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 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}
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
11/*
12 * This allows printing both to /proc/sched_debug and
13 * to the console
14 */
15#define SEQ_printf(m, x...) \
16 do { \
17 if (m) \
18 seq_printf(m, x); \
19 else \
20 pr_cont(x); \
21 } while (0)
22
23/*
24 * Ease the printing of nsec fields:
25 */
26static long long nsec_high(unsigned long long nsec)
27{
28 if ((long long)nsec < 0) {
29 nsec = -nsec;
30 do_div(nsec, 1000000);
31 return -nsec;
32 }
33 do_div(nsec, 1000000);
34
35 return nsec;
36}
37
38static unsigned long nsec_low(unsigned long long nsec)
39{
40 if ((long long)nsec < 0)
41 nsec = -nsec;
42
43 return do_div(nsec, 1000000);
44}
45
46#define SPLIT_NS(x) nsec_high(x), nsec_low(x)
47
48#define SCHED_FEAT(name, enabled) \
49 #name ,
50
51static const char * const sched_feat_names[] = {
52#include "features.h"
53};
54
55#undef SCHED_FEAT
56
57static int sched_feat_show(struct seq_file *m, void *v)
58{
59 int i;
60
61 for (i = 0; i < __SCHED_FEAT_NR; i++) {
62 if (!(sysctl_sched_features & (1UL << i)))
63 seq_puts(m, "NO_");
64 seq_printf(m, "%s ", sched_feat_names[i]);
65 }
66 seq_puts(m, "\n");
67
68 return 0;
69}
70
71#ifdef CONFIG_JUMP_LABEL
72
73#define jump_label_key__true STATIC_KEY_INIT_TRUE
74#define jump_label_key__false STATIC_KEY_INIT_FALSE
75
76#define SCHED_FEAT(name, enabled) \
77 jump_label_key__##enabled ,
78
79struct static_key sched_feat_keys[__SCHED_FEAT_NR] = {
80#include "features.h"
81};
82
83#undef SCHED_FEAT
84
85static void sched_feat_disable(int i)
86{
87 static_key_disable_cpuslocked(&sched_feat_keys[i]);
88}
89
90static void sched_feat_enable(int i)
91{
92 static_key_enable_cpuslocked(&sched_feat_keys[i]);
93}
94#else
95static void sched_feat_disable(int i) { };
96static void sched_feat_enable(int i) { };
97#endif /* CONFIG_JUMP_LABEL */
98
99static int sched_feat_set(char *cmp)
100{
101 int i;
102 int neg = 0;
103
104 if (strncmp(cmp, "NO_", 3) == 0) {
105 neg = 1;
106 cmp += 3;
107 }
108
109 i = match_string(sched_feat_names, __SCHED_FEAT_NR, cmp);
110 if (i < 0)
111 return i;
112
113 if (neg) {
114 sysctl_sched_features &= ~(1UL << i);
115 sched_feat_disable(i);
116 } else {
117 sysctl_sched_features |= (1UL << i);
118 sched_feat_enable(i);
119 }
120
121 return 0;
122}
123
124static ssize_t
125sched_feat_write(struct file *filp, const char __user *ubuf,
126 size_t cnt, loff_t *ppos)
127{
128 char buf[64];
129 char *cmp;
130 int ret;
131 struct inode *inode;
132
133 if (cnt > 63)
134 cnt = 63;
135
136 if (copy_from_user(&buf, ubuf, cnt))
137 return -EFAULT;
138
139 buf[cnt] = 0;
140 cmp = strstrip(buf);
141
142 /* Ensure the static_key remains in a consistent state */
143 inode = file_inode(filp);
144 cpus_read_lock();
145 inode_lock(inode);
146 ret = sched_feat_set(cmp);
147 inode_unlock(inode);
148 cpus_read_unlock();
149 if (ret < 0)
150 return ret;
151
152 *ppos += cnt;
153
154 return cnt;
155}
156
157static int sched_feat_open(struct inode *inode, struct file *filp)
158{
159 return single_open(filp, sched_feat_show, NULL);
160}
161
162static const struct file_operations sched_feat_fops = {
163 .open = sched_feat_open,
164 .write = sched_feat_write,
165 .read = seq_read,
166 .llseek = seq_lseek,
167 .release = single_release,
168};
169
170#ifdef CONFIG_SMP
171
172static ssize_t sched_scaling_write(struct file *filp, const char __user *ubuf,
173 size_t cnt, loff_t *ppos)
174{
175 char buf[16];
176 unsigned int scaling;
177
178 if (cnt > 15)
179 cnt = 15;
180
181 if (copy_from_user(&buf, ubuf, cnt))
182 return -EFAULT;
183 buf[cnt] = '\0';
184
185 if (kstrtouint(buf, 10, &scaling))
