Loading...
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 P_SCHEDSTAT(F) \
373 SEQ_printf(m, " .%-30s: %lld\n", #F, (long long)schedstat_val(F))
374#define PN(F) \
375 SEQ_printf(m, " .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F))
376#define PN_SCHEDSTAT(F) \
377 SEQ_printf(m, " .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)schedstat_val(F)))
378
379 if (!se)
380 return;
381
382 PN(se->exec_start);
383 PN(se->vruntime);
384 PN(se->sum_exec_runtime);
385 if (schedstat_enabled()) {
386 PN_SCHEDSTAT(se->statistics.wait_start);
387 PN_SCHEDSTAT(se->statistics.sleep_start);
388 PN_SCHEDSTAT(se->statistics.block_start);
389 PN_SCHEDSTAT(se->statistics.sleep_max);
390 PN_SCHEDSTAT(se->statistics.block_max);
391 PN_SCHEDSTAT(se->statistics.exec_max);
392 PN_SCHEDSTAT(se->statistics.slice_max);
393 PN_SCHEDSTAT(se->statistics.wait_max);
394 PN_SCHEDSTAT(se->statistics.wait_sum);
395 P_SCHEDSTAT(se->statistics.wait_count);
396 }
397 P(se->load.weight);
398#ifdef CONFIG_SMP
399 P(se->avg.load_avg);
400 P(se->avg.util_avg);
401#endif
402
403#undef PN_SCHEDSTAT
404#undef PN
405#undef P_SCHEDSTAT
406#undef P
407}
408#endif
409
410#ifdef CONFIG_CGROUP_SCHED
411static char group_path[PATH_MAX];
412
413static char *task_group_path(struct task_group *tg)
414{
415 if (autogroup_path(tg, group_path, PATH_MAX))
416 return group_path;
417
418 cgroup_path(tg->css.cgroup, group_path, PATH_MAX);
419 return group_path;
420}
421#endif
422
423static void
424print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
425{
426 if (rq->curr == p)
427 SEQ_printf(m, "R");
428 else
429 SEQ_printf(m, " ");
430
431 SEQ_printf(m, "%15s %5d %9Ld.%06ld %9Ld %5d ",
432 p->comm, task_pid_nr(p),
433 SPLIT_NS(p->se.vruntime),
434 (long long)(p->nvcsw + p->nivcsw),
435 p->prio);
436
437 SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
438 SPLIT_NS(schedstat_val_or_zero(p->se.statistics.wait_sum)),
439 SPLIT_NS(p->se.sum_exec_runtime),
440 SPLIT_NS(schedstat_val_or_zero(p->se.statistics.sum_sleep_runtime)));
441
442#ifdef CONFIG_NUMA_BALANCING
443 SEQ_printf(m, " %d %d", task_node(p), task_numa_group_id(p));
444#endif
445#ifdef CONFIG_CGROUP_SCHED
446 SEQ_printf(m, " %s", task_group_path(task_group(p)));
447#endif
448
449 SEQ_printf(m, "\n");
450}
451
452static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
453{
454 struct task_struct *g, *p;
455
456 SEQ_printf(m,
457 "\nrunnable tasks:\n"
458 " task PID tree-key switches prio"
459 " wait-time sum-exec sum-sleep\n"
460 "------------------------------------------------------"
461 "----------------------------------------------------\n");
462
463 rcu_read_lock();
464 for_each_process_thread(g, p) {
465 if (task_cpu(p) != rq_cpu)
466 continue;
467
468 print_task(m, rq, p);
469 }
470 rcu_read_unlock();
471}
472
473void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
474{
475 s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
476 spread, rq0_min_vruntime, spread0;
477 struct rq *rq = cpu_rq(cpu);
478 struct sched_entity *last;
479 unsigned long flags;
480
481#ifdef CONFIG_FAIR_GROUP_SCHED
482 SEQ_printf(m, "\ncfs_rq[%d]:%s\n", cpu, task_group_path(cfs_rq->tg));
483#else
484 SEQ_printf(m, "\ncfs_rq[%d]:\n", cpu);
485#endif
486 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "exec_clock",
487 SPLIT_NS(cfs_rq->exec_clock));
488
489 raw_spin_lock_irqsave(&rq->lock, flags);
490 if (cfs_rq->rb_leftmost)
491 MIN_vruntime = (__pick_first_entity(cfs_rq))->vruntime;
492 last = __pick_last_entity(cfs_rq);
493 if (last)
494 max_vruntime = last->vruntime;
495 min_vruntime = cfs_rq->min_vruntime;
496 rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
497 raw_spin_unlock_irqrestore(&rq->lock, flags);
498 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "MIN_vruntime",
499 SPLIT_NS(MIN_vruntime));
500 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "min_vruntime",
501 SPLIT_NS(min_vruntime));
502 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "max_vruntime",
503 SPLIT_NS(max_vruntime));
504 spread = max_vruntime - MIN_vruntime;
505 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread",
