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 10 /* 11 * This allows printing both to /sys/kernel/debug/sched/debug and 12 * to the console 13 */ 14 #define SEQ_printf(m, x...) \ 15 do { \ 16 if (m) \ 17 seq_printf(m, x); \ 18 else \ 19 pr_cont(x); \ 20 } while (0) 21 22 /* 23 * Ease the printing of nsec fields: 24 */ 25 static long long nsec_high(unsigned long long nsec) 26 { 27 if ((long long)nsec < 0) { 28 nsec = -nsec; 29 do_div(nsec, 1000000); 30 return -nsec; 31 } 32 do_div(nsec, 1000000); 33 34 return nsec; 35 } 36 37 static unsigned long nsec_low(unsigned long long nsec) 38 { 39 if ((long long)nsec < 0) 40 nsec = -nsec; 41 42 return do_div(nsec, 1000000); 43 } 44 45 #define SPLIT_NS(x) nsec_high(x), nsec_low(x) 46 47 #define SCHED_FEAT(name, enabled) \ 48 #name , 49 50 static const char * const sched_feat_names[] = { 51 #include "features.h" 52 }; 53 54 #undef SCHED_FEAT 55 56 static int sched_feat_show(struct seq_file *m, void *v) 57 { 58 int i; 59 60 for (i = 0; i < __SCHED_FEAT_NR; i++) { 61 if (!(sysctl_sched_features & (1UL << i))) 62 seq_puts(m, "NO_"); 63 seq_printf(m, "%s ", sched_feat_names[i]); 64 } 65 seq_puts(m, "\n"); 66 67 return 0; 68 } 69 70 #ifdef CONFIG_JUMP_LABEL 71 72 #define jump_label_key__true STATIC_KEY_INIT_TRUE 73 #define jump_label_key__false STATIC_KEY_INIT_FALSE 74 75 #define SCHED_FEAT(name, enabled) \ 76 jump_label_key__##enabled , 77 78 struct static_key sched_feat_keys[__SCHED_FEAT_NR] = { 79 #include "features.h" 80 }; 81 82 #undef SCHED_FEAT 83 84 static void sched_feat_disable(int i) 85 { 86 static_key_disable_cpuslocked(&sched_feat_keys[i]); 87 } 88 89 static void sched_feat_enable(int i) 90 { 91 static_key_enable_cpuslocked(&sched_feat_keys[i]); 92 } 93 #else 94 static void sched_feat_disable(int i) { }; 95 static void sched_feat_enable(int i) { }; 96 #endif /* CONFIG_JUMP_LABEL */ 97 98 static int sched_feat_set(char *cmp) 99 { 100 int i; 101 int neg = 0; 102 103 if (strncmp(cmp, "NO_", 3) == 0) { 104 neg = 1; 105 cmp += 3; 106 } 107 108 i = match_string(sched_feat_names, __SCHED_FEAT_NR, cmp); 109 if (i < 0) 110 return i; 111 112 if (neg) { 113 sysctl_sched_features &= ~(1UL << i); 114 sched_feat_disable(i); 115 } else { 116 sysctl_sched_features |= (1UL << i); 117 sched_feat_enable(i); 118 } 119 120 return 0; 121 } 122 123 static ssize_t 124 sched_feat_write(struct file *filp, const char __user *ubuf, 125 size_t cnt, loff_t *ppos) 126 { 127 char buf[64]; 128 char *cmp; 129 int ret; 130 struct inode *inode; 131 132 if (cnt > 63) 133 cnt = 63; 134 135 if (copy_from_user(&buf, ubuf, cnt)) 136 return -EFAULT; 137 138 buf[cnt] = 0; 139 cmp = strstrip(buf); 140 141 /* Ensure the static_key remains in a consistent state */ 142 inode = file_inode(filp); 143 cpus_read_lock(); 144 inode_lock(inode); 145 ret = sched_feat_set(cmp); 146 inode_unlock(inode); 147 cpus_read_unlock(); 148 if (ret < 0) 149 return ret; 150 151 *ppos += cnt; 152 153 return cnt; 154 } 155 156 static int sched_feat_open(struct inode *inode, struct file *filp) 157 { 158 return single_open(filp, sched_feat_show, NULL); 159 } 160 161 static const struct file_operations sched_feat_fops = { 162 .open = sched_feat_open, 163 .write = sched_feat_write, 164 .read = seq_read, 165 .llseek = seq_lseek, 166 .release = single_release, 167 }; 168 169 #ifdef CONFIG_SMP 170 171 static ssize_t sched_scaling_write(struct file *filp, const char __user *ubuf, 172 size_t cnt, loff_t *ppos) 173 { 174 char buf[16]; 175 unsigned int scaling; 176 177 if (cnt > 15) 178 cnt = 15; 179 180 if (copy_from_user(&buf, ubuf, cnt)) 181 return -EFAULT; 182 buf[cnt] = '\0'; 183 184 if (kstrtouint(buf, 10, &scaling)) 185 return -EINVAL; 186 187 if (scaling >= SCHED_TUNABLESCALING_END) 188 return -EINVAL; 189 190 sysctl_sched_tunable_scaling = scaling; 191 if (sched_update_scaling()) 192 return -EINVAL; 193 194 *ppos += cnt; 195 return cnt; 196 } 197 198 static int sched_scaling_show(struct seq_file *m, void *v) 199 { 200 seq_printf(m, "%d\n", sysctl_sched_tunable_scaling); 201 return 0; 202 } 203 204 static int sched_scaling_open(struct inode *inode, struct file *filp) 205 { 206 return single_open(filp, sched_scaling_show, NULL); 207 } 208 209 static const struct file_operations sched_scaling_fops = { 210 .open = sched_scaling_open, 211 .write = sched_scaling_write, 212 .read = seq_read, 213 .llseek = seq_lseek, 214 .release = single_release, 215 }; 216 217 #endif /* SMP */ 218 219 #ifdef CONFIG_PREEMPT_DYNAMIC 220 221 static ssize_t sched_dynamic_write(struct file *filp, const char __user *ubuf, 222 size_t cnt, loff_t *ppos) 223 { 224 char buf[16]; 225 int mode; 226 227 if (cnt > 15) 228 cnt = 15; 229 230 if (copy_from_user(&buf, ubuf, cnt)) 231 return -EFAULT; 232 233 buf[cnt] = 0; 234 mode = sched_dynamic_mode(strstrip(buf)); 235 if (mode < 0) 236 return mode; 237 238 sched_dynamic_update(mode); 239 240 *ppos += cnt; 241 242 return cnt; 243 } 244 245 static int sched_dynamic_show(struct seq_file *m, void *v) 246 { 247 static const char * preempt_modes[] = { 248 "none", "voluntary", "full", "lazy", 249 }; 250 int j = ARRAY_SIZE(preempt_modes) - !IS_ENABLED(CONFIG_ARCH_HAS_PREEMPT_LAZY); 251 int i = IS_ENABLED(CONFIG_PREEMPT_RT) * 2; 252 253 for (; i < j; i++) { 254 if (preempt_dynamic_mode == i) 255 seq_puts(m, "("); 256 seq_puts(m, preempt_modes[i]); 257 if (preempt_dynamic_mode == i) 258 seq_puts(m, ")"); 259 260 seq_puts(m, " "); 261 } 262 263 seq_puts(m, "\n"); 264 return 0; 265 } 266 267 static int sched_dynamic_open(struct inode *inode, struct file *filp) 268 { 269 return single_open(filp, sched_dynamic_show, NULL); 270 } 271 272 static const struct file_operations sched_dynamic_fops = { 273 .open = sched_dynamic_open, 274 .write = sched_dynamic_write, 275 .read = seq_read, 276 .llseek = seq_lseek, 277 .release = single_release, 278 }; 279 280 #endif /* CONFIG_PREEMPT_DYNAMIC */ 281 282 __read_mostly bool sched_debug_verbose; 283 284 #ifdef CONFIG_SMP 285 static struct dentry *sd_dentry; 286 287 288 static ssize_t sched_verbose_write(struct file *filp, const char __user *ubuf, 289 size_t cnt, loff_t *ppos) 290 { 291 ssize_t result; 292 bool orig; 293 294 cpus_read_lock(); 295 mutex_lock(&sched_domains_mutex); 296 297 orig = sched_debug_verbose; 298 result = debugfs_write_file_bool(filp, ubuf, cnt, ppos); 299 300 if (sched_debug_verbose && !orig) 301 update_sched_domain_debugfs(); 302 else if (!sched_debug_verbose && orig) { 303 debugfs_remove(sd_dentry); 304 sd_dentry = NULL; 305 } 306 307 mutex_unlock(&sched_domains_mutex); 308 cpus_read_unlock(); 309 310 return result; 311 } 312 #else 313 #define sched_verbose_write debugfs_write_file_bool 314 #endif 315 316 static const struct file_operations sched_verbose_fops = { 317 .read = debugfs_read_file_bool, 318 .write = sched_verbose_write, 319 .open = simple_open, 320 .llseek = default_llseek, 321 }; 322 323 static const struct seq_operations sched_debug_sops; 324 325 static int sched_debug_open(struct inode *inode, struct file *filp) 326 { 327 return seq_open(filp, &sched_debug_sops); 328 } 329 330 static const struct file_operations sched_debug_fops = { 331 .open = sched_debug_open, 332 .read = seq_read, 333 .llseek = seq_lseek, 334 .release = seq_release, 335 }; 336 337 enum dl_param { 338 DL_RUNTIME = 0, 339 DL_PERIOD, 340 }; 341 342 static unsigned long fair_server_period_max = (1UL << 22) * NSEC_PER_USEC; /* ~4 seconds */ 343 static unsigned long fair_server_period_min = (100) * NSEC_PER_USEC; /* 100 us */ 344 345 static ssize_t sched_fair_server_write(struct file *filp, const char __user *ubuf, 346 size_t cnt, loff_t *ppos, enum dl_param param) 347 { 348 long cpu = (long) ((struct seq_file *) filp->private_data)->private; 349 struct rq *rq = cpu_rq(cpu); 350 u64 runtime, period; 351 size_t err; 352 int retval; 353 u64 value; 354 355 err = kstrtoull_from_user(ubuf, cnt, 10, &value); 356 if (err) 357 return err; 358 359 scoped_guard (rq_lock_irqsave, rq) { 360 runtime = rq->fair_server.dl_runtime; 361 period = rq->fair_server.dl_period; 362 363 switch (param) { 364 case DL_RUNTIME: 365 if (runtime == value) 366 break; 367 runtime = value; 368 break; 369 case DL_PERIOD: 370 if (value == period) 371 break; 372 period = value; 373 break; 374 } 375 376 if (runtime > period || 377 period > fair_server_period_max || 378 period < fair_server_period_min) { 379 return -EINVAL; 380 } 381 382 if (rq->cfs.h_nr_running) { 383 update_rq_clock(rq); 384 dl_server_stop(&rq->fair_server); 385 } 386 387 retval = dl_server_apply_params(&rq->fair_server, runtime, period, 0); 388 if (retval) 389 cnt = retval; 390 391 if (!runtime) 392 printk_deferred("Fair server disabled in CPU %d, system may crash due to starvation.