sched: Fix /proc/sched_stat failure on very very large systems
[linux-2.6-block.git] / kernel / sched / debug.c
CommitLineData
43ae34cb 1/*
391e43da 2 * kernel/sched/debug.c
43ae34cb
IM
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
029632fb
PZ
19#include "sched.h"
20
efe25c2c
BR
21static DEFINE_SPINLOCK(sched_debug_lock);
22
43ae34cb
IM
23/*
24 * This allows printing both to /proc/sched_debug and
25 * to the console
26 */
27#define SEQ_printf(m, x...) \
28 do { \
29 if (m) \
30 seq_printf(m, x); \
31 else \
32 printk(x); \
33 } while (0)
34
ef83a571
IM
35/*
36 * Ease the printing of nsec fields:
37 */
90b2628f 38static long long nsec_high(unsigned long long nsec)
ef83a571 39{
90b2628f 40 if ((long long)nsec < 0) {
ef83a571
IM
41 nsec = -nsec;
42 do_div(nsec, 1000000);
43 return -nsec;
44 }
45 do_div(nsec, 1000000);
46
47 return nsec;
48}
49
90b2628f 50static unsigned long nsec_low(unsigned long long nsec)
ef83a571 51{
90b2628f 52 if ((long long)nsec < 0)
ef83a571
IM
53 nsec = -nsec;
54
55 return do_div(nsec, 1000000);
56}
57
58#define SPLIT_NS(x) nsec_high(x), nsec_low(x)
59
ff9b48c3 60#ifdef CONFIG_FAIR_GROUP_SCHED
5091faa4 61static void print_cfs_group_stats(struct seq_file *m, int cpu, struct task_group *tg)
ff9b48c3
BR
62{
63 struct sched_entity *se = tg->se[cpu];
ff9b48c3
BR
64
65#define P(F) \
66 SEQ_printf(m, " .%-30s: %lld\n", #F, (long long)F)
67#define PN(F) \
68 SEQ_printf(m, " .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F))
69
18bf2805
BS
70 if (!se) {
71 struct sched_avg *avg = &cpu_rq(cpu)->avg;
72 P(avg->runnable_avg_sum);
73 P(avg->runnable_avg_period);
74 return;
75 }
76
77
ff9b48c3
BR
78 PN(se->exec_start);
79 PN(se->vruntime);
80 PN(se->sum_exec_runtime);
81#ifdef CONFIG_SCHEDSTATS
41acab88
LDM
82 PN(se->statistics.wait_start);
83 PN(se->statistics.sleep_start);
84 PN(se->statistics.block_start);
85 PN(se->statistics.sleep_max);
86 PN(se->statistics.block_max);
87 PN(se->statistics.exec_max);
88 PN(se->statistics.slice_max);
89 PN(se->statistics.wait_max);
90 PN(se->statistics.wait_sum);
91 P(se->statistics.wait_count);
ff9b48c3
BR
92#endif
93 P(se->load.weight);
9d85f21c
PT
94#ifdef CONFIG_SMP
95 P(se->avg.runnable_avg_sum);
96 P(se->avg.runnable_avg_period);
2dac754e 97 P(se->avg.load_avg_contrib);
9ee474f5 98 P(se->avg.decay_count);
9d85f21c 99#endif
ff9b48c3
BR
100#undef PN
101#undef P
102}
103#endif
104
efe25c2c
BR
105#ifdef CONFIG_CGROUP_SCHED
106static char group_path[PATH_MAX];
107
108static char *task_group_path(struct task_group *tg)
109{
8ecedd7a
BR
110 if (autogroup_path(tg, group_path, PATH_MAX))
111 return group_path;
112
efe25c2c
BR
113 /*
114 * May be NULL if the underlying cgroup isn't fully-created yet
115 */
116 if (!tg->css.cgroup) {
117 group_path[0] = '\0';
118 return group_path;
119 }
120 cgroup_path(tg->css.cgroup, group_path, PATH_MAX);
121 return group_path;
122}
123#endif
124
43ae34cb 125static void
a48da48b 126print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
43ae34cb
IM
127{
128 if (rq->curr == p)
129 SEQ_printf(m, "R");
130 else
131 SEQ_printf(m, " ");
132
ef83a571 133 SEQ_printf(m, "%15s %5d %9Ld.%06ld %9Ld %5d ",
43ae34cb 134 p->comm, p->pid,
