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 int sd_ctl_doflags(struct ctl_table *table, int write,
 249                           void *buffer, size_t *lenp, loff_t *ppos)
 250 {
 251         unsigned long flags = *(unsigned long *)table->data;
 252         size_t data_size = 0;
 253         size_t len = 0;
 254         char *tmp, *buf;
 255         int idx;
 256
 257         if (write)
 258                 return 0;
 259
 260         for_each_set_bit(idx, &flags, __SD_FLAG_CNT) {
 261                 char *name = sd_flag_debug[idx].name;
 262
 263                 /* Name plus whitespace */
 264                 data_size += strlen(name) + 1;
 265         }
 266
 267         if (*ppos > data_size) {
 268                 *lenp = 0;
 269                 return 0;
 270         }
 271
 272         buf = kcalloc(data_size + 1, sizeof(*buf), GFP_KERNEL);
 273         if (!buf)
 274                 return -ENOMEM;
 275
 276         for_each_set_bit(idx, &flags, __SD_FLAG_CNT) {
 277                 char *name = sd_flag_debug[idx].name;
 278
 279                 len += snprintf(buf + len, strlen(name) + 2, "%s ", name);
 280         }
 281
 282         tmp = buf + *ppos;
 283         len -= *ppos;
 284
 285         if (len > *lenp)
 286                 len = *lenp;
 287         if (len)
 288                 memcpy(buffer, tmp, len);
 289         if (len < *lenp) {
 290                 ((char *)buffer)[len] = '\n';
 291                 len++;
 292         }
 293
 294         *lenp = len;
 295         *ppos += len;
 296
 297         kfree(buf);
 298
 299         return 0;
 300 }
 301
 302 static struct ctl_table *
 303 sd_alloc_ctl_domain_table(struct sched_domain *sd)
 304 {
 305         struct ctl_table *table = sd_alloc_ctl_entry(9);
 306
 307         if (table == NULL)
 308                 return NULL;
 309
 310         set_table_entry(&table[0], "min_interval",        &sd->min_interval,        sizeof(long), 0644, proc_doulongvec_minmax);
 311         set_table_entry(&table[1], "max_interval",        &sd->max_interval,        sizeof(long), 0644, proc_doulongvec_minmax);
 312         set_table_entry(&table[2], "busy_factor",         &sd->busy_factor,         sizeof(int),  0644, proc_dointvec_minmax);
 313         set_table_entry(&table[3], "imbalance_pct",       &sd->imbalance_pct,       sizeof(int),  0644, proc_dointvec_minmax);
 314         set_table_entry(&table[4], "cache_nice_tries",    &sd->cache_nice_tries,    sizeof(int),  0644, proc_dointvec_minmax);
 315         set_table_entry(&table[5], "flags",               &sd->flags,               sizeof(int),  0444, sd_ctl_doflags);
 316         set_table_entry(&table[6], "max_newidle_lb_cost", &sd->max_newidle_lb_cost, sizeof(long), 0644, proc_doulongvec_minmax);
 317         set_table_entry(&table[7], "name",                sd->name,            CORENAME_MAX_SIZE, 0444, proc_dostring);
 318         /* &table[8] is terminator */
 319
 320         return table;
 321 }
 322
 323 static struct ctl_table *sd_alloc_ctl_cpu_table(int cpu)
 324 {
 325         struct ctl_table *entry, *table;
 326         struct sched_domain *sd;
 327         int domain_num = 0, i;
 328         char buf[32];
 329
 330         for_each_domain(cpu, sd)
 331                 domain_num++;
 332         entry = table = sd_alloc_ctl_entry(domain_num + 1);
 333         if (table == NULL)
 334                 return NULL;
 335
 336         i = 0;
 337         for_each_domain(cpu, sd) {
 338                 snprintf(buf, 32, "domain%d", i);
