aboutsummaryrefslogtreecommitdiffstats
path: root/kernel/sched/debug.c
diff options
context:
space:
mode:
Diffstat (limited to 'kernel/sched/debug.c')
-rw-r--r--kernel/sched/debug.c428
1 files changed, 370 insertions, 58 deletions
diff --git a/kernel/sched/debug.c b/kernel/sched/debug.c
index 641511771ae6..0368c393a336 100644
--- a/kernel/sched/debug.c
+++ b/kernel/sched/debug.c
@@ -16,6 +16,7 @@
16#include <linux/kallsyms.h> 16#include <linux/kallsyms.h>
17#include <linux/utsname.h> 17#include <linux/utsname.h>
18#include <linux/mempolicy.h> 18#include <linux/mempolicy.h>
19#include <linux/debugfs.h>
19 20
20#include "sched.h" 21#include "sched.h"
21 22
@@ -58,6 +59,309 @@ static unsigned long nsec_low(unsigned long long nsec)
58 59
59#define SPLIT_NS(x) nsec_high(x), nsec_low(x) 60#define SPLIT_NS(x) nsec_high(x), nsec_low(x)
60 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
61#ifdef CONFIG_FAIR_GROUP_SCHED 365#ifdef CONFIG_FAIR_GROUP_SCHED
62static void print_cfs_group_stats(struct seq_file *m, int cpu, struct task_group *tg) 366static void print_cfs_group_stats(struct seq_file *m, int cpu, struct task_group *tg)
63{ 367{
@@ -75,16 +379,18 @@ static void print_cfs_group_stats(struct seq_file *m, int cpu, struct task_group
75 PN(se->vruntime); 379 PN(se->vruntime);
76 PN(se->sum_exec_runtime); 380 PN(se->sum_exec_runtime);
77#ifdef CONFIG_SCHEDSTATS 381#ifdef CONFIG_SCHEDSTATS
78 PN(se->statistics.wait_start); 382 if (schedstat_enabled()) {
79 PN(se->statistics.sleep_start); 383 PN(se->statistics.wait_start);
80 PN(se->statistics.block_start); 384 PN(se->statistics.sleep_start);
81 PN(se->statistics.sleep_max); 385 PN(se->statistics.block_start);
82 PN(se->statistics.block_max); 386 PN(se->statistics.sleep_max);
83 PN(se->statistics.exec_max); 387 PN(se->statistics.block_max);
84 PN(se->statistics.slice_max); 388 PN(se->statistics.exec_max);
85 PN(se->statistics.wait_max); 389 PN(se->statistics.slice_max);
86 PN(se->statistics.wait_sum); 390 PN(se->statistics.wait_max);
87 P(se->statistics.wait_count); 391 PN(se->statistics.wait_sum);
392 P(se->statistics.wait_count);
393 }
88#endif 394#endif
89 P(se->load.weight); 395 P(se->load.weight);
90#ifdef CONFIG_SMP 396#ifdef CONFIG_SMP
@@ -121,17 +427,12 @@ print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
121 SPLIT_NS(p->se.vruntime), 427 SPLIT_NS(p->se.vruntime),
122 (long long)(p->nvcsw + p->nivcsw), 428 (long long)(p->nvcsw + p->nivcsw),
123 p->prio); 429 p->prio);
124#ifdef CONFIG_SCHEDSTATS 430
125 SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
126 SPLIT_NS(p->se.statistics.wait_sum),
127 SPLIT_NS(p->se.sum_exec_runtime),
128 SPLIT_NS(p->se.statistics.sum_sleep_runtime));
129#else
130 SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld", 431 SEQ_printf(m, "%9Ld.%06ld %9Ld.%06ld %9Ld.%06ld",
131 0LL, 0L, 432 SPLIT_NS(schedstat_val(p, se.statistics.wait_sum)),
132 SPLIT_NS(p->se.sum_exec_runtime), 433 SPLIT_NS(p->se.sum_exec_runtime),
133 0LL, 0L); 434 SPLIT_NS(schedstat_val(p, se.statistics.sum_sleep_runtime)));
134#endif 435
135#ifdef CONFIG_NUMA_BALANCING 436#ifdef CONFIG_NUMA_BALANCING
136 SEQ_printf(m, " %d %d", task_node(p), task_numa_group_id(p)); 437 SEQ_printf(m, " %d %d", task_node(p), task_numa_group_id(p));
137#endif 438#endif
@@ -258,8 +559,17 @@ void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
258 559
259void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq) 560void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq)
260{ 561{
562 struct dl_bw *dl_bw;
563
261 SEQ_printf(m, "\ndl_rq[%d]:\n", cpu); 564 SEQ_printf(m, "\ndl_rq[%d]:\n", cpu);
262 SEQ_printf(m, " .%-30s: %ld\n", "dl_nr_running", dl_rq->dl_nr_running); 565 SEQ_printf(m, " .%-30s: %ld\n", "dl_nr_running", dl_rq->dl_nr_running);
566#ifdef CONFIG_SMP
567 dl_bw = &cpu_rq(cpu)->rd->dl_bw;
568#else
569 dl_bw = &dl_rq->dl_bw;
570#endif
571 SEQ_printf(m, " .%-30s: %lld\n", "dl_bw->bw", dl_bw->bw);
