projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
sched/debug: Make schedstats a runtime tunable that is disabled by default
[deliverable/linux.git] / kernel / sched / debug.c
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 if (schedstat_enabled()) {
79 PN(se->statistics.wait_start);
80 PN(se->statistics.sleep_start);
81 PN(se->statistics.block_start);
82 PN(se->statistics.sleep_max);
83 PN(se->statistics.block_max);
84 PN(se->statistics.exec_max);
85 PN(se->statistics.slice_max);
86 PN(se->statistics.wait_max);
87 PN(se->statistics.wait_sum);
88 P(se->statistics.wait_count);
89 }
90 #endif
91 P(se->load.weight);
92 #ifdef CONFIG_SMP
93 P(se->avg.load_avg);
94 P(se->avg.util_avg);
95 #endif
96 #undef PN
97 #undef P
98 }
99 #endif
100
101 #ifdef CONFIG_CGROUP_SCHED
102 static char group_path[PATH_MAX];
103
104 static char *task_group_path(struct task_group *tg)
105 {
106 if (autogroup_path(tg, group_path, PATH_MAX))
107 return group_path;
108
109 return cgroup_path(tg->css.cgroup, group_path, PATH_MAX);
110 }
111 #endif
112
113 static void
114 print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
115 {
116 if (rq->curr == p)
117 SEQ_printf(m, "R");
118 else
119 SEQ_printf(m, " ");
120
121 SEQ_printf(m, "%15s %5d %9Ld.%06ld %9Ld %5d ",
122 p->comm, task_pid_nr(p),
123 SPLIT_NS(p->se.vruntime),
124 (long long)(p->nvcsw + p->nivcsw),
125 p->prio);
126 #ifdef CONFIG_SCHEDSTATS
127 if (schedstat_enabled()) {
128 SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
129 SPLIT_NS(p->se.statistics.wait_sum),
130 SPLIT_NS(p->se.sum_exec_runtime),
131 SPLIT_NS(p->se.statistics.sum_sleep_runtime));
132 }
133 #else
134 SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
135 0LL, 0L,
136 SPLIT_NS(p->se.sum_exec_runtime),
137 0LL, 0L);
138 #endif
139 #ifdef CONFIG_NUMA_BALANCING
140 SEQ_printf(m, " %d %d", task_node(p), task_numa_group_id(p));
141 #endif
142 #ifdef CONFIG_CGROUP_SCHED
143 SEQ_printf(m, " %s", task_group_path(task_group(p)));
144 #endif
145
146 SEQ_printf(m, "\n");
147 }
148
149 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
150 {
151 struct task_struct *g, *p;
152
153 SEQ_printf(m,
154 "\nrunnable tasks:\n"
155 " task PID tree-key switches prio"
156 " wait-time sum-exec sum-sleep\n"
157 "------------------------------------------------------"
158 "----------------------------------------------------\n");
159
160 rcu_read_lock();
161 for_each_process_thread(g, p) {
162 if (task_cpu(p) != rq_cpu)
163 continue;
164
165 print_task(m, rq, p);
166 }
167 rcu_read_unlock();
168 }
169
170 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
171 {
172 s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
173 spread, rq0_min_vruntime, spread0;
174 struct rq *rq = cpu_rq(cpu);
175 struct sched_entity *last;
176 unsigned long flags;
177
178 #ifdef CONFIG_FAIR_GROUP_SCHED
179 SEQ_printf(m, "\ncfs_rq[%d]:%s\n", cpu, task_group_path(cfs_rq->tg));
180 #else
181 SEQ_printf(m, "\ncfs_rq[%d]:\n", cpu);
182 #endif
183 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "exec_clock",
184 SPLIT_NS(cfs_rq->exec_clock));
185
186 raw_spin_lock_irqsave(&rq->lock, flags);
187 if (cfs_rq->rb_leftmost)
188 MIN_vruntime = (__pick_first_entity(cfs_rq))->vruntime;
189 last = __pick_last_entity(cfs_rq);
