695f9773bb6018fb75d4424d51e91cdbd7c1eb71
[projects/modsched/linux.git] / kernel / sched / cfs / 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                 struct sched_avg *avg = &cpu_rq(cpu)->avg;
73                 P(avg->runnable_avg_sum);
74                 P(avg->runnable_avg_period);
75                 return;
76         }
77
78
79         PN(se->exec_start);
80         PN(se->vruntime);
81         PN(se->sum_exec_runtime);
82 #ifdef CONFIG_SCHEDSTATS
83         PN(se->statistics.wait_start);
84         PN(se->statistics.sleep_start);
85         PN(se->statistics.block_start);
86         PN(se->statistics.sleep_max);
87         PN(se->statistics.block_max);
88         PN(se->statistics.exec_max);
89         PN(se->statistics.slice_max);
90         PN(se->statistics.wait_max);
91         PN(se->statistics.wait_sum);
92         P(se->statistics.wait_count);
93 #endif
94         P(se->load.weight);
95 #ifdef CONFIG_SMP
96         P(se->avg.runnable_avg_sum);
97         P(se->avg.runnable_avg_period);
98         P(se->avg.load_avg_contrib);
99         P(se->avg.decay_count);
100 #endif
101 #undef PN
102 #undef P
103 }
104 #endif
105
106 #ifdef CONFIG_CGROUP_SCHED
107 static char group_path[PATH_MAX];
108
109 static char *task_group_path(struct task_group *tg)
110 {
111         if (autogroup_path(tg, group_path, PATH_MAX))
112                 return group_path;
113
114         return cgroup_path(tg->css.cgroup, group_path, PATH_MAX);
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, task_pid_nr(p),
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_NUMA_BALANCING
141         SEQ_printf(m, " %d", task_node(p));
142 #endif
143 #ifdef CONFIG_CGROUP_SCHED
144         SEQ_printf(m, " %s", task_group_path(task_group(p)));
145 #endif
146
147         SEQ_printf(m, "\n");
148 }
149
150 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
151 {
152         struct task_struct *g, *p;
153         unsigned long flags;
154
155         SEQ_printf(m,
156         "\nrunnable tasks:\n"
157         "            task   PID         tree-key  switches  prio"
158         "     exec-runtime         sum-exec        sum-sleep\n"
159         "------------------------------------------------------"
160         "----------------------------------------------------\n");
161
162         read_lock_irqsave(&tasklist_lock, flags);
163
164         do_each_thread(g, p) {
165                 if (task_cpu(p) != rq_cpu)
166                         continue;
167
168                 print_task(m, rq, p);
169         } while_each_thread(g, p);
170
171         read_unlock_irqrestore(&tasklist_lock, flags);
172 }
173
174 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
175 {
176         s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
177                 spread, rq0_min_vruntime, spread0;
178         struct rq *rq = cpu_rq(cpu);
179         struct sched_entity *last;
180         unsigned long flags;
181
182 #ifdef CONFIG_FAIR_GROUP_SCHED
183         SEQ_printf(m, "\ncfs_rq[%d]:%s\n", cpu, task_group_path(cfs_rq->tg));
184 #else
185         SEQ_printf(m, "\ncfs_rq[%d]:\n", cpu);
186 #endif
187         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "exec_clock",
188                         SPLIT_NS(cfs_rq->exec_clock));
189
190         raw_spin_lock_irqsave(&rq->lock, flags);
191         if (cfs_rq->rb_leftmost)
192                 MIN_vruntime = (__pick_first_entity(cfs_rq))->vruntime;
193         last = __pick_last_entity(cfs_rq);
194         if (last)
195                 max_vruntime = last->vruntime;
196         min_vruntime = cfs_rq->min_vruntime;
197         rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
198         raw_spin_unlock_irqrestore(&rq->lock, flags);
199         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "MIN_vruntime",