ef83a571 135 SPLIT_NS(p->se.vruntime),
43ae34cb 136 (long long)(p->nvcsw + p->nivcsw),
6f605d83 137 p->prio);
6cfb0d5d 138#ifdef CONFIG_SCHEDSTATS
d19ca308 139 SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
ef83a571
IM
140 SPLIT_NS(p->se.vruntime),
141 SPLIT_NS(p->se.sum_exec_runtime),
41acab88 142 SPLIT_NS(p->se.statistics.sum_sleep_runtime));
6cfb0d5d 143#else
d19ca308 144 SEQ_printf(m, "%15Ld %15Ld %15Ld.%06ld %15Ld.%06ld %15Ld.%06ld",
ef83a571 145 0LL, 0LL, 0LL, 0L, 0LL, 0L, 0LL, 0L);
6cfb0d5d 146#endif
efe25c2c
BR
147#ifdef CONFIG_CGROUP_SCHED
148 SEQ_printf(m, " %s", task_group_path(task_group(p)));
149#endif
d19ca308 150
d19ca308 151 SEQ_printf(m, "\n");
43ae34cb
IM
152}
153
a48da48b 154static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
43ae34cb
IM
155{
156 struct task_struct *g, *p;
ab63a633 157 unsigned long flags;
43ae34cb
IM
158
159 SEQ_printf(m,
160 "\nrunnable tasks:\n"
c86da3a3
MG
161 " task PID tree-key switches prio"
162 " exec-runtime sum-exec sum-sleep\n"
1a75b94f 163 "------------------------------------------------------"
c86da3a3 164 "----------------------------------------------------\n");
43ae34cb 165
ab63a633 166 read_lock_irqsave(&tasklist_lock, flags);
43ae34cb
IM
167
168 do_each_thread(g, p) {
fd2f4419 169 if (!p->on_rq || task_cpu(p) != rq_cpu)
43ae34cb
IM
170 continue;
171
a48da48b 172 print_task(m, rq, p);
43ae34cb
IM
173 } while_each_thread(g, p);
174
ab63a633 175 read_unlock_irqrestore(&tasklist_lock, flags);
43ae34cb
IM
176}
177
5cef9eca 178void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
43ae34cb 179{
86d9560c
IM
180 s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
181 spread, rq0_min_vruntime, spread0;
348ec61e 182 struct rq *rq = cpu_rq(cpu);
67e12eac
IM
183 struct sched_entity *last;
184 unsigned long flags;
185
efe25c2c
BR
186#ifdef CONFIG_FAIR_GROUP_SCHED
187 SEQ_printf(m, "\ncfs_rq[%d]:%s\n", cpu, task_group_path(cfs_rq->tg));
188#else
ada18de2 189 SEQ_printf(m, "\ncfs_rq[%d]:\n", cpu);
efe25c2c 190#endif
ef83a571
IM
191 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "exec_clock",
192 SPLIT_NS(cfs_rq->exec_clock));
67e12eac 193
05fa785c 194 raw_spin_lock_irqsave(&rq->lock, flags);
67e12eac 195 if (cfs_rq->rb_leftmost)
ac53db59 196 MIN_vruntime = (__pick_first_entity(cfs_rq))->vruntime;
67e12eac
IM
197 last = __pick_last_entity(cfs_rq);
198 if (last)
199 max_vruntime = last->vruntime;
5ac5c4d6 200 min_vruntime = cfs_rq->min_vruntime;
348ec61e 201 rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
05fa785c 202 raw_spin_unlock_irqrestore(&rq->lock, flags);
ef83a571
IM
203 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "MIN_vruntime",
204 SPLIT_NS(MIN_vruntime));
205 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "min_vruntime",
206 SPLIT_NS(min_vruntime));
207 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "max_vruntime",
208 SPLIT_NS(max_vruntime));
67e12eac 209 spread = max_vruntime - MIN_vruntime;
ef83a571
IM
210 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread",
211 SPLIT_NS(spread));
86d9560c 212 spread0 = min_vruntime - rq0_min_vruntime;
ef83a571
IM
213 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread0",
214 SPLIT_NS(spread0));
