kernel/sched/debug.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * kernel/sched/debug.c
   4  *
   5  * Print the CFS rbtree and other debugging details
   6  *
   7  * Copyright(C) 2007, Red Hat, Inc., Ingo Molnar
   8  */
   9 #include "sched.h"
  10
  11 static DEFINE_SPINLOCK(sched_debug_lock);
  12
  13 /*
  14  * This allows printing both to /proc/sched_debug and
  15  * to the console
  16  */
  17 #define SEQ_printf(m, x...)                     \
  18  do {                                           \
  19         if (m)                                  \
  20                 seq_printf(m, x);               \
  21         else                                    \
  22                 pr_cont(x);                     \
  23  } while (0)
  24
  25 /*
  26  * Ease the printing of nsec fields:
  27  */
  28 static long long nsec_high(unsigned long long nsec)
  29 {
  30         if ((long long)nsec < 0) {
  31                 nsec = -nsec;
  32                 do_div(nsec, 1000000);
  33                 return -nsec;
  34         }
  35         do_div(nsec, 1000000);
  36
  37         return nsec;
  38 }
  39
  40 static unsigned long nsec_low(unsigned long long nsec)
  41 {
  42         if ((long long)nsec < 0)
  43                 nsec = -nsec;
  44
  45         return do_div(nsec, 1000000);
  46 }
  47
  48 #define SPLIT_NS(x) nsec_high(x), nsec_low(x)
  49
  50 #define SCHED_FEAT(name, enabled)       \
  51         #name ,
  52
  53 static const char * const sched_feat_names[] = {
  54 #include "features.h"
  55 };
  56
  57 #undef SCHED_FEAT
  58
  59 static int sched_feat_show(struct seq_file *m, void *v)
  60 {
  61         int i;
  62
  63         for (i = 0; i < __SCHED_FEAT_NR; i++) {
  64                 if (!(sysctl_sched_features & (1UL << i)))
  65                         seq_puts(m, "NO_");
  66                 seq_printf(m, "%s ", sched_feat_names[i]);
  67         }
  68         seq_puts(m, "\n");
  69
  70         return 0;
  71 }
  72
  73 #ifdef CONFIG_JUMP_LABEL
  74
  75 #define jump_label_key__true  STATIC_KEY_INIT_TRUE
  76 #define jump_label_key__false STATIC_KEY_INIT_FALSE
  77
  78 #define SCHED_FEAT(name, enabled)       \
  79         jump_label_key__##enabled ,
  80
  81 struct static_key sched_feat_keys[__SCHED_FEAT_NR] = {
  82 #include "features.h"
  83 };
  84
  85 #undef SCHED_FEAT
  86
  87 static void sched_feat_disable(int i)
  88 {
  89         static_key_disable_cpuslocked(&sched_feat_keys[i]);
  90 }
  91
  92 static void sched_feat_enable(int i)
  93 {
  94         static_key_enable_cpuslocked(&sched_feat_keys[i]);
  95 }
  96 #else
  97 static void sched_feat_disable(int i) { };
  98 static void sched_feat_enable(int i) { };
  99 #endif /* CONFIG_JUMP_LABEL */
 100
 101 static int sched_feat_set(char *cmp)
 102 {
 103         int i;
 104         int neg = 0;
 105
 106         if (strncmp(cmp, "NO_", 3) == 0) {
 107                 neg = 1;
 108                 cmp += 3;
 109         }
 110
 111         i = match_string(sched_feat_names, __SCHED_FEAT_NR, cmp);
 112         if (i < 0)
 113                 return i;
 114
 115         if (neg) {
 116                 sysctl_sched_features &= ~(1UL << i);
 117                 sched_feat_disable(i);
 118         } else {
 119                 sysctl_sched_features |= (1UL << i);
 120                 sched_feat_enable(i);
 121         }
 122
 123         return 0;
 124 }
 125
 126 static ssize_t
 127 sched_feat_write(struct file *filp, const char __user *ubuf,
 128                 size_t cnt, loff_t *ppos)
 129 {
 130         char buf[64];
 131         char *cmp;
 132         int ret;
 133         struct inode *inode;
 134
 135         if (cnt > 63)
 136                 cnt = 63;
 137
 138         if (copy_from_user(&buf, ubuf, cnt))
 139                 return -EFAULT;
 140
 141         buf[cnt] = 0;
 142         cmp = strstrip(buf);
 143
 144         /* Ensure the static_key remains in a consistent state */
 145         inode = file_inode(filp);
 146         cpus_read_lock();
 147         inode_lock(inode);
 148         ret = sched_feat_set(cmp);
 149         inode_unlock(inode);
 150         cpus_read_unlock();
 151         if (ret < 0)
 152                 return ret;
 153
 154         *ppos += cnt;
 155
 156         return cnt;
 157 }
 158
 159 static int sched_feat_open(struct inode *inode, struct file *filp)
 160 {