200                         SPLIT_NS(MIN_vruntime));
201         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "min_vruntime",
202                         SPLIT_NS(min_vruntime));
203         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "max_vruntime",
204                         SPLIT_NS(max_vruntime));
205         spread = max_vruntime - MIN_vruntime;
206         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread",
207                         SPLIT_NS(spread));
208         spread0 = min_vruntime - rq0_min_vruntime;
209         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread0",
210                         SPLIT_NS(spread0));
211         SEQ_printf(m, "  .%-30s: %d\n", "nr_spread_over",
212                         cfs_rq->nr_spread_over);
213         SEQ_printf(m, "  .%-30s: %d\n", "nr_running", cfs_rq->nr_running);
214         SEQ_printf(m, "  .%-30s: %ld\n", "load", cfs_rq->load.weight);
215 #ifdef CONFIG_SMP
216         SEQ_printf(m, "  .%-30s: %ld\n", "runnable_load_avg",
217                         cfs_rq->runnable_load_avg);
218         SEQ_printf(m, "  .%-30s: %ld\n", "blocked_load_avg",
219                         cfs_rq->blocked_load_avg);
220 #ifdef CONFIG_FAIR_GROUP_SCHED
221         SEQ_printf(m, "  .%-30s: %ld\n", "tg_load_contrib",
222                         cfs_rq->tg_load_contrib);
223         SEQ_printf(m, "  .%-30s: %d\n", "tg_runnable_contrib",
224                         cfs_rq->tg_runnable_contrib);
225         SEQ_printf(m, "  .%-30s: %ld\n", "tg_load_avg",
226                         atomic_long_read(&cfs_rq->tg->load_avg));
227         SEQ_printf(m, "  .%-30s: %d\n", "tg->runnable_avg",
228                         atomic_read(&cfs_rq->tg->runnable_avg));
229 #endif
230 #endif
231 #ifdef CONFIG_CFS_BANDWIDTH
232         SEQ_printf(m, "  .%-30s: %d\n", "tg->cfs_bandwidth.timer_active",
233                         cfs_rq->tg->cfs_bandwidth.timer_active);
234         SEQ_printf(m, "  .%-30s: %d\n", "throttled",
235                         cfs_rq->throttled);
236         SEQ_printf(m, "  .%-30s: %d\n", "throttle_count",
237                         cfs_rq->throttle_count);
238 #endif
239
240 #ifdef CONFIG_FAIR_GROUP_SCHED
241         print_cfs_group_stats(m, cpu, cfs_rq->tg);
242 #endif
243 }
244
245 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
246 {
247 #ifdef CONFIG_RT_GROUP_SCHED
248         SEQ_printf(m, "\nrt_rq[%d]:%s\n", cpu, task_group_path(rt_rq->tg));
249 #else
250         SEQ_printf(m, "\nrt_rq[%d]:\n", cpu);
251 #endif
252
253 #define P(x) \
254         SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
255 #define PN(x) \
256         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
257
258         P(rt_nr_running);
259         P(rt_throttled);
260         PN(rt_time);
261         PN(rt_runtime);
262
263 #undef PN
264 #undef P
265 }
266
267 extern __read_mostly int sched_clock_running;
268
269 static void print_cpu(struct seq_file *m, int cpu)
270 {
271         struct rq *rq = cpu_rq(cpu);
272         unsigned long flags;
273
274 #ifdef CONFIG_X86
275         {
276                 unsigned int freq = cpu_khz ? : 1;
277
278                 SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
279                            cpu, freq / 1000, (freq % 1000));
280         }
281 #else
282         SEQ_printf(m, "cpu#%d\n", cpu);
283 #endif
284
285 #define P(x)                                                            \
286 do {                                                                    \
287         if (sizeof(rq->x) == 4)                                         \
288                 SEQ_printf(m, "  .%-30s: %ld\n", #x, (long)(rq->x));    \
289         else                                                            \
290                 SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rq->x));\
291 } while (0)
