sched: _cpu_down(): Don't play with current->cpus_allowed
[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                 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 #else
177         SEQ_printf(m, "\ncfs_rq[%d]:\n", cpu);
178 #endif
179         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "exec_clock",
180                         SPLIT_NS(cfs_rq->exec_clock));
181
182         raw_spin_lock_irqsave(&rq->lock, flags);
183         if (cfs_rq->rb_leftmost)
184                 MIN_vruntime = (__pick_next_entity(cfs_rq))->vruntime;
185         last = __pick_last_entity(cfs_rq);
186         if (last)
187                 max_vruntime = last->vruntime;
188         min_vruntime = cfs_rq->min_vruntime;
189         rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
190         raw_spin_unlock_irqrestore(&rq->lock, flags);
191         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "MIN_vruntime",
192                         SPLIT_NS(MIN_vruntime));
193         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "min_vruntime",
194                         SPLIT_NS(min_vruntime));
195         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "max_vruntime",
196                         SPLIT_NS(max_vruntime));
197         spread = max_vruntime - MIN_vruntime;
198         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread",
199                         SPLIT_NS(spread));
200         spread0 = min_vruntime - rq0_min_vruntime;
201         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread0",
202                         SPLIT_NS(spread0));
203         SEQ_printf(m, "  .%-30s: %ld\n", "nr_running", cfs_rq->nr_running);
204         SEQ_printf(m, "  .%-30s: %ld\n", "load", cfs_rq->load.weight);
205
206         SEQ_printf(m, "  .%-30s: %d\n", "nr_spread_over",
207                         cfs_rq->nr_spread_over);
208 #ifdef CONFIG_FAIR_GROUP_SCHED
209 #ifdef CONFIG_SMP
210         SEQ_printf(m, "  .%-30s: %lu\n", "shares", cfs_rq->shares);
211 #endif
212         print_cfs_group_stats(m, cpu, cfs_rq->tg);
213 #endif
214 }
215
216 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
217 {
218 #if defined(CONFIG_CGROUP_SCHED) && defined(CONFIG_RT_GROUP_SCHED)
219         char path[128];
220         struct task_group *tg = rt_rq->tg;
221
222         task_group_path(tg, path, sizeof(path));
223
224         SEQ_printf(m, "\nrt_rq[%d]:%s\n", cpu, path);
225 #else
226         SEQ_printf(m, "\nrt_rq[%d]:\n", cpu);
227 #endif
228
229
230 #define P(x) \
231         SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
232 #define PN(x) \
233         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
234
235         P(rt_nr_running);
236         P(rt_throttled);
237         PN(rt_time);
238         PN(rt_runtime);
239
240 #undef PN
241 #undef P
242 }
243
244 static void print_cpu(struct seq_file *m, int cpu)
245 {
246         struct rq *rq = cpu_rq(cpu);
247
248 #ifdef CONFIG_X86
249         {
250                 unsigned int freq = cpu_khz ? : 1;
251
252                 SEQ_printf(m, "\ncpu#%d, %u.%03u MHz\n",
253                            cpu, freq / 1000, (freq % 1000));
254         }
255 #else
256         SEQ_printf(m, "\ncpu#%d\n", cpu);
257 #endif
258
259 #define P(x) \
260         SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rq->x))
261 #define PN(x) \
262         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
263
264         P(nr_running);
265         SEQ_printf(m, "  .%-30s: %lu\n", "load",
266                    rq->load.weight);
267         P(nr_switches);
268         P(nr_load_updates);
269         P(nr_uninterruptible);
270         PN(next_balance);
271         P(curr->pid);
272         PN(clock);
273         P(cpu_load[0]);
274         P(cpu_load[1]);
275         P(cpu_load[2]);
276         P(cpu_load[3]);
277         P(cpu_load[4]);
278 #undef P
279 #undef PN
280
281 #ifdef CONFIG_SCHEDSTATS
282 #define P(n) SEQ_printf(m, "  .%-30s: %d\n", #n, rq->n);
283 #define P64(n) SEQ_printf(m, "  .%-30s: %Ld\n", #n, rq->n);
284
285         P(yld_count);
286
287         P(sched_switch);
288         P(sched_count);
289         P(sched_goidle);
290 #ifdef CONFIG_SMP
291         P64(avg_idle);
292 #endif
293
294         P(ttwu_count);
295         P(ttwu_local);
296
297         P(bkl_count);
298
299 #undef P
300 #endif
301         print_cfs_stats(m, cpu);
302         print_rt_stats(m, cpu);
303
304         print_rq(m, rq, cpu);
305 }
306
307 static const char *sched_tunable_scaling_names[] = {
308         "none",
309         "logaritmic",
310         "linear"
311 };
312
313 static int sched_debug_show(struct seq_file *m, void *v)
314 {
315         u64 now = ktime_to_ns(ktime_get());
316         int cpu;
317
318         SEQ_printf(m, "Sched Debug Version: v0.09, %s %.*s\n",
319                 init_utsname()->release,
320                 (int)strcspn(init_utsname()->version, " "),
321                 init_utsname()->version);
322
323         SEQ_printf(m, "now at %Lu.%06ld msecs\n", SPLIT_NS(now));
324
325 #define P(x) \
326         SEQ_printf(m, "  .%-40s: %Ld\n", #x, (long long)(x))
327 #define PN(x) \
328         SEQ_printf(m, "  .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
329         P(jiffies);
330         PN(sysctl_sched_latency);
331         PN(sysctl_sched_min_granularity);
332         PN(sysctl_sched_wakeup_granularity);
333         PN(sysctl_sched_child_runs_first);
334         P(sysctl_sched_features);
335 #undef PN
