cpuquiet: Update averaging of nr_runnables
[linux-3.10.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
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         cgroup_path(tg->css.cgroup, group_path, PATH_MAX);
114         return group_path;
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, p->pid,
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_CGROUP_SCHED
141         SEQ_printf(m, " %s", task_group_path(task_group(p)));
142 #endif
143
144         SEQ_printf(m, "\n");
145 }
146
147 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
148 {
149         struct task_struct *g, *p;
150         unsigned long flags;
151
152         SEQ_printf(m,
153         "\nrunnable tasks:\n"
154         "            task   PID         tree-key  switches  prio"
155         "     exec-runtime         sum-exec        sum-sleep\n"
156         "------------------------------------------------------"
157         "----------------------------------------------------\n");
158
159         read_lock_irqsave(&tasklist_lock, flags);
160
161         do_each_thread(g, p) {
162                 if (!p->on_rq || task_cpu(p) != rq_cpu)
163                         continue;
164
165                 print_task(m, rq, p);
166         } while_each_thread(g, p);
167
168         read_unlock_irqrestore(&tasklist_lock, flags);
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_FAIR_GROUP_SCHED
213 #ifdef CONFIG_SMP
214         SEQ_printf(m, "  .%-30s: %lld\n", "runnable_load_avg",
215                         cfs_rq->runnable_load_avg);
216         SEQ_printf(m, "  .%-30s: %lld\n", "blocked_load_avg",
217                         cfs_rq->blocked_load_avg);
218         SEQ_printf(m, "  .%-30s: %lld\n", "tg_load_avg",
219                         (unsigned long long)atomic64_read(&cfs_rq->tg->load_avg));
220         SEQ_printf(m, "  .%-30s: %lld\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: %d\n", "tg->runnable_avg",
225                         atomic_read(&cfs_rq->tg->runnable_avg));
226 #endif
227
228         print_cfs_group_stats(m, cpu, cfs_rq->tg);
229 #endif
230 }
231
232 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
233 {
234 #ifdef CONFIG_RT_GROUP_SCHED
235         SEQ_printf(m, "\nrt_rq[%d]:%s\n", cpu, task_group_path(rt_rq->tg));
236 #else
237         SEQ_printf(m, "\nrt_rq[%d]:\n", cpu);
238 #endif
239
240 #define P(x) \
241         SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
242 #define PN(x) \
243         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
244
245         P(rt_nr_running);
246         P(rt_throttled);
247         PN(rt_time);
248         PN(rt_runtime);
249
250 #undef PN
251 #undef P
252 }
253
254 extern __read_mostly int sched_clock_running;
255
256 static void print_cpu(struct seq_file *m, int cpu)
257 {
258         struct rq *rq = cpu_rq(cpu);
259         unsigned long flags;
260
261 #ifdef CONFIG_X86
262         {
263                 unsigned int freq = cpu_khz ? : 1;
264
265                 SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
266                            cpu, freq / 1000, (freq % 1000));
267         }
268 #else
269         SEQ_printf(m, "cpu#%d\n", cpu);
270 #endif
271
272 #define P(x)                                                            \
273 do {                                                                    \
274         if (sizeof(rq->x) == 4)                                         \
275                 SEQ_printf(m, "  .%-30s: %ld\n", #x, (long)(rq->x));    \
276         else                                                            \
277                 SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rq->x));\
278 } while (0)
279
280 #define PN(x) \
281         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
282
283         P(nr_running);
284         SEQ_printf(m, "  .%-30s: %lu\n", "load",
285                    rq->load.weight);
286         P(nr_switches);
287         P(nr_load_updates);
288         P(nr_uninterruptible);
289         PN(next_balance);
290         P(curr->pid);
291         PN(clock);
292         P(cpu_load[0]);
293         P(cpu_load[1]);
294         P(cpu_load[2]);
295         P(cpu_load[3]);
296         P(cpu_load[4]);
297 #undef P
298 #undef PN
299
300 #ifdef CONFIG_SCHEDSTATS
301 #define P(n) SEQ_printf(m, "  .%-30s: %d\n", #n, rq->n);
302 #define P64(n) SEQ_printf(m, "  .%-30s: %Ld\n", #n, rq->n);
303