 161         return single_open(filp, sched_feat_show, NULL);
 162 }
 163
 164 static const struct file_operations sched_feat_fops = {
 165         .open           = sched_feat_open,
 166         .write          = sched_feat_write,
 167         .read           = seq_read,
 168         .llseek         = seq_lseek,
 169         .release        = single_release,
 170 };
 171
 172 __read_mostly bool sched_debug_enabled;
 173
 174 static __init int sched_init_debug(void)
 175 {
 176         debugfs_create_file("sched_features", 0644, NULL, NULL,
 177                         &sched_feat_fops);
 178
 179         debugfs_create_bool("sched_debug", 0644, NULL,
 180                         &sched_debug_enabled);
 181
 182         return 0;
 183 }
 184 late_initcall(sched_init_debug);
 185
 186 #ifdef CONFIG_SMP
 187
 188 #ifdef CONFIG_SYSCTL
 189
 190 static struct ctl_table sd_ctl_dir[] = {
 191         {
 192                 .procname       = "sched_domain",
 193                 .mode           = 0555,
 194         },
 195         {}
 196 };
 197
 198 static struct ctl_table sd_ctl_root[] = {
 199         {
 200                 .procname       = "kernel",
 201                 .mode           = 0555,
 202                 .child          = sd_ctl_dir,
 203         },
 204         {}
 205 };
 206
 207 static struct ctl_table *sd_alloc_ctl_entry(int n)
 208 {
 209         struct ctl_table *entry =
 210                 kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL);
 211
 212         return entry;
 213 }
 214
 215 static void sd_free_ctl_entry(struct ctl_table **tablep)
 216 {
 217         struct ctl_table *entry;
 218
 219         /*
 220          * In the intermediate directories, both the child directory and
 221          * procname are dynamically allocated and could fail but the mode
 222          * will always be set. In the lowest directory the names are
 223          * static strings and all have proc handlers.
 224          */
 225         for (entry = *tablep; entry->mode; entry++) {
 226                 if (entry->child)
 227                         sd_free_ctl_entry(&entry->child);
 228                 if (entry->proc_handler == NULL)
 229                         kfree(entry->procname);
 230         }
 231
 232         kfree(*tablep);
 233         *tablep = NULL;
 234 }
 235
 236 static void
 237 set_table_entry(struct ctl_table *entry,
 238                 const char *procname, void *data, int maxlen,
 239                 umode_t mode, proc_handler *proc_handler)
 240 {
 241         entry->procname = procname;
 242         entry->data = data;
 243         entry->maxlen = maxlen;
 244         entry->mode = mode;
 245         entry->proc_handler = proc_handler;
 246 }
 247
 248 static struct ctl_table *
 249 sd_alloc_ctl_domain_table(struct sched_domain *sd)
 250 {
 251         struct ctl_table *table = sd_alloc_ctl_entry(9);
 252
 253         if (table == NULL)
 254                 return NULL;
 255
 256         set_table_entry(&table[0], "min_interval",        &sd->min_interval,        sizeof(long), 0644, proc_doulongvec_minmax);
 257         set_table_entry(&table[1], "max_interval",        &sd->max_interval,        sizeof(long), 0644, proc_doulongvec_minmax);
 258         set_table_entry(&table[2], "busy_factor",         &sd->busy_factor,         sizeof(int),  0644, proc_dointvec_minmax);
 259         set_table_entry(&table[3], "imbalance_pct",       &sd->imbalance_pct,       sizeof(int),  0644, proc_dointvec_minmax);
 260         set_table_entry(&table[4], "cache_nice_tries",    &sd->cache_nice_tries,    sizeof(int),  0644, proc_dointvec_minmax);
 261         set_table_entry(&table[5], "flags",               &sd->flags,               sizeof(int),  0644, proc_dointvec_minmax);
 262         set_table_entry(&table[6], "max_newidle_lb_cost", &sd->max_newidle_lb_cost, sizeof(long), 0644, proc_doulongvec_minmax);
 263         set_table_entry(&table[7], "name",                sd->name,            CORENAME_MAX_SIZE, 0444, proc_dostring);
 264         /* &table[8] is terminator */
 265
 266         return table;
 267 }
 268
 269 static struct ctl_table *sd_alloc_ctl_cpu_table(int cpu)
 270 {
 271         struct ctl_table *entry, *table;
 272         struct sched_domain *sd;
 273         int domain_num = 0, i;
 274         char buf[32];
 275
 276         for_each_domain(cpu, sd)
 277                 domain_num++;
