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 #ifdef CONFIG_SMP 406 P(se->avg.load_avg); 407 P(se->avg.util_avg); 408 P(se->avg.runnable_avg); 409 #endif 410 411 #undef PN_SCHEDSTAT 412 #undef PN 413 #undef P_SCHEDSTAT 414 #undef P 415 } 416 #endif 417 418 #ifdef CONFIG_CGROUP_SCHED 419 static char group_path[PATH_MAX]; 420 421 static char *task_group_path(struct task_group *tg) 422 { 423 if (autogroup_path(tg, group_path, PATH_MAX)) 424 return group_path; 425 426 cgroup_path(tg->css.cgroup, group_path, PATH_MAX); 427 428 return group_path; 429 } 430 #endif 431 432 static void 433 print_task(struct seq_file *m, struct rq *rq, struct task_struct *p) 434 { 435 if (rq->curr == p) 436 SEQ_printf(m, ">R"); 437 else 438 SEQ_printf(m, " %c", task_state_to_char(p)); 439 440 SEQ_printf(m, "%15s %5d %9Ld.%06ld %9Ld %5d ", 441 p->comm, task_pid_nr(p), 442 SPLIT_NS(p->se.vruntime), 443 (long long)(p->nvcsw + p->nivcsw), 444 p->prio); 445 446 SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld", 447 SPLIT_NS(schedstat_val_or_zero(p->se.statistics.wait_sum)), 448 SPLIT_NS(p->se.sum_exec_runtime), 449 SPLIT_NS(schedstat_val_or_zero(p->se.statistics.sum_sleep_runtime))); 450 451 #ifdef CONFIG_NUMA_BALANCING 452 SEQ_printf(m, " %d %d", task_node(p), task_numa_group_id(p)); 453 #endif 454 #ifdef CONFIG_CGROUP_SCHED 455 SEQ_printf(m, " %s", task_group_path(task_group(p))); 456 #endif 457 458 SEQ_printf(m, "\n"); 459 } 460 461 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu) 462 { 463 struct task_struct *g, *p; 464 465 SEQ_printf(m, "\n"); 466 SEQ_printf(m, "runnable tasks:\n"); 467 SEQ_printf(m, " S task PID tree-key switches prio" 468 " wait-time sum-exec sum-sleep\n"); 469 SEQ_printf(m, "-------------------------------------------------------" 470 "----------------------------------------------------\n"); 471 472 rcu_read_lock(); 473 for_each_process_thread(g, p) { 474 if (task_cpu(p) != rq_cpu) 475 continue; 476 477 print_task(m, rq, p); 478 } 479 rcu_read_unlock(); 480 } 481 482 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq) 483 { 484 s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1, 485 spread, rq0_min_vruntime, spread0; 486 struct rq *rq = cpu_rq(cpu); 487 struct sched_entity *last; 488 unsigned long flags; 489 490 #ifdef CONFIG_FAIR_GROUP_SCHED 491 SEQ_printf(m, "\n"); 492 SEQ_printf(m, "cfs_rq[%d]:%s\n", cpu, task_group_path(cfs_rq->tg)); 493 #else 494 SEQ_printf(m, "\n"); 495 SEQ_printf(m, "cfs_rq[%d]:\n", cpu); 496 #endif 497 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "exec_clock", 498 SPLIT_NS(cfs_rq->exec_clock)); 499 500 raw_spin_lock_irqsave(&rq->lock, flags); 501 if (rb_first_cached(&cfs_rq->tasks_timeline)) 502 MIN_vruntime = (__pick_first_entity(cfs_rq))->vruntime; 503 last = __pick_last_entity(cfs_rq); 504 if (last) 505 max_vruntime = last->vruntime; 506 min_vruntime = cfs_rq->min_vruntime; 507 rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime; 508 raw_spin_unlock_irqrestore(&rq->lock, flags); 