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 19 #include "sched.h" 20 21 static DEFINE_SPINLOCK(sched_debug_lock); 22 23 /* 24 * This allows printing both to /proc/sched_debug and 25 * to the console 26 */ 27 #define SEQ_printf(m, x...) \ 28 do { \ 29 if (m) \ 30 seq_printf(m, x); \ 31 else \ 32 printk(x); \ 33 } while (0) 34 35 /* 36 * Ease the printing of nsec fields: 37 */ 38 static long long nsec_high(unsigned long long nsec) 39 { 40 if ((long long)nsec < 0) { 41 nsec = -nsec; 42 do_div(nsec, 1000000); 43 return -nsec; 44 } 45 do_div(nsec, 1000000); 46 47 return nsec; 48 } 49 50 static unsigned long nsec_low(unsigned long long nsec) 51 { 52 if ((long long)nsec < 0) 53 nsec = -nsec; 54 55 return do_div(nsec, 1000000); 56 } 57 58 #define SPLIT_NS(x) nsec_high(x), nsec_low(x) 59 60 #ifdef CONFIG_FAIR_GROUP_SCHED 61 static void print_cfs_group_stats(struct seq_file *m, int cpu, struct task_group *tg) 62 { 63 struct sched_entity *se = tg->se[cpu]; 64 65 #define P(F) \ 66 SEQ_printf(m, " .%-30s: %lld\n", #F, (long long)F) 67 #define PN(F) \ 68 SEQ_printf(m, " .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F)) 69 70 if (!se) { 71 struct sched_avg *avg = &cpu_rq(cpu)->avg; 72 P(avg->runnable_avg_sum); 73 P(avg->runnable_avg_period); 74 return; 75 } 76 77 78 PN(se->exec_start); 79 PN(se->vruntime); 80 PN(se->sum_exec_runtime); 81 #ifdef CONFIG_SCHEDSTATS 82 PN(se->statistics.wait_start); 83 PN(se->statistics.sleep_start); 84 PN(se->statistics.block_start); 85 PN(se->statistics.sleep_max); 86 PN(se->statistics.block_max); 87 PN(se->statistics.exec_max); 88 PN(se->statistics.slice_max); 89 PN(se->statistics.wait_max); 90 PN(se->statistics.wait_sum); 91 P(se->statistics.wait_count); 92 #endif 93 P(se->load.weight); 94 #ifdef CONFIG_SMP 95 P(se->avg.runnable_avg_sum); 96 P(se->avg.runnable_avg_period); 97 P(se->avg.load_avg_contrib); 98 P(se->avg.decay_count); 99 #endif 100 #undef PN 101 #undef P 102 } 103 #endif 104 105 #ifdef CONFIG_CGROUP_SCHED 106 static char group_path[PATH_MAX]; 107 108 static char *task_group_path(struct task_group *tg) 109 { 110 if (autogroup_path(tg, group_path, PATH_MAX)) 111 return group_path; 112 113 /* 114 * May be NULL if the underlying cgroup isn't fully-created yet 115 */ 116 if (!tg->css.cgroup) { 117 group_path[0] = '\0'; 118 return group_path; 119 } 120 cgroup_path(tg->css.cgroup, group_path, PATH_MAX); 121 return group_path; 122 } 123 #endif 124 125 static void 126 print_task(struct seq_file *m, struct rq *rq, struct task_struct *p) 127 { 128 if (rq->curr == p) 129 SEQ_printf(m, "R"); 130 else 131 SEQ_printf(m, " "); 132 133 SEQ_printf(m, "%15s %5d %9Ld.%06ld %9Ld %5d ", 134 p->comm, p->pid, 135 SPLIT_NS(p->se.vruntime), 136 (long long)(p->nvcsw + p->nivcsw), 137 p->prio); 138 #ifdef CONFIG_SCHEDSTATS 139 SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld", 140 SPLIT_NS(p->se.vruntime), 141 SPLIT_NS(p->se.sum_exec_runtime), 142 SPLIT_NS(p->se.statistics.sum_sleep_runtime)); 143 #else 144 SEQ_printf(m, "%15Ld %15Ld %15Ld.%06ld %15Ld.%06ld %15Ld.