 278         entry = table = sd_alloc_ctl_entry(domain_num + 1);
 279         if (table == NULL)
 280                 return NULL;
 281
 282         i = 0;
 283         for_each_domain(cpu, sd) {
 284                 snprintf(buf, 32, "domain%d", i);
 285                 entry->procname = kstrdup(buf, GFP_KERNEL);
 286                 entry->mode = 0555;
 287                 entry->child = sd_alloc_ctl_domain_table(sd);
 288                 entry++;
 289                 i++;
 290         }
 291         return table;
 292 }
 293
 294 static cpumask_var_t            sd_sysctl_cpus;
 295 static struct ctl_table_header  *sd_sysctl_header;
 296
 297 void register_sched_domain_sysctl(void)
 298 {
 299         static struct ctl_table *cpu_entries;
 300         static struct ctl_table **cpu_idx;
 301         static bool init_done = false;
 302         char buf[32];
 303         int i;
 304
 305         if (!cpu_entries) {
 306                 cpu_entries = sd_alloc_ctl_entry(num_possible_cpus() + 1);
 307                 if (!cpu_entries)
 308                         return;
 309
 310                 WARN_ON(sd_ctl_dir[0].child);
 311                 sd_ctl_dir[0].child = cpu_entries;
 312         }
 313
 314         if (!cpu_idx) {
 315                 struct ctl_table *e = cpu_entries;
 316
 317                 cpu_idx = kcalloc(nr_cpu_ids, sizeof(struct ctl_table*), GFP_KERNEL);
 318                 if (!cpu_idx)
 319                         return;
 320
 321                 /* deal with sparse possible map */
 322                 for_each_possible_cpu(i) {
 323                         cpu_idx[i] = e;
 324                         e++;
 325                 }
 326         }
 327
 328         if (!cpumask_available(sd_sysctl_cpus)) {
 329                 if (!alloc_cpumask_var(&sd_sysctl_cpus, GFP_KERNEL))
 330                         return;
 331         }
 332
 333         if (!init_done) {
 334                 init_done = true;
 335                 /* init to possible to not have holes in @cpu_entries */
 336                 cpumask_copy(sd_sysctl_cpus, cpu_possible_mask);
 337         }
 338
 339         for_each_cpu(i, sd_sysctl_cpus) {
 340                 struct ctl_table *e = cpu_idx[i];
 341
 342                 if (e->child)
 343                         sd_free_ctl_entry(&e->child);
 344
 345                 if (!e->procname) {
 346                         snprintf(buf, 32, "cpu%d", i);
 347                         e->procname = kstrdup(buf, GFP_KERNEL);
 348                 }
 349                 e->mode = 0555;
 350                 e->child = sd_alloc_ctl_cpu_table(i);
 351
 352                 __cpumask_clear_cpu(i, sd_sysctl_cpus);
 353         }
 354
 355         WARN_ON(sd_sysctl_header);
 356         sd_sysctl_header = register_sysctl_table(sd_ctl_root);
 357 }
 358
 359 void dirty_sched_domain_sysctl(int cpu)
 360 {
 361         if (cpumask_available(sd_sysctl_cpus))
 362                 __cpumask_set_cpu(cpu, sd_sysctl_cpus);
 363 }
 364
 365 /* may be called multiple times per register */
 366 void unregister_sched_domain_sysctl(void)
 367 {
 368         unregister_sysctl_table(sd_sysctl_header);
 369         sd_sysctl_header = NULL;
 370 }
 371 #endif /* CONFIG_SYSCTL */
 372 #endif /* CONFIG_SMP */
 373
 374 #ifdef CONFIG_FAIR_GROUP_SCHED
 375 static void print_cfs_group_stats(struct seq_file *m, int cpu, struct task_group *tg)
 376 {
 377         struct sched_entity *se = tg->se[cpu];
 378
 379 #define P(F)            SEQ_printf(m, "  .%-30s: %lld\n",       #F, (long long)F)
 380 #define P_SCHEDSTAT(F)  SEQ_printf(m, "  .%-30s: %lld\n",       #F, (long long)schedstat_val(F))
 381 #define PN(F)           SEQ_printf(m, "  .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F))
 382 #define PN_SCHEDSTAT(F) SEQ_printf(m, "  .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)schedstat_val(F)))
 383
 384         if (!se)
 385                 return;
 386
 387         PN(se->exec_start);
 388         PN(se->vruntime);
 389         PN(se->sum_exec_runtime);
 390
 391         if (schedstat_enabled()) {
 392                 PN_SCHEDSTAT(se->statistics.wait_start);
 393                 PN_SCHEDSTAT(se->statistics.sleep_start);
 394                 PN_SCHEDSTAT(se->statistics.block_start);
 395                 PN_SCHEDSTAT(se->statistics.sleep_max);
 396                 PN_SCHEDSTAT(se->statistics.block_max);
