Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * kernel/sched/debug.c
4 *
5 * Print the CFS rbtree and other debugging details
6 *
7 * Copyright(C) 2007, Red Hat, Inc., Ingo Molnar
8 */
9#include "sched.h"
10
11static DEFINE_SPINLOCK(sched_debug_lock);
12
13/*
14 * This allows printing both to /proc/sched_debug and
15 * to the console
16 */
17#define SEQ_printf(m, x...) \
18 do { \
19 if (m) \
20 seq_printf(m, x); \
21 else \
22 pr_cont(x); \
23 } while (0)
24
25/*
26 * Ease the printing of nsec fields:
27 */
28static long long nsec_high(unsigned long long nsec)
29{
30 if ((long long)nsec < 0) {
31 nsec = -nsec;
32 do_div(nsec, 1000000);
33 return -nsec;
34 }
35 do_div(nsec, 1000000);
36
37 return nsec;
38}
39
40static unsigned long nsec_low(unsigned long long nsec)
41{
42 if ((long long)nsec < 0)
43 nsec = -nsec;
44
45 return do_div(nsec, 1000000);
46}
47
48#define SPLIT_NS(x) nsec_high(x), nsec_low(x)
49
50#define SCHED_FEAT(name, enabled) \
51 #name ,
52
53static const char * const sched_feat_names[] = {
54#include "features.h"
55};
56
57#undef SCHED_FEAT
58
59static int sched_feat_show(struct seq_file *m, void *v)
60{
61 int i;
62
63 for (i = 0; i < __SCHED_FEAT_NR; i++) {
64 if (!(sysctl_sched_features & (1UL << i)))
65 seq_puts(m, "NO_");
66 seq_printf(m, "%s ", sched_feat_names[i]);
67 }
68 seq_puts(m, "\n");
69
70 return 0;
71}
72
73#ifdef CONFIG_JUMP_LABEL
74
75#define jump_label_key__true STATIC_KEY_INIT_TRUE
76#define jump_label_key__false STATIC_KEY_INIT_FALSE
77
78#define SCHED_FEAT(name, enabled) \
79 jump_label_key__##enabled ,
80
81struct static_key sched_feat_keys[__SCHED_FEAT_NR] = {
82#include "features.h"
83};
84
85#undef SCHED_FEAT
86
87static void sched_feat_disable(int i)
88{
89 static_key_disable_cpuslocked(&sched_feat_keys[i]);
90}
91
92static void sched_feat_enable(int i)
93{
94 static_key_enable_cpuslocked(&sched_feat_keys[i]);
95}
96#else
97static void sched_feat_disable(int i) { };
98static void sched_feat_enable(int i) { };
99#endif /* CONFIG_JUMP_LABEL */
100
101static int sched_feat_set(char *cmp)
102{
103 int i;
104 int neg = 0;
105
106 if (strncmp(cmp, "NO_", 3) == 0) {
107 neg = 1;
108 cmp += 3;
109 }
110
111 i = match_string(sched_feat_names, __SCHED_FEAT_NR, cmp);
112 if (i < 0)
113 return i;
114
115 if (neg) {
116 sysctl_sched_features &= ~(1UL << i);
117 sched_feat_disable(i);
118 } else {
119 sysctl_sched_features |= (1UL << i);
120 sched_feat_enable(i);
121 }
122
123 return 0;
124}
125
126static ssize_t
127sched_feat_write(struct file *filp, const char __user *ubuf,
128 size_t cnt, loff_t *ppos)
129{
130 char buf[64];
131 char *cmp;
132 int ret;
133 struct inode *inode;
134
135 if (cnt > 63)
136 cnt = 63;
137
138 if (copy_from_user(&buf, ubuf, cnt))
139 return -EFAULT;
140
141 buf[cnt] = 0;
142 cmp = strstrip(buf);
143
144 /* Ensure the static_key remains in a consistent state */
145 inode = file_inode(filp);
146 cpus_read_lock();
147 inode_lock(inode);
148 ret = sched_feat_set(cmp);
149 inode_unlock(inode);
150 cpus_read_unlock();
151 if (ret < 0)
152 return ret;
153
154 *ppos += cnt;
155
156 return cnt;
157}
158
159static int sched_feat_open(struct inode *inode, struct file *filp)
160{
161 return single_open(filp, sched_feat_show, NULL);
162}
163
164static const struct file_operations sched_feat_fops = {
165 .open = sched_feat_open,
166 .write = sched_feat_write,
167 .read = seq_read,
168 .llseek = seq_lseek,
169 .release = single_release,
170};
171
172__read_mostly bool sched_debug_enabled;
173
174static __init int sched_init_debug(void)
175{
176 debugfs_create_file("sched_features", 0644, NULL, NULL,
177 &sched_feat_fops);
178
179 debugfs_create_bool("sched_debug", 0644, NULL,
180 &sched_debug_enabled);
181
182 return 0;
183}
184late_initcall(sched_init_debug);
185
186#ifdef CONFIG_SMP
187
188#ifdef CONFIG_SYSCTL
189
190static struct ctl_table sd_ctl_dir[] = {
191 {
192 .procname = "sched_domain",
193 .mode = 0555,
194 },
195 {}
196};
197
198static struct ctl_table sd_ctl_root[] = {
199 {
200 .procname = "kernel",
201 .mode = 0555,
202 .child = sd_ctl_dir,
203 },
204 {}
205};
206
207static struct ctl_table *sd_alloc_ctl_entry(int n)
208{
209 struct ctl_table *entry =
210 kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL);
211
212 return entry;
213}
214
215static void sd_free_ctl_entry(struct ctl_table **tablep)
216{
217 struct ctl_table *entry;
218
219 /*
220 * In the intermediate directories, both the child directory and
221 * procname are dynamically allocated and could fail but the mode
222 * will always be set. In the lowest directory the names are
223 * static strings and all have proc handlers.
