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 <linux/log2.h> 12 #include "sched.h" 13 14 /* 15 * This allows printing both to /sys/kernel/debug/sched/debug and 16 * to the console 17 */ 18 #define SEQ_printf(m, x...) \ 19 do { \ 20 if (m) \ 21 seq_printf(m, x); \ 22 else \ 23 pr_cont(x); \ 24 } while (0) 25 26 /* 27 * Ease the printing of nsec fields: 28 */ 29 static long long nsec_high(unsigned long long nsec) 30 { 31 if ((long long)nsec < 0) { 32 nsec = -nsec; 33 do_div(nsec, 1000000); 34 return -nsec; 35 } 36 do_div(nsec, 1000000); 37 38 return nsec; 39 } 40 41 static unsigned long nsec_low(unsigned long long nsec) 42 { 43 if ((long long)nsec < 0) 44 nsec = -nsec; 45 46 return do_div(nsec, 1000000); 47 } 48 49 #define SPLIT_NS(x) nsec_high(x), nsec_low(x) 50 51 #define SCHED_FEAT(name, enabled) \ 52 #name , 53 54 static const char * const sched_feat_names[] = { 55 #include "features.h" 56 }; 57 58 #undef SCHED_FEAT 59 60 static int sched_feat_show(struct seq_file *m, void *v) 61 { 62 int i; 63 64 for (i = 0; i < __SCHED_FEAT_NR; i++) { 65 if (!(sysctl_sched_features & (1UL << i))) 66 seq_puts(m, "NO_"); 67 seq_printf(m, "%s ", sched_feat_names[i]); 68 } 69 seq_puts(m, "\n"); 70 71 return 0; 72 } 73 74 #ifdef CONFIG_JUMP_LABEL 75 76 #define jump_label_key__true { .key_true = STATIC_KEY_TRUE_INIT } 77 #define jump_label_key__false { .key_false = STATIC_KEY_FALSE_INIT } 78 79 #define SCHED_FEAT(name, enabled) \ 80 jump_label_key__##enabled , 81 82 union sched_feat_key sched_feat_keys[__SCHED_FEAT_NR] = { 83 #include "features.h" 84 }; 85 86 #undef SCHED_FEAT 87 88 static void sched_feat_disable(int i) 89 { 90 static_branch_disable_cpuslocked(&sched_feat_keys[i].key_true); 91 } 92 93 static void sched_feat_enable(int i) 94 { 95 static_branch_enable_cpuslocked(&sched_feat_keys[i].key_false); 96 } 97 #else /* !CONFIG_JUMP_LABEL: */ 98 static void sched_feat_disable(int i) { }; 99 static void sched_feat_enable(int i) { }; 100 #endif /* !CONFIG_JUMP_LABEL */ 101 102 static int sched_feat_set(char *cmp) 103 { 104 int i; 105 int neg = 0; 106 107 if (strncmp(cmp, "NO_", 3) == 0) { 108 neg = 1; 109 cmp += 3; 110 } 111 112 i = match_string(sched_feat_names, __SCHED_FEAT_NR, cmp); 113 if (i < 0) 114 return i; 115 116 if (neg) { 117 sysctl_sched_features &= ~(1UL << i); 118 sched_feat_disable(i); 119 } else { 120 sysctl_sched_features |= (1UL << i); 121 sched_feat_enable(i); 122 } 123 124 return 0; 125 } 126 127 static ssize_t 128 sched_feat_write(struct file *filp, const char __user *ubuf, 129 size_t cnt, loff_t *ppos) 130 { 131 char buf[64]; 132 char *cmp; 133 int ret; 134 struct inode *inode; 135 136 if (cnt > 63) 137 cnt = 63; 138 139 if (copy_from_user(buf, ubuf, cnt)) 140 return -EFAULT; 141 142 buf[cnt] = 0; 143 cmp = strstrip(buf); 144 145 /* Ensure the static_key remains in a consistent state */ 146 inode = file_inode(filp); 147 cpus_read_lock(); 148 inode_lock(inode); 149 ret = sched_feat_set(cmp); 150 inode_unlock(inode); 151 cpus_read_unlock(); 152 if (ret < 0) 153 return ret; 154 155 *ppos += cnt; 156 157 return cnt; 158 } 159 160 static int sched_feat_open(struct inode *inode, struct file *filp) 161 { 162 return single_open(filp, sched_feat_show, NULL); 163 } 164 165 static const struct file_operations sched_feat_fops = { 166 .open = sched_feat_open, 167 .write = sched_feat_write, 168 .read = seq_read, 169 .llseek = seq_lseek, 170 .release = single_release, 171 }; 172 173 static ssize_t sched_scaling_write(struct file *filp, const char __user *ubuf, 174 size_t cnt, loff_t *ppos) 175 { 176 unsigned int scaling; 177 int ret; 178 179 ret = kstrtouint_from_user(ubuf, cnt, 10, &scaling); 180 if (ret) 181 return ret; 182 183 if (scaling >= SCHED_TUNABLESCALING_END) 184 return -EINVAL; 185 186 sysctl_sched_tunable_scaling = scaling; 187 if (sched_update_scaling()) 188 return -EINVAL; 189 190 *ppos += cnt; 191 return cnt; 192 } 193 194 static int sched_scaling_show(struct seq_file *m, void *v) 195 { 196 seq_printf(m, "%d\n", sysctl_sched_tunable_scaling); 197 return 0; 198 } 199 200 static int sched_scaling_open(struct inode *inode, struct file *filp) 201 { 202 return single_open(filp, sched_scaling_show, NULL); 203 } 204 205 static const struct file_operations sched_scaling_fops = { 206 .open = sched_scaling_open, 207 .write = sched_scaling_write, 208 .read = seq_read, 209 .llseek = seq_lseek, 210 .release = single_release, 211 }; 212 213 #ifdef CONFIG_SCHED_CACHE 214 static ssize_t 215 sched_cache_enable_write(struct file *filp, const char __user *ubuf, 216 size_t cnt, loff_t *ppos) 217 { 218 bool val; 219 int ret; 220 221 ret = kstrtobool_from_user(ubuf, cnt, &val); 222 if (ret) 223 return ret; 224 225 sysctl_sched_cache_user = val; 226 227 sched_cache_active_set(); 228 229 *ppos += cnt; 230 231 return cnt; 232 } 233 234 static int sched_cache_enable_show(struct seq_file *m, void *v) 235 { 236 seq_printf(m, "%d\n", sysctl_sched_cache_user); 237 return 0; 238 } 239 240 static int sched_cache_enable_open(struct inode *inode, 241 struct file *filp) 242 { 243 return single_open(filp, sched_cache_enable_show, NULL); 244 } 245 246 static const struct file_operations sched_cache_enable_fops = { 247 .open = sched_cache_enable_open, 248 .write = sched_cache_enable_write, 249 .read = seq_read, 250 .llseek = seq_lseek, 251 .release = single_release, 252 }; 253 #endif 254 255 #ifdef CONFIG_PREEMPT_DYNAMIC 256 257 static ssize_t sched_dynamic_write(struct file *filp, const char __user *ubuf, 258 size_t cnt, loff_t *ppos) 259 { 260 char buf[16]; 261 int mode; 262 263 if (cnt > 15) 264 cnt = 15; 265 266 if (copy_from_user(buf, ubuf, cnt)) 267 return -EFAULT; 268 269 buf[cnt] = 0; 270 mode = sched_dynamic_mode(strstrip(buf)); 271 if (mode < 0) 272 return mode; 273 274 sched_dynamic_update(mode); 275 276 *ppos += cnt; 277 278 return cnt; 279 } 280 281 static int sched_dynamic_show(struct seq_file *m, void *v) 282 { 283 int mode = READ_ONCE(preempt_dynamic_mode); 284 285 /* Stop at NULL terminator */ 286 for (int i = 0; preempt_modes[i]; i++) { 287 if (mode == i) 288 seq_puts(m, "("); 289 seq_puts(m, preempt_modes[i]); 290 if (mode == i) 291 seq_puts(m, ")"); 292 293 seq_puts(m, " "); 294 } 295 296 seq_puts(m, "\n"); 297 return 0; 298 } 299 300 static int sched_dynamic_open(struct inode *inode, struct file *filp) 301 { 302 return single_open(filp, sched_dynamic_show, NULL); 303 } 304 305 static const struct file_operations sched_dynamic_fops = { 306 .open = sched_dynamic_open, 307 .write = sched_dynamic_write, 308 .read = seq_read, 309 .llseek = seq_lseek, 310 .release = single_release, 311 }; 312 313 #endif /* CONFIG_PREEMPT_DYNAMIC */ 314 315 __read_mostly bool sched_debug_verbose; 316 317 static struct dentry *sd_dentry; 318 319 320 static ssize_t sched_verbose_write(struct file *filp, const char __user *ubuf, 321 size_t cnt, loff_t *ppos) 322 { 323 ssize_t result; 324 bool orig; 325 326 cpus_read_lock(); 327 sched_domains_mutex_lock(); 328 329 orig = sched_debug_verbose; 330 result = debugfs_write_file_bool(filp, ubuf, cnt, ppos); 331 332 if (sched_debug_verbose && !orig) 333 update_sched_domain_debugfs(); 334 else if (!sched_debug_verbose && orig) { 335 debugfs_remove(sd_dentry); 336 sd_dentry = NULL; 337 } 338 339 sched_domains_mutex_unlock(); 340 cpus_read_unlock(); 341 342 return result; 343 } 344 345 static const struct file_operations sched_verbose_fops = { 346 .read = debugfs_read_file_bool, 347 .write = sched_verbose_write, 348 .open = simple_open, 349 .llseek = default_llseek, 350 }; 351 352 static const struct seq_operations sched_debug_sops; 353 354 static int sched_debug_open(struct inode *inode, struct file *filp) 355 { 356 return seq_open(filp, &sched_debug_sops); 357 } 358 359 static const struct file_operations sched_debug_fops = { 360 .open = sched_debug_open, 361 .read = seq_read, 362 .llseek = seq_lseek, 363 .release = seq_release, 364 }; 365 366 enum dl_param { 367 DL_RUNTIME = 0, 368 DL_PERIOD, 369 }; 370 371 static unsigned long dl_server_period_max = (1UL << 22) * NSEC_PER_USEC; /* ~4 seconds */ 372 static unsigned long dl_server_period_min = (100) * NSEC_PER_USEC; /* 100 us */ 373 374 static ssize_t sched_server_write_common(struct file *filp, const char __user *ubuf, 375 size_t cnt, loff_t *ppos, enum dl_param param, 376 void *server) 377 { 378 long cpu = (long) ((struct seq_file *) filp->private_data)->private; 379 struct sched_dl_entity *dl_se = (struct sched_dl_entity *)server; 380 u64 old_runtime, runtime, period; 381 struct rq *rq = cpu_rq(cpu); 382 int retval = 0; 383 size_t err; 384 u64 value; 385 386 err = kstrtoull_from_user(ubuf, cnt, 10, &value); 387 if (err) 388 return err; 389 390 scoped_guard (rq_lock_irqsave, rq) { 391 old_runtime = runtime = dl_se->dl_runtime; 392 period = dl_se->dl_period; 393 394 switch (param) { 395 case DL_RUNTIME: 396 if (runtime == value) 397 break; 398 runtime = value; 399 break; 400 case DL_PERIOD: 401 if (value == period) 402 break; 403 period = value; 404 break; 405 } 406 407 if (runtime > period || 408 period > dl_server_period_max || 409 period < dl_server_period_min) { 410 return -EINVAL; 411 } 412 413 if (!cpu_online(cpu_of(rq))) 414 return -EBUSY; 415 416 update_rq_clock(rq); 417 dl_server_stop(dl_se); 418 retval = dl_server_apply_params(dl_se, runtime, period, 0); 419 dl_server_start(dl_se); 420 421 if (retval < 0) 422 return retval; 423 } 424 425 if (!!old_runtime ^ !!runtime) { 426 pr_info("%s server %sabled on CPU %d%s.\n", 427 server == &rq->fair_server ? "Fair" : "Ext", 428 runtime ? "en" : "dis", 429 cpu_of(rq), 430 runtime ? "" : ", system may malfunction due to starvation"); 431 } 432 433 *ppos += cnt; 434 return cnt; 435 } 436 437 static size_t sched_server_show_common(struct seq_file *m, void *v, enum dl_param param, 438 void *server) 439 { 440 struct sched_dl_entity *dl_se = (struct sched_dl_entity *)server; 441 u64 value; 442 443 switch (param) { 444 case DL_RUNTIME: 445 value = dl_se->dl_runtime; 446 break; 447 case DL_PERIOD: 448 value = dl_se->dl_period; 449 break; 450 } 451 452 seq_printf(m, "%llu\n", value); 453 return 0; 454 } 455 456 static ssize_t 457 sched_fair_server_runtime_write(struct file *filp, const char __user *ubuf, 458 size_t cnt, loff_t *ppos) 459 { 460 long cpu = (long) ((struct seq_file *) filp->private_data)->private; 461 struct rq *rq = cpu_rq(cpu); 462 463 return sched_server_write_common(filp, ubuf, cnt, ppos, DL_RUNTIME, 464 &rq->fair_server); 465 } 466 467 static int sched_fair_server_runtime_show(struct seq_file *m, void *v) 468 { 469 unsigned long