506 SPLIT_NS(spread));
507 spread0 = min_vruntime - rq0_min_vruntime;
508 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread0",
509 SPLIT_NS(spread0));
510 SEQ_printf(m, " .%-30s: %d\n", "nr_spread_over",
511 cfs_rq->nr_spread_over);
512 SEQ_printf(m, " .%-30s: %d\n", "nr_running", cfs_rq->nr_running);
513 SEQ_printf(m, " .%-30s: %ld\n", "load", cfs_rq->load.weight);
514#ifdef CONFIG_SMP
515 SEQ_printf(m, " .%-30s: %lu\n", "load_avg",
516 cfs_rq->avg.load_avg);
517 SEQ_printf(m, " .%-30s: %lu\n", "runnable_load_avg",
518 cfs_rq->runnable_load_avg);
519 SEQ_printf(m, " .%-30s: %lu\n", "util_avg",
520 cfs_rq->avg.util_avg);
521 SEQ_printf(m, " .%-30s: %ld\n", "removed_load_avg",
522 atomic_long_read(&cfs_rq->removed_load_avg));
523 SEQ_printf(m, " .%-30s: %ld\n", "removed_util_avg",
524 atomic_long_read(&cfs_rq->removed_util_avg));
525#ifdef CONFIG_FAIR_GROUP_SCHED
526 SEQ_printf(m, " .%-30s: %lu\n", "tg_load_avg_contrib",
527 cfs_rq->tg_load_avg_contrib);
528 SEQ_printf(m, " .%-30s: %ld\n", "tg_load_avg",
529 atomic_long_read(&cfs_rq->tg->load_avg));
530#endif
531#endif
532#ifdef CONFIG_CFS_BANDWIDTH
533 SEQ_printf(m, " .%-30s: %d\n", "throttled",
534 cfs_rq->throttled);
535 SEQ_printf(m, " .%-30s: %d\n", "throttle_count",
536 cfs_rq->throttle_count);
537#endif
538
539#ifdef CONFIG_FAIR_GROUP_SCHED
540 print_cfs_group_stats(m, cpu, cfs_rq->tg);
541#endif
542}
543
544void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
545{
546#ifdef CONFIG_RT_GROUP_SCHED
547 SEQ_printf(m, "\nrt_rq[%d]:%s\n", cpu, task_group_path(rt_rq->tg));
548#else
549 SEQ_printf(m, "\nrt_rq[%d]:\n", cpu);
550#endif
551
552#define P(x) \
553 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
554#define PN(x) \
555 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
556
557 P(rt_nr_running);
558 P(rt_throttled);
559 PN(rt_time);
560 PN(rt_runtime);
561
562#undef PN
563#undef P
564}
565
566void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq)
567{
568 struct dl_bw *dl_bw;
569
570 SEQ_printf(m, "\ndl_rq[%d]:\n", cpu);
571 SEQ_printf(m, " .%-30s: %ld\n", "dl_nr_running", dl_rq->dl_nr_running);
572#ifdef CONFIG_SMP
573 dl_bw = &cpu_rq(cpu)->rd->dl_bw;
574#else
575 dl_bw = &dl_rq->dl_bw;
576#endif
577 SEQ_printf(m, " .%-30s: %lld\n", "dl_bw->bw", dl_bw->bw);
578 SEQ_printf(m, " .%-30s: %lld\n", "dl_bw->total_bw", dl_bw->total_bw);
579}
580
581extern __read_mostly int sched_clock_running;
582
583static void print_cpu(struct seq_file *m, int cpu)
584{
585 struct rq *rq = cpu_rq(cpu);
586 unsigned long flags;
587
588#ifdef CONFIG_X86
589 {
590 unsigned int freq = cpu_khz ? : 1;
591
592 SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
593 cpu, freq / 1000, (freq % 1000));
594 }
595#else
596 SEQ_printf(m, "cpu#%d\n", cpu);
597#endif
598
599#define P(x) \
600do { \
601 if (sizeof(rq->x) == 4) \
602 SEQ_printf(m, " .%-30s: %ld\n", #x, (long)(rq->x)); \
603 else \
604 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rq->x));\
605} while (0)
606
607#define PN(x) \
608 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
609
610 P(nr_running);
611 SEQ_printf(m, " .%-30s: %lu\n", "load",
612 rq->load.weight);
613 P(nr_switches);
614 P(nr_load_updates);
615 P(nr_uninterruptible);
616 PN(next_balance);
617 SEQ_printf(m, " .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr)));
618 PN(clock);
619 PN(clock_task);
620 P(cpu_load[0]);
621 P(cpu_load[1]);
622 P(cpu_load[2]);
623 P(cpu_load[3]);
624 P(cpu_load[4]);
625#undef P
626#undef PN
627
628#ifdef CONFIG_SMP
629#define P64(n) SEQ_printf(m, " .%-30s: %Ld\n", #n, rq->n);
630 P64(avg_idle);
631 P64(max_idle_balance_cost);
632#undef P64
633#endif
634
635#define P(n) SEQ_printf(m, " .%-30s: %d\n", #n, schedstat_val(rq->n));
636 if (schedstat_enabled()) {
637 P(yld_count);
638 P(sched_count);
639 P(sched_goidle);
640 P(ttwu_count);
641 P(ttwu_local);
642 }
643#undef P
644
645 spin_lock_irqsave(&sched_debug_lock, flags);
646 print_cfs_stats(m, cpu);
647 print_rt_stats(m, cpu);
648 print_dl_stats(m, cpu);
649
650 print_rq(m, rq, cpu);
