sched: debug: add some rt debug output
[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 static void
57 print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
58 {
59         if (rq->curr == p)
60                 SEQ_printf(m, "R");
61         else
62                 SEQ_printf(m, " ");
63
64         SEQ_printf(m, "%15s %5d %9Ld.%06ld %9Ld %5d ",
65                 p->comm, p->pid,
66                 SPLIT_NS(p->se.vruntime),
67                 (long long)(p->nvcsw + p->nivcsw),
68                 p->prio);
69 #ifdef CONFIG_SCHEDSTATS
70         SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
71                 SPLIT_NS(p->se.vruntime),
72                 SPLIT_NS(p->se.sum_exec_runtime),
73                 SPLIT_NS(p->se.sum_sleep_runtime));
74 #else
75         SEQ_printf(m, "%15Ld %15Ld %15Ld.%06ld %15Ld.%06ld %15Ld.%06ld",
76                 0LL, 0LL, 0LL, 0L, 0LL, 0L, 0LL, 0L);
77 #endif
78
79 #ifdef CONFIG_CGROUP_SCHED
80         {
81                 char path[64];
82
83                 cgroup_path(task_group(p)->css.cgroup, path, sizeof(path));
84                 SEQ_printf(m, " %s", path);
85         }
86 #endif
87         SEQ_printf(m, "\n");
88 }
89
90 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
91 {
92         struct task_struct *g, *p;
93         unsigned long flags;
94
95         SEQ_printf(m,
96         "\nrunnable tasks:\n"
97         "            task   PID         tree-key  switches  prio"
98         "     exec-runtime         sum-exec        sum-sleep\n"
99         "------------------------------------------------------"
100         "----------------------------------------------------\n");
101
102         read_lock_irqsave(&tasklist_lock, flags);
103
104         do_each_thread(g, p) {
105                 if (!p->se.on_rq || task_cpu(p) != rq_cpu)
106                         continue;
107
108                 print_task(m, rq, p);
109         } while_each_thread(g, p);
110
111         read_unlock_irqrestore(&tasklist_lock, flags);
112 }
113
114 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
115 {
116         s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
117                 spread, rq0_min_vruntime, spread0;
118         struct rq *rq = &per_cpu(runqueues, cpu);
119         struct sched_entity *last;
120         unsigned long flags;
121
122 #if defined(CONFIG_CGROUP_SCHED) && defined(CONFIG_FAIR_GROUP_SCHED)
123         char path[128] = "";
124         struct cgroup *cgroup = NULL;
125         struct task_group *tg = cfs_rq->tg;
126
127         if (tg)
128                 cgroup = tg->css.cgroup;
129
130         if (cgroup)
131                 cgroup_path(cgroup, path, sizeof(path));
132
133         SEQ_printf(m, "\ncfs_rq[%d]:%s\n", cpu, path);
134 #else
135         SEQ_printf(m, "\ncfs_rq[%d]:\n", cpu);
136 #endif
137
138         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "exec_clock",
139                         SPLIT_NS(cfs_rq->exec_clock));
140
141         spin_lock_irqsave(&rq->lock, flags);
142         if (cfs_rq->rb_leftmost)
143                 MIN_vruntime = (__pick_next_entity(cfs_rq))->vruntime;
144         last = __pick_last_entity(cfs_rq);
145         if (last)
146                 max_vruntime = last->vruntime;
147         min_vruntime = rq->cfs.min_vruntime;
148         rq0_min_vruntime = per_cpu(runqueues, 0).cfs.min_vruntime;
149         spin_unlock_irqrestore(&rq->lock, flags);
150         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "MIN_vruntime",
151                         SPLIT_NS(MIN_vruntime));
152         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "min_vruntime",
153                         SPLIT_NS(min_vruntime));
154         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "max_vruntime",
155                         SPLIT_NS(max_vruntime));
156         spread = max_vruntime - MIN_vruntime;
157         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread",
158                         SPLIT_NS(spread));
159         spread0 = min_vruntime - rq0_min_vruntime;
160         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread0",
161                         SPLIT_NS(spread0));
162         SEQ_printf(m, "  .%-30s: %ld\n", "nr_running", cfs_rq->nr_running);
163         SEQ_printf(m, "  .%-30s: %ld\n", "load", cfs_rq->load.weight);
164 #ifdef CONFIG_SCHEDSTATS
165         SEQ_printf(m, "  .%-30s: %d\n", "bkl_count",
166                         rq->bkl_count);
167 #endif
168         SEQ_printf(m, "  .%-30s: %ld\n", "nr_spread_over",
169                         cfs_rq->nr_spread_over);
170 }
171
172 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