186 return -EINVAL;
187
188 if (scaling >= SCHED_TUNABLESCALING_END)
189 return -EINVAL;
190
191 sysctl_sched_tunable_scaling = scaling;
192 if (sched_update_scaling())
193 return -EINVAL;
194
195 *ppos += cnt;
196 return cnt;
197}
198
199static int sched_scaling_show(struct seq_file *m, void *v)
200{
201 seq_printf(m, "%d\n", sysctl_sched_tunable_scaling);
202 return 0;
203}
204
205static int sched_scaling_open(struct inode *inode, struct file *filp)
206{
207 return single_open(filp, sched_scaling_show, NULL);
208}
209
210static const struct file_operations sched_scaling_fops = {
211 .open = sched_scaling_open,
212 .write = sched_scaling_write,
213 .read = seq_read,
214 .llseek = seq_lseek,
215 .release = single_release,
216};
217
218#endif /* SMP */
219
220#ifdef CONFIG_PREEMPT_DYNAMIC
221
222static ssize_t sched_dynamic_write(struct file *filp, const char __user *ubuf,
223 size_t cnt, loff_t *ppos)
224{
225 char buf[16];
226 int mode;
227
228 if (cnt > 15)
229 cnt = 15;
230
231 if (copy_from_user(&buf, ubuf, cnt))
232 return -EFAULT;
233
234 buf[cnt] = 0;
235 mode = sched_dynamic_mode(strstrip(buf));
236 if (mode < 0)
237 return mode;
238
239 sched_dynamic_update(mode);
240
241 *ppos += cnt;
242
243 return cnt;
244}
245
246static int sched_dynamic_show(struct seq_file *m, void *v)
247{
248 static const char * preempt_modes[] = {
249 "none", "voluntary", "full"
250 };
251 int i;
252
253 for (i = 0; i < ARRAY_SIZE(preempt_modes); i++) {
254 if (preempt_dynamic_mode == i)
255 seq_puts(m, "(");
256 seq_puts(m, preempt_modes[i]);
257 if (preempt_dynamic_mode == i)
258 seq_puts(m, ")");
259
260 seq_puts(m, " ");
261 }
262
263 seq_puts(m, "\n");
264 return 0;
265}
266
267static int sched_dynamic_open(struct inode *inode, struct file *filp)
268{
269 return single_open(filp, sched_dynamic_show, NULL);
270}
271
272static const struct file_operations sched_dynamic_fops = {
273 .open = sched_dynamic_open,
274 .write = sched_dynamic_write,
275 .read = seq_read,
276 .llseek = seq_lseek,
277 .release = single_release,
278};
279
280#endif /* CONFIG_PREEMPT_DYNAMIC */
281
282__read_mostly bool sched_debug_verbose;
283
284static const struct seq_operations sched_debug_sops;
285
286static int sched_debug_open(struct inode *inode, struct file *filp)
287{
288 return seq_open(filp, &sched_debug_sops);
289}
290
291static const struct file_operations sched_debug_fops = {
292 .open = sched_debug_open,
293 .read = seq_read,
294 .llseek = seq_lseek,
295 .release = seq_release,
296};
297
298static struct dentry *debugfs_sched;
299
300static __init int sched_init_debug(void)
301{
302 struct dentry __maybe_unused *numa;
303
304 debugfs_sched = debugfs_create_dir("sched", NULL);
305
306 debugfs_create_file("features", 0644, debugfs_sched, NULL, &sched_feat_fops);
307 debugfs_create_bool("verbose", 0644, debugfs_sched, &sched_debug_verbose);
308#ifdef CONFIG_PREEMPT_DYNAMIC
309 debugfs_create_file("preempt", 0644, debugfs_sched, NULL, &sched_dynamic_fops);
310#endif
311
312 debugfs_create_u32("latency_ns", 0644, debugfs_sched, &sysctl_sched_latency);
313 debugfs_create_u32("min_granularity_ns", 0644, debugfs_sched, &sysctl_sched_min_granularity);
314 debugfs_create_u32("wakeup_granularity_ns", 0644, debugfs_sched, &sysctl_sched_wakeup_granularity);
315
316 debugfs_create_u32("latency_warn_ms", 0644, debugfs_sched, &sysctl_resched_latency_warn_ms);
317 debugfs_create_u32("latency_warn_once", 0644, debugfs_sched, &sysctl_resched_latency_warn_once);
318
319#ifdef CONFIG_SMP
320 debugfs_create_file("tunable_scaling", 0644, debugfs_sched, NULL, &sched_scaling_fops);
321 debugfs_create_u32("migration_cost_ns", 0644, debugfs_sched, &sysctl_sched_migration_cost);