651 spin_unlock_irqrestore(&sched_debug_lock, flags);
652 SEQ_printf(m, "\n");
653}
654
655static const char *sched_tunable_scaling_names[] = {
656 "none",
657 "logaritmic",
658 "linear"
659};
660
661static void sched_debug_header(struct seq_file *m)
662{
663 u64 ktime, sched_clk, cpu_clk;
664 unsigned long flags;
665
666 local_irq_save(flags);
667 ktime = ktime_to_ns(ktime_get());
668 sched_clk = sched_clock();
669 cpu_clk = local_clock();
670 local_irq_restore(flags);
671
672 SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n",
673 init_utsname()->release,
674 (int)strcspn(init_utsname()->version, " "),
675 init_utsname()->version);
676
677#define P(x) \
678 SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
679#define PN(x) \
680 SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
681 PN(ktime);
682 PN(sched_clk);
683 PN(cpu_clk);
684 P(jiffies);
685#ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
686 P(sched_clock_stable());
687#endif
688#undef PN
689#undef P
690
691 SEQ_printf(m, "\n");
692 SEQ_printf(m, "sysctl_sched\n");
693
694#define P(x) \
695 SEQ_printf(m, " .%-40s: %Ld\n", #x, (long long)(x))
696#define PN(x) \
697 SEQ_printf(m, " .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
698 PN(sysctl_sched_latency);
699 PN(sysctl_sched_min_granularity);
700 PN(sysctl_sched_wakeup_granularity);
701 P(sysctl_sched_child_runs_first);
702 P(sysctl_sched_features);
703#undef PN
704#undef P
705
706 SEQ_printf(m, " .%-40s: %d (%s)\n",
707 "sysctl_sched_tunable_scaling",
708 sysctl_sched_tunable_scaling,
709 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
710 SEQ_printf(m, "\n");
711}
712
713static int sched_debug_show(struct seq_file *m, void *v)
714{
715 int cpu = (unsigned long)(v - 2);
716
717 if (cpu != -1)
718 print_cpu(m, cpu);
719 else
720 sched_debug_header(m);
721
722 return 0;
723}
724
725void sysrq_sched_debug_show(void)
726{
727 int cpu;
728
729 sched_debug_header(NULL);
730 for_each_online_cpu(cpu)
731 print_cpu(NULL, cpu);
732
733}
734
735/*
736 * This itererator needs some explanation.
737 * It returns 1 for the header position.
738 * This means 2 is cpu 0.
739 * In a hotplugged system some cpus, including cpu 0, may be missing so we have
740 * to use cpumask_* to iterate over the cpus.
741 */
742static void *sched_debug_start(struct seq_file *file, loff_t *offset)
743{
744 unsigned long n = *offset;
745
746 if (n == 0)
747 return (void *) 1;
748
749 n--;
750
751 if (n > 0)
752 n = cpumask_next(n - 1, cpu_online_mask);
753 else
754 n = cpumask_first(cpu_online_mask);
755
756 *offset = n + 1;
757
758 if (n < nr_cpu_ids)
759 return (void *)(unsigned long)(n + 2);
760 return NULL;
761}
762
763static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
764{
765 (*offset)++;
766 return sched_debug_start(file, offset);
767}
768
769static void sched_debug_stop(struct seq_file *file, void *data)
770{
771}
772
773static const struct seq_operations sched_debug_sops = {
774 .start = sched_debug_start,
775 .next = sched_debug_next,
776 .stop = sched_debug_stop,
777 .show = sched_debug_show,
778};
779
780static int sched_debug_release(struct inode *inode, struct file *file)
781{
782 seq_release(inode, file);
783
784 return 0;
785}
786
787static int sched_debug_open(struct inode *inode, struct file *filp)
788{
789 int ret = 0;
790
791 ret = seq_open(filp, &sched_debug_sops);
792
793 return ret;
794}
795
796static const struct file_operations sched_debug_fops = {
797 .open = sched_debug_open,
798 .read = seq_read,
799 .llseek = seq_lseek,
800 .release = sched_debug_release,
801};
802
803static int __init init_sched_debug_procfs(void)
804{
805 struct proc_dir_entry *pe;
806
807 pe = proc_create("sched_debug", 0444, NULL, &sched_debug_fops);
808 if (!pe)
809 return -ENOMEM;
810 return 0;
811}
812
813__initcall(init_sched_debug_procfs);
814
815#define __P(F) \
816 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)F)
817#define P(F) \
818 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)p->F)
819#define __PN(F) \
820 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
821#define PN(F) \