224 */
225 for (entry = *tablep; entry->mode; entry++) {
226 if (entry->child)
227 sd_free_ctl_entry(&entry->child);
228 if (entry->proc_handler == NULL)
229 kfree(entry->procname);
230 }
231
232 kfree(*tablep);
233 *tablep = NULL;
234}
235
236static void
237set_table_entry(struct ctl_table *entry,
238 const char *procname, void *data, int maxlen,
239 umode_t mode, proc_handler *proc_handler)
240{
241 entry->procname = procname;
242 entry->data = data;
243 entry->maxlen = maxlen;
244 entry->mode = mode;
245 entry->proc_handler = proc_handler;
246}
247
248static struct ctl_table *
249sd_alloc_ctl_domain_table(struct sched_domain *sd)
250{
251 struct ctl_table *table = sd_alloc_ctl_entry(9);
252
253 if (table == NULL)
254 return NULL;
255
256 set_table_entry(&table[0], "min_interval", &sd->min_interval, sizeof(long), 0644, proc_doulongvec_minmax);
257 set_table_entry(&table[1], "max_interval", &sd->max_interval, sizeof(long), 0644, proc_doulongvec_minmax);
258 set_table_entry(&table[2], "busy_factor", &sd->busy_factor, sizeof(int), 0644, proc_dointvec_minmax);
259 set_table_entry(&table[3], "imbalance_pct", &sd->imbalance_pct, sizeof(int), 0644, proc_dointvec_minmax);
260 set_table_entry(&table[4], "cache_nice_tries", &sd->cache_nice_tries, sizeof(int), 0644, proc_dointvec_minmax);
261 set_table_entry(&table[5], "flags", &sd->flags, sizeof(int), 0644, proc_dointvec_minmax);
262 set_table_entry(&table[6], "max_newidle_lb_cost", &sd->max_newidle_lb_cost, sizeof(long), 0644, proc_doulongvec_minmax);
263 set_table_entry(&table[7], "name", sd->name, CORENAME_MAX_SIZE, 0444, proc_dostring);
264 /* &table[8] is terminator */
265
266 return table;
267}
268
269static struct ctl_table *sd_alloc_ctl_cpu_table(int cpu)
270{
271 struct ctl_table *entry, *table;
272 struct sched_domain *sd;
273 int domain_num = 0, i;
274 char buf[32];
275
276 for_each_domain(cpu, sd)
277 domain_num++;
278 entry = table = sd_alloc_ctl_entry(domain_num + 1);
279 if (table == NULL)
280 return NULL;
281
282 i = 0;
283 for_each_domain(cpu, sd) {
284 snprintf(buf, 32, "domain%d", i);
285 entry->procname = kstrdup(buf, GFP_KERNEL);
286 entry->mode = 0555;
287 entry->child = sd_alloc_ctl_domain_table(sd);
288 entry++;
289 i++;
290 }
291 return table;
292}
293
294static cpumask_var_t sd_sysctl_cpus;
295static struct ctl_table_header *sd_sysctl_header;
296
297void register_sched_domain_sysctl(void)
298{
299 static struct ctl_table *cpu_entries;
300 static struct ctl_table **cpu_idx;
301 static bool init_done = false;
302 char buf[32];
303 int i;
304
305 if (!cpu_entries) {
306 cpu_entries = sd_alloc_ctl_entry(num_possible_cpus() + 1);
307 if (!cpu_entries)
308 return;
309
310 WARN_ON(sd_ctl_dir[0].child);
311 sd_ctl_dir[0].child = cpu_entries;
312 }
313
314 if (!cpu_idx) {
315 struct ctl_table *e = cpu_entries;
316
317 cpu_idx = kcalloc(nr_cpu_ids, sizeof(struct ctl_table*), GFP_KERNEL);
318 if (!cpu_idx)
319 return;
320
321 /* deal with sparse possible map */
322 for_each_possible_cpu(i) {
323 cpu_idx[i] = e;
324 e++;
325 }
326 }
327
328 if (!cpumask_available(sd_sysctl_cpus)) {
329 if (!alloc_cpumask_var(&sd_sysctl_cpus, GFP_KERNEL))
330 return;
331 }
332
333 if (!init_done) {
334 init_done = true;
335 /* init to possible to not have holes in @cpu_entries */
336 cpumask_copy(sd_sysctl_cpus, cpu_possible_mask);
337 }
338
339 for_each_cpu(i, sd_sysctl_cpus) {
340 struct ctl_table *e = cpu_idx[i];
341
342 if (e->child)
343 sd_free_ctl_entry(&e->child);
344
345 if (!e->procname) {
346 snprintf(buf, 32, "cpu%d", i);
347 e->procname = kstrdup(buf, GFP_KERNEL);
348 }
349 e->mode = 0555;
350 e->child = sd_alloc_ctl_cpu_table(i);
351
352 __cpumask_clear_cpu(i, sd_sysctl_cpus);
353 }
354
355 WARN_ON(sd_sysctl_header);
356 sd_sysctl_header = register_sysctl_table(sd_ctl_root);
357}
358
359void dirty_sched_domain_sysctl(int cpu)
360{
361 if (cpumask_available(sd_sysctl_cpus))
362 __cpumask_set_cpu(cpu, sd_sysctl_cpus);
363}
364
365/* may be called multiple times per register */
366void unregister_sched_domain_sysctl(void)
367{
368 unregister_sysctl_table(sd_sysctl_header);
369 sd_sysctl_header = NULL;
370}
371#endif /* CONFIG_SYSCTL */
372#endif /* CONFIG_SMP */
373
374#ifdef CONFIG_FAIR_GROUP_SCHED
375static void print_cfs_group_stats(struct seq_file *m, int cpu, struct task_group *tg)
376{
377 struct sched_entity *se = tg->se[cpu];
378
379#define P(F) SEQ_printf(m, " .%-30s: %lld\n", #F, (long long)F)
380#define P_SCHEDSTAT(F) SEQ_printf(m, " .%-30s: %lld\n", #F, (long long)schedstat_val(F))
381#define PN(F) SEQ_printf(m, " .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F))
382#define PN_SCHEDSTAT(F) SEQ_printf(m, " .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)schedstat_val(F)))
383
384 if (!se)
385 return;
386
387 PN(se->exec_start);
388 PN(se->vruntime);
389 PN(se->sum_exec_runtime);
390
391 if (schedstat_enabled()) {
392 PN_SCHEDSTAT(se->statistics.wait_start);
393 PN_SCHEDSTAT(se->statistics.sleep_start);
394 PN_SCHEDSTAT(se->statistics.block_start);
395 PN_SCHEDSTAT(se->statistics.sleep_max);
396 PN_SCHEDSTAT(se->statistics.block_max);
397 PN_SCHEDSTAT(se->statistics.exec_max);
398 PN_SCHEDSTAT(se->statistics.slice_max);
399 PN_SCHEDSTAT(se->statistics.wait_max);
400 PN_SCHEDSTAT(se->statistics.wait_sum);
401 P_SCHEDSTAT(se->statistics.wait_count);
402 }
403
404 P(se->load.weight);
405 P(se->runnable_weight);
406#ifdef CONFIG_SMP
407 P(se->avg.load_avg);
408 P(se->avg.util_avg);
409 P(se->avg.runnable_load_avg);
410#endif
411
412#undef PN_SCHEDSTAT
413#undef PN
414#undef P_SCHEDSTAT
415#undef P
416}
417#endif
418
419#ifdef CONFIG_CGROUP_SCHED
420static char group_path[PATH_MAX];
421
422static char *task_group_path(struct task_group *tg)
423{
424 if (autogroup_path(tg, group_path, PATH_MAX))
425 return group_path;
426
427 cgroup_path(tg->css.cgroup, group_path, PATH_MAX);
428
429 return group_path;
430}
431#endif
432
433static void
434print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
435{
436 if (rq->curr == p)
437 SEQ_printf(m, ">R");
438 else
439 SEQ_printf(m, " %c", task_state_to_char(p));
440
441 SEQ_printf(m, "%15s %5d %9Ld.%06ld %9Ld %5d ",
442 p->comm, task_pid_nr(p),
443 SPLIT_NS(p->se.vruntime),
444 (long long)(p->nvcsw + p->nivcsw),
445 p->prio);
446
447 SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
448 SPLIT_NS(schedstat_val_or_zero(p->se.statistics.wait_sum)),
449 SPLIT_NS(p->se.sum_exec_runtime),
450 SPLIT_NS(schedstat_val_or_zero(p->se.statistics.sum_sleep_runtime)));
451
452#ifdef CONFIG_NUMA_BALANCING
453 SEQ_printf(m, " %d %d", task_node(p), task_numa_group_id(p));
454#endif
455#ifdef CONFIG_CGROUP_SCHED
456 SEQ_printf(m, " %s", task_group_path(task_group(p)));
457#endif
458
459 SEQ_printf(m, "\n");
460}
461
462static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
463{
464 struct task_struct *g, *p;
465
466 SEQ_printf(m, "\n");
467 SEQ_printf(m, "runnable tasks:\n");
468 SEQ_printf(m, " S task PID tree-key switches prio"
469 " wait-time sum-exec sum-sleep\n");
470 SEQ_printf(m, "-------------------------------------------------------"
471 "----------------------------------------------------\n");
472
473 rcu_read_lock();
474 for_each_process_thread(g, p) {
475 if (task_cpu(p) != rq_cpu)
476 continue;
477
478 print_task(m, rq, p);
479 }
480 rcu_read_unlock();
481}
482
483void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
484{
485 s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
486 spread, rq0_min_vruntime, spread0;
487 struct rq *rq = cpu_rq(cpu);
488 struct sched_entity *last;
489 unsigned long flags;