 339                 entry->procname = kstrdup(buf, GFP_KERNEL);
 340                 entry->mode = 0555;
 341                 entry->child = sd_alloc_ctl_domain_table(sd);
 342                 entry++;
 343                 i++;
 344         }
 345         return table;
 346 }
 347
 348 static cpumask_var_t            sd_sysctl_cpus;
 349 static struct ctl_table_header  *sd_sysctl_header;
 350
 351 void register_sched_domain_sysctl(void)
 352 {
 353         static struct ctl_table *cpu_entries;
 354         static struct ctl_table **cpu_idx;
 355         static bool init_done = false;
 356         char buf[32];
 357         int i;
 358
 359         if (!cpu_entries) {
 360                 cpu_entries = sd_alloc_ctl_entry(num_possible_cpus() + 1);
 361                 if (!cpu_entries)
 362                         return;
 363
 364                 WARN_ON(sd_ctl_dir[0].child);
 365                 sd_ctl_dir[0].child = cpu_entries;
 366         }
 367
 368         if (!cpu_idx) {
 369                 struct ctl_table *e = cpu_entries;
 370
 371                 cpu_idx = kcalloc(nr_cpu_ids, sizeof(struct ctl_table*), GFP_KERNEL);
 372                 if (!cpu_idx)
 373                         return;
 374
 375                 /* deal with sparse possible map */
 376                 for_each_possible_cpu(i) {
 377                         cpu_idx[i] = e;
 378                         e++;
 379                 }
 380         }
 381
 382         if (!cpumask_available(sd_sysctl_cpus)) {
 383                 if (!alloc_cpumask_var(&sd_sysctl_cpus, GFP_KERNEL))
 384                         return;
 385         }
 386
 387         if (!init_done) {
 388                 init_done = true;
 389                 /* init to possible to not have holes in @cpu_entries */
 390                 cpumask_copy(sd_sysctl_cpus, cpu_possible_mask);
 391         }
 392
 393         for_each_cpu(i, sd_sysctl_cpus) {
 394                 struct ctl_table *e = cpu_idx[i];
 395
 396                 if (e->child)
 397                         sd_free_ctl_entry(&e->child);
 398
 399                 if (!e->procname) {
 400                         snprintf(buf, 32, "cpu%d", i);
 401                         e->procname = kstrdup(buf, GFP_KERNEL);
 402                 }
 403                 e->mode = 0555;
 404                 e->child = sd_alloc_ctl_cpu_table(i);
 405
 406                 __cpumask_clear_cpu(i, sd_sysctl_cpus);
 407         }
 408
 409         WARN_ON(sd_sysctl_header);
 410         sd_sysctl_header = register_sysctl_table(sd_ctl_root);
 411 }
 412
 413 void dirty_sched_domain_sysctl(int cpu)
 414 {
 415         if (cpumask_available(sd_sysctl_cpus))
 416                 __cpumask_set_cpu(cpu, sd_sysctl_cpus);
 417 }
 418
 419 /* may be called multiple times per register */
 420 void unregister_sched_domain_sysctl(void)
 421 {
 422         unregister_sysctl_table(sd_sysctl_header);
 423         sd_sysctl_header = NULL;
 424 }
 425 #endif /* CONFIG_SYSCTL */
 426 #endif /* CONFIG_SMP */
 427
 428 #ifdef CONFIG_FAIR_GROUP_SCHED
 429 static void print_cfs_group_stats(struct seq_file *m, int cpu, struct task_group *tg)
 430 {
 431         struct sched_entity *se = tg->se[cpu];
 432
 433 #define P(F)            SEQ_printf(m, "  .%-30s: %lld\n",       #F, (long long)F)
 434 #define P_SCHEDSTAT(F)  SEQ_printf(m, "  .%-30s: %lld\n",       #F, (long long)schedstat_val(F))
 435 #define PN(F)           SEQ_printf(m, "  .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F))