822 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
823
824
825#ifdef CONFIG_NUMA_BALANCING
826void print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
827 unsigned long tpf, unsigned long gsf, unsigned long gpf)
828{
829 SEQ_printf(m, "numa_faults node=%d ", node);
830 SEQ_printf(m, "task_private=%lu task_shared=%lu ", tsf, tpf);
831 SEQ_printf(m, "group_private=%lu group_shared=%lu\n", gsf, gpf);
832}
833#endif
834
835
836static void sched_show_numa(struct task_struct *p, struct seq_file *m)
837{
838#ifdef CONFIG_NUMA_BALANCING
839 struct mempolicy *pol;
840
841 if (p->mm)
842 P(mm->numa_scan_seq);
843
844 task_lock(p);
845 pol = p->mempolicy;
846 if (pol && !(pol->flags & MPOL_F_MORON))
847 pol = NULL;
848 mpol_get(pol);
849 task_unlock(p);
850
851 P(numa_pages_migrated);
852 P(numa_preferred_nid);
853 P(total_numa_faults);
854 SEQ_printf(m, "current_node=%d, numa_group_id=%d\n",
855 task_node(p), task_numa_group_id(p));
856 show_numa_stats(p, m);
857 mpol_put(pol);
858#endif
859}
860
861void proc_sched_show_task(struct task_struct *p, struct seq_file *m)
862{
863 unsigned long nr_switches;
864
865 SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr(p),
866 get_nr_threads(p));
867 SEQ_printf(m,
868 "---------------------------------------------------------"
869 "----------\n");
870#define __P(F) \
871 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)F)
872#define P(F) \
873 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)p->F)
874#define P_SCHEDSTAT(F) \
875 SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)schedstat_val(p->F))
876#define __PN(F) \
877 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
878#define PN(F) \
879 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
880#define PN_SCHEDSTAT(F) \
881 SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)schedstat_val(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 P(se.nr_migrations);
890
891 if (schedstat_enabled()) {
892 u64 avg_atom, avg_per_cpu;
893
894 PN_SCHEDSTAT(se.statistics.sum_sleep_runtime);
895 PN_SCHEDSTAT(se.statistics.wait_start);
896 PN_SCHEDSTAT(se.statistics.sleep_start);
897 PN_SCHEDSTAT(se.statistics.block_start);
898 PN_SCHEDSTAT(se.statistics.sleep_max);
899 PN_SCHEDSTAT(se.statistics.block_max);
900 PN_SCHEDSTAT(se.statistics.exec_max);
901 PN_SCHEDSTAT(se.statistics.slice_max);
902 PN_SCHEDSTAT(se.statistics.wait_max);
903 PN_SCHEDSTAT(se.statistics.wait_sum);
904 P_SCHEDSTAT(se.statistics.wait_count);
905 PN_SCHEDSTAT(se.statistics.iowait_sum);
906 P_SCHEDSTAT(se.statistics.iowait_count);
907 P_SCHEDSTAT(se.statistics.nr_migrations_cold);
908 P_SCHEDSTAT(se.statistics.nr_failed_migrations_affine);
909 P_SCHEDSTAT(se.statistics.nr_failed_migrations_running);
910 P_SCHEDSTAT(se.statistics.nr_failed_migrations_hot);
911 P_SCHEDSTAT(se.statistics.nr_forced_migrations);
912 P_SCHEDSTAT(se.statistics.nr_wakeups);
913 P_SCHEDSTAT(se.statistics.nr_wakeups_sync);
914 P_SCHEDSTAT(se.statistics.nr_wakeups_migrate);
915 P_SCHEDSTAT(se.statistics.nr_wakeups_local);
916 P_SCHEDSTAT(se.statistics.nr_wakeups_remote);
917 P_SCHEDSTAT(se.statistics.nr_wakeups_affine);
918 P_SCHEDSTAT(se.statistics.nr_wakeups_affine_attempts);
919 P_SCHEDSTAT(se.statistics.nr_wakeups_passive);
920 P_SCHEDSTAT(se.statistics.nr_wakeups_idle);
921
922 avg_atom = p->se.sum_exec_runtime;
923 if (nr_switches)
924 avg_atom = div64_ul(avg_atom, nr_switches);
925 else
926 avg_atom = -1LL;
927
928 avg_per_cpu = p->se.sum_exec_runtime;
929 if (p->se.nr_migrations) {
930 avg_per_cpu = div64_u64(avg_per_cpu,
931 p->se.nr_migrations);
932 } else {
933 avg_per_cpu = -1LL;
934 }
935
936 __PN(avg_atom);
937 __PN(avg_per_cpu);
938 }
939
940 __P(nr_switches);
941 SEQ_printf(m, "%-45s:%21Ld\n",
942 "nr_voluntary_switches", (long long)p->nvcsw);
943 SEQ_printf(m, "%-45s:%21Ld\n",
944 "nr_involuntary_switches", (long long)p->nivcsw);
945
946 P(se.load.weight);
947#ifdef CONFIG_SMP
948 P(se.avg.load_sum);
949 P(se.avg.util_sum);
950 P(se.avg.load_avg);
951 P(se.avg.util_avg);
952 P(se.avg.last_update_time);
953#endif
954 P(policy);
955 P(prio);
956#undef PN_SCHEDSTAT