490
491#ifdef CONFIG_FAIR_GROUP_SCHED
492 SEQ_printf(m, "\n");
493 SEQ_printf(m, "cfs_rq[%d]:%s\n", cpu, task_group_path(cfs_rq->tg));
494#else
495 SEQ_printf(m, "\n");
496 SEQ_printf(m, "cfs_rq[%d]:\n", cpu);
497#endif
498 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "exec_clock",
499 SPLIT_NS(cfs_rq->exec_clock));
500
501 raw_spin_lock_irqsave(&rq->lock, flags);
502 if (rb_first_cached(&cfs_rq->tasks_timeline))
503 MIN_vruntime = (__pick_first_entity(cfs_rq))->vruntime;
504 last = __pick_last_entity(cfs_rq);
505 if (last)
506 max_vruntime = last->vruntime;
507 min_vruntime = cfs_rq->min_vruntime;
508 rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
509 raw_spin_unlock_irqrestore(&rq->lock, flags);
510 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "MIN_vruntime",
511 SPLIT_NS(MIN_vruntime));
512 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "min_vruntime",
513 SPLIT_NS(min_vruntime));
514 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "max_vruntime",
515 SPLIT_NS(max_vruntime));
516 spread = max_vruntime - MIN_vruntime;
517 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread",
518 SPLIT_NS(spread));
519 spread0 = min_vruntime - rq0_min_vruntime;
520 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread0",
521 SPLIT_NS(spread0));
522 SEQ_printf(m, " .%-30s: %d\n", "nr_spread_over",
523 cfs_rq->nr_spread_over);
524 SEQ_printf(m, " .%-30s: %d\n", "nr_running", cfs_rq->nr_running);
525 SEQ_printf(m, " .%-30s: %ld\n", "load", cfs_rq->load.weight);
526#ifdef CONFIG_SMP
527 SEQ_printf(m, " .%-30s: %ld\n", "runnable_weight", cfs_rq->runnable_weight);
528 SEQ_printf(m, " .%-30s: %lu\n", "load_avg",
529 cfs_rq->avg.load_avg);
530 SEQ_printf(m, " .%-30s: %lu\n", "runnable_load_avg",
531 cfs_rq->avg.runnable_load_avg);
532 SEQ_printf(m, " .%-30s: %lu\n", "util_avg",
533 cfs_rq->avg.util_avg);
534 SEQ_printf(m, " .%-30s: %u\n", "util_est_enqueued",
535 cfs_rq->avg.util_est.enqueued);
536 SEQ_printf(m, " .%-30s: %ld\n", "removed.load_avg",
537 cfs_rq->removed.load_avg);
538 SEQ_printf(m, " .%-30s: %ld\n", "removed.util_avg",
539 cfs_rq->removed.util_avg);
540 SEQ_printf(m, " .%-30s: %ld\n", "removed.runnable_sum",
541 cfs_rq->removed.runnable_sum);
542#ifdef CONFIG_FAIR_GROUP_SCHED
543 SEQ_printf(m, " .%-30s: %lu\n", "tg_load_avg_contrib",
544 cfs_rq->tg_load_avg_contrib);
545 SEQ_printf(m, " .%-30s: %ld\n", "tg_load_avg",
546 atomic_long_read(&cfs_rq->tg->load_avg));
547#endif
548#endif
549#ifdef CONFIG_CFS_BANDWIDTH
550 SEQ_printf(m, " .%-30s: %d\n", "throttled",
551 cfs_rq->throttled);
552 SEQ_printf(m, " .%-30s: %d\n", "throttle_count",
553 cfs_rq->throttle_count);
554#endif
555
556#ifdef CONFIG_FAIR_GROUP_SCHED
557 print_cfs_group_stats(m, cpu, cfs_rq->tg);
558#endif
559}
560
561void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
562{
563#ifdef CONFIG_RT_GROUP_SCHED
564 SEQ_printf(m, "\n");
565 SEQ_printf(m, "rt_rq[%d]:%s\n", cpu, task_group_path(rt_rq->tg));
566#else
567 SEQ_printf(m, "\n");
568 SEQ_printf(m, "rt_rq[%d]:\n", cpu);
569#endif
570
571#define P(x) \
572 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
573#define PU(x) \
574 SEQ_printf(m, " .%-30s: %lu\n", #x, (unsigned long)(rt_rq->x))
575#define PN(x) \
576 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
577
578 PU(rt_nr_running);
579#ifdef CONFIG_SMP
580 PU(rt_nr_migratory);
581#endif
582 P(rt_throttled);
583 PN(rt_time);
584 PN(rt_runtime);
585
586#undef PN
587#undef PU
588#undef P
589}
590
591void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq)
592{
593 struct dl_bw *dl_bw;
594
595 SEQ_printf(m, "\n");
596 SEQ_printf(m, "dl_rq[%d]:\n", cpu);
597
598#define PU(x) \
599 SEQ_printf(m, " .%-30s: %lu\n", #x, (unsigned long)(dl_rq->x))
600
601 PU(dl_nr_running);
602#ifdef CONFIG_SMP
603 PU(dl_nr_migratory);
604 dl_bw = &cpu_rq(cpu)->rd->dl_bw;
605#else
606 dl_bw = &dl_rq->dl_bw;
607#endif
608 SEQ_printf(m, " .%-30s: %lld\n", "dl_bw->bw", dl_bw->bw);
609 SEQ_printf(m, " .%-30s: %lld\n", "dl_bw->total_bw", dl_bw->total_bw);
610
611#undef PU
612}
613
614static void print_cpu(struct seq_file *m, int cpu)
615{
616 struct rq *rq = cpu_rq(cpu);
617 unsigned long flags;
618
619#ifdef CONFIG_X86
620 {
621 unsigned int freq = cpu_khz ? : 1;
622
623 SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
624 cpu, freq / 1000, (freq % 1000));
625 }
626#else
627 SEQ_printf(m, "cpu#%d\n", cpu);
628#endif
629
630#define P(x) \
631do { \
632 if (sizeof(rq->x) == 4) \
633 SEQ_printf(m, " .%-30s: %ld\n", #x, (long)(rq->x)); \
634 else \
635 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rq->x));\
636} while (0)
637
638#define PN(x) \
639 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
640
641 P(nr_running);
642 P(nr_switches);
643 P(nr_load_updates);
644 P(nr_uninterruptible);
645 PN(next_balance);
646 SEQ_printf(m, " .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr)));
647 PN(clock);
648 PN(clock_task);
649#undef P
650#undef PN
651
652#ifdef CONFIG_SMP
653#define P64(n) SEQ_printf(m, " .%-30s: %Ld\n", #n, rq->n);
654 P64(avg_idle);
655 P64(max_idle_balance_cost);
656#undef P64
657#endif
658
659#define P(n) SEQ_printf(m, " .%-30s: %d\n", #n, schedstat_val(rq->n));
660 if (schedstat_enabled()) {
661 P(yld_count);
662 P(sched_count);
663 P(sched_goidle);
664 P(ttwu_count);
665 P(ttwu_local);
666 }
667#undef P
668
669 spin_lock_irqsave(&sched_debug_lock, flags);
670 print_cfs_stats(m, cpu);
671 print_rt_stats(m, cpu);
672 print_dl_stats(m, cpu);
673
674 print_rq(m, rq, cpu);
675 spin_unlock_irqrestore(&sched_debug_lock, flags);
676 SEQ_printf(m, "\n");
677}
678
679static const char *sched_tunable_scaling_names[] = {
680 "none",
681 "logarithmic",
682 "linear"
683};
684
685static void sched_debug_header(struct seq_file *m)
686{
687 u64 ktime, sched_clk, cpu_clk;
688 unsigned long flags;
689
690 local_irq_save(flags);
691 ktime = ktime_to_ns(ktime_get());
692 sched_clk = sched_clock();
693 cpu_clk = local_clock();
694 local_irq_restore(flags);
695
696 SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n",
697 init_utsname()->release,
698 (int)strcspn(init_utsname()->version, " "),
699 init_utsname()->version);
700
701#define P(x) \
702 SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
703#define PN(x) \
704 SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
705 PN(ktime);
706 PN(sched_clk);
707 PN(cpu_clk);
708 P(jiffies);
709#ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
710 P(sched_clock_stable());
711#endif
712#undef PN
713#undef P
714
715 SEQ_printf(m, "\n");
716 SEQ_printf(m, "sysctl_sched\n");
717
718#define P(x) \
719 SEQ_printf(m, " .%-40s: %Ld\n", #x, (long long)(x))
720#define PN(x) \
721 SEQ_printf(m, " .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
722 PN(sysctl_sched_latency);
723 PN(sysctl_sched_min_granularity);
724 PN(sysctl_sched_wakeup_granularity);
725 P(sysctl_sched_child_runs_first);
726 P(sysctl_sched_features);
727#undef PN
728#undef P
729
730 SEQ_printf(m, " .%-40s: %d (%s)\n",
731 "sysctl_sched_tunable_scaling",
732 sysctl_sched_tunable_scaling,
733 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
734 SEQ_printf(m, "\n");
735}
736
737static int sched_debug_show(struct seq_file *m, void *v)
738{
739 int cpu = (unsigned long)(v - 2);
740
741 if (cpu != -1)
742 print_cpu(m, cpu);
743 else
744 sched_debug_header(m);
745
746 return 0;
747}
748
749void sysrq_sched_debug_show(void)
750{
751 int cpu;
752
753 sched_debug_header(NULL);
754 for_each_online_cpu(cpu)
755 print_cpu(NULL, cpu);
756
757}
758
759/*
760 * This itererator needs some explanation.
