xref: /openbmc/linux/kernel/sched/debug.c (revision b9ccfda2)
1 /*
2  * kernel/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 #include "sched.h"
20 
21 static DEFINE_SPINLOCK(sched_debug_lock);
22 
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 
35 /*
36  * Ease the printing of nsec fields:
37  */
38 static long long nsec_high(unsigned long long nsec)
39 {
40 	if ((long long)nsec < 0) {
41 		nsec = -nsec;
42 		do_div(nsec, 1000000);
43 		return -nsec;
44 	}
45 	do_div(nsec, 1000000);
46 
47 	return nsec;
48 }
49 
50 static unsigned long nsec_low(unsigned long long nsec)
51 {
52 	if ((long long)nsec < 0)
53 		nsec = -nsec;
54 
55 	return do_div(nsec, 1000000);
56 }
57 
58 #define SPLIT_NS(x) nsec_high(x), nsec_low(x)
59 
60 #ifdef CONFIG_FAIR_GROUP_SCHED
61 static void print_cfs_group_stats(struct seq_file *m, int cpu, struct task_group *tg)
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
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);
86 #endif
87 	P(se->load.weight);
88 #undef PN
89 #undef P
90 }
91 #endif
92 
93 #ifdef CONFIG_CGROUP_SCHED
94 static char group_path[PATH_MAX];
95 
96 static char *task_group_path(struct task_group *tg)
97 {
98 	if (autogroup_path(tg, group_path, PATH_MAX))
99 		return group_path;
100 
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 
113 static void
114 print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
115 {
116 	if (rq->curr == p)
117 		SEQ_printf(m, "R");
118 	else
119 		SEQ_printf(m, " ");
120 
121 	SEQ_printf(m, "%15s %5d %9Ld.%06ld %9Ld %5d ",
122 		p->comm, p->pid,
123 		SPLIT_NS(p->se.vruntime),
124 		(long long)(p->nvcsw + p->nivcsw),
125 		p->prio);
126 #ifdef CONFIG_SCHEDSTATS
127 	SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
128 		SPLIT_NS(p->se.vruntime),
129 		SPLIT_NS(p->se.sum_exec_runtime),
130 		SPLIT_NS(p->se.statistics.sum_sleep_runtime));
131 #else
132 	SEQ_printf(m, "%15Ld %15Ld %15Ld.%06ld %15Ld.%06ld %15Ld.%06ld",
133 		0LL, 0LL, 0LL, 0L, 0LL, 0L, 0LL, 0L);
134 #endif
135 #ifdef CONFIG_CGROUP_SCHED
136 	SEQ_printf(m, " %s", task_group_path(task_group(p)));
137 #endif
138 
139 	SEQ_printf(m, "\n");
140 }
141 
142 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
143 {
144 	struct task_struct *g, *p;
145 	unsigned long flags;
146 
147 	SEQ_printf(m,
148 	"\nrunnable tasks:\n"
149 	"            task   PID         tree-key  switches  prio"
150 	"     exec-runtime         sum-exec        sum-sleep\n"
151 	"------------------------------------------------------"
152 	"----------------------------------------------------\n");
153 
154 	read_lock_irqsave(&tasklist_lock, flags);
155 
156 	do_each_thread(g, p) {
157 		if (!p->on_rq || task_cpu(p) != rq_cpu)
158 			continue;
159 
160 		print_task(m, rq, p);
161 	} while_each_thread(g, p);
162 
163 	read_unlock_irqrestore(&tasklist_lock, flags);
164 }
165 
166 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
167 {
168 	s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
169 		spread, rq0_min_vruntime, spread0;
170 	struct rq *rq = cpu_rq(cpu);
171 	struct sched_entity *last;
172 	unsigned long flags;
173 
174 #ifdef CONFIG_FAIR_GROUP_SCHED
175 	SEQ_printf(m, "\ncfs_rq[%d]:%s\n", cpu, task_group_path(cfs_rq->tg));
176 #else
177 	SEQ_printf(m, "\ncfs_rq[%d]:\n", cpu);
178 #endif
179 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "exec_clock",
180 			SPLIT_NS(cfs_rq->exec_clock));
181 
182 	raw_spin_lock_irqsave(&rq->lock, flags);
183 	if (cfs_rq->rb_leftmost)
184 		MIN_vruntime = (__pick_first_entity(cfs_rq))->vruntime;
185 	last = __pick_last_entity(cfs_rq);
186 	if (last)
187 		max_vruntime = last->vruntime;
188 	min_vruntime = cfs_rq->min_vruntime;
189 	rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
190 	raw_spin_unlock_irqrestore(&rq->lock, flags);
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));
197 	spread = max_vruntime - MIN_vruntime;
198 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread",
