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 /*
  12  * This allows printing both to /proc/sched_debug and
  13  * to the console
  14  */
  15 #define SEQ_printf(m, x...)                     \
  16  do {                                           \
  17         if (m)                                  \
  18                 seq_printf(m, x);               \
  19         else                                    \
  20                 pr_cont(x);                     \
  21  } while (0)
  22
  23 /*
  24  * Ease the printing of nsec fields:
  25  */
  26 static long long nsec_high(unsigned long long nsec)
  27 {
  28         if ((long long)nsec < 0) {
  29                 nsec = -nsec;
  30                 do_div(nsec, 1000000);
  31                 return -nsec;
  32         }
  33         do_div(nsec, 1000000);
  34
  35         return nsec;
  36 }
  37
  38 static unsigned long nsec_low(unsigned long long nsec)
  39 {
  40         if ((long long)nsec < 0)
  41                 nsec = -nsec;
  42
  43         return do_div(nsec, 1000000);
  44 }
  45
  46 #define SPLIT_NS(x) nsec_high(x), nsec_low(x)
  47
  48 #define SCHED_FEAT(name, enabled)       \
  49         #name ,
  50
  51 static const char * const sched_feat_names[] = {
  52 #include "features.h"
  53 };
  54
  55 #undef SCHED_FEAT
  56
  57 static int sched_feat_show(struct seq_file *m, void *v)
  58 {
  59         int i;
  60
  61         for (i = 0; i < __SCHED_FEAT_NR; i++) {
  62                 if (!(sysctl_sched_features & (1UL << i)))
  63                         seq_puts(m, "NO_");
  64                 seq_printf(m, "%s ", sched_feat_names[i]);
  65         }
  66         seq_puts(m, "\n");
  67
  68         return 0;
  69 }
  70
  71 #ifdef CONFIG_JUMP_LABEL
  72
  73 #define jump_label_key__true  STATIC_KEY_INIT_TRUE
  74 #define jump_label_key__false STATIC_KEY_INIT_FALSE
  75
  76 #define SCHED_FEAT(name, enabled)       \
  77         jump_label_key__##enabled ,
  78
  79 struct static_key sched_feat_keys[__SCHED_FEAT_NR] = {
  80 #include "features.h"
  81 };
  82
  83 #undef SCHED_FEAT
  84
  85 static void sched_feat_disable(int i)
  86 {
  87         static_key_disable_cpuslocked(&sched_feat_keys[i]);
  88 }
  89
  90 static void sched_feat_enable(int i)
  91 {
  92         static_key_enable_cpuslocked(&sched_feat_keys[i]);
  93 }
  94 #else
  95 static void sched_feat_disable(int i) { };
  96 static void sched_feat_enable(int i) { };
  97 #endif /* CONFIG_JUMP_LABEL */
  98
  99 static int sched_feat_set(char *cmp)
 100 {
 101         int i;
 102         int neg = 0;
 103
 104         if (strncmp(cmp, "NO_", 3) == 0) {
 105                 neg = 1;
 106                 cmp += 3;
 107         }
 108
 109         i = match_string(sched_feat_names, __SCHED_FEAT_NR, cmp);
 110         if (i < 0)
 111                 return i;
 112
 113         if (neg) {
 114                 sysctl_sched_features &= ~(1UL << i);
 115                 sched_feat_disable(i);
 116         } else {
 117                 sysctl_sched_features |= (1UL << i);
 118                 sched_feat_enable(i);
 119         }
 120
 121         return 0;
 122 }
 123
 124 static ssize_t
 125 sched_feat_write(struct file *filp, const char __user *ubuf,
 126                 size_t cnt, loff_t *ppos)
 127 {
 128         char buf[64];
 129         char *cmp;
 130         int ret;
 131         struct inode *inode;
 132
 133         if (cnt > 63)
 134                 cnt = 63;
 135
 136         if (copy_from_user(&buf, ubuf, cnt))
 137                 return -EFAULT;
 138
 139         buf[cnt] = 0;
 140         cmp = strstrip(buf);
 141
 142         /* Ensure the static_key remains in a consistent state */
 143         inode = file_inode(filp);
 144         cpus_read_lock();
 145         inode_lock(inode);
 146         ret = sched_feat_set(cmp);
 147         inode_unlock(inode);
 148         cpus_read_unlock();
 149         if (ret < 0)
 150                 return ret;
 151
 152         *ppos += cnt;
 153
 154         return cnt;
 155 }
 156
 157 static int sched_feat_open(struct inode *inode, struct file *filp)
 158 {
 159         return single_open(filp, sched_feat_show, NULL);
 160 }
 161
 162 static const struct file_operations sched_feat_fops = {
 163         .open           = sched_feat_open,
 164         .write          = sched_feat_write,
 165         .read           = seq_read,
 166         .llseek         = seq_lseek,
 167         .release        = single_release,
 168 };
 169
 170 __read_mostly bool sched_debug_enabled;
 171
 172 static __init int sched_init_debug(void)
 173 {
 174         debugfs_create_file("sched_features", 0644, NULL, NULL,
 175                         &sched_feat_fops);
 176
 177         debugfs_create_bool("sched_debug", 0644, NULL,
 178                         &sched_debug_enabled);
 179
 180         return 0;
 181 }
 182 late_initcall(sched_init_debug);
 183
 184 #ifdef CONFIG_SMP
 185
 186 #ifdef CONFIG_SYSCTL
 187
 188 static struct ctl_table sd_ctl_dir[] = {
 189         {
 190                 .procname       = "sched_domain",
 191                 .mode           = 0555,
 192         },
 193         {}
 194 };
 195
 196 static struct ctl_table sd_ctl_root[] = {
 197         {
 198                 .procname       = "kernel",
 199                 .mode           = 0555,
 200                 .child          = sd_ctl_dir,
 201         },
 202         {}
 203 };
 204
 205 static struct ctl_table *sd_alloc_ctl_entry(int n)
 206 {
 207         struct ctl_table *entry =
 208                 kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL);
 209
 210         return entry;
 211 }
 212
 213 static void sd_free_ctl_entry(struct ctl_table **tablep)
 214 {
 215         struct ctl_table *entry;
 216
 217         /*
 218          * In the intermediate directories, both the child directory and
 219          * procname are dynamically allocated and could fail but the mode
 220          * will always be set. In the lowest directory the names are
 221          * static strings and all have proc handlers.