304         P(yld_count);
305
306         P(sched_count);
307         P(sched_goidle);
308 #ifdef CONFIG_SMP
309         P64(avg_idle);
310 #endif
311
312         P(ttwu_count);
313         P(ttwu_local);
314
315 #undef P
316 #undef P64
317 #endif
318         spin_lock_irqsave(&sched_debug_lock, flags);
319         print_cfs_stats(m, cpu);
320         print_rt_stats(m, cpu);
321
322         rcu_read_lock();
323         print_rq(m, rq, cpu);
324         rcu_read_unlock();
325         spin_unlock_irqrestore(&sched_debug_lock, flags);
326         SEQ_printf(m, "\n");
327 }
328
329 static const char *sched_tunable_scaling_names[] = {
330         "none",
331         "logaritmic",
332         "linear"
333 };
334
335 static void sched_debug_header(struct seq_file *m)
336 {
337         u64 ktime, sched_clk, cpu_clk;
338         unsigned long flags;
339
340         local_irq_save(flags);
341         ktime = ktime_to_ns(ktime_get());
342         sched_clk = sched_clock();
343         cpu_clk = local_clock();
344         local_irq_restore(flags);
345
346         SEQ_printf(m, "Sched Debug Version: v0.10, %s %.*s\n",
347                 init_utsname()->release,
348                 (int)strcspn(init_utsname()->version, " "),
349                 init_utsname()->version);
350
351 #define P(x) \
352         SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
353 #define PN(x) \
354         SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
355         PN(ktime);
356         PN(sched_clk);
357         PN(cpu_clk);
358         P(jiffies);
359 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
360         P(sched_clock_stable);
361 #endif
362 #undef PN
363 #undef P
364
365         SEQ_printf(m, "\n");
366         SEQ_printf(m, "sysctl_sched\n");
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(sysctl_sched_latency);
373         PN(sysctl_sched_min_granularity);
374         PN(sysctl_sched_wakeup_granularity);
375         P(sysctl_sched_child_runs_first);
376         P(sysctl_sched_features);
377 #undef PN
378 #undef P
379
380         SEQ_printf(m, "  .%-40s: %d (%s)\n",
381                 "sysctl_sched_tunable_scaling",
382                 sysctl_sched_tunable_scaling,
383                 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
384         SEQ_printf(m, "\n");
385 }
386
387 static int sched_debug_show(struct seq_file *m, void *v)
388 {
389         int cpu = (unsigned long)(v - 2);
390
391         if (cpu != -1)
392                 print_cpu(m, cpu);
393         else
394                 sched_debug_header(m);
395
396         return 0;
397 }
398
399 void sysrq_sched_debug_show(void)
400 {
401         int cpu;
402
403         sched_debug_header(NULL);
404         for_each_online_cpu(cpu)
405                 print_cpu(NULL, cpu);
406
407 }
408
409 /*
410  * This itererator needs some explanation.
411  * It returns 1 for the header position.
412  * This means 2 is cpu 0.
413  * In a hotplugged system some cpus, including cpu 0, may be missing so we have
414  * to use cpumask_* to iterate over the cpus.
415  */
416 static void *sched_debug_start(struct seq_file *file, loff_t *offset)
417 {
418         unsigned long n = *offset;
419
420         if (n == 0)
421                 return (void *) 1;
422
423         n--;
424
425         if (n > 0)
426                 n = cpumask_next(n - 1, cpu_online_mask);
427         else
428                 n = cpumask_first(cpu_online_mask);
429
430         *offset = n + 1;
431
432         if (n < nr_cpu_ids)
433                 return (void *)(unsigned long)(n + 2);
434         return NULL;
435 }
436
437 static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
438 {
439         (*offset)++;
440         return sched_debug_start(file, offset);
441 }
442
443 static void sched_debug_stop(struct seq_file *file, void *data)
444 {
445 }
446
447 static const struct seq_operations sched_debug_sops = {
448         .start = sched_debug_start,
449         .next = sched_debug_next,
450         .stop = sched_debug_stop,
451         .show = sched_debug_show,
452 };
453
454 static int sched_debug_release(struct inode *inode, struct file *file)
455 {
456         seq_release(inode, file);
457
458         return 0;
459 }
460
461 static int sched_debug_open(struct inode *inode, struct file *filp)
462 {
463         int ret = 0;
464
465         ret = seq_open(filp, &sched_debug_sops);
466
467         return ret;