322 debugfs_create_u32("nr_migrate", 0644, debugfs_sched, &sysctl_sched_nr_migrate);
323
324 mutex_lock(&sched_domains_mutex);
325 update_sched_domain_debugfs();
326 mutex_unlock(&sched_domains_mutex);
327#endif
328
329#ifdef CONFIG_NUMA_BALANCING
330 numa = debugfs_create_dir("numa_balancing", debugfs_sched);
331
332 debugfs_create_u32("scan_delay_ms", 0644, numa, &sysctl_numa_balancing_scan_delay);
333 debugfs_create_u32("scan_period_min_ms", 0644, numa, &sysctl_numa_balancing_scan_period_min);
334 debugfs_create_u32("scan_period_max_ms", 0644, numa, &sysctl_numa_balancing_scan_period_max);
335 debugfs_create_u32("scan_size_mb", 0644, numa, &sysctl_numa_balancing_scan_size);
336#endif
337
338 debugfs_create_file("debug", 0444, debugfs_sched, NULL, &sched_debug_fops);
339
340 return 0;
341}
342late_initcall(sched_init_debug);
343
344#ifdef CONFIG_SMP
345
346static cpumask_var_t sd_sysctl_cpus;
347static struct dentry *sd_dentry;
348
349static int sd_flags_show(struct seq_file *m, void *v)
350{
351 unsigned long flags = *(unsigned int *)m->private;
352 int idx;
353
354 for_each_set_bit(idx, &flags, __SD_FLAG_CNT) {
355 seq_puts(m, sd_flag_debug[idx].name);
356 seq_puts(m, " ");
357 }
358 seq_puts(m, "\n");
359
360 return 0;
361}
362
363static int sd_flags_open(struct inode *inode, struct file *file)
364{
365 return single_open(file, sd_flags_show, inode->i_private);
366}
367
368static const struct file_operations sd_flags_fops = {
369 .open = sd_flags_open,
370 .read = seq_read,
371 .llseek = seq_lseek,
372 .release = single_release,
373};
374
375static void register_sd(struct sched_domain *sd, struct dentry *parent)
376{
377#define SDM(type, mode, member) \
378 debugfs_create_##type(#member, mode, parent, &sd->member)
379
380 SDM(ulong, 0644, min_interval);
381 SDM(ulong, 0644, max_interval);
382 SDM(u64, 0644, max_newidle_lb_cost);
383 SDM(u32, 0644, busy_factor);
384 SDM(u32, 0644, imbalance_pct);
385 SDM(u32, 0644, cache_nice_tries);
386 SDM(str, 0444, name);
387
388#undef SDM
389
390 debugfs_create_file("flags", 0444, parent, &sd->flags, &sd_flags_fops);
391}
392
393void update_sched_domain_debugfs(void)
394{
395 int cpu, i;
396
397 /*
398 * This can unfortunately be invoked before sched_debug_init() creates
399 * the debug directory. Don't touch sd_sysctl_cpus until then.
400 */
401 if (!debugfs_sched)
402 return;
403
404 if (!cpumask_available(sd_sysctl_cpus)) {
405 if (!alloc_cpumask_var(&sd_sysctl_cpus, GFP_KERNEL))
406 return;
407 cpumask_copy(sd_sysctl_cpus, cpu_possible_mask);
408 }
409
410 if (!sd_dentry)
411 sd_dentry = debugfs_create_dir("domains", debugfs_sched);
412
413 for_each_cpu(cpu, sd_sysctl_cpus) {
414 struct sched_domain *sd;
415 struct dentry *d_cpu;
416 char buf[32];
417
418 snprintf(buf, sizeof(buf), "cpu%d", cpu);
419 debugfs_remove(debugfs_lookup(buf, sd_dentry));
420 d_cpu = debugfs_create_dir(buf, sd_dentry);
421
422 i = 0;
423 for_each_domain(cpu, sd) {
424 struct dentry *d_sd;
425
426 snprintf(buf, sizeof(buf), "domain%d", i);
427 d_sd = debugfs_create_dir(buf, d_cpu);
428
429 register_sd(sd, d_sd);
430 i++;
431 }
432
433 __cpumask_clear_cpu(cpu, sd_sysctl_cpus);
434 }
435}
436
437void dirty_sched_domain_sysctl(int cpu)
438{
439 if (cpumask_available(sd_sysctl_cpus))
440 __cpumask_set_cpu(cpu, sd_sysctl_cpus);
441}
442
443#endif /* CONFIG_SMP */
444
445#ifdef CONFIG_FAIR_GROUP_SCHED
446static void print_cfs_group_stats(struct seq_file *m, int cpu, struct task_group *tg)
447{
448 struct sched_entity *se = tg->se[cpu];
449
450#define P(F) SEQ_printf(m, " .%-30s: %lld\n", #F, (long long)F)
451#define P_SCHEDSTAT(F) SEQ_printf(m, " .%-30s: %lld\n", #F, (long long)schedstat_val(F))
452#define PN(F) SEQ_printf(m, " .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F))
