new buffering logic part 1
[cor_2_6_31.git] / kernel / sched_debug.c
blob70c7e0b79946eb7f8f79769e82e87db677c9196a
1 /*
2 * kernel/time/sched_debug.c
4 * Print the CFS rbtree
6 * Copyright(C) 2007, Red Hat, Inc., Ingo Molnar
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.
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>
20 * This allows printing both to /proc/sched_debug and
21 * to the console
23 #define SEQ_printf(m, x...) \
24 do { \
25 if (m) \
26 seq_printf(m, x); \
27 else \
28 printk(x); \
29 } while (0)
32 * Ease the printing of nsec fields:
34 static long long nsec_high(unsigned long long nsec)
36 if ((long long)nsec < 0) {
37 nsec = -nsec;
38 do_div(nsec, 1000000);
39 return -nsec;
41 do_div(nsec, 1000000);
43 return nsec;
46 static unsigned long nsec_low(unsigned long long nsec)
48 if ((long long)nsec < 0)
49 nsec = -nsec;
51 return do_div(nsec, 1000000);
54 #define SPLIT_NS(x) nsec_high(x), nsec_low(x)
56 #ifdef CONFIG_FAIR_GROUP_SCHED
57 static void print_cfs_group_stats(struct seq_file *m, int cpu,
58 struct task_group *tg)
60 struct sched_entity *se = tg->se[cpu];
61 if (!se)
62 return;
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))
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
88 #endif
90 static void
91 print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
93 if (rq->curr == p)
94 SEQ_printf(m, "R");
95 else
96 SEQ_printf(m, " ");
98 SEQ_printf(m, "%15s %5d %9Ld.%06ld %9Ld %5d ",
99 p->comm, p->pid,
100 SPLIT_NS(p->se.vruntime),
101 (long long)(p->nvcsw + p->nivcsw),
102 p->prio);
103 #ifdef CONFIG_SCHEDSTATS
104 SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
105 SPLIT_NS(p->se.vruntime),
106 SPLIT_NS(p->se.sum_exec_runtime),
107 SPLIT_NS(p->se.sum_sleep_runtime));
108 #else
109 SEQ_printf(m, "%15Ld %15Ld %15Ld.%06ld %15Ld.%06ld %15Ld.%06ld",
110 0LL, 0LL, 0LL, 0L, 0LL, 0L, 0LL, 0L);
111 #endif
113 #ifdef CONFIG_CGROUP_SCHED
115 char path[64];
117 cgroup_path(task_group(p)->css.cgroup, path, sizeof(path));
118 SEQ_printf(m, " %s", path);
120 #endif
121 SEQ_printf(m, "\n");
124 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
126 struct task_struct *g, *p;
127 unsigned long flags;
129 SEQ_printf(m,
130 "\nrunnable tasks:\n"
131 " task PID tree-key switches prio"
132 " exec-runtime sum-exec sum-sleep\n"
133 "------------------------------------------------------"
134 "----------------------------------------------------\n");
136 read_lock_irqsave(&tasklist_lock, flags);
138 do_each_thread(g, p) {
139 if (!p->se.on_rq || task_cpu(p) != rq_cpu)
140 continue;
142 print_task(m, rq, p);
143 } while_each_thread(g, p);
145 read_unlock_irqrestore(&tasklist_lock, flags);
148 #if defined(CONFIG_CGROUP_SCHED) && \
149 (defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED))
150 static void task_group_path(struct task_group *tg, char *buf, int buflen)
152 /* may be NULL if the underlying cgroup isn't fully-created yet */
153 if (!tg->css.cgroup) {
154 buf[0] = '\0';
155 return;
157 cgroup_path(tg->css.cgroup, buf, buflen);
159 #endif
161 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
163 s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
164 spread, rq0_min_vruntime, spread0;
165 struct rq *rq = cpu_rq(cpu);
166 struct sched_entity *last;
167 unsigned long flags;
169 #if defined(CONFIG_CGROUP_SCHED) && defined(CONFIG_FAIR_GROUP_SCHED)
170 char path[128];
171 struct task_group *tg = cfs_rq->tg;
173 task_group_path(tg, path, sizeof(path));
175 SEQ_printf(m, "\ncfs_rq[%d]:%s\n", cpu, path);
176 #elif defined(CONFIG_USER_SCHED) && defined(CONFIG_FAIR_GROUP_SCHED)
178 uid_t uid = cfs_rq->tg->uid;
179 SEQ_printf(m, "\ncfs_rq[%d] for UID: %u\n", cpu, uid);
181 #else
182 SEQ_printf(m, "\ncfs_rq[%d]:\n", cpu);
183 #endif
184 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "exec_clock",
185 SPLIT_NS(cfs_rq->exec_clock));
187 spin_lock_irqsave(&rq->lock, flags);
188 if (cfs_rq->rb_leftmost)
189 MIN_vruntime = (__pick_next_entity(cfs_rq))->vruntime;