%06ld", 145 0LL, 0LL, 0LL, 0L, 0LL, 0L, 0LL, 0L); 146 #endif 147 #ifdef CONFIG_CGROUP_SCHED 148 SEQ_printf(m, " %s", task_group_path(task_group(p))); 149 #endif 150 151 SEQ_printf(m, "\n"); 152 } 153 154 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu) 155 { 156 struct task_struct *g, *p; 157 unsigned long flags; 158 159 SEQ_printf(m, 160 "\nrunnable tasks:\n" 161 " task PID tree-key switches prio" 162 " exec-runtime sum-exec sum-sleep\n" 163 "------------------------------------------------------" 164 "----------------------------------------------------\n"); 165 166 read_lock_irqsave(&tasklist_lock, flags); 167 168 do_each_thread(g, p) { 169 if (!p->on_rq || task_cpu(p) != rq_cpu) 170 continue; 171 172 print_task(m, rq, p); 173 } while_each_thread(g, p); 174 175 read_unlock_irqrestore(&tasklist_lock, flags); 176 } 177 178 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq) 179 { 180 s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1, 181 spread, rq0_min_vruntime, spread0; 182 struct rq *rq = cpu_rq(cpu); 183 struct sched_entity *last; 184 unsigned long flags; 185 186 #ifdef CONFIG_FAIR_GROUP_SCHED 187 SEQ_printf(m, "\ncfs_rq[%d]:%s\n", cpu, task_group_path(cfs_rq->tg)); 188 #else 189 SEQ_printf(m, "\ncfs_rq[%d]:\n", cpu); 190 #endif 191 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "exec_clock", 192 SPLIT_NS(cfs_rq->exec_clock)); 193 194 raw_spin_lock_irqsave(&rq->lock, flags); 195 if (cfs_rq->rb_leftmost) 196 MIN_vruntime = (__pick_first_entity(cfs_rq))->vruntime; 197 last = __pick_last_entity(cfs_rq); 198 if (last) 199 max_vruntime = last->vruntime; 200 min_vruntime = cfs_rq->min_vruntime; 201 rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime; 202 raw_spin_unlock_irqrestore(&rq->lock, flags); 203 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "MIN_vruntime", 204 SPLIT_NS(MIN_vruntime)); 205 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "min_vruntime", 206 SPLIT_NS(min_vruntime)); 207 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "max_vruntime", 208 SPLIT_NS(max_vruntime)); 209 spread = max_vruntime - MIN_vruntime; 210 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread", 211 SPLIT_NS(spread)); 212 spread0 = min_vruntime - rq0_min_vruntime; 213 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread0", 214 SPLIT_NS(spread0)); 215 SEQ_printf(m, " .%-30s: %d\n", "nr_spread_over", 216 cfs_rq->nr_spread_over); 217 SEQ_printf(m, " .%-30s: %d\n", "nr_running", cfs_rq->nr_running); 218 SEQ_printf(m, " .%-30s: %ld\n", "load", cfs_rq->load.weight); 219 #ifdef CONFIG_FAIR_GROUP_SCHED 220 #ifdef CONFIG_SMP 221 SEQ_printf(m, " .%-30s: %lld\n", "runnable_load_avg", 222 cfs_rq->runnable_load_avg); 223 SEQ_printf(m, " .%-30s: %lld\n", "blocked_load_avg", 224 cfs_rq->blocked_load_avg); 225 SEQ_printf(m, " .%-30s: %lld\n", "tg_load_avg", 226 (unsigned long long)atomic64_read(&cfs_rq->tg->load_avg)); 227 SEQ_printf(m, " .%-30s: %lld\n", "tg_load_contrib", 228 cfs_rq->tg_load_contrib); 229 SEQ_printf(m, " .%-30s: %d\n", "tg_runnable_contrib", 230 cfs_rq->tg_runnable_contrib); 231 SEQ_printf(m, " .%-30s: %d\n", "tg->runnable_avg", 232 atomic_read(&cfs_rq->tg->runnable_avg)); 233 #endif 234 235 print_cfs_group_stats(m, cpu, cfs_rq->tg); 236 #endif 237 } 238 239 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq) 240 { 241 #ifdef CONFIG_RT_GROUP_SCHED 242 SEQ_printf(m, "\nrt_rq[%d]:%s\n", cpu, task_group_path(rt_rq->tg)); 243 #else 244 SEQ_printf(m, "\nrt_rq[%d]:\n", cpu); 245 #endif 246 247 #define P(x) \ 248 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rt_rq->x)) 249 #define PN(x) \ 250 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x)) 251 252 P(rt_nr_running); 253 P(rt_throttled); 254 PN(rt_time); 255 PN(rt_runtime); 256 257 #undef PN 258 #undef P 259 } 260 261 extern __read_mostly int sched_clock_running; 262 263 static void print_cpu(struct seq_file *m, int cpu) 264 { 265 struct rq *rq = cpu_rq(cpu); 266 unsigned long flags; 267 268 #ifdef CONFIG_X86 269 { 270 unsigned int freq = cpu_khz ? : 1; 271 272 SEQ_printf(m, "\ncpu#%d, %u.