292
293 #define PN(x) \
294         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
295
296         P(nr_running);
297         SEQ_printf(m, "  .%-30s: %lu\n", "load",
298                    rq->load.weight);
299         P(nr_switches);
300         P(nr_load_updates);
301         P(nr_uninterruptible);
302         PN(next_balance);
303         SEQ_printf(m, "  .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr)));
304         PN(clock);
305         P(cpu_load[0]);
306         P(cpu_load[1]);
307         P(cpu_load[2]);
308         P(cpu_load[3]);
309         P(cpu_load[4]);
310 #undef P
311 #undef PN
312
313 #ifdef CONFIG_SCHEDSTATS
314 #define P(n) SEQ_printf(m, "  .%-30s: %d\n", #n, rq->n);
315 #define P64(n) SEQ_printf(m, "  .%-30s: %Ld\n", #n, rq->n);
316
317         P(yld_count);
318
319         P(sched_count);
320         P(sched_goidle);
321 #ifdef CONFIG_SMP
322         P64(avg_idle);
323         P64(max_idle_balance_cost);
324 #endif
325
326         P(ttwu_count);
327         P(ttwu_local);
328
329 #undef P
330 #undef P64
331 #endif
332         spin_lock_irqsave(&sched_debug_lock, flags);
333         print_cfs_stats(m, cpu);
334         print_rt_stats(m, cpu);
335
336         rcu_read_lock();
337         print_rq(m, rq, cpu);
338         rcu_read_unlock();
339         spin_unlock_irqrestore(&sched_debug_lock, flags);
340         SEQ_printf(m, "\n");
341 }
342
343 static const char *sched_tunable_scaling_names[] = {
344         "none",
345         "logaritmic",
346         "linear"
347 };
348
349 static void sched_debug_header(struct seq_file *m)
350 {
351         u64 ktime, sched_clk, cpu_clk;
352         unsigned long flags;
353
354         local_irq_save(flags);
355         ktime = ktime_to_ns(ktime_get());
356         sched_clk = sched_clock();
357         cpu_clk = local_clock();
358         local_irq_restore(flags);
359
360         SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n",
361                 init_utsname()->release,
362                 (int)strcspn(init_utsname()->version, " "),
363                 init_utsname()->version);
364
365 #define P(x) \
366         SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
367 #define PN(x) \
368         SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
369         PN(ktime);
370         PN(sched_clk);
371         PN(cpu_clk);
372         P(jiffies);
373 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
374         P(sched_clock_stable());
375 #endif
376 #undef PN
377 #undef P
378
379         SEQ_printf(m, "\n");
380         SEQ_printf(m, "sysctl_sched\n");
381
382 #define P(x) \
383         SEQ_printf(m, "  .%-40s: %Ld\n", #x, (long long)(x))
384 #define PN(x) \
385         SEQ_printf(m, "  .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
386         PN(sysctl_sched_latency);
387         PN(sysctl_sched_min_granularity);
388         PN(sysctl_sched_wakeup_granularity);
389         P(sysctl_sched_child_runs_first);
390         P(sysctl_sched_features);
391 #undef PN
392 #undef P
393
394         SEQ_printf(m, "  .%-40s: %d (%s)\n",
395                 "sysctl_sched_tunable_scaling",
396                 sysctl_sched_tunable_scaling,
397                 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
398         SEQ_printf(m, "\n");
399 }
400
401 static int sched_debug_show(struct seq_file *m, void *v)
402 {
403         int cpu = (unsigned long)(v - 2);
404
405         if (cpu != -1)
406                 print_cpu(m, cpu);
407         else
408                 sched_debug_header(m);
409
410         return 0;
411 }
412
413 void sysrq_sched_debug_show(void)
414 {
415         int cpu;
416
417         sched_debug_header(NULL);
418         for_each_online_cpu(cpu)
419                 print_cpu(NULL, cpu);
420
421 }
422
423 /*
424  * This itererator needs some explanation.