5ac5c4d6 215 SEQ_printf(m, " .%-30s: %d\n", "nr_spread_over",
ddc97297 216 cfs_rq->nr_spread_over);
c82513e5 217 SEQ_printf(m, " .%-30s: %d\n", "nr_running", cfs_rq->nr_running);
2069dd75 218 SEQ_printf(m, " .%-30s: %ld\n", "load", cfs_rq->load.weight);
c09595f6
PZ
219#ifdef CONFIG_FAIR_GROUP_SCHED
220#ifdef CONFIG_SMP
2dac754e
PT
221 SEQ_printf(m, " .%-30s: %lld\n", "runnable_load_avg",
222 cfs_rq->runnable_load_avg);
9ee474f5
PT
223 SEQ_printf(m, " .%-30s: %lld\n", "blocked_load_avg",
224 cfs_rq->blocked_load_avg);
cff3c124
AB
225 SEQ_printf(m, " .%-30s: %lld\n", "tg_load_avg",
226 (unsigned long long)atomic64_read(&cfs_rq->tg->load_avg));
c566e8e9
PT
227 SEQ_printf(m, " .%-30s: %lld\n", "tg_load_contrib",
228 cfs_rq->tg_load_contrib);
bb17f655
PT
229 SEQ_printf(m, " .%-30s: %d\n", "tg_runnable_contrib",
230 cfs_rq->tg_runnable_contrib);
231 SEQ_printf(m, " .%-30s: %d\n", "tg->runnable_avg",
232 atomic_read(&cfs_rq->tg->runnable_avg));
c09595f6 233#endif
2069dd75 234
ff9b48c3 235 print_cfs_group_stats(m, cpu, cfs_rq->tg);
c09595f6 236#endif
43ae34cb
IM
237}
238
ada18de2
PZ
239void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
240{
efe25c2c
BR
241#ifdef CONFIG_RT_GROUP_SCHED
242 SEQ_printf(m, "\nrt_rq[%d]:%s\n", cpu, task_group_path(rt_rq->tg));
243#else
ada18de2 244 SEQ_printf(m, "\nrt_rq[%d]:\n", cpu);
efe25c2c 245#endif
ada18de2
PZ
246
247#define P(x) \
248 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
249#define PN(x) \
250 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
251
252 P(rt_nr_running);
253 P(rt_throttled);
254 PN(rt_time);
255 PN(rt_runtime);
256
257#undef PN
258#undef P
259}
260
5bb6b1ea
PZ
261extern __read_mostly int sched_clock_running;
262
a48da48b 263static void print_cpu(struct seq_file *m, int cpu)
43ae34cb 264{
348ec61e 265 struct rq *rq = cpu_rq(cpu);
efe25c2c 266 unsigned long flags;
43ae34cb
IM
267
268#ifdef CONFIG_X86
269 {
270 unsigned int freq = cpu_khz ? : 1;
271
272 SEQ_printf(m, "\ncpu#%d, %u.%03u MHz\n",
273 cpu, freq / 1000, (freq % 1000));
274 }
275#else
276 SEQ_printf(m, "\ncpu#%d\n", cpu);
277#endif
278
13e099d2
PZ
279#define P(x) \
280do { \
281 if (sizeof(rq->x) == 4) \
282 SEQ_printf(m, " .%-30s: %ld\n", #x, (long)(rq->x)); \
283 else \
284 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rq->x));\
285} while (0)
286
ef83a571
IM
287#define PN(x) \
288 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
43ae34cb
IM
289
290 P(nr_running);
291 SEQ_printf(m, " .%-30s: %lu\n", "load",
495eca49 292 rq->load.weight);
43ae34cb
IM
293 P(nr_switches);
294 P(nr_load_updates);
295 P(nr_uninterruptible);
ef83a571 296 PN(next_balance);
43ae34cb 297 P(curr->pid);
ef83a571 298 PN(clock);
43ae34cb
IM
299 P(cpu_load[0]);
300 P(cpu_load[1]);
301 P(cpu_load[2]);
302 P(cpu_load[3]);
303 P(cpu_load[4]);
304#undef P
ef83a571 305#undef PN
43ae34cb 306
5ac5c4d6
PZ
307#ifdef CONFIG_SCHEDSTATS
308#define P(n) SEQ_printf(m, " .%-30s: %d\n", #n, rq->n);
1b9508f6 309#define P64(n) SEQ_printf(m, " .%-30s: %Ld\n", #n, rq->n);
5ac5c4d6 310
5ac5c4d6
PZ
311 P(yld_count);
312
5ac5c4d6
PZ
313 P(sched_count);
314 P(sched_goidle);
1b9508f6
MG
315#ifdef CONFIG_SMP
316 P64(avg_idle);
317#endif
5ac5c4d6
PZ
