1/* 2 * kernel/sched/debug.c 3 * 4 * Print the CFS rbtree 5 * 6 * Copyright(C) 2007, Red Hat, Inc., Ingo Molnar 7 * 8 * This program is free software; you can redistribute it and/or modify 9 * it under the terms of the GNU General Public License version 2 as 10 * published by the Free Software Foundation. 11 */ 12 13#include <linux/proc_fs.h> 14#include <linux/sched.h> 15#include <linux/seq_file.h> 16#include <linux/kallsyms.h> 17#include <linux/utsname.h> 18#include <linux/mempolicy.h> 19 20#include "sched.h" 21 22static DEFINE_SPINLOCK(sched_debug_lock); 23 24/* 25 * This allows printing both to /proc/sched_debug and 26 * to the console 27 */ 28#define SEQ_printf(m, x...) \ 29 do { \ 30 if (m) \ 31 seq_printf(m, x); \ 32 else \ 33 printk(x); \ 34 } while (0) 35 36/* 37 * Ease the printing of nsec fields: 38 */ 39static long long nsec_high(unsigned long long nsec) 40{ 41 if ((long long)nsec < 0) { 42 nsec = -nsec; 43 do_div(nsec, 1000000); 44 return -nsec; 45 } 46 do_div(nsec, 1000000); 47 48 return nsec; 49} 50 51static unsigned long nsec_low(unsigned long long nsec) 52{ 53 if ((long long)nsec < 0) 54 nsec = -nsec; 55 56 return do_div(nsec, 1000000); 57} 58 59#define SPLIT_NS(x) nsec_high(x), nsec_low(x) 60 61#ifdef CONFIG_FAIR_GROUP_SCHED 62static void print_cfs_group_stats(struct seq_file *m, int cpu, struct task_group *tg) 63{ 64 struct sched_entity *se = tg->se[cpu]; 65 66#define P(F) \ 67 SEQ_printf(m, " .%-30s: %lld\n", #F, (long long)F) 68#define PN(F) \ 69 SEQ_printf(m, " .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F)) 70 71 if (!se) { 72 struct sched_avg *avg = &cpu_rq(cpu)->avg; 73 P(avg->runnable_avg_sum); 74 P(avg->avg_period); 75 return; 76 } 77 78 79 PN(se->exec_start); 80 PN(se->vruntime); 81 PN(se->sum_exec_runtime); 82#ifdef CONFIG_SCHEDSTATS 83 PN(se->statistics.wait_start); 84 PN(se->statistics.sleep_start); 85 PN(se->statistics.block_start); 86 PN(se->statistics.sleep_max); 87 PN(se->statistics.block_max); 88 PN(se->statistics.exec_max); 89 PN(se->statistics.slice_max); 90 PN(se->statistics.wait_max); 91 PN(se->statistics.wait_sum); 92 P(se->statistics.wait_count); 93#endif 94 P(se->load.weight); 95#ifdef CONFIG_SMP 96 P(se->avg.runnable_avg_sum); 97 P(se->avg.running_avg_sum); 98 P(se->avg.avg_period); 99 P(se->avg.load_avg_contrib); 100 P(se->avg.utilization_avg_contrib); 101 P(se->avg.decay_count); 102#endif 103#undef PN 104#undef P 105} 106#endif 107 108#ifdef CONFIG_CGROUP_SCHED 109static char group_path[PATH_MAX]; 110 111static char *task_group_path(struct task_group *tg) 112{ 113 if (autogroup_path(tg, group_path, PATH_MAX)) 114 return group_path; 115 116 return cgroup_path(tg->css.cgroup, group_path, PATH_MAX); 117} 118#endif 119 120static void 121print_task(struct seq_file *m, struct rq *rq, struct task_struct *p) 122{ 123 if (rq->curr == p) 124 SEQ_printf(m, "R"); 125 else 126 SEQ_printf(m, " "); 127 128 SEQ_printf(m, "%15s %5d %9Ld.%06ld %9Ld %5d ", 129 p->comm, task_pid_nr(p), 130 SPLIT_NS(p->se.vruntime), 131 (long long)(p->nvcsw + p->nivcsw), 132 p->prio); 133#ifdef CONFIG_SCHEDSTATS 134 SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld", 135 SPLIT_NS(p->se.vruntime), 136 SPLIT_NS(p->se.sum_exec_runtime), 137 SPLIT_NS(p->se.statistics.sum_sleep_runtime)); 138#else 139 SEQ_printf(m, "%15Ld %15Ld %15Ld.%06ld %15Ld.%06ld %15Ld.