\n", 393 cpu_of(rq)); 394 395 if (rq->cfs.h_nr_running) 396 dl_server_start(&rq->fair_server); 397 } 398 399 *ppos += cnt; 400 return cnt; 401 } 402 403 static size_t sched_fair_server_show(struct seq_file *m, void *v, enum dl_param param) 404 { 405 unsigned long cpu = (unsigned long) m->private; 406 struct rq *rq = cpu_rq(cpu); 407 u64 value; 408 409 switch (param) { 410 case DL_RUNTIME: 411 value = rq->fair_server.dl_runtime; 412 break; 413 case DL_PERIOD: 414 value = rq->fair_server.dl_period; 415 break; 416 } 417 418 seq_printf(m, "%llu\n", value); 419 return 0; 420 421 } 422 423 static ssize_t 424 sched_fair_server_runtime_write(struct file *filp, const char __user *ubuf, 425 size_t cnt, loff_t *ppos) 426 { 427 return sched_fair_server_write(filp, ubuf, cnt, ppos, DL_RUNTIME); 428 } 429 430 static int sched_fair_server_runtime_show(struct seq_file *m, void *v) 431 { 432 return sched_fair_server_show(m, v, DL_RUNTIME); 433 } 434 435 static int sched_fair_server_runtime_open(struct inode *inode, struct file *filp) 436 { 437 return single_open(filp, sched_fair_server_runtime_show, inode->i_private); 438 } 439 440 static const struct file_operations fair_server_runtime_fops = { 441 .open = sched_fair_server_runtime_open, 442 .write = sched_fair_server_runtime_write, 443 .read = seq_read, 444 .llseek = seq_lseek, 445 .release = single_release, 446 }; 447 448 static ssize_t 449 sched_fair_server_period_write(struct file *filp, const char __user *ubuf, 450 size_t cnt, loff_t *ppos) 451 { 452 return sched_fair_server_write(filp, ubuf, cnt, ppos, DL_PERIOD); 453 } 454 455 static int sched_fair_server_period_show(struct seq_file *m, void *v) 456 { 457 return sched_fair_server_show(m, v, DL_PERIOD); 458 } 459 460 static int sched_fair_server_period_open(struct inode *inode, struct file *filp) 461 { 462 return single_open(filp, sched_fair_server_period_show, inode->i_private); 463 } 464 465 static const struct file_operations fair_server_period_fops = { 466 .open = sched_fair_server_period_open, 467 .write = sched_fair_server_period_write, 468 .read = seq_read, 469 .llseek = seq_lseek, 470 .release = single_release, 471 }; 472 473 static struct dentry *debugfs_sched; 474 475 static void debugfs_fair_server_init(void) 476 { 477 struct dentry *d_fair; 478 unsigned long cpu; 479 480 d_fair = debugfs_create_dir("fair_server", debugfs_sched); 481 if (!d_fair) 482 return; 483 484 for_each_possible_cpu(cpu) { 485 struct dentry *d_cpu; 486 char buf[32]; 487 488 snprintf(buf, sizeof(buf), "cpu%lu", cpu); 489 d_cpu = debugfs_create_dir(buf, d_fair); 490 491 debugfs_create_file("runtime", 0644, d_cpu, (void *) cpu, &fair_server_runtime_fops); 492 debugfs_create_file("period", 0644, d_cpu, (void *) cpu, &fair_server_period_fops); 493 } 494 } 495 496 static __init int sched_init_debug(void) 497 { 498 struct dentry __maybe_unused *numa; 499 500 debugfs_sched = debugfs_create_dir("sched", NULL); 501 502 debugfs_create_file("features", 0644, debugfs_sched, NULL, &sched_feat_fops); 503 debugfs_create_file_unsafe("verbose", 0644, debugfs_sched, &sched_debug_verbose, &sched_verbose_fops); 504 #ifdef CONFIG_PREEMPT_DYNAMIC 505 debugfs_create_file("preempt", 0644, debugfs_sched, NULL, &sched_dynamic_fops); 506 #endif 507 508 debugfs_create_u32("base_slice_ns", 0644, debugfs_sched, &sysctl_sched_base_slice); 509 510 debugfs_create_u32("latency_warn_ms", 0644, debugfs_sched, &sysctl_resched_latency_warn_ms); 511 debugfs_create_u32("latency_warn_once", 0644, debugfs_sched, &sysctl_resched_latency_warn_once); 512 513 #ifdef CONFIG_SMP 514 debugfs_create_file("tunable_scaling", 0644, debugfs_sched, NULL, &sched_scaling_fops); 515 debugfs_create_u32("migration_cost_ns", 0644, debugfs_sched, &sysctl_sched_migration_cost); 516 debugfs_create_u32("nr_migrate", 0644, debugfs_sched, &sysctl_sched_nr_migrate); 517 518 mutex_lock(&sched_domains_mutex); 519 update_sched_domain_debugfs(); 520 mutex_unlock(&sched_domains_mutex); 521 #endif 522 523 #ifdef CONFIG_NUMA_BALANCING 524 numa = debugfs_create_dir("numa_balancing", debugfs_sched); 525 526 debugfs_create_u32("scan_delay_ms", 0644, numa, &sysctl_numa_balancing_scan_delay); 527 debugfs_create_u32("scan_period_min_ms", 0644, numa, &sysctl_numa_balancing_scan_period_min); 528 debugfs_create_u32("scan_period_max_ms", 0644, numa, &sysctl_numa_balancing_scan_period_max); 529 debugfs_create_u32("scan_size_mb", 0644, numa, &sysctl_numa_balancing_scan_size); 