sched: include group statistics in /proc/sched_debug
[safe/jmp/linux-2.6] / kernel / sched_debug.c
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 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
149 {
150         s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
151                 spread, rq0_min_vruntime, spread0;
152         struct rq *rq = &per_cpu(runqueues, cpu);
153         struct sched_entity *last;
154         unsigned long flags;
155
156 #if defined(CONFIG_CGROUP_SCHED) && defined(CONFIG_FAIR_GROUP_SCHED)
157         char path[128] = "";
158         struct task_group *tg = cfs_rq->tg;
159
160         cgroup_path(tg->css.cgroup, path, sizeof(path));
161
162         SEQ_printf(m, "\ncfs_rq[%d]:%s\n", cpu, path);
163 #else
164         SEQ_printf(m, "\ncfs_rq[%d]:\n", cpu);
165 #endif
166
167         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "exec_clock",
168                         SPLIT_NS(cfs_rq->exec_clock));
169
170         spin_lock_irqsave(&rq->lock, flags);
171         if (cfs_rq->rb_leftmost)
172                 MIN_vruntime = (__pick_next_entity(cfs_rq))->vruntime;
173         last = __pick_last_entity(cfs_rq);
174         if (last)
175                 max_vruntime = last->vruntime;
176         min_vruntime = rq->cfs.min_vruntime;
177         rq0_min_vruntime = per_cpu(runqueues, 0).cfs.min_vruntime;
178         spin_unlock_irqrestore(&rq->lock, flags);
179         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "MIN_vruntime",
180                         SPLIT_NS(MIN_vruntime));
181         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "min_vruntime",
182                         SPLIT_NS(min_vruntime));
183         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "max_vruntime",
184                         SPLIT_NS(max_vruntime));
185         spread = max_vruntime - MIN_vruntime;
186         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread",
187                         SPLIT_NS(spread));
188         spread0 = min_vruntime - rq0_min_vruntime;
189         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread0",
190                         SPLIT_NS(spread0));
191         SEQ_printf(m, "  .%-30s: %ld\n", "nr_running", cfs_rq->nr_running);
192         SEQ_printf(m, "  .%-30s: %ld\n", "load", cfs_rq->load.weight);
193 #ifdef CONFIG_SCHEDSTATS
194 #define P(n) SEQ_printf(m, "  .%-30s: %d\n", #n, rq->n);
195
196         P(yld_exp_empty);
197         P(yld_act_empty);
198         P(yld_both_empty);
199         P(yld_count);
200
201         P(sched_switch);
202         P(sched_count);
203         P(sched_goidle);
204
205         P(ttwu_count);
206         P(ttwu_local);
207
208         P(bkl_count);
209
210 #undef P
211 #endif
212         SEQ_printf(m, "  .%-30s: %ld\n", "nr_spread_over",
213                         cfs_rq->nr_spread_over);
214 #ifdef CONFIG_FAIR_GROUP_SCHED
215 #ifdef CONFIG_SMP
216         SEQ_printf(m, "  .%-30s: %lu\n", "shares", cfs_rq->shares);
217 #endif
218         print_cfs_group_stats(m, cpu, cfs_rq->tg);
219 #endif
220 }
221
222 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
223 {
224 #if defined(CONFIG_CGROUP_SCHED) && defined(CONFIG_RT_GROUP_SCHED)
225         char path[128] = "";
226         struct task_group *tg = rt_rq->tg;
227
228         cgroup_path(tg->css.cgroup, path, sizeof(path));
229
230         SEQ_printf(m, "\nrt_rq[%d]:%s\n", cpu, path);
231 #else
232         SEQ_printf(m, "\nrt_rq[%d]:\n", cpu);
233 #endif
234
235
236 #define P(x) \
237         SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
238 #define PN(x) \
239         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
240
241         P(rt_nr_running);
242         P(rt_throttled);
243         PN(rt_time);
244         PN(rt_runtime);
245
246 #undef PN
247 #undef P
248 }
249
250 static void print_cpu(struct seq_file *m, int cpu)
251 {
252         struct rq *rq = &per_cpu(runqueues, cpu);
253
254 #ifdef CONFIG_X86
255         {
256                 unsigned int freq = cpu_khz ? : 1;
257
258                 SEQ_printf(m, "\ncpu#%d, %u.%03u MHz\n",