 436 #define PN_SCHEDSTAT(F) SEQ_printf(m, "  .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)schedstat_val(F)))
 437
 438         if (!se)
 439                 return;
 440
 441         PN(se->exec_start);
 442         PN(se->vruntime);
 443         PN(se->sum_exec_runtime);
 444
 445         if (schedstat_enabled()) {
 446                 PN_SCHEDSTAT(se->statistics.wait_start);
 447                 PN_SCHEDSTAT(se->statistics.sleep_start);
 448                 PN_SCHEDSTAT(se->statistics.block_start);
 449                 PN_SCHEDSTAT(se->statistics.sleep_max);
 450                 PN_SCHEDSTAT(se->statistics.block_max);
 451                 PN_SCHEDSTAT(se->statistics.exec_max);
 452                 PN_SCHEDSTAT(se->statistics.slice_max);
 453                 PN_SCHEDSTAT(se->statistics.wait_max);
 454                 PN_SCHEDSTAT(se->statistics.wait_sum);
 455                 P_SCHEDSTAT(se->statistics.wait_count);
 456         }
 457
 458         P(se->load.weight);
 459 #ifdef CONFIG_SMP
 460         P(se->avg.load_avg);
 461         P(se->avg.util_avg);
 462         P(se->avg.runnable_avg);
 463 #endif
 464
 465 #undef PN_SCHEDSTAT
 466 #undef PN
 467 #undef P_SCHEDSTAT
 468 #undef P
 469 }
 470 #endif
 471
 472 #ifdef CONFIG_CGROUP_SCHED
 473 static char group_path[PATH_MAX];
 474
 475 static char *task_group_path(struct task_group *tg)
 476 {
 477         if (autogroup_path(tg, group_path, PATH_MAX))
 478                 return group_path;
 479
 480         cgroup_path(tg->css.cgroup, group_path, PATH_MAX);
 481
 482         return group_path;
 483 }
 484 #endif
 485
 486 static void
 487 print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
 488 {
 489         if (task_current(rq, p))
 490                 SEQ_printf(m, ">R");
 491         else
 492                 SEQ_printf(m, " %c", task_state_to_char(p));
 493
 494         SEQ_printf(m, " %15s %5d %9Ld.%06ld %9Ld %5d ",
 495                 p->comm, task_pid_nr(p),
 496                 SPLIT_NS(p->se.vruntime),
 497                 (long long)(p->nvcsw + p->nivcsw),
 498                 p->prio);
 499
 500         SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
 501                 SPLIT_NS(schedstat_val_or_zero(p->se.statistics.wait_sum)),
 502                 SPLIT_NS(p->se.sum_exec_runtime),
 503                 SPLIT_NS(schedstat_val_or_zero(p->se.statistics.sum_sleep_runtime)));
 504
 505 #ifdef CONFIG_NUMA_BALANCING
 506         SEQ_printf(m, " %d %d", task_node(p), task_numa_group_id(p));
 507 #endif
 508 #ifdef CONFIG_CGROUP_SCHED
 509         SEQ_printf(m, " %s", task_group_path(task_group(p)));
 510 #endif
 511
 512         SEQ_printf(m, "\n");
 513 }
 514
 515 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
 516 {
 517         struct task_struct *g, *p;
 518
 519         SEQ_printf(m, "\n");
 520         SEQ_printf(m, "runnable tasks:\n");
 521         SEQ_printf(m, " S            task   PID         tree-key  switches  prio"
 522                    "     wait-time             sum-exec        sum-sleep\n");
 523         SEQ_printf(m, "-------------------------------------------------------"
 524                    "------------------------------------------------------\n");
 525
 526         rcu_read_lock();
 527         for_each_process_thread(g, p) {
 528                 if (task_cpu(p) != rq_cpu)
 529                         continue;
 530
 531                 print_task(m, rq, p);
 532         }
 533         rcu_read_unlock();
 534 }
 535
 536 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
 537 {
