Merge branch 'core-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
[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->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 #if defined(CONFIG_CGROUP_SCHED) && \
149         (defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED))
150 static void task_group_path(struct task_group *tg, char *buf, int buflen)
151 {
152         /* may be NULL if the underlying cgroup isn't fully-created yet */
153         if (!tg->css.cgroup) {
154                 buf[0] = '\0';
155                 return;
156         }
157         cgroup_path(tg->css.cgroup, buf, buflen);
158 }
159 #endif
160
161 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
162 {
163         s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
164                 spread, rq0_min_vruntime, spread0;
165         struct rq *rq = cpu_rq(cpu);
166         struct sched_entity *last;
167         unsigned long flags;
168
169 #if defined(CONFIG_CGROUP_SCHED) && defined(CONFIG_FAIR_GROUP_SCHED)
170         char path[128];
171         struct task_group *tg = cfs_rq->tg;
172
173         task_group_path(tg, path, sizeof(path));
174
175         SEQ_printf(m, "\ncfs_rq[%d]:%s\n", cpu, path);
176 #elif defined(CONFIG_USER_SCHED) && defined(CONFIG_FAIR_GROUP_SCHED)
177         {
178                 uid_t uid = cfs_rq->tg->uid;
179                 SEQ_printf(m, "\ncfs_rq[%d] for UID: %u\n", cpu, uid);
180         }
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_next_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: %ld\n", "nr_running", cfs_rq->nr_running);
209         SEQ_printf(m, "  .%-30s: %ld\n", "load", cfs_rq->load.weight);
210
211         SEQ_printf(m, "  .%-30s: %d\n", "nr_spread_over",
212                         cfs_rq->nr_spread_over);
213 #ifdef CONFIG_FAIR_GROUP_SCHED
214 #ifdef CONFIG_SMP
215         SEQ_printf(m, "  .%-30s: %lu\n", "shares", cfs_rq->shares);
216 #endif
217         print_cfs_group_stats(m, cpu, cfs_rq->tg);
218 #endif
219 }
220
221 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
222 {
223 #if defined(CONFIG_CGROUP_SCHED) && defined(CONFIG_RT_GROUP_SCHED)
224         char path[128];
225         struct task_group *tg = rt_rq->tg;
226
227         task_group_path(tg, path, sizeof(path));
228
229         SEQ_printf(m, "\nrt_rq[%d]:%s\n", cpu, path);
230 #else
231         SEQ_printf(m, "\nrt_rq[%d]:\n", cpu);
232 #endif
233
234
235 #define P(x) \
236         SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
237 #define PN(x) \
238         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
239
240         P(rt_nr_running);
241         P(rt_throttled);
242         PN(rt_time);
243         PN(rt_runtime);
244
245 #undef PN
246 #undef P
247 }
248
249 static void print_cpu(struct seq_file *m, int cpu)
250 {
251         struct rq *rq = cpu_rq(cpu);
252
253 #ifdef CONFIG_X86
254         {
255                 unsigned int freq = cpu_khz ? : 1;
256
257                 SEQ_printf(m, "\ncpu#%d, %u.%03u MHz\n",
258                            cpu, freq / 1000, (freq % 1000));
259         }
260 #else
261         SEQ_printf(m, "\ncpu#%d\n", cpu);
262 #endif
263
264 #define P(x) \
265         SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rq->x))
266 #define PN(x) \
267         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
268
269         P(nr_running);
270         SEQ_printf(m, "  .%-30s: %lu\n", "load",
271                    rq->load.weight);
272         P(nr_switches);
273         P(nr_load_updates);
274         P(nr_uninterruptible);
275         PN(next_balance);
276         P(curr->pid);
277         PN(clock);
278         P(cpu_load[0]);
279         P(cpu_load[1]);
280         P(cpu_load[2]);
281         P(cpu_load[3]);
282         P(cpu_load[4]);
283 #undef P
284 #undef PN
285
286 #ifdef CONFIG_SCHEDSTATS
287 #define P(n) SEQ_printf(m, "  .%-30s: %d\n", #n, rq->n);
288 #define P64(n) SEQ_printf(m, "  .%-30s: %Ld\n", #n, rq->n);
289
290         P(yld_count);
291
292         P(sched_switch);