cpu = (unsigned long) m->private; 470 struct rq *rq = cpu_rq(cpu); 471 472 return sched_server_show_common(m, v, DL_RUNTIME, &rq->fair_server); 473 } 474 475 static int sched_fair_server_runtime_open(struct inode *inode, struct file *filp) 476 { 477 return single_open(filp, sched_fair_server_runtime_show, inode->i_private); 478 } 479 480 static const struct file_operations fair_server_runtime_fops = { 481 .open = sched_fair_server_runtime_open, 482 .write = sched_fair_server_runtime_write, 483 .read = seq_read, 484 .llseek = seq_lseek, 485 .release = single_release, 486 }; 487 488 static struct dentry *debugfs_sched; 489 490 #ifdef CONFIG_SCHED_CLASS_EXT 491 static ssize_t 492 sched_ext_server_runtime_write(struct file *filp, const char __user *ubuf, 493 size_t cnt, loff_t *ppos) 494 { 495 long cpu = (long) ((struct seq_file *) filp->private_data)->private; 496 struct rq *rq = cpu_rq(cpu); 497 498 return sched_server_write_common(filp, ubuf, cnt, ppos, DL_RUNTIME, 499 &rq->ext_server); 500 } 501 502 static int sched_ext_server_runtime_show(struct seq_file *m, void *v) 503 { 504 unsigned long cpu = (unsigned long) m->private; 505 struct rq *rq = cpu_rq(cpu); 506 507 return sched_server_show_common(m, v, DL_RUNTIME, &rq->ext_server); 508 } 509 510 static int sched_ext_server_runtime_open(struct inode *inode, struct file *filp) 511 { 512 return single_open(filp, sched_ext_server_runtime_show, inode->i_private); 513 } 514 515 static const struct file_operations ext_server_runtime_fops = { 516 .open = sched_ext_server_runtime_open, 517 .write = sched_ext_server_runtime_write, 518 .read = seq_read, 519 .llseek = seq_lseek, 520 .release = single_release, 521 }; 522 523 static ssize_t 524 sched_ext_server_period_write(struct file *filp, const char __user *ubuf, 525 size_t cnt, loff_t *ppos) 526 { 527 long cpu = (long) ((struct seq_file *) filp->private_data)->private; 528 struct rq *rq = cpu_rq(cpu); 529 530 return sched_server_write_common(filp, ubuf, cnt, ppos, DL_PERIOD, 531 &rq->ext_server); 532 } 533 534 static int sched_ext_server_period_show(struct seq_file *m, void *v) 535 { 536 unsigned long cpu = (unsigned long) m->private; 537 struct rq *rq = cpu_rq(cpu); 538 539 return sched_server_show_common(m, v, DL_PERIOD, &rq->ext_server); 540 } 541 542 static int sched_ext_server_period_open(struct inode *inode, struct file *filp) 543 { 544 return single_open(filp, sched_ext_server_period_show, inode->i_private); 545 } 546 547 static const struct file_operations ext_server_period_fops = { 548 .open = sched_ext_server_period_open, 549 .write = sched_ext_server_period_write, 550 .read = seq_read, 551 .llseek = seq_lseek, 552 .release = single_release, 553 }; 554 555 static void debugfs_ext_server_init(void) 556 { 557 struct dentry *d_ext; 558 unsigned long cpu; 559 560 d_ext = debugfs_create_dir("ext_server", debugfs_sched); 561 if (!d_ext) 562 return; 563 564 for_each_possible_cpu(cpu) { 565 struct dentry *d_cpu; 566 char buf[32]; 567 568 snprintf(buf, sizeof(buf), "cpu%lu", cpu); 569 d_cpu = debugfs_create_dir(buf, d_ext); 570 571 debugfs_create_file("runtime", 0644, d_cpu, (void *) cpu, &ext_server_runtime_fops); 572 debugfs_create_file("period", 0644, d_cpu, (void *) cpu, &ext_server_period_fops); 573 } 574 } 575 #endif /* CONFIG_SCHED_CLASS_EXT */ 576 577 static ssize_t 578 sched_fair_server_period_write(struct file *filp, const char __user *ubuf, 579 size_t cnt, loff_t *ppos) 580 { 581 long cpu = (long) ((struct seq_file *) filp->private_data)->private; 582 struct rq *rq = cpu_rq(cpu); 583 584 return sched_server_write_common(filp, ubuf, cnt, ppos, DL_PERIOD, 585 &rq->fair_server); 586 } 587 588 static int sched_fair_server_period_show(struct seq_file *m, void *v) 589 { 590 unsigned long cpu = (unsigned long) m->private; 591 struct rq *rq = cpu_rq(cpu); 592 593 return sched_server_show_common(m, v, DL_PERIOD, &rq->fair_server); 594 } 595 596 static int sched_fair_server_period_open(struct inode *inode, struct file *filp) 597 { 598 return single_open(filp, sched_fair_server_period_show, inode->i_private); 599 } 600 601 static const struct file_operations fair_server_period_fops = { 602 .open = sched_fair_server_period_open, 603 .write = sched_fair_server_period_write, 604 .read = seq_read, 605 .llseek = seq_lseek, 606 .release = single_release, 607 }; 608 609 static void debugfs_fair_server_init(void) 610 { 611 struct dentry *d_fair; 612 unsigned long cpu; 613 614 d_fair = debugfs_create_dir("fair_server", debugfs_sched); 615 if (!d_fair) 616 return; 617 618 for_each_possible_cpu(cpu) { 619 struct dentry *d_cpu; 620 char buf[32]; 621 622 snprintf(buf, sizeof(buf), "cpu%lu", cpu); 623 d_cpu = debugfs_create_dir(buf, d_fair); 624 625 debugfs_create_file("runtime", 0644, d_cpu, (void *) cpu, &fair_server_runtime_fops); 626 debugfs_create_file("period", 0644, d_cpu, (void *) cpu, &fair_server_period_fops); 627 } 628 } 629 630 #ifdef CONFIG_FAIR_GROUP_SCHED 631 static int cgroup_mode = 2; 632 633 /* See __sched_cgroup_mode_update(). */ 634 static const char *cgroup_mode_str[] = { 635 "up", 636 "smp", 637 "concur", 638 "max", 639 "tasks", 640 }; 641 642 static int sched_cgroup_mode(const char *str) 643 { 644 for (int i = 0; i < ARRAY_SIZE(cgroup_mode_str); i++) { 645 if (!strcmp(str, cgroup_mode_str[i])) 