259                            cpu, freq / 1000, (freq % 1000));
260         }
261 #else
262         SEQ_printf(m, "\ncpu#%d\n", cpu);
263 #endif
264
265 #define P(x) \
266         SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rq->x))
267 #define PN(x) \
268         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
269
270         P(nr_running);
271         SEQ_printf(m, "  .%-30s: %lu\n", "load",
272                    rq->load.weight);
273         P(nr_switches);
274         P(nr_load_updates);
275         P(nr_uninterruptible);
276         SEQ_printf(m, "  .%-30s: %lu\n", "jiffies", jiffies);
277         PN(next_balance);
278         P(curr->pid);
279         PN(clock);
280         P(cpu_load[0]);
281         P(cpu_load[1]);
282         P(cpu_load[2]);
283         P(cpu_load[3]);
284         P(cpu_load[4]);
285 #undef P
286 #undef PN
287
288         print_cfs_stats(m, cpu);
289         print_rt_stats(m, cpu);
290
291         print_rq(m, rq, cpu);
292 }
293
294 static int sched_debug_show(struct seq_file *m, void *v)
295 {
296         u64 now = ktime_to_ns(ktime_get());
297         int cpu;
298
299         SEQ_printf(m, "Sched Debug Version: v0.08, %s %.*s\n",
300                 init_utsname()->release,
301                 (int)strcspn(init_utsname()->version, " "),
302                 init_utsname()->version);
303
304         SEQ_printf(m, "now at %Lu.%06ld msecs\n", SPLIT_NS(now));
305
306 #define P(x) \
307         SEQ_printf(m, "  .%-40s: %Ld\n", #x, (long long)(x))
308 #define PN(x) \
309         SEQ_printf(m, "  .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
310         PN(sysctl_sched_latency);
311         PN(sysctl_sched_min_granularity);
312         PN(sysctl_sched_wakeup_granularity);
313         PN(sysctl_sched_child_runs_first);
314         P(sysctl_sched_features);
315 #undef PN
316 #undef P
317
318         for_each_online_cpu(cpu)
319                 print_cpu(m, cpu);
320
321         SEQ_printf(m, "\n");
322
323         return 0;
324 }
325
326 static void sysrq_sched_debug_show(void)
327 {
328         sched_debug_show(NULL, NULL);
329 }
330
331 static int sched_debug_open(struct inode *inode, struct file *filp)
332 {
333         return single_open(filp, sched_debug_show, NULL);
334 }
335
336 static const struct file_operations sched_debug_fops = {
337         .open           = sched_debug_open,
338         .read           = seq_read,
339         .llseek         = seq_lseek,
340         .release        = single_release,
341 };
342
343 static int __init init_sched_debug_procfs(void)
344 {
345         struct proc_dir_entry *pe;
346
347         pe = proc_create("sched_debug", 0444, NULL, &sched_debug_fops);
348         if (!pe)
349                 return -ENOMEM;
350         return 0;
351 }
352
353 __initcall(init_sched_debug_procfs);
354
355 void proc_sched_show_task(struct task_struct *p, struct seq_file *m)
356 {
357         unsigned long nr_switches;
358         unsigned long flags;
359         int num_threads = 1;
360
361         if (lock_task_sighand(p, &flags)) {
362                 num_threads = atomic_read(&p->signal->count);
363                 unlock_task_sighand(p, &flags);
364         }
365
366         SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, p->pid, num_threads);
367         SEQ_printf(m,
368                 "---------------------------------------------------------\n");
369 #define __P(F) \
370         SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)F)
371 #define P(F) \
372         SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)p->F)
373 #define __PN(F) \
374         SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
375 #define PN(F) \
376         SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
377
378         PN(se.exec_start);
379         PN(se.vruntime);
380         PN(se.sum_exec_runtime);
381         PN(se.avg_overlap);
382
383         nr_switches = p->nvcsw + p->nivcsw;
384
385 #ifdef CONFIG_SCHEDSTATS
386         PN(se.wait_start);
387         PN(se.sleep_start);