190 if (last)
191 max_vruntime = last->vruntime;
192 min_vruntime = cfs_rq->min_vruntime;
193 rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
194 raw_spin_unlock_irqrestore(&rq->lock, flags);
195 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "MIN_vruntime",
196 SPLIT_NS(MIN_vruntime));
197 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "min_vruntime",
198 SPLIT_NS(min_vruntime));
199 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "max_vruntime",
200 SPLIT_NS(max_vruntime));
201 spread = max_vruntime - MIN_vruntime;
202 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread",
203 SPLIT_NS(spread));
204 spread0 = min_vruntime - rq0_min_vruntime;
205 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread0",
206 SPLIT_NS(spread0));
207 SEQ_printf(m, " .%-30s: %d\n", "nr_spread_over",
208 cfs_rq->nr_spread_over);
209 SEQ_printf(m, " .%-30s: %d\n", "nr_running", cfs_rq->nr_running);
210 SEQ_printf(m, " .%-30s: %ld\n", "load", cfs_rq->load.weight);
211 #ifdef CONFIG_SMP
212 SEQ_printf(m, " .%-30s: %lu\n", "load_avg",
213 cfs_rq->avg.load_avg);
214 SEQ_printf(m, " .%-30s: %lu\n", "runnable_load_avg",
215 cfs_rq->runnable_load_avg);
216 SEQ_printf(m, " .%-30s: %lu\n", "util_avg",
217 cfs_rq->avg.util_avg);
218 SEQ_printf(m, " .%-30s: %ld\n", "removed_load_avg",
219 atomic_long_read(&cfs_rq->removed_load_avg));
220 SEQ_printf(m, " .%-30s: %ld\n", "removed_util_avg",
221 atomic_long_read(&cfs_rq->removed_util_avg));
222 #ifdef CONFIG_FAIR_GROUP_SCHED
223 SEQ_printf(m, " .%-30s: %lu\n", "tg_load_avg_contrib",
224 cfs_rq->tg_load_avg_contrib);
225 SEQ_printf(m, " .%-30s: %ld\n", "tg_load_avg",
226 atomic_long_read(&cfs_rq->tg->load_avg));
227 #endif
228 #endif
229 #ifdef CONFIG_CFS_BANDWIDTH
230 SEQ_printf(m, " .%-30s: %d\n", "throttled",
231 cfs_rq->throttled);
232 SEQ_printf(m, " .%-30s: %d\n", "throttle_count",
233 cfs_rq->throttle_count);
234 #endif
235
236 #ifdef CONFIG_FAIR_GROUP_SCHED
237 print_cfs_group_stats(m, cpu, cfs_rq->tg);
238 #endif
239 }
240
241 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
242 {
243 #ifdef CONFIG_RT_GROUP_SCHED
244 SEQ_printf(m, "\nrt_rq[%d]:%s\n", cpu, task_group_path(rt_rq->tg));
245 #else
246 SEQ_printf(m, "\nrt_rq[%d]:\n", cpu);
247 #endif
248
249 #define P(x) \
250 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
251 #define PN(x) \
252 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
253
254 P(rt_nr_running);
255 P(rt_throttled);
256 PN(rt_time);
257 PN(rt_runtime);
258
259 #undef PN
260 #undef P
261 }
262
263 void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq)
264 {
265 SEQ_printf(m, "\ndl_rq[%d]:\n", cpu);
266 SEQ_printf(m, " .%-30s: %ld\n", "dl_nr_running", dl_rq->dl_nr_running);
267 }
268
269 extern __read_mostly int sched_clock_running;
270
271 static void print_cpu(struct seq_file *m, int cpu)
272 {
273 struct rq *rq = cpu_rq(cpu);
274 unsigned long flags;
275
276 #ifdef CONFIG_X86
277 {
278 unsigned int freq = cpu_khz ? : 1;
279
280 SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
281 cpu, freq / 1000, (freq % 1000));
282 }
283 #else
284 SEQ_printf(m, "cpu#%d\n", cpu);
285 #endif
286
287 #define P(x) \
288 do { \
289 if (sizeof(rq->x) == 4) \
290 SEQ_printf(m, " .%-30s: %ld\n", #x, (long)(rq->x)); \
291 else \