646 return i; 647 } 648 return -EINVAL; 649 } 650 651 static ssize_t sched_cgroup_write(struct file *filp, const char __user *ubuf, 652 size_t cnt, loff_t *ppos) 653 { 654 char buf[16]; 655 int mode; 656 657 if (cnt > 15) 658 cnt = 15; 659 660 if (copy_from_user(buf, ubuf, cnt)) 661 return -EFAULT; 662 663 buf[cnt] = 0; 664 mode = sched_cgroup_mode(strstrip(buf)); 665 if (mode < 0) 666 return mode; 667 668 __sched_cgroup_mode_update(mode); 669 WRITE_ONCE(cgroup_mode, mode); 670 671 *ppos += cnt; 672 return cnt; 673 } 674 675 static int sched_cgroup_show(struct seq_file *m, void *v) 676 { 677 int mode = READ_ONCE(cgroup_mode); 678 679 for (int i = 0; i < ARRAY_SIZE(cgroup_mode_str); i++) { 680 if (mode == i) 681 seq_puts(m, "("); 682 seq_puts(m, cgroup_mode_str[i]); 683 if (mode == i) 684 seq_puts(m, ")"); 685 686 seq_puts(m, " "); 687 } 688 seq_puts(m, "\n"); 689 return 0; 690 } 691 692 static int sched_cgroup_open(struct inode *inode, struct file *filp) 693 { 694 return single_open(filp, sched_cgroup_show, NULL); 695 } 696 697 static const struct file_operations sched_cgroup_fops = { 698 .open = sched_cgroup_open, 699 .write = sched_cgroup_write, 700 .read = seq_read, 701 .llseek = seq_lseek, 702 .release = single_release, 703 }; 704 #endif 705 706 static __init int sched_init_debug(void) 707 { 708 struct dentry __maybe_unused *numa, *llc; 709 710 debugfs_sched = debugfs_create_dir("sched", NULL); 711 712 debugfs_create_file("features", 0644, debugfs_sched, NULL, &sched_feat_fops); 713 debugfs_create_file_unsafe("verbose", 0644, debugfs_sched, &sched_debug_verbose, &sched_verbose_fops); 714 #ifdef CONFIG_PREEMPT_DYNAMIC 715 debugfs_create_file("preempt", 0644, debugfs_sched, NULL, &sched_dynamic_fops); 716 #endif 717 718 debugfs_create_u32("base_slice_ns", 0644, debugfs_sched, &sysctl_sched_base_slice); 719 720 debugfs_create_u32("latency_warn_ms", 0644, debugfs_sched, &sysctl_resched_latency_warn_ms); 721 debugfs_create_u32("latency_warn_once", 0644, debugfs_sched, &sysctl_resched_latency_warn_once); 722 723 debugfs_create_file("tunable_scaling", 0644, debugfs_sched, NULL, &sched_scaling_fops); 724 debugfs_create_u32("migration_cost_ns", 0644, debugfs_sched, &sysctl_sched_migration_cost); 725 debugfs_create_u32("nr_migrate", 0644, debugfs_sched, &sysctl_sched_nr_migrate); 726 727 sched_domains_mutex_lock(); 728 update_sched_domain_debugfs(); 729 sched_domains_mutex_unlock(); 730 731 #ifdef CONFIG_NUMA_BALANCING 732 numa = debugfs_create_dir("numa_balancing", debugfs_sched); 733 734 debugfs_create_u32("scan_delay_ms", 0644, numa, &sysctl_numa_balancing_scan_delay); 735 debugfs_create_u32("scan_period_min_ms", 0644, numa, &sysctl_numa_balancing_scan_period_min); 736 debugfs_create_u32("scan_period_max_ms", 0644, numa, &sysctl_numa_balancing_scan_period_max); 737 debugfs_create_u32("scan_size_mb", 0644, numa, &sysctl_numa_balancing_scan_size); 738 debugfs_create_u32("hot_threshold_ms", 0644, numa, &sysctl_numa_balancing_hot_threshold); 739 #endif /* CONFIG_NUMA_BALANCING */ 740 741 #ifdef CONFIG_SCHED_CACHE 742 llc = debugfs_create_dir("llc_balancing", debugfs_sched); 743 debugfs_create_file("enabled", 0644, llc, NULL, 744 &sched_cache_enable_fops); 745 debugfs_create_u32("aggr_tolerance", 0644, llc, 746 &llc_aggr_tolerance); 747 debugfs_create_u32("epoch_period", 0644, llc, 748 &llc_epoch_period); 749 debugfs_create_u32("epoch_affinity_timeout", 0644, llc, 750 &llc_epoch_affinity_timeout); 751 debugfs_create_u32("overaggr_pct", 0644, llc, 752 &llc_overaggr_pct); 753 debugfs_create_u32("imb_pct", 0644, llc, 754 &llc_imb_pct); 755 #endif 756 757 debugfs_create_file("debug", 0444, debugfs_sched, NULL, &sched_debug_fops); 758 759 #ifdef CONFIG_FAIR_GROUP_SCHED 760 debugfs_create_file("cgroup_mode", 0644, debugfs_sched, NULL, &sched_cgroup_fops); 761 #endif 762 763 debugfs_fair_server_init(); 764 #ifdef CONFIG_SCHED_CLASS_EXT 765 debugfs_ext_server_init(); 766 #endif 767 768 return 0; 769 } 770 late_initcall(sched_init_debug); 771 772 static cpumask_var_t sd_sysctl_cpus; 773 774 static int sd_flags_show(struct seq_file *m, void *v) 775 { 776 unsigned long flags = *(unsigned int *)m->private; 777 int idx; 778 779 for_each_set_bit(idx, &flags, __SD_FLAG_CNT) { 780 seq_puts(m, sd_flag_debug[idx].name); 781 seq_puts(m, " "); 782 } 783 seq_puts(m, "\n"); 784 785 return 0; 786 } 787 788 static int sd_flags_open(struct inode *inode, struct file *file) 789 { 790 return single_open(file, sd_flags_show, inode->i_private); 791 } 792 793 static const struct file_operations sd_flags_fops = { 794 .open = sd_flags_open, 795 .read = seq_read, 796 .llseek = seq_lseek, 797 .release = single_release, 798 }; 799 800 static void register_sd(struct sched_domain *sd, struct dentry *parent) 801 { 802 #define SDM(type, mode, member) \ 803 debugfs_create_##type(#member, mode, parent, &sd->member) 804 805 SDM(ulong, 0644, min_interval); 806 SDM(ulong, 0644, max_interval); 807 SDM(u64, 0644, max_newidle_lb_cost); 808 SDM(u32, 0644, busy_factor); 809 SDM(u32, 0644, imbalance_pct); 810 SDM(u32, 0644, cache_nice_tries); 811 SDM(str, 0444, name); 812 813 #undef SDM 814 815 debugfs_create_file("flags", 0444, parent, &sd->flags, &sd_flags_fops); 816 debugfs_create_file("groups_flags", 