sched: move double_unlock_balance() higher
[linux-2.6-block.git] / kernel / sched_debug.c
CommitLineData
43ae34cb
IM
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
ef83a571
IM
31/*
32 * Ease the printing of nsec fields:
33 */
90b2628f 34static long long nsec_high(unsigned long long nsec)
ef83a571 35{
90b2628f 36 if ((long long)nsec < 0) {
ef83a571
IM
37 nsec = -nsec;
38 do_div(nsec, 1000000);
39 return -nsec;
40 }
41 do_div(nsec, 1000000);
42
43 return nsec;
44}
45
90b2628f 46static unsigned long nsec_low(unsigned long long nsec)
ef83a571 47{
90b2628f 48 if ((long long)nsec < 0)
ef83a571
IM
49 nsec = -nsec;
50
51 return do_div(nsec, 1000000);
52}
53
54#define SPLIT_NS(x) nsec_high(x), nsec_low(x)
55
ff9b48c3
BR
56#ifdef CONFIG_FAIR_GROUP_SCHED
57static 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
43ae34cb 90static void
a48da48b 91print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
43ae34cb
IM
92{
93 if (rq->curr == p)
94 SEQ_printf(m, "R");
95 else
96 SEQ_printf(m, " ");
97
ef83a571 98 SEQ_printf(m, "%15s %5d %9Ld.%06ld %9Ld %5d ",
43ae34cb 99 p->comm, p->pid,
ef83a571 100 SPLIT_NS(p->se.vruntime),
43ae34cb 101 (long long)(p->nvcsw + p->nivcsw),
6f605d83 102 p->prio);
6cfb0d5d 103#ifdef CONFIG_SCHEDSTATS
d19ca308 104 SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
ef83a571
IM
105 SPLIT_NS(p->se.vruntime),
106 SPLIT_NS(p->se.sum_exec_runtime),
107 SPLIT_NS(p->se.sum_sleep_runtime));
6cfb0d5d 108#else
d19ca308 109 SEQ_printf(m, "%15Ld %15Ld %15Ld.%06ld %15Ld.%06ld %15Ld.%06ld",
ef83a571 110 0LL, 0LL, 0LL, 0L, 0LL, 0L, 0LL, 0L);
6cfb0d5d 111#endif
d19ca308
PZ
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");
43ae34cb
IM
122}
123
a48da48b 124static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
43ae34cb
IM
125{
126 struct task_struct *g, *p;
ab63a633 127 unsigned long flags;
43ae34cb
IM
128
129 SEQ_printf(m,
130 "\nrunnable tasks:\n"
c86da3a3
MG
131 " task PID tree-key switches prio"
132 " exec-runtime sum-exec sum-sleep\n"
1a75b94f 133 "------------------------------------------------------"
c86da3a3 134 "----------------------------------------------------\n");
43ae34cb 135
ab63a633 136 read_lock_irqsave(&tasklist_lock, flags);
43ae34cb
IM
137
138 do_each_thread(g, p) {
139 if (!p->se.on_rq || task_cpu(p) != rq_cpu)
140 continue;
141
a48da48b 142 print_task(m, rq, p);
43ae34cb
IM
143 } while_each_thread(g, p);
144
ab63a633 145 read_unlock_irqrestore(&tasklist_lock, flags);
43ae34cb
IM
146}
147
5cef9eca 148void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
43ae34cb 149{
86d9560c
IM
150 s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
151 spread, rq0_min_vruntime, spread0;
67e12eac
IM
152 struct rq *rq = &per_cpu(runqueues, cpu);
153 struct sched_entity *last;
154 unsigned long flags;
155
ada18de2 156#if defined(CONFIG_CGROUP_SCHED) && defined(CONFIG_FAIR_GROUP_SCHED)
d19ca308 157 char path[128] = "";
d19ca308
PZ
158 struct task_group *tg = cfs_rq->tg;
159
0a0db8f5 160 cgroup_path(tg->css.cgroup, path, sizeof(path));
d19ca308
PZ
161
162 SEQ_printf(m, "\ncfs_rq[%d]:%s\n", cpu, path);
ada18de2
PZ
163#else
164 SEQ_printf(m, "\ncfs_rq[%d]:\n", cpu);
d19ca308 165#endif
43ae34cb 166
ef83a571
IM
167 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "exec_clock",