292 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rq->x));\
293 } while (0)
294
295 #define PN(x) \
296 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
297
298 P(nr_running);
299 SEQ_printf(m, " .%-30s: %lu\n", "load",
300 rq->load.weight);
301 P(nr_switches);
302 P(nr_load_updates);
303 P(nr_uninterruptible);
304 PN(next_balance);
305 SEQ_printf(m, " .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr)));
306 PN(clock);
307 PN(clock_task);
308 P(cpu_load[0]);
309 P(cpu_load[1]);
310 P(cpu_load[2]);
311 P(cpu_load[3]);
312 P(cpu_load[4]);
313 #undef P
314 #undef PN
315
316 #ifdef CONFIG_SCHEDSTATS
317 #define P(n) SEQ_printf(m, " .%-30s: %d\n", #n, rq->n);
318 #define P64(n) SEQ_printf(m, " .%-30s: %Ld\n", #n, rq->n);
319
320 #ifdef CONFIG_SMP
321 P64(avg_idle);
322 P64(max_idle_balance_cost);
323 #endif
324
325 if (schedstat_enabled()) {
326 P(yld_count);
327 P(sched_count);
328 P(sched_goidle);
329 P(ttwu_count);
330 P(ttwu_local);
331 }
332
333 #undef P
334 #undef P64
335 #endif
336 spin_lock_irqsave(&sched_debug_lock, flags);
337 print_cfs_stats(m, cpu);
338 print_rt_stats(m, cpu);
339 print_dl_stats(m, cpu);
340
341 print_rq(m, rq, cpu);
342 spin_unlock_irqrestore(&sched_debug_lock, flags);
343 SEQ_printf(m, "\n");
344 }
345
346 static const char *sched_tunable_scaling_names[] = {
347 "none",
348 "logaritmic",
349 "linear"
350 };
351
352 static void sched_debug_header(struct seq_file *m)
353 {
354 u64 ktime, sched_clk, cpu_clk;
355 unsigned long flags;
356
357 local_irq_save(flags);
358 ktime = ktime_to_ns(ktime_get());
359 sched_clk = sched_clock();
360 cpu_clk = local_clock();
361 local_irq_restore(flags);
362
363 SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n",
364 init_utsname()->release,
365 (int)strcspn(init_utsname()->version, " "),
366 init_utsname()->version);
367
368 #define P(x) \
369 SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
370 #define PN(x) \
371 SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
372 PN(ktime);
373 PN(sched_clk);
374 PN(cpu_clk);
375 P(jiffies);
376 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
377 P(sched_clock_stable());
378 #endif
379 #undef PN
380 #undef P
381
382 SEQ_printf(m, "\n");
383 SEQ_printf(m, "sysctl_sched\n");
384
385 #define P(x) \
386 SEQ_printf(m, " .%-40s: %Ld\n", #x, (long long)(x))
387 #define PN(x) \
388 SEQ_printf(m, " .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
389 PN(sysctl_sched_latency);
390 PN(sysctl_sched_min_granularity);
391 PN(sysctl_sched_wakeup_granularity);
392 P(sysctl_sched_child_runs_first);
393 P(sysctl_sched_features);
394 #undef PN
395 #undef P
396
397 SEQ_printf(m, " .%-40s: %d (%s)\n",
398 "sysctl_sched_tunable_scaling",
399 sysctl_sched_tunable_scaling,
400 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
401 SEQ_printf(m, "\n");
402 }
403
404 static int sched_debug_show(struct seq_file *m, void *v)
405 {
406 int cpu = (unsigned long)(v - 2);
407
408 if (cpu != -1)
409 print_cpu(m, cpu);
410 else
411 sched_debug_header(m);
412
413 return 0;
414 }
415
416 void sysrq_sched_debug_show(void)
417 {
418 int cpu;
419
420 sched_debug_header(NULL);
421 for_each_online_cpu(cpu)
422 print_cpu(NULL, cpu);
423
424 }
425
426 /*
427 * This itererator needs some explanation.