530 debugfs_create_u32("hot_threshold_ms", 0644, numa, &sysctl_numa_balancing_hot_threshold); 531 #endif 532 533 debugfs_create_file("debug", 0444, debugfs_sched, NULL, &sched_debug_fops); 534 535 debugfs_fair_server_init(); 536 537 return 0; 538 } 539 late_initcall(sched_init_debug); 540 541 #ifdef CONFIG_SMP 542 543 static cpumask_var_t sd_sysctl_cpus; 544 545 static int sd_flags_show(struct seq_file *m, void *v) 546 { 547 unsigned long flags = *(unsigned int *)m->private; 548 int idx; 549 550 for_each_set_bit(idx, &flags, __SD_FLAG_CNT) { 551 seq_puts(m, sd_flag_debug[idx].name); 552 seq_puts(m, " "); 553 } 554 seq_puts(m, "\n"); 555 556 return 0; 557 } 558 559 static int sd_flags_open(struct inode *inode, struct file *file) 560 { 561 return single_open(file, sd_flags_show, inode->i_private); 562 } 563 564 static const struct file_operations sd_flags_fops = { 565 .open = sd_flags_open, 566 .read = seq_read, 567 .llseek = seq_lseek, 568 .release = single_release, 569 }; 570 571 static void register_sd(struct sched_domain *sd, struct dentry *parent) 572 { 573 #define SDM(type, mode, member) \ 574 debugfs_create_##type(#member, mode, parent, &sd->member) 575 576 SDM(ulong, 0644, min_interval); 577 SDM(ulong, 0644, max_interval); 578 SDM(u64, 0644, max_newidle_lb_cost); 579 SDM(u32, 0644, busy_factor); 580 SDM(u32, 0644, imbalance_pct); 581 SDM(u32, 0644, cache_nice_tries); 582 SDM(str, 0444, name); 583 584 #undef SDM 585 586 debugfs_create_file("flags", 0444, parent, &sd->flags, &sd_flags_fops); 587 debugfs_create_file("groups_flags", 0444, parent, &sd->groups->flags, &sd_flags_fops); 588 debugfs_create_u32("level", 0444, parent, (u32 *)&sd->level); 589 } 590 591 void update_sched_domain_debugfs(void) 592 { 593 int cpu, i; 594 595 /* 596 * This can unfortunately be invoked before sched_debug_init() creates 597 * the debug directory. Don't touch sd_sysctl_cpus until then. 598 */ 599 if (!debugfs_sched) 600 return; 601 602 if (!sched_debug_verbose) 603 return; 604 605 if (!cpumask_available(sd_sysctl_cpus)) { 606 if (!alloc_cpumask_var(&sd_sysctl_cpus, GFP_KERNEL)) 607 return; 608 cpumask_copy(sd_sysctl_cpus, cpu_possible_mask); 609 } 610 611 if (!sd_dentry) { 612 sd_dentry = debugfs_create_dir("domains", debugfs_sched); 613 614 /* rebuild sd_sysctl_cpus if empty since it gets cleared below */ 615 if (cpumask_empty(sd_sysctl_cpus)) 616 cpumask_copy(sd_sysctl_cpus, cpu_online_mask); 617 } 618 619 for_each_cpu(cpu, sd_sysctl_cpus) { 620 struct sched_domain *sd; 621 struct dentry *d_cpu; 622 char buf[32]; 623 624 snprintf(buf, sizeof(buf), "cpu%d", cpu); 625 debugfs_lookup_and_remove(buf, sd_dentry); 626 d_cpu = debugfs_create_dir(buf, sd_dentry); 627 628 i = 0; 629 for_each_domain(cpu, sd) { 630 struct dentry *d_sd; 631 632 snprintf(buf, sizeof(buf), "domain%d", i); 633 d_sd = debugfs_create_dir(buf, d_cpu); 634 635 register_sd(sd, d_sd); 636 i++; 637 } 638 639 __cpumask_clear_cpu(cpu, sd_sysctl_cpus); 640 } 641 } 642 643 void dirty_sched_domain_sysctl(int cpu) 644 { 645 if (cpumask_available(sd_sysctl_cpus)) 646 __cpumask_set_cpu(cpu, sd_sysctl_cpus); 647 } 648 649 #endif /* CONFIG_SMP */ 650 651 #ifdef CONFIG_FAIR_GROUP_SCHED 652 static void print_cfs_group_stats(struct seq_file *m, int cpu, struct task_group *tg) 653 { 654 struct sched_entity *se = tg->se[cpu]; 655 656 #define P(F) SEQ_printf(m, " .%-30s: %lld\n", #F, (long long)F) 657 #define P_SCHEDSTAT(F) SEQ_printf(m, " .%-30s: %lld\n", \ 658 #F, (long long)schedstat_val(stats->F)) 659 #define PN(F) SEQ_printf(m, " .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F)) 660 #define PN_SCHEDSTAT(F) SEQ_printf(m, " .%-30s: %lld.%06ld\n", \ 661 #F, SPLIT_NS((long long)schedstat_val(stats->F))) 662 663 if (!se) 664 return; 665 666 PN(se->exec_start); 667 PN(se->vruntime); 668 PN(se->sum_exec_runtime); 669 670 if (schedstat_enabled()) { 671 struct sched_statistics *stats; 672 stats = __schedstats_from_se(se); 673 674 PN_SCHEDSTAT(wait_start); 675 PN_SCHEDSTAT(sleep_start); 676 PN_SCHEDSTAT(block_start); 677 PN_SCHEDSTAT(sleep_max); 678 PN_SCHEDSTAT(block_max); 679 PN_SCHEDSTAT(exec_max); 680 PN_SCHEDSTAT(slice_max); 681 PN_SCHEDSTAT(wait_max); 682 PN_SCHEDSTAT(wait_sum); 683 P_SCHEDSTAT(wait_count); 684 } 685 686 P(se->load.weight); 687 #ifdef CONFIG_SMP 688 P(se->avg.load_avg); 689 P(se->avg.util_avg); 690 P(se->avg.runnable_avg); 