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