428 * It returns 1 for the header position.
429 * This means 2 is cpu 0.
430 * In a hotplugged system some cpus, including cpu 0, may be missing so we have
431 * to use cpumask_* to iterate over the cpus.
432 */
433 static void *sched_debug_start(struct seq_file *file, loff_t *offset)
434 {
435 unsigned long n = *offset;
436
437 if (n == 0)
438 return (void *) 1;
439
440 n--;
441
442 if (n > 0)
443 n = cpumask_next(n - 1, cpu_online_mask);
444 else
445 n = cpumask_first(cpu_online_mask);
446
447 *offset = n + 1;
448
449 if (n < nr_cpu_ids)
450 return (void *)(unsigned long)(n + 2);
451 return NULL;
452 }
453
454 static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
455 {
456 (*offset)++;
457 return sched_debug_start(file, offset);
458 }
459
460 static void sched_debug_stop(struct seq_file *file, void *data)
461 {
462 }
463
464 static const struct seq_operations sched_debug_sops = {
465 .start = sched_debug_start,
466 .next = sched_debug_next,
467 .stop = sched_debug_stop,
468 .show = sched_debug_show,
469 };
470
471 static int sched_debug_release(struct inode *inode, struct file *file)
472 {
473 seq_release(inode, file);
474
475 return 0;
476 }
477
478 static int sched_debug_open(struct inode *inode, struct file *filp)
479 {
480 int ret = 0;
481
482 ret = seq_open(filp, &sched_debug_sops);
483
484 return ret;
485 }
486
487 static const struct file_operations sched_debug_fops = {
488 .open = sched_debug_open,
489 .read = seq_read,
490 .llseek = seq_lseek,
491 .release = sched_debug_release,
492 };
493
494 static int __init init_sched_debug_procfs(void)
495 {
496 struct proc_dir_entry *pe;
497
498 pe = proc_create("sched_debug", 0444, NULL, &sched_debug_fops);
499 if (!pe)
500 return -ENOMEM;
501 return 0;
502 }
503
504 __initcall(init_sched_debug_procfs);
505
506 #define __P(F) \
507 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)F)
508 #define P(F) \
509 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)p->F)
510 #define __PN(F) \
511 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
512 #define PN(F) \
513 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
514
515
516 #ifdef CONFIG_NUMA_BALANCING
517 void print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
518 unsigned long tpf, unsigned long gsf, unsigned long gpf)
519 {
520 SEQ_printf(m, "numa_faults node=%d ", node);
521 SEQ_printf(m, "task_private=%lu task_shared=%lu ", tsf, tpf);
522 SEQ_printf(m, "group_private=%lu group_shared=%lu\n", gsf, gpf);
523 }
524 #endif
525
526
527 static void sched_show_numa(struct task_struct *p, struct seq_file *m)
528 {
529 #ifdef CONFIG_NUMA_BALANCING
530 struct mempolicy *pol;
531
532 if (p->mm)
533 P(mm->numa_scan_seq);
534
535 task_lock(p);
536 pol = p->mempolicy;
537 if (pol && !(pol->flags & MPOL_F_MORON))
538 pol = NULL;
539 mpol_get(pol);
540 task_unlock(p);
541
542 P(numa_pages_migrated);
543 P(numa_preferred_nid);
544 P(total_numa_faults);
545 SEQ_printf(m, "current_node=%d, numa_group_id=%d\n",
546 task_node(p), task_numa_group_id(p));