 397                 PN_SCHEDSTAT(se->statistics.exec_max);
 398                 PN_SCHEDSTAT(se->statistics.slice_max);
 399                 PN_SCHEDSTAT(se->statistics.wait_max);
 400                 PN_SCHEDSTAT(se->statistics.wait_sum);
 401                 P_SCHEDSTAT(se->statistics.wait_count);
 402         }
 403
 404         P(se->load.weight);
 405         P(se->runnable_weight);
 406 #ifdef CONFIG_SMP
 407         P(se->avg.load_avg);
 408         P(se->avg.util_avg);
 409         P(se->avg.runnable_load_avg);
 410 #endif
 411
 412 #undef PN_SCHEDSTAT
 413 #undef PN
 414 #undef P_SCHEDSTAT
 415 #undef P
 416 }
 417 #endif
 418
 419 #ifdef CONFIG_CGROUP_SCHED
 420 static char group_path[PATH_MAX];
 421
 422 static char *task_group_path(struct task_group *tg)
 423 {
 424         if (autogroup_path(tg, group_path, PATH_MAX))
 425                 return group_path;
 426
 427         cgroup_path(tg->css.cgroup, group_path, PATH_MAX);
 428
 429         return group_path;
 430 }
 431 #endif
 432
 433 static void
 434 print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
 435 {
 436         if (rq->curr == p)
 437                 SEQ_printf(m, ">R");
 438         else
 439                 SEQ_printf(m, " %c", task_state_to_char(p));
 440
 441         SEQ_printf(m, "%15s %5d %9Ld.%06ld %9Ld %5d ",
 442                 p->comm, task_pid_nr(p),
 443                 SPLIT_NS(p->se.vruntime),
 444                 (long long)(p->nvcsw + p->nivcsw),
 445                 p->prio);
 446
 447         SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
 448                 SPLIT_NS(schedstat_val_or_zero(p->se.statistics.wait_sum)),
 449                 SPLIT_NS(p->se.sum_exec_runtime),
 450                 SPLIT_NS(schedstat_val_or_zero(p->se.statistics.sum_sleep_runtime)));
 451
 452 #ifdef CONFIG_NUMA_BALANCING
 453         SEQ_printf(m, " %d %d", task_node(p), task_numa_group_id(p));
 454 #endif
 455 #ifdef CONFIG_CGROUP_SCHED
 456         SEQ_printf(m, " %s", task_group_path(task_group(p)));
 457 #endif
 458
 459         SEQ_printf(m, "\n");
 460 }
 461
 462 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
 463 {
 464         struct task_struct *g, *p;
 465
 466         SEQ_printf(m, "\n");
 467         SEQ_printf(m, "runnable tasks:\n");
 468         SEQ_printf(m, " S           task   PID         tree-key  switches  prio"
 469                    "     wait-time             sum-exec        sum-sleep\n");
 470         SEQ_printf(m, "-------------------------------------------------------"
 471                    "----------------------------------------------------\n");
 472
 473         rcu_read_lock();
 474         for_each_process_thread(g, p) {
 475                 if (task_cpu(p) != rq_cpu)
 476                         continue;
 477
 478                 print_task(m, rq, p);
 479         }
 480         rcu_read_unlock();
 481 }
 482
 483 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
 484 {
 485         s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
 486                 spread, rq0_min_vruntime, spread0;
 487         struct rq *rq = cpu_rq(cpu);
 488         struct sched_entity *last;
 489         unsigned long flags;
 490
 491 #ifdef CONFIG_FAIR_GROUP_SCHED
 492         SEQ_printf(m, "\n");
 493         SEQ_printf(m, "cfs_rq[%d]:%s\n", cpu, task_group_path(cfs_rq->tg));
 494 #else
 495         SEQ_printf(m, "\n");
 496         SEQ_printf(m, "cfs_rq[%d]:\n", cpu);
 497 #endif
 498         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "exec_clock",
 499                         SPLIT_NS(cfs_rq->exec_clock));
 500
 501         raw_spin_lock_irqsave(&rq->lock, flags);
 502         if (rb_first_cached(&cfs_rq->tasks_timeline))
 503                 MIN_vruntime = (__pick_first_entity(cfs_rq))->vruntime;
 504         last = __pick_last_entity(cfs_rq);
 505         if (last)
 506                 max_vruntime = last->vruntime;
 507         min_vruntime = cfs_rq->min_vruntime;
 508         rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
 509         raw_spin_unlock_irqrestore(&rq->lock, flags);
 510         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "MIN_vruntime",
 511                         SPLIT_NS(MIN_vruntime));
 512         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "min_vruntime",
 513                         SPLIT_NS(min_vruntime));
 514         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "max_vruntime",