 538         s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
 539                 spread, rq0_min_vruntime, spread0;
 540         struct rq *rq = cpu_rq(cpu);
 541         struct sched_entity *last;
 542         unsigned long flags;
 543
 544 #ifdef CONFIG_FAIR_GROUP_SCHED
 545         SEQ_printf(m, "\n");
 546         SEQ_printf(m, "cfs_rq[%d]:%s\n", cpu, task_group_path(cfs_rq->tg));
 547 #else
 548         SEQ_printf(m, "\n");
 549         SEQ_printf(m, "cfs_rq[%d]:\n", cpu);
 550 #endif
 551         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "exec_clock",
 552                         SPLIT_NS(cfs_rq->exec_clock));
 553
 554         raw_spin_lock_irqsave(&rq->lock, flags);
 555         if (rb_first_cached(&cfs_rq->tasks_timeline))
 556                 MIN_vruntime = (__pick_first_entity(cfs_rq))->vruntime;
 557         last = __pick_last_entity(cfs_rq);
 558         if (last)
 559                 max_vruntime = last->vruntime;
 560         min_vruntime = cfs_rq->min_vruntime;
 561         rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
 562         raw_spin_unlock_irqrestore(&rq->lock, flags);
 563         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "MIN_vruntime",
 564                         SPLIT_NS(MIN_vruntime));
 565         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "min_vruntime",
 566                         SPLIT_NS(min_vruntime));
 567         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "max_vruntime",
 568                         SPLIT_NS(max_vruntime));
 569         spread = max_vruntime - MIN_vruntime;
 570         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread",
 571                         SPLIT_NS(spread));
 572         spread0 = min_vruntime - rq0_min_vruntime;
 573         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread0",
 574                         SPLIT_NS(spread0));
 575         SEQ_printf(m, "  .%-30s: %d\n", "nr_spread_over",
 576                         cfs_rq->nr_spread_over);
 577         SEQ_printf(m, "  .%-30s: %d\n", "nr_running", cfs_rq->nr_running);
 578         SEQ_printf(m, "  .%-30s: %ld\n", "load", cfs_rq->load.weight);
 579 #ifdef CONFIG_SMP
 580         SEQ_printf(m, "  .%-30s: %lu\n", "load_avg",
 581                         cfs_rq->avg.load_avg);
 582         SEQ_printf(m, "  .%-30s: %lu\n", "runnable_avg",
 583                         cfs_rq->avg.runnable_avg);
 584         SEQ_printf(m, "  .%-30s: %lu\n", "util_avg",
 585                         cfs_rq->avg.util_avg);
 586         SEQ_printf(m, "  .%-30s: %u\n", "util_est_enqueued",
 587                         cfs_rq->avg.util_est.enqueued);
 588         SEQ_printf(m, "  .%-30s: %ld\n", "removed.load_avg",
 589                         cfs_rq->removed.load_avg);
 590         SEQ_printf(m, "  .%-30s: %ld\n", "removed.util_avg",
 591                         cfs_rq->removed.util_avg);
 592         SEQ_printf(m, "  .%-30s: %ld\n", "removed.runnable_avg",
 593                         cfs_rq->removed.runnable_avg);
 594 #ifdef CONFIG_FAIR_GROUP_SCHED
 595         SEQ_printf(m, "  .%-30s: %lu\n", "tg_load_avg_contrib",
 596                         cfs_rq->tg_load_avg_contrib);
 597         SEQ_printf(m, "  .%-30s: %ld\n", "tg_load_avg",
 598                         atomic_long_read(&cfs_rq->tg->load_avg));
 599 #endif
 600 #endif
 601 #ifdef CONFIG_CFS_BANDWIDTH
 602         SEQ_printf(m, "  .%-30s: %d\n", "throttled",
 603                         cfs_rq->throttled);
 604         SEQ_printf(m, "  .%-30s: %d\n", "throttle_count",