168 SPLIT_NS(cfs_rq->exec_clock));
67e12eac
IM
169
170 spin_lock_irqsave(&rq->lock, flags);
171 if (cfs_rq->rb_leftmost)
172 MIN_vruntime = (__pick_next_entity(cfs_rq))->vruntime;
173 last = __pick_last_entity(cfs_rq);
174 if (last)
175 max_vruntime = last->vruntime;
5ac5c4d6 176 min_vruntime = cfs_rq->min_vruntime;
86d9560c 177 rq0_min_vruntime = per_cpu(runqueues, 0).cfs.min_vruntime;
67e12eac 178 spin_unlock_irqrestore(&rq->lock, flags);
ef83a571
IM
179 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "MIN_vruntime",
180 SPLIT_NS(MIN_vruntime));
181 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "min_vruntime",
182 SPLIT_NS(min_vruntime));
183 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "max_vruntime",
184 SPLIT_NS(max_vruntime));
67e12eac 185 spread = max_vruntime - MIN_vruntime;
ef83a571
IM
186 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread",
187 SPLIT_NS(spread));
86d9560c 188 spread0 = min_vruntime - rq0_min_vruntime;
ef83a571
IM
189 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread0",
190 SPLIT_NS(spread0));
545f3b18
SV
191 SEQ_printf(m, " .%-30s: %ld\n", "nr_running", cfs_rq->nr_running);
192 SEQ_printf(m, " .%-30s: %ld\n", "load", cfs_rq->load.weight);
32df2ee8 193
5ac5c4d6 194 SEQ_printf(m, " .%-30s: %d\n", "nr_spread_over",
ddc97297 195 cfs_rq->nr_spread_over);
c09595f6
PZ
196#ifdef CONFIG_FAIR_GROUP_SCHED
197#ifdef CONFIG_SMP
198 SEQ_printf(m, " .%-30s: %lu\n", "shares", cfs_rq->shares);
199#endif
ff9b48c3 200 print_cfs_group_stats(m, cpu, cfs_rq->tg);
c09595f6 201#endif
43ae34cb
IM
202}
203
ada18de2
PZ
204void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
205{
206#if defined(CONFIG_CGROUP_SCHED) && defined(CONFIG_RT_GROUP_SCHED)
207 char path[128] = "";
ada18de2
PZ
208 struct task_group *tg = rt_rq->tg;
209
0a0db8f5 210 cgroup_path(tg->css.cgroup, path, sizeof(path));
ada18de2
PZ
211
212 SEQ_printf(m, "\nrt_rq[%d]:%s\n", cpu, path);
213#else
214 SEQ_printf(m, "\nrt_rq[%d]:\n", cpu);
215#endif
216
217
218#define P(x) \
219 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
220#define PN(x) \
221 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
222
223 P(rt_nr_running);
224 P(rt_throttled);
225 PN(rt_time);
226 PN(rt_runtime);
227
228#undef PN
229#undef P
230}
231
a48da48b 232static void print_cpu(struct seq_file *m, int cpu)
43ae34cb
IM
233{
234 struct rq *rq = &per_cpu(runqueues, cpu);
235
236#ifdef CONFIG_X86
237 {
238 unsigned int freq = cpu_khz ? : 1;
239
240 SEQ_printf(m, "\ncpu#%d, %u.%03u MHz\n",
241 cpu, freq / 1000, (freq % 1000));
242 }
243#else
244 SEQ_printf(m, "\ncpu#%d\n", cpu);
245#endif
246
247#define P(x) \
248 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rq->x))
ef83a571
IM
249#define PN(x) \
250 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
43ae34cb
IM
251
252 P(nr_running);
253 SEQ_printf(m, " .%-30s: %lu\n", "load",
495eca49 254 rq->load.weight);
43ae34cb
IM
255 P(nr_switches);
256 P(nr_load_updates);
257 P(nr_uninterruptible);
258 SEQ_printf(m, " .%-30s: %lu\n", "jiffies", jiffies);
ef83a571 259 PN(next_balance);
43ae34cb 260 P(curr->pid);
ef83a571 261 PN(clock);
43ae34cb
IM
262 P(cpu_load[0]);
263 P(cpu_load[1]);
264 P(cpu_load[2]);
265 P(cpu_load[3]);