761 * It returns 1 for the header position.
762 * This means 2 is CPU 0.
763 * In a hotplugged system some CPUs, including CPU 0, may be missing so we have
764 * to use cpumask_* to iterate over the CPUs.
765 */
766static void *sched_debug_start(struct seq_file *file, loff_t *offset)
767{
768 unsigned long n = *offset;
769
770 if (n == 0)
771 return (void *) 1;
772
773 n--;
774
775 if (n > 0)
776 n = cpumask_next(n - 1, cpu_online_mask);
777 else
778 n = cpumask_first(cpu_online_mask);
779
780 *offset = n + 1;
781
782 if (n < nr_cpu_ids)
783 return (void *)(unsigned long)(n + 2);
784
785 return NULL;
786}
787
788static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
789{
790 (*offset)++;
791 return sched_debug_start(file, offset);
792}
793
794static void sched_debug_stop(struct seq_file *file, void *data)
795{
796}
797
798static const struct seq_operations sched_debug_sops = {
799 .start = sched_debug_start,
800 .next = sched_debug_next,
801 .stop = sched_debug_stop,
802 .show = sched_debug_show,
803};
804
805static int __init init_sched_debug_procfs(void)
806{
807 if (!proc_create_seq("sched_debug", 0444, NULL, &sched_debug_sops))
808 return -ENOMEM;
809 return 0;
810}
811
812__initcall(init_sched_debug_procfs);
813
814#define __P(F) SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)F)
815#define P(F) SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)p->F)
816#define __PN(F) SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
817#define PN(F) SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
818
819
820#ifdef CONFIG_NUMA_BALANCING
821void print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
822 unsigned long tpf, unsigned long gsf, unsigned long gpf)
823{
824 SEQ_printf(m, "numa_faults node=%d ", node);
825 SEQ_printf(m, "task_private=%lu task_shared=%lu ", tpf, tsf);
826 SEQ_printf(m, "group_private=%lu group_shared=%lu\n", gpf, gsf);
827}
828#endif
829
830
831static void sched_show_numa(struct task_struct *p, struct seq_file *m)
832{
833#ifdef CONFIG_NUMA_BALANCING
834 struct mempolicy *pol;
835
836 if (p->mm)
837 P(mm->numa_scan_seq);
838
839 task_lock(p);
840 pol = p->mempolicy;
841 if (pol && !(pol->flags & MPOL_F_MORON))
842 pol = NULL;
843 mpol_get(pol);
844 task_unlock(p);
845
846 P(numa_pages_migrated);
847 P(numa_preferred_nid);
848 P(total_numa_faults);
849 SEQ_printf(m, "current_node=%d, numa_group_id=%d\n",
850 task_node(p), task_numa_group_id(p));
851 show_numa_stats(p, m);
852 mpol_put(pol);
853#endif
854}
855
856void proc_sched_show_task(struct task_struct *p, struct pid_namespace *ns,
857 struct seq_file *m)
858{
859 unsigned long nr_switches;
860
861 SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr_ns(p, ns),
862 get_nr_threads(p));
863 SEQ_printf(m,
864 "---------------------------------------------------------"
865 "----------\n");
866#define __P(F) \
867 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)F)
868#define P(F) \
869 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)p->F)
870#define P_SCHEDSTAT(F) \
871 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)schedstat_val(p->F))
872#define __PN(F) \
873 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
874#define PN(F) \
875 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
876#define PN_SCHEDSTAT(F) \
877 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)schedstat_val(p->F)))
878
879 PN(se.exec_start);
880 PN(se.vruntime);
881 PN(se.sum_exec_runtime);
882
883 nr_switches = p->nvcsw + p->nivcsw;
884
885 P(se.nr_migrations);
886
887 if (schedstat_enabled()) {
888 u64 avg_atom, avg_per_cpu;
889
890 PN_SCHEDSTAT(se.statistics.sum_sleep_runtime);
891 PN_SCHEDSTAT(se.statistics.wait_start);
892 PN_SCHEDSTAT(se.statistics.sleep_start);
893 PN_SCHEDSTAT(se.statistics.block_start);
894 PN_SCHEDSTAT(se.statistics.sleep_max);
895 PN_SCHEDSTAT(se.statistics.block_max);
896 PN_SCHEDSTAT(se.statistics.exec_max);
897 PN_SCHEDSTAT(se.statistics.slice_max);
898 PN_SCHEDSTAT(se.statistics.wait_max);
899 PN_SCHEDSTAT(se.statistics.wait_sum);
900 P_SCHEDSTAT(se.statistics.wait_count);
901 PN_SCHEDSTAT(se.statistics.iowait_sum);
902 P_SCHEDSTAT(se.statistics.iowait_count);
903 P_SCHEDSTAT(se.statistics.nr_migrations_cold);
904 P_SCHEDSTAT(se.statistics.nr_failed_migrations_affine);
905 P_SCHEDSTAT(se.statistics.nr_failed_migrations_running);
906 P_SCHEDSTAT(se.statistics.nr_failed_migrations_hot);
907 P_SCHEDSTAT(se.statistics.nr_forced_migrations);
908 P_SCHEDSTAT(se.statistics.nr_wakeups);
909 P_SCHEDSTAT(se.statistics.nr_wakeups_sync);
910 P_SCHEDSTAT(se.statistics.nr_wakeups_migrate);
911 P_SCHEDSTAT(se.statistics.nr_wakeups_local);
912 P_SCHEDSTAT(se.statistics.nr_wakeups_remote);
913 P_SCHEDSTAT(se.statistics.nr_wakeups_affine);
914 P_SCHEDSTAT(se.statistics.nr_wakeups_affine_attempts);
915 P_SCHEDSTAT(se.statistics.nr_wakeups_passive);
916 P_SCHEDSTAT(se.statistics.nr_wakeups_idle);
917
918 avg_atom = p->se.sum_exec_runtime;
919 if (nr_switches)
920 avg_atom = div64_ul(avg_atom, nr_switches);
921 else
922 avg_atom = -1LL;
923
924 avg_per_cpu = p->se.sum_exec_runtime;
925 if (p->se.nr_migrations) {
926 avg_per_cpu = div64_u64(avg_per_cpu,
927 p->se.nr_migrations);
928 } else {
929 avg_per_cpu = -1LL;
930 }
931
932 __PN(avg_atom);
933 __PN(avg_per_cpu);
934 }
935
936 __P(nr_switches);
937 SEQ_printf(m, "%-45s:%21Ld\n",
938 "nr_voluntary_switches", (long long)p->nvcsw);
939 SEQ_printf(m, "%-45s:%21Ld\n",
