sched/fair: replace cfs_rq->rb_leftmost
[sfrench/cifs-2.6.git] / kernel / sched / debug.c
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/mm.h>
15 #include <linux/sched/task.h>
16 #include <linux/seq_file.h>
17 #include <linux/kallsyms.h>
18 #include <linux/utsname.h>
19 #include <linux/mempolicy.h>
20 #include <linux/debugfs.h>
21
22 #include "sched.h"
23
24 static DEFINE_SPINLOCK(sched_debug_lock);
25
26 /*
27  * This allows printing both to /proc/sched_debug and
28  * to the console
29  */
30 #define SEQ_printf(m, x...)                     \
31  do {                                           \
32         if (m)                                  \
33                 seq_printf(m, x);               \
34         else                                    \
35                 printk(x);                      \
36  } while (0)
37
38 /*
39  * Ease the printing of nsec fields:
40  */
41 static long long nsec_high(unsigned long long nsec)
42 {
43         if ((long long)nsec < 0) {
44                 nsec = -nsec;
45                 do_div(nsec, 1000000);
46                 return -nsec;
47         }
48         do_div(nsec, 1000000);
49
50         return nsec;
51 }
52
53 static unsigned long nsec_low(unsigned long long nsec)
54 {
55         if ((long long)nsec < 0)
56                 nsec = -nsec;
57
58         return do_div(nsec, 1000000);
59 }
60
61 #define SPLIT_NS(x) nsec_high(x), nsec_low(x)
62
63 #define SCHED_FEAT(name, enabled)       \
64         #name ,
65
66 static const char * const sched_feat_names[] = {
67 #include "features.h"
68 };
69
70 #undef SCHED_FEAT
71
72 static int sched_feat_show(struct seq_file *m, void *v)
73 {
74         int i;
75
76         for (i = 0; i < __SCHED_FEAT_NR; i++) {
77                 if (!(sysctl_sched_features & (1UL << i)))
78                         seq_puts(m, "NO_");
79                 seq_printf(m, "%s ", sched_feat_names[i]);
80         }
81         seq_puts(m, "\n");
82
83         return 0;
84 }
85
86 #ifdef HAVE_JUMP_LABEL
87
88 #define jump_label_key__true  STATIC_KEY_INIT_TRUE
89 #define jump_label_key__false STATIC_KEY_INIT_FALSE
90
91 #define SCHED_FEAT(name, enabled)       \
92         jump_label_key__##enabled ,
93
94 struct static_key sched_feat_keys[__SCHED_FEAT_NR] = {
95 #include "features.h"
96 };
97
98 #undef SCHED_FEAT
99
100 static void sched_feat_disable(int i)
101 {
102         static_key_disable(&sched_feat_keys[i]);
103 }
104
105 static void sched_feat_enable(int i)
106 {
107         static_key_enable(&sched_feat_keys[i]);
108 }
109 #else
110 static void sched_feat_disable(int i) { };
111 static void sched_feat_enable(int i) { };
112 #endif /* HAVE_JUMP_LABEL */
113
114 static int sched_feat_set(char *cmp)
115 {
116         int i;
117         int neg = 0;
118
119         if (strncmp(cmp, "NO_", 3) == 0) {
120                 neg = 1;
121                 cmp += 3;
122         }
123
124         for (i = 0; i < __SCHED_FEAT_NR; i++) {
125                 if (strcmp(cmp, sched_feat_names[i]) == 0) {
126                         if (neg) {
127                                 sysctl_sched_features &= ~(1UL << i);
128                                 sched_feat_disable(i);
129                         } else {
130                                 sysctl_sched_features |= (1UL << i);
131                                 sched_feat_enable(i);
132                         }
133                         break;
134                 }
135         }
136
137         return i;
138 }
139
140 static ssize_t
141 sched_feat_write(struct file *filp, const char __user *ubuf,
142                 size_t cnt, loff_t *ppos)
143 {
144         char buf[64];
145         char *cmp;
146         int i;
147         struct inode *inode;
148
149         if (cnt > 63)
150                 cnt = 63;
151
152         if (copy_from_user(&buf, ubuf, cnt))
153                 return -EFAULT;
154
155         buf[cnt] = 0;
156         cmp = strstrip(buf);
157
158         /* Ensure the static_key remains in a consistent state */
159         inode = file_inode(filp);
160         inode_lock(inode);
161         i = sched_feat_set(cmp);
162         inode_unlock(inode);
163         if (i == __SCHED_FEAT_NR)
164                 return -EINVAL;
165
166         *ppos += cnt;
167
168         return cnt;
169 }
170
171 static int sched_feat_open(struct inode *inode, struct file *filp)
172 {
