| 1 | /* |
| 2 | * kernel/time/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 | /* |
| 20 | * This allows printing both to /proc/sched_debug and |
| 21 | * to the console |
| 22 | */ |
| 23 | #define SEQ_printf(m, x...) \ |
| 24 | do { \ |
| 25 | if (m) \ |
| 26 | seq_printf(m, x); \ |
| 27 | else \ |
| 28 | printk(x); \ |
| 29 | } while (0) |
| 30 | |
| 31 | /* |
| 32 | * Ease the printing of nsec fields: |
| 33 | */ |
| 34 | static long long nsec_high(unsigned long long nsec) |
| 35 | { |
| 36 | if ((long long)nsec < 0) { |
| 37 | nsec = -nsec; |
| 38 | do_div(nsec, 1000000); |
| 39 | return -nsec; |
| 40 | } |
| 41 | do_div(nsec, 1000000); |
| 42 | |
| 43 | return nsec; |
| 44 | } |
| 45 | |
| 46 | static unsigned long nsec_low(unsigned long long nsec) |
| 47 | { |
| 48 | if ((long long)nsec < 0) |
| 49 | nsec = -nsec; |
| 50 | |
| 51 | return do_div(nsec, 1000000); |
| 52 | } |
| 53 | |
| 54 | #define SPLIT_NS(x) nsec_high(x), nsec_low(x) |
| 55 | |
| 56 | #ifdef CONFIG_FAIR_GROUP_SCHED |
| 57 | static void print_cfs_group_stats(struct seq_file *m, int cpu, |
| 58 | struct task_group *tg) |
| 59 | { |
| 60 | struct sched_entity *se = tg->se[cpu]; |
| 61 | if (!se) |
| 62 | return; |
| 63 | |
| 64 | #define P(F) \ |
| 65 | SEQ_printf(m, " .%-30s: %lld\n", #F, (long long)F) |
| 66 | #define PN(F) \ |
| 67 | SEQ_printf(m, " .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F)) |
| 68 | |
| 69 | PN(se->exec_start); |
| 70 | PN(se->vruntime); |
| 71 | PN(se->sum_exec_runtime); |
| 72 | #ifdef CONFIG_SCHEDSTATS |
| 73 | PN(se->wait_start); |
| 74 | PN(se->sleep_start); |
| 75 | PN(se->block_start); |
| 76 | PN(se->sleep_max); |
| 77 | PN(se->block_max); |
| 78 | PN(se->exec_max); |
| 79 | PN(se->slice_max); |
| 80 | PN(se->wait_max); |
| 81 | PN(se->wait_sum); |
| 82 | P(se->wait_count); |
| 83 | #endif |
| 84 | P(se->load.weight); |
| 85 | #undef PN |
| 86 | #undef P |
| 87 | } |
| 88 | #endif |
| 89 | |
| 90 | static void |
| 91 | print_task(struct seq_file *m, struct rq *rq, struct task_struct *p) |
| 92 | { |
| 93 | if (rq->curr == p) |
| 94 | SEQ_printf(m, "R"); |
| 95 | else |
| 96 | SEQ_printf(m, " "); |
| 97 | |
| 98 | SEQ_printf(m, "%15s %5d %9Ld.%06ld %9Ld %5d ", |
| 99 | p->comm, p->pid, |
| 100 | SPLIT_NS(p->se.vruntime), |
| 101 | (long long)(p->nvcsw + p->nivcsw), |
| 102 | p->prio); |
| 103 | #ifdef CONFIG_SCHEDSTATS |
| 104 | SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld", |
| 105 | SPLIT_NS(p->se.vruntime), |
| 106 | SPLIT_NS(p->se.sum_exec_runtime), |
| 107 | SPLIT_NS(p->se.sum_sleep_runtime)); |
| 108 | #else |
| 109 | SEQ_printf(m, "%15Ld %15Ld %15Ld.%06ld %15Ld.%06ld %15Ld.%06ld", |
| 110 | 0LL, 0LL, 0LL, 0L, 0LL, 0L, 0LL, 0L); |
| 111 | #endif |
| 112 | |
| 113 | #ifdef CONFIG_CGROUP_SCHED |
| 114 | { |
| 115 | char path[64]; |
| 116 | |
| 117 | cgroup_path(task_group(p)->css.cgroup, path, sizeof(path)); |
| 118 | SEQ_printf(m, " %s", path); |
| 119 | } |
| 120 | #endif |