453#define PN_SCHEDSTAT(F) SEQ_printf(m, " .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)schedstat_val(F)))
454
455 if (!se)
456 return;
457
458 PN(se->exec_start);
459 PN(se->vruntime);
460 PN(se->sum_exec_runtime);
461
462 if (schedstat_enabled()) {
463 PN_SCHEDSTAT(se->statistics.wait_start);
464 PN_SCHEDSTAT(se->statistics.sleep_start);
465 PN_SCHEDSTAT(se->statistics.block_start);
466 PN_SCHEDSTAT(se->statistics.sleep_max);
467 PN_SCHEDSTAT(se->statistics.block_max);
468 PN_SCHEDSTAT(se->statistics.exec_max);
469 PN_SCHEDSTAT(se->statistics.slice_max);
470 PN_SCHEDSTAT(se->statistics.wait_max);
471 PN_SCHEDSTAT(se->statistics.wait_sum);
472 P_SCHEDSTAT(se->statistics.wait_count);
473 }
474
475 P(se->load.weight);
476#ifdef CONFIG_SMP
477 P(se->avg.load_avg);
478 P(se->avg.util_avg);
479 P(se->avg.runnable_avg);
480#endif
481
482#undef PN_SCHEDSTAT
483#undef PN
484#undef P_SCHEDSTAT
485#undef P
486}
487#endif
488
489#ifdef CONFIG_CGROUP_SCHED
490static DEFINE_SPINLOCK(sched_debug_lock);
491static char group_path[PATH_MAX];
492
493static void task_group_path(struct task_group *tg, char *path, int plen)
494{
495 if (autogroup_path(tg, path, plen))
496 return;
497
498 cgroup_path(tg->css.cgroup, path, plen);
499}
500
501/*
502 * Only 1 SEQ_printf_task_group_path() caller can use the full length
503 * group_path[] for cgroup path. Other simultaneous callers will have
504 * to use a shorter stack buffer. A "..." suffix is appended at the end
505 * of the stack buffer so that it will show up in case the output length
506 * matches the given buffer size to indicate possible path name truncation.
507 */
508#define SEQ_printf_task_group_path(m, tg, fmt...) \
509{ \
510 if (spin_trylock(&sched_debug_lock)) { \
511 task_group_path(tg, group_path, sizeof(group_path)); \
512 SEQ_printf(m, fmt, group_path); \
513 spin_unlock(&sched_debug_lock); \
514 } else { \
515 char buf[128]; \
516 char *bufend = buf + sizeof(buf) - 3; \
517 task_group_path(tg, buf, bufend - buf); \
518 strcpy(bufend - 1, "..."); \
519 SEQ_printf(m, fmt, buf); \
520 } \
521}
522#endif
523
524static void
525print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
526{
527 if (task_current(rq, p))
528 SEQ_printf(m, ">R");
529 else
530 SEQ_printf(m, " %c", task_state_to_char(p));
531
532 SEQ_printf(m, " %15s %5d %9Ld.%06ld %9Ld %5d ",
533 p->comm, task_pid_nr(p),
534 SPLIT_NS(p->se.vruntime),
535 (long long)(p->nvcsw + p->nivcsw),
536 p->prio);
537
538 SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
539 SPLIT_NS(schedstat_val_or_zero(p->se.statistics.wait_sum)),
540 SPLIT_NS(p->se.sum_exec_runtime),
541 SPLIT_NS(schedstat_val_or_zero(p->se.statistics.sum_sleep_runtime)));
542
543#ifdef CONFIG_NUMA_BALANCING
544 SEQ_printf(m, " %d %d", task_node(p), task_numa_group_id(p));
545#endif
546#ifdef CONFIG_CGROUP_SCHED
547 SEQ_printf_task_group_path(m, task_group(p), " %s")
548#endif
549
550 SEQ_printf(m, "\n");
551}
552
553static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
554{
555 struct task_struct *g, *p;
556
557 SEQ_printf(m, "\n");
558 SEQ_printf(m, "runnable tasks:\n");
559 SEQ_printf(m, " S task PID tree-key switches prio"
560 " wait-time sum-exec sum-sleep\n");
561 SEQ_printf(m, "-------------------------------------------------------"
562 "------------------------------------------------------\n");
563
564 rcu_read_lock();
565 for_each_process_thread(g, p) {
566 if (task_cpu(p) != rq_cpu)
567 continue;
568
569 print_task(m, rq, p);
570 }
571 rcu_read_unlock();
572}
573
574void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
575{
576 s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
577 spread, rq0_min_vruntime, spread0;
578 struct rq *rq = cpu_rq(cpu);