 515                         SPLIT_NS(max_vruntime));
 516         spread = max_vruntime - MIN_vruntime;
 517         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread",
 518                         SPLIT_NS(spread));
 519         spread0 = min_vruntime - rq0_min_vruntime;
 520         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread0",
 521                         SPLIT_NS(spread0));
 522         SEQ_printf(m, "  .%-30s: %d\n", "nr_spread_over",
 523                         cfs_rq->nr_spread_over);
 524         SEQ_printf(m, "  .%-30s: %d\n", "nr_running", cfs_rq->nr_running);
 525         SEQ_printf(m, "  .%-30s: %ld\n", "load", cfs_rq->load.weight);
 526 #ifdef CONFIG_SMP
 527         SEQ_printf(m, "  .%-30s: %ld\n", "runnable_weight", cfs_rq->runnable_weight);
 528         SEQ_printf(m, "  .%-30s: %lu\n", "load_avg",
 529                         cfs_rq->avg.load_avg);
 530         SEQ_printf(m, "  .%-30s: %lu\n", "runnable_load_avg",
 531                         cfs_rq->avg.runnable_load_avg);
 532         SEQ_printf(m, "  .%-30s: %lu\n", "util_avg",
 533                         cfs_rq->avg.util_avg);
 534         SEQ_printf(m, "  .%-30s: %u\n", "util_est_enqueued",
 535                         cfs_rq->avg.util_est.enqueued);
 536         SEQ_printf(m, "  .%-30s: %ld\n", "removed.load_avg",
 537                         cfs_rq->removed.load_avg);
 538         SEQ_printf(m, "  .%-30s: %ld\n", "removed.util_avg",
 539                         cfs_rq->removed.util_avg);
 540         SEQ_printf(m, "  .%-30s: %ld\n", "removed.runnable_sum",
 541                         cfs_rq->removed.runnable_sum);
 542 #ifdef CONFIG_FAIR_GROUP_SCHED
 543         SEQ_printf(m, "  .%-30s: %lu\n", "tg_load_avg_contrib",
 544                         cfs_rq->tg_load_avg_contrib);
 545         SEQ_printf(m, "  .%-30s: %ld\n", "tg_load_avg",
 546                         atomic_long_read(&cfs_rq->tg->load_avg));
 547 #endif
 548 #endif
 549 #ifdef CONFIG_CFS_BANDWIDTH
 550         SEQ_printf(m, "  .%-30s: %d\n", "throttled",
 551                         cfs_rq->throttled);
 552         SEQ_printf(m, "  .%-30s: %d\n", "throttle_count",
 553                         cfs_rq->throttle_count);
 554 #endif
 555
 556 #ifdef CONFIG_FAIR_GROUP_SCHED
 557         print_cfs_group_stats(m, cpu, cfs_rq->tg);
 558 #endif
 559 }
 560
 561 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
 562 {
 563 #ifdef CONFIG_RT_GROUP_SCHED
 564         SEQ_printf(m, "\n");
 565         SEQ_printf(m, "rt_rq[%d]:%s\n", cpu, task_group_path(rt_rq->tg));
 566 #else
 567         SEQ_printf(m, "\n");
 568         SEQ_printf(m, "rt_rq[%d]:\n", cpu);
 569 #endif
 570
 571 #define P(x) \
 572         SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
 573 #define PU(x) \
 574         SEQ_printf(m, "  .%-30s: %lu\n", #x, (unsigned long)(rt_rq->x))
 575 #define PN(x) \
 576         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
 577
 578         PU(rt_nr_running);
 579 #ifdef CONFIG_SMP
 580         PU(rt_nr_migratory);
 581 #endif
 582         P(rt_throttled);
 583         PN(rt_time);
 584         PN(rt_runtime);
 585
 586 #undef PN
 587 #undef PU
 588 #undef P
 589 }
 590
 591 void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq)
 592 {
 593         struct dl_bw *dl_bw;
 594
 595         SEQ_printf(m, "\n");
 596         SEQ_printf(m, "dl_rq[%d]:\n", cpu);
 597
 598 #define PU(x) \
 599         SEQ_printf(m, "  .%-30s: %lu\n", #x, (unsigned long)(dl_rq->x))
 600
 601         PU(dl_nr_running);
 602 #ifdef CONFIG_SMP
 603         PU(dl_nr_migratory);
 604         dl_bw = &cpu_rq(cpu)->rd->dl_bw;
 605 #else
 606         dl_bw = &dl_rq->dl_bw;
 607 #endif
 608         SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->bw", dl_bw->bw);
 609         SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->total_bw", dl_bw->total_bw);
 610
 611 #undef PU
 612 }
 613
 614 static void print_cpu(struct seq_file *m, int cpu)
 615 {
 616         struct rq *rq = cpu_rq(cpu);
 617         unsigned long flags;
 618
 619 #ifdef CONFIG_X86
 620         {
 621                 unsigned int freq = cpu_khz ? : 1;
 622
 623                 SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
 624                            cpu, freq / 1000, (freq % 1000));
 625         }
 626 #else
 627         SEQ_printf(m, "cpu#%d\n", cpu);
 628 #endif