388         PN(se.block_start);
389         PN(se.sleep_max);
390         PN(se.block_max);
391         PN(se.exec_max);
392         PN(se.slice_max);
393         PN(se.wait_max);
394         PN(se.wait_sum);
395         P(se.wait_count);
396         P(sched_info.bkl_count);
397         P(se.nr_migrations);
398         P(se.nr_migrations_cold);
399         P(se.nr_failed_migrations_affine);
400         P(se.nr_failed_migrations_running);
401         P(se.nr_failed_migrations_hot);
402         P(se.nr_forced_migrations);
403         P(se.nr_forced2_migrations);
404         P(se.nr_wakeups);
405         P(se.nr_wakeups_sync);
406         P(se.nr_wakeups_migrate);
407         P(se.nr_wakeups_local);
408         P(se.nr_wakeups_remote);
409         P(se.nr_wakeups_affine);
410         P(se.nr_wakeups_affine_attempts);
411         P(se.nr_wakeups_passive);
412         P(se.nr_wakeups_idle);
413
414         {
415                 u64 avg_atom, avg_per_cpu;
416
417                 avg_atom = p->se.sum_exec_runtime;
418                 if (nr_switches)
419                         do_div(avg_atom, nr_switches);
420                 else
421                         avg_atom = -1LL;
422
423                 avg_per_cpu = p->se.sum_exec_runtime;
424                 if (p->se.nr_migrations) {
425                         avg_per_cpu = div64_u64(avg_per_cpu,
426                                                 p->se.nr_migrations);
427                 } else {
428                         avg_per_cpu = -1LL;
429                 }
430
431                 __PN(avg_atom);
432                 __PN(avg_per_cpu);
433         }
434 #endif
435         __P(nr_switches);
436         SEQ_printf(m, "%-35s:%21Ld\n",
437                    "nr_voluntary_switches", (long long)p->nvcsw);
438         SEQ_printf(m, "%-35s:%21Ld\n",
439                    "nr_involuntary_switches", (long long)p->nivcsw);
440
441         P(se.load.weight);
442         P(policy);
443         P(prio);
444 #undef PN
445 #undef __PN
446 #undef P
447 #undef __P
448
449         {
450                 u64 t0, t1;
451
452                 t0 = sched_clock();
453                 t1 = sched_clock();
454                 SEQ_printf(m, "%-35s:%21Ld\n",
455                            "clock-delta", (long long)(t1-t0));
456         }
457 }
458
459 void proc_sched_set_task(struct task_struct *p)
460 {
461 #ifdef CONFIG_SCHEDSTATS
462         p->se.wait_max                          = 0;
463         p->se.wait_sum                          = 0;
464         p->se.wait_count                        = 0;
465         p->se.sleep_max                         = 0;
466         p->se.sum_sleep_runtime                 = 0;
467         p->se.block_max                         = 0;
468         p->se.exec_max                          = 0;
469         p->se.slice_max                         = 0;
470         p->se.nr_migrations                     = 0;
471         p->se.nr_migrations_cold                = 0;
472         p->se.nr_failed_migrations_affine       = 0;
473         p->se.nr_failed_migrations_running      = 0;
474         p->se.nr_failed_migrations_hot          = 0;
475         p->se.nr_forced_migrations              = 0;
476         p->se.nr_forced2_migrations             = 0;
477         p->se.nr_wakeups                        = 0;
478         p->se.nr_wakeups_sync                   = 0;
479         p->se.nr_wakeups_migrate                = 0;
480         p->se.nr_wakeups_local                  = 0;
481         p->se.nr_wakeups_remote                 = 0;
482         p->se.nr_wakeups_affine                 = 0;
483         p->se.nr_wakeups_affine_attempts        = 0;
484         p->se.nr_wakeups_passive                = 0;
485         p->se.nr_wakeups_idle                   = 0;
486         p->sched_info.bkl_count                 = 0;
487 #endif
488         p->se.sum_exec_runtime                  = 0;
489         p->se.prev_sum_exec_runtime             = 0;
490         p->nvcsw                                = 0;
491         p->nivcsw                               = 0;
492 }