691 #endif 692 693 #undef PN_SCHEDSTAT 694 #undef PN 695 #undef P_SCHEDSTAT 696 #undef P 697 } 698 #endif 699 700 #ifdef CONFIG_CGROUP_SCHED 701 static DEFINE_SPINLOCK(sched_debug_lock); 702 static char group_path[PATH_MAX]; 703 704 static void task_group_path(struct task_group *tg, char *path, int plen) 705 { 706 if (autogroup_path(tg, path, plen)) 707 return; 708 709 cgroup_path(tg->css.cgroup, path, plen); 710 } 711 712 /* 713 * Only 1 SEQ_printf_task_group_path() caller can use the full length 714 * group_path[] for cgroup path. Other simultaneous callers will have 715 * to use a shorter stack buffer. A "..." suffix is appended at the end 716 * of the stack buffer so that it will show up in case the output length 717 * matches the given buffer size to indicate possible path name truncation. 718 */ 719 #define SEQ_printf_task_group_path(m, tg, fmt...) \ 720 { \ 721 if (spin_trylock(&sched_debug_lock)) { \ 722 task_group_path(tg, group_path, sizeof(group_path)); \ 723 SEQ_printf(m, fmt, group_path); \ 724 spin_unlock(&sched_debug_lock); \ 725 } else { \ 726 char buf[128]; \ 727 char *bufend = buf + sizeof(buf) - 3; \ 728 task_group_path(tg, buf, bufend - buf); \ 729 strcpy(bufend - 1, "..."); \ 730 SEQ_printf(m, fmt, buf); \ 731 } \ 732 } 733 #endif 734 735 static void 736 print_task(struct seq_file *m, struct rq *rq, struct task_struct *p) 737 { 738 if (task_current(rq, p)) 739 SEQ_printf(m, ">R"); 740 else 741 SEQ_printf(m, " %c", task_state_to_char(p)); 742 743 SEQ_printf(m, " %15s %5d %9Ld.%06ld %c %9Ld.%06ld %c %9Ld.%06ld %9Ld.%06ld %9Ld %5d ", 744 p->comm, task_pid_nr(p), 745 SPLIT_NS(p->se.vruntime), 746 entity_eligible(cfs_rq_of(&p->se), &p->se) ? 'E' : 'N', 747 SPLIT_NS(p->se.deadline), 748 p->se.custom_slice ? 'S' : ' ', 749 SPLIT_NS(p->se.slice), 750 SPLIT_NS(p->se.sum_exec_runtime), 751 (long long)(p->nvcsw + p->nivcsw), 752 p->prio); 753 754 SEQ_printf(m, "%9lld.%06ld %9lld.%06ld %9lld.%06ld", 755 SPLIT_NS(schedstat_val_or_zero(p->stats.wait_sum)), 756 SPLIT_NS(schedstat_val_or_zero(p->stats.sum_sleep_runtime)), 757 SPLIT_NS(schedstat_val_or_zero(p->stats.sum_block_runtime))); 758 759 #ifdef CONFIG_NUMA_BALANCING 760 SEQ_printf(m, " %d %d", task_node(p), task_numa_group_id(p)); 761 #endif 762 #ifdef CONFIG_CGROUP_SCHED 763 SEQ_printf_task_group_path(m, task_group(p), " %s") 764 #endif 765 766 SEQ_printf(m, "\n"); 767 } 768 769 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu) 770 { 771 struct task_struct *g, *p; 772 773 SEQ_printf(m, "\n"); 774 SEQ_printf(m, "runnable tasks:\n"); 775 SEQ_printf(m, " S task PID vruntime eligible " 776 "deadline slice sum-exec switches " 777 "prio wait-time sum-sleep sum-block" 778 #ifdef CONFIG_NUMA_BALANCING 779 " node group-id" 780 #endif 781 #ifdef CONFIG_CGROUP_SCHED 782 " group-path" 783 #endif 784 "\n"); 785 SEQ_printf(m, "-------------------------------------------------------" 786 "------------------------------------------------------" 787 "------------------------------------------------------" 788 #ifdef CONFIG_NUMA_BALANCING 789 "--------------" 790 #endif 791 #ifdef CONFIG_CGROUP_SCHED 792 "--------------" 793 #endif 794 "\n"); 795 796 rcu_read_lock(); 797 for_each_process_thread(g, p) { 798 if (task_cpu(p) != rq_cpu) 799 continue; 800 801 print_task(m, rq, p); 802 } 803 rcu_read_unlock(); 804 } 805 806 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq) 807 { 808 s64 left_vruntime = -1, min_vruntime, right_vruntime = -1, left_deadline = -1, spread; 809 struct sched_entity *last, *first, *root; 810 struct rq *rq = cpu_rq(cpu); 811 unsigned long flags; 812 813 #ifdef CONFIG_FAIR_GROUP_SCHED 814 SEQ_printf(m, "\n"); 815 SEQ_printf_task_group_path(m, cfs_rq->tg, "cfs_rq[%d]:%s\n", cpu); 816 #else 817 SEQ_printf(m, "\n"); 818 SEQ_printf(m, "cfs_rq[%d]:\n", cpu); 819 #endif 820 821 raw_spin_rq_lock_irqsave(rq, flags); 822 root = __pick_root_entity(cfs_rq); 823 if (root) 824 left_vruntime = root->min_vruntime; 825 first = __pick_first_entity(cfs_rq); 826 if (first) 827 left_deadline = first->deadline; 828 last = __pick_last_entity(cfs_rq); 829 if (last) 830 right_vruntime = last->vruntime; 831 min_vruntime = cfs_rq->min_vruntime; 832 raw_spin_rq_unlock_irqrestore(rq, flags); 833 834 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "left_deadline", 835 SPLIT_NS(left_deadline)); 836 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "left_vruntime", 837 SPLIT_NS(left_vruntime)); 838 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "min_vruntime", 839 SPLIT_NS(min_vruntime)); 840 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "avg_vruntime", 841 SPLIT_NS(avg_vruntime(cfs_rq))); 842 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "right_vruntime", 843 SPLIT_NS(right_vruntime)); 844 spread = right_vruntime - left_vruntime; 845 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread", SPLIT_NS(spread)); 846 SEQ_printf(m, " .