x86/numa: Check for nonsensical topologies on real hw as well
[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];
64 if (!se)
65 return;
66
67#define P(F) \
68 SEQ_printf(m, " .%-30s: %lld\n", #F, (long long)F)
69#define PN(F) \
70 SEQ_printf(m, " .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F))
71
72 PN(se->exec_start);
73 PN(se->vruntime);
74 PN(se->sum_exec_runtime);
75#ifdef CONFIG_SCHEDSTATS
41acab88
LDM
76 PN(se->statistics.wait_start);
77 PN(se->statistics.sleep_start);
78 PN(se->statistics.block_start);
79 PN(se->statistics.sleep_max);
80 PN(se->statistics.block_max);
81 PN(se->statistics.exec_max);
82 PN(se->statistics.slice_max);
83 PN(se->statistics.wait_max);
84 PN(se->statistics.wait_sum);
85 P(se->statistics.wait_count);
ff9b48c3
BR
86#endif
87 P(se->load.weight);
88#undef PN
89#undef P
90}
91#endif
92
efe25c2c
BR
93#ifdef CONFIG_CGROUP_SCHED
94static char group_path[PATH_MAX];
95
96static char *task_group_path(struct task_group *tg)
97{
8ecedd7a
BR
98 if (autogroup_path(tg, group_path, PATH_MAX))
99 return group_path;
100
efe25c2c
BR
101 /*
102 * May be NULL if the underlying cgroup isn't fully-created yet
103 */
104 if (!tg->css.cgroup) {
105 group_path[0] = '\0';
106 return group_path;
107 }
108 cgroup_path(tg->css.cgroup, group_path, PATH_MAX);
109 return group_path;
110}
111#endif
112
43ae34cb 113static void
a48da48b 114print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
43ae34cb
IM
115{
116 if (rq->curr == p)
117 SEQ_printf(m, "R");
118 else
119 SEQ_printf(m, " ");
120
ef83a571 121 SEQ_printf(m, "%15s %5d %9Ld.%06ld %9Ld %5d ",
43ae34cb 122 p->comm, p->pid,
ef83a571 123 SPLIT_NS(p->se.vruntime),
43ae34cb 124 (long long)(p->nvcsw + p->nivcsw),
6f605d83 125 p->prio);
6cfb0d5d 126#ifdef CONFIG_SCHEDSTATS
d19ca308 127 SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
ef83a571
IM
128 SPLIT_NS(p->se.vruntime),
129 SPLIT_NS(p->se.sum_exec_runtime),
41acab88 130 SPLIT_NS(p->se.statistics.sum_sleep_runtime));
6cfb0d5d 131#else
d19ca308 132 SEQ_printf(m, "%15Ld %15Ld %15Ld.%06ld %15Ld.%06ld %15Ld.%06ld",
ef83a571 133 0LL, 0LL, 0LL, 0L, 0LL, 0L, 0LL, 0L);
6cfb0d5d 134#endif
efe25c2c
BR
135#ifdef CONFIG_CGROUP_SCHED
136 SEQ_printf(m, " %s", task_group_path(task_group(p)));
137#endif
d19ca308 138
d19ca308 139 SEQ_printf(m, "\n");
43ae34cb
IM
140}
141
a48da48b 142static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
43ae34cb
IM
143{
144 struct task_struct *g, *p;
ab63a633 145 unsigned long flags;
43ae34cb
IM
146
147 SEQ_printf(m,
148 "\nrunnable tasks:\n"
c86da3a3
MG
149 " task PID tree-key switches prio"
150 " exec-runtime sum-exec sum-sleep\n"
1a75b94f 151 "------------------------------------------------------"
c86da3a3 152 "----------------------------------------------------\n");
43ae34cb 153
ab63a633 154 read_lock_irqsave(&tasklist_lock, flags);
43ae34cb
IM
155
156 do_each_thread(g, p) {
fd2f4419 157 if (!p->on_rq || task_cpu(p) != rq_cpu)
43ae34cb
IM
158 continue;
159
a48da48b 160 print_task(m, rq, p);
43ae34cb
IM
161 } while_each_thread(g, p);
162
ab63a633 163 read_unlock_irqrestore(&tasklist_lock, flags);
43ae34cb
IM
164}
165
5cef9eca 166void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
43ae34cb 167{
86d9560c
IM
168 s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
169 spread, rq0_min_vruntime, spread0;
348ec61e 170 struct rq *rq = cpu_rq(cpu);
67e12eac
IM
171 struct sched_entity *last;
172 unsigned long flags;
173
efe25c2c