266 P(cpu_load[4]);
267#undef P
ef83a571 268#undef PN
43ae34cb 269
5ac5c4d6
PZ
270#ifdef CONFIG_SCHEDSTATS
271#define P(n) SEQ_printf(m, " .%-30s: %d\n", #n, rq->n);
272
273 P(yld_exp_empty);
274 P(yld_act_empty);
275 P(yld_both_empty);
276 P(yld_count);
277
278 P(sched_switch);
279 P(sched_count);
280 P(sched_goidle);
281
282 P(ttwu_count);
283 P(ttwu_local);
284
285 P(bkl_count);
286
287#undef P
288#endif
5cef9eca 289 print_cfs_stats(m, cpu);
ada18de2 290 print_rt_stats(m, cpu);
43ae34cb 291
a48da48b 292 print_rq(m, rq, cpu);
43ae34cb
IM
293}
294
295static int sched_debug_show(struct seq_file *m, void *v)
296{
297 u64 now = ktime_to_ns(ktime_get());
298 int cpu;
299
ff9b48c3 300 SEQ_printf(m, "Sched Debug Version: v0.08, %s %.*s\n",
43ae34cb
IM
301 init_utsname()->release,
302 (int)strcspn(init_utsname()->version, " "),
303 init_utsname()->version);
304
ef83a571 305 SEQ_printf(m, "now at %Lu.%06ld msecs\n", SPLIT_NS(now));
43ae34cb 306
1aa4731e 307#define P(x) \
d822cece 308 SEQ_printf(m, " .%-40s: %Ld\n", #x, (long long)(x))
1aa4731e 309#define PN(x) \
d822cece 310 SEQ_printf(m, " .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
1aa4731e 311 PN(sysctl_sched_latency);
b2be5e96 312 PN(sysctl_sched_min_granularity);
1aa4731e 313 PN(sysctl_sched_wakeup_granularity);
1aa4731e
IM
314 PN(sysctl_sched_child_runs_first);
315 P(sysctl_sched_features);
316#undef PN
317#undef P
318
43ae34cb 319 for_each_online_cpu(cpu)
a48da48b 320 print_cpu(m, cpu);
43ae34cb
IM
321
322 SEQ_printf(m, "\n");
323
324 return 0;
325}
326
f3373461 327static void sysrq_sched_debug_show(void)
43ae34cb
IM
328{
329 sched_debug_show(NULL, NULL);
330}
331
332static int sched_debug_open(struct inode *inode, struct file *filp)
333{
334 return single_open(filp, sched_debug_show, NULL);
335}
336
0dbee3a6 337static const struct file_operations sched_debug_fops = {
43ae34cb
IM
338 .open = sched_debug_open,
339 .read = seq_read,
340 .llseek = seq_lseek,
5ea473a1 341 .release = single_release,
43ae34cb
IM
342};
343
344static int __init init_sched_debug_procfs(void)
345{
346 struct proc_dir_entry *pe;
347
a9cf4ddb 348 pe = proc_create("sched_debug", 0444, NULL, &sched_debug_fops);
43ae34cb
IM
349 if (!pe)
350 return -ENOMEM;
43ae34cb
IM
351 return 0;
352}
353
354__initcall(init_sched_debug_procfs);
355
356void proc_sched_show_task(struct task_struct *p, struct seq_file *m)
357{
cc367732 358 unsigned long nr_switches;
43ae34cb
IM
359 unsigned long flags;
360 int num_threads = 1;
361
43ae34cb
IM
362 if (lock_task_sighand(p, &flags)) {
363 num_threads = atomic_read(&p->signal->count);
364 unlock_task_sighand(p, &flags);
365 }
43ae34cb
IM
366
367 SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, p->pid, num_threads);
2d92f227
IM
368 SEQ_printf(m,
369 "---------------------------------------------------------\n");
cc367732
IM
370#define __P(F) \
371 SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)F)
43ae34cb 372#define P(F) \
2d92f227 373 SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)p->F)
cc367732
IM
374#define __PN(F) \
375 SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
ef83a571 376#define PN(F) \
2d92f227 377 SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
43ae34cb 378
ef83a571