 605                         cfs_rq->throttle_count);
 606 #endif
 607
 608 #ifdef CONFIG_FAIR_GROUP_SCHED
 609         print_cfs_group_stats(m, cpu, cfs_rq->tg);
 610 #endif
 611 }
 612
 613 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
 614 {
 615 #ifdef CONFIG_RT_GROUP_SCHED
 616         SEQ_printf(m, "\n");
 617         SEQ_printf(m, "rt_rq[%d]:%s\n", cpu, task_group_path(rt_rq->tg));
 618 #else
 619         SEQ_printf(m, "\n");
 620         SEQ_printf(m, "rt_rq[%d]:\n", cpu);
 621 #endif
 622
 623 #define P(x) \
 624         SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
 625 #define PU(x) \
 626         SEQ_printf(m, "  .%-30s: %lu\n", #x, (unsigned long)(rt_rq->x))
 627 #define PN(x) \
 628         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
 629
 630         PU(rt_nr_running);
 631 #ifdef CONFIG_SMP
 632         PU(rt_nr_migratory);
 633 #endif
 634         P(rt_throttled);
 635         PN(rt_time);
 636         PN(rt_runtime);
 637
 638 #undef PN
 639 #undef PU
 640 #undef P
 641 }
 642
 643 void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq)
 644 {
 645         struct dl_bw *dl_bw;
 646
 647         SEQ_printf(m, "\n");
 648         SEQ_printf(m, "dl_rq[%d]:\n", cpu);
 649
 650 #define PU(x) \
 651         SEQ_printf(m, "  .%-30s: %lu\n", #x, (unsigned long)(dl_rq->x))
 652
 653         PU(dl_nr_running);
 654 #ifdef CONFIG_SMP
 655         PU(dl_nr_migratory);
 656         dl_bw = &cpu_rq(cpu)->rd->dl_bw;
 657 #else
 658         dl_bw = &dl_rq->dl_bw;
 659 #endif
 660         SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->bw", dl_bw->bw);
 661         SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->total_bw", dl_bw->total_bw);
 662
 663 #undef PU
 664 }
 665
 666 static void print_cpu(struct seq_file *m, int cpu)
 667 {
 668         struct rq *rq = cpu_rq(cpu);
 669         unsigned long flags;
 670
 671 #ifdef CONFIG_X86
 672         {
 673                 unsigned int freq = cpu_khz ? : 1;
 674
 675                 SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
 676                            cpu, freq / 1000, (freq % 1000));
 677         }
 678 #else
 679         SEQ_printf(m, "cpu#%d\n", cpu);
 680 #endif
 681
 682 #define P(x)                                                            \
 683 do {                                                                    \
 684         if (sizeof(rq->x) == 4)                                         \
 685                 SEQ_printf(m, "  .%-30s: %ld\n", #x, (long)(rq->x));    \
 686         else                                                            \
 687                 SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rq->x));\
 688 } while (0)
 689
 690 #define PN(x) \
 691         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
 692
 693         P(nr_running);
 694         P(nr_switches);
 695         P(nr_uninterruptible);
 696         PN(next_balance);
 697         SEQ_printf(m, "  .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr)));
 698         PN(clock);
 699         PN(clock_task);
 700 #undef P
 701 #undef PN
 702
 703 #ifdef CONFIG_SMP
 704 #define P64(n) SEQ_printf(m, "  .%-30s: %Ld\n", #n, rq->n);
 705         P64(avg_idle);
 706         P64(max_idle_balance_cost);
 707 #undef P64
 708 #endif
 709
 710 #define P(n) SEQ_printf(m, "  .%-30s: %d\n", #n, schedstat_val(rq->n));
 711         if (schedstat_enabled()) {
 712                 P(yld_count);
 713                 P(sched_count);
 714                 P(sched_goidle);