| 121 | SEQ_printf(m, "\n"); |
| 122 | } |
| 123 | |
| 124 | static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu) |
| 125 | { |
| 126 | struct task_struct *g, *p; |
| 127 | unsigned long flags; |
| 128 | |
| 129 | SEQ_printf(m, |
| 130 | "\nrunnable tasks:\n" |
| 131 | " task PID tree-key switches prio" |
| 132 | " exec-runtime sum-exec sum-sleep\n" |
| 133 | "------------------------------------------------------" |
| 134 | "----------------------------------------------------\n"); |
| 135 | |
| 136 | read_lock_irqsave(&tasklist_lock, flags); |
| 137 | |
| 138 | do_each_thread(g, p) { |
| 139 | if (!p->se.on_rq || task_cpu(p) != rq_cpu) |
| 140 | continue; |
| 141 | |
| 142 | print_task(m, rq, p); |
| 143 | } while_each_thread(g, p); |
| 144 | |
| 145 | read_unlock_irqrestore(&tasklist_lock, flags); |
| 146 | } |
| 147 | |
| 148 | #if defined(CONFIG_CGROUP_SCHED) && \ |
| 149 | (defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)) |
| 150 | static void task_group_path(struct task_group *tg, char *buf, int buflen) |
| 151 | { |
| 152 | /* may be NULL if the underlying cgroup isn't fully-created yet */ |
| 153 | if (!tg->css.cgroup) { |
| 154 | buf[0] = '\0'; |
| 155 | return; |
| 156 | } |
| 157 | cgroup_path(tg->css.cgroup, buf, buflen); |
| 158 | } |
| 159 | #endif |
| 160 | |
| 161 | void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq) |
| 162 | { |
| 163 | s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1, |
| 164 | spread, rq0_min_vruntime, spread0; |
| 165 | struct rq *rq = cpu_rq(cpu); |
| 166 | struct sched_entity *last; |
| 167 | unsigned long flags; |
| 168 | |
| 169 | #if defined(CONFIG_CGROUP_SCHED) && defined(CONFIG_FAIR_GROUP_SCHED) |
| 170 | char path[128]; |
| 171 | struct task_group *tg = cfs_rq->tg; |
| 172 | |
| 173 | task_group_path(tg, path, sizeof(path)); |
| 174 | |
| 175 | SEQ_printf(m, "\ncfs_rq[%d]:%s\n", cpu, path); |
| 176 | #elif defined(CONFIG_USER_SCHED) && defined(CONFIG_FAIR_GROUP_SCHED) |
| 177 | { |
| 178 | uid_t uid = cfs_rq->tg->uid; |
| 179 | SEQ_printf(m, "\ncfs_rq[%d] for UID: %u\n", cpu, uid); |
| 180 | } |
| 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 | spin_lock_irqsave(&rq->lock, flags); |
| 188 | if (cfs_rq->rb_leftmost) |
| 189 | MIN_vruntime = (__pick_next_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 | 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: %ld\n", "nr_running", cfs_rq->nr_running); |
| 209 | SEQ_printf(m, " .%-30s: %ld\n", "load", cfs_rq->load.weight); |
| 210 | |
| 211 | SEQ_printf(m, " .%-30s: %d\n", "nr_spread_over", |
| 212 | cfs_rq->nr_spread_over); |
| 213 | #ifdef CONFIG_FAIR_GROUP_SCHED |
| 214 | #ifdef CONFIG_SMP |
| 215 | SEQ_printf(m, " .%-30s: %lu\n", "shares", cfs_rq->shares); |
| 216 | #endif |
| 217 | print_cfs_group_stats(m, cpu, cfs_rq->tg); |
| 218 | #endif |
| 219 | } |
| 220 | |
| 221 | void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq) |
| 222 | { |