199 			SPLIT_NS(spread));
200 	spread0 = min_vruntime - rq0_min_vruntime;
201 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread0",
202 			SPLIT_NS(spread0));
203 	SEQ_printf(m, "  .%-30s: %d\n", "nr_spread_over",
204 			cfs_rq->nr_spread_over);
205 	SEQ_printf(m, "  .%-30s: %d\n", "nr_running", cfs_rq->nr_running);
206 	SEQ_printf(m, "  .%-30s: %ld\n", "load", cfs_rq->load.weight);
207 #ifdef CONFIG_FAIR_GROUP_SCHED
208 #ifdef CONFIG_SMP
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",
216 			atomic_read(&cfs_rq->tg->load_weight));
217 #endif
218 
219 	print_cfs_group_stats(m, cpu, cfs_rq->tg);
220 #endif
221 }
222 
223 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
224 {
225 #ifdef CONFIG_RT_GROUP_SCHED
226 	SEQ_printf(m, "\nrt_rq[%d]:%s\n", cpu, task_group_path(rt_rq->tg));
227 #else
228 	SEQ_printf(m, "\nrt_rq[%d]:\n", cpu);
229 #endif
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 
245 extern __read_mostly int sched_clock_running;
246 
247 static void print_cpu(struct seq_file *m, int cpu)
248 {
249 	struct rq *rq = cpu_rq(cpu);
250 	unsigned long flags;
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 do {									\
265 	if (sizeof(rq->x) == 4)						\
266 		SEQ_printf(m, "  .%-30s: %ld\n", #x, (long)(rq->x));	\
267 	else								\
268 		SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rq->x));\
269 } while (0)
270 
271 #define PN(x) \
272 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
273 
274 	P(nr_running);
275 	SEQ_printf(m, "  .%-30s: %lu\n", "load",
276 		   rq->load.weight);
277 	P(nr_switches);
278 	P(nr_load_updates);
279 	P(nr_uninterruptible);
280 	PN(next_balance);
281 	P(curr->pid);
282 	PN(clock);
283 	P(cpu_load[0]);
284 	P(cpu_load[1]);
285 	P(cpu_load[2]);
286 	P(cpu_load[3]);
287 	P(cpu_load[4]);
288 #undef P
289 #undef PN
290 
291 #ifdef CONFIG_SCHEDSTATS
292 #define P(n) SEQ_printf(m, "  .%-30s: %d\n", #n, rq->n);
293 #define P64(n) SEQ_printf(m, "  .%-30s: %Ld\n", #n, rq->n);
294 
295 	P(yld_count);
296 
297 	P(sched_count);
298 	P(sched_goidle);
299 #ifdef CONFIG_SMP
300 	P64(avg_idle);
301 #endif
302 
303 	P(ttwu_count);
304 	P(ttwu_local);
305 
306 #undef P
307 #undef P64
308 #endif
309 	spin_lock_irqsave(&sched_debug_lock, flags);
310 	print_cfs_stats(m, cpu);
311 	print_rt_stats(m, cpu);
312 
313 	rcu_read_lock();
314 	print_rq(m, rq, cpu);
315 	rcu_read_unlock();
316 	spin_unlock_irqrestore(&sched_debug_lock, flags);
317 }
318 
319 static const char *sched_tunable_scaling_names[] = {
320 	"none",
321 	"logaritmic",
322 	"linear"
323 };
324 
325 static int sched_debug_show(struct seq_file *m, void *v)
326 {
327 	u64 ktime, sched_clk, cpu_clk;
328 	unsigned long flags;
329 	int cpu;
330 
331 	local_irq_save(flags);
332 	ktime = ktime_to_ns(ktime_get());
333 	sched_clk = sched_clock();
334 	cpu_clk = local_clock();
335 	local_irq_restore(flags);
336 
337 	SEQ_printf(m, "Sched Debug Version: v0.10, %s %.*s\n",
338 		init_utsname()->release,
339 		(int)strcspn(init_utsname()->version, " "),
340 		init_utsname()->version);
341 
342 #define P(x) \
343 	SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
344 #define PN(x) \
345 	SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
346 	PN(ktime);
347 	PN(sched_clk);
348 	PN(cpu_clk);
349 	P(jiffies);
350 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
351 	P(sched_clock_stable);
352 #endif
353 #undef PN
354 #undef P
355 
356 	SEQ_printf(m, "\n");
357 	SEQ_printf(m, "sysctl_sched\n");
358 
359 #define P(x) \
360 	SEQ_printf(m, "  .%-40s: %Ld\n", #x, (long long)(x))
361 #define PN(x) \
362 	SEQ_printf(m, "  .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
363 	PN(sysctl_sched_latency);
364 	PN(sysctl_sched_min_granularity);
365 	PN(sysctl_sched_wakeup_granularity);
366 	P(sysctl_sched_child_runs_first);
367 	P(sysctl_sched_features);
368 #undef PN
369 #undef P
370 
371 	SEQ_printf(m, "  .%-40s: %d (%s)\n", "sysctl_sched_tunable_scaling",
372 		sysctl_sched_tunable_scaling,
