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