940 "nr_involuntary_switches", (long long)p->nivcsw);
941
942 P(se.load.weight);
943 P(se.runnable_weight);
944#ifdef CONFIG_SMP
945 P(se.avg.load_sum);
946 P(se.avg.runnable_load_sum);
947 P(se.avg.util_sum);
948 P(se.avg.load_avg);
949 P(se.avg.runnable_load_avg);
950 P(se.avg.util_avg);
951 P(se.avg.last_update_time);
952 P(se.avg.util_est.ewma);
953 P(se.avg.util_est.enqueued);
954#endif
955 P(policy);
956 P(prio);
957 if (task_has_dl_policy(p)) {
958 P(dl.runtime);
959 P(dl.deadline);
960 }
961#undef PN_SCHEDSTAT
962#undef PN
963#undef __PN
964#undef P_SCHEDSTAT
965#undef P
966#undef __P
967
968 {
969 unsigned int this_cpu = raw_smp_processor_id();
970 u64 t0, t1;
971
972 t0 = cpu_clock(this_cpu);
973 t1 = cpu_clock(this_cpu);
974 SEQ_printf(m, "%-45s:%21Ld\n",
975 "clock-delta", (long long)(t1-t0));
976 }
977
978 sched_show_numa(p, m);
979}
980
981void proc_sched_set_task(struct task_struct *p)
982{
983#ifdef CONFIG_SCHEDSTATS
984 memset(&p->se.statistics, 0, sizeof(p->se.statistics));
985#endif
986}
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#include <linux/mempolicy.h>
19#include <linux/debugfs.h>
20
21#include "sched.h"
22
23static DEFINE_SPINLOCK(sched_debug_lock);
24
25/*
26 * This allows printing both to /proc/sched_debug and
27 * to the console
28 */
29#define SEQ_printf(m, x...) \
30 do { \
31 if (m) \
32 seq_printf(m, x); \
33 else \
34 printk(x); \
35 } while (0)
36
37/*
38 * Ease the printing of nsec fields:
39 */
40static long long nsec_high(unsigned long long nsec)
41{
42 if ((long long)nsec < 0) {
43 nsec = -nsec;
44 do_div(nsec, 1000000);
45 return -nsec;
46 }
47 do_div(nsec, 1000000);
48
49 return nsec;
50}
51
52static unsigned long nsec_low(unsigned long long nsec)
53{
54 if ((long long)nsec < 0)
55 nsec = -nsec;
56
57 return do_div(nsec, 1000000);
58}
59
60#define SPLIT_NS(x) nsec_high(x), nsec_low(x)
61
62#define SCHED_FEAT(name, enabled) \
63 #name ,
64
65static const char * const sched_feat_names[] = {
66#include "features.h"
67};
68
69#undef SCHED_FEAT
70
71static int sched_feat_show(struct seq_file *m, void *v)
72{
73 int i;
74
75 for (i = 0; i < __SCHED_FEAT_NR; i++) {
76 if (!(sysctl_sched_features & (1UL << i)))
77 seq_puts(m, "NO_");
78 seq_printf(m, "%s ", sched_feat_names[i]);
79 }
80 seq_puts(m, "\n");
81
82 return 0;
83}
84
85#ifdef HAVE_JUMP_LABEL
86
87#define jump_label_key__true STATIC_KEY_INIT_TRUE
88#define jump_label_key__false STATIC_KEY_INIT_FALSE
89
90#define SCHED_FEAT(name, enabled) \
91 jump_label_key__##enabled ,
92
93struct static_key sched_feat_keys[__SCHED_FEAT_NR] = {
94#include "features.h"
95};
96
97#undef SCHED_FEAT
98
99static void sched_feat_disable(int i)
100{
101 static_key_disable(&sched_feat_keys[i]);
102}
103
104static void sched_feat_enable(int i)
105{
106 static_key_enable(&sched_feat_keys[i]);
107}
108#else
109static void sched_feat_disable(int i) { };
110static void sched_feat_enable(int i) { };
111#endif /* HAVE_JUMP_LABEL */
112
113static int sched_feat_set(char *cmp)
114{
115 int i;
116 int neg = 0;
117
118 if (strncmp(cmp, "NO_", 3) == 0) {
119 neg = 1;
120 cmp += 3;
121 }
122
123 for (i = 0; i < __SCHED_FEAT_NR; i++) {
124 if (strcmp(cmp, sched_feat_names[i]) == 0) {
125 if (neg) {
126 sysctl_sched_features &= ~(1UL << i);
127 sched_feat_disable(i);
128 } else {
129 sysctl_sched_features |= (1UL << i);
130 sched_feat_enable(i);
131 }
132 break;
133 }
134 }
135
136 return i;
137}
138
139static ssize_t
140sched_feat_write(struct file *filp, const char __user *ubuf,
141 size_t cnt, loff_t *ppos)
142{
143 char buf[64];
144 char *cmp;
145 int i;
146 struct inode *inode;
147
148 if (cnt > 63)
149 cnt = 63;
150
151 if (copy_from_user(&buf, ubuf, cnt))
152 return -EFAULT;
153
154 buf[cnt] = 0;
155 cmp = strstrip(buf);
156
157 /* Ensure the static_key remains in a consistent state */
158 inode = file_inode(filp);
159 inode_lock(inode);
160 i = sched_feat_set(cmp);
161 inode_unlock(inode);
162 if (i == __SCHED_FEAT_NR)
163 return -EINVAL;
164
165 *ppos += cnt;
166
167 return cnt;
168}
169
170static int sched_feat_open(struct inode *inode, struct file *filp)
171{
172 return single_open(filp, sched_feat_show, NULL);
173}
174
175static const struct file_operations sched_feat_fops = {
176 .open = sched_feat_open,
177 .write = sched_feat_write,
178 .read = seq_read,
179 .llseek = seq_lseek,
180 .release = single_release,
181};
182
183static __init int sched_init_debug(void)
184{
185 debugfs_create_file("sched_features", 0644, NULL, NULL,
186 &sched_feat_fops);
187
188 return 0;
189}
190late_initcall(sched_init_debug);
191
192#ifdef CONFIG_SMP
193
194#ifdef CONFIG_SYSCTL
195
196static struct ctl_table sd_ctl_dir[] = {
197 {
198 .procname = "sched_domain",
199 .mode = 0555,
200 },
201 {}
202};
203
204static struct ctl_table sd_ctl_root[] = {
205 {
206 .procname = "kernel",
207 .mode = 0555,
208 .child = sd_ctl_dir,
209 },
210 {}
211};
212
213static struct ctl_table *sd_alloc_ctl_entry(int n)
214{
215 struct ctl_table *entry =
216 kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL);
217
218 return entry;
219}
220
221static void sd_free_ctl_entry(struct ctl_table **tablep)
222{
223 struct ctl_table *entry;
224
225 /*
226 * In the intermediate directories, both the child directory and
227 * procname are dynamically allocated and could fail but the mode
228 * will always be set. In the lowest directory the names are
229 * static strings and all have proc handlers.