957#undef PN
958#undef __PN
959#undef P_SCHEDSTAT
960#undef P
961#undef __P
962
963 {
964 unsigned int this_cpu = raw_smp_processor_id();
965 u64 t0, t1;
966
967 t0 = cpu_clock(this_cpu);
968 t1 = cpu_clock(this_cpu);
969 SEQ_printf(m, "%-45s:%21Ld\n",
970 "clock-delta", (long long)(t1-t0));
971 }
972
973 sched_show_numa(p, m);
974}
975
976void proc_sched_set_task(struct task_struct *p)
977{
978#ifdef CONFIG_SCHEDSTATS
979 memset(&p->se.statistics, 0, sizeof(p->se.statistics));
980#endif
981}
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), 0444, 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#ifdef CONFIG_SMP
406 P(se->avg.load_avg);
407 P(se->avg.util_avg);
408 P(se->avg.runnable_avg);
409#endif
410
411#undef PN_SCHEDSTAT
412#undef PN
413#undef P_SCHEDSTAT
414#undef P
415}
416#endif
417
418#ifdef CONFIG_CGROUP_SCHED
419static char group_path[PATH_MAX];
420
421static char *task_group_path(struct task_group *tg)
422{
423 if (autogroup_path(tg, group_path, PATH_MAX))
424 return group_path;
425
426 cgroup_path(tg->css.cgroup, group_path, PATH_MAX);
427
428 return group_path;
429}
430#endif
431
432static void
433print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
434{
435 if (rq->curr == p)
436 SEQ_printf(m, ">R");
437 else
438 SEQ_printf(m, " %c", task_state_to_char(p));
439
440 SEQ_printf(m, " %15s %5d %9Ld.%06ld %9Ld %5d ",
441 p->comm, task_pid_nr(p),
442 SPLIT_NS(p->se.vruntime),
443 (long long)(p->nvcsw + p->nivcsw),
444 p->prio);
445
446 SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
447 SPLIT_NS(schedstat_val_or_zero(p->se.statistics.wait_sum)),
448 SPLIT_NS(p->se.sum_exec_runtime),
449 SPLIT_NS(schedstat_val_or_zero(p->se.statistics.sum_sleep_runtime)));
450
451#ifdef CONFIG_NUMA_BALANCING
452 SEQ_printf(m, " %d %d", task_node(p), task_numa_group_id(p));
453#endif
454#ifdef CONFIG_CGROUP_SCHED
455 SEQ_printf(m, " %s", task_group_path(task_group(p)));
456#endif
457
458 SEQ_printf(m, "\n");
459}
460
461static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
462{
463 struct task_struct *g, *p;
464
465 SEQ_printf(m, "\n");
466 SEQ_printf(m, "runnable tasks:\n");
467 SEQ_printf(m, " S task PID tree-key switches prio"
468 " wait-time sum-exec sum-sleep\n");
469 SEQ_printf(m, "-------------------------------------------------------"
470 "------------------------------------------------------\n");
471
472 rcu_read_lock();
473 for_each_process_thread(g, p) {
474 if (task_cpu(p) != rq_cpu)
475 continue;
476
477 print_task(m, rq, p);
478 }
479 rcu_read_unlock();
480}
481
482void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
483{
484 s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
485 spread, rq0_min_vruntime, spread0;
486 struct rq *rq = cpu_rq(cpu);
487 struct sched_entity *last;
488 unsigned long flags;
489
490#ifdef CONFIG_FAIR_GROUP_SCHED
491 SEQ_printf(m, "\n");
492 SEQ_printf(m, "cfs_rq[%d]:%s\n", cpu, task_group_path(cfs_rq->tg));
493#else
494 SEQ_printf(m, "\n");
495 SEQ_printf(m, "cfs_rq[%d]:\n", cpu);
496#endif
497 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "exec_clock",
498 SPLIT_NS(cfs_rq->exec_clock));
499
500 raw_spin_lock_irqsave(&rq->lock, flags);
501 if (rb_first_cached(&cfs_rq->tasks_timeline))
502 MIN_vruntime = (__pick_first_entity(cfs_rq))->vruntime;
503 last = __pick_last_entity(cfs_rq);
504 if (last)
505 max_vruntime = last->vruntime;
506 min_vruntime = cfs_rq->min_vruntime;
507 rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
508 raw_spin_unlock_irqrestore(&rq->lock, flags);
509 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "MIN_vruntime",
510 SPLIT_NS(MIN_vruntime));
511 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "min_vruntime",
512 SPLIT_NS(min_vruntime));
513 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "max_vruntime",
514 SPLIT_NS(max_vruntime));
515 spread = max_vruntime - MIN_vruntime;
516 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread",
517 SPLIT_NS(spread));
518 spread0 = min_vruntime - rq0_min_vruntime;
519 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread0",
520 SPLIT_NS(spread0));