 629
 630 #define P(x)                                                            \
 631 do {                                                                    \
 632         if (sizeof(rq->x) == 4)                                         \
 633                 SEQ_printf(m, "  .%-30s: %ld\n", #x, (long)(rq->x));    \
 634         else                                                            \
 635                 SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rq->x));\
 636 } while (0)
 637
 638 #define PN(x) \
 639         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
 640
 641         P(nr_running);
 642         P(nr_switches);
 643         P(nr_load_updates);
 644         P(nr_uninterruptible);
 645         PN(next_balance);
 646         SEQ_printf(m, "  .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr)));
 647         PN(clock);
 648         PN(clock_task);
 649 #undef P
 650 #undef PN
 651
 652 #ifdef CONFIG_SMP
 653 #define P64(n) SEQ_printf(m, "  .%-30s: %Ld\n", #n, rq->n);
 654         P64(avg_idle);
 655         P64(max_idle_balance_cost);
 656 #undef P64
 657 #endif
 658
 659 #define P(n) SEQ_printf(m, "  .%-30s: %d\n", #n, schedstat_val(rq->n));
 660         if (schedstat_enabled()) {
 661                 P(yld_count);
 662                 P(sched_count);
 663                 P(sched_goidle);
 664                 P(ttwu_count);
 665                 P(ttwu_local);
 666         }
 667 #undef P
 668
 669         spin_lock_irqsave(&sched_debug_lock, flags);
 670         print_cfs_stats(m, cpu);
 671         print_rt_stats(m, cpu);
 672         print_dl_stats(m, cpu);
 673
 674         print_rq(m, rq, cpu);
 675         spin_unlock_irqrestore(&sched_debug_lock, flags);
 676         SEQ_printf(m, "\n");
 677 }
 678
 679 static const char *sched_tunable_scaling_names[] = {
 680         "none",
 681         "logarithmic",
 682         "linear"
 683 };
 684
 685 static void sched_debug_header(struct seq_file *m)
 686 {
 687         u64 ktime, sched_clk, cpu_clk;
 688         unsigned long flags;
 689
 690         local_irq_save(flags);
 691         ktime = ktime_to_ns(ktime_get());
 692         sched_clk = sched_clock();
 693         cpu_clk = local_clock();
 694         local_irq_restore(flags);
 695
 696         SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n",
 697                 init_utsname()->release,
 698                 (int)strcspn(init_utsname()->version, " "),
 699                 init_utsname()->version);
 700
 701 #define P(x) \
 702         SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
 703 #define PN(x) \
 704         SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
 705         PN(ktime);
 706         PN(sched_clk);
 707         PN(cpu_clk);
 708         P(jiffies);
 709 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
 710         P(sched_clock_stable());
 711 #endif
 712 #undef PN
 713 #undef P
 714
 715         SEQ_printf(m, "\n");
 716         SEQ_printf(m, "sysctl_sched\n");
 717
 718 #define P(x) \
 719         SEQ_printf(m, "  .%-40s: %Ld\n", #x, (long long)(x))
 720 #define PN(x) \
 721         SEQ_printf(m, "  .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
 722         PN(sysctl_sched_latency);
 723         PN(sysctl_sched_min_granularity);
 724         PN(sysctl_sched_wakeup_granularity);
 725         P(sysctl_sched_child_runs_first);
 726         P(sysctl_sched_features);
 727 #undef PN
 728 #undef P
 729
 730         SEQ_printf(m, "  .%-40s: %d (%s)\n",
 731                 "sysctl_sched_tunable_scaling",
 732                 sysctl_sched_tunable_scaling,
 733                 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
 734         SEQ_printf(m, "\n");
 735 }
 736
 737 static int sched_debug_show(struct seq_file *m, void *v)
 738 {
 739         int cpu = (unsigned long)(v - 2);
 740
 741         if (cpu != -1)
 742                 print_cpu(m, cpu);
 743         else
 744                 sched_debug_header(m);
 745
 746         return 0;
 747 }
 748
 749 void sysrq_sched_debug_show(void)
 750 {
 751         int cpu;
 752
 753         sched_debug_header(NULL);
 754         for_each_online_cpu(cpu) {
 755                 /*
 756                  * Need to reset softlockup watchdogs on all CPUs, because
 757                  * another CPU might be blocked waiting for us to process
 758                  * an IPI or stop_machine.