%-30s: %d\n", "nr_running", cfs_rq->nr_running); 847 SEQ_printf(m, " .%-30s: %d\n", "h_nr_running", cfs_rq->h_nr_running); 848 SEQ_printf(m, " .%-30s: %d\n", "h_nr_delayed", cfs_rq->h_nr_delayed); 849 SEQ_printf(m, " .%-30s: %d\n", "idle_nr_running", 850 cfs_rq->idle_nr_running); 851 SEQ_printf(m, " .%-30s: %d\n", "idle_h_nr_running", 852 cfs_rq->idle_h_nr_running); 853 SEQ_printf(m, " .%-30s: %ld\n", "load", cfs_rq->load.weight); 854 #ifdef CONFIG_SMP 855 SEQ_printf(m, " .%-30s: %lu\n", "load_avg", 856 cfs_rq->avg.load_avg); 857 SEQ_printf(m, " .%-30s: %lu\n", "runnable_avg", 858 cfs_rq->avg.runnable_avg); 859 SEQ_printf(m, " .%-30s: %lu\n", "util_avg", 860 cfs_rq->avg.util_avg); 861 SEQ_printf(m, " .%-30s: %u\n", "util_est", 862 cfs_rq->avg.util_est); 863 SEQ_printf(m, " .%-30s: %ld\n", "removed.load_avg", 864 cfs_rq->removed.load_avg); 865 SEQ_printf(m, " .%-30s: %ld\n", "removed.util_avg", 866 cfs_rq->removed.util_avg); 867 SEQ_printf(m, " .%-30s: %ld\n", "removed.runnable_avg", 868 cfs_rq->removed.runnable_avg); 869 #ifdef CONFIG_FAIR_GROUP_SCHED 870 SEQ_printf(m, " .%-30s: %lu\n", "tg_load_avg_contrib", 871 cfs_rq->tg_load_avg_contrib); 872 SEQ_printf(m, " .%-30s: %ld\n", "tg_load_avg", 873 atomic_long_read(&cfs_rq->tg->load_avg)); 874 #endif 875 #endif 876 #ifdef CONFIG_CFS_BANDWIDTH 877 SEQ_printf(m, " .%-30s: %d\n", "throttled", 878 cfs_rq->throttled); 879 SEQ_printf(m, " .%-30s: %d\n", "throttle_count", 880 cfs_rq->throttle_count); 881 #endif 882 883 #ifdef CONFIG_FAIR_GROUP_SCHED 884 print_cfs_group_stats(m, cpu, cfs_rq->tg); 885 #endif 886 } 887 888 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq) 889 { 890 #ifdef CONFIG_RT_GROUP_SCHED 891 SEQ_printf(m, "\n"); 892 SEQ_printf_task_group_path(m, rt_rq->tg, "rt_rq[%d]:%s\n", cpu); 893 #else 894 SEQ_printf(m, "\n"); 895 SEQ_printf(m, "rt_rq[%d]:\n", cpu); 896 #endif 897 898 #define P(x) \ 899 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rt_rq->x)) 900 #define PU(x) \ 901 SEQ_printf(m, " .%-30s: %lu\n", #x, (unsigned long)(rt_rq->x)) 902 #define PN(x) \ 903 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x)) 904 905 PU(rt_nr_running); 906 907 #ifdef CONFIG_RT_GROUP_SCHED 908 P(rt_throttled); 909 PN(rt_time); 910 PN(rt_runtime); 911 #endif 912 913 #undef PN 914 #undef PU 915 #undef P 916 } 917 918 void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq) 919 { 920 struct dl_bw *dl_bw; 921 922 SEQ_printf(m, "\n"); 923 SEQ_printf(m, "dl_rq[%d]:\n", cpu); 924 925 #define PU(x) \ 926 SEQ_printf(m, " .%-30s: %lu\n", #x, (unsigned long)(dl_rq->x)) 927 928 PU(dl_nr_running); 929 #ifdef CONFIG_SMP 930 dl_bw = &cpu_rq(cpu)->rd->dl_bw; 931 #else 932 dl_bw = &dl_rq->dl_bw; 933 #endif 934 SEQ_printf(m, " .%-30s: %lld\n", "dl_bw->bw", dl_bw->bw); 935 SEQ_printf(m, " .%-30s: %lld\n", "dl_bw->total_bw", dl_bw->total_bw); 936 937 #undef PU 938 } 939 940 static void print_cpu(struct seq_file *m, int cpu) 941 { 942 struct rq *rq = cpu_rq(cpu); 943 944 #ifdef CONFIG_X86 945 { 946 unsigned int freq = cpu_khz ? : 1; 947 948 SEQ_printf(m, "cpu#%d, %u.%03u MHz\n", 949 cpu, freq / 1000, (freq % 1000)); 950 } 951 #else 952 SEQ_printf(m, "cpu#%d\n", cpu); 953 #endif 954 955 #define P(x) \ 956 do { \ 957 if (sizeof(rq->x) == 4) \ 958 SEQ_printf(m, " .%-30s: %d\n", #x, (int)(rq->x)); \ 959 else \ 960 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rq->x));\ 961 } while (0) 962 963 #define PN(x) \ 964 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x)) 965 966 P(nr_running); 967 P(nr_switches); 968 P(nr_uninterruptible); 969 PN(next_balance); 970 SEQ_printf(m, " .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr))); 971 PN(clock); 972 PN(clock_task); 973 #undef P 974 #undef PN 975 976 #ifdef CONFIG_SMP 977 #define P64(n) SEQ_printf(m, " .%-30s: %Ld\n", #n, rq->n); 978 P64(avg_idle); 979 P64(max_idle_balance_cost); 980 #undef P64 981 #endif 982 983 #define P(n) SEQ_printf(m, " .