425  * It returns 1 for the header position.
426  * This means 2 is cpu 0.
427  * In a hotplugged system some cpus, including cpu 0, may be missing so we have
428  * to use cpumask_* to iterate over the cpus.
429  */
430 static void *sched_debug_start(struct seq_file *file, loff_t *offset)
431 {
432         unsigned long n = *offset;
433
434         if (n == 0)
435                 return (void *) 1;
436
437         n--;
438
439         if (n > 0)
440                 n = cpumask_next(n - 1, cpu_online_mask);
441         else
442                 n = cpumask_first(cpu_online_mask);
443
444         *offset = n + 1;
445
446         if (n < nr_cpu_ids)
447                 return (void *)(unsigned long)(n + 2);
448         return NULL;
449 }
450
451 static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
452 {
453         (*offset)++;
454         return sched_debug_start(file, offset);
455 }
456
457 static void sched_debug_stop(struct seq_file *file, void *data)
458 {
459 }
460
461 static const struct seq_operations sched_debug_sops = {
462         .start = sched_debug_start,
463         .next = sched_debug_next,
464         .stop = sched_debug_stop,
465         .show = sched_debug_show,
466 };
467
468 static int sched_debug_release(struct inode *inode, struct file *file)
469 {
470         seq_release(inode, file);
471
472         return 0;
473 }
474
475 static int sched_debug_open(struct inode *inode, struct file *filp)
476 {
477         int ret = 0;
478
479         ret = seq_open(filp, &sched_debug_sops);
480
481         return ret;
482 }
483
484 static const struct file_operations sched_debug_fops = {
485         .open           = sched_debug_open,
486         .read           = seq_read,
487         .llseek         = seq_lseek,
488         .release        = sched_debug_release,
489 };
490
491 static int __init init_sched_debug_procfs(void)
492 {
493         struct proc_dir_entry *pe;
494
495         pe = proc_create("sched_debug", 0444, NULL, &sched_debug_fops);
496         if (!pe)
497                 return -ENOMEM;
498         return 0;
499 }
500
501 __initcall(init_sched_debug_procfs);
502
503 #define __P(F) \
504         SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)F)
505 #define P(F) \
506         SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)p->F)
507 #define __PN(F) \
508         SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
509 #define PN(F) \
510         SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
511
512
513 static void sched_show_numa(struct task_struct *p, struct seq_file *m)
514 {
515 #ifdef CONFIG_NUMA_BALANCING
516         struct mempolicy *pol;
517         int node, i;
518
519         if (p->mm)
520                 P(mm->numa_scan_seq);
521
522         task_lock(p);
523         pol = p->mempolicy;
524         if (pol && !(pol->flags & MPOL_F_MORON))
525                 pol = NULL;
526         mpol_get(pol);
527         task_unlock(p);
528
529         SEQ_printf(m, "numa_migrations, %ld\n", xchg(&p->numa_pages_migrated, 0));
530
531         for_each_online_node(node) {
532                 for (i = 0; i < 2; i++) {
533                         unsigned long nr_faults = -1;
534                         int cpu_current, home_node;
535
536                         if (p->numa_faults_memory)
537                                 nr_faults = p->numa_faults_memory[2*node + i];
538
539                         cpu_current = !i ? (task_node(p) == node) :
540                                 (pol && node_isset(node, pol->v.nodes));
541
542                         home_node = (p->numa_preferred_nid == node);
543
544                         SEQ_printf(m, "numa_faults_memory, %d, %d, %d, %d, %ld\n",
545                                 i, node, cpu_current, home_node, nr_faults);
546                 }
547         }
548
549         mpol_put(pol);
550 #endif
551 }
552
553 void proc_sched_show_task(struct task_struct *p, struct seq_file *m)
554 {
555         unsigned long nr_switches;