 715                 P(ttwu_count);
 716                 P(ttwu_local);
 717         }
 718 #undef P
 719
 720         spin_lock_irqsave(&sched_debug_lock, flags);
 721         print_cfs_stats(m, cpu);
 722         print_rt_stats(m, cpu);
 723         print_dl_stats(m, cpu);
 724
 725         print_rq(m, rq, cpu);
 726         spin_unlock_irqrestore(&sched_debug_lock, flags);
 727         SEQ_printf(m, "\n");
 728 }
 729
 730 static const char *sched_tunable_scaling_names[] = {
 731         "none",
 732         "logarithmic",
 733         "linear"
 734 };
 735
 736 static void sched_debug_header(struct seq_file *m)
 737 {
 738         u64 ktime, sched_clk, cpu_clk;
 739         unsigned long flags;
 740
 741         local_irq_save(flags);
 742         ktime = ktime_to_ns(ktime_get());
 743         sched_clk = sched_clock();
 744         cpu_clk = local_clock();
 745         local_irq_restore(flags);
 746
 747         SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n",
 748                 init_utsname()->release,
 749                 (int)strcspn(init_utsname()->version, " "),
 750                 init_utsname()->version);
 751
 752 #define P(x) \
 753         SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
 754 #define PN(x) \
 755         SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
 756         PN(ktime);
 757         PN(sched_clk);
 758         PN(cpu_clk);
 759         P(jiffies);
 760 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
 761         P(sched_clock_stable());
 762 #endif
 763 #undef PN
 764 #undef P
 765
 766         SEQ_printf(m, "\n");
 767         SEQ_printf(m, "sysctl_sched\n");
 768
 769 #define P(x) \
 770         SEQ_printf(m, "  .%-40s: %Ld\n", #x, (long long)(x))
 771 #define PN(x) \
 772         SEQ_printf(m, "  .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
 773         PN(sysctl_sched_latency);
 774         PN(sysctl_sched_min_granularity);
 775         PN(sysctl_sched_wakeup_granularity);
 776         P(sysctl_sched_child_runs_first);
 777         P(sysctl_sched_features);
 778 #undef PN
 779 #undef P
 780
 781         SEQ_printf(m, "  .%-40s: %d (%s)\n",
 782                 "sysctl_sched_tunable_scaling",
 783                 sysctl_sched_tunable_scaling,
 784                 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
 785         SEQ_printf(m, "\n");
 786 }
 787
 788 static int sched_debug_show(struct seq_file *m, void *v)
 789 {
 790         int cpu = (unsigned long)(v - 2);
 791
 792         if (cpu != -1)
 793                 print_cpu(m, cpu);
 794         else
 795                 sched_debug_header(m);
 796
 797         return 0;
 798 }
 799
 800 void sysrq_sched_debug_show(void)
 801 {
 802         int cpu;
 803
 804         sched_debug_header(NULL);
 805         for_each_online_cpu(cpu) {
 806                 /*
 807                  * Need to reset softlockup watchdogs on all CPUs, because
 808                  * another CPU might be blocked waiting for us to process
 809                  * an IPI or stop_machine.
 810                  */
 811                 touch_nmi_watchdog();
 812                 touch_all_softlockup_watchdogs();
 813                 print_cpu(NULL, cpu);
 814         }
 815 }
 816
 817 /*
 818  * This itererator needs some explanation.
 819  * It returns 1 for the header position.
 820  * This means 2 is CPU 0.
 821  * In a hotplugged system some CPUs, including CPU 0, may be missing so we have
 822  * to use cpumask_* to iterate over the CPUs.