293         P(sched_count);
294         P(sched_goidle);
295 #ifdef CONFIG_SMP
296         P64(avg_idle);
297 #endif
298
299         P(ttwu_count);
300         P(ttwu_local);
301
302         P(bkl_count);
303
304 #undef P
305 #endif
306         print_cfs_stats(m, cpu);
307         print_rt_stats(m, cpu);
308
309         print_rq(m, rq, cpu);
310 }
311
312 static const char *sched_tunable_scaling_names[] = {
313         "none",
314         "logaritmic",
315         "linear"
316 };
317
318 static int sched_debug_show(struct seq_file *m, void *v)
319 {
320         u64 now = ktime_to_ns(ktime_get());
321         int cpu;
322
323         SEQ_printf(m, "Sched Debug Version: v0.09, %s %.*s\n",
324                 init_utsname()->release,
325                 (int)strcspn(init_utsname()->version, " "),
326                 init_utsname()->version);
327
328         SEQ_printf(m, "now at %Lu.%06ld msecs\n", SPLIT_NS(now));
329
330 #define P(x) \
331         SEQ_printf(m, "  .%-40s: %Ld\n", #x, (long long)(x))
332 #define PN(x) \
333         SEQ_printf(m, "  .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
334         P(jiffies);
335         PN(sysctl_sched_latency);
336         PN(sysctl_sched_min_granularity);
337         PN(sysctl_sched_wakeup_granularity);
338         PN(sysctl_sched_child_runs_first);
339         P(sysctl_sched_features);
340 #undef PN
341 #undef P
342
343         SEQ_printf(m, "  .%-40s: %d (%s)\n", "sysctl_sched_tunable_scaling",
344                 sysctl_sched_tunable_scaling,
345                 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
346
347         for_each_online_cpu(cpu)
348                 print_cpu(m, cpu);
349
350         SEQ_printf(m, "\n");
351
352         return 0;
353 }
354
355 static void sysrq_sched_debug_show(void)
356 {
357         sched_debug_show(NULL, NULL);
358 }
359
360 static int sched_debug_open(struct inode *inode, struct file *filp)
361 {
362         return single_open(filp, sched_debug_show, NULL);
363 }
364
365 static const struct file_operations sched_debug_fops = {
366         .open           = sched_debug_open,
367         .read           = seq_read,
368         .llseek         = seq_lseek,
369         .release        = single_release,
370 };
371
372 static int __init init_sched_debug_procfs(void)
373 {
374         struct proc_dir_entry *pe;
375
376         pe = proc_create("sched_debug", 0444, NULL, &sched_debug_fops);
377         if (!pe)
378                 return -ENOMEM;
379         return 0;
380 }
381
382 __initcall(init_sched_debug_procfs);
383
384 void proc_sched_show_task(struct task_struct *p, struct seq_file *m)
385 {
386         unsigned long nr_switches;
387         unsigned long flags;
388         int num_threads = 1;
389
390         if (lock_task_sighand(p, &flags)) {
391                 num_threads = atomic_read(&p->signal->count);
392                 unlock_task_sighand(p, &flags);
393         }
394
395         SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, p->pid, num_threads);
396         SEQ_printf(m,
397                 "---------------------------------------------------------\n");
398 #define __P(F) \
399         SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)F)
400 #define P(F) \
401         SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)p->F)
402 #define __PN(F) \
403         SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
404 #define PN(F) \
405         SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
406
407         PN(se.exec_start);
408         PN(se.vruntime);
409         PN(se.sum_exec_runtime);
410         PN(se.avg_overlap);
411         PN(se.avg_wakeup);
412
413         nr_switches = p->nvcsw + p->nivcsw;
414
415 #ifdef CONFIG_SCHEDSTATS
416         PN(se.wait_start);
417         PN(se.sleep_start);
418         PN(se.block_start);
419         PN(se.sleep_max);
420         PN(se.block_max);
421         PN(se.exec_max);
422         PN(se.slice_max);
423         PN(se.wait_max);
424         PN(se.wait_sum);