%06ld", 140 0LL, 0LL, 0LL, 0L, 0LL, 0L, 0LL, 0L); 141#endif 142#ifdef CONFIG_NUMA_BALANCING 143 SEQ_printf(m, " %d", task_node(p)); 144#endif 145#ifdef CONFIG_CGROUP_SCHED 146 SEQ_printf(m, " %s", task_group_path(task_group(p))); 147#endif 148 149 SEQ_printf(m, "\n"); 150} 151 152static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu) 153{ 154 struct task_struct *g, *p; 155 156 SEQ_printf(m, 157 "\nrunnable tasks:\n" 158 " task PID tree-key switches prio" 159 " exec-runtime sum-exec sum-sleep\n" 160 "------------------------------------------------------" 161 "----------------------------------------------------\n"); 162 163 rcu_read_lock(); 164 for_each_process_thread(g, p) { 165 if (task_cpu(p) != rq_cpu) 166 continue; 167 168 print_task(m, rq, p); 169 } 170 rcu_read_unlock(); 171} 172 173void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq) 174{ 175 s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1, 176 spread, rq0_min_vruntime, spread0; 177 struct rq *rq = cpu_rq(cpu); 178 struct sched_entity *last; 179 unsigned long flags; 180 181#ifdef CONFIG_FAIR_GROUP_SCHED 182 SEQ_printf(m, "\ncfs_rq[%d]:%s\n", cpu, task_group_path(cfs_rq->tg)); 183#else 184 SEQ_printf(m, "\ncfs_rq[%d]:\n", cpu); 185#endif 186 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "exec_clock", 187 SPLIT_NS(cfs_rq->exec_clock)); 188 189 raw_spin_lock_irqsave(&rq->lock, flags); 190 if (cfs_rq->rb_leftmost) 191 MIN_vruntime = (__pick_first_entity(cfs_rq))->vruntime; 192 last = __pick_last_entity(cfs_rq); 193 if (last) 194 max_vruntime = last->vruntime; 195 min_vruntime = cfs_rq->min_vruntime; 196 rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime; 197 raw_spin_unlock_irqrestore(&rq->lock, flags); 198 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "MIN_vruntime", 199 SPLIT_NS(MIN_vruntime)); 200 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "min_vruntime", 201 SPLIT_NS(min_vruntime)); 202 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "max_vruntime", 203 SPLIT_NS(max_vruntime)); 204 spread = max_vruntime - MIN_vruntime; 205 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread", 206 SPLIT_NS(spread)); 207 spread0 = min_vruntime - rq0_min_vruntime; 208 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread0", 209 SPLIT_NS(spread0)); 210 SEQ_printf(m, " .%-30s: %d\n", "nr_spread_over", 211 cfs_rq->nr_spread_over); 212 SEQ_printf(m, " .%-30s: %d\n", "nr_running", cfs_rq->nr_running); 213 SEQ_printf(m, " .%-30s: %ld\n", "load", cfs_rq->load.weight); 214#ifdef CONFIG_SMP 215 SEQ_printf(m, " .%-30s: %ld\n", "runnable_load_avg", 216 cfs_rq->runnable_load_avg); 217 SEQ_printf(m, " .%-30s: %ld\n", "blocked_load_avg", 218 cfs_rq->blocked_load_avg); 219 SEQ_printf(m, " .%-30s: %ld\n", "utilization_load_avg", 220 cfs_rq->utilization_load_avg); 221#ifdef CONFIG_FAIR_GROUP_SCHED 222 SEQ_printf(m, " .%-30s: %ld\n", "tg_load_contrib", 223 cfs_rq->tg_load_contrib); 224 SEQ_printf(m, " .%-30s: %d\n", "tg_runnable_contrib", 225 cfs_rq->tg_runnable_contrib); 226 SEQ_printf(m, " .%-30s: %ld\n", "tg_load_avg", 227 atomic_long_read(&cfs_rq->tg->load_avg)); 228 SEQ_printf(m, " .%-30s: %d\n", "tg->runnable_avg", 229 atomic_read(&cfs_rq->tg->runnable_avg)); 230#endif 231#endif 232#ifdef CONFIG_CFS_BANDWIDTH 233 SEQ_printf(m, " .%-30s: %d\n", "tg->cfs_bandwidth.timer_active", 234 cfs_rq->tg->cfs_bandwidth.timer_active); 235 SEQ_printf(m, " .%-30s: %d\n", "throttled", 236 cfs_rq->throttled); 237 SEQ_printf(m, " .