BR
174#ifdef CONFIG_FAIR_GROUP_SCHED
175 SEQ_printf(m, "\ncfs_rq[%d]:%s\n", cpu, task_group_path(cfs_rq->tg));
176#else
ada18de2 177 SEQ_printf(m, "\ncfs_rq[%d]:\n", cpu);
efe25c2c 178#endif
ef83a571
IM
179 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "exec_clock",
180 SPLIT_NS(cfs_rq->exec_clock));
67e12eac 181
05fa785c 182 raw_spin_lock_irqsave(&rq->lock, flags);
67e12eac 183 if (cfs_rq->rb_leftmost)
ac53db59 184 MIN_vruntime = (__pick_first_entity(cfs_rq))->vruntime;
67e12eac
IM
185 last = __pick_last_entity(cfs_rq);
186 if (last)
187 max_vruntime = last->vruntime;
5ac5c4d6 188 min_vruntime = cfs_rq->min_vruntime;
348ec61e 189 rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
05fa785c 190 raw_spin_unlock_irqrestore(&rq->lock, flags);
ef83a571
IM
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));
67e12eac 197 spread = max_vruntime - MIN_vruntime;
ef83a571
IM
198 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread",
199 SPLIT_NS(spread));
86d9560c 200 spread0 = min_vruntime - rq0_min_vruntime;
ef83a571
IM
201 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread0",
202 SPLIT_NS(spread0));
5ac5c4d6 203 SEQ_printf(m, " .%-30s: %d\n", "nr_spread_over",
ddc97297 204 cfs_rq->nr_spread_over);
2069dd75
PZ
205 SEQ_printf(m, " .%-30s: %ld\n", "nr_running", cfs_rq->nr_running);
206 SEQ_printf(m, " .%-30s: %ld\n", "load", cfs_rq->load.weight);
c09595f6
PZ
207#ifdef CONFIG_FAIR_GROUP_SCHED
208#ifdef CONFIG_SMP
2069dd75
PZ
209 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "load_avg",
210 SPLIT_NS(cfs_rq->load_avg));
211 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "load_period",
212 SPLIT_NS(cfs_rq->load_period));
213 SEQ_printf(m, " .%-30s: %ld\n", "load_contrib",
214 cfs_rq->load_contribution);
215 SEQ_printf(m, " .%-30s: %d\n", "load_tg",
5091faa4 216 atomic_read(&cfs_rq->tg->load_weight));
c09595f6 217#endif
2069dd75 218
ff9b48c3 219 print_cfs_group_stats(m, cpu, cfs_rq->tg);
c09595f6 220#endif
43ae34cb
IM
221}
222
ada18de2
PZ
223void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
224{
efe25c2c
BR
225#ifdef CONFIG_RT_GROUP_SCHED
226 SEQ_printf(m, "\nrt_rq[%d]:%s\n", cpu, task_group_path(rt_rq->tg));
227#else
ada18de2 228 SEQ_printf(m, "\nrt_rq[%d]:\n", cpu);
efe25c2c 229#endif
ada18de2
PZ
230
231#define P(x) \
232 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
233#define PN(x) \
234 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
235
236 P(rt_nr_running);
237 P(rt_throttled);
238 PN(rt_time);
239 PN(rt_runtime);
240
241#undef PN
242#undef P
243}
244
5bb6b1ea
PZ
245extern __read_mostly int sched_clock_running;
246
a48da48b 247static void print_cpu(struct seq_file *m, int cpu)
43ae34cb 248{
348ec61e 249 struct rq *rq = cpu_rq(cpu);
efe25c2c 250 unsigned long flags;
43ae34cb
IM
251
252#ifdef CONFIG_X86
253 {
254 unsigned int freq = cpu_khz ? : 1;
255
256 SEQ_printf(m, "\ncpu#%d, %u.%03u MHz\n",
257 cpu, freq / 1000, (freq % 1000));
258 }
259#else
260 SEQ_printf(m, "\ncpu#%d\n", cpu);
261#endif
262
263#define P(x) \
264 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rq->x))
ef83a571
IM
265#define PN(x) \
266 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
43ae34cb
IM
267
268 P(nr_running);
269 SEQ_printf(m, " .%-30s: %lu\n", "load",
495eca49 270 rq->load.weight);
43ae34cb
IM
271 P(nr_switches);
272 P(nr_load_updates);
273 P(nr_uninterruptible);
ef83a571 274 PN(next_balance);
43ae34cb 275 P(curr->pid);
ef83a571 276 PN(clock);
