sched: reintroduce the sched_min_granularity tunable
[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(long long nsec)
35 {
36         if (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(long long nsec)
47 {
48         if (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\n",
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\n",
76                 0LL, 0LL, 0LL, 0L, 0LL, 0L, 0LL, 0L);
77 #endif
78 }
79
80 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
81 {
82         struct task_struct *g, *p;
83         unsigned long flags;
84
85         SEQ_printf(m,
86         "\nrunnable tasks:\n"
87         "            task   PID         tree-key  switches  prio"
88         "     exec-runtime         sum-exec        sum-sleep\n"
89         "------------------------------------------------------"
90         "----------------------------------------------------\n");
91
92         read_lock_irqsave(&tasklist_lock, flags);
93
94         do_each_thread(g, p) {
95                 if (!p->se.on_rq || task_cpu(p) != rq_cpu)
96                         continue;
97
98                 print_task(m, rq, p);
99         } while_each_thread(g, p);
100
101         read_unlock_irqrestore(&tasklist_lock, flags);
102 }
103
104 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
105 {
106         s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
107                 spread, rq0_min_vruntime, spread0;
108         struct rq *rq = &per_cpu(runqueues, cpu);
109         struct sched_entity *last;
110         unsigned long flags;
111
112         SEQ_printf(m, "\ncfs_rq\n");
113
114         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "exec_clock",
115                         SPLIT_NS(cfs_rq->exec_clock));
116
117         spin_lock_irqsave(&rq->lock, flags);
118         if (cfs_rq->rb_leftmost)
119                 MIN_vruntime = (__pick_next_entity(cfs_rq))->vruntime;
120         last = __pick_last_entity(cfs_rq);
121         if (last)
122                 max_vruntime = last->vruntime;
123         min_vruntime = rq->cfs.min_vruntime;
124         rq0_min_vruntime = per_cpu(runqueues, 0).cfs.min_vruntime;
125         spin_unlock_irqrestore(&rq->lock, flags);
126         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "MIN_vruntime",
127                         SPLIT_NS(MIN_vruntime));
128         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "min_vruntime",
129                         SPLIT_NS(min_vruntime));
130         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "max_vruntime",
131                         SPLIT_NS(max_vruntime));
132         spread = max_vruntime - MIN_vruntime;
133         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread",
134                         SPLIT_NS(spread));
135         spread0 = min_vruntime - rq0_min_vruntime;
136         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread0",
137                         SPLIT_NS(spread0));
138         SEQ_printf(m, "  .%-30s: %ld\n", "nr_running", cfs_rq->nr_running);
139         SEQ_printf(m, "  .%-30s: %ld\n", "load", cfs_rq->load.weight);
140 #ifdef CONFIG_SCHEDSTATS
141         SEQ_printf(m, "  .%-30s: %d\n", "bkl_count",
142                         rq->bkl_count);
143 #endif
144         SEQ_printf(m, "  .%-30s: %ld\n", "nr_spread_over",
145                         cfs_rq->nr_spread_over);
146 }
147
148 static void print_cpu(struct seq_file *m, int cpu)
149 {
150         struct rq *rq = &per_cpu(runqueues, cpu);
151
152 #ifdef CONFIG_X86
153         {
154                 unsigned int freq = cpu_khz ? : 1;
155
156                 SEQ_printf(m, "\ncpu#%d, %u.%03u MHz\n",
157                            cpu, freq / 1000, (freq % 1000));
158         }
159 #else
160         SEQ_printf(m, "\ncpu#%d\n", cpu);
161 #endif
162
163 #define P(x) \
164         SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rq->x))
165 #define PN(x) \
