nick piggin: change email address
[linux-2.6.git] / 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->statistics.wait_start);
74         PN(se->statistics.sleep_start);
75         PN(se->statistics.block_start);
76         PN(se->statistics.sleep_max);
77         PN(se->statistics.block_max);
78         PN(se->statistics.exec_max);
79         PN(se->statistics.slice_max);
80         PN(se->statistics.wait_max);
81         PN(se->statistics.wait_sum);
82         P(se->statistics.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.statistics.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                 rcu_read_lock();
118                 cgroup_path(task_group(p)->css.cgroup, path, sizeof(path));
119                 rcu_read_unlock();
120                 SEQ_printf(m, " %s", path);
121         }
122 #endif
123         SEQ_printf(m, "\n");
124 }
125
126 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
127 {
128         struct task_struct *g, *p;
129         unsigned long flags;
130
131         SEQ_printf(m,
132         "\nrunnable tasks:\n"
133         "            task   PID         tree-key  switches  prio"
134         "     exec-runtime         sum-exec        sum-sleep\n"
135         "------------------------------------------------------"
136         "----------------------------------------------------\n");
137
138         read_lock_irqsave(&tasklist_lock, flags);
139
140         do_each_thread(g, p) {
141                 if (!p->se.on_rq || task_cpu(p) != rq_cpu)
142                         continue;
143
144                 print_task(m, rq, p);
145         } while_each_thread(g, p);
146
147         read_unlock_irqrestore(&tasklist_lock, flags);
148 }
149
150 #if defined(CONFIG_CGROUP_SCHED) && \
151         (defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED))
152 static void task_group_path(struct task_group *tg, char *buf, int buflen)
153 {
154         /* may be NULL if the underlying cgroup isn't fully-created yet */
155         if (!tg->css.cgroup) {
156                 buf[0] = '\0';
157                 return;
158         }
159         cgroup_path(tg->css.cgroup, buf, buflen);
160 }
161 #endif
162
163 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
164 {
165         s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
166                 spread, rq0_min_vruntime, spread0;
167         struct rq *rq = cpu_rq(cpu);
168         struct sched_entity *last;
169         unsigned long flags;
170
171 #if defined(CONFIG_CGROUP_SCHED) && defined(CONFIG_FAIR_GROUP_SCHED)
172         char path[128];
173         struct task_group *tg = cfs_rq->tg;
174
175         task_group_path(tg, path, sizeof(path));
176
177         SEQ_printf(m, "\ncfs_rq[%d]:%s\n", cpu, path);
178 #else
179         SEQ_printf(m, "\ncfs_rq[%d]:\n", cpu);
180 #endif
181         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "exec_clock",
182                         SPLIT_NS(cfs_rq->exec_clock));
183
184         raw_spin_lock_irqsave(&rq->lock, flags);
185         if (cfs_rq->rb_leftmost)
186                 MIN_vruntime = (__pick_next_entity(cfs_rq))->vruntime;
187         last = __pick_last_entity(cfs_rq);
188         if (last)
189                 max_vruntime = last->vruntime;
190         min_vruntime = cfs_rq->min_vruntime;
191         rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
192         raw_spin_unlock_irqrestore(&rq->lock, flags);
193         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "MIN_vruntime",
194                         SPLIT_NS(MIN_vruntime));
195         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "min_vruntime",
196                         SPLIT_NS(min_vruntime));
197         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "max_vruntime",
198                         SPLIT_NS(max_vruntime));
199         spread = max_vruntime - MIN_vruntime;
200         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread",
201                         SPLIT_NS(spread));
202         spread0 = min_vruntime - rq0_min_vruntime;
203         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread0",
204                         SPLIT_NS(spread0));
205         SEQ_printf(m, "  .%-30s: %ld\n", "nr_running", cfs_rq->nr_running);
206         SEQ_printf(m, "  .%-30s: %ld\n", "load", cfs_rq->load.weight);
207
208         SEQ_printf(m, "  .%-30s: %d\n", "nr_spread_over",