| 223 | #if defined(CONFIG_CGROUP_SCHED) && defined(CONFIG_RT_GROUP_SCHED) |
| 224 | char path[128]; |
| 225 | struct task_group *tg = rt_rq->tg; |
| 226 | |
| 227 | task_group_path(tg, path, sizeof(path)); |
| 228 | |
| 229 | SEQ_printf(m, "\nrt_rq[%d]:%s\n", cpu, path); |
| 230 | #else |
| 231 | SEQ_printf(m, "\nrt_rq[%d]:\n", cpu); |
| 232 | #endif |
| 233 | |
| 234 | |
| 235 | #define P(x) \ |
| 236 | SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rt_rq->x)) |
| 237 | #define PN(x) \ |
| 238 | SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x)) |
| 239 | |
| 240 | P(rt_nr_running); |
| 241 | P(rt_throttled); |
| 242 | PN(rt_time); |
| 243 | PN(rt_runtime); |
| 244 | |
| 245 | #undef PN |
| 246 | #undef P |
| 247 | } |
| 248 | |
| 249 | static void print_cpu(struct seq_file *m, int cpu) |
| 250 | { |
| 251 | struct rq *rq = cpu_rq(cpu); |
| 252 | |
| 253 | #ifdef CONFIG_X86 |
| 254 | { |
| 255 | unsigned int freq = cpu_khz ? : 1; |
| 256 | |
| 257 | SEQ_printf(m, "\ncpu#%d, %u.%03u MHz\n", |
| 258 | cpu, freq / 1000, (freq % 1000)); |
| 259 | } |
| 260 | #else |
| 261 | SEQ_printf(m, "\ncpu#%d\n", cpu); |
| 262 | #endif |
| 263 | |
| 264 | #define P(x) \ |
| 265 | SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rq->x)) |
| 266 | #define PN(x) \ |
| 267 | SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x)) |
| 268 | |
| 269 | P(nr_running); |
| 270 | SEQ_printf(m, " .%-30s: %lu\n", "load", |
| 271 | rq->load.weight); |
| 272 | P(nr_switches); |
| 273 | P(nr_load_updates); |
| 274 | P(nr_uninterruptible); |
| 275 | PN(next_balance); |
| 276 | P(curr->pid); |
| 277 | PN(clock); |
| 278 | P(cpu_load[0]); |
| 279 | P(cpu_load[1]); |
| 280 | P(cpu_load[2]); |
| 281 | P(cpu_load[3]); |
| 282 | P(cpu_load[4]); |
| 283 | #undef P |
| 284 | #undef PN |
| 285 | |
| 286 | #ifdef CONFIG_SCHEDSTATS |
| 287 | #define P(n) SEQ_printf(m, " .%-30s: %d\n", #n, rq->n); |
| 288 | |
| 289 | P(yld_count); |
| 290 | |
| 291 | P(sched_switch); |
| 292 | P(sched_count); |
| 293 | P(sched_goidle); |
| 294 | |
| 295 | P(ttwu_count); |
| 296 | P(ttwu_local); |
| 297 | |
| 298 | P(bkl_count); |
| 299 | |
| 300 | #undef P |
| 301 | #endif |
| 302 | print_cfs_stats(m, cpu); |
| 303 | print_rt_stats(m, cpu); |
| 304 | |
| 305 | print_rq(m, rq, cpu); |
| 306 | } |
| 307 | |
| 308 | static int sched_debug_show(struct seq_file *m, void *v) |
| 309 | { |
| 310 | u64 now = ktime_to_ns(ktime_get()); |
| 311 | int cpu; |
| 312 | |
| 313 | SEQ_printf(m, "Sched Debug Version: v0.09, %s %.*s\n", |
| 314 | init_utsname()->release, |
| 315 | (int)strcspn(init_utsname()->version, " "), |
| 316 | init_utsname()->version); |
| 317 | |
| 318 | SEQ_printf(m, "now at %Lu.%06ld msecs\n", SPLIT_NS(now)); |
| 319 | |
| 320 | #define P(x) \ |
| 321 | SEQ_printf(m, " .%-40s: %Ld\n", #x, (long long)(x)) |
| 322 | #define PN(x) \ |
| 323 | SEQ_printf(m, " .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x)) |
| 324 | P(jiffies); |
| 325 | PN(sysctl_sched_latency); |