0444, parent, &sd->groups->flags, &sd_flags_fops); 817 debugfs_create_u32("level", 0444, parent, (u32 *)&sd->level); 818 819 if (sd->flags & SD_ASYM_PACKING) 820 debugfs_create_u32("group_asym_prefer_cpu", 0444, parent, 821 (u32 *)&sd->groups->asym_prefer_cpu); 822 } 823 824 void update_sched_domain_debugfs(void) 825 { 826 int cpu, i; 827 828 /* 829 * This can unfortunately be invoked before sched_debug_init() creates 830 * the debug directory. Don't touch sd_sysctl_cpus until then. 831 */ 832 if (!debugfs_sched) 833 return; 834 835 if (!sched_debug_verbose) 836 return; 837 838 if (!cpumask_available(sd_sysctl_cpus)) { 839 if (!alloc_cpumask_var(&sd_sysctl_cpus, GFP_KERNEL)) 840 return; 841 cpumask_copy(sd_sysctl_cpus, cpu_possible_mask); 842 } 843 844 if (!sd_dentry) { 845 sd_dentry = debugfs_create_dir("domains", debugfs_sched); 846 847 /* rebuild sd_sysctl_cpus if empty since it gets cleared below */ 848 if (cpumask_empty(sd_sysctl_cpus)) 849 cpumask_copy(sd_sysctl_cpus, cpu_online_mask); 850 } 851 852 for_each_cpu(cpu, sd_sysctl_cpus) { 853 struct sched_domain *sd; 854 struct dentry *d_cpu; 855 char buf[32]; 856 857 snprintf(buf, sizeof(buf), "cpu%d", cpu); 858 debugfs_lookup_and_remove(buf, sd_dentry); 859 d_cpu = debugfs_create_dir(buf, sd_dentry); 860 861 i = 0; 862 for_each_domain(cpu, sd) { 863 struct dentry *d_sd; 864 865 snprintf(buf, sizeof(buf), "domain%d", i); 866 d_sd = debugfs_create_dir(buf, d_cpu); 867 868 register_sd(sd, d_sd); 869 i++; 870 } 871 872 __cpumask_clear_cpu(cpu, sd_sysctl_cpus); 873 } 874 } 875 876 void dirty_sched_domain_sysctl(int cpu) 877 { 878 if (cpumask_available(sd_sysctl_cpus)) 879 __cpumask_set_cpu(cpu, sd_sysctl_cpus); 880 } 881 882 #ifdef CONFIG_FAIR_GROUP_SCHED 883 static void print_cfs_group_stats(struct seq_file *m, int cpu, struct task_group *tg) 884 { 885 struct sched_entity *se = tg_se(tg, cpu); 886 887 #define P(F) SEQ_printf(m, " .%-30s: %lld\n", #F, (long long)F) 888 #define P_SCHEDSTAT(F) SEQ_printf(m, " .%-30s: %lld\n", \ 889 #F, (long long)schedstat_val(stats->F)) 890 #define PN(F) SEQ_printf(m, " .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F)) 891 #define PN_SCHEDSTAT(F) SEQ_printf(m, " .%-30s: %lld.%06ld\n", \ 892 #F, SPLIT_NS((long long)schedstat_val(stats->F))) 893 894 if (!se) 895 return; 896 897 PN(se->exec_start); 898 PN(se->vruntime); 899 PN(se->sum_exec_runtime); 900 901 if (schedstat_enabled()) { 902 struct sched_statistics *stats; 903 stats = __schedstats_from_se(se); 904 905 PN_SCHEDSTAT(wait_start); 906 PN_SCHEDSTAT(sleep_start); 907 PN_SCHEDSTAT(block_start); 908 PN_SCHEDSTAT(sleep_max); 909 PN_SCHEDSTAT(block_max); 910 PN_SCHEDSTAT(exec_max); 911 PN_SCHEDSTAT(slice_max); 912 PN_SCHEDSTAT(wait_max); 913 PN_SCHEDSTAT(wait_sum); 914 P_SCHEDSTAT(wait_count); 915 } 916 917 P(se->load.weight); 918 P(se->avg.load_avg); 919 P(se->avg.util_avg); 920 P(se->avg.runnable_avg); 921 922 #undef PN_SCHEDSTAT 923 #undef PN 924 #undef P_SCHEDSTAT 925 #undef P 926 } 927 #endif /* CONFIG_FAIR_GROUP_SCHED */ 928 929 #ifdef CONFIG_CGROUP_SCHED 930 static DEFINE_SPINLOCK(sched_debug_lock); 931 static char group_path[PATH_MAX]; 932 933 static void task_group_path(struct task_group *tg, char *path, int plen) 934 { 935 if (autogroup_path(tg, path, plen)) 936 return; 937 938 cgroup_path(tg->css.cgroup, path, plen); 939 } 940 941 /* 942 * Only 1 SEQ_printf_task_group_path() caller can use the full length 943 * group_path[] for cgroup path. Other simultaneous callers will have 944 * to use a shorter stack buffer. A "..." suffix is appended at the end 945 * of the stack buffer so that it will show up in case the output length 946 * matches the given buffer size to indicate possible path name truncation. 947 */ 948 #define SEQ_printf_task_group_path(m, tg, fmt...) \ 949 { \ 950 if (spin_trylock(&sched_debug_lock)) { \ 951 task_group_path(tg, group_path, sizeof(group_path)); \ 952 SEQ_printf(m, fmt, group_path); \ 953 spin_unlock(&sched_debug_lock); \ 954 } else { \ 955 char buf[128]; \ 956 char *bufend = buf + sizeof(buf) - 3; \ 957 task_group_path(tg, buf, bufend - buf); \ 958 strcpy(bufend - 1, "..."); \ 959 SEQ_printf(m, fmt, buf); \ 960 } \ 961 } 962 #endif 963 964 static void 965 print_task(struct seq_file *m, struct rq *rq, struct task_struct *p) 966 { 967 if (task_current(rq, p)) 968 SEQ_printf(m, ">R"); 969 else 970 SEQ_printf(m, " %c", task_state_to_char(p)); 971 972 SEQ_printf(m, " %15s %5d %10ld %9Ld.%06ld %c %9Ld.%06ld %c %9Ld.%06ld %9Ld.%06ld %9Ld %5d ", 973 p->comm, task_pid_nr(p), 974 p->se.h_load.weight, 975 SPLIT_NS(p->se.vruntime), 976 entity_eligible(&rq->cfs, &p->se) ? 'E' : 'N', 977 SPLIT_NS(p->se.deadline), 978 p->se.custom_slice ? 'S' : ' ', 979 SPLIT_NS(p->se.slice), 980 SPLIT_NS(p->se.sum_exec_runtime), 981 (long long)(p->nvcsw + p->nivcsw), 982 p->prio); 983 984 SEQ_printf(m, "%9lld.%06ld %9lld.%06ld %9lld.