%-30s: %d\n", "throttle_count", 238 cfs_rq->throttle_count); 239#endif 240 241#ifdef CONFIG_FAIR_GROUP_SCHED 242 print_cfs_group_stats(m, cpu, cfs_rq->tg); 243#endif 244} 245 246void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq) 247{ 248#ifdef CONFIG_RT_GROUP_SCHED 249 SEQ_printf(m, "\nrt_rq[%d]:%s\n", cpu, task_group_path(rt_rq->tg)); 250#else 251 SEQ_printf(m, "\nrt_rq[%d]:\n", cpu); 252#endif 253 254#define P(x) \ 255 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rt_rq->x)) 256#define PN(x) \ 257 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x)) 258 259 P(rt_nr_running); 260 P(rt_throttled); 261 PN(rt_time); 262 PN(rt_runtime); 263 264#undef PN 265#undef P 266} 267 268void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq) 269{ 270 SEQ_printf(m, "\ndl_rq[%d]:\n", cpu); 271 SEQ_printf(m, " .%-30s: %ld\n", "dl_nr_running", dl_rq->dl_nr_running); 272} 273 274extern __read_mostly int sched_clock_running; 275 276static void print_cpu(struct seq_file *m, int cpu) 277{ 278 struct rq *rq = cpu_rq(cpu); 279 unsigned long flags; 280 281#ifdef CONFIG_X86 282 { 283 unsigned int freq = cpu_khz ? : 1; 284 285 SEQ_printf(m, "cpu#%d, %u.%03u MHz\n", 286 cpu, freq / 1000, (freq % 1000)); 287 } 288#else 289 SEQ_printf(m, "cpu#%d\n", cpu); 290#endif 291 292#define P(x) \ 293do { \ 294 if (sizeof(rq->x) == 4) \ 295 SEQ_printf(m, " .%-30s: %ld\n", #x, (long)(rq->x)); \ 296 else \ 297 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rq->x));\ 298} while (0) 299 300#define PN(x) \ 301 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x)) 302 303 P(nr_running); 304 SEQ_printf(m, " .%-30s: %lu\n", "load", 305 rq->load.weight); 306 P(nr_switches); 307 P(nr_load_updates); 308 P(nr_uninterruptible); 309 PN(next_balance); 310 SEQ_printf(m, " .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr))); 311 PN(clock); 312 PN(clock_task); 313 P(cpu_load[0]); 314 P(cpu_load[1]); 315 P(cpu_load[2]); 316 P(cpu_load[3]); 317 P(cpu_load[4]); 318#undef P 319#undef PN 320 321#ifdef CONFIG_SCHEDSTATS 322#define P(n) SEQ_printf(m, " .%-30s: %d\n", #n, rq->n); 323#define P64(n) SEQ_printf(m, " .%-30s: %Ld\n", #n, rq->n); 324 325 P(yld_count); 326 327 P(sched_count); 328 P(sched_goidle); 329#ifdef CONFIG_SMP 330 P64(avg_idle); 331 P64(max_idle_balance_cost); 332#endif 333 334 P(ttwu_count); 335 P(ttwu_local); 336 337#undef P 338#undef P64 339#endif 340 spin_lock_irqsave(&sched_debug_lock, flags); 341 print_cfs_stats(m, cpu); 342 print_rt_stats(m, cpu); 343 print_dl_stats(m, cpu); 344 345 print_rq(m, rq, cpu); 346 spin_unlock_irqrestore(&sched_debug_lock, flags); 347 SEQ_printf(m, "\n"); 348} 349 350static const char *sched_tunable_scaling_names[] = { 351 "none", 352 "logaritmic", 353 "linear" 354}; 355 356static void sched_debug_header(struct seq_file *m) 357{ 358 u64 ktime, sched_clk, cpu_clk; 359 unsigned long flags; 360 361 local_irq_save(flags); 362 ktime = ktime_to_ns(ktime_get()); 363 sched_clk = sched_clock(); 364 cpu_clk = local_clock(); 365 local_irq_restore(flags); 366 367 SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n", 368 init_utsname()->release, 369 (int)strcspn(init_utsname()->version, " "), 370 init_utsname()->version); 371 372#define P(x) \ 373 SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x)) 374#define PN(x) \ 375 SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x)) 376 PN(ktime); 377 PN(sched_clk); 378 PN(cpu_clk); 379 P(jiffies); 380#ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK 381 P(sched_clock_stable()); 382#endif 383#undef PN 384#undef P 385 386 SEQ_printf(m, "\n"); 387 SEQ_printf(m, "sysctl_sched\n"); 388 389#define P(x) \ 390 SEQ_printf(m, " .%-40s: %Ld\n", #x, (long long)(x)) 391#define PN(x) \ 392 SEQ_printf(m, " .