556
557         SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr(p),
558                                                 get_nr_threads(p));
559         SEQ_printf(m,
560                 "---------------------------------------------------------"
561                 "----------\n");
562 #define __P(F) \
563         SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)F)
564 #define P(F) \
565         SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)p->F)
566 #define __PN(F) \
567         SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
568 #define PN(F) \
569         SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
570
571         PN(se.exec_start);
572         PN(se.vruntime);
573         PN(se.sum_exec_runtime);
574
575         nr_switches = p->nvcsw + p->nivcsw;
576
577 #ifdef CONFIG_SCHEDSTATS
578         PN(se.statistics.wait_start);
579         PN(se.statistics.sleep_start);
580         PN(se.statistics.block_start);
581         PN(se.statistics.sleep_max);
582         PN(se.statistics.block_max);
583         PN(se.statistics.exec_max);
584         PN(se.statistics.slice_max);
585         PN(se.statistics.wait_max);
586         PN(se.statistics.wait_sum);
587         P(se.statistics.wait_count);
588         PN(se.statistics.iowait_sum);
589         P(se.statistics.iowait_count);
590         P(se.nr_migrations);
591         P(se.statistics.nr_migrations_cold);
592         P(se.statistics.nr_failed_migrations_affine);
593         P(se.statistics.nr_failed_migrations_running);
594         P(se.statistics.nr_failed_migrations_hot);
595         P(se.statistics.nr_forced_migrations);
596         P(se.statistics.nr_wakeups);
597         P(se.statistics.nr_wakeups_sync);
598         P(se.statistics.nr_wakeups_migrate);
599         P(se.statistics.nr_wakeups_local);
600         P(se.statistics.nr_wakeups_remote);
601         P(se.statistics.nr_wakeups_affine);
602         P(se.statistics.nr_wakeups_affine_attempts);
603         P(se.statistics.nr_wakeups_passive);
604         P(se.statistics.nr_wakeups_idle);
605
606         {
607                 u64 avg_atom, avg_per_cpu;
608
609                 avg_atom = p->se.sum_exec_runtime;
610                 if (nr_switches)
611                         do_div(avg_atom, nr_switches);
612                 else
613                         avg_atom = -1LL;
614
615                 avg_per_cpu = p->se.sum_exec_runtime;
616                 if (p->se.nr_migrations) {
617                         avg_per_cpu = div64_u64(avg_per_cpu,
618                                                 p->se.nr_migrations);
619                 } else {
620                         avg_per_cpu = -1LL;
621                 }
622
623                 __PN(avg_atom);
624                 __PN(avg_per_cpu);
625         }
626 #endif
627         __P(nr_switches);
628         SEQ_printf(m, "%-45s:%21Ld\n",
629                    "nr_voluntary_switches", (long long)p->nvcsw);
630         SEQ_printf(m, "%-45s:%21Ld\n",
631                    "nr_involuntary_switches", (long long)p->nivcsw);
632
633         P(se.load.weight);
634 #ifdef CONFIG_SMP
635         P(se.avg.runnable_avg_sum);
636         P(se.avg.runnable_avg_period);
637         P(se.avg.load_avg_contrib);
638         P(se.avg.decay_count);
639 #endif
640         P(policy);
641         P(prio);
642 #undef PN
643 #undef __PN
644 #undef P
645 #undef __P
646
647         {
648                 unsigned int this_cpu = raw_smp_processor_id();
649                 u64 t0, t1;
650
651                 t0 = cpu_clock(this_cpu);
652                 t1 = cpu_clock(this_cpu);
653                 SEQ_printf(m, "%-45s:%21Ld\n",
654                            "clock-delta", (long long)(t1-t0));
655         }
656
657         sched_show_numa(p, m);
658 }
659
660 void proc_sched_set_task(struct task_struct *p)
661 {
662 #ifdef CONFIG_SCHEDSTATS
663         memset(&p->se.statistics, 0, sizeof(p->se.statistics));
664 #endif
665 }