SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
 857 #define   PN(F) SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
 858
 859
 860 #ifdef CONFIG_NUMA_BALANCING
 861 void print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
 862                 unsigned long tpf, unsigned long gsf, unsigned long gpf)
 863 {
 864         SEQ_printf(m, "numa_faults node=%d ", node);
 865         SEQ_printf(m, "task_private=%lu task_shared=%lu ", tpf, tsf);
 866         SEQ_printf(m, "group_private=%lu group_shared=%lu\n", gpf, gsf);
 867 }
 868 #endif
 869
 870
 871 static void sched_show_numa(struct task_struct *p, struct seq_file *m)
 872 {
 873 #ifdef CONFIG_NUMA_BALANCING
 874         struct mempolicy *pol;
 875
 876         if (p->mm)
 877                 P(mm->numa_scan_seq);
 878
 879         task_lock(p);
 880         pol = p->mempolicy;
 881         if (pol && !(pol->flags & MPOL_F_MORON))
 882                 pol = NULL;
 883         mpol_get(pol);
 884         task_unlock(p);
 885
 886         P(numa_pages_migrated);
 887         P(numa_preferred_nid);
 888         P(total_numa_faults);
 889         SEQ_printf(m, "current_node=%d, numa_group_id=%d\n",
 890                         task_node(p), task_numa_group_id(p));
 891         show_numa_stats(p, m);
 892         mpol_put(pol);
 893 #endif
 894 }
 895
 896 void proc_sched_show_task(struct task_struct *p, struct pid_namespace *ns,
 897                                                   struct seq_file *m)
 898 {
 899         unsigned long nr_switches;
 900
 901         SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr_ns(p, ns),
 902                                                 get_nr_threads(p));
 903         SEQ_printf(m,
 904                 "---------------------------------------------------------"
 905                 "----------\n");
 906 #define __P(F) \
 907         SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)F)
 908 #define P(F) \
 909         SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)p->F)
 910 #define P_SCHEDSTAT(F) \
 911         SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)schedstat_val(p->F))
 912 #define __PN(F) \
 913         SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
 914 #define PN(F) \
 915         SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
 916 #define PN_SCHEDSTAT(F) \
 917         SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)schedstat_val(p->F)))
 918
 919         PN(se.exec_start);
 920         PN(se.vruntime);
 921         PN(se.sum_exec_runtime);
 922
 923         nr_switches = p->nvcsw + p->nivcsw;
 924
 925         P(se.nr_migrations);
 926
 927         if (schedstat_enabled()) {
 928                 u64 avg_atom, avg_per_cpu;
 929
 930                 PN_SCHEDSTAT(se.statistics.sum_sleep_runtime);
 931                 PN_SCHEDSTAT(se.statistics.wait_start);
 932                 PN_SCHEDSTAT(se.statistics.sleep_start);
 933                 PN_SCHEDSTAT(se.statistics.block_start);
 934                 PN_SCHEDSTAT(se.statistics.sleep_max);
 935                 PN_SCHEDSTAT(se.statistics.block_max);
 936                 PN_SCHEDSTAT(se.statistics.exec_max);
 937                 PN_SCHEDSTAT(se.statistics.slice_max);
 938                 PN_SCHEDSTAT(se.statistics.wait_max);
 939                 PN_SCHEDSTAT(se.statistics.wait_sum);
 940                 P_SCHEDSTAT(se.statistics.wait_count);
 941                 PN_SCHEDSTAT(se.statistics.iowait_sum);
 942                 P_SCHEDSTAT(se.statistics.iowait_count);
 943                 P_SCHEDSTAT(se.statistics.nr_migrations_cold);
 944                 P_SCHEDSTAT(se.statistics.nr_failed_migrations_affine);
 945                 P_SCHEDSTAT(se.statistics.nr_failed_migrations_running);
 946                 P_SCHEDSTAT(se.statistics.nr_failed_migrations_hot);
 947                 P_SCHEDSTAT(se.statistics.nr_forced_migrations);
 948                 P_SCHEDSTAT(se.statistics.nr_wakeups);
 949                 P_SCHEDSTAT(se.statistics.nr_wakeups_sync);
 950                 P_SCHEDSTAT(se.statistics.nr_wakeups_migrate);
 951                 P_SCHEDSTAT(se.statistics.nr_wakeups_local);
 952                 P_SCHEDSTAT(se.statistics.nr_wakeups_remote);
 953                 P_SCHEDSTAT(se.statistics.nr_wakeups_affine);
 954                 P_SCHEDSTAT(se.statistics.nr_wakeups_affine_attempts);
 955                 P_SCHEDSTAT(se.statistics.nr_wakeups_passive);
 956                 P_SCHEDSTAT(se.statistics.nr_wakeups_idle);
 957
 958                 avg_atom = p->se.sum_exec_runtime;
 959                 if (nr_switches)
 960                         avg_atom = div64_ul(avg_atom, nr_switches);
 961                 else
 962                         avg_atom = -1LL;
 963
 964                 avg_per_cpu = p->se.sum_exec_runtime;
 965                 if (p->se.nr_migrations) {
 966                         avg_per_cpu = div64_u64(avg_per_cpu,
 967                                                 p->se.nr_migrations);
 968                 } else {
 969                         avg_per_cpu = -1LL;
 970                 }
 971
 972                 __PN(avg_atom);
 973                 __PN(avg_per_cpu);
 974         }
 975
 976         __P(nr_switches);
 977         SEQ_printf(m, "%-45s:%21Ld\n",
 978                    "nr_voluntary_switches", (long long)p->nvcsw);
 979         SEQ_printf(m, "%-45s:%21Ld\n",
 980                    "nr_involuntary_switches", (long long)p->nivcsw);
 981
 982         P(se.load.weight);
 983         P(se.runnable_weight);
 984 #ifdef CONFIG_SMP
 985         P(se.avg.load_sum);
 986         P(se.avg.runnable_load_sum);
 987         P(se.avg.util_sum);
 988         P(se.avg.load_avg);
 989         P(se.avg.runnable_load_avg);
 990         P(se.avg.util_avg);
 991         P(se.avg.last_update_time);
 992         P(se.avg.util_est.ewma);
 993         P(se.avg.util_est.enqueued);
 994 #endif
 995         P(policy);
 996         P(prio);
 997         if (p->policy == SCHED_DEADLINE) {
 998                 P(dl.runtime);
 999                 P(dl.deadline);
1000         }
1001 #undef PN_SCHEDSTAT
1002 #undef PN
1003 #undef __PN
1004 #undef P_SCHEDSTAT
1005 #undef P
1006 #undef __P
1007
1008         {
1009                 unsigned int this_cpu = raw_smp_processor_id();
1010                 u64 t0, t1;
1011
1012                 t0 = cpu_clock(this_cpu);
1013                 t1 = cpu_clock(this_cpu);
1014                 SEQ_printf(m, "%-45s:%21Ld\n",
1015                            "clock-delta", (long long)(t1-t0));
1016         }
1017
1018         sched_show_numa(p, m);
1019 }
1020
1021 void proc_sched_set_task(struct task_struct *p)
1022 {
1023 #ifdef CONFIG_SCHEDSTATS
1024         memset(&p->se.statistics, 0, sizeof(p->se.statistics));
1025 #endif
1026 }