%06ld", 985 SPLIT_NS(schedstat_val_or_zero(p->stats.wait_sum)), 986 SPLIT_NS(schedstat_val_or_zero(p->stats.sum_sleep_runtime)), 987 SPLIT_NS(schedstat_val_or_zero(p->stats.sum_block_runtime))); 988 989 #ifdef CONFIG_NUMA_BALANCING 990 SEQ_printf(m, " %d %d", task_node(p), task_numa_group_id(p)); 991 #endif 992 #ifdef CONFIG_CGROUP_SCHED 993 SEQ_printf_task_group_path(m, task_group(p), " %s") 994 #endif 995 996 SEQ_printf(m, "\n"); 997 } 998 999 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu) 1000 { 1001 struct task_struct *g, *p; 1002 1003 SEQ_printf(m, "\n"); 1004 SEQ_printf(m, "runnable tasks:\n"); 1005 SEQ_printf(m, " S task PID weight vruntime eligible " 1006 "deadline slice sum-exec switches " 1007 "prio wait-time sum-sleep sum-block" 1008 #ifdef CONFIG_NUMA_BALANCING 1009 " node group-id" 1010 #endif 1011 #ifdef CONFIG_CGROUP_SCHED 1012 " group-path" 1013 #endif 1014 "\n"); 1015 SEQ_printf(m, "-------------------------------------------------------" 1016 "------------------------------------------------------" 1017 "------------------------------------------------------" 1018 #ifdef CONFIG_NUMA_BALANCING 1019 "--------------" 1020 #endif 1021 #ifdef CONFIG_CGROUP_SCHED 1022 "--------------" 1023 #endif 1024 "\n"); 1025 1026 rcu_read_lock(); 1027 for_each_process_thread(g, p) { 1028 if (task_cpu(p) != rq_cpu) 1029 continue; 1030 1031 print_task(m, rq, p); 1032 } 1033 rcu_read_unlock(); 1034 } 1035 1036 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq) 1037 { 1038 s64 left_vruntime = -1, right_vruntime = -1, left_deadline = -1, spread; 1039 s64 zero_vruntime = -1, sum_w_vruntime = -1; 1040 u64 avruntime; 1041 struct sched_entity *last, *first, *root; 1042 struct rq *rq = cpu_rq(cpu); 1043 unsigned int sum_shift; 1044 unsigned long flags; 1045 u64 sum_weight; 1046 1047 #ifdef CONFIG_FAIR_GROUP_SCHED 1048 SEQ_printf(m, "\n"); 1049 SEQ_printf_task_group_path(m, cfs_rq->tg, "cfs_rq[%d]:%s\n", cpu); 1050 #else 1051 SEQ_printf(m, "\n"); 1052 SEQ_printf(m, "cfs_rq[%d]:\n", cpu); 1053 #endif 1054 1055 raw_spin_rq_lock_irqsave(rq, flags); 1056 root = __pick_root_entity(cfs_rq); 1057 if (root) 1058 left_vruntime = root->min_vruntime; 1059 first = __pick_first_entity(cfs_rq); 1060 if (first) 1061 left_deadline = first->deadline; 1062 last = __pick_last_entity(cfs_rq); 1063 if (last) 1064 right_vruntime = last->vruntime; 1065 zero_vruntime = cfs_rq->zero_vruntime; 1066 sum_w_vruntime = cfs_rq->sum_w_vruntime; 1067 sum_weight = cfs_rq->sum_weight; 1068 sum_shift = cfs_rq->sum_shift; 1069 avruntime = avg_vruntime(cfs_rq); 1070 raw_spin_rq_unlock_irqrestore(rq, flags); 1071 1072 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "left_deadline", 1073 SPLIT_NS(left_deadline)); 1074 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "left_vruntime", 1075 SPLIT_NS(left_vruntime)); 1076 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "zero_vruntime", 1077 SPLIT_NS(zero_vruntime)); 1078 SEQ_printf(m, " .%-30s: %Ld (%d bits)\n", "sum_w_vruntime", 1079 sum_w_vruntime, ilog2(abs(sum_w_vruntime))); 1080 SEQ_printf(m, " .%-30s: %Lu\n", "sum_weight", 1081 sum_weight); 1082 SEQ_printf(m, " .%-30s: %u\n", "sum_shift", sum_shift); 1083 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "avg_vruntime", 1084 SPLIT_NS(avruntime)); 1085 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "right_vruntime", 1086 SPLIT_NS(right_vruntime)); 1087 spread = right_vruntime - left_vruntime; 1088 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread", SPLIT_NS(spread)); 1089 SEQ_printf(m, " .%-30s: %d\n", "nr_queued", cfs_rq->nr_queued); 1090 SEQ_printf(m, " .%-30s: %d\n", "h_nr_runnable", cfs_rq->h_nr_runnable); 1091 SEQ_printf(m, " .%-30s: %d\n", "h_nr_queued", cfs_rq->h_nr_queued); 1092 SEQ_printf(m, " .%-30s: %d\n", "h_nr_idle", cfs_rq->h_nr_idle); 1093 SEQ_printf(m, " .%-30s: %ld\n", "load", cfs_rq->load.weight); 1094 SEQ_printf(m, " .%-30s: %lu\n", "load_avg", 1095 cfs_rq->avg.load_avg); 1096 SEQ_printf(m, " .%-30s: %lu\n", "runnable_avg", 1097 cfs_rq->avg.runnable_avg); 1098 SEQ_printf(m, " .%-30s: %lu\n", "util_avg", 1099 cfs_rq->avg.util_avg); 1100 SEQ_printf(m, " .%-30s: %u\n", "util_est", 1101 cfs_rq->avg.util_est); 1102 SEQ_printf(m, " .%-30s: %ld\n", "removed.load_avg", 1103 cfs_rq->removed.load_avg); 1104 SEQ_printf(m, " .%-30s: %ld\n", "removed.util_avg", 1105 cfs_rq->removed.util_avg); 1106 SEQ_printf(m, " .%-30s: %ld\n", "removed.runnable_avg", 1107 cfs_rq->removed.runnable_avg); 1108 #ifdef CONFIG_FAIR_GROUP_SCHED 1109 SEQ_printf(m, " .%-30s: %lu\n", "tg_load_avg_contrib", 1110 cfs_rq->tg_load_avg_contrib); 1111 SEQ_printf(m, " .%-30s: %ld\n", "tg_load_avg", 1112 atomic_long_read(&cfs_rq->tg->load_avg)); 1113 SEQ_printf(m, " .%-30s: %lu\n", "h_load", 1114 cfs_rq->h_load); 1115 #endif /* CONFIG_FAIR_GROUP_SCHED */ 1116 #ifdef CONFIG_CFS_BANDWIDTH 1117 SEQ_printf(m, " .%-30s: %d\n", "throttled", 1118 cfs_rq->throttled); 1119 SEQ_printf(m, " .%-30s: %d\n", "throttle_count", 1120 cfs_rq->throttle_count); 1121 #endif 1122 1123 #ifdef CONFIG_FAIR_GROUP_SCHED 1124 print_cfs_group_stats(m, cpu, cfs_rq->tg); 1125 #endif 1126 } 1127 1128 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq) 1129 { 1130 #ifdef CONFIG_RT_GROUP_SCHED 1131 SEQ_printf(m, "\n"); 1132 SEQ_printf_task_group_path(m, rt_rq->tg, "rt_rq[%d]:%s\n", cpu); 1133 #else 1134 SEQ_printf(m, "\n"); 1135 SEQ_printf(m, "rt_rq[%d]:\n", cpu); 1136 #endif 1137 1138 #define P(x) \ 1139 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rt_rq->x)) 1140 #define PU(x) \ 1141 SEQ_printf(m, " .