509 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "MIN_vruntime", 510 SPLIT_NS(MIN_vruntime)); 511 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "min_vruntime", 512 SPLIT_NS(min_vruntime)); 513 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "max_vruntime", 514 SPLIT_NS(max_vruntime)); 515 spread = max_vruntime - MIN_vruntime; 516 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread", 517 SPLIT_NS(spread)); 518 spread0 = min_vruntime - rq0_min_vruntime; 519 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread0", 520 SPLIT_NS(spread0)); 521 SEQ_printf(m, " .%-30s: %d\n", "nr_spread_over", 522 cfs_rq->nr_spread_over); 523 SEQ_printf(m, " .%-30s: %d\n", "nr_running", cfs_rq->nr_running); 524 SEQ_printf(m, " .%-30s: %ld\n", "load", cfs_rq->load.weight); 525 #ifdef CONFIG_SMP 526 SEQ_printf(m, " .%-30s: %lu\n", "load_avg", 527 cfs_rq->avg.load_avg); 528 SEQ_printf(m, " .%-30s: %lu\n", "runnable_avg", 529 cfs_rq->avg.runnable_avg); 530 SEQ_printf(m, " .%-30s: %lu\n", "util_avg", 531 cfs_rq->avg.util_avg); 532 SEQ_printf(m, " .%-30s: %u\n", "util_est_enqueued", 533 cfs_rq->avg.util_est.enqueued); 534 SEQ_printf(m, " .%-30s: %ld\n", "removed.load_avg", 535 cfs_rq->removed.load_avg); 536 SEQ_printf(m, " .%-30s: %ld\n", "removed.util_avg", 537 cfs_rq->removed.util_avg); 538 SEQ_printf(m, " .%-30s: %ld\n", "removed.runnable_avg", 539 cfs_rq->removed.runnable_avg); 540 #ifdef CONFIG_FAIR_GROUP_SCHED 541 SEQ_printf(m, " .%-30s: %lu\n", "tg_load_avg_contrib", 542 cfs_rq->tg_load_avg_contrib); 543 SEQ_printf(m, " .%-30s: %ld\n", "tg_load_avg", 544 atomic_long_read(&cfs_rq->tg->load_avg)); 545 #endif 546 #endif 547 #ifdef CONFIG_CFS_BANDWIDTH 548 SEQ_printf(m, " .%-30s: %d\n", "throttled", 549 cfs_rq->throttled); 550 SEQ_printf(m, " .%-30s: %d\n", "throttle_count", 551 cfs_rq->throttle_count); 552 #endif 553 554 #ifdef CONFIG_FAIR_GROUP_SCHED 555 print_cfs_group_stats(m, cpu, cfs_rq->tg); 556 #endif 557 } 558 559 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq) 560 { 561 #ifdef CONFIG_RT_GROUP_SCHED 562 SEQ_printf(m, "\n"); 563 SEQ_printf(m, "rt_rq[%d]:%s\n", cpu, task_group_path(rt_rq->tg)); 564 #else 565 SEQ_printf(m, "\n"); 566 SEQ_printf(m, "rt_rq[%d]:\n", cpu); 567 #endif 568 569 #define P(x) \ 570 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rt_rq->x)) 571 #define PU(x) \ 572 SEQ_printf(m, " .%-30s: %lu\n", #x, (unsigned long)(rt_rq->x)) 573 #define PN(x) \ 574 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x)) 575 576 PU(rt_nr_running); 577 #ifdef CONFIG_SMP 578 PU(rt_nr_migratory); 579 #endif 580 P(rt_throttled); 581 PN(rt_time); 582 PN(rt_runtime); 583 584 #undef PN 585 #undef PU 586 #undef P 587 } 588 589 void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq) 590 { 591 struct dl_bw *dl_bw; 592 593 SEQ_printf(m, "\n"); 594 SEQ_printf(m, "dl_rq[%d]:\n", cpu); 595 596 #define PU(x) \ 597 SEQ_printf(m, " .