521 SEQ_printf(m, " .%-30s: %d\n", "nr_spread_over",
522 cfs_rq->nr_spread_over);
523 SEQ_printf(m, " .%-30s: %d\n", "nr_running", cfs_rq->nr_running);
524 SEQ_printf(m, " .%-30s: %ld\n", "load", cfs_rq->load.weight);
525#ifdef CONFIG_SMP
526 SEQ_printf(m, " .%-30s: %lu\n", "load_avg",
527 cfs_rq->avg.load_avg);
528 SEQ_printf(m, " .%-30s: %lu\n", "runnable_avg",
529 cfs_rq->avg.runnable_avg);
530 SEQ_printf(m, " .%-30s: %lu\n", "util_avg",
531 cfs_rq->avg.util_avg);
532 SEQ_printf(m, " .%-30s: %u\n", "util_est_enqueued",
533 cfs_rq->avg.util_est.enqueued);
534 SEQ_printf(m, " .%-30s: %ld\n", "removed.load_avg",
535 cfs_rq->removed.load_avg);
536 SEQ_printf(m, " .%-30s: %ld\n", "removed.util_avg",
537 cfs_rq->removed.util_avg);
538 SEQ_printf(m, " .%-30s: %ld\n", "removed.runnable_avg",
539 cfs_rq->removed.runnable_avg);
540#ifdef CONFIG_FAIR_GROUP_SCHED
541 SEQ_printf(m, " .%-30s: %lu\n", "tg_load_avg_contrib",
542 cfs_rq->tg_load_avg_contrib);
543 SEQ_printf(m, " .%-30s: %ld\n", "tg_load_avg",
544 atomic_long_read(&cfs_rq->tg->load_avg));
545#endif
546#endif
547#ifdef CONFIG_CFS_BANDWIDTH
548 SEQ_printf(m, " .%-30s: %d\n", "throttled",
549 cfs_rq->throttled);
550 SEQ_printf(m, " .%-30s: %d\n", "throttle_count",
551 cfs_rq->throttle_count);
552#endif
553
554#ifdef CONFIG_FAIR_GROUP_SCHED
555 print_cfs_group_stats(m, cpu, cfs_rq->tg);
556#endif
557}
558
559void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
560{
561#ifdef CONFIG_RT_GROUP_SCHED
562 SEQ_printf(m, "\n");
563 SEQ_printf(m, "rt_rq[%d]:%s\n", cpu, task_group_path(rt_rq->tg));
564#else
565 SEQ_printf(m, "\n");
566 SEQ_printf(m, "rt_rq[%d]:\n", cpu);
567#endif
568
569#define P(x) \
570 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
571#define PU(x) \
572 SEQ_printf(m, " .%-30s: %lu\n", #x, (unsigned long)(rt_rq->x))
573#define PN(x) \
574 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
575
576 PU(rt_nr_running);
577#ifdef CONFIG_SMP
578 PU(rt_nr_migratory);
579#endif
580 P(rt_throttled);
581 PN(rt_time);
582 PN(rt_runtime);
583
584#undef PN
585#undef PU
586#undef P
587}
588
589void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq)
590{
591 struct dl_bw *dl_bw;
592
593 SEQ_printf(m, "\n");
594 SEQ_printf(m, "dl_rq[%d]:\n", cpu);
595
596#define PU(x) \
597 SEQ_printf(m, " .%-30s: %lu\n", #x, (unsigned long)(dl_rq->x))
598
599 PU(dl_nr_running);
600#ifdef CONFIG_SMP
601 PU(dl_nr_migratory);
602 dl_bw = &cpu_rq(cpu)->rd->dl_bw;
603#else
604 dl_bw = &dl_rq->dl_bw;
605#endif
606 SEQ_printf(m, " .%-30s: %lld\n", "dl_bw->bw", dl_bw->bw);
607 SEQ_printf(m, " .%-30s: %lld\n", "dl_bw->total_bw", dl_bw->total_bw);
608
609#undef PU
610}
611
612static void print_cpu(struct seq_file *m, int cpu)
613{
614 struct rq *rq = cpu_rq(cpu);
615 unsigned long flags;
616
617#ifdef CONFIG_X86
618 {
619 unsigned int freq = cpu_khz ? : 1;
620
621 SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
622 cpu, freq / 1000, (freq % 1000));
623 }
624#else
625 SEQ_printf(m, "cpu#%d\n", cpu);
626#endif
627
628#define P(x) \
629do { \
630 if (sizeof(rq->x) == 4) \
631 SEQ_printf(m, " .%-30s: %ld\n", #x, (long)(rq->x)); \
632 else \
633 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rq->x));\
634} while (0)
635
636#define PN(x) \
637 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
638
639 P(nr_running);
640 P(nr_switches);
641 P(nr_uninterruptible);
642 PN(next_balance);
643 SEQ_printf(m, " .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr)));
644 PN(clock);
645 PN(clock_task);
646#undef P
647#undef PN
648
649#ifdef CONFIG_SMP
650#define P64(n) SEQ_printf(m, " .%-30s: %Ld\n", #n, rq->n);
651 P64(avg_idle);
652 P64(max_idle_balance_cost);
653#undef P64
654#endif
655
656#define P(n) SEQ_printf(m, " .%-30s: %d\n", #n, schedstat_val(rq->n));
657 if (schedstat_enabled()) {
658 P(yld_count);
659 P(sched_count);
660 P(sched_goidle);
661 P(ttwu_count);
662 P(ttwu_local);
663 }
664#undef P
665
666 spin_lock_irqsave(&sched_debug_lock, flags);
667 print_cfs_stats(m, cpu);
668 print_rt_stats(m, cpu);
669 print_dl_stats(m, cpu);
670
671 print_rq(m, rq, cpu);