%03u MHz\n", 273 cpu, freq / 1000, (freq % 1000)); 274 } 275 #else 276 SEQ_printf(m, "\ncpu#%d\n", cpu); 277 #endif 278 279 #define P(x) \ 280 do { \ 281 if (sizeof(rq->x) == 4) \ 282 SEQ_printf(m, " .%-30s: %ld\n", #x, (long)(rq->x)); \ 283 else \ 284 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rq->x));\ 285 } while (0) 286 287 #define PN(x) \ 288 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x)) 289 290 P(nr_running); 291 SEQ_printf(m, " .%-30s: %lu\n", "load", 292 rq->load.weight); 293 P(nr_switches); 294 P(nr_load_updates); 295 P(nr_uninterruptible); 296 PN(next_balance); 297 P(curr->pid); 298 PN(clock); 299 P(cpu_load[0]); 300 P(cpu_load[1]); 301 P(cpu_load[2]); 302 P(cpu_load[3]); 303 P(cpu_load[4]); 304 #undef P 305 #undef PN 306 307 #ifdef CONFIG_SCHEDSTATS 308 #define P(n) SEQ_printf(m, " .%-30s: %d\n", #n, rq->n); 309 #define P64(n) SEQ_printf(m, " .%-30s: %Ld\n", #n, rq->n); 310 311 P(yld_count); 312 313 P(sched_count); 314 P(sched_goidle); 315 #ifdef CONFIG_SMP 316 P64(avg_idle); 317 #endif 318 319 P(ttwu_count); 320 P(ttwu_local); 321 322 #undef P 323 #undef P64 324 #endif 325 spin_lock_irqsave(&sched_debug_lock, flags); 326 print_cfs_stats(m, cpu); 327 print_rt_stats(m, cpu); 328 329 rcu_read_lock(); 330 print_rq(m, rq, cpu); 331 rcu_read_unlock(); 332 spin_unlock_irqrestore(&sched_debug_lock, flags); 333 } 334 335 static const char *sched_tunable_scaling_names[] = { 336 "none", 337 "logaritmic", 338 "linear" 339 }; 340 341 static int sched_debug_show(struct seq_file *m, void *v) 342 { 343 u64 ktime, sched_clk, cpu_clk; 344 unsigned long flags; 345 int cpu; 346 347 local_irq_save(flags); 348 ktime = ktime_to_ns(ktime_get()); 349 sched_clk = sched_clock(); 350 cpu_clk = local_clock(); 351 local_irq_restore(flags); 352 353 SEQ_printf(m, "Sched Debug Version: v0.10, %s %.*s\n", 354 init_utsname()->release, 355 (int)strcspn(init_utsname()->version, " "), 356 init_utsname()->version); 357 358 #define P(x) \ 359 SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x)) 360 #define PN(x) \ 361 SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x)) 362 PN(ktime); 363 PN(sched_clk); 364 PN(cpu_clk); 365 P(jiffies); 366 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK 367 P(sched_clock_stable); 368 #endif 369 #undef PN 370 #undef P 371 372 SEQ_printf(m, "\n"); 373 SEQ_printf(m, "sysctl_sched\n"); 374 375 #define P(x) \ 376 SEQ_printf(m, " .%-40s: %Ld\n", #x, (long long)(x)) 377 #define PN(x) \ 378 SEQ_printf(m, " .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x)) 379 PN(sysctl_sched_latency); 380 PN(sysctl_sched_min_granularity); 381 PN(sysctl_sched_wakeup_granularity); 382 P(sysctl_sched_child_runs_first); 383 P(sysctl_sched_features); 384 #undef PN 385 #undef P 386 387 SEQ_printf(m, " .