173 {
174 #if defined(CONFIG_CGROUP_SCHED) && defined(CONFIG_RT_GROUP_SCHED)
175         char path[128] = "";
176         struct cgroup *cgroup = NULL;
177         struct task_group *tg = rt_rq->tg;
178
179         if (tg)
180                 cgroup = tg->css.cgroup;
181
182         if (cgroup)
183                 cgroup_path(cgroup, path, sizeof(path));
184
185         SEQ_printf(m, "\nrt_rq[%d]:%s\n", cpu, path);
186 #else
187         SEQ_printf(m, "\nrt_rq[%d]:\n", cpu);
188 #endif
189
190
191 #define P(x) \
192         SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
193 #define PN(x) \
194         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
195
196         P(rt_nr_running);
197         P(rt_throttled);
198         PN(rt_time);
199         PN(rt_runtime);
200
201 #undef PN
202 #undef P
203 }
204
205 static void print_cpu(struct seq_file *m, int cpu)
206 {
207         struct rq *rq = &per_cpu(runqueues, cpu);
208
209 #ifdef CONFIG_X86
210         {
211                 unsigned int freq = cpu_khz ? : 1;
212
213                 SEQ_printf(m, "\ncpu#%d, %u.%03u MHz\n",
214                            cpu, freq / 1000, (freq % 1000));
215         }
216 #else
217         SEQ_printf(m, "\ncpu#%d\n", cpu);
218 #endif
219
220 #define P(x) \
221         SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rq->x))
222 #define PN(x) \
223         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
224
225         P(nr_running);
226         SEQ_printf(m, "  .%-30s: %lu\n", "load",
227                    rq->load.weight);
228         P(nr_switches);
229         P(nr_load_updates);
230         P(nr_uninterruptible);
231         SEQ_printf(m, "  .%-30s: %lu\n", "jiffies", jiffies);
232         PN(next_balance);
233         P(curr->pid);
234         PN(clock);
235         P(cpu_load[0]);
236         P(cpu_load[1]);
237         P(cpu_load[2]);
238         P(cpu_load[3]);
239         P(cpu_load[4]);
240 #undef P
241 #undef PN
242
243         print_cfs_stats(m, cpu);
244         print_rt_stats(m, cpu);
245
246         print_rq(m, rq, cpu);
247 }
248
249 static int sched_debug_show(struct seq_file *m, void *v)
250 {
251         u64 now = ktime_to_ns(ktime_get());
252         int cpu;
253
254         SEQ_printf(m, "Sched Debug Version: v0.07, %s %.*s\n",
255                 init_utsname()->release,
256                 (int)strcspn(init_utsname()->version, " "),
257                 init_utsname()->version);
258
259         SEQ_printf(m, "now at %Lu.%06ld msecs\n", SPLIT_NS(now));
260
261 #define P(x) \
262         SEQ_printf(m, "  .%-40s: %Ld\n", #x, (long long)(x))
263 #define PN(x) \
264         SEQ_printf(m, "  .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
265         PN(sysctl_sched_latency);
266         PN(sysctl_sched_min_granularity);
267         PN(sysctl_sched_wakeup_granularity);
268         PN(sysctl_sched_child_runs_first);
269         P(sysctl_sched_features);
270 #undef PN
271 #undef P
272
273         for_each_online_cpu(cpu)
274                 print_cpu(m, cpu);
275
276         SEQ_printf(m, "\n");
277
278         return 0;
279 }
280
281 static void sysrq_sched_debug_show(void)
282 {
283         sched_debug_show(NULL, NULL);
284 }
285
286 static int sched_debug_open(struct inode *inode, struct file *filp)
287 {
288         return single_open(filp, sched_debug_show, NULL);
289 }
290
291 static const struct file_operations sched_debug_fops = {
292         .open           = sched_debug_open,
293         .read           = seq_read,
294         .llseek         = seq_lseek,
295         .release        = single_release,
296 };
297
298 static int __init init_sched_debug_procfs(void)
299 {
300         struct proc_dir_entry *pe;
301
302         pe = proc_create("sched_debug", 0644, NULL, &sched_debug_fops);
303         if (!pe)
304                 return -ENOMEM;
305         return 0;
306 }
307
308 __initcall(init_sched_debug_procfs);
309
310 void proc_sched_show_task(struct task_struct *p, struct seq_file *m)
311 {
312         unsigned long nr_switches;
313         unsigned long flags;
314         int num_threads = 1;
315
316         rcu_read_lock();
317         if (lock_task_sighand(p, &flags)) {
318                 num_threads = atomic_read(&p->signal->count);
319                 unlock_task_sighand(p, &flags);
320         }
321         rcu_read_unlock();
322
323         SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, p->pid, num_threads);
324         SEQ_printf(m,
325                 "---------------------------------------------------------\n");
326 #define __P(F) \
327         SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)F)
