sched: debug: add some debug code to handle the full hierarchy
[safe/jmp/linux-2.6] / 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 #ifndef CONFIG_CGROUP_SCHED
123         SEQ_printf(m, "\ncfs_rq[%d]:\n", cpu);
124 #else
125         char path[128] = "";
126         struct cgroup *cgroup = NULL;
127         struct task_group *tg = cfs_rq->tg;
128
129         if (tg)
130                 cgroup = tg->css.cgroup;
131
132         if (cgroup)
133                 cgroup_path(cgroup, path, sizeof(path));
134
135         SEQ_printf(m, "\ncfs_rq[%d]:%s\n", cpu, path);
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 #ifdef CONFIG_FAIR_GROUP_SCHED
171 #ifdef CONFIG_SMP
172         SEQ_printf(m, "  .%-30s: %lu\n", "shares", cfs_rq->shares);
173 #endif
174 #endif
175 }
176
177 static void print_cpu(struct seq_file *m, int cpu)
178 {
179         struct rq *rq = &per_cpu(runqueues, cpu);
180
181 #ifdef CONFIG_X86
182         {
183                 unsigned int freq = cpu_khz ? : 1;
184
185                 SEQ_printf(m, "\ncpu#%d, %u.%03u MHz\n",
186                            cpu, freq / 1000, (freq % 1000));
187         }
188 #else
189         SEQ_printf(m, "\ncpu#%d\n", cpu);
190 #endif
191
192 #define P(x) \
193         SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rq->x))
194 #define PN(x) \
195         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
196
197         P(nr_running);
198         SEQ_printf(m, "  .%-30s: %lu\n", "load",
199                    rq->load.weight);
200         P(nr_switches);
201         P(nr_load_updates);
202         P(nr_uninterruptible);
203         SEQ_printf(m, "  .%-30s: %lu\n", "jiffies", jiffies);
204         PN(next_balance);
205         P(curr->pid);
206         PN(clock);
207         PN(idle_clock);
208         PN(prev_clock_raw);
209         P(clock_warps);
210         P(clock_overflows);
211         P(clock_underflows);
212         P(clock_deep_idle_events);
213         PN(clock_max_delta);
214         P(cpu_load[0]);
215         P(cpu_load[1]);
216         P(cpu_load[2]);
217         P(cpu_load[3]);
218         P(cpu_load[4]);
219 #undef P
220 #undef PN
221
222         print_cfs_stats(m, cpu);
223
224         print_rq(m, rq, cpu);
225 }
226
227 static int sched_debug_show(struct seq_file *m, void *v)
228 {
229         u64 now = ktime_to_ns(ktime_get());
230         int cpu;
231
232         SEQ_printf(m, "Sched Debug Version: v0.07, %s %.*s\n",
233                 init_utsname()->release,
234                 (int)strcspn(init_utsname()->version, " "),
235                 init_utsname()->version);
236
237         SEQ_printf(m, "now at %Lu.%06ld msecs\n", SPLIT_NS(now));
238
239 #define P(x) \
240         SEQ_printf(m, "  .%-40s: %Ld\n", #x, (long long)(x))
241 #define PN(x) \
242         SEQ_printf(m, "  .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
243         PN(sysctl_sched_latency);
244         PN(sysctl_sched_min_granularity);
245         PN(sysctl_sched_wakeup_granularity);
246         PN(sysctl_sched_child_runs_first);
247         P(sysctl_sched_features);
248 #undef PN
249 #undef P
250
251         for_each_online_cpu(cpu)
252                 print_cpu(m, cpu);
253
254         SEQ_printf(m, "\n");
255
256         return 0;
257 }
258
259 static void sysrq_sched_debug_show(void)
260 {
261         sched_debug_show(NULL, NULL);
262 }
263
264 static int sched_debug_open(struct inode *inode, struct file *filp)
265 {
266         return single_open(filp, sched_debug_show, NULL);
267 }
268
269 static const struct file_operations sched_debug_fops = {
270         .open           = sched_debug_open,
271         .read           = seq_read,
272         .llseek         = seq_lseek,
273         .release        = single_release,
274 };
275
276 static int __init init_sched_debug_procfs(void)
277 {
278         struct proc_dir_entry *pe;
279
280         pe = create_proc_entry("sched_debug", 0644, NULL);
281         if (!pe)
282                 return -ENOMEM;
283
284         pe->proc_fops = &sched_debug_fops;
285
286         return 0;
287 }
288
289 __initcall(init_sched_debug_procfs);
290
291 void proc_sched_show_task(struct task_struct *p, struct seq_file *m)
292 {
293         unsigned long nr_switches;
294         unsigned long flags;
295         int num_threads = 1;
296
297         rcu_read_lock();
298         if (lock_task_sighand(p, &flags)) {
299                 num_threads = atomic_read(&p->signal->count);
300                 unlock_task_sighand(p, &flags);
301         }
302         rcu_read_unlock();
303
304         SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, p->pid, num_threads);
305         SEQ_printf(m,
306                 "---------------------------------------------------------\n");