43ae34cb
IM
277 P(cpu_load[0]);
278 P(cpu_load[1]);
279 P(cpu_load[2]);
280 P(cpu_load[3]);
281 P(cpu_load[4]);
282#undef P
ef83a571 283#undef PN
43ae34cb 284
5ac5c4d6
PZ
285#ifdef CONFIG_SCHEDSTATS
286#define P(n) SEQ_printf(m, " .%-30s: %d\n", #n, rq->n);
1b9508f6 287#define P64(n) SEQ_printf(m, " .%-30s: %Ld\n", #n, rq->n);
5ac5c4d6 288
5ac5c4d6
PZ
289 P(yld_count);
290
5ac5c4d6
PZ
291 P(sched_count);
292 P(sched_goidle);
1b9508f6
MG
293#ifdef CONFIG_SMP
294 P64(avg_idle);
295#endif
5ac5c4d6
PZ
296
297 P(ttwu_count);
298 P(ttwu_local);
299
5ac5c4d6 300#undef P
fce20979 301#undef P64
5ac5c4d6 302#endif
efe25c2c 303 spin_lock_irqsave(&sched_debug_lock, flags);
5cef9eca 304 print_cfs_stats(m, cpu);
ada18de2 305 print_rt_stats(m, cpu);
43ae34cb 306
efe25c2c 307 rcu_read_lock();
a48da48b 308 print_rq(m, rq, cpu);
efe25c2c
BR
309 rcu_read_unlock();
310 spin_unlock_irqrestore(&sched_debug_lock, flags);
43ae34cb
IM
311}
312
1983a922
CE
313static const char *sched_tunable_scaling_names[] = {
314 "none",
315 "logaritmic",
316 "linear"
317};
318
43ae34cb
IM
319static int sched_debug_show(struct seq_file *m, void *v)
320{
5bb6b1ea
PZ
321 u64 ktime, sched_clk, cpu_clk;
322 unsigned long flags;
43ae34cb
IM
323 int cpu;
324
5bb6b1ea
PZ
325 local_irq_save(flags);
326 ktime = ktime_to_ns(ktime_get());
327 sched_clk = sched_clock();
328 cpu_clk = local_clock();
329 local_irq_restore(flags);
330
331 SEQ_printf(m, "Sched Debug Version: v0.10, %s %.*s\n",
43ae34cb
IM
332 init_utsname()->release,
333 (int)strcspn(init_utsname()->version, " "),
334 init_utsname()->version);
335
5bb6b1ea
PZ
336#define P(x) \
337 SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
338#define PN(x) \
339 SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
340 PN(ktime);
341 PN(sched_clk);
342 PN(cpu_clk);
343 P(jiffies);
344#ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
345 P(sched_clock_stable);
346#endif
347#undef PN
348#undef P
349
350 SEQ_printf(m, "\n");
351 SEQ_printf(m, "sysctl_sched\n");
43ae34cb 352
1aa4731e 353#define P(x) \
d822cece 354 SEQ_printf(m, " .%-40s: %Ld\n", #x, (long long)(x))
1aa4731e 355#define PN(x) \
d822cece 356 SEQ_printf(m, " .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
1aa4731e 357 PN(sysctl_sched_latency);
b2be5e96 358 PN(sysctl_sched_min_granularity);
1aa4731e 359 PN(sysctl_sched_wakeup_granularity);
eebef746 360 P(sysctl_sched_child_runs_first);
1aa4731e
IM
361 P(sysctl_sched_features);
362#undef PN
363#undef P
364
1983a922
CE
365 SEQ_printf(m, " .%-40s: %d (%s)\n", "sysctl_sched_tunable_scaling",
366 sysctl_sched_tunable_scaling,
367 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
368
43ae34cb 369 for_each_online_cpu(cpu)
a48da48b 370 print_cpu(m, cpu);
43ae34cb
IM
371
372 SEQ_printf(m, "\n");
373
374 return 0;
375}
376
029632fb 377void sysrq_sched_debug_show(void)
43ae34cb
IM
378{
379 sched_debug_show(NULL, NULL);
380}
381
382static int sched_debug_open(struct inode *inode, struct file *filp)
383{
384 return single_open(filp, sched_debug_show, NULL);
385}
386
0dbee3a6 387static const struct file_operations sched_debug_fops = {
43ae34cb
IM
388 .open = sched_debug_open,
389 .read = seq_read,
390 .llseek = seq_lseek,
5ea473a1 391 .release = single_release,
43ae34cb
IM
392};
393
394static int __init init_sched_debug_procfs(void)
395{
396 struct proc_dir_entry *pe;
397
a9cf4ddb 398 pe = proc_create("sched_debug", 0444, NULL, &sched_debug_fops);
43ae34cb
IM