 823  */
 824 static void *sched_debug_start(struct seq_file *file, loff_t *offset)
 825 {
 826         unsigned long n = *offset;
 827
 828         if (n == 0)
 829                 return (void *) 1;
 830
 831         n--;
 832
 833         if (n > 0)
 834                 n = cpumask_next(n - 1, cpu_online_mask);
 835         else
 836                 n = cpumask_first(cpu_online_mask);
 837
 838         *offset = n + 1;
 839
 840         if (n < nr_cpu_ids)
 841                 return (void *)(unsigned long)(n + 2);
 842
 843         return NULL;
 844 }
 845
 846 static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
 847 {
 848         (*offset)++;
 849         return sched_debug_start(file, offset);
 850 }
 851
 852 static void sched_debug_stop(struct seq_file *file, void *data)
 853 {
 854 }
 855
 856 static const struct seq_operations sched_debug_sops = {
 857         .start          = sched_debug_start,
 858         .next           = sched_debug_next,
 859         .stop           = sched_debug_stop,
 860         .show           = sched_debug_show,
 861 };
 862
 863 static int __init init_sched_debug_procfs(void)
 864 {
 865         if (!proc_create_seq("sched_debug", 0444, NULL, &sched_debug_sops))
 866                 return -ENOMEM;
 867         return 0;
 868 }
 869
 870 __initcall(init_sched_debug_procfs);
 871
 872 #define __PS(S, F) SEQ_printf(m, "%-45s:%21Ld\n", S, (long long)(F))
 873 #define __P(F) __PS(#F, F)
 874 #define   P(F) __PS(#F, p->F)
 875 #define __PSN(S, F) SEQ_printf(m, "%-45s:%14Ld.%06ld\n", S, SPLIT_NS((long long)(F)))
 876 #define __PN(F) __PSN(#F, F)
 877 #define   PN(F) __PSN(#F, p->F)
 878
 879
 880 #ifdef CONFIG_NUMA_BALANCING
 881 void print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
 882                 unsigned long tpf, unsigned long gsf, unsigned long gpf)
 883 {
 884         SEQ_printf(m, "numa_faults node=%d ", node);
 885         SEQ_printf(m, "task_private=%lu task_shared=%lu ", tpf, tsf);
 886         SEQ_printf(m, "group_private=%lu group_shared=%lu\n", gpf, gsf);
 887 }
 888 #endif
 889
 890
 891 static void sched_show_numa(struct task_struct *p, struct seq_file *m)
 892 {
 893 #ifdef CONFIG_NUMA_BALANCING
 894         struct mempolicy *pol;
 895
 896         if (p->mm)
 897                 P(mm->numa_scan_seq);
 898
 899         task_lock(p);
 900         pol = p->mempolicy;
 901         if (pol && !(pol->flags & MPOL_F_MORON))
 902                 pol = NULL;
 903         mpol_get(pol);
 904         task_unlock(p);
 905
 906         P(numa_pages_migrated);
 907         P(numa_preferred_nid);
 908         P(total_numa_faults);
 909         SEQ_printf(m, "current_node=%d, numa_group_id=%d\n",
 910                         task_node(p), task_numa_group_id(p));
 911         show_numa_stats(p, m);
 912         mpol_put(pol);
 913 #endif
 914 }
 915
 916 void proc_sched_show_task(struct task_struct *p, struct pid_namespace *ns,
 917                                                   struct seq_file *m)
 918 {
 919         unsigned long nr_switches;
 920
 921         SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr_ns(p, ns),
 922                                                 get_nr_threads(p));
 923         SEQ_printf(m,
 924                 "---------------------------------------------------------"
 925                 "----------\n");
 926
 927 #define P_SCHEDSTAT(F)  __PS(#F, schedstat_val(p->F))
 928 #define PN_SCHEDSTAT(F) __PSN(#F, schedstat_val(p->F))
 929
 930         PN(se.exec_start);
 931         PN(se.vruntime);
 932         PN(se.sum_exec_runtime);
 933
 934         nr_switches = p->nvcsw + p->nivcsw;
 935
 936         P(se.nr_migrations);
 937
 938         if (schedstat_enabled()) {
 939                 u64 avg_atom, avg_per_cpu;
 940
 941                 PN_SCHEDSTAT(se.statistics.sum_sleep_runtime);
 942                 PN_SCHEDSTAT(se.statistics.wait_start);