336 #undef P
337
338         SEQ_printf(m, "  .%-40s: %d (%s)\n", "sysctl_sched_tunable_scaling",
339                 sysctl_sched_tunable_scaling,
340                 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
341
342         for_each_online_cpu(cpu)
343                 print_cpu(m, cpu);
344
345         SEQ_printf(m, "\n");
346
347         return 0;
348 }
349
350 static void sysrq_sched_debug_show(void)
351 {
352         sched_debug_show(NULL, NULL);
353 }
354
355 static int sched_debug_open(struct inode *inode, struct file *filp)
356 {
357         return single_open(filp, sched_debug_show, NULL);
358 }
359
360 static const struct file_operations sched_debug_fops = {
361         .open           = sched_debug_open,
362         .read           = seq_read,
363         .llseek         = seq_lseek,
364         .release        = single_release,
365 };
366
367 static int __init init_sched_debug_procfs(void)
368 {
369         struct proc_dir_entry *pe;
370
371         pe = proc_create("sched_debug", 0444, NULL, &sched_debug_fops);
372         if (!pe)
373                 return -ENOMEM;
374         return 0;
375 }
376
377 __initcall(init_sched_debug_procfs);
378
379 void proc_sched_show_task(struct task_struct *p, struct seq_file *m)
380 {
381         unsigned long nr_switches;
382         unsigned long flags;
383         int num_threads = 1;
384
385         if (lock_task_sighand(p, &flags)) {
386                 num_threads = atomic_read(&p->signal->count);
387                 unlock_task_sighand(p, &flags);
388         }
389
390         SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, p->pid, num_threads);
391         SEQ_printf(m,
392                 "---------------------------------------------------------\n");
393 #define __P(F) \
394         SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)F)
395 #define P(F) \
396         SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)p->F)
397 #define __PN(F) \
398         SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
399 #define PN(F) \
400         SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
401
402         PN(se.exec_start);
403         PN(se.vruntime);
404         PN(se.sum_exec_runtime);
405
406         nr_switches = p->nvcsw + p->nivcsw;
407
408 #ifdef CONFIG_SCHEDSTATS
409         PN(se.statistics.wait_start);
410         PN(se.statistics.sleep_start);
411         PN(se.statistics.block_start);
412         PN(se.statistics.sleep_max);
413         PN(se.statistics.block_max);
414         PN(se.statistics.exec_max);
415         PN(se.statistics.slice_max);
416         PN(se.statistics.wait_max);
417         PN(se.statistics.wait_sum);
418         P(se.statistics.wait_count);
419         PN(se.statistics.iowait_sum);
420         P(se.statistics.iowait_count);
421         P(sched_info.bkl_count);
422         P(se.nr_migrations);
423         P(se.statistics.nr_migrations_cold);
424         P(se.statistics.nr_failed_migrations_affine);
425         P(se.statistics.nr_failed_migrations_running);
426         P(se.statistics.nr_failed_migrations_hot);
427         P(se.statistics.nr_forced_migrations);
428         P(se.statistics.nr_wakeups);
429         P(se.statistics.nr_wakeups_sync);
430         P(se.statistics.nr_wakeups_migrate);
431         P(se.statistics.nr_wakeups_local);
432         P(se.statistics.nr_wakeups_remote);
433         P(se.statistics.nr_wakeups_affine);
434         P(se.statistics.nr_wakeups_affine_attempts);
435         P(se.statistics.nr_wakeups_passive);
436         P(se.statistics.nr_wakeups_idle);
437
438         {
439                 u64 avg_atom, avg_per_cpu;
440
441                 avg_atom = p->se.sum_exec_runtime;
442                 if (nr_switches)
443                         do_div(avg_atom, nr_switches);
444                 else
445                         avg_atom = -1LL;
446
447                 avg_per_cpu = p->se.sum_exec_runtime;
448                 if (p->se.nr_migrations) {
449                         avg_per_cpu = div64_u64(avg_per_cpu,
450                                                 p->se.nr_migrations);
451                 } else {
452                         avg_per_cpu = -1LL;
453                 }
454
455                 __PN(avg_atom);
456                 __PN(avg_per_cpu);
457         }
458 #endif
459         __P(nr_switches);
460         SEQ_printf(m, "%-35s:%21Ld\n",
461                    "nr_voluntary_switches", (long long)p->nvcsw);
462         SEQ_printf(m, "%-35s:%21Ld\n",
463                    "nr_involuntary_switches", (long long)p->nivcsw);
464
465         P(se.load.weight);
466         P(policy);
467         P(prio);
468 #undef PN
469 #undef __PN
470 #undef P
471 #undef __P
472
473         {
474                 unsigned int this_cpu = raw_smp_processor_id();
475                 u64 t0, t1;
476
477                 t0 = cpu_clock(this_cpu);
478                 t1 = cpu_clock(this_cpu);
479                 SEQ_printf(m, "%-35s:%21Ld\n",
480                            "clock-delta", (long long)(t1-t0));
481         }
482 }
483
484 void proc_sched_set_task(struct task_struct *p)
485 {
486 #ifdef CONFIG_SCHEDSTATS
487         memset(&p->se.statistics, 0, sizeof(p->se.statistics));
488 #endif
489         p->se.sum_exec_runtime                  = 0;
490         p->se.prev_sum_exec_runtime             = 0;
491         p->nvcsw                                = 0;
492         p->nivcsw                               = 0;
493 }