307 #define __P(F) \
308         SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)F)
309 #define P(F) \
310         SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)p->F)
311 #define __PN(F) \
312         SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
313 #define PN(F) \
314         SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
315
316         PN(se.exec_start);
317         PN(se.vruntime);
318         PN(se.sum_exec_runtime);
319         PN(se.avg_overlap);
320
321         nr_switches = p->nvcsw + p->nivcsw;
322
323 #ifdef CONFIG_SCHEDSTATS
324         PN(se.wait_start);
325         PN(se.sleep_start);
326         PN(se.block_start);
327         PN(se.sleep_max);
328         PN(se.block_max);
329         PN(se.exec_max);
330         PN(se.slice_max);
331         PN(se.wait_max);
332         PN(se.wait_sum);
333         P(se.wait_count);
334         P(sched_info.bkl_count);
335         P(se.nr_migrations);
336         P(se.nr_migrations_cold);
337         P(se.nr_failed_migrations_affine);
338         P(se.nr_failed_migrations_running);
339         P(se.nr_failed_migrations_hot);
340         P(se.nr_forced_migrations);
341         P(se.nr_forced2_migrations);
342         P(se.nr_wakeups);
343         P(se.nr_wakeups_sync);
344         P(se.nr_wakeups_migrate);
345         P(se.nr_wakeups_local);
346         P(se.nr_wakeups_remote);
347         P(se.nr_wakeups_affine);
348         P(se.nr_wakeups_affine_attempts);
349         P(se.nr_wakeups_passive);
350         P(se.nr_wakeups_idle);
351
352         {
353                 u64 avg_atom, avg_per_cpu;
354
355                 avg_atom = p->se.sum_exec_runtime;
356                 if (nr_switches)
357                         do_div(avg_atom, nr_switches);
358                 else
359                         avg_atom = -1LL;
360
361                 avg_per_cpu = p->se.sum_exec_runtime;
362                 if (p->se.nr_migrations) {
363                         avg_per_cpu = div64_64(avg_per_cpu,
364                                                p->se.nr_migrations);
365                 } else {
366                         avg_per_cpu = -1LL;
367                 }
368
369                 __PN(avg_atom);
370                 __PN(avg_per_cpu);
371         }
372 #endif
373         __P(nr_switches);
374         SEQ_printf(m, "%-35s:%21Ld\n",
375                    "nr_voluntary_switches", (long long)p->nvcsw);
376         SEQ_printf(m, "%-35s:%21Ld\n",
377                    "nr_involuntary_switches", (long long)p->nivcsw);
378
379         P(se.load.weight);
380         P(policy);
381         P(prio);
382 #undef PN
383 #undef __PN
384 #undef P
385 #undef __P
386
387         {
388                 u64 t0, t1;
389
390                 t0 = sched_clock();
391                 t1 = sched_clock();
392                 SEQ_printf(m, "%-35s:%21Ld\n",
393                            "clock-delta", (long long)(t1-t0));
394         }
395 }
396
397 void proc_sched_set_task(struct task_struct *p)
398 {
399 #ifdef CONFIG_SCHEDSTATS
400         p->se.wait_max                          = 0;
401         p->se.wait_sum                          = 0;
402         p->se.wait_count                        = 0;
403         p->se.sleep_max                         = 0;
404         p->se.sum_sleep_runtime                 = 0;
405         p->se.block_max                         = 0;
406         p->se.exec_max                          = 0;
407         p->se.slice_max                         = 0;
408         p->se.nr_migrations                     = 0;
409         p->se.nr_migrations_cold                = 0;
410         p->se.nr_failed_migrations_affine       = 0;
411         p->se.nr_failed_migrations_running      = 0;
412         p->se.nr_failed_migrations_hot          = 0;
413         p->se.nr_forced_migrations              = 0;
414         p->se.nr_forced2_migrations             = 0;
415         p->se.nr_wakeups                        = 0;
416         p->se.nr_wakeups_sync                   = 0;
417         p->se.nr_wakeups_migrate                = 0;
418         p->se.nr_wakeups_local                  = 0;
419         p->se.nr_wakeups_remote                 = 0;
420         p->se.nr_wakeups_affine                 = 0;
421         p->se.nr_wakeups_affine_attempts        = 0;
422         p->se.nr_wakeups_passive                = 0;
423         p->se.nr_wakeups_idle                   = 0;
424         p->sched_info.bkl_count                 = 0;
425 #endif
426         p->se.sum_exec_runtime                  = 0;
427         p->se.prev_sum_exec_runtime             = 0;
428         p->nvcsw                                = 0;
429         p->nivcsw                               = 0;
430 }