230 */
231 for (entry = *tablep; entry->mode; entry++) {
232 if (entry->child)
233 sd_free_ctl_entry(&entry->child);
234 if (entry->proc_handler == NULL)
235 kfree(entry->procname);
236 }
237
238 kfree(*tablep);
239 *tablep = NULL;
240}
241
242static int min_load_idx = 0;
243static int max_load_idx = CPU_LOAD_IDX_MAX-1;
244
245static void
246set_table_entry(struct ctl_table *entry,
247 const char *procname, void *data, int maxlen,
248 umode_t mode, proc_handler *proc_handler,
249 bool load_idx)
250{
251 entry->procname = procname;
252 entry->data = data;
253 entry->maxlen = maxlen;
254 entry->mode = mode;
255 entry->proc_handler = proc_handler;
256
257 if (load_idx) {
258 entry->extra1 = &min_load_idx;
259 entry->extra2 = &max_load_idx;
260 }
261}
262
263static struct ctl_table *
264sd_alloc_ctl_domain_table(struct sched_domain *sd)
265{
266 struct ctl_table *table = sd_alloc_ctl_entry(14);
267
268 if (table == NULL)
269 return NULL;
270
271 set_table_entry(&table[0], "min_interval", &sd->min_interval,
272 sizeof(long), 0644, proc_doulongvec_minmax, false);
273 set_table_entry(&table[1], "max_interval", &sd->max_interval,
274 sizeof(long), 0644, proc_doulongvec_minmax, false);
275 set_table_entry(&table[2], "busy_idx", &sd->busy_idx,
276 sizeof(int), 0644, proc_dointvec_minmax, true);
277 set_table_entry(&table[3], "idle_idx", &sd->idle_idx,
278 sizeof(int), 0644, proc_dointvec_minmax, true);
279 set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx,
280 sizeof(int), 0644, proc_dointvec_minmax, true);
281 set_table_entry(&table[5], "wake_idx", &sd->wake_idx,
282 sizeof(int), 0644, proc_dointvec_minmax, true);
283 set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx,
284 sizeof(int), 0644, proc_dointvec_minmax, true);
285 set_table_entry(&table[7], "busy_factor", &sd->busy_factor,
286 sizeof(int), 0644, proc_dointvec_minmax, false);
287 set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct,
288 sizeof(int), 0644, proc_dointvec_minmax, false);
289 set_table_entry(&table[9], "cache_nice_tries",
290 &sd->cache_nice_tries,
291 sizeof(int), 0644, proc_dointvec_minmax, false);
292 set_table_entry(&table[10], "flags", &sd->flags,
293 sizeof(int), 0644, proc_dointvec_minmax, false);
294 set_table_entry(&table[11], "max_newidle_lb_cost",
295 &sd->max_newidle_lb_cost,
296 sizeof(long), 0644, proc_doulongvec_minmax, false);
297 set_table_entry(&table[12], "name", sd->name,
298 CORENAME_MAX_SIZE, 0444, proc_dostring, false);
299 /* &table[13] is terminator */
300
301 return table;
302}
303
304static struct ctl_table *sd_alloc_ctl_cpu_table(int cpu)
305{
306 struct ctl_table *entry, *table;
307 struct sched_domain *sd;
308 int domain_num = 0, i;
309 char buf[32];
310
311 for_each_domain(cpu, sd)
312 domain_num++;
313 entry = table = sd_alloc_ctl_entry(domain_num + 1);
314 if (table == NULL)
315 return NULL;
316
317 i = 0;
318 for_each_domain(cpu, sd) {
319 snprintf(buf, 32, "domain%d", i);
320 entry->procname = kstrdup(buf, GFP_KERNEL);
321 entry->mode = 0555;
322 entry->child = sd_alloc_ctl_domain_table(sd);
323 entry++;
324 i++;
325 }
326 return table;
327}
328
329static struct ctl_table_header *sd_sysctl_header;
330void register_sched_domain_sysctl(void)
331{
332 int i, cpu_num = num_possible_cpus();
333 struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1);
334 char buf[32];
335
336 WARN_ON(sd_ctl_dir[0].child);
337 sd_ctl_dir[0].child = entry;
338
339 if (entry == NULL)
340 return;
341
342 for_each_possible_cpu(i) {
343 snprintf(buf, 32, "cpu%d", i);
344 entry->procname = kstrdup(buf, GFP_KERNEL);
345 entry->mode = 0555;
346 entry->child = sd_alloc_ctl_cpu_table(i);
347 entry++;
348 }
349
350 WARN_ON(sd_sysctl_header);
351 sd_sysctl_header = register_sysctl_table(sd_ctl_root);
352}
353
354/* may be called multiple times per register */
355void unregister_sched_domain_sysctl(void)
356{
357 unregister_sysctl_table(sd_sysctl_header);
358 sd_sysctl_header = NULL;
359 if (sd_ctl_dir[0].child)
360 sd_free_ctl_entry(&sd_ctl_dir[0].child);
361}
362#endif /* CONFIG_SYSCTL */
363#endif /* CONFIG_SMP */
364
365#ifdef CONFIG_FAIR_GROUP_SCHED
366static void print_cfs_group_stats(struct seq_file *m, int cpu, struct task_group *tg)
367{
368 struct sched_entity *se = tg->se[cpu];
369
370#define P(F) \
371 SEQ_printf(m, " .%-30s: %lld\n", #F, (long long)F)
372#define PN(F) \
373 SEQ_printf(m, " .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F))
374
375 if (!se)
376 return;
377
378 PN(se->exec_start);
379 PN(se->vruntime);
380 PN(se->sum_exec_runtime);
381#ifdef CONFIG_SCHEDSTATS
382 if (schedstat_enabled()) {
383 PN(se->statistics.wait_start);
384 PN(se->statistics.sleep_start);
385 PN(se->statistics.block_start);
386 PN(se->statistics.sleep_max);
387 PN(se->statistics.block_max);
388 PN(se->statistics.exec_max);
389 PN(se->statistics.slice_max);
390 PN(se->statistics.wait_max);
391 PN(se->statistics.wait_sum);
392 P(se->statistics.wait_count);
393 }
394#endif
395 P(se->load.weight);
396#ifdef CONFIG_SMP
397 P(se->avg.load_avg);
398 P(se->avg.util_avg);
399#endif
400#undef PN
401#undef P
402}
403#endif
404
405#ifdef CONFIG_CGROUP_SCHED
406static char group_path[PATH_MAX];
407
408static char *task_group_path(struct task_group *tg)
409{
410 if (autogroup_path(tg, group_path, PATH_MAX))
411 return group_path;
412
413 return cgroup_path(tg->css.cgroup, group_path, PATH_MAX);
414}
415#endif
416
417static void
418print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
419{
420 if (rq->curr == p)
421 SEQ_printf(m, "R");
422 else
423 SEQ_printf(m, " ");
424
425 SEQ_printf(m, "%15s %5d %9Ld.%06ld %9Ld %5d ",
426 p->comm, task_pid_nr(p),
427 SPLIT_NS(p->se.vruntime),
428 (long long)(p->nvcsw + p->nivcsw),
429 p->prio);
430#ifdef CONFIG_SCHEDSTATS
431 if (schedstat_enabled()) {
432 SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
433 SPLIT_NS(p->se.statistics.wait_sum),
434 SPLIT_NS(p->se.sum_exec_runtime),
435 SPLIT_NS(p->se.statistics.sum_sleep_runtime));
436 }
437#else
438 SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
439 0LL, 0L,
440 SPLIT_NS(p->se.sum_exec_runtime),
441 0LL, 0L);
442#endif
443#ifdef CONFIG_NUMA_BALANCING
444 SEQ_printf(m, " %d %d", task_node(p), task_numa_group_id(p));
445#endif
446#ifdef CONFIG_CGROUP_SCHED
447 SEQ_printf(m, " %s", task_group_path(task_group(p)));
448#endif
449
450 SEQ_printf(m, "\n");
451}
452
453static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
454{
455 struct task_struct *g, *p;
456
457 SEQ_printf(m,
458 "\nrunnable tasks:\n"
459 " task PID tree-key switches prio"
460 " wait-time sum-exec sum-sleep\n"
461 "------------------------------------------------------"
462 "----------------------------------------------------\n");
463
464 rcu_read_lock();
465 for_each_process_thread(g, p) {
466 if (task_cpu(p) != rq_cpu)
467 continue;
468
469 print_task(m, rq, p);
470 }
471 rcu_read_unlock();
472}
473
474void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
475{
476 s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
477 spread, rq0_min_vruntime, spread0;
478 struct rq *rq = cpu_rq(cpu);
479 struct sched_entity *last;
480 unsigned long flags;
481
482#ifdef CONFIG_FAIR_GROUP_SCHED