 759                  */
 760                 touch_nmi_watchdog();
 761                 touch_all_softlockup_watchdogs();
 762                 print_cpu(NULL, cpu);
 763         }
 764 }
 765
 766 /*
 767  * This itererator needs some explanation.
 768  * It returns 1 for the header position.
 769  * This means 2 is CPU 0.
 770  * In a hotplugged system some CPUs, including CPU 0, may be missing so we have
 771  * to use cpumask_* to iterate over the CPUs.
 772  */
 773 static void *sched_debug_start(struct seq_file *file, loff_t *offset)
 774 {
 775         unsigned long n = *offset;
 776
 777         if (n == 0)
 778                 return (void *) 1;
 779
 780         n--;
 781
 782         if (n > 0)
 783                 n = cpumask_next(n - 1, cpu_online_mask);
 784         else
 785                 n = cpumask_first(cpu_online_mask);
 786
 787         *offset = n + 1;
 788
 789         if (n < nr_cpu_ids)
 790                 return (void *)(unsigned long)(n + 2);
 791
 792         return NULL;
 793 }
 794
 795 static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
 796 {
 797         (*offset)++;
 798         return sched_debug_start(file, offset);
 799 }
 800
 801 static void sched_debug_stop(struct seq_file *file, void *data)
 802 {
 803 }
 804
 805 static const struct seq_operations sched_debug_sops = {
 806         .start          = sched_debug_start,
 807         .next           = sched_debug_next,
 808         .stop           = sched_debug_stop,
 809         .show           = sched_debug_show,
 810 };
 811
 812 static int __init init_sched_debug_procfs(void)
 813 {
 814         if (!proc_create_seq("sched_debug", 0444, NULL, &sched_debug_sops))
 815                 return -ENOMEM;
 816         return 0;
 817 }
 818
 819 __initcall(init_sched_debug_procfs);
 820
 821 #define __P(F)  SEQ_printf(m, "%-45s:%21Ld\n",       #F, (long long)F)
 822 #define   P(F)  SEQ_printf(m, "%-45s:%21Ld\n",       #F, (long long)p->F)
 823 #define __PN(F) SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
 824 #define   PN(F) SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
 825
 826
 827 #ifdef CONFIG_NUMA_BALANCING
 828 void print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
 829                 unsigned long tpf, unsigned long gsf, unsigned long gpf)
 830 {
 831         SEQ_printf(m, "numa_faults node=%d ", node);
 832         SEQ_printf(m, "task_private=%lu task_shared=%lu ", tpf, tsf);
 833         SEQ_printf(m, "group_private=%lu group_shared=%lu\n", gpf, gsf);
 834 }
 835 #endif
 836
 837
 838 static void sched_show_numa(struct task_struct *p, struct seq_file *m)
 839 {
 840 #ifdef CONFIG_NUMA_BALANCING
 841         struct mempolicy *pol;
 842
 843         if (p->mm)
 844                 P(mm->numa_scan_seq);
 845
 846         task_lock(p);
 847         pol = p->mempolicy;
 848         if (pol && !(pol->flags & MPOL_F_MORON))
 849                 pol = NULL;
 850         mpol_get(pol);
 851         task_unlock(p);
 852
 853         P(numa_pages_migrated);
 854         P(numa_preferred_nid);
 855         P(total_numa_faults);
 856         SEQ_printf(m, "current_node=%d, numa_group_id=%d\n",
 857                         task_node(p), task_numa_group_id(p));
 858         show_numa_stats(p, m);
 859         mpol_put(pol);
 860 #endif
 861 }
 862
 863 void proc_sched_show_task(struct task_struct *p, struct pid_namespace *ns,
 864                                                   struct seq_file *m)
 865 {
 866         unsigned long nr_switches;
 867
 868         SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr_ns(p, ns),
 869                                                 get_nr_threads(p));
 870         SEQ_printf(m,
 871                 "---------------------------------------------------------"
 872                 "----------\n");
 873 #define __P(F) \
 874         SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)F)
 875 #define P(F) \
 876         SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)p->F)
 877 #define P_SCHEDSTAT(F) \
 878         SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)schedstat_val(p->F))
 879 #define __PN(F) \
 880         SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
 881 #define PN(F) \
 882         SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
 883 #define PN_SCHEDSTAT(F) \