 222          */
 223         for (entry = *tablep; entry->mode; entry++) {
 224                 if (entry->child)
 225                         sd_free_ctl_entry(&entry->child);
 226                 if (entry->proc_handler == NULL)
 227                         kfree(entry->procname);
 228         }
 229
 230         kfree(*tablep);
 231         *tablep = NULL;
 232 }
 233
 234 static void
 235 set_table_entry(struct ctl_table *entry,
 236                 const char *procname, void *data, int maxlen,
 237                 umode_t mode, proc_handler *proc_handler)
 238 {
 239         entry->procname = procname;
 240         entry->data = data;
 241         entry->maxlen = maxlen;
 242         entry->mode = mode;
 243         entry->proc_handler = proc_handler;
 244 }
 245
 246 static int sd_ctl_doflags(struct ctl_table *table, int write,
 247                           void *buffer, size_t *lenp, loff_t *ppos)
 248 {
 249         unsigned long flags = *(unsigned long *)table->data;
 250         size_t data_size = 0;
 251         size_t len = 0;
 252         char *tmp, *buf;
 253         int idx;
 254
 255         if (write)
 256                 return 0;
 257
 258         for_each_set_bit(idx, &flags, __SD_FLAG_CNT) {
 259                 char *name = sd_flag_debug[idx].name;
 260
 261                 /* Name plus whitespace */
 262                 data_size += strlen(name) + 1;
 263         }
 264
 265         if (*ppos > data_size) {
 266                 *lenp = 0;
 267                 return 0;
 268         }
 269
 270         buf = kcalloc(data_size + 1, sizeof(*buf), GFP_KERNEL);
 271         if (!buf)
 272                 return -ENOMEM;
 273
 274         for_each_set_bit(idx, &flags, __SD_FLAG_CNT) {
 275                 char *name = sd_flag_debug[idx].name;
 276
 277                 len += snprintf(buf + len, strlen(name) + 2, "%s ", name);
 278         }
 279
 280         tmp = buf + *ppos;
 281         len -= *ppos;
 282
 283         if (len > *lenp)
 284                 len = *lenp;
 285         if (len)
 286                 memcpy(buffer, tmp, len);
 287         if (len < *lenp) {
 288                 ((char *)buffer)[len] = '\n';
 289                 len++;
 290         }
 291
 292         *lenp = len;
 293         *ppos += len;
 294
 295         kfree(buf);
 296
 297         return 0;
 298 }
 299
 300 static struct ctl_table *
 301 sd_alloc_ctl_domain_table(struct sched_domain *sd)
 302 {
 303         struct ctl_table *table = sd_alloc_ctl_entry(9);
 304
 305         if (table == NULL)
 306                 return NULL;
 307
 308         set_table_entry(&table[0], "min_interval",        &sd->min_interval,        sizeof(long), 0644, proc_doulongvec_minmax);
 309         set_table_entry(&table[1], "max_interval",        &sd->max_interval,        sizeof(long), 0644, proc_doulongvec_minmax);
 310         set_table_entry(&table[2], "busy_factor",         &sd->busy_factor,         sizeof(int),  0644, proc_dointvec_minmax);
 311         set_table_entry(&table[3], "imbalance_pct",       &sd->imbalance_pct,       sizeof(int),  0644, proc_dointvec_minmax);
 312         set_table_entry(&table[4], "cache_nice_tries",    &sd->cache_nice_tries,    sizeof(int),  0644, proc_dointvec_minmax);
 313         set_table_entry(&table[5], "flags",               &sd->flags,               sizeof(int),  0444, sd_ctl_doflags);
 314         set_table_entry(&table[6], "max_newidle_lb_cost", &sd->max_newidle_lb_cost, sizeof(long), 0644, proc_doulongvec_minmax);
 315         set_table_entry(&table[7], "name",                sd->name,            CORENAME_MAX_SIZE, 0444, proc_dostring);
 316         /* &table[8] is terminator */
 317
 318         return table;
 319 }
 320
 321 static struct ctl_table *sd_alloc_ctl_cpu_table(int cpu)
 322 {
 323         struct ctl_table *entry, *table;
 324         struct sched_domain *sd;
 325         int domain_num = 0, i;
 326         char buf[32];
 327
 328         for_each_domain(cpu, sd)
 329                 domain_num++;
 330         entry = table = sd_alloc_ctl_entry(domain_num + 1);
 331         if (table == NULL)
 332                 return NULL;
 333
 334         i = 0;
 335         for_each_domain(cpu, sd) {
 336                 snprintf(buf, 32, "domain%d", i);
 337                 entry->procname = kstrdup(buf, GFP_KERNEL);
 338                 entry->mode = 0555;
 339                 entry->child = sd_alloc_ctl_domain_table(sd);
 340                 entry++;
 341                 i++;
 342         }
 343         return table;
 344 }
 345