173         return single_open(filp, sched_feat_show, NULL);
174 }
175
176 static const struct file_operations sched_feat_fops = {
177         .open           = sched_feat_open,
178         .write          = sched_feat_write,
179         .read           = seq_read,
180         .llseek         = seq_lseek,
181         .release        = single_release,
182 };
183
184 static __init int sched_init_debug(void)
185 {
186         debugfs_create_file("sched_features", 0644, NULL, NULL,
187                         &sched_feat_fops);
188
189         return 0;
190 }
191 late_initcall(sched_init_debug);
192
193 #ifdef CONFIG_SMP
194
195 #ifdef CONFIG_SYSCTL
196
197 static struct ctl_table sd_ctl_dir[] = {
198         {
199                 .procname       = "sched_domain",
200                 .mode           = 0555,
201         },
202         {}
203 };
204
205 static struct ctl_table sd_ctl_root[] = {
206         {
207                 .procname       = "kernel",
208                 .mode           = 0555,
209                 .child          = sd_ctl_dir,
210         },
211         {}
212 };
213
214 static struct ctl_table *sd_alloc_ctl_entry(int n)
215 {
216         struct ctl_table *entry =
217                 kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL);
218
219         return entry;
220 }
221
222 static void sd_free_ctl_entry(struct ctl_table **tablep)
223 {
224         struct ctl_table *entry;
225
226         /*
227          * In the intermediate directories, both the child directory and
228          * procname are dynamically allocated and could fail but the mode
229          * will always be set. In the lowest directory the names are
230          * static strings and all have proc handlers.
231          */
232         for (entry = *tablep; entry->mode; entry++) {
233                 if (entry->child)
234                         sd_free_ctl_entry(&entry->child);
235                 if (entry->proc_handler == NULL)
236                         kfree(entry->procname);
237         }
238
239         kfree(*tablep);
240         *tablep = NULL;
241 }
242
243 static int min_load_idx = 0;
244 static int max_load_idx = CPU_LOAD_IDX_MAX-1;
245
246 static void
247 set_table_entry(struct ctl_table *entry,
248                 const char *procname, void *data, int maxlen,
249                 umode_t mode, proc_handler *proc_handler,
250                 bool load_idx)
251 {
252         entry->procname = procname;
253         entry->data = data;
254         entry->maxlen = maxlen;
255         entry->mode = mode;
256         entry->proc_handler = proc_handler;
257
258         if (load_idx) {
259                 entry->extra1 = &min_load_idx;
260                 entry->extra2 = &max_load_idx;
261         }
262 }
263
264 static struct ctl_table *
265 sd_alloc_ctl_domain_table(struct sched_domain *sd)
266 {
267         struct ctl_table *table = sd_alloc_ctl_entry(14);
268
269         if (table == NULL)
270                 return NULL;
271
272         set_table_entry(&table[0], "min_interval", &sd->min_interval,
273                 sizeof(long), 0644, proc_doulongvec_minmax, false);
274         set_table_entry(&table[1], "max_interval", &sd->max_interval,
275                 sizeof(long), 0644, proc_doulongvec_minmax, false);
276         set_table_entry(&table[2], "busy_idx", &sd->busy_idx,
277                 sizeof(int), 0644, proc_dointvec_minmax, true);
278         set_table_entry(&table[3], "idle_idx", &sd->idle_idx,
279                 sizeof(int), 0644, proc_dointvec_minmax, true);
280         set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx,
281                 sizeof(int), 0644, proc_dointvec_minmax, true);
282         set_table_entry(&table[5], "wake_idx", &sd->wake_idx,
283                 sizeof(int), 0644, proc_dointvec_minmax, true);
284         set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx,
285                 sizeof(int), 0644, proc_dointvec_minmax, true);
286         set_table_entry(&table[7], "busy_factor", &sd->busy_factor,
287                 sizeof(int), 0644, proc_dointvec_minmax, false);
288         set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct,
289                 sizeof(int), 0644, proc_dointvec_minmax, false);
290         set_table_entry(&table[9], "cache_nice_tries",
291                 &sd->cache_nice_tries,
292                 sizeof(int), 0644, proc_dointvec_minmax, false);