%-30s: %lu\n", #x, (unsigned long)(dl_rq->x)) 598 599 PU(dl_nr_running); 600 #ifdef CONFIG_SMP 601 PU(dl_nr_migratory); 602 dl_bw = &cpu_rq(cpu)->rd->dl_bw; 603 #else 604 dl_bw = &dl_rq->dl_bw; 605 #endif 606 SEQ_printf(m, " .%-30s: %lld\n", "dl_bw->bw", dl_bw->bw); 607 SEQ_printf(m, " .%-30s: %lld\n", "dl_bw->total_bw", dl_bw->total_bw); 608 609 #undef PU 610 } 611 612 static void print_cpu(struct seq_file *m, int cpu) 613 { 614 struct rq *rq = cpu_rq(cpu); 615 unsigned long flags; 616 617 #ifdef CONFIG_X86 618 { 619 unsigned int freq = cpu_khz ? : 1; 620 621 SEQ_printf(m, "cpu#%d, %u.%03u MHz\n", 622 cpu, freq / 1000, (freq % 1000)); 623 } 624 #else 625 SEQ_printf(m, "cpu#%d\n", cpu); 626 #endif 627 628 #define P(x) \ 629 do { \ 630 if (sizeof(rq->x) == 4) \ 631 SEQ_printf(m, " .%-30s: %ld\n", #x, (long)(rq->x)); \ 632 else \ 633 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rq->x));\ 634 } while (0) 635 636 #define PN(x) \ 637 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x)) 638 639 P(nr_running); 640 P(nr_switches); 641 P(nr_load_updates); 642 P(nr_uninterruptible); 643 PN(next_balance); 644 SEQ_printf(m, " .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr))); 645 PN(clock); 646 PN(clock_task); 647 #undef P 648 #undef PN 649 650 #ifdef CONFIG_SMP 651 #define P64(n) SEQ_printf(m, " .%-30s: %Ld\n", #n, rq->n); 652 P64(avg_idle); 653 P64(max_idle_balance_cost); 654 #undef P64 655 #endif 656 657 #define P(n) SEQ_printf(m, " .%-30s: %d\n", #n, schedstat_val(rq->n)); 658 if (schedstat_enabled()) { 659 P(yld_count); 660 P(sched_count); 661 P(sched_goidle); 662 P(ttwu_count); 663 P(ttwu_local); 664 } 665 #undef P 666 667 spin_lock_irqsave(&sched_debug_lock, flags); 668 print_cfs_stats(m, cpu); 669 print_rt_stats(m, cpu); 670 print_dl_stats(m, cpu); 671 672 print_rq(m, rq, cpu); 673 spin_unlock_irqrestore(&sched_debug_lock, flags); 674 SEQ_printf(m, "\n"); 675 } 676 677 static const char *sched_tunable_scaling_names[] = { 678 "none", 679 "logarithmic", 680 "linear" 681 }; 682 683 static void sched_debug_header(struct seq_file *m) 684 { 685 u64 ktime, sched_clk, cpu_clk; 686 unsigned long flags; 687 688 local_irq_save(flags); 689 ktime = ktime_to_ns(ktime_get()); 690 sched_clk = sched_clock(); 691 cpu_clk = local_clock(); 692 local_irq_restore(flags); 693 694 SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n", 695 init_utsname()->release, 696 (int)strcspn(init_utsname()->version, " "), 697 init_utsname()->version); 698 699 #define P(x) \ 700 SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x)) 701 #define PN(x) \ 702 SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x)) 703 PN(ktime); 704 PN(sched_clk); 705 PN(cpu_clk); 706 P(jiffies); 707 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK 708 P(sched_clock_stable()); 709 #endif 710 #undef PN 711 #undef P 712 713 SEQ_printf(m, "\n"); 714 SEQ_printf(m, "sysctl_sched\n"); 715 716 #define P(x) \ 717 SEQ_printf(m, " .%-40s: %Ld\n", #x, (long long)(x)) 718 #define PN(x) \ 719 SEQ_printf(m, " .