190 last = __pick_last_entity(cfs_rq);
191 if (last)
192 max_vruntime = last->vruntime;
193 min_vruntime = cfs_rq->min_vruntime;
194 rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
195 spin_unlock_irqrestore(&rq->lock, flags);
196 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "MIN_vruntime",
197 SPLIT_NS(MIN_vruntime));
198 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "min_vruntime",
199 SPLIT_NS(min_vruntime));
200 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "max_vruntime",
201 SPLIT_NS(max_vruntime));
202 spread = max_vruntime - MIN_vruntime;
203 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread",
204 SPLIT_NS(spread));
205 spread0 = min_vruntime - rq0_min_vruntime;
206 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread0",
207 SPLIT_NS(spread0));
208 SEQ_printf(m, " .%-30s: %ld\n", "nr_running", cfs_rq->nr_running);
209 SEQ_printf(m, " .%-30s: %ld\n", "load", cfs_rq->load.weight);
211 SEQ_printf(m, " .%-30s: %d\n", "nr_spread_over",
212 cfs_rq->nr_spread_over);
213 #ifdef CONFIG_FAIR_GROUP_SCHED
214 #ifdef CONFIG_SMP
215 SEQ_printf(m, " .%-30s: %lu\n", "shares", cfs_rq->shares);
216 #endif
217 print_cfs_group_stats(m, cpu, cfs_rq->tg);
218 #endif
221 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
223 #if defined(CONFIG_CGROUP_SCHED) && defined(CONFIG_RT_GROUP_SCHED)
224 char path[128];
225 struct task_group *tg = rt_rq->tg;
227 task_group_path(tg, path, sizeof(path));
229 SEQ_printf(m, "\nrt_rq[%d]:%s\n", cpu, path);
230 #else
231 SEQ_printf(m, "\nrt_rq[%d]:\n", cpu);
232 #endif
235 #define P(x) \
236 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
237 #define PN(x) \
238 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
240 P(rt_nr_running);
241 P(rt_throttled);
242 PN(rt_time);
243 PN(rt_runtime);
245 #undef PN
246 #undef P
249 static void print_cpu(struct seq_file *m, int cpu)
251 struct rq *rq = cpu_rq(cpu);
253 #ifdef CONFIG_X86
255 unsigned int freq = cpu_khz ? : 1;
257 SEQ_printf(m, "\ncpu#%d, %u.%03u MHz\n",
258 cpu, freq / 1000, (freq % 1000));
260 #else
261 SEQ_printf(m, "\ncpu#%d\n", cpu);
262 #endif
264 #define P(x) \
265 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rq->x))
266 #define PN(x) \
267 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
269 P(nr_running);
270 SEQ_printf(m, " .%-30s: %lu\n", "load",
271 rq->load.weight);
272 P(nr_switches);
273 P(nr_load_updates);
274 P(nr_uninterruptible);
275 PN(next_balance);
276 P(curr->pid);
277 PN(clock);
278 P(cpu_load[0]);
279 P(cpu_load[1]);
280 P(cpu_load[2]);
281 P(cpu_load[3]);
282 P(cpu_load[4]);
283 #undef P
284 #undef PN
286 #ifdef CONFIG_SCHEDSTATS
287 #define P(n) SEQ_printf(m, " .%-30s: %d\n", #n, rq->n);
289 P(yld_count);
291 P(sched_switch);
292 P(sched_count);
293 P(sched_goidle);
295 P(ttwu_count);
296 P(ttwu_local);
298 P(bkl_count);
300 #undef P
301 #endif
302 print_cfs_stats(m, cpu);
303 print_rt_stats(m, cpu);
305 print_rq(m, rq, cpu);
308 static int sched_debug_show(struct seq_file *m, void *v)
310 u64 now = ktime_to_ns(ktime_get());
311 int cpu;
313 SEQ_printf(m, "Sched Debug Version: v0.09, %s %.*s\n",
314 init_utsname()->release,
315 (int)strcspn(init_utsname()->version, " "),
316 init_utsname()->version);
318 SEQ_printf(m, "now at %Lu.%06ld msecs\n", SPLIT_NS(now));
320 #define P(x) \
321 SEQ_printf(m, " .%-40s: %Ld\n", #x, (long long)(x))
322 #define PN(x) \
323 SEQ_printf(m, " .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
324 P(jiffies);
325 PN(sysctl_sched_latency);
326 PN(sysctl_sched_min_granularity);
327 PN(sysctl_sched_wakeup_granularity);
328 PN(sysctl_sched_child_runs_first);
329 P(sysctl_sched_features);
330 #undef PN
331 #undef P
333 for_each_online_cpu(cpu)
334 print_cpu(m, cpu);
336 SEQ_printf(m, "\n");
338 return 0;
341 static void sysrq_sched_debug_show(void)
343 sched_debug_show(NULL, NULL);
346 static int sched_debug_open(struct inode *inode, struct file *filp)
348 return single_open(filp, sched_debug_show, NULL);