| 326 | PN(sysctl_sched_min_granularity); |
| 327 | PN(sysctl_sched_wakeup_granularity); |
| 328 | PN(sysctl_sched_child_runs_first); |
| 329 | P(sysctl_sched_features); |
| 330 | #undef PN |
| 331 | #undef P |
| 332 | |
| 333 | for_each_online_cpu(cpu) |
| 334 | print_cpu(m, cpu); |
| 335 | |
| 336 | SEQ_printf(m, "\n"); |
| 337 | |
| 338 | return 0; |
| 339 | } |
| 340 | |
| 341 | static void sysrq_sched_debug_show(void) |
| 342 | { |
| 343 | sched_debug_show(NULL, NULL); |
| 344 | } |
| 345 | |
| 346 | static int sched_debug_open(struct inode *inode, struct file *filp) |
| 347 | { |
| 348 | return single_open(filp, sched_debug_show, NULL); |
| 349 | } |
| 350 | |
| 351 | static const struct file_operations sched_debug_fops = { |
| 352 | .open = sched_debug_open, |
| 353 | .read = seq_read, |
| 354 | .llseek = seq_lseek, |
| 355 | .release = single_release, |
| 356 | }; |
| 357 | |
| 358 | static int __init init_sched_debug_procfs(void) |
| 359 | { |
| 360 | struct proc_dir_entry *pe; |
| 361 | |
| 362 | pe = proc_create("sched_debug", 0444, NULL, &sched_debug_fops); |
| 363 | if (!pe) |
| 364 | return -ENOMEM; |
| 365 | return 0; |
| 366 | } |
| 367 | |
| 368 | __initcall(init_sched_debug_procfs); |
| 369 | |
| 370 | void proc_sched_show_task(struct task_struct *p, struct seq_file *m) |
| 371 | { |
| 372 | unsigned long nr_switches; |
| 373 | unsigned long flags; |
| 374 | int num_threads = 1; |
| 375 | |
| 376 | if (lock_task_sighand(p, &flags)) { |
| 377 | num_threads = atomic_read(&p->signal->count); |
| 378 | unlock_task_sighand(p, &flags); |
| 379 | } |
| 380 | |
| 381 | SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, p->pid, num_threads); |
| 382 | SEQ_printf(m, |
| 383 | "---------------------------------------------------------\n"); |
| 384 | #define __P(F) \ |
| 385 | SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)F) |
| 386 | #define P(F) \ |
| 387 | SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)p->F) |
| 388 | #define __PN(F) \ |
| 389 | SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F)) |
| 390 | #define PN(F) \ |
| 391 | SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F)) |
| 392 | |
| 393 | PN(se.exec_start); |
| 394 | PN(se.vruntime); |
| 395 | PN(se.sum_exec_runtime); |
| 396 | PN(se.avg_overlap); |
| 397 | PN(se.avg_wakeup); |
| 398 | PN(se.avg_running); |
| 399 | |
| 400 | nr_switches = p->nvcsw + p->nivcsw; |
| 401 | |
| 402 | #ifdef CONFIG_SCHEDSTATS |
| 403 | PN(se.wait_start); |
| 404 | PN(se.sleep_start); |
| 405 | PN(se.block_start); |
| 406 | PN(se.sleep_max); |
| 407 | PN(se.block_max); |
| 408 | PN(se.exec_max); |
| 409 | PN(se.slice_max); |
| 410 | PN(se.wait_max); |
| 411 | PN(se.wait_sum); |
| 412 | P(se.wait_count); |
| 413 | PN(se.iowait_sum); |
| 414 | P(se.iowait_count); |
| 415 | P(sched_info.bkl_count); |
| 416 | P(se.nr_migrations); |
| 417 | P(se.nr_migrations_cold); |
| 418 | P(se.nr_failed_migrations_affine); |
| 419 | P(se.nr_failed_migrations_running); |