483 SEQ_printf(m, "\ncfs_rq[%d]:%s\n", cpu, task_group_path(cfs_rq->tg));
484#else
485 SEQ_printf(m, "\ncfs_rq[%d]:\n", cpu);
486#endif
487 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "exec_clock",
488 SPLIT_NS(cfs_rq->exec_clock));
489
490 raw_spin_lock_irqsave(&rq->lock, flags);
491 if (cfs_rq->rb_leftmost)
492 MIN_vruntime = (__pick_first_entity(cfs_rq))->vruntime;
493 last = __pick_last_entity(cfs_rq);
494 if (last)
495 max_vruntime = last->vruntime;
496 min_vruntime = cfs_rq->min_vruntime;
497 rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
498 raw_spin_unlock_irqrestore(&rq->lock, flags);
499 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "MIN_vruntime",
500 SPLIT_NS(MIN_vruntime));
501 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "min_vruntime",
502 SPLIT_NS(min_vruntime));
503 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "max_vruntime",
504 SPLIT_NS(max_vruntime));
505 spread = max_vruntime - MIN_vruntime;
506 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread",
507 SPLIT_NS(spread));
508 spread0 = min_vruntime - rq0_min_vruntime;
509 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread0",
510 SPLIT_NS(spread0));
511 SEQ_printf(m, " .%-30s: %d\n", "nr_spread_over",
512 cfs_rq->nr_spread_over);
513 SEQ_printf(m, " .%-30s: %d\n", "nr_running", cfs_rq->nr_running);
514 SEQ_printf(m, " .%-30s: %ld\n", "load", cfs_rq->load.weight);
515#ifdef CONFIG_SMP
516 SEQ_printf(m, " .%-30s: %lu\n", "load_avg",
517 cfs_rq->avg.load_avg);
518 SEQ_printf(m, " .%-30s: %lu\n", "runnable_load_avg",
519 cfs_rq->runnable_load_avg);
520 SEQ_printf(m, " .%-30s: %lu\n", "util_avg",
521 cfs_rq->avg.util_avg);
522 SEQ_printf(m, " .%-30s: %ld\n", "removed_load_avg",
523 atomic_long_read(&cfs_rq->removed_load_avg));
524 SEQ_printf(m, " .%-30s: %ld\n", "removed_util_avg",
525 atomic_long_read(&cfs_rq->removed_util_avg));
526#ifdef CONFIG_FAIR_GROUP_SCHED
527 SEQ_printf(m, " .%-30s: %lu\n", "tg_load_avg_contrib",
528 cfs_rq->tg_load_avg_contrib);
529 SEQ_printf(m, " .%-30s: %ld\n", "tg_load_avg",
530 atomic_long_read(&cfs_rq->tg->load_avg));
531#endif
532#endif
533#ifdef CONFIG_CFS_BANDWIDTH
534 SEQ_printf(m, " .%-30s: %d\n", "throttled",
535 cfs_rq->throttled);
536 SEQ_printf(m, " .%-30s: %d\n", "throttle_count",
537 cfs_rq->throttle_count);
538#endif
539
540#ifdef CONFIG_FAIR_GROUP_SCHED
541 print_cfs_group_stats(m, cpu, cfs_rq->tg);
542#endif
543}
544
545void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
546{
547#ifdef CONFIG_RT_GROUP_SCHED
548 SEQ_printf(m, "\nrt_rq[%d]:%s\n", cpu, task_group_path(rt_rq->tg));
549#else
550 SEQ_printf(m, "\nrt_rq[%d]:\n", cpu);
551#endif
552
553#define P(x) \
554 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
555#define PN(x) \
556 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
557
558 P(rt_nr_running);
559 P(rt_throttled);
560 PN(rt_time);
561 PN(rt_runtime);
562
563#undef PN
564#undef P
565}
566
567void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq)
568{
569 struct dl_bw *dl_bw;
570
571 SEQ_printf(m, "\ndl_rq[%d]:\n", cpu);
572 SEQ_printf(m, " .%-30s: %ld\n", "dl_nr_running", dl_rq->dl_nr_running);
573#ifdef CONFIG_SMP
574 dl_bw = &cpu_rq(cpu)->rd->dl_bw;
575#else
576 dl_bw = &dl_rq->dl_bw;
577#endif
578 SEQ_printf(m, " .%-30s: %lld\n", "dl_bw->bw", dl_bw->bw);
579 SEQ_printf(m, " .%-30s: %lld\n", "dl_bw->total_bw", dl_bw->total_bw);
580}
581
582extern __read_mostly int sched_clock_running;
583
584static void print_cpu(struct seq_file *m, int cpu)
585{
586 struct rq *rq = cpu_rq(cpu);
587 unsigned long flags;
588
589#ifdef CONFIG_X86
590 {
591 unsigned int freq = cpu_khz ? : 1;
592
593 SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
594 cpu, freq / 1000, (freq % 1000));
595 }
596#else
597 SEQ_printf(m, "cpu#%d\n", cpu);
598#endif
599
600#define P(x) \
601do { \
602 if (sizeof(rq->x) == 4) \
603 SEQ_printf(m, " .%-30s: %ld\n", #x, (long)(rq->x)); \
604 else \
605 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rq->x));\
606} while (0)
607
608#define PN(x) \
609 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
610
611 P(nr_running);
612 SEQ_printf(m, " .%-30s: %lu\n", "load",
613 rq->load.weight);
614 P(nr_switches);
615 P(nr_load_updates);
616 P(nr_uninterruptible);
617 PN(next_balance);
618 SEQ_printf(m, " .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr)));
619 PN(clock);
620 PN(clock_task);
621 P(cpu_load[0]);
622 P(cpu_load[1]);
623 P(cpu_load[2]);
624 P(cpu_load[3]);
625 P(cpu_load[4]);
626#undef P
627#undef PN
628
629#ifdef CONFIG_SCHEDSTATS
630#define P(n) SEQ_printf(m, " .%-30s: %d\n", #n, rq->n);
631#define P64(n) SEQ_printf(m, " .%-30s: %Ld\n", #n, rq->n);
632
633#ifdef CONFIG_SMP
634 P64(avg_idle);
635 P64(max_idle_balance_cost);
636#endif
637
638 if (schedstat_enabled()) {
639 P(yld_count);
640 P(sched_count);
641 P(sched_goidle);
642 P(ttwu_count);
643 P(ttwu_local);
644 }
645
646#undef P
647#undef P64
648#endif
649 spin_lock_irqsave(&sched_debug_lock, flags);
650 print_cfs_stats(m, cpu);
651 print_rt_stats(m, cpu);
652 print_dl_stats(m, cpu);
653
654 print_rq(m, rq, cpu);
655 spin_unlock_irqrestore(&sched_debug_lock, flags);
656 SEQ_printf(m, "\n");
657}
658
659static const char *sched_tunable_scaling_names[] = {
660 "none",
661 "logaritmic",
662 "linear"
663};
664
665static void sched_debug_header(struct seq_file *m)
666{
667 u64 ktime, sched_clk, cpu_clk;
668 unsigned long flags;
669
670 local_irq_save(flags);
671 ktime = ktime_to_ns(ktime_get());
672 sched_clk = sched_clock();
673 cpu_clk = local_clock();
674 local_irq_restore(flags);
675
676 SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n",
677 init_utsname()->release,
678 (int)strcspn(init_utsname()->version, " "),
679 init_utsname()->version);
680
681#define P(x) \
682 SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
683#define PN(x) \
684 SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
685 PN(ktime);
686 PN(sched_clk);
687 PN(cpu_clk);
688 P(jiffies);
689#ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
690 P(sched_clock_stable());
691#endif
692#undef PN
693#undef P
694
695 SEQ_printf(m, "\n");
696 SEQ_printf(m, "sysctl_sched\n");
697
698#define P(x) \
699 SEQ_printf(m, " .%-40s: %Ld\n", #x, (long long)(x))
700#define PN(x) \
701 SEQ_printf(m, " .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
702 PN(sysctl_sched_latency);
703 PN(sysctl_sched_min_granularity);
704 PN(sysctl_sched_wakeup_granularity);
705 P(sysctl_sched_child_runs_first);
706 P(sysctl_sched_features);
707#undef PN
708#undef P
709
710 SEQ_printf(m, " .%-40s: %d (%s)\n",
711 "sysctl_sched_tunable_scaling",
712 sysctl_sched_tunable_scaling,
713 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
714 SEQ_printf(m, "\n");
715}
716
717static int sched_debug_show(struct seq_file *m, void *v)
718{
719 int cpu = (unsigned long)(v - 2);
720
721 if (cpu != -1)
722 print_cpu(m, cpu);
723 else
724 sched_debug_header(m);
725
726 return 0;
727}
728
729void sysrq_sched_debug_show(void)
730{
731 int cpu;
732
733 sched_debug_header(NULL);
734 for_each_online_cpu(cpu)
735 print_cpu(NULL, cpu);
736
737}
738
739/*
740 * This itererator needs some explanation.