293         set_table_entry(&table[10], "flags", &sd->flags,
294                 sizeof(int), 0644, proc_dointvec_minmax, false);
295         set_table_entry(&table[11], "max_newidle_lb_cost",
296                 &sd->max_newidle_lb_cost,
297                 sizeof(long), 0644, proc_doulongvec_minmax, false);
298         set_table_entry(&table[12], "name", sd->name,
299                 CORENAME_MAX_SIZE, 0444, proc_dostring, false);
300         /* &table[13] is terminator */
301
302         return table;
303 }
304
305 static struct ctl_table *sd_alloc_ctl_cpu_table(int cpu)
306 {
307         struct ctl_table *entry, *table;
308         struct sched_domain *sd;
309         int domain_num = 0, i;
310         char buf[32];
311
312         for_each_domain(cpu, sd)
313                 domain_num++;
314         entry = table = sd_alloc_ctl_entry(domain_num + 1);
315         if (table == NULL)
316                 return NULL;
317
318         i = 0;
319         for_each_domain(cpu, sd) {
320                 snprintf(buf, 32, "domain%d", i);
321                 entry->procname = kstrdup(buf, GFP_KERNEL);
322                 entry->mode = 0555;
323                 entry->child = sd_alloc_ctl_domain_table(sd);
324                 entry++;
325                 i++;
326         }
327         return table;
328 }
329
330 static cpumask_var_t sd_sysctl_cpus;
331 static struct ctl_table_header *sd_sysctl_header;
332
333 void register_sched_domain_sysctl(void)
334 {
335         static struct ctl_table *cpu_entries;
336         static struct ctl_table **cpu_idx;
337         char buf[32];
338         int i;
339
340         if (!cpu_entries) {
341                 cpu_entries = sd_alloc_ctl_entry(num_possible_cpus() + 1);
342                 if (!cpu_entries)
343                         return;
344
345                 WARN_ON(sd_ctl_dir[0].child);
346                 sd_ctl_dir[0].child = cpu_entries;
347         }
348
349         if (!cpu_idx) {
350                 struct ctl_table *e = cpu_entries;
351
352                 cpu_idx = kcalloc(nr_cpu_ids, sizeof(struct ctl_table*), GFP_KERNEL);
353                 if (!cpu_idx)
354                         return;
355
356                 /* deal with sparse possible map */
357                 for_each_possible_cpu(i) {
358                         cpu_idx[i] = e;
359                         e++;
360                 }
361         }
362
363         if (!cpumask_available(sd_sysctl_cpus)) {
364                 if (!alloc_cpumask_var(&sd_sysctl_cpus, GFP_KERNEL))
365                         return;
366
367                 /* init to possible to not have holes in @cpu_entries */
368                 cpumask_copy(sd_sysctl_cpus, cpu_possible_mask);
369         }
370
371         for_each_cpu(i, sd_sysctl_cpus) {
372                 struct ctl_table *e = cpu_idx[i];
373
374                 if (e->child)
375                         sd_free_ctl_entry(&e->child);
376
377                 if (!e->procname) {
378                         snprintf(buf, 32, "cpu%d", i);
379                         e->procname = kstrdup(buf, GFP_KERNEL);
380                 }
381                 e->mode = 0555;
382                 e->child = sd_alloc_ctl_cpu_table(i);
383
384                 __cpumask_clear_cpu(i, sd_sysctl_cpus);
385         }
386
387         WARN_ON(sd_sysctl_header);
388         sd_sysctl_header = register_sysctl_table(sd_ctl_root);
389 }
390
391 void dirty_sched_domain_sysctl(int cpu)
392 {
393         if (cpumask_available(sd_sysctl_cpus))
394                 __cpumask_set_cpu(cpu, sd_sysctl_cpus);
395 }
396
397 /* may be called multiple times per register */
398 void unregister_sched_domain_sysctl(void)
399 {
400         unregister_sysctl_table(sd_sysctl_header);
401         sd_sysctl_header = NULL;
402 }
403 #endif /* CONFIG_SYSCTL */
404 #endif /* CONFIG_SMP */
405
406 #ifdef CONFIG_FAIR_GROUP_SCHED
407 static void print_cfs_group_stats(struct seq_file *m, int cpu, struct task_group *tg)
408 {
409         struct sched_entity *se = tg->se[cpu];
410
411 #define P(F) \
412         SEQ_printf(m, "  .%-30s: %lld\n", #F, (long long)F)
413 #define P_SCHEDSTAT(F) \
414         SEQ_printf(m, "  .%-30s: %lld\n", #F, (long long)schedstat_val(F))
415 #define PN(F) \
416         SEQ_printf(m, "  .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F))
417 #define PN_SCHEDSTAT(F) \