%-30s: %lu\n", #x, (unsigned long)(rt_rq->x)) 1142 #define PN(x) \ 1143 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x)) 1144 1145 PU(rt_nr_running); 1146 1147 #ifdef CONFIG_RT_GROUP_SCHED 1148 P(rt_throttled); 1149 PN(rt_time); 1150 PN(rt_runtime); 1151 #endif 1152 1153 #undef PN 1154 #undef PU 1155 #undef P 1156 } 1157 1158 void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq) 1159 { 1160 struct dl_bw *dl_bw; 1161 1162 SEQ_printf(m, "\n"); 1163 SEQ_printf(m, "dl_rq[%d]:\n", cpu); 1164 1165 #define PU(x) \ 1166 SEQ_printf(m, " .%-30s: %lu\n", #x, (unsigned long)(dl_rq->x)) 1167 1168 PU(dl_nr_running); 1169 dl_bw = &cpu_rq(cpu)->rd->dl_bw; 1170 SEQ_printf(m, " .%-30s: %lld\n", "dl_bw->bw", dl_bw->bw); 1171 SEQ_printf(m, " .%-30s: %lld\n", "dl_bw->total_bw", dl_bw->total_bw); 1172 1173 #undef PU 1174 } 1175 1176 static void print_cpu(struct seq_file *m, int cpu) 1177 { 1178 struct rq *rq = cpu_rq(cpu); 1179 1180 #ifdef CONFIG_X86 1181 { 1182 unsigned int freq = cpu_khz ? : 1; 1183 1184 SEQ_printf(m, "cpu#%d, %u.%03u MHz\n", 1185 cpu, freq / 1000, (freq % 1000)); 1186 } 1187 #else /* !CONFIG_X86: */ 1188 SEQ_printf(m, "cpu#%d\n", cpu); 1189 #endif /* !CONFIG_X86 */ 1190 1191 #define P(x) \ 1192 do { \ 1193 if (sizeof(rq->x) == 4) \ 1194 SEQ_printf(m, " .%-30s: %d\n", #x, (int)(rq->x)); \ 1195 else \ 1196 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rq->x));\ 1197 } while (0) 1198 1199 #define PN(x) \ 1200 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x)) 1201 1202 P(nr_running); 1203 P(nr_switches); 1204 P(nr_uninterruptible); 1205 PN(next_balance); 1206 SEQ_printf(m, " .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr))); 1207 PN(clock); 1208 PN(clock_task); 1209 #undef P 1210 #undef PN 1211 1212 #define P64(n) SEQ_printf(m, " .%-30s: %Ld\n", #n, rq->n); 1213 P64(avg_idle); 1214 P64(max_idle_balance_cost); 1215 #undef P64 1216 1217 #define P(n) SEQ_printf(m, " .%-30s: %d\n", #n, schedstat_val(rq->n)); 1218 if (schedstat_enabled()) { 1219 P(yld_count); 1220 P(sched_count); 1221 P(sched_goidle); 1222 P(ttwu_count); 1223 P(ttwu_local); 1224 } 1225 #undef P 1226 1227 print_cfs_stats(m, cpu); 1228 print_rt_stats(m, cpu); 1229 print_dl_stats(m, cpu); 1230 1231 print_rq(m, rq, cpu); 1232 SEQ_printf(m, "\n"); 1233 } 1234 1235 static const char *sched_tunable_scaling_names[] = { 1236 "none", 1237 "logarithmic", 1238 "linear" 1239 }; 1240 1241 static void sched_debug_header(struct seq_file *m) 1242 { 1243 u64 ktime, sched_clk, cpu_clk; 1244 unsigned long flags; 1245 1246 local_irq_save(flags); 1247 ktime = ktime_to_ns(ktime_get()); 1248 sched_clk = sched_clock(); 1249 cpu_clk = local_clock(); 1250 local_irq_restore(flags); 1251 1252 SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n", 1253 init_utsname()->release, 1254 (int)strcspn(init_utsname()->version, " "), 1255 init_utsname()->version); 1256 1257 #define P(x) \ 1258 SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x)) 1259 #define PN(x) \ 1260 SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x)) 1261 PN(ktime); 1262 PN(sched_clk); 1263 PN(cpu_clk); 1264 P(jiffies); 1265 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK 1266 P(sched_clock_stable()); 1267 #endif 1268 #undef PN 1269 #undef P 1270 1271 SEQ_printf(m, "\n"); 1272 SEQ_printf(m, "sysctl_sched\n"); 1273 1274 #define P(x) \ 1275 SEQ_printf(m, " .%-40s: %Ld\n", #x, (long long)(x)) 1276 #define PN(x) \ 1277 SEQ_printf(m, " .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x)) 1278 PN(sysctl_sched_base_slice); 1279 P(sysctl_sched_features); 1280 #undef PN 1281 #undef P 1282 1283 SEQ_printf(m, " .%-40s: %d (%s)\n", 1284 "sysctl_sched_tunable_scaling", 1285 sysctl_sched_tunable_scaling, 1286 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]); 1287 SEQ_printf(m, "\n"); 1288 } 1289 1290 static int sched_debug_show(struct seq_file *m, void *v) 1291 { 1292 int cpu = (unsigned long)(v - 2); 1293 1294 if (cpu != -1) 1295 print_cpu(m, cpu); 1296 else 1297 sched_debug_header(m); 1298 1299 return 0; 1300 } 1301 1302 void sysrq_sched_debug_show(void) 1303 { 1304 int cpu; 1305 1306 sched_debug_header(NULL); 1307 for_each_online_cpu(cpu) { 1308 /* 1309 * Need to reset softlockup watchdogs on all CPUs, because 1310 * another CPU might be blocked waiting for us to process 1311 * an IPI or stop_machine. 1312 */ 1313 touch_nmi_watchdog(); 1314 touch_all_softlockup_watchdogs(); 1315 print_cpu(NULL, cpu); 1316 } 1317 } 1318 1319 /* 1320 * This iterator needs some explanation. 1321 * It returns 1 for the header position. 1322 * This means 2 is CPU 0. 1323 * In a hotplugged system some CPUs, including CPU 0, may be missing so we have 1324 * to use cpumask_* to iterate over the CPUs. 1325 */ 1326 static void *sched_debug_start(struct seq_file *file, loff_t *offset) 1327 { 1328 unsigned long n = *offset; 1329 1330 if (n == 0) 1331 return (void *) 1; 1332 1333 n--; 1334 1335 if (n > 0) 1336 n = cpumask_next(n - 1, cpu_online_mask); 1337 else 1338 n = cpumask_first(cpu_online_mask); 1339 1340 *offset = n + 1; 1341 1342 if (n < nr_cpu_ids) 1343 return (void *)(unsigned long)(n + 2); 1344 1345 return NULL; 1346 } 1347 1348 static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset) 1349 { 1350 (*offset)++; 1351 return sched_debug_start(file, offset); 1352 } 1353 1354 static void sched_debug_stop(struct seq_file *file, void *data) 1355 { 1356 } 1357 1358 static const struct seq_operations sched_debug_sops = { 1359 .start = sched_debug_start, 1360 .next = sched_debug_next, 1361 .stop = sched_debug_stop, 1362 .show = sched_debug_show, 1363 }; 1364 1365 #define __PS(S, F) SEQ_printf(m, "%-45s:%21Ld\n", S, (long long)(F)) 1366 #define __P(F) __PS(#F, F) 1367 #define P(F) __PS(#F, p->F) 1368 #define PM(F, M) __PS(#F, p->F & (M)) 1369 #define __PSN(S, F) SEQ_printf(m, "%-45s:%14Ld.