 346 static cpumask_var_t            sd_sysctl_cpus;
 347 static struct ctl_table_header  *sd_sysctl_header;
 348
 349 void register_sched_domain_sysctl(void)
 350 {
 351         static struct ctl_table *cpu_entries;
 352         static struct ctl_table **cpu_idx;
 353         static bool init_done = false;
 354         char buf[32];
 355         int i;
 356
 357         if (!cpu_entries) {
 358                 cpu_entries = sd_alloc_ctl_entry(num_possible_cpus() + 1);
 359                 if (!cpu_entries)
 360                         return;
 361
 362                 WARN_ON(sd_ctl_dir[0].child);
 363                 sd_ctl_dir[0].child = cpu_entries;
 364         }
 365
 366         if (!cpu_idx) {
 367                 struct ctl_table *e = cpu_entries;
 368
 369                 cpu_idx = kcalloc(nr_cpu_ids, sizeof(struct ctl_table*), GFP_KERNEL);
 370                 if (!cpu_idx)
 371                         return;
 372
 373                 /* deal with sparse possible map */
 374                 for_each_possible_cpu(i) {
 375                         cpu_idx[i] = e;
 376                         e++;
 377                 }
 378         }
 379
 380         if (!cpumask_available(sd_sysctl_cpus)) {
 381                 if (!alloc_cpumask_var(&sd_sysctl_cpus, GFP_KERNEL))
 382                         return;
 383         }
 384
 385         if (!init_done) {
 386                 init_done = true;
 387                 /* init to possible to not have holes in @cpu_entries */
 388                 cpumask_copy(sd_sysctl_cpus, cpu_possible_mask);
 389         }
 390
 391         for_each_cpu(i, sd_sysctl_cpus) {
 392                 struct ctl_table *e = cpu_idx[i];
 393
 394                 if (e->child)
 395                         sd_free_ctl_entry(&e->child);
 396
 397                 if (!e->procname) {
 398                         snprintf(buf, 32, "cpu%d", i);
 399                         e->procname = kstrdup(buf, GFP_KERNEL);
 400                 }
 401                 e->mode = 0555;
 402                 e->child = sd_alloc_ctl_cpu_table(i);
 403
 404                 __cpumask_clear_cpu(i, sd_sysctl_cpus);
 405         }
 406
 407         WARN_ON(sd_sysctl_header);
 408         sd_sysctl_header = register_sysctl_table(sd_ctl_root);
 409 }
 410
 411 void dirty_sched_domain_sysctl(int cpu)
 412 {
 413         if (cpumask_available(sd_sysctl_cpus))
 414                 __cpumask_set_cpu(cpu, sd_sysctl_cpus);
 415 }
 416
 417 /* may be called multiple times per register */
 418 void unregister_sched_domain_sysctl(void)
 419 {
 420         unregister_sysctl_table(sd_sysctl_header);
 421         sd_sysctl_header = NULL;
 422 }
 423 #endif /* CONFIG_SYSCTL */
 424 #endif /* CONFIG_SMP */
 425
 426 #ifdef CONFIG_FAIR_GROUP_SCHED
 427 static void print_cfs_group_stats(struct seq_file *m, int cpu, struct task_group *tg)
 428 {
 429         struct sched_entity *se = tg->se[cpu];
 430
 431 #define P(F)            SEQ_printf(m, "  .%-30s: %lld\n",       #F, (long long)F)
 432 #define P_SCHEDSTAT(F)  SEQ_printf(m, "  .%-30s: %lld\n",       #F, (long long)schedstat_val(F))
 433 #define PN(F)           SEQ_printf(m, "  .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F))
 434 #define PN_SCHEDSTAT(F) SEQ_printf(m, "  .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)schedstat_val(F)))
 435
 436         if (!se)
 437                 return;
 438
 439         PN(se->exec_start);
 440         PN(se->vruntime);
 441         PN(se->sum_exec_runtime);
 442
 443         if (schedstat_enabled()) {
 444                 PN_SCHEDSTAT(se->statistics.wait_start);
 445                 PN_SCHEDSTAT(se->statistics.sleep_start);
 446                 PN_SCHEDSTAT(se->statistics.block_start);
 447                 PN_SCHEDSTAT(se->statistics.sleep_max);
 448                 PN_SCHEDSTAT(se->statistics.block_max);
 449                 PN_SCHEDSTAT(se->statistics.exec_max);
 450                 PN_SCHEDSTAT(se->statistics.slice_max);
 451                 PN_SCHEDSTAT(se->statistics.wait_max);
 452                 PN_SCHEDSTAT(se->statistics.wait_sum);
 453                 P_SCHEDSTAT(se->statistics.wait_count);
 454         }
 455
 456         P(se->load.weight);
 457 #ifdef CONFIG_SMP
 458         P(se->avg.load_avg);
 459         P(se->avg.util_avg);
 460         P(se->avg.runnable_avg);
 461 #endif
 462
 463 #undef PN_SCHEDSTAT
 464 #undef PN
 465 #undef P_SCHEDSTAT
 466 #undef P
 467 }
 468 #endif
 469
 470 #ifdef CONFIG_CGROUP_SCHED
 471 static DEFINE_SPINLOCK(sched_debug_lock);