418         SEQ_printf(m, "  .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)schedstat_val(F)))
419
420         if (!se)
421                 return;
422
423         PN(se->exec_start);
424         PN(se->vruntime);
425         PN(se->sum_exec_runtime);
426         if (schedstat_enabled()) {
427                 PN_SCHEDSTAT(se->statistics.wait_start);
428                 PN_SCHEDSTAT(se->statistics.sleep_start);
429                 PN_SCHEDSTAT(se->statistics.block_start);
430                 PN_SCHEDSTAT(se->statistics.sleep_max);
431                 PN_SCHEDSTAT(se->statistics.block_max);
432                 PN_SCHEDSTAT(se->statistics.exec_max);
433                 PN_SCHEDSTAT(se->statistics.slice_max);
434                 PN_SCHEDSTAT(se->statistics.wait_max);
435                 PN_SCHEDSTAT(se->statistics.wait_sum);
436                 P_SCHEDSTAT(se->statistics.wait_count);
437         }
438         P(se->load.weight);
439 #ifdef CONFIG_SMP
440         P(se->avg.load_avg);
441         P(se->avg.util_avg);
442 #endif
443
444 #undef PN_SCHEDSTAT
445 #undef PN
446 #undef P_SCHEDSTAT
447 #undef P
448 }
449 #endif
450
451 #ifdef CONFIG_CGROUP_SCHED
452 static char group_path[PATH_MAX];
453
454 static char *task_group_path(struct task_group *tg)
455 {
456         if (autogroup_path(tg, group_path, PATH_MAX))
457                 return group_path;
458
459         cgroup_path(tg->css.cgroup, group_path, PATH_MAX);
460         return group_path;
461 }
462 #endif
463
464 static const char stat_nam[] = TASK_STATE_TO_CHAR_STR;
465
466 static void
467 print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
468 {
469         if (rq->curr == p)
470                 SEQ_printf(m, ">R");
471         else
472                 SEQ_printf(m, " %c", task_state_to_char(p));
473
474         SEQ_printf(m, "%15s %5d %9Ld.%06ld %9Ld %5d ",
475                 p->comm, task_pid_nr(p),
476                 SPLIT_NS(p->se.vruntime),
477                 (long long)(p->nvcsw + p->nivcsw),
478                 p->prio);
479
480         SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
481                 SPLIT_NS(schedstat_val_or_zero(p->se.statistics.wait_sum)),
482                 SPLIT_NS(p->se.sum_exec_runtime),
483                 SPLIT_NS(schedstat_val_or_zero(p->se.statistics.sum_sleep_runtime)));
484
485 #ifdef CONFIG_NUMA_BALANCING
486         SEQ_printf(m, " %d %d", task_node(p), task_numa_group_id(p));
487 #endif
488 #ifdef CONFIG_CGROUP_SCHED
489         SEQ_printf(m, " %s", task_group_path(task_group(p)));
490 #endif
491
492         SEQ_printf(m, "\n");
493 }
494
495 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
496 {
497         struct task_struct *g, *p;
498
499         SEQ_printf(m,
500         "\nrunnable tasks:\n"
501         " S           task   PID         tree-key  switches  prio"
502         "     wait-time             sum-exec        sum-sleep\n"
503         "-------------------------------------------------------"
504         "----------------------------------------------------\n");
505
506         rcu_read_lock();
507         for_each_process_thread(g, p) {
508                 if (task_cpu(p) != rq_cpu)
509                         continue;
510
511                 print_task(m, rq, p);
512         }
513         rcu_read_unlock();
514 }
515
516 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
517 {
518         s64 MIN_vruntime = -1, min_vruntime, max_vruntime = -1,
519                 spread, rq0_min_vruntime, spread0;
520         struct rq *rq = cpu_rq(cpu);
521         struct sched_entity *last;
522         unsigned long flags;
523
524 #ifdef CONFIG_FAIR_GROUP_SCHED
525         SEQ_printf(m, "\ncfs_rq[%d]:%s\n", cpu, task_group_path(cfs_rq->tg));
526 #else
527         SEQ_printf(m, "\ncfs_rq[%d]:\n", cpu);
528 #endif
529         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "exec_clock",
530                         SPLIT_NS(cfs_rq->exec_clock));
531
532         raw_spin_lock_irqsave(&rq->lock, flags);
533         if (rb_first_cached(&cfs_rq->tasks_timeline))
534                 MIN_vruntime = (__pick_first_entity(cfs_rq))->vruntime;
535         last = __pick_last_entity(cfs_rq);
536         if (last)
537                 max_vruntime = last->vruntime;
538         min_vruntime = cfs_rq->min_vruntime;
539         rq0_min_vruntime = cpu_rq(0)->cfs.min_vruntime;