425         P(se.wait_count);
426         PN(se.iowait_sum);
427         P(se.iowait_count);
428         P(sched_info.bkl_count);
429         P(se.nr_migrations);
430         P(se.nr_migrations_cold);
431         P(se.nr_failed_migrations_affine);
432         P(se.nr_failed_migrations_running);
433         P(se.nr_failed_migrations_hot);
434         P(se.nr_forced_migrations);
435         P(se.nr_wakeups);
436         P(se.nr_wakeups_sync);
437         P(se.nr_wakeups_migrate);
438         P(se.nr_wakeups_local);
439         P(se.nr_wakeups_remote);
440         P(se.nr_wakeups_affine);
441         P(se.nr_wakeups_affine_attempts);
442         P(se.nr_wakeups_passive);
443         P(se.nr_wakeups_idle);
444
445         {
446                 u64 avg_atom, avg_per_cpu;
447
448                 avg_atom = p->se.sum_exec_runtime;
449                 if (nr_switches)
450                         do_div(avg_atom, nr_switches);
451                 else
452                         avg_atom = -1LL;
453
454                 avg_per_cpu = p->se.sum_exec_runtime;
455                 if (p->se.nr_migrations) {
456                         avg_per_cpu = div64_u64(avg_per_cpu,
457                                                 p->se.nr_migrations);
458                 } else {
459                         avg_per_cpu = -1LL;
460                 }
461
462                 __PN(avg_atom);
463                 __PN(avg_per_cpu);
464         }
465 #endif
466         __P(nr_switches);
467         SEQ_printf(m, "%-35s:%21Ld\n",
468                    "nr_voluntary_switches", (long long)p->nvcsw);
469         SEQ_printf(m, "%-35s:%21Ld\n",
470                    "nr_involuntary_switches", (long long)p->nivcsw);
471
472         P(se.load.weight);
473         P(policy);
474         P(prio);
475 #undef PN
476 #undef __PN
477 #undef P
478 #undef __P
479
480         {
481                 unsigned int this_cpu = raw_smp_processor_id();
482                 u64 t0, t1;
483
484                 t0 = cpu_clock(this_cpu);
485                 t1 = cpu_clock(this_cpu);
486                 SEQ_printf(m, "%-35s:%21Ld\n",
487                            "clock-delta", (long long)(t1-t0));
488         }
489 }
490
491 void proc_sched_set_task(struct task_struct *p)
492 {
493 #ifdef CONFIG_SCHEDSTATS
494         p->se.wait_max                          = 0;
495         p->se.wait_sum                          = 0;
496         p->se.wait_count                        = 0;
497         p->se.iowait_sum                        = 0;
498         p->se.iowait_count                      = 0;
499         p->se.sleep_max                         = 0;
500         p->se.sum_sleep_runtime                 = 0;
501         p->se.block_max                         = 0;
502         p->se.exec_max                          = 0;
503         p->se.slice_max                         = 0;
504         p->se.nr_migrations                     = 0;
505         p->se.nr_migrations_cold                = 0;
506         p->se.nr_failed_migrations_affine       = 0;
507         p->se.nr_failed_migrations_running      = 0;
508         p->se.nr_failed_migrations_hot          = 0;
509         p->se.nr_forced_migrations              = 0;
510         p->se.nr_wakeups                        = 0;
511         p->se.nr_wakeups_sync                   = 0;
512         p->se.nr_wakeups_migrate                = 0;
513         p->se.nr_wakeups_local                  = 0;
514         p->se.nr_wakeups_remote                 = 0;
515         p->se.nr_wakeups_affine                 = 0;
516         p->se.nr_wakeups_affine_attempts        = 0;
517         p->se.nr_wakeups_passive                = 0;
518         p->se.nr_wakeups_idle                   = 0;
519         p->sched_info.bkl_count                 = 0;
520 #endif
521         p->se.sum_exec_runtime                  = 0;
522         p->se.prev_sum_exec_runtime             = 0;
523         p->nvcsw                                = 0;
524         p->nivcsw                               = 0;
525 }