572 SEQ_printf(m, " .%-30s: %lld\n", "dl_bw->total_bw", dl_bw->total_bw);
263} 573}
264 574
265extern __read_mostly int sched_clock_running; 575extern __read_mostly int sched_clock_running;
@@ -309,24 +619,25 @@ do { \
309#undef P 619#undef P
310#undef PN 620#undef PN
311 621
312#ifdef CONFIG_SCHEDSTATS
313#define P(n) SEQ_printf(m, " .%-30s: %d\n", #n, rq->n);
314#define P64(n) SEQ_printf(m, " .%-30s: %Ld\n", #n, rq->n);
315
316 P(yld_count);
317
318 P(sched_count);
319 P(sched_goidle);
320#ifdef CONFIG_SMP 622#ifdef CONFIG_SMP
623#define P64(n) SEQ_printf(m, " .%-30s: %Ld\n", #n, rq->n);
321 P64(avg_idle); 624 P64(avg_idle);
322 P64(max_idle_balance_cost); 625 P64(max_idle_balance_cost);
626#undef P64
323#endif 627#endif
324 628
325 P(ttwu_count); 629#ifdef CONFIG_SCHEDSTATS
326 P(ttwu_local); 630#define P(n) SEQ_printf(m, " .%-30s: %d\n", #n, rq->n);
631
632 if (schedstat_enabled()) {
633 P(yld_count);
634 P(sched_count);
635 P(sched_goidle);
636 P(ttwu_count);
637 P(ttwu_local);
638 }
327 639
328#undef P 640#undef P
329#undef P64
330#endif 641#endif
331 spin_lock_irqsave(&sched_debug_lock, flags); 642 spin_lock_irqsave(&sched_debug_lock, flags);
332 print_cfs_stats(m, cpu); 643 print_cfs_stats(m, cpu);
@@ -569,38 +880,39 @@ void proc_sched_show_task(struct task_struct *p, struct seq_file *m)
569 nr_switches = p->nvcsw + p->nivcsw; 880 nr_switches = p->nvcsw + p->nivcsw;
570 881
571#ifdef CONFIG_SCHEDSTATS 882#ifdef CONFIG_SCHEDSTATS
572 PN(se.statistics.sum_sleep_runtime);
573 PN(se.statistics.wait_start);
574 PN(se.statistics.sleep_start);
575 PN(se.statistics.block_start);
576 PN(se.statistics.sleep_max);
577 PN(se.statistics.block_max);
578 PN(se.statistics.exec_max);
579 PN(se.statistics.slice_max);
580 PN(se.statistics.wait_max);
581 PN(se.statistics.wait_sum);
582 P(se.statistics.wait_count);
583 PN(se.statistics.iowait_sum);
584 P(se.statistics.iowait_count);
585 P(se.nr_migrations); 883 P(se.nr_migrations);
586 P(se.statistics.nr_migrations_cold);
587 P(se.statistics.nr_failed_migrations_affine);
588 P(se.statistics.nr_failed_migrations_running);
589 P(se.statistics.nr_failed_migrations_hot);
590 P(se.statistics.nr_forced_migrations);
591 P(se.statistics.nr_wakeups);
592 P(se.statistics.nr_wakeups_sync);
593 P(se.statistics.nr_wakeups_migrate);
594 P(se.statistics.nr_wakeups_local);
595 P(se.statistics.nr_wakeups_remote);
596 P(se.statistics.nr_wakeups_affine);
597 P(se.statistics.nr_wakeups_affine_attempts);
598 P(se.statistics.nr_wakeups_passive);
599 P(se.statistics.nr_wakeups_idle);
600 884
601 { 885 if (schedstat_enabled()) {
602 u64 avg_atom, avg_per_cpu; 886 u64 avg_atom, avg_per_cpu;
603 887
888 PN(se.statistics.sum_sleep_runtime);
889 PN(se.statistics.wait_start);
890 PN(se.statistics.sleep_start);
891 PN(se.statistics.block_start);
892 PN(se.statistics.sleep_max);
893 PN(se.statistics.block_max);
894 PN(se.statistics.exec_max);
895 PN(se.statistics.slice_max);
896 PN(se.statistics.wait_max);
897 PN(se.statistics.wait_sum);
898 P(se.statistics.wait_count);
899 PN(se.statistics.iowait_sum);
900 P(se.statistics.iowait_count);
901 P(se.statistics.nr_migrations_cold);
902 P(se.statistics.nr_failed_migrations_affine);
903 P(se.statistics.nr_failed_migrations_running);
904 P(se.statistics.nr_failed_migrations_hot);
905 P(se.statistics.nr_forced_migrations);
906 P(se.statistics.nr_wakeups);
907 P(se.statistics.nr_wakeups_sync);
908 P(se.statistics.nr_wakeups_migrate);
909 P(se.statistics.nr_wakeups_local);
910 P(se.statistics.nr_wakeups_remote);
911 P(se.statistics.nr_wakeups_affine);
912 P(se.statistics.nr_wakeups_affine_attempts);
913 P(se.statistics.nr_wakeups_passive);
914 P(se.statistics.nr_wakeups_idle);
915
604 avg_atom = p->se.sum_exec_runtime; 916 avg_atom = p->se.sum_exec_runtime;
605 if (nr_switches) 917 if (nr_switches)
606 avg_atom = div64_ul(avg_atom, nr_switches); 918 avg_atom = div64_ul(avg_atom, nr_switches);
generated by cgit v1.2.2 (git 2.25.0) at 2025-04-27 01:43:19 -0400