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x)) 720 PN(sysctl_sched_latency); 721 PN(sysctl_sched_min_granularity); 722 PN(sysctl_sched_wakeup_granularity); 723 P(sysctl_sched_child_runs_first); 724 P(sysctl_sched_features); 725 #undef PN 726 #undef P 727 728 SEQ_printf(m, " .%-40s: %d (%s)\n", 729 "sysctl_sched_tunable_scaling", 730 sysctl_sched_tunable_scaling, 731 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]); 732 SEQ_printf(m, "\n"); 733 } 734 735 static int sched_debug_show(struct seq_file *m, void *v) 736 { 737 int cpu = (unsigned long)(v - 2); 738 739 if (cpu != -1) 740 print_cpu(m, cpu); 741 else 742 sched_debug_header(m); 743 744 return 0; 745 } 746 747 void sysrq_sched_debug_show(void) 748 { 749 int cpu; 750 751 sched_debug_header(NULL); 752 for_each_online_cpu(cpu) { 753 /* 754 * Need to reset softlockup watchdogs on all CPUs, because 755 * another CPU might be blocked waiting for us to process 756 * an IPI or stop_machine. 757 */ 758 touch_nmi_watchdog(); 759 touch_all_softlockup_watchdogs(); 760 print_cpu(NULL, cpu); 761 } 762 } 763 764 /* 765 * This itererator needs some explanation. 766 * It returns 1 for the header position. 767 * This means 2 is CPU 0. 768 * In a hotplugged system some CPUs, including CPU 0, may be missing so we have 769 * to use cpumask_* to iterate over the CPUs. 770 */ 771 static void *sched_debug_start(struct seq_file *file, loff_t *offset) 772 { 773 unsigned long n = *offset; 774 775 if (n == 0) 776 return (void *) 1; 777 778 n--; 779 780 if (n > 0) 781 n = cpumask_next(n - 1, cpu_online_mask); 782 else 783 n = cpumask_first(cpu_online_mask); 784 785 *offset = n + 1; 786 787 if (n < nr_cpu_ids) 788 return (void *)(unsigned long)(n + 2); 789 790 return NULL; 791 } 792 793 static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset) 794 { 795 (*offset)++; 796 return sched_debug_start(file, offset); 797 } 798 799 static void sched_debug_stop(struct seq_file *file, void *data) 800 { 801 } 802 803 static const struct seq_operations sched_debug_sops = { 804 .start = sched_debug_start, 805 .next = sched_debug_next, 806 .stop = sched_debug_stop, 807 .show = sched_debug_show, 808 }; 809 810 static int __init init_sched_debug_procfs(void) 811 { 812 if (!proc_create_seq("sched_debug", 0444, NULL, &sched_debug_sops)) 813 return -ENOMEM; 814 return 0; 815 } 816 817 __initcall(init_sched_debug_procfs); 818 819 #define __PS(S, F) SEQ_printf(m, "%-45s:%21Ld\n", S, (long long)(F)) 820 #define __P(F) __PS(#F, F) 821 #define P(F) __PS(#F, p->F) 822 #define __PSN(S, F) SEQ_printf(m, "%-45s:%14Ld.