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x)) 393 PN(sysctl_sched_latency); 394 PN(sysctl_sched_min_granularity); 395 PN(sysctl_sched_wakeup_granularity); 396 P(sysctl_sched_child_runs_first); 397 P(sysctl_sched_features); 398#undef PN 399#undef P 400 401 SEQ_printf(m, " .%-40s: %d (%s)\n", 402 "sysctl_sched_tunable_scaling", 403 sysctl_sched_tunable_scaling, 404 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]); 405 SEQ_printf(m, "\n"); 406} 407 408static int sched_debug_show(struct seq_file *m, void *v) 409{ 410 int cpu = (unsigned long)(v - 2); 411 412 if (cpu != -1) 413 print_cpu(m, cpu); 414 else 415 sched_debug_header(m); 416 417 return 0; 418} 419 420void sysrq_sched_debug_show(void) 421{ 422 int cpu; 423 424 sched_debug_header(NULL); 425 for_each_online_cpu(cpu) 426 print_cpu(NULL, cpu); 427 428} 429 430/* 431 * This itererator needs some explanation. 432 * It returns 1 for the header position. 433 * This means 2 is cpu 0. 434 * In a hotplugged system some cpus, including cpu 0, may be missing so we have 435 * to use cpumask_* to iterate over the cpus. 436 */ 437static void *sched_debug_start(struct seq_file *file, loff_t *offset) 438{ 439 unsigned long n = *offset; 440 441 if (n == 0) 442 return (void *) 1; 443 444 n--; 445 446 if (n > 0) 447 n = cpumask_next(n - 1, cpu_online_mask); 448 else 449 n = cpumask_first(cpu_online_mask); 450 451 *offset = n + 1; 452 453 if (n < nr_cpu_ids) 454 return (void *)(unsigned long)(n + 2); 455 return NULL; 456} 457 458static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset) 459{ 460 (*offset)++; 461 return sched_debug_start(file, offset); 462} 463 464static void sched_debug_stop(struct seq_file *file, void *data) 465{ 466} 467 468static const struct seq_operations sched_debug_sops = { 469 .start = sched_debug_start, 470 .next = sched_debug_next, 471 .stop = sched_debug_stop, 472 .show = sched_debug_show, 473}; 474 475static int sched_debug_release(struct inode *inode, struct file *file) 476{ 477 seq_release(inode, file); 478 479 return 0; 480} 481 482static int sched_debug_open(struct inode *inode, struct file *filp) 483{ 484 int ret = 0; 485 486 ret = seq_open(filp, &sched_debug_sops); 487 488 return ret; 489} 490 491static const struct file_operations sched_debug_fops = { 492 .open = sched_debug_open, 493 .read = seq_read, 494 .llseek = seq_lseek, 495 .release = sched_debug_release, 496}; 497 498static int __init init_sched_debug_procfs(void) 499{ 500 struct proc_dir_entry *pe; 501 502 pe = proc_create("sched_debug", 0444, NULL, &sched_debug_fops); 503 if (!pe) 504 return -ENOMEM; 505 return 0; 506} 507 508__initcall(init_sched_debug_procfs); 509 510#define __P(F) \ 511 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)F) 512#define P(F) \ 513 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)p->F) 514#define __PN(F) \ 515 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F)) 516#define PN(F) \ 517 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F)) 518 519 520static void sched_show_numa(struct task_struct *p, struct seq_file *m) 521{ 522#ifdef CONFIG_NUMA_BALANCING 523 struct mempolicy *pol; 524 int node, i; 525 526 if (p->mm) 527 P(mm->numa_scan_seq); 528 529 task_lock(p); 530 pol = p->mempolicy; 531 if (pol && !