%-40s: %d (%s)\n", "sysctl_sched_tunable_scaling", 388 sysctl_sched_tunable_scaling, 389 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]); 390 391 for_each_online_cpu(cpu) 392 print_cpu(m, cpu); 393 394 SEQ_printf(m, "\n"); 395 396 return 0; 397 } 398 399 void sysrq_sched_debug_show(void) 400 { 401 sched_debug_show(NULL, NULL); 402 } 403 404 static int sched_debug_open(struct inode *inode, struct file *filp) 405 { 406 return single_open(filp, sched_debug_show, NULL); 407 } 408 409 static const struct file_operations sched_debug_fops = { 410 .open = sched_debug_open, 411 .read = seq_read, 412 .llseek = seq_lseek, 413 .release = single_release, 414 }; 415 416 static int __init init_sched_debug_procfs(void) 417 { 418 struct proc_dir_entry *pe; 419 420 pe = proc_create("sched_debug", 0444, NULL, &sched_debug_fops); 421 if (!pe) 422 return -ENOMEM; 423 return 0; 424 } 425 426 __initcall(init_sched_debug_procfs); 427 428 void proc_sched_show_task(struct task_struct *p, struct seq_file *m) 429 { 430 unsigned long nr_switches; 431 432 SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, p->pid, 433 get_nr_threads(p)); 434 SEQ_printf(m, 435 "---------------------------------------------------------\n"); 436 #define __P(F) \ 437 SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)F) 438 #define P(F) \ 439 SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)p->F) 440 #define __PN(F) \ 441 SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F)) 442 #define PN(F) \ 443 SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F)) 444 445 PN(se.exec_start); 446 PN(se.vruntime); 447 PN(se.sum_exec_runtime); 448 449 nr_switches = p->nvcsw + p->nivcsw; 450 451 #ifdef CONFIG_SCHEDSTATS 452 PN(se.statistics.wait_start); 453 PN(se.statistics.sleep_start); 454 PN(se.statistics.block_start); 455 PN(se.statistics.sleep_max); 456 PN(se.statistics.block_max); 457 PN(se.statistics.exec_max); 458 PN(se.statistics.slice_max); 459 PN(se.statistics.wait_max); 460 PN(se.statistics.wait_sum); 461 P(se.statistics.wait_count); 462 PN(se.statistics.iowait_sum); 463 P(se.statistics.iowait_count); 464 P(se.nr_migrations); 465 P(se.statistics.nr_migrations_cold); 466 P(se.statistics.nr_failed_migrations_affine); 467 P(se.statistics.nr_failed_migrations_running); 468 P(se.statistics.nr_failed_migrations_hot); 469 P(se.statistics.nr_forced_migrations); 470 P(se.statistics.nr_wakeups); 471 P(se.statistics.nr_wakeups_sync); 472 P(se.statistics.nr_wakeups_migrate); 473 P(se.statistics.nr_wakeups_local); 474 P(se.statistics.nr_wakeups_remote); 475 P(se.statistics.nr_wakeups_affine); 476 P(se.statistics.nr_wakeups_affine_attempts); 477 P(se.statistics.nr_wakeups_passive); 478 P(se.statistics.nr_wakeups_idle); 479 480 { 481 u64 avg_atom, avg_per_cpu; 482 483 avg_atom = p->se.sum_exec_runtime; 484 if (nr_switches) 485 do_div(avg_atom, nr_switches); 486 else 487 avg_atom = -1LL; 488 489 avg_per_cpu = p->se.sum_exec_runtime; 490 if (p->se.nr_migrations) { 491 avg_per_cpu = div64_u64(avg_per_cpu, 492 p->se.nr_migrations); 493 } else { 494 avg_per_cpu = -1LL; 495 } 496 497 __PN(avg_atom); 498 __PN(avg_per_cpu); 499 } 500 #endif 501 __P(nr_switches); 502 SEQ_printf(m, "%-35s:%21Ld\n", 503 "nr_voluntary_switches", (long long)p->nvcsw); 504 SEQ_printf(m, "%-35s:%21Ld\n", 505 "nr_involuntary_switches", (long long)p->nivcsw); 506 507 P(se.load.weight); 508 P(policy); 509 P(prio); 510 #undef PN 511 #undef __PN 512 #undef P 513 #undef __P 514 515 { 516 unsigned int this_cpu = raw_smp_processor_id(); 517 u64 t0, t1; 518 519 t0 = cpu_clock(this_cpu); 520 t1 = cpu_clock(this_cpu); 521 SEQ_printf(m, "%-35s:%21Ld\n", 522 "clock-delta", (long long)(t1-t0)); 523 } 524 } 525 526 void proc_sched_set_task(struct task_struct *p) 527 { 528 #ifdef CONFIG_SCHEDSTATS 529 memset(&p->se.statistics, 0, sizeof(p->se.statistics)); 530 #endif 531 } 532