672 spin_unlock_irqrestore(&sched_debug_lock, flags);
673 SEQ_printf(m, "\n");
674}
675
676static const char *sched_tunable_scaling_names[] = {
677 "none",
678 "logarithmic",
679 "linear"
680};
681
682static void sched_debug_header(struct seq_file *m)
683{
684 u64 ktime, sched_clk, cpu_clk;
685 unsigned long flags;
686
687 local_irq_save(flags);
688 ktime = ktime_to_ns(ktime_get());
689 sched_clk = sched_clock();
690 cpu_clk = local_clock();
691 local_irq_restore(flags);
692
693 SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n",
694 init_utsname()->release,
695 (int)strcspn(init_utsname()->version, " "),
696 init_utsname()->version);
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(ktime);
703 PN(sched_clk);
704 PN(cpu_clk);
705 P(jiffies);
706#ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
707 P(sched_clock_stable());
708#endif
709#undef PN
710#undef P
711
712 SEQ_printf(m, "\n");
713 SEQ_printf(m, "sysctl_sched\n");
714
715#define P(x) \
716 SEQ_printf(m, " .%-40s: %Ld\n", #x, (long long)(x))
717#define PN(x) \
718 SEQ_printf(m, " .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
719 PN(sysctl_sched_latency);
720 PN(sysctl_sched_min_granularity);
721 PN(sysctl_sched_wakeup_granularity);
722 P(sysctl_sched_child_runs_first);
723 P(sysctl_sched_features);
724#undef PN
725#undef P
726
727 SEQ_printf(m, " .%-40s: %d (%s)\n",
728 "sysctl_sched_tunable_scaling",
729 sysctl_sched_tunable_scaling,
730 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
731 SEQ_printf(m, "\n");
732}
733
734static int sched_debug_show(struct seq_file *m, void *v)
735{
736 int cpu = (unsigned long)(v - 2);
737
738 if (cpu != -1)
739 print_cpu(m, cpu);
740 else
741 sched_debug_header(m);
742
743 return 0;
744}
745
746void sysrq_sched_debug_show(void)
747{
748 int cpu;
749
750 sched_debug_header(NULL);
751 for_each_online_cpu(cpu) {
752 /*
753 * Need to reset softlockup watchdogs on all CPUs, because
754 * another CPU might be blocked waiting for us to process
755 * an IPI or stop_machine.
756 */
757 touch_nmi_watchdog();
758 touch_all_softlockup_watchdogs();
759 print_cpu(NULL, cpu);
760 }
761}
762
763/*
764 * This itererator needs some explanation.
765 * It returns 1 for the header position.
766 * This means 2 is CPU 0.
767 * In a hotplugged system some CPUs, including CPU 0, may be missing so we have
768 * to use cpumask_* to iterate over the CPUs.
769 */
770static void *sched_debug_start(struct seq_file *file, loff_t *offset)
771{
772 unsigned long n = *offset;
773
774 if (n == 0)
775 return (void *) 1;
776
777 n--;
778
779 if (n > 0)
780 n = cpumask_next(n - 1, cpu_online_mask);
781 else
782 n = cpumask_first(cpu_online_mask);
783
784 *offset = n + 1;
785
786 if (n < nr_cpu_ids)
787 return (void *)(unsigned long)(n + 2);
788
789 return NULL;
790}
791
792static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
793{
794 (*offset)++;
795 return sched_debug_start(file, offset);
796}
797
798static void sched_debug_stop(struct seq_file *file, void *data)
799{
800}
801
802static const struct seq_operations sched_debug_sops = {
803 .start = sched_debug_start,
804 .next = sched_debug_next,
805 .stop = sched_debug_stop,
806 .show = sched_debug_show,
807};
808
809static int __init init_sched_debug_procfs(void)
810{
811 if (!proc_create_seq("sched_debug", 0444, NULL, &sched_debug_sops))
812 return -ENOMEM;
813 return 0;
814}
815
816__initcall(init_sched_debug_procfs);
817
818#define __PS(S, F) SEQ_printf(m, "%-45s:%21Ld\n", S, (long long)(F))
819#define __P(F) __PS(#F, F)
820#define P(F) __PS(#F, p->F)
821#define __PSN(S, F) SEQ_printf(m, "%-45s:%14Ld.%06ld\n", S, SPLIT_NS((long long)(F)))
822#define __PN(F) __PSN(#F, F)
823#define PN(F) __PSN(#F, p->F)
824
825
826#ifdef CONFIG_NUMA_BALANCING
827void print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
828 unsigned long tpf, unsigned long gsf, unsigned long gpf)
829{
830 SEQ_printf(m, "numa_faults node=%d ", node);
831 SEQ_printf(m, "task_private=%lu task_shared=%lu ", tpf, tsf);
832 SEQ_printf(m, "group_private=%lu group_shared=%lu\n", gpf, gsf);
833}
834#endif