579 struct sched_entity *last;
580 unsigned long flags;
581
582#ifdef CONFIG_FAIR_GROUP_SCHED
583 SEQ_printf(m, "\n");
584 SEQ_printf_task_group_path(m, cfs_rq->tg, "cfs_rq[%d]:%s\n", cpu);
585#else
586 SEQ_printf(m, "\n");
587 SEQ_printf(m, "cfs_rq[%d]:\n", cpu);
588#endif
589 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "exec_clock",
590 SPLIT_NS(cfs_rq->exec_clock));
591
592 raw_spin_rq_lock_irqsave(rq, flags);
593 if (rb_first_cached(&cfs_rq->tasks_timeline))
594 MIN_vruntime = (__pick_first_entity(cfs_rq))->vruntime;
595 last = __pick_last_entity(cfs_rq);
596 if (last)
597 max_vruntime = last->vruntime;
598 min_vruntime = cfs_rq->min_vruntime;
599 rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
600 raw_spin_rq_unlock_irqrestore(rq, flags);
601 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "MIN_vruntime",
602 SPLIT_NS(MIN_vruntime));
603 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "min_vruntime",
604 SPLIT_NS(min_vruntime));
605 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "max_vruntime",
606 SPLIT_NS(max_vruntime));
607 spread = max_vruntime - MIN_vruntime;
608 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread",
609 SPLIT_NS(spread));
610 spread0 = min_vruntime - rq0_min_vruntime;
611 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread0",
612 SPLIT_NS(spread0));
613 SEQ_printf(m, " .%-30s: %d\n", "nr_spread_over",
614 cfs_rq->nr_spread_over);
615 SEQ_printf(m, " .%-30s: %d\n", "nr_running", cfs_rq->nr_running);
616 SEQ_printf(m, " .%-30s: %ld\n", "load", cfs_rq->load.weight);
617#ifdef CONFIG_SMP
618 SEQ_printf(m, " .%-30s: %lu\n", "load_avg",
619 cfs_rq->avg.load_avg);
620 SEQ_printf(m, " .%-30s: %lu\n", "runnable_avg",
621 cfs_rq->avg.runnable_avg);
622 SEQ_printf(m, " .%-30s: %lu\n", "util_avg",
623 cfs_rq->avg.util_avg);
624 SEQ_printf(m, " .%-30s: %u\n", "util_est_enqueued",
625 cfs_rq->avg.util_est.enqueued);
626 SEQ_printf(m, " .%-30s: %ld\n", "removed.load_avg",
627 cfs_rq->removed.load_avg);
628 SEQ_printf(m, " .%-30s: %ld\n", "removed.util_avg",
629 cfs_rq->removed.util_avg);
630 SEQ_printf(m, " .%-30s: %ld\n", "removed.runnable_avg",
631 cfs_rq->removed.runnable_avg);
632#ifdef CONFIG_FAIR_GROUP_SCHED
633 SEQ_printf(m, " .%-30s: %lu\n", "tg_load_avg_contrib",
634 cfs_rq->tg_load_avg_contrib);
635 SEQ_printf(m, " .%-30s: %ld\n", "tg_load_avg",
636 atomic_long_read(&cfs_rq->tg->load_avg));
637#endif
638#endif
639#ifdef CONFIG_CFS_BANDWIDTH
640 SEQ_printf(m, " .%-30s: %d\n", "throttled",
641 cfs_rq->throttled);
642 SEQ_printf(m, " .%-30s: %d\n", "throttle_count",
643 cfs_rq->throttle_count);
644#endif
645
646#ifdef CONFIG_FAIR_GROUP_SCHED
647 print_cfs_group_stats(m, cpu, cfs_rq->tg);
648#endif
649}
650
651void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
652{
653#ifdef CONFIG_RT_GROUP_SCHED
654 SEQ_printf(m, "\n");
655 SEQ_printf_task_group_path(m, rt_rq->tg, "rt_rq[%d]:%s\n", cpu);
656#else
657 SEQ_printf(m, "\n");
658 SEQ_printf(m, "rt_rq[%d]:\n", cpu);
659#endif
660
661#define P(x) \
662 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
663#define PU(x) \
664 SEQ_printf(m, " .%-30s: %lu\n", #x, (unsigned long)(rt_rq->x))
665#define PN(x) \
666 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
667
668 PU(rt_nr_running);
669#ifdef CONFIG_SMP
670 PU(rt_nr_migratory);
671#endif
672 P(rt_throttled);
673 PN(rt_time);
674 PN(rt_runtime);
675
676#undef PN
677#undef PU
678#undef P
679}
680
681void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq)
682{
683 struct dl_bw *dl_bw;
684
685 SEQ_printf(m, "\n");
686 SEQ_printf(m, "dl_rq[%d]:\n", cpu);
687
688#define PU(x) \
689 SEQ_printf(m, " .%-30s: %lu\n", #x, (unsigned long)(dl_rq->x))
690
691 PU(dl_nr_running);