 472 static char group_path[PATH_MAX];
 473
 474 static void task_group_path(struct task_group *tg, char *path, int plen)
 475 {
 476         if (autogroup_path(tg, path, plen))
 477                 return;
 478
 479         cgroup_path(tg->css.cgroup, path, plen);
 480 }
 481
 482 /*
 483  * Only 1 SEQ_printf_task_group_path() caller can use the full length
 484  * group_path[] for cgroup path. Other simultaneous callers will have
 485  * to use a shorter stack buffer. A "..." suffix is appended at the end
 486  * of the stack buffer so that it will show up in case the output length
 487  * matches the given buffer size to indicate possible path name truncation.
 488  */
 489 #define SEQ_printf_task_group_path(m, tg, fmt...)                       \
 490 {                                                                       \
 491         if (spin_trylock(&sched_debug_lock)) {                          \
 492                 task_group_path(tg, group_path, sizeof(group_path));    \
 493                 SEQ_printf(m, fmt, group_path);                         \
 494                 spin_unlock(&sched_debug_lock);                         \
 495         } else {                                                        \
 496                 char buf[128];                                          \
 497                 char *bufend = buf + sizeof(buf) - 3;                   \
 498                 task_group_path(tg, buf, bufend - buf);                 \
 499                 strcpy(bufend - 1, "...");                              \
 500                 SEQ_printf(m, fmt, buf);                                \
 501         }                                                               \
 502 }
 503 #endif
 504
 505 static void
 506 print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
 507 {
 508         if (rq->curr == p)
 509                 SEQ_printf(m, ">R");
 510         else
 511                 SEQ_printf(m, " %c", task_state_to_char(p));
 512
 513         SEQ_printf(m, " %15s %5d %9Ld.%06ld %9Ld %5d ",
 514                 p->comm, task_pid_nr(p),
 515                 SPLIT_NS(p->se.vruntime),
 516                 (long long)(p->nvcsw + p->nivcsw),
 517                 p->prio);
 518
 519         SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
 520                 SPLIT_NS(schedstat_val_or_zero(p->se.statistics.wait_sum)),
 521                 SPLIT_NS(p->se.sum_exec_runtime),
 522                 SPLIT_NS(schedstat_val_or_zero(p->se.statistics.sum_sleep_runtime)));
 523
 524 #ifdef CONFIG_NUMA_BALANCING
 525         SEQ_printf(m, " %d %d", task_node(p), task_numa_group_id(p));
 526 #endif
 527 #ifdef CONFIG_CGROUP_SCHED
 528         SEQ_printf_task_group_path(m, task_group(p), " %s")
 529 #endif
 530
 531         SEQ_printf(m, "\n");
 532 }
 533
 534 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
 535 {
 536         struct task_struct *g, *p;
 537
 538         SEQ_printf(m, "\n");
 539         SEQ_printf(m, "runnable tasks:\n");
 540         SEQ_printf(m, " S            task   PID         tree-key  switches  prio"
 541                    "     wait-time             sum-exec        sum-sleep\n");
 542         SEQ_printf(m, "-------------------------------------------------------"
 543                    "------------------------------------------------------\n");
 544
 545         rcu_read_lock();
 546         for_each_process_thread(g, p) {
 547                 if (task_cpu(p) != rq_cpu)
 548                         continue;
 549
 550                 print_task(m, rq, p);
 551         }
 552         rcu_read_unlock();
 553 }
 554
 555 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
 556 {
 557         s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
 558                 spread, rq0_min_vruntime, spread0;
 559         struct rq *rq = cpu_rq(cpu);
 560         struct sched_entity *last;
 561         unsigned long flags;
 562
 563 #ifdef CONFIG_FAIR_GROUP_SCHED
 564         SEQ_printf(m, "\n");
 565         SEQ_printf_task_group_path(m, cfs_rq->tg, "cfs_rq[%d]:%s\n", cpu);
 566 #else
 567         SEQ_printf(m, "\n");
 568         SEQ_printf(m, "cfs_rq[%d]:\n", cpu);
 569 #endif
 570         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "exec_clock",
 571                         SPLIT_NS(cfs_rq->exec_clock));
 572
 573         raw_spin_lock_irqsave(&rq->lock, flags);
 574         if (rb_first_cached(&cfs_rq->tasks_timeline))
 575                 MIN_vruntime = (__pick_first_entity(cfs_rq))->vruntime;
 576         last = __pick_last_entity(cfs_rq);