835
836
837static void sched_show_numa(struct task_struct *p, struct seq_file *m)
838{
839#ifdef CONFIG_NUMA_BALANCING
840 struct mempolicy *pol;
841
842 if (p->mm)
843 P(mm->numa_scan_seq);
844
845 task_lock(p);
846 pol = p->mempolicy;
847 if (pol && !(pol->flags & MPOL_F_MORON))
848 pol = NULL;
849 mpol_get(pol);
850 task_unlock(p);
851
852 P(numa_pages_migrated);
853 P(numa_preferred_nid);
854 P(total_numa_faults);
855 SEQ_printf(m, "current_node=%d, numa_group_id=%d\n",
856 task_node(p), task_numa_group_id(p));
857 show_numa_stats(p, m);
858 mpol_put(pol);
859#endif
860}
861
862void proc_sched_show_task(struct task_struct *p, struct pid_namespace *ns,
863 struct seq_file *m)
864{
865 unsigned long nr_switches;
866
867 SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr_ns(p, ns),
868 get_nr_threads(p));
869 SEQ_printf(m,
870 "---------------------------------------------------------"
871 "----------\n");
872
873#define P_SCHEDSTAT(F) __PS(#F, schedstat_val(p->F))
874#define PN_SCHEDSTAT(F) __PSN(#F, schedstat_val(p->F))
875
876 PN(se.exec_start);
877 PN(se.vruntime);
878 PN(se.sum_exec_runtime);
879
880 nr_switches = p->nvcsw + p->nivcsw;
881
882 P(se.nr_migrations);
883
884 if (schedstat_enabled()) {
885 u64 avg_atom, avg_per_cpu;
886
887 PN_SCHEDSTAT(se.statistics.sum_sleep_runtime);
888 PN_SCHEDSTAT(se.statistics.wait_start);
889 PN_SCHEDSTAT(se.statistics.sleep_start);
890 PN_SCHEDSTAT(se.statistics.block_start);
891 PN_SCHEDSTAT(se.statistics.sleep_max);
892 PN_SCHEDSTAT(se.statistics.block_max);
893 PN_SCHEDSTAT(se.statistics.exec_max);
894 PN_SCHEDSTAT(se.statistics.slice_max);
895 PN_SCHEDSTAT(se.statistics.wait_max);
896 PN_SCHEDSTAT(se.statistics.wait_sum);
897 P_SCHEDSTAT(se.statistics.wait_count);
898 PN_SCHEDSTAT(se.statistics.iowait_sum);
899 P_SCHEDSTAT(se.statistics.iowait_count);
900 P_SCHEDSTAT(se.statistics.nr_migrations_cold);
901 P_SCHEDSTAT(se.statistics.nr_failed_migrations_affine);
902 P_SCHEDSTAT(se.statistics.nr_failed_migrations_running);
903 P_SCHEDSTAT(se.statistics.nr_failed_migrations_hot);
904 P_SCHEDSTAT(se.statistics.nr_forced_migrations);
905 P_SCHEDSTAT(se.statistics.nr_wakeups);
906 P_SCHEDSTAT(se.statistics.nr_wakeups_sync);
907 P_SCHEDSTAT(se.statistics.nr_wakeups_migrate);
908 P_SCHEDSTAT(se.statistics.nr_wakeups_local);
909 P_SCHEDSTAT(se.statistics.nr_wakeups_remote);
910 P_SCHEDSTAT(se.statistics.nr_wakeups_affine);
911 P_SCHEDSTAT(se.statistics.nr_wakeups_affine_attempts);
912 P_SCHEDSTAT(se.statistics.nr_wakeups_passive);
913 P_SCHEDSTAT(se.statistics.nr_wakeups_idle);
914
915 avg_atom = p->se.sum_exec_runtime;
916 if (nr_switches)
917 avg_atom = div64_ul(avg_atom, nr_switches);
918 else
919 avg_atom = -1LL;
920
921 avg_per_cpu = p->se.sum_exec_runtime;
922 if (p->se.nr_migrations) {
923 avg_per_cpu = div64_u64(avg_per_cpu,
924 p->se.nr_migrations);
925 } else {
926 avg_per_cpu = -1LL;
927 }
928
929 __PN(avg_atom);
930 __PN(avg_per_cpu);
931 }
932
933 __P(nr_switches);
934 __PS("nr_voluntary_switches", p->nvcsw);
935 __PS("nr_involuntary_switches", p->nivcsw);
936
937 P(se.load.weight);
938#ifdef CONFIG_SMP
939 P(se.avg.load_sum);
940 P(se.avg.runnable_sum);
941 P(se.avg.util_sum);
942 P(se.avg.load_avg);
943 P(se.avg.runnable_avg);
944 P(se.avg.util_avg);
945 P(se.avg.last_update_time);
946 P(se.avg.util_est.ewma);
947 P(se.avg.util_est.enqueued);
948#endif
949#ifdef CONFIG_UCLAMP_TASK
950 __PS("uclamp.min", p->uclamp_req[UCLAMP_MIN].value);
951 __PS("uclamp.max", p->uclamp_req[UCLAMP_MAX].value);
952 __PS("effective uclamp.min", uclamp_eff_value(p, UCLAMP_MIN));
953 __PS("effective uclamp.max", uclamp_eff_value(p, UCLAMP_MAX));
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 P_SCHEDSTAT
963
964 {
965 unsigned int this_cpu = raw_smp_processor_id();
966 u64 t0, t1;
967
968 t0 = cpu_clock(this_cpu);
969 t1 = cpu_clock(this_cpu);
970 __PS("clock-delta", t1-t0);
971 }
972
973 sched_show_numa(p, m);
974}
975
976void proc_sched_set_task(struct task_struct *p)
977{
978#ifdef CONFIG_SCHEDSTATS
979 memset(&p->se.statistics, 0, sizeof(p->se.statistics));
980#endif
981}