 884         SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)schedstat_val(p->F)))
 885
 886         PN(se.exec_start);
 887         PN(se.vruntime);
 888         PN(se.sum_exec_runtime);
 889
 890         nr_switches = p->nvcsw + p->nivcsw;
 891
 892         P(se.nr_migrations);
 893
 894         if (schedstat_enabled()) {
 895                 u64 avg_atom, avg_per_cpu;
 896
 897                 PN_SCHEDSTAT(se.statistics.sum_sleep_runtime);
 898                 PN_SCHEDSTAT(se.statistics.wait_start);
 899                 PN_SCHEDSTAT(se.statistics.sleep_start);
 900                 PN_SCHEDSTAT(se.statistics.block_start);
 901                 PN_SCHEDSTAT(se.statistics.sleep_max);
 902                 PN_SCHEDSTAT(se.statistics.block_max);
 903                 PN_SCHEDSTAT(se.statistics.exec_max);
 904                 PN_SCHEDSTAT(se.statistics.slice_max);
 905                 PN_SCHEDSTAT(se.statistics.wait_max);
 906                 PN_SCHEDSTAT(se.statistics.wait_sum);
 907                 P_SCHEDSTAT(se.statistics.wait_count);
 908                 PN_SCHEDSTAT(se.statistics.iowait_sum);
 909                 P_SCHEDSTAT(se.statistics.iowait_count);
 910                 P_SCHEDSTAT(se.statistics.nr_migrations_cold);
 911                 P_SCHEDSTAT(se.statistics.nr_failed_migrations_affine);
 912                 P_SCHEDSTAT(se.statistics.nr_failed_migrations_running);
 913                 P_SCHEDSTAT(se.statistics.nr_failed_migrations_hot);
 914                 P_SCHEDSTAT(se.statistics.nr_forced_migrations);
 915                 P_SCHEDSTAT(se.statistics.nr_wakeups);
 916                 P_SCHEDSTAT(se.statistics.nr_wakeups_sync);
 917                 P_SCHEDSTAT(se.statistics.nr_wakeups_migrate);
 918                 P_SCHEDSTAT(se.statistics.nr_wakeups_local);
 919                 P_SCHEDSTAT(se.statistics.nr_wakeups_remote);
 920                 P_SCHEDSTAT(se.statistics.nr_wakeups_affine);
 921                 P_SCHEDSTAT(se.statistics.nr_wakeups_affine_attempts);
 922                 P_SCHEDSTAT(se.statistics.nr_wakeups_passive);
 923                 P_SCHEDSTAT(se.statistics.nr_wakeups_idle);
 924
 925                 avg_atom = p->se.sum_exec_runtime;
 926                 if (nr_switches)
 927                         avg_atom = div64_ul(avg_atom, nr_switches);
 928                 else
 929                         avg_atom = -1LL;
 930
 931                 avg_per_cpu = p->se.sum_exec_runtime;
 932                 if (p->se.nr_migrations) {
 933                         avg_per_cpu = div64_u64(avg_per_cpu,
 934                                                 p->se.nr_migrations);
 935                 } else {
 936                         avg_per_cpu = -1LL;
 937                 }
 938
 939                 __PN(avg_atom);
 940                 __PN(avg_per_cpu);
 941         }
 942
 943         __P(nr_switches);
 944         SEQ_printf(m, "%-45s:%21Ld\n",
 945                    "nr_voluntary_switches", (long long)p->nvcsw);
 946         SEQ_printf(m, "%-45s:%21Ld\n",
 947                    "nr_involuntary_switches", (long long)p->nivcsw);
 948
 949         P(se.load.weight);
 950         P(se.runnable_weight);
 951 #ifdef CONFIG_SMP
 952         P(se.avg.load_sum);
 953         P(se.avg.runnable_load_sum);
 954         P(se.avg.util_sum);
 955         P(se.avg.load_avg);
 956         P(se.avg.runnable_load_avg);
 957         P(se.avg.util_avg);
 958         P(se.avg.last_update_time);
 959         P(se.avg.util_est.ewma);
 960         P(se.avg.util_est.enqueued);
 961 #endif
 962         P(policy);
 963         P(prio);
 964         if (task_has_dl_policy(p)) {
 965                 P(dl.runtime);
 966                 P(dl.deadline);
 967         }
 968 #undef PN_SCHEDSTAT
 969 #undef PN
 970 #undef __PN
 971 #undef P_SCHEDSTAT
 972 #undef P
 973 #undef __P
 974
 975         {
 976                 unsigned int this_cpu = raw_smp_processor_id();
 977                 u64 t0, t1;
 978
 979                 t0 = cpu_clock(this_cpu);
 980                 t1 = cpu_clock(this_cpu);
 981                 SEQ_printf(m, "%-45s:%21Ld\n",
 982                            "clock-delta", (long long)(t1-t0));
 983         }
 984
 985         sched_show_numa(p, m);
 986 }
 987
 988 void proc_sched_set_task(struct task_struct *p)
 989 {
 990 #ifdef CONFIG_SCHEDSTATS
 991         memset(&p->se.statistics, 0, sizeof(p->se.statistics));
 992 #endif
 993 }