 577         if (last)
 578                 max_vruntime = last->vruntime;
 579         min_vruntime = cfs_rq->min_vruntime;
 580         rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
 581         raw_spin_unlock_irqrestore(&rq->lock, flags);
 582         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "MIN_vruntime",
 583                         SPLIT_NS(MIN_vruntime));
 584         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "min_vruntime",
 585                         SPLIT_NS(min_vruntime));
 586         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "max_vruntime",
 587                         SPLIT_NS(max_vruntime));
 588         spread = max_vruntime - MIN_vruntime;
 589         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread",
 590                         SPLIT_NS(spread));
 591         spread0 = min_vruntime - rq0_min_vruntime;
 592         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread0",
 593                         SPLIT_NS(spread0));
 594         SEQ_printf(m, "  .%-30s: %d\n", "nr_spread_over",
 595                         cfs_rq->nr_spread_over);
 596         SEQ_printf(m, "  .%-30s: %d\n", "nr_running", cfs_rq->nr_running);
 597         SEQ_printf(m, "  .%-30s: %ld\n", "load", cfs_rq->load.weight);
 598 #ifdef CONFIG_SMP
 599         SEQ_printf(m, "  .%-30s: %lu\n", "load_avg",
 600                         cfs_rq->avg.load_avg);
 601         SEQ_printf(m, "  .%-30s: %lu\n", "runnable_avg",
 602                         cfs_rq->avg.runnable_avg);
 603         SEQ_printf(m, "  .%-30s: %lu\n", "util_avg",
 604                         cfs_rq->avg.util_avg);
 605         SEQ_printf(m, "  .%-30s: %u\n", "util_est_enqueued",
 606                         cfs_rq->avg.util_est.enqueued);
 607         SEQ_printf(m, "  .%-30s: %ld\n", "removed.load_avg",
 608                         cfs_rq->removed.load_avg);
 609         SEQ_printf(m, "  .%-30s: %ld\n", "removed.util_avg",
 610                         cfs_rq->removed.util_avg);
 611         SEQ_printf(m, "  .%-30s: %ld\n", "removed.runnable_avg",
 612                         cfs_rq->removed.runnable_avg);
 613 #ifdef CONFIG_FAIR_GROUP_SCHED
 614         SEQ_printf(m, "  .%-30s: %lu\n", "tg_load_avg_contrib",
 615                         cfs_rq->tg_load_avg_contrib);
 616         SEQ_printf(m, "  .%-30s: %ld\n", "tg_load_avg",
 617                         atomic_long_read(&cfs_rq->tg->load_avg));
 618 #endif
 619 #endif
 620 #ifdef CONFIG_CFS_BANDWIDTH
 621         SEQ_printf(m, "  .%-30s: %d\n", "throttled",
 622                         cfs_rq->throttled);
 623         SEQ_printf(m, "  .%-30s: %d\n", "throttle_count",
 624                         cfs_rq->throttle_count);
 625 #endif
 626
 627 #ifdef CONFIG_FAIR_GROUP_SCHED
 628         print_cfs_group_stats(m, cpu, cfs_rq->tg);
 629 #endif
 630 }
 631
 632 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
 633 {
 634 #ifdef CONFIG_RT_GROUP_SCHED
 635         SEQ_printf(m, "\n");
 636         SEQ_printf_task_group_path(m, rt_rq->tg, "rt_rq[%d]:%s\n", cpu);
 637 #else
 638         SEQ_printf(m, "\n");
 639         SEQ_printf(m, "rt_rq[%d]:\n", cpu);
 640 #endif
 641
 642 #define P(x) \
 643         SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
 644 #define PU(x) \
 645         SEQ_printf(m, "  .%-30s: %lu\n", #x, (unsigned long)(rt_rq->x))
 646 #define PN(x) \
 647         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
 648
 649         PU(rt_nr_running);
 650 #ifdef CONFIG_SMP
 651         PU(rt_nr_migratory);
 652 #endif
 653         P(rt_throttled);
 654         PN(rt_time);
 655         PN(rt_runtime);
 656
 657 #undef PN
 658 #undef PU
 659 #undef P
 660 }
 661
 662 void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq)
 663 {
 664         struct dl_bw *dl_bw;
 665
 666         SEQ_printf(m, "\n");
 667         SEQ_printf(m, "dl_rq[%d]:\n", cpu);
 668
 669 #define PU(x) \
 670         SEQ_printf(m, "  .%-30s: %lu\n", #x, (unsigned long)(dl_rq->x))
 671
 672         PU(dl_nr_running);
 673 #ifdef CONFIG_SMP
 674         PU(dl_nr_migratory);
 675         dl_bw = &cpu_rq(cpu)->rd->dl_bw;
 676 #else
 677         dl_bw = &dl_rq->dl_bw;
 678 #endif
 679         SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->bw", dl_bw->bw);
 680         SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->total_bw", dl_bw->total_bw);
 681
 682 #undef PU
 683 }
 684
 685 static void print_cpu(struct seq_file *m, int cpu)
 686 {
 687         struct rq *rq = cpu_rq(cpu);
 688
 689 #ifdef CONFIG_X86