692#ifdef CONFIG_SMP
693 PU(dl_nr_migratory);
694 dl_bw = &cpu_rq(cpu)->rd->dl_bw;
695#else
696 dl_bw = &dl_rq->dl_bw;
697#endif
698 SEQ_printf(m, " .%-30s: %lld\n", "dl_bw->bw", dl_bw->bw);
699 SEQ_printf(m, " .%-30s: %lld\n", "dl_bw->total_bw", dl_bw->total_bw);
700
701#undef PU
702}
703
704static void print_cpu(struct seq_file *m, int cpu)
705{
706 struct rq *rq = cpu_rq(cpu);
707
708#ifdef CONFIG_X86
709 {
710 unsigned int freq = cpu_khz ? : 1;
711
712 SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
713 cpu, freq / 1000, (freq % 1000));
714 }
715#else
716 SEQ_printf(m, "cpu#%d\n", cpu);
717#endif
718
719#define P(x) \
720do { \
721 if (sizeof(rq->x) == 4) \
722 SEQ_printf(m, " .%-30s: %ld\n", #x, (long)(rq->x)); \
723 else \
724 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rq->x));\
725} while (0)
726
727#define PN(x) \
728 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
729
730 P(nr_running);
731 P(nr_switches);
732 P(nr_uninterruptible);
733 PN(next_balance);
734 SEQ_printf(m, " .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr)));
735 PN(clock);
736 PN(clock_task);
737#undef P
738#undef PN
739
740#ifdef CONFIG_SMP
741#define P64(n) SEQ_printf(m, " .%-30s: %Ld\n", #n, rq->n);
742 P64(avg_idle);
743 P64(max_idle_balance_cost);
744#undef P64
745#endif
746
747#define P(n) SEQ_printf(m, " .%-30s: %d\n", #n, schedstat_val(rq->n));
748 if (schedstat_enabled()) {
749 P(yld_count);
750 P(sched_count);
751 P(sched_goidle);
752 P(ttwu_count);
753 P(ttwu_local);
754 }
755#undef P
756
757 print_cfs_stats(m, cpu);
758 print_rt_stats(m, cpu);
759 print_dl_stats(m, cpu);
760
761 print_rq(m, rq, cpu);
762 SEQ_printf(m, "\n");
763}
764
765static const char *sched_tunable_scaling_names[] = {
766 "none",
767 "logarithmic",
768 "linear"
769};
770
771static void sched_debug_header(struct seq_file *m)
772{
773 u64 ktime, sched_clk, cpu_clk;
774 unsigned long flags;
775
776 local_irq_save(flags);
777 ktime = ktime_to_ns(ktime_get());
778 sched_clk = sched_clock();
779 cpu_clk = local_clock();
780 local_irq_restore(flags);
781
782 SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n",
783 init_utsname()->release,
784 (int)strcspn(init_utsname()->version, " "),
785 init_utsname()->version);
786
787#define P(x) \
788 SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
789#define PN(x) \
790 SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
791 PN(ktime);
792 PN(sched_clk);
793 PN(cpu_clk);
794 P(jiffies);
795#ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
796 P(sched_clock_stable());
797#endif
798#undef PN
799#undef P
800
801 SEQ_printf(m, "\n");
802 SEQ_printf(m, "sysctl_sched\n");
803
804#define P(x) \
805 SEQ_printf(m, " .%-40s: %Ld\n", #x, (long long)(x))
806#define PN(x) \
807 SEQ_printf(m, " .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
808 PN(sysctl_sched_latency);
809 PN(sysctl_sched_min_granularity);
810 PN(sysctl_sched_wakeup_granularity);
811 P(sysctl_sched_child_runs_first);
812 P(sysctl_sched_features);
813#undef PN
814#undef P
815
816 SEQ_printf(m, " .%-40s: %d (%s)\n",
817 "sysctl_sched_tunable_scaling",
818 sysctl_sched_tunable_scaling,
819 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
820 SEQ_printf(m, "\n");
821}
822
823static int sched_debug_show(struct seq_file *m, void *v)
824{
825 int cpu = (unsigned long)(v - 2);
826
827 if (cpu != -1)
828 print_cpu(m, cpu);
829 else
830 sched_debug_header(m);
831
832 return 0;
833}
834
835void sysrq_sched_debug_show(void)
836{
837 int cpu;
838
839 sched_debug_header(NULL);
840 for_each_online_cpu(cpu) {
841 /*
842 * Need to reset softlockup watchdogs on all CPUs, because
843 * another CPU might be blocked waiting for us to process
844 * an IPI or stop_machine.