318
319 P(ttwu_count);
320 P(ttwu_local);
321
5ac5c4d6 322#undef P
fce20979 323#undef P64
5ac5c4d6 324#endif
efe25c2c 325 spin_lock_irqsave(&sched_debug_lock, flags);
5cef9eca 326 print_cfs_stats(m, cpu);
ada18de2 327 print_rt_stats(m, cpu);
43ae34cb 328
efe25c2c 329 rcu_read_lock();
a48da48b 330 print_rq(m, rq, cpu);
efe25c2c
BR
331 rcu_read_unlock();
332 spin_unlock_irqrestore(&sched_debug_lock, flags);
43ae34cb
IM
333}
334
1983a922
CE
335static const char *sched_tunable_scaling_names[] = {
336 "none",
337 "logaritmic",
338 "linear"
339};
340
43ae34cb
IM
341static int sched_debug_show(struct seq_file *m, void *v)
342{
5bb6b1ea
PZ
343 u64 ktime, sched_clk, cpu_clk;
344 unsigned long flags;
43ae34cb
IM
345 int cpu;
346
5bb6b1ea
PZ
347 local_irq_save(flags);
348 ktime = ktime_to_ns(ktime_get());
349 sched_clk = sched_clock();
350 cpu_clk = local_clock();
351 local_irq_restore(flags);
352
353 SEQ_printf(m, "Sched Debug Version: v0.10, %s %.*s\n",
43ae34cb
IM
354 init_utsname()->release,
355 (int)strcspn(init_utsname()->version, " "),
356 init_utsname()->version);
357
5bb6b1ea
PZ
358#define P(x) \
359 SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
360#define PN(x) \
361 SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
362 PN(ktime);
363 PN(sched_clk);
364 PN(cpu_clk);
365 P(jiffies);
366#ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
367 P(sched_clock_stable);
368#endif
369#undef PN
370#undef P
371
372 SEQ_printf(m, "\n");
373 SEQ_printf(m, "sysctl_sched\n");
43ae34cb 374
1aa4731e 375#define P(x) \
d822cece 376 SEQ_printf(m, " .%-40s: %Ld\n", #x, (long long)(x))
1aa4731e 377#define PN(x) \
d822cece 378 SEQ_printf(m, " .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
1aa4731e 379 PN(sysctl_sched_latency);
b2be5e96 380 PN(sysctl_sched_min_granularity);
1aa4731e 381 PN(sysctl_sched_wakeup_granularity);
eebef746 382 P(sysctl_sched_child_runs_first);
1aa4731e
IM
383 P(sysctl_sched_features);
384#undef PN
385#undef P
386
1983a922
CE
387 SEQ_printf(m, " .%-40s: %d (%s)\n", "sysctl_sched_tunable_scaling",
388 sysctl_sched_tunable_scaling,
389 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
390
43ae34cb 391 for_each_online_cpu(cpu)
a48da48b 392 print_cpu(m, cpu);
43ae34cb
IM
393
394 SEQ_printf(m, "\n");
395
396 return 0;
397}
398
029632fb 399void sysrq_sched_debug_show(void)
43ae34cb
IM
400{
401 sched_debug_show(NULL, NULL);
402}
403
404static int sched_debug_open(struct inode *inode, struct file *filp)
405{
406 return single_open(filp, sched_debug_show, NULL);
407}
408
0dbee3a6 409static const struct file_operations sched_debug_fops = {
43ae34cb
IM
410 .open = sched_debug_open,
411 .read = seq_read,
412 .llseek = seq_lseek,
5ea473a1 413 .release = single_release,
43ae34cb
IM
414};
415
416static int __init init_sched_debug_procfs(void)
417{
418 struct proc_dir_entry *pe;
419
a9cf4ddb 420 pe = proc_create("sched_debug", 0444, NULL, &sched_debug_fops);
43ae34cb
IM
421 if (!pe)
422 return -ENOMEM;
43ae34cb
IM
423 return 0;
424}
425
426__initcall(init_sched_debug_procfs);
427
428void proc_sched_show_task(struct task_struct *p, struct seq_file *m)
429{
cc367732 430 unsigned long nr_switches;
43ae34cb 431
5089a976
ON