%06ld\n", S, SPLIT_NS((long long)(F))) 1370 #define __PN(F) __PSN(#F, F) 1371 #define PN(F) __PSN(#F, p->F) 1372 1373 1374 #ifdef CONFIG_NUMA_BALANCING 1375 void print_numa_stats(struct seq_file *m, int node, unsigned long tsf, 1376 unsigned long tpf, unsigned long gsf, unsigned long gpf) 1377 { 1378 SEQ_printf(m, "numa_faults node=%d ", node); 1379 SEQ_printf(m, "task_private=%lu task_shared=%lu ", tpf, tsf); 1380 SEQ_printf(m, "group_private=%lu group_shared=%lu\n", gpf, gsf); 1381 } 1382 #endif 1383 1384 1385 static void sched_show_numa(struct task_struct *p, struct seq_file *m) 1386 { 1387 #ifdef CONFIG_NUMA_BALANCING 1388 if (p->mm) 1389 P(mm->numa_scan_seq); 1390 1391 P(numa_pages_migrated); 1392 P(numa_preferred_nid); 1393 P(total_numa_faults); 1394 SEQ_printf(m, "current_node=%d, numa_group_id=%d\n", 1395 task_node(p), task_numa_group_id(p)); 1396 show_numa_stats(p, m); 1397 #endif /* CONFIG_NUMA_BALANCING */ 1398 } 1399 1400 void proc_sched_show_task(struct task_struct *p, struct pid_namespace *ns, 1401 struct seq_file *m) 1402 { 1403 unsigned long nr_switches; 1404 1405 SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr_ns(p, ns), 1406 get_nr_threads(p)); 1407 SEQ_printf(m, 1408 "---------------------------------------------------------" 1409 "----------\n"); 1410 1411 #define P_SCHEDSTAT(F) __PS(#F, schedstat_val(p->stats.F)) 1412 #define PN_SCHEDSTAT(F) __PSN(#F, schedstat_val(p->stats.F)) 1413 1414 PN(se.exec_start); 1415 PN(se.vruntime); 1416 PN(se.sum_exec_runtime); 1417 1418 nr_switches = p->nvcsw + p->nivcsw; 1419 1420 P(se.nr_migrations); 1421 1422 if (schedstat_enabled()) { 1423 u64 avg_atom, avg_per_cpu; 1424 1425 PN_SCHEDSTAT(sum_sleep_runtime); 1426 PN_SCHEDSTAT(sum_block_runtime); 1427 PN_SCHEDSTAT(wait_start); 1428 PN_SCHEDSTAT(sleep_start); 1429 PN_SCHEDSTAT(block_start); 1430 PN_SCHEDSTAT(sleep_max); 1431 PN_SCHEDSTAT(block_max); 1432 PN_SCHEDSTAT(exec_max); 1433 PN_SCHEDSTAT(slice_max); 1434 PN_SCHEDSTAT(wait_max); 1435 PN_SCHEDSTAT(wait_sum); 1436 P_SCHEDSTAT(wait_count); 1437 PN_SCHEDSTAT(iowait_sum); 1438 P_SCHEDSTAT(iowait_count); 1439 P_SCHEDSTAT(nr_failed_migrations_affine); 1440 P_SCHEDSTAT(nr_failed_migrations_running); 1441 P_SCHEDSTAT(nr_failed_migrations_hot); 1442 P_SCHEDSTAT(nr_forced_migrations); 1443 P_SCHEDSTAT(nr_migrations_cpu_non_preferred); 1444 P_SCHEDSTAT(nr_wakeups); 1445 P_SCHEDSTAT(nr_wakeups_sync); 1446 P_SCHEDSTAT(nr_wakeups_migrate); 1447 P_SCHEDSTAT(nr_wakeups_local); 1448 P_SCHEDSTAT(nr_wakeups_remote); 1449 P_SCHEDSTAT(nr_wakeups_affine); 1450 P_SCHEDSTAT(nr_wakeups_affine_attempts); 1451 1452 avg_atom = p->se.sum_exec_runtime; 1453 if (nr_switches) 1454 avg_atom = div64_ul(avg_atom, nr_switches); 1455 else 1456 avg_atom = -1LL; 1457 1458 avg_per_cpu = p->se.sum_exec_runtime; 1459 if (p->se.nr_migrations) { 1460 avg_per_cpu = div64_u64(avg_per_cpu, 1461 p->se.nr_migrations); 1462 } else { 1463 avg_per_cpu = -1LL; 1464 } 1465 1466 __PN(avg_atom); 1467 __PN(avg_per_cpu); 1468 1469 #ifdef CONFIG_SCHED_CORE 1470 PN_SCHEDSTAT(core_forceidle_sum); 1471 #endif 1472 } 1473 1474 __P(nr_switches); 1475 __PS("nr_voluntary_switches", p->nvcsw); 1476 __PS("nr_involuntary_switches", p->nivcsw); 1477 1478 P(se.load.weight); 1479 P(se.avg.load_sum); 1480 P(se.avg.runnable_sum); 1481 P(se.avg.util_sum); 1482 P(se.avg.load_avg); 1483 P(se.avg.runnable_avg); 1484 P(se.avg.util_avg); 1485 P(se.avg.last_update_time); 1486 PM(se.avg.util_est, ~UTIL_AVG_UNCHANGED); 1487 #ifdef CONFIG_UCLAMP_TASK 1488 __PS("uclamp.min", p->uclamp_req[UCLAMP_MIN].value); 1489 __PS("uclamp.max", p->uclamp_req[UCLAMP_MAX].value); 1490 __PS("effective uclamp.min", uclamp_eff_value(p, UCLAMP_MIN)); 1491 __PS("effective uclamp.max", uclamp_eff_value(p, UCLAMP_MAX)); 1492 #endif /* CONFIG_UCLAMP_TASK */ 1493 P(policy); 1494 P(prio); 1495 if (task_has_dl_policy(p)) { 1496 P(dl.runtime); 1497 P(dl.deadline); 1498 } else if (fair_policy(p->policy)) { 1499 P(se.slice); 1500 } 1501 #ifdef CONFIG_SCHED_CLASS_EXT 1502 __PS("ext.enabled", task_on_scx(p)); 1503 #endif 1504 #undef PN_SCHEDSTAT 1505 #undef P_SCHEDSTAT 1506 1507 { 1508 unsigned int this_cpu = raw_smp_processor_id(); 1509 u64 t0, t1; 1510 1511 t0 = cpu_clock(this_cpu); 1512 t1 = cpu_clock(this_cpu); 1513 __PS("clock-delta", t1-t0); 1514 } 1515 1516 sched_show_numa(p, m); 1517 } 1518 1519 void proc_sched_set_task(struct task_struct *p) 1520 { 1521 #ifdef CONFIG_SCHEDSTATS 1522 memset(&p->stats, 0, sizeof(p->stats)); 1523 #endif 1524 } 1525 1526 void resched_latency_warn(int cpu, u64 latency) 1527 { 1528 static DEFINE_RATELIMIT_STATE(latency_check_ratelimit, 60 * 60 * HZ, 1); 1529 1530 if (likely(!__ratelimit(&latency_check_ratelimit))) 1531 return; 1532 1533 pr_err("sched: CPU %d need_resched set for > %llu ns (%d ticks) without schedule\n", 1534 cpu, latency, cpu_rq(cpu)->ticks_without_resched); 1535 dump_stack(); 1536 } 1537