 690         {
 691                 unsigned int freq = cpu_khz ? : 1;
 692
 693                 SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
 694                            cpu, freq / 1000, (freq % 1000));
 695         }
 696 #else
 697         SEQ_printf(m, "cpu#%d\n", cpu);
 698 #endif
 699
 700 #define P(x)                                                            \
 701 do {                                                                    \
 702         if (sizeof(rq->x) == 4)                                         \
 703                 SEQ_printf(m, "  .%-30s: %ld\n", #x, (long)(rq->x));    \
 704         else                                                            \
 705                 SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rq->x));\
 706 } while (0)
 707
 708 #define PN(x) \
 709         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
 710
 711         P(nr_running);
 712         P(nr_switches);
 713         P(nr_uninterruptible);
 714         PN(next_balance);
 715         SEQ_printf(m, "  .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr)));
 716         PN(clock);
 717         PN(clock_task);
 718 #undef P
 719 #undef PN
 720
 721 #ifdef CONFIG_SMP
 722 #define P64(n) SEQ_printf(m, "  .%-30s: %Ld\n", #n, rq->n);
 723         P64(avg_idle);
 724         P64(max_idle_balance_cost);
 725 #undef P64
 726 #endif
 727
 728 #define P(n) SEQ_printf(m, "  .%-30s: %d\n", #n, schedstat_val(rq->n));
 729         if (schedstat_enabled()) {
 730                 P(yld_count);
 731                 P(sched_count);
 732                 P(sched_goidle);
 733                 P(ttwu_count);
 734                 P(ttwu_local);
 735         }
 736 #undef P
 737
 738         print_cfs_stats(m, cpu);
 739         print_rt_stats(m, cpu);
 740         print_dl_stats(m, cpu);
 741
 742         print_rq(m, rq, cpu);
 743         SEQ_printf(m, "\n");
 744 }
 745
 746 static const char *sched_tunable_scaling_names[] = {
 747         "none",
 748         "logarithmic",
 749         "linear"
 750 };
 751
 752 static void sched_debug_header(struct seq_file *m)
 753 {
 754         u64 ktime, sched_clk, cpu_clk;
 755         unsigned long flags;
 756
 757         local_irq_save(flags);
 758         ktime = ktime_to_ns(ktime_get());
 759         sched_clk = sched_clock();
 760         cpu_clk = local_clock();
 761         local_irq_restore(flags);
 762
 763         SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n",
 764                 init_utsname()->release,
 765                 (int)strcspn(init_utsname()->version, " "),
 766                 init_utsname()->version);
 767
 768 #define P(x) \
 769         SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
 770 #define PN(x) \
 771         SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
 772         PN(ktime);
 773         PN(sched_clk);
 774         PN(cpu_clk);
 775         P(jiffies);
 776 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
 777         P(sched_clock_stable());
 778 #endif
 779 #undef PN
 780 #undef P
 781
 782         SEQ_printf(m, "\n");
 783         SEQ_printf(m, "sysctl_sched\n");
 784
 785 #define P(x) \
 786         SEQ_printf(m, "  .%-40s: %Ld\n", #x, (long long)(x))
 787 #define PN(x) \
 788         SEQ_printf(m, "  .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
 789         PN(sysctl_sched_latency);
 790         PN(sysctl_sched_min_granularity);
 791         PN(sysctl_sched_wakeup_granularity);
 792         P(sysctl_sched_child_runs_first);
 793         P(sysctl_sched_features);
 794 #undef PN
 795 #undef P
 796
 797         SEQ_printf(m, "  .%-40s: %d (%s)\n",
 798                 "sysctl_sched_tunable_scaling",
 799                 sysctl_sched_tunable_scaling,
 800                 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
 801         SEQ_printf(m, "\n");
 802 }
 803
 804 static int sched_debug_show(struct seq_file *m, void *v)
 805 {
 806         int cpu = (unsigned long)(v - 2);
 807
 808         if (cpu != -1)
 809                 print_cpu(m, cpu);
 810         else
 811                 sched_debug_header(m);
 812
 813         return 0;
 814 }
 815
 816 void sysrq_sched_debug_show(void)
 817 {
 818         int cpu;
 819
 820         sched_debug_header(NULL);
 821         for_each_online_cpu(cpu) {
 822                 /*
 823                  * Need to reset softlockup watchdogs on all CPUs, because
 824                  * another CPU might be blocked waiting for us to process
 825                  * an IPI or stop_machine.
