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