741 * It returns 1 for the header position.
742 * This means 2 is cpu 0.
743 * In a hotplugged system some cpus, including cpu 0, may be missing so we have
744 * to use cpumask_* to iterate over the cpus.
745 */
746static void *sched_debug_start(struct seq_file *file, loff_t *offset)
747{
748 unsigned long n = *offset;
749
750 if (n == 0)
751 return (void *) 1;
752
753 n--;
754
755 if (n > 0)
756 n = cpumask_next(n - 1, cpu_online_mask);
757 else
758 n = cpumask_first(cpu_online_mask);
759
760 *offset = n + 1;
761
762 if (n < nr_cpu_ids)
763 return (void *)(unsigned long)(n + 2);
764 return NULL;
765}
766
767static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
768{
769 (*offset)++;
770 return sched_debug_start(file, offset);
771}
772
773static void sched_debug_stop(struct seq_file *file, void *data)
774{
775}
776
777static const struct seq_operations sched_debug_sops = {
778 .start = sched_debug_start,
779 .next = sched_debug_next,
780 .stop = sched_debug_stop,
781 .show = sched_debug_show,
782};
783
784static int sched_debug_release(struct inode *inode, struct file *file)
785{
786 seq_release(inode, file);
787
788 return 0;
789}
790
791static int sched_debug_open(struct inode *inode, struct file *filp)
792{
793 int ret = 0;
794
795 ret = seq_open(filp, &sched_debug_sops);
796
797 return ret;
798}
799
800static const struct file_operations sched_debug_fops = {
801 .open = sched_debug_open,
802 .read = seq_read,
803 .llseek = seq_lseek,
804 .release = sched_debug_release,
805};
806
807static int __init init_sched_debug_procfs(void)
808{
809 struct proc_dir_entry *pe;
810
811 pe = proc_create("sched_debug", 0444, NULL, &sched_debug_fops);
812 if (!pe)
813 return -ENOMEM;
814 return 0;
815}
816
817__initcall(init_sched_debug_procfs);
818
819#define __P(F) \
820 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)F)
821#define P(F) \
822 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)p->F)
823#define __PN(F) \
824 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
825#define PN(F) \
826 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
827
828
829#ifdef CONFIG_NUMA_BALANCING
830void print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
831 unsigned long tpf, unsigned long gsf, unsigned long gpf)
832{
833 SEQ_printf(m, "numa_faults node=%d ", node);
834 SEQ_printf(m, "task_private=%lu task_shared=%lu ", tsf, tpf);
835 SEQ_printf(m, "group_private=%lu group_shared=%lu\n", gsf, gpf);
836}
837#endif
838
839
840static void sched_show_numa(struct task_struct *p, struct seq_file *m)
841{
842#ifdef CONFIG_NUMA_BALANCING
843 struct mempolicy *pol;
844
845 if (p->mm)
846 P(mm->numa_scan_seq);
847
848 task_lock(p);
849 pol = p->mempolicy;
850 if (pol && !(pol->flags & MPOL_F_MORON))
851 pol = NULL;
852 mpol_get(pol);
853 task_unlock(p);
854
855 P(numa_pages_migrated);
856 P(numa_preferred_nid);
857 P(total_numa_faults);
858 SEQ_printf(m, "current_node=%d, numa_group_id=%d\n",
859 task_node(p), task_numa_group_id(p));
860 show_numa_stats(p, m);
861 mpol_put(pol);
862#endif
863}
864
865void proc_sched_show_task(struct task_struct *p, struct seq_file *m)
866{
867 unsigned long nr_switches;
868
869 SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr(p),
870 get_nr_threads(p));
871 SEQ_printf(m,
872 "---------------------------------------------------------"
873 "----------\n");
874#define __P(F) \
875 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)F)
876#define P(F) \
877 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)p->F)
878#define __PN(F) \
879 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
880#define PN(F) \
881 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
882
883 PN(se.exec_start);
884 PN(se.vruntime);
885 PN(se.sum_exec_runtime);
886
887 nr_switches = p->nvcsw + p->nivcsw;
888
889#ifdef CONFIG_SCHEDSTATS
890 P(se.nr_migrations);
891
892 if (schedstat_enabled()) {
893 u64 avg_atom, avg_per_cpu;
894
895 PN(se.statistics.sum_sleep_runtime);
896 PN(se.statistics.wait_start);
897 PN(se.statistics.sleep_start);
898 PN(se.statistics.block_start);
899 PN(se.statistics.sleep_max);
900 PN(se.statistics.block_max);
901 PN(se.statistics.exec_max);
902 PN(se.statistics.slice_max);
903 PN(se.statistics.wait_max);
904 PN(se.statistics.wait_sum);
905 P(se.statistics.wait_count);
906 PN(se.statistics.iowait_sum);
907 P(se.statistics.iowait_count);
908 P(se.statistics.nr_migrations_cold);
909 P(se.statistics.nr_failed_migrations_affine);
910 P(se.statistics.nr_failed_migrations_running);
911 P(se.statistics.nr_failed_migrations_hot);
912 P(se.statistics.nr_forced_migrations);
913 P(se.statistics.nr_wakeups);
914 P(se.statistics.nr_wakeups_sync);
915 P(se.statistics.nr_wakeups_migrate);
916 P(se.statistics.nr_wakeups_local);
917 P(se.statistics.nr_wakeups_remote);
918 P(se.statistics.nr_wakeups_affine);
919 P(se.statistics.nr_wakeups_affine_attempts);
920 P(se.statistics.nr_wakeups_passive);
921 P(se.statistics.nr_wakeups_idle);
922
923 avg_atom = p->se.sum_exec_runtime;
924 if (nr_switches)
925 avg_atom = div64_ul(avg_atom, nr_switches);
926 else
927 avg_atom = -1LL;
928
929 avg_per_cpu = p->se.sum_exec_runtime;
930 if (p->se.nr_migrations) {
931 avg_per_cpu = div64_u64(avg_per_cpu,
932 p->se.nr_migrations);
933 } else {
934 avg_per_cpu = -1LL;
935 }
936
937 __PN(avg_atom);
938 __PN(avg_per_cpu);
939 }
940#endif
941 __P(nr_switches);
942 SEQ_printf(m, "%-45s:%21Ld\n",
943 "nr_voluntary_switches", (long long)p->nvcsw);
944 SEQ_printf(m, "%-45s:%21Ld\n",
945 "nr_involuntary_switches", (long long)p->nivcsw);
946
947 P(se.load.weight);
948#ifdef CONFIG_SMP
949 P(se.avg.load_sum);
950 P(se.avg.util_sum);
951 P(se.avg.load_avg);
952 P(se.avg.util_avg);
953 P(se.avg.last_update_time);
954#endif
955 P(policy);
956 P(prio);
957#undef PN
958#undef __PN
959#undef P
960#undef __P
961
962 {
963 unsigned int this_cpu = raw_smp_processor_id();
964 u64 t0, t1;
965
966 t0 = cpu_clock(this_cpu);
967 t1 = cpu_clock(this_cpu);
968 SEQ_printf(m, "%-45s:%21Ld\n",
969 "clock-delta", (long long)(t1-t0));
970 }
971
972 sched_show_numa(p, m);
973}
974
975void proc_sched_set_task(struct task_struct *p)
976{
977#ifdef CONFIG_SCHEDSTATS
978 memset(&p->se.statistics, 0, sizeof(p->se.statistics));
979#endif
980}