166         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
167
168         P(nr_running);
169         SEQ_printf(m, "  .%-30s: %lu\n", "load",
170                    rq->load.weight);
171         P(nr_switches);
172         P(nr_load_updates);
173         P(nr_uninterruptible);
174         SEQ_printf(m, "  .%-30s: %lu\n", "jiffies", jiffies);
175         PN(next_balance);
176         P(curr->pid);
177         PN(clock);
178         PN(idle_clock);
179         PN(prev_clock_raw);
180         P(clock_warps);
181         P(clock_overflows);
182         P(clock_deep_idle_events);
183         PN(clock_max_delta);
184         P(cpu_load[0]);
185         P(cpu_load[1]);
186         P(cpu_load[2]);
187         P(cpu_load[3]);
188         P(cpu_load[4]);
189 #undef P
190 #undef PN
191
192         print_cfs_stats(m, cpu);
193
194         print_rq(m, rq, cpu);
195 }
196
197 static int sched_debug_show(struct seq_file *m, void *v)
198 {
199         u64 now = ktime_to_ns(ktime_get());
200         int cpu;
201
202         SEQ_printf(m, "Sched Debug Version: v0.06-v22, %s %.*s\n",
203                 init_utsname()->release,
204                 (int)strcspn(init_utsname()->version, " "),
205                 init_utsname()->version);
206
207         SEQ_printf(m, "now at %Lu.%06ld msecs\n", SPLIT_NS(now));
208
209 #define P(x) \
210         SEQ_printf(m, "  .%-40s: %Ld\n", #x, (long long)(x))
211 #define PN(x) \
212         SEQ_printf(m, "  .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
213         PN(sysctl_sched_latency);
214         PN(sysctl_sched_min_granularity);
215         PN(sysctl_sched_wakeup_granularity);
216         PN(sysctl_sched_batch_wakeup_granularity);
217         PN(sysctl_sched_child_runs_first);
218         P(sysctl_sched_features);
219 #undef PN
220 #undef P
221
222         for_each_online_cpu(cpu)
223                 print_cpu(m, cpu);
224
225         SEQ_printf(m, "\n");
226
227         return 0;
228 }
229
230 static void sysrq_sched_debug_show(void)
231 {
232         sched_debug_show(NULL, NULL);
233 }
234
235 static int sched_debug_open(struct inode *inode, struct file *filp)
236 {
237         return single_open(filp, sched_debug_show, NULL);
238 }
239
240 static const struct file_operations sched_debug_fops = {
241         .open           = sched_debug_open,
242         .read           = seq_read,
243         .llseek         = seq_lseek,
244         .release        = single_release,
245 };
246
247 static int __init init_sched_debug_procfs(void)
248 {
249         struct proc_dir_entry *pe;
250
251         pe = create_proc_entry("sched_debug", 0644, NULL);
252         if (!pe)
253                 return -ENOMEM;
254
255         pe->proc_fops = &sched_debug_fops;
256
257         return 0;
258 }
259
260 __initcall(init_sched_debug_procfs);
261
262 void proc_sched_show_task(struct task_struct *p, struct seq_file *m)
263 {
264         unsigned long nr_switches;
265         unsigned long flags;
266         int num_threads = 1;
267
268         rcu_read_lock();
269         if (lock_task_sighand(p, &flags)) {
270                 num_threads = atomic_read(&p->signal->count);
271                 unlock_task_sighand(p, &flags);
272         }
273         rcu_read_unlock();
274
275         SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, p->pid, num_threads);
276         SEQ_printf(m,
277                 "---------------------------------------------------------\n");
278 #define __P(F) \
279         SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)F)
280 #define P(F) \
281         SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)p->F)
282 #define __PN(F) \
283         SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
284 #define PN(F) \
285         SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
286
287         PN(se.exec_start);
288         PN(se.vruntime);
289         PN(se.sum_exec_runtime);
290
291         nr_switches = p->nvcsw + p->nivcsw;