540         raw_spin_unlock_irqrestore(&rq->lock, flags);
541         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "MIN_vruntime",
542                         SPLIT_NS(MIN_vruntime));
543         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "min_vruntime",
544                         SPLIT_NS(min_vruntime));
545         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "max_vruntime",
546                         SPLIT_NS(max_vruntime));
547         spread = max_vruntime - MIN_vruntime;
548         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread",
549                         SPLIT_NS(spread));
550         spread0 = min_vruntime - rq0_min_vruntime;
551         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", "spread0",
552                         SPLIT_NS(spread0));
553         SEQ_printf(m, "  .%-30s: %d\n", "nr_spread_over",
554                         cfs_rq->nr_spread_over);
555         SEQ_printf(m, "  .%-30s: %d\n", "nr_running", cfs_rq->nr_running);
556         SEQ_printf(m, "  .%-30s: %ld\n", "load", cfs_rq->load.weight);
557 #ifdef CONFIG_SMP
558         SEQ_printf(m, "  .%-30s: %lu\n", "load_avg",
559                         cfs_rq->avg.load_avg);
560         SEQ_printf(m, "  .%-30s: %lu\n", "runnable_load_avg",
561                         cfs_rq->runnable_load_avg);
562         SEQ_printf(m, "  .%-30s: %lu\n", "util_avg",
563                         cfs_rq->avg.util_avg);
564         SEQ_printf(m, "  .%-30s: %ld\n", "removed_load_avg",
565                         atomic_long_read(&cfs_rq->removed_load_avg));
566         SEQ_printf(m, "  .%-30s: %ld\n", "removed_util_avg",
567                         atomic_long_read(&cfs_rq->removed_util_avg));
568 #ifdef CONFIG_FAIR_GROUP_SCHED
569         SEQ_printf(m, "  .%-30s: %lu\n", "tg_load_avg_contrib",
570                         cfs_rq->tg_load_avg_contrib);
571         SEQ_printf(m, "  .%-30s: %ld\n", "tg_load_avg",
572                         atomic_long_read(&cfs_rq->tg->load_avg));
573 #endif
574 #endif
575 #ifdef CONFIG_CFS_BANDWIDTH
576         SEQ_printf(m, "  .%-30s: %d\n", "throttled",
577                         cfs_rq->throttled);
578         SEQ_printf(m, "  .%-30s: %d\n", "throttle_count",
579                         cfs_rq->throttle_count);
580 #endif
581
582 #ifdef CONFIG_FAIR_GROUP_SCHED
583         print_cfs_group_stats(m, cpu, cfs_rq->tg);
584 #endif
585 }
586
587 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
588 {
589 #ifdef CONFIG_RT_GROUP_SCHED
590         SEQ_printf(m, "\nrt_rq[%d]:%s\n", cpu, task_group_path(rt_rq->tg));
591 #else
592         SEQ_printf(m, "\nrt_rq[%d]:\n", cpu);
593 #endif
594
595 #define P(x) \
596         SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
597 #define PU(x) \
598         SEQ_printf(m, "  .%-30s: %lu\n", #x, (unsigned long)(rt_rq->x))
599 #define PN(x) \
600         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
601
602         PU(rt_nr_running);
603 #ifdef CONFIG_SMP
604         PU(rt_nr_migratory);
605 #endif
606         P(rt_throttled);
607         PN(rt_time);
608         PN(rt_runtime);
609
610 #undef PN
611 #undef PU
612 #undef P
613 }
614
615 void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq)
616 {
617         struct dl_bw *dl_bw;
618
619         SEQ_printf(m, "\ndl_rq[%d]:\n", cpu);
620
621 #define PU(x) \
622         SEQ_printf(m, "  .%-30s: %lu\n", #x, (unsigned long)(dl_rq->x))
623
624         PU(dl_nr_running);
625 #ifdef CONFIG_SMP
626         PU(dl_nr_migratory);
627         dl_bw = &cpu_rq(cpu)->rd->dl_bw;
628 #else
629         dl_bw = &dl_rq->dl_bw;
630 #endif
631         SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->bw", dl_bw->bw);
632         SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->total_bw", dl_bw->total_bw);
633
634 #undef PU
635 }
636
637 extern __read_mostly int sched_clock_running;
638
639 static void print_cpu(struct seq_file *m, int cpu)
640 {
641         struct rq *rq = cpu_rq(cpu);
642         unsigned long flags;
643
644 #ifdef CONFIG_X86
645         {
646                 unsigned int freq = cpu_khz ? : 1;
647
648                 SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
649                            cpu, freq / 1000, (freq % 1000));
650         }
651 #else
652         SEQ_printf(m, "cpu#%d\n", cpu);
653 #endif
654
655 #define P(x)                                                            \
