sched: fix kernel warning on /proc/sched_debug access
[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 = cfs_rq->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
165         SEQ_printf(m, "  .%-30s: %d\n", "nr_spread_over",
166                         cfs_rq->nr_spread_over);
167 #ifdef CONFIG_FAIR_GROUP_SCHED
168 #ifdef CONFIG_SMP
169         SEQ_printf(m, "  .%-30s: %lu\n", "shares", cfs_rq->shares);
170 #endif
171 #endif
172 }
173
174 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
175 {
176 #if defined(CONFIG_CGROUP_SCHED) && defined(CONFIG_RT_GROUP_SCHED)
177         char path[128] = "";
178         struct cgroup *cgroup = NULL;
179         struct task_group *tg = rt_rq->tg;
180
181         if (tg)
182                 cgroup = tg->css.cgroup;
183
184         if (cgroup)
185                 cgroup_path(cgroup, path, sizeof(path));
186
187         SEQ_printf(m, "\nrt_rq[%d]:%s\n", cpu, path);
188 #else
189         SEQ_printf(m, "\nrt_rq[%d]:\n", cpu);
190 #endif
191
192
193 #define P(x) \
194         SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
195 #define PN(x) \
196         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
197
198         P(rt_nr_running);
199         P(rt_throttled);
200         PN(rt_time);
201         PN(rt_runtime);
202
203 #undef PN
204 #undef P
205 }
206
207 static void print_cpu(struct seq_file *m, int cpu)
208 {
209         struct rq *rq = &per_cpu(runqueues, cpu);
210
211 #ifdef CONFIG_X86
212         {
213                 unsigned int freq = cpu_khz ? : 1;
214
215                 SEQ_printf(m, "\ncpu#%d, %u.%03u MHz\n",
216                            cpu, freq / 1000, (freq % 1000));
217         }
218 #else
219         SEQ_printf(m, "\ncpu#%d\n", cpu);
220 #endif
221
222 #define P(x) \
223         SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rq->x))
224 #define PN(x) \
225         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
226
227         P(nr_running);
228         SEQ_printf(m, "  .%-30s: %lu\n", "load",
229                    rq->load.weight);
230         P(nr_switches);
231         P(nr_load_updates);
232         P(nr_uninterruptible);
233         SEQ_printf(m, "  .%-30s: %lu\n", "jiffies", jiffies);
234         PN(next_balance);
235         P(curr->pid);
236         PN(clock);
237         P(cpu_load[0]);
238         P(cpu_load[1]);
239         P(cpu_load[2]);
240         P(cpu_load[3]);
241         P(cpu_load[4]);
242 #undef P
243 #undef PN
244
245 #ifdef CONFIG_SCHEDSTATS
246 #define P(n) SEQ_printf(m, "  .%-30s: %d\n", #n, rq->n);
247
248         P(yld_exp_empty);
249         P(yld_act_empty);
250         P(yld_both_empty);
251         P(yld_count);
252
253         P(sched_switch);
254         P(sched_count);
255         P(sched_goidle);
256
257         P(ttwu_count);
258         P(ttwu_local);
259
260         P(bkl_count);
261
262 #undef P
263 #endif
264         print_cfs_stats(m, cpu);
265         print_rt_stats(m, cpu);
266
267         print_rq(m, rq, cpu);
268 }
269
270 static int sched_debug_show(struct seq_file *m, void *v)
271 {
272         u64 now = ktime_to_ns(ktime_get());
273         int cpu;
274
275         SEQ_printf(m, "Sched Debug Version: v0.07, %s %.*s\n",
276                 init_utsname()->release,
277                 (int)strcspn(init_utsname()->version, " "),
278                 init_utsname()->version);
279
280         SEQ_printf(m, "now at %Lu.%06ld msecs\n", SPLIT_NS(now));
281
282 #define P(x) \
283         SEQ_printf(m, "  .%-40s: %Ld\n", #x, (long long)(x))
284 #define PN(x) \
285         SEQ_printf(m, "  .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
286         PN(sysctl_sched_latency);
287         PN(sysctl_sched_min_granularity);
288         PN(sysctl_sched_wakeup_granularity);
289         PN(sysctl_sched_child_runs_first);
290         P(sysctl_sched_features);
291 #undef PN
292 #undef P
293
294         for_each_online_cpu(cpu)
295                 print_cpu(m, cpu);
296
297         SEQ_printf(m, "\n");
298
299         return 0;
300 }
301
302 static void sysrq_sched_debug_show(void)
303 {
304         sched_debug_show(NULL, NULL);
305 }
306
307 static int sched_debug_open(struct inode *inode, struct file *filp)
308 {
309         return single_open(filp, sched_debug_show, NULL);
310 }
311
312 static const struct file_operations sched_debug_fops = {
313         .open           = sched_debug_open,
314         .read           = seq_read,
315         .llseek         = seq_lseek,
316         .release        = single_release,
317 };
318
319 static int __init init_sched_debug_procfs(void)
320 {
321         struct proc_dir_entry *pe;
322
323         pe = proc_create("sched_debug", 0444, NULL, &sched_debug_fops);
324         if (!pe)
325                 return -ENOMEM;
326         return 0;
327 }
328
329 __initcall(init_sched_debug_procfs);
330
331 void proc_sched_show_task(struct task_struct *p, struct seq_file *m)
332 {
333         unsigned long nr_switches;
334         unsigned long flags;
335         int num_threads = 1;
336
337         if (lock_task_sighand(p, &flags)) {