292
293 #ifdef CONFIG_SCHEDSTATS
294         PN(se.wait_start);
295         PN(se.sleep_start);
296         PN(se.block_start);
297         PN(se.sleep_max);
298         PN(se.block_max);
299         PN(se.exec_max);
300         PN(se.slice_max);
301         PN(se.wait_max);
302         P(sched_info.bkl_count);
303         P(se.nr_migrations);
304         P(se.nr_migrations_cold);
305         P(se.nr_failed_migrations_affine);
306         P(se.nr_failed_migrations_running);
307         P(se.nr_failed_migrations_hot);
308         P(se.nr_forced_migrations);
309         P(se.nr_forced2_migrations);
310         P(se.nr_wakeups);
311         P(se.nr_wakeups_sync);
312         P(se.nr_wakeups_migrate);
313         P(se.nr_wakeups_local);
314         P(se.nr_wakeups_remote);
315         P(se.nr_wakeups_affine);
316         P(se.nr_wakeups_affine_attempts);
317         P(se.nr_wakeups_passive);
318         P(se.nr_wakeups_idle);
319
320         {
321                 u64 avg_atom, avg_per_cpu;
322
323                 avg_atom = p->se.sum_exec_runtime;
324                 if (nr_switches)
325                         do_div(avg_atom, nr_switches);
326                 else
327                         avg_atom = -1LL;
328
329                 avg_per_cpu = p->se.sum_exec_runtime;
330                 if (p->se.nr_migrations)
331                         avg_per_cpu = div64_64(avg_per_cpu, p->se.nr_migrations);
332                 else
333                         avg_per_cpu = -1LL;
334
335                 __PN(avg_atom);
336                 __PN(avg_per_cpu);
337         }
338 #endif
339         __P(nr_switches);
340         SEQ_printf(m, "%-35s:%21Ld\n",
341                    "nr_voluntary_switches", (long long)p->nvcsw);
342         SEQ_printf(m, "%-35s:%21Ld\n",
343                    "nr_involuntary_switches", (long long)p->nivcsw);
344
345         P(se.load.weight);
346         P(policy);
347         P(prio);
348 #undef PN
349 #undef __PN
350 #undef P
351 #undef __P
352
353         {
354                 u64 t0, t1;
355
356                 t0 = sched_clock();
357                 t1 = sched_clock();
358                 SEQ_printf(m, "%-35s:%21Ld\n",
359                            "clock-delta", (long long)(t1-t0));
360         }
361 }
362
363 void proc_sched_set_task(struct task_struct *p)
364 {
365 #ifdef CONFIG_SCHEDSTATS
366         p->se.wait_max                          = 0;
367         p->se.sleep_max                         = 0;
368         p->se.sum_sleep_runtime                 = 0;
369         p->se.block_max                         = 0;
370         p->se.exec_max                          = 0;
371         p->se.slice_max                         = 0;
372         p->se.nr_migrations                     = 0;
373         p->se.nr_migrations_cold                = 0;
374         p->se.nr_failed_migrations_affine       = 0;
375         p->se.nr_failed_migrations_running      = 0;
376         p->se.nr_failed_migrations_hot          = 0;
377         p->se.nr_forced_migrations              = 0;
378         p->se.nr_forced2_migrations             = 0;
379         p->se.nr_wakeups                        = 0;
380         p->se.nr_wakeups_sync                   = 0;
381         p->se.nr_wakeups_migrate                = 0;
382         p->se.nr_wakeups_local                  = 0;
383         p->se.nr_wakeups_remote                 = 0;
384         p->se.nr_wakeups_affine                 = 0;
385         p->se.nr_wakeups_affine_attempts        = 0;
386         p->se.nr_wakeups_passive                = 0;
387         p->se.nr_wakeups_idle                   = 0;
388         p->sched_info.bkl_count                 = 0;
389 #endif
390         p->se.sum_exec_runtime                  = 0;
391         p->se.prev_sum_exec_runtime             = 0;
392         p->nvcsw                                = 0;
393         p->nivcsw                               = 0;
394 }