656 do {                                                                    \
657         if (sizeof(rq->x) == 4)                                         \
658                 SEQ_printf(m, "  .%-30s: %ld\n", #x, (long)(rq->x));    \
659         else                                                            \
660                 SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rq->x));\
661 } while (0)
662
663 #define PN(x) \
664         SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))
665
666         P(nr_running);
667         SEQ_printf(m, "  .%-30s: %lu\n", "load",
668                    rq->load.weight);
669         P(nr_switches);
670         P(nr_load_updates);
671         P(nr_uninterruptible);
672         PN(next_balance);
673         SEQ_printf(m, "  .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr)));
674         PN(clock);
675         PN(clock_task);
676         P(cpu_load[0]);
677         P(cpu_load[1]);
678         P(cpu_load[2]);
679         P(cpu_load[3]);
680         P(cpu_load[4]);
681 #undef P
682 #undef PN
683
684 #ifdef CONFIG_SMP
685 #define P64(n) SEQ_printf(m, "  .%-30s: %Ld\n", #n, rq->n);
686         P64(avg_idle);
687         P64(max_idle_balance_cost);
688 #undef P64
689 #endif
690
691 #define P(n) SEQ_printf(m, "  .%-30s: %d\n", #n, schedstat_val(rq->n));
692         if (schedstat_enabled()) {
693                 P(yld_count);
694                 P(sched_count);
695                 P(sched_goidle);
696                 P(ttwu_count);
697                 P(ttwu_local);
698         }
699 #undef P
700
701         spin_lock_irqsave(&sched_debug_lock, flags);
702         print_cfs_stats(m, cpu);
703         print_rt_stats(m, cpu);
704         print_dl_stats(m, cpu);
705
706         print_rq(m, rq, cpu);
707         spin_unlock_irqrestore(&sched_debug_lock, flags);
708         SEQ_printf(m, "\n");
709 }
710
711 static const char *sched_tunable_scaling_names[] = {
712         "none",
713         "logaritmic",
714         "linear"
715 };
716
717 static void sched_debug_header(struct seq_file *m)
718 {
719         u64 ktime, sched_clk, cpu_clk;
720         unsigned long flags;
721
722         local_irq_save(flags);
723         ktime = ktime_to_ns(ktime_get());
724         sched_clk = sched_clock();
725         cpu_clk = local_clock();
726         local_irq_restore(flags);
727
728         SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n",
729                 init_utsname()->release,
730                 (int)strcspn(init_utsname()->version, " "),
731                 init_utsname()->version);
732
733 #define P(x) \
734         SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x))
735 #define PN(x) \
736         SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
737         PN(ktime);
738         PN(sched_clk);
739         PN(cpu_clk);
740         P(jiffies);
741 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
742         P(sched_clock_stable());
743 #endif
744 #undef PN
745 #undef P
746
747         SEQ_printf(m, "\n");
748         SEQ_printf(m, "sysctl_sched\n");
749
750 #define P(x) \
751         SEQ_printf(m, "  .%-40s: %Ld\n", #x, (long long)(x))
752 #define PN(x) \
753         SEQ_printf(m, "  .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
754         PN(sysctl_sched_latency);
755         PN(sysctl_sched_min_granularity);
756         PN(sysctl_sched_wakeup_granularity);
757         P(sysctl_sched_child_runs_first);
758         P(sysctl_sched_features);
759 #undef PN
760 #undef P
761
762         SEQ_printf(m, "  .%-40s: %d (%s)\n",
763                 "sysctl_sched_tunable_scaling",
764                 sysctl_sched_tunable_scaling,
765                 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
766         SEQ_printf(m, "\n");
767 }
768
769 static int sched_debug_show(struct seq_file *m, void *v)
770 {
771         int cpu = (unsigned long)(v - 2);
772
773         if (cpu != -1)
774                 print_cpu(m, cpu);
775         else
776                 sched_debug_header(m);
777
778         return 0;
779 }
780
781 void sysrq_sched_debug_show(void)
782 {
783         int cpu;
784
785         sched_debug_header(NULL);
786         for_each_online_cpu(cpu)
787                 print_cpu(NULL, cpu);
788
789 }
790
791 /*
792  * This itererator needs some explanation.
793  * It returns 1 for the header position.