338                 num_threads = atomic_read(&p->signal->count);
339                 unlock_task_sighand(p, &flags);
340         }
341
342         SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, p->pid, num_threads);
343         SEQ_printf(m,
344                 "---------------------------------------------------------\n");
345 #define __P(F) \
346         SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)F)
347 #define P(F) \
348         SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)p->F)
349 #define __PN(F) \
350         SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
351 #define PN(F) \
352         SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
353
354         PN(se.exec_start);
355         PN(se.vruntime);
356         PN(se.sum_exec_runtime);
357         PN(se.avg_overlap);
358
359         nr_switches = p->nvcsw + p->nivcsw;
360
361 #ifdef CONFIG_SCHEDSTATS
362         PN(se.wait_start);
363         PN(se.sleep_start);
364         PN(se.block_start);
365         PN(se.sleep_max);
366         PN(se.block_max);
367         PN(se.exec_max);
368         PN(se.slice_max);
369         PN(se.wait_max);
370         PN(se.wait_sum);
371         P(se.wait_count);
372         P(sched_info.bkl_count);
373         P(se.nr_migrations);
374         P(se.nr_migrations_cold);
375         P(se.nr_failed_migrations_affine);
376         P(se.nr_failed_migrations_running);
377         P(se.nr_failed_migrations_hot);
378         P(se.nr_forced_migrations);
379         P(se.nr_forced2_migrations);
380         P(se.nr_wakeups);
381         P(se.nr_wakeups_sync);
382         P(se.nr_wakeups_migrate);
383         P(se.nr_wakeups_local);
384         P(se.nr_wakeups_remote);
385         P(se.nr_wakeups_affine);
386         P(se.nr_wakeups_affine_attempts);
387         P(se.nr_wakeups_passive);
388         P(se.nr_wakeups_idle);
389
390         {
391                 u64 avg_atom, avg_per_cpu;
392
393                 avg_atom = p->se.sum_exec_runtime;
394                 if (nr_switches)
395                         do_div(avg_atom, nr_switches);
396                 else
397                         avg_atom = -1LL;
398
399                 avg_per_cpu = p->se.sum_exec_runtime;
400                 if (p->se.nr_migrations) {
401                         avg_per_cpu = div64_u64(avg_per_cpu,
402                                                 p->se.nr_migrations);
403                 } else {
404                         avg_per_cpu = -1LL;
405                 }
406
407                 __PN(avg_atom);
408                 __PN(avg_per_cpu);
409         }
410 #endif
411         __P(nr_switches);
412         SEQ_printf(m, "%-35s:%21Ld\n",
413                    "nr_voluntary_switches", (long long)p->nvcsw);
414         SEQ_printf(m, "%-35s:%21Ld\n",
415                    "nr_involuntary_switches", (long long)p->nivcsw);
416
417         P(se.load.weight);
418         P(policy);
419         P(prio);
420 #undef PN
421 #undef __PN
422 #undef P
423 #undef __P
424
425         {
426                 unsigned int this_cpu = raw_smp_processor_id();
427                 u64 t0, t1;
428
429                 t0 = cpu_clock(this_cpu);
430                 t1 = cpu_clock(this_cpu);
431                 SEQ_printf(m, "%-35s:%21Ld\n",
432                            "clock-delta", (long long)(t1-t0));
433         }
434 }
435
436 void proc_sched_set_task(struct task_struct *p)
437 {
438 #ifdef CONFIG_SCHEDSTATS
439         p->se.wait_max                          = 0;
440         p->se.wait_sum                          = 0;
441         p->se.wait_count                        = 0;
442         p->se.sleep_max                         = 0;
443         p->se.sum_sleep_runtime                 = 0;
444         p->se.block_max                         = 0;
445         p->se.exec_max                          = 0;
446         p->se.slice_max                         = 0;
447         p->se.nr_migrations                     = 0;
448         p->se.nr_migrations_cold                = 0;
449         p->se.nr_failed_migrations_affine       = 0;
450         p->se.nr_failed_migrations_running      = 0;
451         p->se.nr_failed_migrations_hot          = 0;
452         p->se.nr_forced_migrations              = 0;
453         p->se.nr_forced2_migrations             = 0;
454         p->se.nr_wakeups                        = 0;
455         p->se.nr_wakeups_sync                   = 0;
456         p->se.nr_wakeups_migrate                = 0;
457         p->se.nr_wakeups_local                  = 0;
458         p->se.nr_wakeups_remote                 = 0;
459         p->se.nr_wakeups_affine                 = 0;
460         p->se.nr_wakeups_affine_attempts        = 0;
461         p->se.nr_wakeups_passive                = 0;
462         p->se.nr_wakeups_idle                   = 0;
463         p->sched_info.bkl_count                 = 0;
464 #endif
465         p->se.sum_exec_runtime                  = 0;
466         p->se.prev_sum_exec_runtime             = 0;
467         p->nvcsw                                = 0;
468         p->nivcsw                               = 0;
469 }