209                         cfs_rq->nr_spread_over);
210 #ifdef CONFIG_FAIR_GROUP_SCHED
211 #ifdef CONFIG_SMP
212         SEQ_printf(m, "  .%-30s: %lu\n", "shares", cfs_rq->shares);
213 #endif
214         print_cfs_group_stats(m, cpu, cfs_rq->tg);
215 #endif
216 }
217
218 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
219 {
220 #if defined(CONFIG_CGROUP_SCHED) && defined(CONFIG_RT_GROUP_SCHED)
221         char path[128];
222         struct task_group *tg = rt_rq->tg;
223
224         task_group_path(tg, path, sizeof(path));
225
226         SEQ_printf(m, "\nrt_rq[%d]:%s\n", cpu, path);
227 #else
228         SEQ_printf(m, "\nrt_rq[%d]:\n", cpu);
229 #endif
230
231
232 #define P(x) \
233         SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
234 #define PN(x) \
235         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
236
237         P(rt_nr_running);
238         P(rt_throttled);
239         PN(rt_time);
240         PN(rt_runtime);
241
242 #undef PN
243 #undef P
244 }
245
246 static void print_cpu(struct seq_file *m, int cpu)
247 {
248         struct rq *rq = cpu_rq(cpu);
249
250 #ifdef CONFIG_X86
251         {
252                 unsigned int freq = cpu_khz ? : 1;
253
254                 SEQ_printf(m, "\ncpu#%d, %u.%03u MHz\n",
255                            cpu, freq / 1000, (freq % 1000));
256         }
257 #else
258         SEQ_printf(m, "\ncpu#%d\n", cpu);
259 #endif
260
261 #define P(x) \
262         SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rq->x))
263 #define PN(x) \
264         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
265
266         P(nr_running);
267         SEQ_printf(m, "  .%-30s: %lu\n", "load",
268                    rq->load.weight);
269         P(nr_switches);
270         P(nr_load_updates);
271         P(nr_uninterruptible);
272         PN(next_balance);
273         P(curr->pid);
274         PN(clock);
275         P(cpu_load[0]);
276         P(cpu_load[1]);
277         P(cpu_load[2]);
278         P(cpu_load[3]);
279         P(cpu_load[4]);
280 #undef P
281 #undef PN
282
283 #ifdef CONFIG_SCHEDSTATS
284 #define P(n) SEQ_printf(m, "  .%-30s: %d\n", #n, rq->n);
285 #define P64(n) SEQ_printf(m, "  .%-30s: %Ld\n", #n, rq->n);
286
287         P(yld_count);
288
289         P(sched_switch);
290         P(sched_count);
291         P(sched_goidle);
292 #ifdef CONFIG_SMP
293         P64(avg_idle);
294 #endif
295
296         P(ttwu_count);
297         P(ttwu_local);
298
299         P(bkl_count);
300
301 #undef P
302 #endif
303         print_cfs_stats(m, cpu);
304         print_rt_stats(m, cpu);
305
306         print_rq(m, rq, cpu);
307 }
308
309 static const char *sched_tunable_scaling_names[] = {
310         "none",
311         "logaritmic",
312         "linear"
313 };
314
315 static int sched_debug_show(struct seq_file *m, void *v)
316 {
317         u64 now = ktime_to_ns(ktime_get());
318         int cpu;
319
320         SEQ_printf(m, "Sched Debug Version: v0.09, %s %.*s\n",
321                 init_utsname()->release,
322                 (int)strcspn(init_utsname()->version, " "),
323                 init_utsname()->version);
324
325         SEQ_printf(m, "now at %Lu.%06ld msecs\n", SPLIT_NS(now));
326
327 #define P(x) \
328         SEQ_printf(m, "  .%-40s: %Ld\n", #x, (long long)(x))
329 #define PN(x) \
330         SEQ_printf(m, "  .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
331         P(jiffies);
332         PN(sysctl_sched_latency);
333         PN(sysctl_sched_min_granularity);
334         PN(sysctl_sched_wakeup_granularity);
335         PN(sysctl_sched_child_runs_first);
336         P(sysctl_sched_features);
337 #undef PN
338 #undef P
339
340         SEQ_printf(m, "  .%-40s: %d (%s)\n", "sysctl_sched_tunable_scaling",
341                 sysctl_sched_tunable_scaling,
342                 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
343
344         for_each_online_cpu(cpu)
345                 print_cpu(m, cpu);
346
347         SEQ_printf(m, "\n");
348
349         return 0;
350 }
351
352 static void sysrq_sched_debug_show(void)
353 {
354         sched_debug_show(NULL, NULL);
355 }
356
357 static int sched_debug_open(struct inode *inode, struct file *filp)
358 {
359         return single_open(filp, sched_debug_show, NULL);
360 }
361
362 static const struct file_operations sched_debug_fops = {