794  * This means 2 is cpu 0.
795  * In a hotplugged system some cpus, including cpu 0, may be missing so we have
796  * to use cpumask_* to iterate over the cpus.
797  */
798 static void *sched_debug_start(struct seq_file *file, loff_t *offset)
799 {
800         unsigned long n = *offset;
801
802         if (n == 0)
803                 return (void *) 1;
804
805         n--;
806
807         if (n > 0)
808                 n = cpumask_next(n - 1, cpu_online_mask);
809         else
810                 n = cpumask_first(cpu_online_mask);
811
812         *offset = n + 1;
813
814         if (n < nr_cpu_ids)
815                 return (void *)(unsigned long)(n + 2);
816         return NULL;
817 }
818
819 static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
820 {
821         (*offset)++;
822         return sched_debug_start(file, offset);
823 }
824
825 static void sched_debug_stop(struct seq_file *file, void *data)
826 {
827 }
828
829 static const struct seq_operations sched_debug_sops = {
830         .start = sched_debug_start,
831         .next = sched_debug_next,
832         .stop = sched_debug_stop,
833         .show = sched_debug_show,
834 };
835
836 static int sched_debug_release(struct inode *inode, struct file *file)
837 {
838         seq_release(inode, file);
839
840         return 0;
841 }
842
843 static int sched_debug_open(struct inode *inode, struct file *filp)
844 {
845         int ret = 0;
846
847         ret = seq_open(filp, &sched_debug_sops);
848
849         return ret;
850 }
851
852 static const struct file_operations sched_debug_fops = {
853         .open           = sched_debug_open,
854         .read           = seq_read,
855         .llseek         = seq_lseek,
856         .release        = sched_debug_release,
857 };
858
859 static int __init init_sched_debug_procfs(void)
860 {
861         struct proc_dir_entry *pe;
862
863         pe = proc_create("sched_debug", 0444, NULL, &sched_debug_fops);
864         if (!pe)
865                 return -ENOMEM;
866         return 0;
867 }
868
869 __initcall(init_sched_debug_procfs);
870
871 #define __P(F) \
872         SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)F)
873 #define P(F) \
874         SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)p->F)
875 #define __PN(F) \
876         SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
877 #define PN(F) \
878         SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
879
880
881 #ifdef CONFIG_NUMA_BALANCING
882 void print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
883                 unsigned long tpf, unsigned long gsf, unsigned long gpf)
884 {
885         SEQ_printf(m, "numa_faults node=%d ", node);
886         SEQ_printf(m, "task_private=%lu task_shared=%lu ", tsf, tpf);
887         SEQ_printf(m, "group_private=%lu group_shared=%lu\n", gsf, gpf);
888 }
889 #endif
890
891
892 static void sched_show_numa(struct task_struct *p, struct seq_file *m)
893 {
894 #ifdef CONFIG_NUMA_BALANCING
895         struct mempolicy *pol;
896
897         if (p->mm)
898                 P(mm->numa_scan_seq);
899
900         task_lock(p);
901         pol = p->mempolicy;
902         if (pol && !(pol->flags & MPOL_F_MORON))
903                 pol = NULL;
904         mpol_get(pol);
905         task_unlock(p);
906
907         P(numa_pages_migrated);
908         P(numa_preferred_nid);
909         P(total_numa_faults);
910         SEQ_printf(m, "current_node=%d, numa_group_id=%d\n",
911                         task_node(p), task_numa_group_id(p));
912         show_numa_stats(p, m);
913         mpol_put(pol);
914 #endif
915 }
916
917 void proc_sched_show_task(struct task_struct *p, struct pid_namespace *ns,
918                                                   struct seq_file *m)
919 {
920         unsigned long nr_switches;
921
922         SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr_ns(p, ns),
923                                                 get_nr_threads(p));
924         SEQ_printf(m,
925                 "---------------------------------------------------------"
926                 "----------\n");
927 #define __P(F) \
928         SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)F)
929 #define P(F) \
930         SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)p->F)
931 #define P_SCHEDSTAT(F) \
932         SEQ_printf(m, "%-45s:%21Ld\n", #F, (long long)schedstat_val(p->F))