547 show_numa_stats(p, m);
548 mpol_put(pol);
549 #endif
550 }
551
552 void proc_sched_show_task(struct task_struct *p, struct seq_file *m)
553 {
554 unsigned long nr_switches;
555
556 SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr(p),
557 get_nr_threads(p));
558 SEQ_printf(m,
559 "---------------------------------------------------------"
560 "----------\n");
561 #define __P(F) \
562 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)F)
563 #define P(F) \
564 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)p->F)
565 #define __PN(F) \
566 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
567 #define PN(F) \
568 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
569
570 PN(se.exec_start);
571 PN(se.vruntime);
572 PN(se.sum_exec_runtime);
573
574 nr_switches = p->nvcsw + p->nivcsw;
575
576 #ifdef CONFIG_SCHEDSTATS
577 P(se.nr_migrations);
578
579 if (schedstat_enabled()) {
580 u64 avg_atom, avg_per_cpu;
581
582 PN(se.statistics.sum_sleep_runtime);
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.statistics.nr_migrations_cold);
596 P(se.statistics.nr_failed_migrations_affine);
597 P(se.statistics.nr_failed_migrations_running);
598 P(se.statistics.nr_failed_migrations_hot);
599 P(se.statistics.nr_forced_migrations);
600 P(se.statistics.nr_wakeups);
601 P(se.statistics.nr_wakeups_sync);
602 P(se.statistics.nr_wakeups_migrate);
603 P(se.statistics.nr_wakeups_local);
604 P(se.statistics.nr_wakeups_remote);
605 P(se.statistics.nr_wakeups_affine);
606 P(se.statistics.nr_wakeups_affine_attempts);
607 P(se.statistics.nr_wakeups_passive);
608 P(se.statistics.nr_wakeups_idle);
609
610 avg_atom = p->se.sum_exec_runtime;
611 if (nr_switches)
612 avg_atom = div64_ul(avg_atom, nr_switches);
613 else
614 avg_atom = -1LL;
615
616 avg_per_cpu = p->se.sum_exec_runtime;
617 if (p->se.nr_migrations) {
618 avg_per_cpu = div64_u64(avg_per_cpu,
619 p->se.nr_migrations);
620 } else {
621 avg_per_cpu = -1LL;
622 }
623
624 __PN(avg_atom);
625 __PN(avg_per_cpu);
626 }
627 #endif
628 __P(nr_switches);
629 SEQ_printf(m, "%-45s:%21Ld\n",
630 "nr_voluntary_switches", (long long)p->nvcsw);
631 SEQ_printf(m, "%-45s:%21Ld\n",
632 "nr_involuntary_switches", (long long)p->nivcsw);
633
634 P(se.load.weight);
635 #ifdef CONFIG_SMP
636 P(se.avg.load_sum);
637 P(se.avg.util_sum);
638 P(se.avg.load_avg);
639 P(se.avg.util_avg);
640 P(se.avg.last_update_time);
641 #endif
642 P(policy);
643 P(prio);
644 #undef PN
645 #undef __PN
646 #undef P
647 #undef __P
648
649 {
650 unsigned int this_cpu = raw_smp_processor_id();
651 u64 t0, t1;
652
653 t0 = cpu_clock(this_cpu);
654 t1 = cpu_clock(this_cpu);
655 SEQ_printf(m, "%-45s:%21Ld\n",
656 "clock-delta", (long long)(t1-t0));
657 }
658
659 sched_show_numa(p, m);
660 }
661
662 void proc_sched_set_task(struct task_struct *p)
663 {
664 #ifdef CONFIG_SCHEDSTATS
665 memset(&p->se.statistics, 0, sizeof(p->se.statistics));
666 #endif
667 }
Linux kernel - Deliverables
This page took 0.053712 seconds and 5 git commands to generate.