845 */
846 touch_nmi_watchdog();
847 touch_all_softlockup_watchdogs();
848 print_cpu(NULL, cpu);
849 }
850}
851
852/*
853 * This iterator needs some explanation.
854 * It returns 1 for the header position.
855 * This means 2 is CPU 0.
856 * In a hotplugged system some CPUs, including CPU 0, may be missing so we have
857 * to use cpumask_* to iterate over the CPUs.
858 */
859static void *sched_debug_start(struct seq_file *file, loff_t *offset)
860{
861 unsigned long n = *offset;
862
863 if (n == 0)
864 return (void *) 1;
865
866 n--;
867
868 if (n > 0)
869 n = cpumask_next(n - 1, cpu_online_mask);
870 else
871 n = cpumask_first(cpu_online_mask);
872
873 *offset = n + 1;
874
875 if (n < nr_cpu_ids)
876 return (void *)(unsigned long)(n + 2);
877
878 return NULL;
879}
880
881static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
882{
883 (*offset)++;
884 return sched_debug_start(file, offset);
885}
886
887static void sched_debug_stop(struct seq_file *file, void *data)
888{
889}
890
891static const struct seq_operations sched_debug_sops = {
892 .start = sched_debug_start,
893 .next = sched_debug_next,
894 .stop = sched_debug_stop,
895 .show = sched_debug_show,
896};
897
898#define __PS(S, F) SEQ_printf(m, "%-45s:%21Ld\n", S, (long long)(F))
899#define __P(F) __PS(#F, F)
900#define P(F) __PS(#F, p->F)
901#define PM(F, M) __PS(#F, p->F & (M))
902#define __PSN(S, F) SEQ_printf(m, "%-45s:%14Ld.%06ld\n", S, SPLIT_NS((long long)(F)))
903#define __PN(F) __PSN(#F, F)
904#define PN(F) __PSN(#F, p->F)
905
906
907#ifdef CONFIG_NUMA_BALANCING
908void print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
909 unsigned long tpf, unsigned long gsf, unsigned long gpf)
910{
911 SEQ_printf(m, "numa_faults node=%d ", node);
912 SEQ_printf(m, "task_private=%lu task_shared=%lu ", tpf, tsf);
913 SEQ_printf(m, "group_private=%lu group_shared=%lu\n", gpf, gsf);
914}
915#endif
916
917
918static void sched_show_numa(struct task_struct *p, struct seq_file *m)
919{
920#ifdef CONFIG_NUMA_BALANCING
921 struct mempolicy *pol;
922
923 if (p->mm)
924 P(mm->numa_scan_seq);
925
926 task_lock(p);
927 pol = p->mempolicy;
928 if (pol && !(pol->flags & MPOL_F_MORON))
929 pol = NULL;
930 mpol_get(pol);
931 task_unlock(p);
932
933 P(numa_pages_migrated);
934 P(numa_preferred_nid);
935 P(total_numa_faults);
936 SEQ_printf(m, "current_node=%d, numa_group_id=%d\n",
937 task_node(p), task_numa_group_id(p));
938 show_numa_stats(p, m);
939 mpol_put(pol);
940#endif
941}
942
943void proc_sched_show_task(struct task_struct *p, struct pid_namespace *ns,
944 struct seq_file *m)
945{
946 unsigned long nr_switches;
947
948 SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr_ns(p, ns),
949 get_nr_threads(p));
950 SEQ_printf(m,
951 "---------------------------------------------------------"
952 "----------\n");
953
954#define P_SCHEDSTAT(F) __PS(#F, schedstat_val(p->F))
955#define PN_SCHEDSTAT(F) __PSN(#F, schedstat_val(p->F))
956
957 PN(se.exec_start);
958 PN(se.vruntime);
959 PN(se.sum_exec_runtime);
960
961 nr_switches = p->nvcsw + p->nivcsw;
962
963 P(se.nr_migrations);
964
965 if (schedstat_enabled()) {
966 u64 avg_atom, avg_per_cpu;
967
968 PN_SCHEDSTAT(se.statistics.sum_sleep_runtime);
969 PN_SCHEDSTAT(se.statistics.wait_start);