 943                 PN_SCHEDSTAT(se.statistics.sleep_start);
 944                 PN_SCHEDSTAT(se.statistics.block_start);
 945                 PN_SCHEDSTAT(se.statistics.sleep_max);
 946                 PN_SCHEDSTAT(se.statistics.block_max);
 947                 PN_SCHEDSTAT(se.statistics.exec_max);
 948                 PN_SCHEDSTAT(se.statistics.slice_max);
 949                 PN_SCHEDSTAT(se.statistics.wait_max);
 950                 PN_SCHEDSTAT(se.statistics.wait_sum);
 951                 P_SCHEDSTAT(se.statistics.wait_count);
 952                 PN_SCHEDSTAT(se.statistics.iowait_sum);
 953                 P_SCHEDSTAT(se.statistics.iowait_count);
 954                 P_SCHEDSTAT(se.statistics.nr_migrations_cold);
 955                 P_SCHEDSTAT(se.statistics.nr_failed_migrations_affine);
 956                 P_SCHEDSTAT(se.statistics.nr_failed_migrations_running);
 957                 P_SCHEDSTAT(se.statistics.nr_failed_migrations_hot);
 958                 P_SCHEDSTAT(se.statistics.nr_forced_migrations);
 959                 P_SCHEDSTAT(se.statistics.nr_wakeups);
 960                 P_SCHEDSTAT(se.statistics.nr_wakeups_sync);
 961                 P_SCHEDSTAT(se.statistics.nr_wakeups_migrate);
 962                 P_SCHEDSTAT(se.statistics.nr_wakeups_local);
 963                 P_SCHEDSTAT(se.statistics.nr_wakeups_remote);
 964                 P_SCHEDSTAT(se.statistics.nr_wakeups_affine);
 965                 P_SCHEDSTAT(se.statistics.nr_wakeups_affine_attempts);
 966                 P_SCHEDSTAT(se.statistics.nr_wakeups_passive);
 967                 P_SCHEDSTAT(se.statistics.nr_wakeups_idle);
 968
 969                 avg_atom = p->se.sum_exec_runtime;
 970                 if (nr_switches)
 971                         avg_atom = div64_ul(avg_atom, nr_switches);
 972                 else
 973                         avg_atom = -1LL;
 974
 975                 avg_per_cpu = p->se.sum_exec_runtime;
 976                 if (p->se.nr_migrations) {
 977                         avg_per_cpu = div64_u64(avg_per_cpu,
 978                                                 p->se.nr_migrations);
 979                 } else {
 980                         avg_per_cpu = -1LL;
 981                 }
 982
 983                 __PN(avg_atom);
 984                 __PN(avg_per_cpu);
 985         }
 986
 987         __P(nr_switches);
 988         __PS("nr_voluntary_switches", p->nvcsw);
 989         __PS("nr_involuntary_switches", p->nivcsw);
 990
 991         P(se.load.weight);
 992 #ifdef CONFIG_SMP
 993         P(se.avg.load_sum);
 994         P(se.avg.runnable_sum);
 995         P(se.avg.util_sum);
 996         P(se.avg.load_avg);
 997         P(se.avg.runnable_avg);
 998         P(se.avg.util_avg);
 999         P(se.avg.last_update_time);
1000         P(se.avg.util_est.ewma);
1001         P(se.avg.util_est.enqueued);
1002 #endif
1003 #ifdef CONFIG_UCLAMP_TASK
1004         __PS("uclamp.min", p->uclamp_req[UCLAMP_MIN].value);
1005         __PS("uclamp.max", p->uclamp_req[UCLAMP_MAX].value);
1006         __PS("effective uclamp.min", uclamp_eff_value(p, UCLAMP_MIN));
1007         __PS("effective uclamp.max", uclamp_eff_value(p, UCLAMP_MAX));
1008 #endif
1009         P(policy);
1010         P(prio);
1011         if (task_has_dl_policy(p)) {
1012                 P(dl.runtime);
1013                 P(dl.deadline);
1014         }
1015 #undef PN_SCHEDSTAT
1016 #undef P_SCHEDSTAT
1017
1018         {
1019                 unsigned int this_cpu = raw_smp_processor_id();
1020                 u64 t0, t1;
1021
1022                 t0 = cpu_clock(this_cpu);
1023                 t1 = cpu_clock(this_cpu);
1024                 __PS("clock-delta", t1-t0);
1025         }
1026
1027         sched_show_numa(p, m);
1028 }
1029
1030 void proc_sched_set_task(struct task_struct *p)
1031 {
1032 #ifdef CONFIG_SCHEDSTATS
1033         memset(&p->se.statistics, 0, sizeof(p->se.statistics));
1034 #endif
1035 }