| 420 | P(se.nr_failed_migrations_hot); |
| 421 | P(se.nr_forced_migrations); |
| 422 | P(se.nr_forced2_migrations); |
| 423 | P(se.nr_wakeups); |
| 424 | P(se.nr_wakeups_sync); |
| 425 | P(se.nr_wakeups_migrate); |
| 426 | P(se.nr_wakeups_local); |
| 427 | P(se.nr_wakeups_remote); |
| 428 | P(se.nr_wakeups_affine); |
| 429 | P(se.nr_wakeups_affine_attempts); |
| 430 | P(se.nr_wakeups_passive); |
| 431 | P(se.nr_wakeups_idle); |
| 432 | |
| 433 | { |
| 434 | u64 avg_atom, avg_per_cpu; |
| 435 | |
| 436 | avg_atom = p->se.sum_exec_runtime; |
| 437 | if (nr_switches) |
| 438 | do_div(avg_atom, nr_switches); |
| 439 | else |
| 440 | avg_atom = -1LL; |
| 441 | |
| 442 | avg_per_cpu = p->se.sum_exec_runtime; |
| 443 | if (p->se.nr_migrations) { |
| 444 | avg_per_cpu = div64_u64(avg_per_cpu, |
| 445 | p->se.nr_migrations); |
| 446 | } else { |
| 447 | avg_per_cpu = -1LL; |
| 448 | } |
| 449 | |
| 450 | __PN(avg_atom); |
| 451 | __PN(avg_per_cpu); |
| 452 | } |
| 453 | #endif |
| 454 | __P(nr_switches); |
| 455 | SEQ_printf(m, "%-35s:%21Ld\n", |
| 456 | "nr_voluntary_switches", (long long)p->nvcsw); |
| 457 | SEQ_printf(m, "%-35s:%21Ld\n", |
| 458 | "nr_involuntary_switches", (long long)p->nivcsw); |
| 459 | |
| 460 | P(se.load.weight); |
| 461 | P(policy); |
| 462 | P(prio); |
| 463 | #undef PN |
| 464 | #undef __PN |
| 465 | #undef P |
| 466 | #undef __P |
| 467 | |
| 468 | { |
| 469 | unsigned int this_cpu = raw_smp_processor_id(); |
| 470 | u64 t0, t1; |
| 471 | |
| 472 | t0 = cpu_clock(this_cpu); |
| 473 | t1 = cpu_clock(this_cpu); |
| 474 | SEQ_printf(m, "%-35s:%21Ld\n", |
| 475 | "clock-delta", (long long)(t1-t0)); |
| 476 | } |
| 477 | } |
| 478 | |
| 479 | void proc_sched_set_task(struct task_struct *p) |
| 480 | { |
| 481 | #ifdef CONFIG_SCHEDSTATS |
| 482 | p->se.wait_max = 0; |
| 483 | p->se.wait_sum = 0; |
| 484 | p->se.wait_count = 0; |
| 485 | p->se.iowait_sum = 0; |
| 486 | p->se.iowait_count = 0; |
| 487 | p->se.sleep_max = 0; |
| 488 | p->se.sum_sleep_runtime = 0; |
| 489 | p->se.block_max = 0; |
| 490 | p->se.exec_max = 0; |
| 491 | p->se.slice_max = 0; |
| 492 | p->se.nr_migrations = 0; |
| 493 | p->se.nr_migrations_cold = 0; |
| 494 | p->se.nr_failed_migrations_affine = 0; |
| 495 | p->se.nr_failed_migrations_running = 0; |
| 496 | p->se.nr_failed_migrations_hot = 0; |
| 497 | p->se.nr_forced_migrations = 0; |
| 498 | p->se.nr_forced2_migrations = 0; |
| 499 | p->se.nr_wakeups = 0; |
| 500 | p->se.nr_wakeups_sync = 0; |
| 501 | p->se.nr_wakeups_migrate = 0; |
| 502 | p->se.nr_wakeups_local = 0; |
| 503 | p->se.nr_wakeups_remote = 0; |
| 504 | p->se.nr_wakeups_affine = 0; |
| 505 | p->se.nr_wakeups_affine_attempts = 0; |
| 506 | p->se.nr_wakeups_passive = 0; |
| 507 | p->se.nr_wakeups_idle = 0; |
| 508 | p->sched_info.bkl_count = 0; |
| 509 | #endif |
| 510 | p->se.sum_exec_runtime = 0; |
| 511 | p->se.prev_sum_exec_runtime = 0; |
| 512 | p->nvcsw = 0; |
| 513 | p->nivcsw = 0; |
| 514 | } |