970 PN_SCHEDSTAT(se.statistics.sleep_start);
971 PN_SCHEDSTAT(se.statistics.block_start);
972 PN_SCHEDSTAT(se.statistics.sleep_max);
973 PN_SCHEDSTAT(se.statistics.block_max);
974 PN_SCHEDSTAT(se.statistics.exec_max);
975 PN_SCHEDSTAT(se.statistics.slice_max);
976 PN_SCHEDSTAT(se.statistics.wait_max);
977 PN_SCHEDSTAT(se.statistics.wait_sum);
978 P_SCHEDSTAT(se.statistics.wait_count);
979 PN_SCHEDSTAT(se.statistics.iowait_sum);
980 P_SCHEDSTAT(se.statistics.iowait_count);
981 P_SCHEDSTAT(se.statistics.nr_migrations_cold);
982 P_SCHEDSTAT(se.statistics.nr_failed_migrations_affine);
983 P_SCHEDSTAT(se.statistics.nr_failed_migrations_running);
984 P_SCHEDSTAT(se.statistics.nr_failed_migrations_hot);
985 P_SCHEDSTAT(se.statistics.nr_forced_migrations);
986 P_SCHEDSTAT(se.statistics.nr_wakeups);
987 P_SCHEDSTAT(se.statistics.nr_wakeups_sync);
988 P_SCHEDSTAT(se.statistics.nr_wakeups_migrate);
989 P_SCHEDSTAT(se.statistics.nr_wakeups_local);
990 P_SCHEDSTAT(se.statistics.nr_wakeups_remote);
991 P_SCHEDSTAT(se.statistics.nr_wakeups_affine);
992 P_SCHEDSTAT(se.statistics.nr_wakeups_affine_attempts);
993 P_SCHEDSTAT(se.statistics.nr_wakeups_passive);
994 P_SCHEDSTAT(se.statistics.nr_wakeups_idle);
995
996 avg_atom = p->se.sum_exec_runtime;
997 if (nr_switches)
998 avg_atom = div64_ul(avg_atom, nr_switches);
999 else
1000 avg_atom = -1LL;
1001
1002 avg_per_cpu = p->se.sum_exec_runtime;
1003 if (p->se.nr_migrations) {
1004 avg_per_cpu = div64_u64(avg_per_cpu,
1005 p->se.nr_migrations);
1006 } else {
1007 avg_per_cpu = -1LL;
1008 }
1009
1010 __PN(avg_atom);
1011 __PN(avg_per_cpu);
1012 }
1013
1014 __P(nr_switches);
1015 __PS("nr_voluntary_switches", p->nvcsw);
1016 __PS("nr_involuntary_switches", p->nivcsw);
1017
1018 P(se.load.weight);
1019#ifdef CONFIG_SMP
1020 P(se.avg.load_sum);
1021 P(se.avg.runnable_sum);
1022 P(se.avg.util_sum);
1023 P(se.avg.load_avg);
1024 P(se.avg.runnable_avg);
1025 P(se.avg.util_avg);
1026 P(se.avg.last_update_time);
1027 P(se.avg.util_est.ewma);
1028 PM(se.avg.util_est.enqueued, ~UTIL_AVG_UNCHANGED);
1029#endif
1030#ifdef CONFIG_UCLAMP_TASK
1031 __PS("uclamp.min", p->uclamp_req[UCLAMP_MIN].value);
1032 __PS("uclamp.max", p->uclamp_req[UCLAMP_MAX].value);
1033 __PS("effective uclamp.min", uclamp_eff_value(p, UCLAMP_MIN));
1034 __PS("effective uclamp.max", uclamp_eff_value(p, UCLAMP_MAX));
1035#endif
1036 P(policy);
1037 P(prio);
1038 if (task_has_dl_policy(p)) {
1039 P(dl.runtime);
1040 P(dl.deadline);
1041 }
1042#undef PN_SCHEDSTAT
1043#undef P_SCHEDSTAT
1044
1045 {
1046 unsigned int this_cpu = raw_smp_processor_id();
1047 u64 t0, t1;
1048
1049 t0 = cpu_clock(this_cpu);
1050 t1 = cpu_clock(this_cpu);
1051 __PS("clock-delta", t1-t0);
1052 }
1053
1054 sched_show_numa(p, m);
1055}
1056
1057void proc_sched_set_task(struct task_struct *p)
1058{
1059#ifdef CONFIG_SCHEDSTATS
1060 memset(&p->se.statistics, 0, sizeof(p->se.statistics));
1061#endif
1062}
1063
1064void resched_latency_warn(int cpu, u64 latency)
1065{
1066 static DEFINE_RATELIMIT_STATE(latency_check_ratelimit, 60 * 60 * HZ, 1);
1067
1068 WARN(__ratelimit(&latency_check_ratelimit),
1069 "sched: CPU %d need_resched set for > %llu ns (%d ticks) "
1070 "without schedule\n",
1071 cpu, latency, cpu_rq(cpu)->ticks_without_resched);
1072}