%06ld\n", S, SPLIT_NS((long long)(F))) 823 #define __PN(F) __PSN(#F, F) 824 #define PN(F) __PSN(#F, 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"); 873 874 #define P_SCHEDSTAT(F) __PS(#F, schedstat_val(p->F)) 875 #define PN_SCHEDSTAT(F) __PSN(#F, schedstat_val(p->F)) 876 877 PN(se.exec_start); 878 PN(se.vruntime); 879 PN(se.sum_exec_runtime); 880 881 nr_switches = p->nvcsw + p->nivcsw; 882 883 P(se.nr_migrations); 884 885 if (schedstat_enabled()) { 886 u64 avg_atom, avg_per_cpu; 887 888 PN_SCHEDSTAT(se.statistics.sum_sleep_runtime); 889 PN_SCHEDSTAT(se.statistics.wait_start); 890 PN_SCHEDSTAT(se.statistics.sleep_start); 891 PN_SCHEDSTAT(se.statistics.block_start); 892 PN_SCHEDSTAT(se.statistics.sleep_max); 893 PN_SCHEDSTAT(se.statistics.block_max); 894 PN_SCHEDSTAT(se.statistics.exec_max); 895 PN_SCHEDSTAT(se.statistics.slice_max); 896 PN_SCHEDSTAT(se.statistics.wait_max); 897 PN_SCHEDSTAT(se.statistics.wait_sum); 898 P_SCHEDSTAT(se.statistics.wait_count); 899 PN_SCHEDSTAT(se.statistics.iowait_sum); 900 P_SCHEDSTAT(se.statistics.iowait_count); 901 P_SCHEDSTAT(se.statistics.nr_migrations_cold); 902 P_SCHEDSTAT(se.statistics.nr_failed_migrations_affine); 903 P_SCHEDSTAT(se.statistics.nr_failed_migrations_running); 904 P_SCHEDSTAT(se.statistics.nr_failed_migrations_hot); 905 P_SCHEDSTAT(se.statistics.nr_forced_migrations); 906 P_SCHEDSTAT(se.statistics.nr_wakeups); 907 P_SCHEDSTAT(se.statistics.nr_wakeups_sync); 908 P_SCHEDSTAT(se.statistics.nr_wakeups_migrate); 909 P_SCHEDSTAT(se.statistics.nr_wakeups_local); 910 P_SCHEDSTAT(se.statistics.nr_wakeups_remote); 911 P_SCHEDSTAT(se.statistics.nr_wakeups_affine); 912 P_SCHEDSTAT(se.statistics.nr_wakeups_affine_attempts); 913 P_SCHEDSTAT(se.statistics.nr_wakeups_passive); 914 P_SCHEDSTAT(se.statistics.nr_wakeups_idle); 915 916 avg_atom = p->se.sum_exec_runtime; 917 if (nr_switches) 918 avg_atom = div64_ul(avg_atom, nr_switches); 919 else 920 avg_atom = -1LL; 921 922 avg_per_cpu = p->se.sum_exec_runtime; 923 if (p->se.nr_migrations) { 924 avg_per_cpu = div64_u64(avg_per_cpu, 925 p->se.nr_migrations); 926 } else { 927 avg_per_cpu = -1LL; 928 } 929 930 __PN(avg_atom); 931 __PN(avg_per_cpu); 932 } 933 934 __P(nr_switches); 935 __PS("nr_voluntary_switches", p->nvcsw); 936 __PS("nr_involuntary_switches", p->nivcsw); 937 938 P(se.load.weight); 939 #ifdef CONFIG_SMP 940 P(se.avg.load_sum); 941 P(se.avg.runnable_sum); 942 P(se.avg.util_sum); 943 P(se.avg.load_avg); 944 P(se.avg.runnable_avg); 945 P(se.avg.util_avg); 946 P(se.avg.last_update_time); 947 P(se.avg.util_est.ewma); 948 P(se.avg.util_est.enqueued); 949 #endif 950 #ifdef CONFIG_UCLAMP_TASK 951 __PS("uclamp.min", p->uclamp[UCLAMP_MIN].value); 952 __PS("uclamp.max", p->uclamp[UCLAMP_MAX].value); 953 __PS("effective uclamp.min", uclamp_eff_value(p, UCLAMP_MIN)); 954 __PS("effective uclamp.max", uclamp_eff_value(p, UCLAMP_MAX)); 955 #endif 956 P(policy); 957 P(prio); 958 if (task_has_dl_policy(p)) { 959 P(dl.runtime); 960 P(dl.deadline); 961 } 962 #undef PN_SCHEDSTAT 963 #undef P_SCHEDSTAT 964 965 { 966 unsigned int this_cpu = raw_smp_processor_id(); 967 u64 t0, t1; 968 969 t0 = cpu_clock(this_cpu); 970 t1 = cpu_clock(this_cpu); 971 __PS("clock-delta", t1-t0); 972 } 973 974 sched_show_numa(p, m); 975 } 976 977 void proc_sched_set_task(struct task_struct *p) 978 { 979 #ifdef CONFIG_SCHEDSTATS 980 memset(&p->se.statistics, 0, sizeof(p->se.statistics)); 981 #endif 982 } 983