351 static const struct file_operations sched_debug_fops = {
352 .open = sched_debug_open,
353 .read = seq_read,
354 .llseek = seq_lseek,
355 .release = single_release,
358 static int __init init_sched_debug_procfs(void)
360 struct proc_dir_entry *pe;
362 pe = proc_create("sched_debug", 0444, NULL, &sched_debug_fops);
363 if (!pe)
364 return -ENOMEM;
365 return 0;
368 __initcall(init_sched_debug_procfs);
370 void proc_sched_show_task(struct task_struct *p, struct seq_file *m)
372 unsigned long nr_switches;
373 unsigned long flags;
374 int num_threads = 1;
376 if (lock_task_sighand(p, &flags)) {
377 num_threads = atomic_read(&p->signal->count);
378 unlock_task_sighand(p, &flags);
381 SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, p->pid, num_threads);
382 SEQ_printf(m,
383 "---------------------------------------------------------\n");
384 #define __P(F) \
385 SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)F)
386 #define P(F) \
387 SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)p->F)
388 #define __PN(F) \
389 SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
390 #define PN(F) \
391 SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
393 PN(se.exec_start);
394 PN(se.vruntime);
395 PN(se.sum_exec_runtime);
396 PN(se.avg_overlap);
397 PN(se.avg_wakeup);
399 nr_switches = p->nvcsw + p->nivcsw;
401 #ifdef CONFIG_SCHEDSTATS
402 PN(se.wait_start);
403 PN(se.sleep_start);
404 PN(se.block_start);
405 PN(se.sleep_max);
406 PN(se.block_max);
407 PN(se.exec_max);
408 PN(se.slice_max);
409 PN(se.wait_max);
410 PN(se.wait_sum);
411 P(se.wait_count);
412 P(sched_info.bkl_count);
413 P(se.nr_migrations);
414 P(se.nr_migrations_cold);
415 P(se.nr_failed_migrations_affine);
416 P(se.nr_failed_migrations_running);
417 P(se.nr_failed_migrations_hot);
418 P(se.nr_forced_migrations);
419 P(se.nr_forced2_migrations);
420 P(se.nr_wakeups);
421 P(se.nr_wakeups_sync);
422 P(se.nr_wakeups_migrate);
423 P(se.nr_wakeups_local);
424 P(se.nr_wakeups_remote);
425 P(se.nr_wakeups_affine);
426 P(se.nr_wakeups_affine_attempts);
427 P(se.nr_wakeups_passive);
428 P(se.nr_wakeups_idle);
431 u64 avg_atom, avg_per_cpu;
433 avg_atom = p->se.sum_exec_runtime;
434 if (nr_switches)
435 do_div(avg_atom, nr_switches);
436 else
437 avg_atom = -1LL;
439 avg_per_cpu = p->se.sum_exec_runtime;
440 if (p->se.nr_migrations) {
441 avg_per_cpu = div64_u64(avg_per_cpu,
442 p->se.nr_migrations);
443 } else {
444 avg_per_cpu = -1LL;
447 __PN(avg_atom);
448 __PN(avg_per_cpu);
450 #endif
451 __P(nr_switches);
452 SEQ_printf(m, "%-35s:%21Ld\n",
453 "nr_voluntary_switches", (long long)p->nvcsw);
454 SEQ_printf(m, "%-35s:%21Ld\n",
455 "nr_involuntary_switches", (long long)p->nivcsw);
457 P(se.load.weight);
458 P(policy);
459 P(prio);
460 #undef PN
461 #undef __PN
462 #undef P
463 #undef __P
466 unsigned int this_cpu = raw_smp_processor_id();
467 u64 t0, t1;
469 t0 = cpu_clock(this_cpu);
470 t1 = cpu_clock(this_cpu);
471 SEQ_printf(m, "%-35s:%21Ld\n",
472 "clock-delta", (long long)(t1-t0));
476 void proc_sched_set_task(struct task_struct *p)
478 #ifdef CONFIG_SCHEDSTATS
479 p->se.wait_max = 0;
480 p->se.wait_sum = 0;
481 p->se.wait_count = 0;
482 p->se.sleep_max = 0;
483 p->se.sum_sleep_runtime = 0;
484 p->se.block_max = 0;
485 p->se.exec_max = 0;
486 p->se.slice_max = 0;
487 p->se.nr_migrations = 0;
488 p->se.nr_migrations_cold = 0;
489 p->se.nr_failed_migrations_affine = 0;
490 p->se.nr_failed_migrations_running = 0;
491 p->se.nr_failed_migrations_hot = 0;
492 p->se.nr_forced_migrations = 0;
493 p->se.nr_forced2_migrations = 0;
494 p->se.nr_wakeups = 0;
495 p->se.nr_wakeups_sync = 0;
496 p->se.nr_wakeups_migrate = 0;
497 p->se.nr_wakeups_local = 0;
498 p->se.nr_wakeups_remote = 0;
499 p->se.nr_wakeups_affine = 0;
500 p->se.nr_wakeups_affine_attempts = 0;
501 p->se.nr_wakeups_passive = 0;
502 p->se.nr_wakeups_idle = 0;
503 p->sched_info.bkl_count = 0;
504 #endif
505 p->se.sum_exec_runtime = 0;
506 p->se.prev_sum_exec_runtime = 0;
507 p->nvcsw = 0;
508 p->nivcsw = 0;