 826                  */
 827                 touch_nmi_watchdog();
 828                 touch_all_softlockup_watchdogs();
 829                 print_cpu(NULL, cpu);
 830         }
 831 }
 832
 833 /*
 834  * This itererator needs some explanation.
 835  * It returns 1 for the header position.
 836  * This means 2 is CPU 0.
 837  * In a hotplugged system some CPUs, including CPU 0, may be missing so we have
 838  * to use cpumask_* to iterate over the CPUs.
 839  */
 840 static void *sched_debug_start(struct seq_file *file, loff_t *offset)
 841 {
 842         unsigned long n = *offset;
 843
 844         if (n == 0)
 845                 return (void *) 1;
 846
 847         n--;
 848
 849         if (n > 0)
 850                 n = cpumask_next(n - 1, cpu_online_mask);
 851         else
 852                 n = cpumask_first(cpu_online_mask);
 853
 854         *offset = n + 1;
 855
 856         if (n < nr_cpu_ids)
 857                 return (void *)(unsigned long)(n + 2);
 858
 859         return NULL;
 860 }
 861
 862 static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
 863 {
 864         (*offset)++;
 865         return sched_debug_start(file, offset);
 866 }
 867
 868 static void sched_debug_stop(struct seq_file *file, void *data)
 869 {
 870 }
 871
 872 static const struct seq_operations sched_debug_sops = {
 873         .start          = sched_debug_start,
 874         .next           = sched_debug_next,
 875         .stop           = sched_debug_stop,
 876         .show           = sched_debug_show,
 877 };
 878
 879 static int __init init_sched_debug_procfs(void)
 880 {
 881         if (!proc_create_seq("sched_debug", 0444, NULL, &sched_debug_sops))
 882                 return -ENOMEM;
 883         return 0;
 884 }
 885
 886 __initcall(init_sched_debug_procfs);
 887
 888 #define __PS(S, F) SEQ_printf(m, "%-45s:%21Ld\n", S, (long long)(F))
 889 #define __P(F) __PS(#F, F)
 890 #define   P(F) __PS(#F, p->F)
 891 #define   PM(F, M) __PS(#F, p->F & (M))
 892 #define __PSN(S, F) SEQ_printf(m, "%-45s:%14Ld.%06ld\n", S, SPLIT_NS((long long)(F)))
 893 #define __PN(F) __PSN(#F, F)
 894 #define   PN(F) __PSN(#F, p->F)
 895
 896
 897 #ifdef CONFIG_NUMA_BALANCING
 898 void print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
 899                 unsigned long tpf, unsigned long gsf, unsigned long gpf)
 900 {
 901         SEQ_printf(m, "numa_faults node=%d ", node);
 902         SEQ_printf(m, "task_private=%lu task_shared=%lu ", tpf, tsf);
 903         SEQ_printf(m, "group_private=%lu group_shared=%lu\n", gpf, gsf);
 904 }
 905 #endif
 906
 907
 908 static void sched_show_numa(struct task_struct *p, struct seq_file *m)
 909 {
 910 #ifdef CONFIG_NUMA_BALANCING
 911         if (p->mm)
 912                 P(mm->numa_scan_seq);
 913
 914         P(numa_pages_migrated);
 915         P(numa_preferred_nid);
 916         P(total_numa_faults);
 917         SEQ_printf(m, "current_node=%d, numa_group_id=%d\n",
 918                         task_node(p), task_numa_group_id(p));
 919         show_numa_stats(p, m);
 920 #endif
 921 }
 922
 923 void proc_sched_show_task(struct task_struct *p, struct pid_namespace *ns,
 924                                                   struct seq_file *m)
 925 {
 926         unsigned long nr_switches;
 927
 928         SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr_ns(p, ns),
 929                                                 get_nr_threads(p));
 930         SEQ_printf(m,
 931                 "---------------------------------------------------------"
 932                 "----------\n");
 933
 934 #define P_SCHEDSTAT(F)  __PS(#F, schedstat_val(p->F))
 935 #define PN_SCHEDSTAT(F) __PSN(#F, schedstat_val(p->F))
 936
 937         PN(se.exec_start);
 938         PN(se.vruntime);
 939         PN(se.sum_exec_runtime);
 940
 941         nr_switches = p->nvcsw + p->nivcsw;
 942
 943         P(se.nr_migrations);
 944
 945         if (schedstat_enabled()) {
 946                 u64 avg_atom, avg_per_cpu;
 947
 948                 PN_SCHEDSTAT(se.statistics.sum_sleep_runtime);