%-30s: %d\n", #n, schedstat_val(rq->n)); 984 if (schedstat_enabled()) { 985 P(yld_count); 986 P(sched_count); 987 P(sched_goidle); 988 P(ttwu_count); 989 P(ttwu_local); 990 } 991 #undef P 992 993 print_cfs_stats(m, cpu); 994 print_rt_stats(m, cpu); 995 print_dl_stats(m, cpu); 996 997 print_rq(m, rq, cpu); 998 SEQ_printf(m, "\n"); 999 } 1000 1001 static const char *sched_tunable_scaling_names[] = { 1002 "none", 1003 "logarithmic", 1004 "linear" 1005 }; 1006 1007 static void sched_debug_header(struct seq_file *m) 1008 { 1009 u64 ktime, sched_clk, cpu_clk; 1010 unsigned long flags; 1011 1012 local_irq_save(flags); 1013 ktime = ktime_to_ns(ktime_get()); 1014 sched_clk = sched_clock(); 1015 cpu_clk = local_clock(); 1016 local_irq_restore(flags); 1017 1018 SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n", 1019 init_utsname()->release, 1020 (int)strcspn(init_utsname()->version, " "), 1021 init_utsname()->version); 1022 1023 #define P(x) \ 1024 SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x)) 1025 #define PN(x) \ 1026 SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x)) 1027 PN(ktime); 1028 PN(sched_clk); 1029 PN(cpu_clk); 1030 P(jiffies); 1031 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK 1032 P(sched_clock_stable()); 1033 #endif 1034 #undef PN 1035 #undef P 1036 1037 SEQ_printf(m, "\n"); 1038 SEQ_printf(m, "sysctl_sched\n"); 1039 1040 #define P(x) \ 1041 SEQ_printf(m, " .%-40s: %Ld\n", #x, (long long)(x)) 1042 #define PN(x) \ 1043 SEQ_printf(m, " .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x)) 1044 PN(sysctl_sched_base_slice); 1045 P(sysctl_sched_features); 1046 #undef PN 1047 #undef P 1048 1049 SEQ_printf(m, " .%-40s: %d (%s)\n", 1050 "sysctl_sched_tunable_scaling", 1051 sysctl_sched_tunable_scaling, 1052 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]); 1053 SEQ_printf(m, "\n"); 1054 } 1055 1056 static int sched_debug_show(struct seq_file *m, void *v) 1057 { 1058 int cpu = (unsigned long)(v - 2); 1059 1060 if (cpu != -1) 1061 print_cpu(m, cpu); 1062 else 1063 sched_debug_header(m); 1064 1065 return 0; 1066 } 1067 1068 void sysrq_sched_debug_show(void) 1069 { 1070 int cpu; 1071 1072 sched_debug_header(NULL); 1073 for_each_online_cpu(cpu) { 1074 /* 1075 * Need to reset softlockup watchdogs on all CPUs, because 1076 * another CPU might be blocked waiting for us to process 1077 * an IPI or stop_machine. 1078 */ 1079 touch_nmi_watchdog(); 1080 touch_all_softlockup_watchdogs(); 1081 print_cpu(NULL, cpu); 1082 } 1083 } 1084 1085 /* 1086 * This iterator needs some explanation. 1087 * It returns 1 for the header position. 1088 * This means 2 is CPU 0. 1089 * In a hotplugged system some CPUs, including CPU 0, may be missing so we have 1090 * to use cpumask_* to iterate over the CPUs. 1091 */ 1092 static void *sched_debug_start(struct seq_file *file, loff_t *offset) 1093 { 1094 unsigned long n = *offset; 1095 1096 if (n == 0) 1097 return (void *) 1; 1098 1099 n--; 1100 1101 if (n > 0) 1102 n = cpumask_next(n - 1, cpu_online_mask); 1103 else 1104 n = cpumask_first(cpu_online_mask); 1105 1106 *offset = n + 1; 1107 1108 if (n < nr_cpu_ids) 1109 return (void *)(unsigned long)(n + 2); 1110 1111 return NULL; 1112 } 1113 1114 static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset) 1115 { 1116 (*offset)++; 1117 return sched_debug_start(file, offset); 1118 } 1119 1120 static void sched_debug_stop(struct seq_file *file, void *data) 1121 { 1122 } 1123 1124 static const struct seq_operations sched_debug_sops = { 1125 .start = sched_debug_start, 1126 .next = sched_debug_next, 1127 .stop = sched_debug_stop, 1128 .show = sched_debug_show, 1129 }; 1130 1131 #define __PS(S, F) SEQ_printf(m, "%-45s:%21Ld\n", S, (long long)(F)) 1132 #define __P(F) __PS(#F, F) 1133 #define P(F) __PS(#F, p->F) 1134 #define PM(F, M) __PS(#F, p->F & (M)) 1135 #define __PSN(S, F) SEQ_printf(m, "%-45s:%14Ld.