(pol->flags & MPOL_F_MORON)) 532 pol = NULL; 533 mpol_get(pol); 534 task_unlock(p); 535 536 SEQ_printf(m, "numa_migrations, %ld\n", xchg(&p->numa_pages_migrated, 0)); 537 538 for_each_online_node(node) { 539 for (i = 0; i < 2; i++) { 540 unsigned long nr_faults = -1; 541 int cpu_current, home_node; 542 543 if (p->numa_faults) 544 nr_faults = p->numa_faults[2*node + i]; 545 546 cpu_current = !i ? (task_node(p) == node) : 547 (pol && node_isset(node, pol->v.nodes)); 548 549 home_node = (p->numa_preferred_nid == node); 550 551 SEQ_printf(m, "numa_faults_memory, %d, %d, %d, %d, %ld\n", 552 i, node, cpu_current, home_node, nr_faults); 553 } 554 } 555 556 mpol_put(pol); 557#endif 558} 559 560void proc_sched_show_task(struct task_struct *p, struct seq_file *m) 561{ 562 unsigned long nr_switches; 563 564 SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr(p), 565 get_nr_threads(p)); 566 SEQ_printf(m, 567 "---------------------------------------------------------" 568 "----------\n"); 569#define __P(F) \ 570 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)F) 571#define P(F) \ 572 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)p->F) 573#define __PN(F) \ 574 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F)) 575#define PN(F) \ 576 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F)) 577 578 PN(se.exec_start); 579 PN(se.vruntime); 580 PN(se.sum_exec_runtime); 581 582 nr_switches = p->nvcsw + p->nivcsw; 583 584#ifdef CONFIG_SCHEDSTATS 585 PN(se.statistics.wait_start); 586 PN(se.statistics.sleep_start); 587 PN(se.statistics.block_start); 588 PN(se.statistics.sleep_max); 589 PN(se.statistics.block_max); 590 PN(se.statistics.exec_max); 591 PN(se.statistics.slice_max); 592 PN(se.statistics.wait_max); 593 PN(se.statistics.wait_sum); 594 P(se.statistics.wait_count); 595 PN(se.statistics.iowait_sum); 596 P(se.statistics.iowait_count); 597 P(se.nr_migrations); 598 P(se.statistics.nr_migrations_cold); 599 P(se.statistics.nr_failed_migrations_affine); 600 P(se.statistics.nr_failed_migrations_running); 601 P(se.statistics.nr_failed_migrations_hot); 602 P(se.statistics.nr_forced_migrations); 603 P(se.statistics.nr_wakeups); 604 P(se.statistics.nr_wakeups_sync); 605 P(se.statistics.nr_wakeups_migrate); 606 P(se.statistics.nr_wakeups_local); 607 P(se.statistics.nr_wakeups_remote); 608 P(se.statistics.nr_wakeups_affine); 609 P(se.statistics.nr_wakeups_affine_attempts); 610 P(se.statistics.nr_wakeups_passive); 611 P(se.statistics.nr_wakeups_idle); 612 613 { 614 u64 avg_atom, avg_per_cpu; 615 616 avg_atom = p->se.sum_exec_runtime; 617 if (nr_switches) 618 avg_atom = div64_ul(avg_atom, nr_switches); 619 else 620 avg_atom = -1LL; 621 622 avg_per_cpu = p->se.sum_exec_runtime; 623 if (p->se.nr_migrations) { 624 avg_per_cpu = div64_u64(avg_per_cpu, 625 p->se.nr_migrations); 626 } else { 627 avg_per_cpu = -1LL; 628 } 629 630 __PN(avg_atom); 631 __PN(avg_per_cpu); 632 } 633#endif 634 __P(nr_switches); 635 SEQ_printf(m, "%-45s:%21Ld\n", 636 "nr_voluntary_switches", (long long)p->nvcsw); 637 SEQ_printf(m, "%-45s:%21Ld\n", 638 "nr_involuntary_switches", (long long)p->nivcsw); 639 640 P(se.load.weight); 641#ifdef CONFIG_SMP 642 P(se.avg.runnable_avg_sum); 643 P(se.avg.running_avg_sum); 644 P(se.avg.avg_period); 645 P(se.avg.load_avg_contrib); 646 P(se.avg.utilization_avg_contrib); 647 P(se.avg.decay_count); 648#endif 649 P(policy); 650 P(prio); 651#undef PN 652#undef __PN 653#undef P 654#undef __P 655 656 { 657 unsigned int this_cpu = raw_smp_processor_id(); 658 u64 t0, t1; 659 660 t0 = cpu_clock(this_cpu); 661 t1 = cpu_clock(this_cpu); 662 SEQ_printf(m, "%-45s:%21Ld\n", 663 "clock-delta", (long long)(t1-t0)); 664 } 665 666 sched_show_numa(p, m); 667} 668 669void proc_sched_set_task(struct task_struct *p) 670{ 671#ifdef CONFIG_SCHEDSTATS 672 memset(&p->se.statistics, 0, sizeof(p->se.statistics)); 673#endif 674} 675