373 		sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
374 
375 	for_each_online_cpu(cpu)
376 		print_cpu(m, cpu);
377 
378 	SEQ_printf(m, "\n");
379 
380 	return 0;
381 }
382 
383 void sysrq_sched_debug_show(void)
384 {
385 	sched_debug_show(NULL, NULL);
386 }
387 
388 static int sched_debug_open(struct inode *inode, struct file *filp)
389 {
390 	return single_open(filp, sched_debug_show, NULL);
391 }
392 
393 static const struct file_operations sched_debug_fops = {
394 	.open		= sched_debug_open,
395 	.read		= seq_read,
396 	.llseek		= seq_lseek,
397 	.release	= single_release,
398 };
399 
400 static int __init init_sched_debug_procfs(void)
401 {
402 	struct proc_dir_entry *pe;
403 
404 	pe = proc_create("sched_debug", 0444, NULL, &sched_debug_fops);
405 	if (!pe)
406 		return -ENOMEM;
407 	return 0;
408 }
409 
410 __initcall(init_sched_debug_procfs);
411 
412 void proc_sched_show_task(struct task_struct *p, struct seq_file *m)
413 {
414 	unsigned long nr_switches;
415 
416 	SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, p->pid,
417 						get_nr_threads(p));
418 	SEQ_printf(m,
419 		"---------------------------------------------------------\n");
420 #define __P(F) \
421 	SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)F)
422 #define P(F) \
423 	SEQ_printf(m, "%-35s:%21Ld\n", #F, (long long)p->F)
424 #define __PN(F) \
425 	SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
426 #define PN(F) \
427 	SEQ_printf(m, "%-35s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
428 
429 	PN(se.exec_start);
430 	PN(se.vruntime);
431 	PN(se.sum_exec_runtime);
432 
433 	nr_switches = p->nvcsw + p->nivcsw;
434 
435 #ifdef CONFIG_SCHEDSTATS
436 	PN(se.statistics.wait_start);
437 	PN(se.statistics.sleep_start);
438 	PN(se.statistics.block_start);
439 	PN(se.statistics.sleep_max);
440 	PN(se.statistics.block_max);
441 	PN(se.statistics.exec_max);
442 	PN(se.statistics.slice_max);
443 	PN(se.statistics.wait_max);
444 	PN(se.statistics.wait_sum);
445 	P(se.statistics.wait_count);
446 	PN(se.statistics.iowait_sum);
447 	P(se.statistics.iowait_count);
448 	P(se.nr_migrations);
449 	P(se.statistics.nr_migrations_cold);
450 	P(se.statistics.nr_failed_migrations_affine);
451 	P(se.statistics.nr_failed_migrations_running);
452 	P(se.statistics.nr_failed_migrations_hot);
453 	P(se.statistics.nr_forced_migrations);
454 	P(se.statistics.nr_wakeups);
455 	P(se.statistics.nr_wakeups_sync);
456 	P(se.statistics.nr_wakeups_migrate);
457 	P(se.statistics.nr_wakeups_local);
458 	P(se.statistics.nr_wakeups_remote);
459 	P(se.statistics.nr_wakeups_affine);
460 	P(se.statistics.nr_wakeups_affine_attempts);
461 	P(se.statistics.nr_wakeups_passive);
462 	P(se.statistics.nr_wakeups_idle);
463 
464 	{
465 		u64 avg_atom, avg_per_cpu;
466 
467 		avg_atom = p->se.sum_exec_runtime;
468 		if (nr_switches)
469 			do_div(avg_atom, nr_switches);
470 		else
471 			avg_atom = -1LL;
472 
473 		avg_per_cpu = p->se.sum_exec_runtime;
474 		if (p->se.nr_migrations) {
475 			avg_per_cpu = div64_u64(avg_per_cpu,
476 						p->se.nr_migrations);
477 		} else {
478 			avg_per_cpu = -1LL;
479 		}
480 
481 		__PN(avg_atom);
482 		__PN(avg_per_cpu);
483 	}
484 #endif
485 	__P(nr_switches);
486 	SEQ_printf(m, "%-35s:%21Ld\n",
487 		   "nr_voluntary_switches", (long long)p->nvcsw);
488 	SEQ_printf(m, "%-35s:%21Ld\n",
489 		   "nr_involuntary_switches", (long long)p->nivcsw);
490 
491 	P(se.load.weight);
492 	P(policy);
493 	P(prio);
494 #undef PN
495 #undef __PN
496 #undef P
497 #undef __P
498 
499 	{
500 		unsigned int this_cpu = raw_smp_processor_id();
501 		u64 t0, t1;
502 
503 		t0 = cpu_clock(this_cpu);
504 		t1 = cpu_clock(this_cpu);
505 		SEQ_printf(m, "%-35s:%21Ld\n",
506 			   "clock-delta", (long long)(t1-t0));
507 	}
508 }
509 
510 void proc_sched_set_task(struct task_struct *p)
511 {
512 #ifdef CONFIG_SCHEDSTATS
513 	memset(&p->se.statistics, 0, sizeof(p->se.statistics));
514 #endif
515 }
516