933 #define __PN(F) \
934         SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)F))
935 #define PN(F) \
936         SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
937 #define PN_SCHEDSTAT(F) \
938         SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)schedstat_val(p->F)))
939
940         PN(se.exec_start);
941         PN(se.vruntime);
942         PN(se.sum_exec_runtime);
943
944         nr_switches = p->nvcsw + p->nivcsw;
945
946         P(se.nr_migrations);
947
948         if (schedstat_enabled()) {
949                 u64 avg_atom, avg_per_cpu;
950
951                 PN_SCHEDSTAT(se.statistics.sum_sleep_runtime);
952                 PN_SCHEDSTAT(se.statistics.wait_start);
953                 PN_SCHEDSTAT(se.statistics.sleep_start);
954                 PN_SCHEDSTAT(se.statistics.block_start);
955                 PN_SCHEDSTAT(se.statistics.sleep_max);
956                 PN_SCHEDSTAT(se.statistics.block_max);
957                 PN_SCHEDSTAT(se.statistics.exec_max);
958                 PN_SCHEDSTAT(se.statistics.slice_max);
959                 PN_SCHEDSTAT(se.statistics.wait_max);
960                 PN_SCHEDSTAT(se.statistics.wait_sum);
961                 P_SCHEDSTAT(se.statistics.wait_count);
962                 PN_SCHEDSTAT(se.statistics.iowait_sum);
963                 P_SCHEDSTAT(se.statistics.iowait_count);
964                 P_SCHEDSTAT(se.statistics.nr_migrations_cold);
965                 P_SCHEDSTAT(se.statistics.nr_failed_migrations_affine);
966                 P_SCHEDSTAT(se.statistics.nr_failed_migrations_running);
967                 P_SCHEDSTAT(se.statistics.nr_failed_migrations_hot);
968                 P_SCHEDSTAT(se.statistics.nr_forced_migrations);
969                 P_SCHEDSTAT(se.statistics.nr_wakeups);
970                 P_SCHEDSTAT(se.statistics.nr_wakeups_sync);
971                 P_SCHEDSTAT(se.statistics.nr_wakeups_migrate);
972                 P_SCHEDSTAT(se.statistics.nr_wakeups_local);
973                 P_SCHEDSTAT(se.statistics.nr_wakeups_remote);
974                 P_SCHEDSTAT(se.statistics.nr_wakeups_affine);
975                 P_SCHEDSTAT(se.statistics.nr_wakeups_affine_attempts);
976                 P_SCHEDSTAT(se.statistics.nr_wakeups_passive);
977                 P_SCHEDSTAT(se.statistics.nr_wakeups_idle);
978
979                 avg_atom = p->se.sum_exec_runtime;
980                 if (nr_switches)
981                         avg_atom = div64_ul(avg_atom, nr_switches);
982                 else
983                         avg_atom = -1LL;
984
985                 avg_per_cpu = p->se.sum_exec_runtime;
986                 if (p->se.nr_migrations) {
987                         avg_per_cpu = div64_u64(avg_per_cpu,
988                                                 p->se.nr_migrations);
989                 } else {
990                         avg_per_cpu = -1LL;
991                 }
992
993                 __PN(avg_atom);
994                 __PN(avg_per_cpu);
995         }
996
997         __P(nr_switches);
998         SEQ_printf(m, "%-45s:%21Ld\n",
999                    "nr_voluntary_switches", (long long)p->nvcsw);
1000         SEQ_printf(m, "%-45s:%21Ld\n",
1001                    "nr_involuntary_switches", (long long)p->nivcsw);
1002
1003         P(se.load.weight);
1004 #ifdef CONFIG_SMP
1005         P(se.avg.load_sum);
1006         P(se.avg.util_sum);
1007         P(se.avg.load_avg);
1008         P(se.avg.util_avg);
1009         P(se.avg.last_update_time);
1010 #endif
1011         P(policy);
1012         P(prio);
1013         if (p->policy == SCHED_DEADLINE) {
1014                 P(dl.runtime);
1015                 P(dl.deadline);
1016         }
1017 #undef PN_SCHEDSTAT
1018 #undef PN
1019 #undef __PN
1020 #undef P_SCHEDSTAT
1021 #undef P
1022 #undef __P
1023
1024         {
1025                 unsigned int this_cpu = raw_smp_processor_id();
1026                 u64 t0, t1;
1027
1028                 t0 = cpu_clock(this_cpu);
1029                 t1 = cpu_clock(this_cpu);
1030                 SEQ_printf(m, "%-45s:%21Ld\n",
1031                            "clock-delta", (long long)(t1-t0));
1032         }
1033
1034         sched_show_numa(p, m);
1035 }
1036
1037 void proc_sched_set_task(struct task_struct *p)
1038 {
1039 #ifdef CONFIG_SCHEDSTATS
1040         memset(&p->se.statistics, 0, sizeof(p->se.statistics));
1041 #endif
1042 }