468 }
469
470 static const struct file_operations sched_debug_fops = {
471         .open           = sched_debug_open,
472         .read           = seq_read,
473         .llseek         = seq_lseek,
474         .release        = sched_debug_release,
475 };
476
477 static int __init init_sched_debug_procfs(void)
478 {
479         struct proc_dir_entry *pe;
480
481         pe = proc_create("sched_debug", 0444, NULL, &sched_debug_fops);
482         if (!pe)
483                 return -ENOMEM;
484         return 0;
485 }
486
487 __initcall(init_sched_debug_procfs);
488
489 void proc_sched_show_task(struct task_struct *p, struct seq_file *m)
490 {
491         unsigned long nr_switches;
492
493         SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, p->pid,
494                                                 get_nr_threads(p));
495         SEQ_printf(m,
496                 "---------------------------------------------------------\n");
497 #define __P(F) \
498         SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)F)
499 #define P(F) \
500         SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)p->F)
501 #define __PN(F) \
502         SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
503 #define PN(F) \
504         SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
505
506         PN(se.exec_start);
507         PN(se.vruntime);
508         PN(se.sum_exec_runtime);
509
510         nr_switches = p->nvcsw + p->nivcsw;
511
512 #ifdef CONFIG_SCHEDSTATS
513         PN(se.statistics.wait_start);
514         PN(se.statistics.sleep_start);
515         PN(se.statistics.block_start);
516         PN(se.statistics.sleep_max);
517         PN(se.statistics.block_max);
518         PN(se.statistics.exec_max);
519         PN(se.statistics.slice_max);
520         PN(se.statistics.wait_max);
521         PN(se.statistics.wait_sum);
522         P(se.statistics.wait_count);
523         PN(se.statistics.iowait_sum);
524         P(se.statistics.iowait_count);
525         P(se.nr_migrations);
526         P(se.statistics.nr_migrations_cold);
527         P(se.statistics.nr_failed_migrations_affine);
528         P(se.statistics.nr_failed_migrations_running);
529         P(se.statistics.nr_failed_migrations_hot);
530         P(se.statistics.nr_forced_migrations);
531         P(se.statistics.nr_wakeups);
532         P(se.statistics.nr_wakeups_sync);
533         P(se.statistics.nr_wakeups_migrate);
534         P(se.statistics.nr_wakeups_local);
535         P(se.statistics.nr_wakeups_remote);
536         P(se.statistics.nr_wakeups_affine);
537         P(se.statistics.nr_wakeups_affine_attempts);
538         P(se.statistics.nr_wakeups_passive);
539         P(se.statistics.nr_wakeups_idle);
540
541         {
542                 u64 avg_atom, avg_per_cpu;
543
544                 avg_atom = p->se.sum_exec_runtime;
545                 if (nr_switches)
546                         do_div(avg_atom, nr_switches);
547                 else
548                         avg_atom = -1LL;
549
550                 avg_per_cpu = p->se.sum_exec_runtime;
551                 if (p->se.nr_migrations) {
552                         avg_per_cpu = div64_u64(avg_per_cpu,
553                                                 p->se.nr_migrations);
554                 } else {
555                         avg_per_cpu = -1LL;
556                 }
557
558                 __PN(avg_atom);
559                 __PN(avg_per_cpu);
560         }
561 #endif
562         __P(nr_switches);
563         SEQ_printf(m, "%-35s:%21Ld\n",
564                    "nr_voluntary_switches", (long long)p->nvcsw);
565         SEQ_printf(m, "%-35s:%21Ld\n",
566                    "nr_involuntary_switches", (long long)p->nivcsw);
567
568         P(se.load.weight);
569         P(policy);
570         P(prio);
571 #undef PN
572 #undef __PN
573 #undef P
574 #undef __P
575
576         {
577                 unsigned int this_cpu = raw_smp_processor_id();
578                 u64 t0, t1;
579
580                 t0 = cpu_clock(this_cpu);
581                 t1 = cpu_clock(this_cpu);
582                 SEQ_printf(m, "%-35s:%21Ld\n",
583                            "clock-delta", (long long)(t1-t0));
584         }
585 }
586
587 void proc_sched_set_task(struct task_struct *p)
588 {
589 #ifdef CONFIG_SCHEDSTATS
590         memset(&p->se.statistics, 0, sizeof(p->se.statistics));
591 #endif
592 }