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