432 SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, p->pid,
433 get_nr_threads(p));
2d92f227
IM
434 SEQ_printf(m,
435 "---------------------------------------------------------\n");
cc367732
IM
436#define __P(F) \
437 SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)F)
43ae34cb 438#define P(F) \
2d92f227 439 SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)p->F)
cc367732
IM
440#define __PN(F) \
441 SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
ef83a571 442#define PN(F) \
2d92f227 443 SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
43ae34cb 444
ef83a571
IM
445 PN(se.exec_start);
446 PN(se.vruntime);
447 PN(se.sum_exec_runtime);
6cfb0d5d 448
cc367732
IM
449 nr_switches = p->nvcsw + p->nivcsw;
450
6cfb0d5d 451#ifdef CONFIG_SCHEDSTATS
41acab88
LDM
452 PN(se.statistics.wait_start);
453 PN(se.statistics.sleep_start);
454 PN(se.statistics.block_start);
455 PN(se.statistics.sleep_max);
456 PN(se.statistics.block_max);
457 PN(se.statistics.exec_max);
458 PN(se.statistics.slice_max);
459 PN(se.statistics.wait_max);
460 PN(se.statistics.wait_sum);
461 P(se.statistics.wait_count);
462 PN(se.statistics.iowait_sum);
463 P(se.statistics.iowait_count);
cc367732 464 P(se.nr_migrations);
41acab88
LDM
465 P(se.statistics.nr_migrations_cold);
466 P(se.statistics.nr_failed_migrations_affine);
467 P(se.statistics.nr_failed_migrations_running);
468 P(se.statistics.nr_failed_migrations_hot);
469 P(se.statistics.nr_forced_migrations);
470 P(se.statistics.nr_wakeups);
471 P(se.statistics.nr_wakeups_sync);
472 P(se.statistics.nr_wakeups_migrate);
473 P(se.statistics.nr_wakeups_local);
474 P(se.statistics.nr_wakeups_remote);
475 P(se.statistics.nr_wakeups_affine);
476 P(se.statistics.nr_wakeups_affine_attempts);
477 P(se.statistics.nr_wakeups_passive);
478 P(se.statistics.nr_wakeups_idle);
cc367732
IM
479
480 {
481 u64 avg_atom, avg_per_cpu;
482
483 avg_atom = p->se.sum_exec_runtime;
484 if (nr_switches)
485 do_div(avg_atom, nr_switches);
486 else
487 avg_atom = -1LL;
488
489 avg_per_cpu = p->se.sum_exec_runtime;
c1a89740 490 if (p->se.nr_migrations) {
6f6d6a1a
RZ
491 avg_per_cpu = div64_u64(avg_per_cpu,
492 p->se.nr_migrations);
c1a89740 493 } else {
cc367732 494 avg_per_cpu = -1LL;
c1a89740 495 }
cc367732
IM
496
497 __PN(avg_atom);
498 __PN(avg_per_cpu);
499 }
6cfb0d5d 500#endif
cc367732 501 __P(nr_switches);
2d92f227 502 SEQ_printf(m, "%-35s:%21Ld\n",
cc367732
IM
503 "nr_voluntary_switches", (long long)p->nvcsw);
504 SEQ_printf(m, "%-35s:%21Ld\n",
505 "nr_involuntary_switches", (long long)p->nivcsw);
506
43ae34cb
IM
507 P(se.load.weight);
508 P(policy);
509 P(prio);
ef83a571 510#undef PN
cc367732
IM
511#undef __PN
512#undef P
513#undef __P
43ae34cb
IM
514
515 {
29d7b90c 516 unsigned int this_cpu = raw_smp_processor_id();
43ae34cb
IM
517 u64 t0, t1;
518
29d7b90c
IM
519 t0 = cpu_clock(this_cpu);
520 t1 = cpu_clock(this_cpu);
2d92f227 521 SEQ_printf(m, "%-35s:%21Ld\n",
43ae34cb
IM
522 "clock-delta", (long long)(t1-t0));
523 }
524}
525
526void proc_sched_set_task(struct task_struct *p)
527{
6cfb0d5d 528#ifdef CONFIG_SCHEDSTATS
41acab88 529 memset(&p->se.statistics, 0, sizeof(p->se.statistics));
6cfb0d5d 530#endif
43ae34cb 531}