IM
379 PN(se.exec_start);
380 PN(se.vruntime);
381 PN(se.sum_exec_runtime);
4ae7d5ce 382 PN(se.avg_overlap);
6cfb0d5d 383
cc367732
IM
384 nr_switches = p->nvcsw + p->nivcsw;
385
6cfb0d5d 386#ifdef CONFIG_SCHEDSTATS
ef83a571
IM
387 PN(se.wait_start);
388 PN(se.sleep_start);
389 PN(se.block_start);
390 PN(se.sleep_max);
391 PN(se.block_max);
392 PN(se.exec_max);
393 PN(se.slice_max);
394 PN(se.wait_max);
6d082592
AV
395 PN(se.wait_sum);
396 P(se.wait_count);
2d72376b 397 P(sched_info.bkl_count);
cc367732
IM
398 P(se.nr_migrations);
399 P(se.nr_migrations_cold);
400 P(se.nr_failed_migrations_affine);
401 P(se.nr_failed_migrations_running);
402 P(se.nr_failed_migrations_hot);
403 P(se.nr_forced_migrations);
404 P(se.nr_forced2_migrations);
405 P(se.nr_wakeups);
406 P(se.nr_wakeups_sync);
407 P(se.nr_wakeups_migrate);
408 P(se.nr_wakeups_local);
409 P(se.nr_wakeups_remote);
410 P(se.nr_wakeups_affine);
411 P(se.nr_wakeups_affine_attempts);
412 P(se.nr_wakeups_passive);
413 P(se.nr_wakeups_idle);
414
415 {
416 u64 avg_atom, avg_per_cpu;
417
418 avg_atom = p->se.sum_exec_runtime;
419 if (nr_switches)
420 do_div(avg_atom, nr_switches);
421 else
422 avg_atom = -1LL;
423
424 avg_per_cpu = p->se.sum_exec_runtime;
c1a89740 425 if (p->se.nr_migrations) {
6f6d6a1a
RZ
426 avg_per_cpu = div64_u64(avg_per_cpu,
427 p->se.nr_migrations);
c1a89740 428 } else {
cc367732 429 avg_per_cpu = -1LL;
c1a89740 430 }
cc367732
IM
431
432 __PN(avg_atom);
433 __PN(avg_per_cpu);
434 }
6cfb0d5d 435#endif
cc367732 436 __P(nr_switches);
2d92f227 437 SEQ_printf(m, "%-35s:%21Ld\n",
cc367732
IM
438 "nr_voluntary_switches", (long long)p->nvcsw);
439 SEQ_printf(m, "%-35s:%21Ld\n",
440 "nr_involuntary_switches", (long long)p->nivcsw);
441
43ae34cb
IM
442 P(se.load.weight);
443 P(policy);
444 P(prio);
ef83a571 445#undef PN
cc367732
IM
446#undef __PN
447#undef P
448#undef __P
43ae34cb
IM
449
450 {
29d7b90c 451 unsigned int this_cpu = raw_smp_processor_id();
43ae34cb
IM
452 u64 t0, t1;
453
29d7b90c
IM
454 t0 = cpu_clock(this_cpu);
455 t1 = cpu_clock(this_cpu);
2d92f227 456 SEQ_printf(m, "%-35s:%21Ld\n",
43ae34cb
IM
457 "clock-delta", (long long)(t1-t0));
458 }
459}
460
461void proc_sched_set_task(struct task_struct *p)
462{
6cfb0d5d 463#ifdef CONFIG_SCHEDSTATS
cc367732 464 p->se.wait_max = 0;
6d082592
AV
465 p->se.wait_sum = 0;
466 p->se.wait_count = 0;
cc367732
IM
467 p->se.sleep_max = 0;
468 p->se.sum_sleep_runtime = 0;
469 p->se.block_max = 0;
470 p->se.exec_max = 0;
471 p->se.slice_max = 0;
472 p->se.nr_migrations = 0;
473 p->se.nr_migrations_cold = 0;
474 p->se.nr_failed_migrations_affine = 0;
475 p->se.nr_failed_migrations_running = 0;
476 p->se.nr_failed_migrations_hot = 0;
477 p->se.nr_forced_migrations = 0;
478 p->se.nr_forced2_migrations = 0;
479 p->se.nr_wakeups = 0;
480 p->se.nr_wakeups_sync = 0;
481 p->se.nr_wakeups_migrate = 0;
482 p->se.nr_wakeups_local = 0;
483 p->se.nr_wakeups_remote = 0;
484 p->se.nr_wakeups_affine = 0;
485 p->se.nr_wakeups_affine_attempts = 0;
486 p->se.nr_wakeups_passive = 0;
487 p->se.nr_wakeups_idle = 0;
488 p->sched_info.bkl_count = 0;
6cfb0d5d 489#endif
cc367732
IM
490 p->se.sum_exec_runtime = 0;
491 p->se.prev_sum_exec_runtime = 0;
492 p->nvcsw = 0;
493 p->nivcsw = 0;
43ae34cb 494}