 949                 PN_SCHEDSTAT(se.statistics.wait_start);
 950                 PN_SCHEDSTAT(se.statistics.sleep_start);
 951                 PN_SCHEDSTAT(se.statistics.block_start);
 952                 PN_SCHEDSTAT(se.statistics.sleep_max);
 953                 PN_SCHEDSTAT(se.statistics.block_max);
 954                 PN_SCHEDSTAT(se.statistics.exec_max);
 955                 PN_SCHEDSTAT(se.statistics.slice_max);
 956                 PN_SCHEDSTAT(se.statistics.wait_max);
 957                 PN_SCHEDSTAT(se.statistics.wait_sum);
 958                 P_SCHEDSTAT(se.statistics.wait_count);
 959                 PN_SCHEDSTAT(se.statistics.iowait_sum);
 960                 P_SCHEDSTAT(se.statistics.iowait_count);
 961                 P_SCHEDSTAT(se.statistics.nr_migrations_cold);
 962                 P_SCHEDSTAT(se.statistics.nr_failed_migrations_affine);
 963                 P_SCHEDSTAT(se.statistics.nr_failed_migrations_running);
 964                 P_SCHEDSTAT(se.statistics.nr_failed_migrations_hot);
 965                 P_SCHEDSTAT(se.statistics.nr_forced_migrations);
 966                 P_SCHEDSTAT(se.statistics.nr_wakeups);
 967                 P_SCHEDSTAT(se.statistics.nr_wakeups_sync);
 968                 P_SCHEDSTAT(se.statistics.nr_wakeups_migrate);
 969                 P_SCHEDSTAT(se.statistics.nr_wakeups_local);
 970                 P_SCHEDSTAT(se.statistics.nr_wakeups_remote);
 971                 P_SCHEDSTAT(se.statistics.nr_wakeups_affine);
 972                 P_SCHEDSTAT(se.statistics.nr_wakeups_affine_attempts);
 973                 P_SCHEDSTAT(se.statistics.nr_wakeups_passive);
 974                 P_SCHEDSTAT(se.statistics.nr_wakeups_idle);
 975
 976                 avg_atom = p->se.sum_exec_runtime;
 977                 if (nr_switches)
 978                         avg_atom = div64_ul(avg_atom, nr_switches);
 979                 else
 980                         avg_atom = -1LL;
 981
 982                 avg_per_cpu = p->se.sum_exec_runtime;
 983                 if (p->se.nr_migrations) {
 984                         avg_per_cpu = div64_u64(avg_per_cpu,
 985                                                 p->se.nr_migrations);
 986                 } else {
 987                         avg_per_cpu = -1LL;
 988                 }
 989
 990                 __PN(avg_atom);
 991                 __PN(avg_per_cpu);
 992         }
 993
 994         __P(nr_switches);
 995         __PS("nr_voluntary_switches", p->nvcsw);
 996         __PS("nr_involuntary_switches", p->nivcsw);
 997
 998         P(se.load.weight);
 999 #ifdef CONFIG_SMP
1000         P(se.avg.load_sum);
1001         P(se.avg.runnable_sum);
1002         P(se.avg.util_sum);
1003         P(se.avg.load_avg);
1004         P(se.avg.runnable_avg);
1005         P(se.avg.util_avg);
1006         P(se.avg.last_update_time);
1007         P(se.avg.util_est.ewma);
1008         PM(se.avg.util_est.enqueued, ~UTIL_AVG_UNCHANGED);
1009 #endif
1010 #ifdef CONFIG_UCLAMP_TASK
1011         __PS("uclamp.min", p->uclamp_req[UCLAMP_MIN].value);
1012         __PS("uclamp.max", p->uclamp_req[UCLAMP_MAX].value);
1013         __PS("effective uclamp.min", uclamp_eff_value(p, UCLAMP_MIN));
1014         __PS("effective uclamp.max", uclamp_eff_value(p, UCLAMP_MAX));
1015 #endif
1016         P(policy);
1017         P(prio);
1018         if (task_has_dl_policy(p)) {
1019                 P(dl.runtime);
1020                 P(dl.deadline);
1021         }
1022 #undef PN_SCHEDSTAT
1023 #undef P_SCHEDSTAT
1024
1025         {
1026                 unsigned int this_cpu = raw_smp_processor_id();
1027                 u64 t0, t1;
1028
1029                 t0 = cpu_clock(this_cpu);
1030                 t1 = cpu_clock(this_cpu);
1031                 __PS("clock-delta", t1-t0);
1032         }
1033
1034         sched_show_numa(p, m);
1035 }
1036
1037 void proc_sched_set_task(struct task_struct *p)
1038 {
1039 #ifdef CONFIG_SCHEDSTATS
1040         memset(&p->se.statistics, 0, sizeof(p->se.statistics));
1041 #endif
1042 }