399 if (!pe)
400 return -ENOMEM;
43ae34cb
IM
401 return 0;
402}
403
404__initcall(init_sched_debug_procfs);
405
406void proc_sched_show_task(struct task_struct *p, struct seq_file *m)
407{
cc367732 408 unsigned long nr_switches;
43ae34cb 409
5089a976
ON
410 SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, p->pid,
411 get_nr_threads(p));
2d92f227
IM
412 SEQ_printf(m,
413 "---------------------------------------------------------\n");
cc367732
IM
414#define __P(F) \
415 SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)F)
43ae34cb 416#define P(F) \
2d92f227 417 SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)p->F)
cc367732
IM
418#define __PN(F) \
419 SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
ef83a571 420#define PN(F) \
2d92f227 421 SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
43ae34cb 422
ef83a571
IM
423 PN(se.exec_start);
424 PN(se.vruntime);
425 PN(se.sum_exec_runtime);
6cfb0d5d 426
cc367732
IM
427 nr_switches = p->nvcsw + p->nivcsw;
428
6cfb0d5d 429#ifdef CONFIG_SCHEDSTATS
41acab88
LDM
430 PN(se.statistics.wait_start);
431 PN(se.statistics.sleep_start);
432 PN(se.statistics.block_start);
433 PN(se.statistics.sleep_max);
434 PN(se.statistics.block_max);
435 PN(se.statistics.exec_max);
436 PN(se.statistics.slice_max);
437 PN(se.statistics.wait_max);
438 PN(se.statistics.wait_sum);
439 P(se.statistics.wait_count);
440 PN(se.statistics.iowait_sum);
441 P(se.statistics.iowait_count);
cc367732 442 P(se.nr_migrations);
41acab88
LDM
443 P(se.statistics.nr_migrations_cold);
444 P(se.statistics.nr_failed_migrations_affine);
445 P(se.statistics.nr_failed_migrations_running);
446 P(se.statistics.nr_failed_migrations_hot);
447 P(se.statistics.nr_forced_migrations);
448 P(se.statistics.nr_wakeups);
449 P(se.statistics.nr_wakeups_sync);
450 P(se.statistics.nr_wakeups_migrate);
451 P(se.statistics.nr_wakeups_local);
452 P(se.statistics.nr_wakeups_remote);
453 P(se.statistics.nr_wakeups_affine);
454 P(se.statistics.nr_wakeups_affine_attempts);
455 P(se.statistics.nr_wakeups_passive);
456 P(se.statistics.nr_wakeups_idle);
cc367732
IM
457
458 {
459 u64 avg_atom, avg_per_cpu;
460
461 avg_atom = p->se.sum_exec_runtime;
462 if (nr_switches)
463 do_div(avg_atom, nr_switches);
464 else
465 avg_atom = -1LL;
466
467 avg_per_cpu = p->se.sum_exec_runtime;
c1a89740 468 if (p->se.nr_migrations) {
6f6d6a1a
RZ
469 avg_per_cpu = div64_u64(avg_per_cpu,
470 p->se.nr_migrations);
c1a89740 471 } else {
cc367732 472 avg_per_cpu = -1LL;
c1a89740 473 }
cc367732
IM
474
475 __PN(avg_atom);
476 __PN(avg_per_cpu);
477 }
6cfb0d5d 478#endif
cc367732 479 __P(nr_switches);
2d92f227 480 SEQ_printf(m, "%-35s:%21Ld\n",
cc367732
IM
481 "nr_voluntary_switches", (long long)p->nvcsw);
482 SEQ_printf(m, "%-35s:%21Ld\n",
483 "nr_involuntary_switches", (long long)p->nivcsw);
484
43ae34cb
IM
485 P(se.load.weight);
486 P(policy);
487 P(prio);
ef83a571 488#undef PN
cc367732
IM
489#undef __PN
490#undef P
491#undef __P
43ae34cb
IM
492
493 {
29d7b90c 494 unsigned int this_cpu = raw_smp_processor_id();
43ae34cb
IM
495 u64 t0, t1;
496
29d7b90c
IM
497 t0 = cpu_clock(this_cpu);
498 t1 = cpu_clock(this_cpu);
2d92f227 499 SEQ_printf(m, "%-35s:%21Ld\n",
43ae34cb
IM
500 "clock-delta", (long long)(t1-t0));
501 }
502}
503
504void proc_sched_set_task(struct task_struct *p)
505{
6cfb0d5d 506#ifdef CONFIG_SCHEDSTATS
41acab88 507 memset(&p->se.statistics, 0, sizeof(p->se.statistics));
6cfb0d5d 508#endif
43ae34cb 509}