363         .open           = sched_debug_open,
364         .read           = seq_read,
365         .llseek         = seq_lseek,
366         .release        = single_release,
367 };
368
369 static int __init init_sched_debug_procfs(void)
370 {
371         struct proc_dir_entry *pe;
372
373         pe = proc_create("sched_debug", 0444, NULL, &sched_debug_fops);
374         if (!pe)
375                 return -ENOMEM;
376         return 0;
377 }
378
379 __initcall(init_sched_debug_procfs);
380
381 void proc_sched_show_task(struct task_struct *p, struct seq_file *m)
382 {
383         unsigned long nr_switches;
384
385         SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, p->pid,
386                                                 get_nr_threads(p));
387         SEQ_printf(m,
388                 "---------------------------------------------------------\n");
389 #define __P(F) \
390         SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)F)
391 #define P(F) \
392         SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)p->F)
393 #define __PN(F) \
394         SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
395 #define PN(F) \
396         SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
397
398         PN(se.exec_start);
399         PN(se.vruntime);
400         PN(se.sum_exec_runtime);
401
402         nr_switches = p->nvcsw + p->nivcsw;
403
404 #ifdef CONFIG_SCHEDSTATS
405         PN(se.statistics.wait_start);
406         PN(se.statistics.sleep_start);
407         PN(se.statistics.block_start);
408         PN(se.statistics.sleep_max);
409         PN(se.statistics.block_max);
410         PN(se.statistics.exec_max);
411         PN(se.statistics.slice_max);
412         PN(se.statistics.wait_max);
413         PN(se.statistics.wait_sum);
414         P(se.statistics.wait_count);
415         PN(se.statistics.iowait_sum);
416         P(se.statistics.iowait_count);
417         P(sched_info.bkl_count);
418         P(se.nr_migrations);
419         P(se.statistics.nr_migrations_cold);
420         P(se.statistics.nr_failed_migrations_affine);
421         P(se.statistics.nr_failed_migrations_running);
422         P(se.statistics.nr_failed_migrations_hot);
423         P(se.statistics.nr_forced_migrations);
424         P(se.statistics.nr_wakeups);
425         P(se.statistics.nr_wakeups_sync);
426         P(se.statistics.nr_wakeups_migrate);
427         P(se.statistics.nr_wakeups_local);
428         P(se.statistics.nr_wakeups_remote);
429         P(se.statistics.nr_wakeups_affine);
430         P(se.statistics.nr_wakeups_affine_attempts);
431         P(se.statistics.nr_wakeups_passive);
432         P(se.statistics.nr_wakeups_idle);
433
434         {
435                 u64 avg_atom, avg_per_cpu;
436
437                 avg_atom = p->se.sum_exec_runtime;
438                 if (nr_switches)
439                         do_div(avg_atom, nr_switches);
440                 else
441                         avg_atom = -1LL;
442
443                 avg_per_cpu = p->se.sum_exec_runtime;
444                 if (p->se.nr_migrations) {
445                         avg_per_cpu = div64_u64(avg_per_cpu,
446                                                 p->se.nr_migrations);
447                 } else {
448                         avg_per_cpu = -1LL;
449                 }
450
451                 __PN(avg_atom);
452                 __PN(avg_per_cpu);
453         }
454 #endif
455         __P(nr_switches);
456         SEQ_printf(m, "%-35s:%21Ld\n",
457                    "nr_voluntary_switches", (long long)p->nvcsw);
458         SEQ_printf(m, "%-35s:%21Ld\n",
459                    "nr_involuntary_switches", (long long)p->nivcsw);
460
461         P(se.load.weight);
462         P(policy);
463         P(prio);
464 #undef PN
465 #undef __PN
466 #undef P
467 #undef __P
468
469         {
470                 unsigned int this_cpu = raw_smp_processor_id();
471                 u64 t0, t1;
472
473                 t0 = cpu_clock(this_cpu);
474                 t1 = cpu_clock(this_cpu);
475                 SEQ_printf(m, "%-35s:%21Ld\n",
476                            "clock-delta", (long long)(t1-t0));
477         }
478 }
479
480 void proc_sched_set_task(struct task_struct *p)
481 {
482 #ifdef CONFIG_SCHEDSTATS
483         memset(&p->se.statistics, 0, sizeof(p->se.statistics));
484 #endif
485 }