%06ld\n", S, SPLIT_NS((long long)(F))) 1136 #define __PN(F) __PSN(#F, F) 1137 #define PN(F) __PSN(#F, p->F) 1138 1139 1140 #ifdef CONFIG_NUMA_BALANCING 1141 void print_numa_stats(struct seq_file *m, int node, unsigned long tsf, 1142 unsigned long tpf, unsigned long gsf, unsigned long gpf) 1143 { 1144 SEQ_printf(m, "numa_faults node=%d ", node); 1145 SEQ_printf(m, "task_private=%lu task_shared=%lu ", tpf, tsf); 1146 SEQ_printf(m, "group_private=%lu group_shared=%lu\n", gpf, gsf); 1147 } 1148 #endif 1149 1150 1151 static void sched_show_numa(struct task_struct *p, struct seq_file *m) 1152 { 1153 #ifdef CONFIG_NUMA_BALANCING 1154 if (p->mm) 1155 P(mm->numa_scan_seq); 1156 1157 P(numa_pages_migrated); 1158 P(numa_preferred_nid); 1159 P(total_numa_faults); 1160 SEQ_printf(m, "current_node=%d, numa_group_id=%d\n", 1161 task_node(p), task_numa_group_id(p)); 1162 show_numa_stats(p, m); 1163 #endif 1164 } 1165 1166 void proc_sched_show_task(struct task_struct *p, struct pid_namespace *ns, 1167 struct seq_file *m) 1168 { 1169 unsigned long nr_switches; 1170 1171 SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr_ns(p, ns), 1172 get_nr_threads(p)); 1173 SEQ_printf(m, 1174 "---------------------------------------------------------" 1175 "----------\n"); 1176 1177 #define P_SCHEDSTAT(F) __PS(#F, schedstat_val(p->stats.F)) 1178 #define PN_SCHEDSTAT(F) __PSN(#F, schedstat_val(p->stats.F)) 1179 1180 PN(se.exec_start); 1181 PN(se.vruntime); 1182 PN(se.sum_exec_runtime); 1183 1184 nr_switches = p->nvcsw + p->nivcsw; 1185 1186 P(se.nr_migrations); 1187 1188 if (schedstat_enabled()) { 1189 u64 avg_atom, avg_per_cpu; 1190 1191 PN_SCHEDSTAT(sum_sleep_runtime); 1192 PN_SCHEDSTAT(sum_block_runtime); 1193 PN_SCHEDSTAT(wait_start); 1194 PN_SCHEDSTAT(sleep_start); 1195 PN_SCHEDSTAT(block_start); 1196 PN_SCHEDSTAT(sleep_max); 1197 PN_SCHEDSTAT(block_max); 1198 PN_SCHEDSTAT(exec_max); 1199 PN_SCHEDSTAT(slice_max); 1200 PN_SCHEDSTAT(wait_max); 1201 PN_SCHEDSTAT(wait_sum); 1202 P_SCHEDSTAT(wait_count); 1203 PN_SCHEDSTAT(iowait_sum); 1204 P_SCHEDSTAT(iowait_count); 1205 P_SCHEDSTAT(nr_migrations_cold); 1206 P_SCHEDSTAT(nr_failed_migrations_affine); 1207 P_SCHEDSTAT(nr_failed_migrations_running); 1208 P_SCHEDSTAT(nr_failed_migrations_hot); 1209 P_SCHEDSTAT(nr_forced_migrations); 1210 P_SCHEDSTAT(nr_wakeups); 1211 P_SCHEDSTAT(nr_wakeups_sync); 1212 P_SCHEDSTAT(nr_wakeups_migrate); 1213 P_SCHEDSTAT(nr_wakeups_local); 1214 P_SCHEDSTAT(nr_wakeups_remote); 1215 P_SCHEDSTAT(nr_wakeups_affine); 1216 P_SCHEDSTAT(nr_wakeups_affine_attempts); 1217 P_SCHEDSTAT(nr_wakeups_passive); 1218 P_SCHEDSTAT(nr_wakeups_idle); 1219 1220 avg_atom = p->se.sum_exec_runtime; 1221 if (nr_switches) 1222 avg_atom = div64_ul(avg_atom, nr_switches); 1223 else 1224 avg_atom = -1LL; 1225 1226 avg_per_cpu = p->se.sum_exec_runtime; 1227 if (p->se.nr_migrations) { 1228 avg_per_cpu = div64_u64(avg_per_cpu, 1229 p->se.nr_migrations); 1230 } else { 1231 avg_per_cpu = -1LL; 1232 } 1233 1234 __PN(avg_atom); 1235 __PN(avg_per_cpu); 1236 1237 #ifdef CONFIG_SCHED_CORE 1238 PN_SCHEDSTAT(core_forceidle_sum); 1239 #endif 1240 } 1241 1242 __P(nr_switches); 1243 __PS("nr_voluntary_switches", p->nvcsw); 1244 __PS("nr_involuntary_switches", p->nivcsw); 1245 1246 P(se.load.weight); 1247 #ifdef CONFIG_SMP 1248 P(se.avg.load_sum); 1249 P(se.avg.runnable_sum); 1250 P(se.avg.util_sum); 1251 P(se.avg.load_avg); 1252 P(se.avg.runnable_avg); 1253 P(se.avg.util_avg); 1254 P(se.avg.last_update_time); 1255 PM(se.avg.util_est, ~UTIL_AVG_UNCHANGED); 1256 #endif 1257 #ifdef CONFIG_UCLAMP_TASK 1258 __PS("uclamp.min", p->uclamp_req[UCLAMP_MIN].value); 1259 __PS("uclamp.max", p->uclamp_req[UCLAMP_MAX].value); 1260 __PS("effective uclamp.min", uclamp_eff_value(p, UCLAMP_MIN)); 1261 __PS("effective uclamp.max", uclamp_eff_value(p, UCLAMP_MAX)); 1262 #endif 1263 P(policy); 1264 P(prio); 1265 if (task_has_dl_policy(p)) { 1266 P(dl.runtime); 1267 P(dl.deadline); 1268 } 1269 #ifdef CONFIG_SCHED_CLASS_EXT 1270 __PS("ext.enabled", task_on_scx(p)); 1271 #endif 1272 #undef PN_SCHEDSTAT 1273 #undef P_SCHEDSTAT 1274 1275 { 1276 unsigned int this_cpu = raw_smp_processor_id(); 1277 u64 t0, t1; 1278 1279 t0 = cpu_clock(this_cpu); 1280 t1 = cpu_clock(this_cpu); 1281 __PS("clock-delta", t1-t0); 1282 } 1283 1284 sched_show_numa(p, m); 1285 } 1286 1287 void proc_sched_set_task(struct task_struct *p) 1288 { 1289 #ifdef CONFIG_SCHEDSTATS 1290 memset(&p->stats, 0, sizeof(p->stats)); 1291 #endif 1292 } 1293 1294 void resched_latency_warn(int cpu, u64 latency) 1295 { 1296 static DEFINE_RATELIMIT_STATE(latency_check_ratelimit, 60 * 60 * HZ, 1); 1297 1298 WARN(__ratelimit(&latency_check_ratelimit), 1299 "sched: CPU %d need_resched set for > %llu ns (%d ticks) " 1300 "without schedule\n", 1301 cpu, latency, cpu_rq(cpu)->ticks_without_resched); 1302 } 1303