328 #define P(F) \
329         SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)p->F)
330 #define __PN(F) \
331         SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
332 #define PN(F) \
333         SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
334
335         PN(se.exec_start);
336         PN(se.vruntime);
337         PN(se.sum_exec_runtime);
338         PN(se.avg_overlap);
339
340         nr_switches = p->nvcsw + p->nivcsw;
341
342 #ifdef CONFIG_SCHEDSTATS
343         PN(se.wait_start);
344         PN(se.sleep_start);
345         PN(se.block_start);
346         PN(se.sleep_max);
347         PN(se.block_max);
348         PN(se.exec_max);
349         PN(se.slice_max);
350         PN(se.wait_max);
351         PN(se.wait_sum);
352         P(se.wait_count);
353         P(sched_info.bkl_count);
354         P(se.nr_migrations);
355         P(se.nr_migrations_cold);
356         P(se.nr_failed_migrations_affine);
357         P(se.nr_failed_migrations_running);
358         P(se.nr_failed_migrations_hot);
359         P(se.nr_forced_migrations);
360         P(se.nr_forced2_migrations);
361         P(se.nr_wakeups);
362         P(se.nr_wakeups_sync);
363         P(se.nr_wakeups_migrate);
364         P(se.nr_wakeups_local);
365         P(se.nr_wakeups_remote);
366         P(se.nr_wakeups_affine);
367         P(se.nr_wakeups_affine_attempts);
368         P(se.nr_wakeups_passive);
369         P(se.nr_wakeups_idle);
370
371         {
372                 u64 avg_atom, avg_per_cpu;
373
374                 avg_atom = p->se.sum_exec_runtime;
375                 if (nr_switches)
376                         do_div(avg_atom, nr_switches);
377                 else
378                         avg_atom = -1LL;
379
380                 avg_per_cpu = p->se.sum_exec_runtime;
381                 if (p->se.nr_migrations) {
382                         avg_per_cpu = div64_u64(avg_per_cpu,
383                                                 p->se.nr_migrations);
384                 } else {
385                         avg_per_cpu = -1LL;
386                 }
387
388                 __PN(avg_atom);
389                 __PN(avg_per_cpu);
390         }
391 #endif
392         __P(nr_switches);
393         SEQ_printf(m, "%-35s:%21Ld\n",
394                    "nr_voluntary_switches", (long long)p->nvcsw);
395         SEQ_printf(m, "%-35s:%21Ld\n",
396                    "nr_involuntary_switches", (long long)p->nivcsw);
397
398         P(se.load.weight);
399         P(policy);
400         P(prio);
401 #undef PN
402 #undef __PN
403 #undef P
404 #undef __P
405
406         {
407                 u64 t0, t1;
408
409                 t0 = sched_clock();
410                 t1 = sched_clock();
411                 SEQ_printf(m, "%-35s:%21Ld\n",
412                            "clock-delta", (long long)(t1-t0));
413         }
414 }
415
416 void proc_sched_set_task(struct task_struct *p)
417 {
418 #ifdef CONFIG_SCHEDSTATS
419         p->se.wait_max                          = 0;
420         p->se.wait_sum                          = 0;
421         p->se.wait_count                        = 0;
422         p->se.sleep_max                         = 0;
423         p->se.sum_sleep_runtime                 = 0;
424         p->se.block_max                         = 0;
425         p->se.exec_max                          = 0;
426         p->se.slice_max                         = 0;
427         p->se.nr_migrations                     = 0;
428         p->se.nr_migrations_cold                = 0;
429         p->se.nr_failed_migrations_affine       = 0;
430         p->se.nr_failed_migrations_running      = 0;
431         p->se.nr_failed_migrations_hot          = 0;
432         p->se.nr_forced_migrations              = 0;
433         p->se.nr_forced2_migrations             = 0;
434         p->se.nr_wakeups                        = 0;
435         p->se.nr_wakeups_sync                   = 0;
436         p->se.nr_wakeups_migrate                = 0;
437         p->se.nr_wakeups_local                  = 0;
438         p->se.nr_wakeups_remote                 = 0;
439         p->se.nr_wakeups_affine                 = 0;
440         p->se.nr_wakeups_affine_attempts        = 0;
441         p->se.nr_wakeups_passive                = 0;
442         p->se.nr_wakeups_idle                   = 0;
443         p->sched_info.bkl_count                 = 0;
444 #endif
445         p->se.sum_exec_runtime                  = 0;
446         p->se.prev_sum_exec_runtime             = 0;
447         p->nvcsw                                = 0;
448         p->nivcsw                               = 0;
449 }