History

sched,debug: Convert sysctl sched_domains to debugfs

Stop polluting sysctl, move to debugfs for SCHED_DEBUG stuff.

Peter Zijlstra committed 5 years ago

Commit 3b87f131 parent 9af04403 files · +59 −211Browse files

Unified · Split

3 files
  1. kernel/sched/debug.c +55 −199
  2. kernel/sched/sched.h +3 −7
  3. kernel/sched/topology.c +1 −5
kernel/sched/debug.c+55−199
@@ -299,6 +299,10 @@ static __init int sched_init_debug(void) · 298 unmodified lines
299 debugfs_create_file("tunable_scaling", 0644, debugfs_sched, NULL, &sched_scaling_fops);299 debugfs_create_file("tunable_scaling", 0644, debugfs_sched, NULL, &sched_scaling_fops);
300 debugfs_create_u32("migration_cost_ns", 0644, debugfs_sched, &sysctl_sched_migration_cost);300 debugfs_create_u32("migration_cost_ns", 0644, debugfs_sched, &sysctl_sched_migration_cost);
301 debugfs_create_u32("nr_migrate", 0644, debugfs_sched, &sysctl_sched_nr_migrate);301 debugfs_create_u32("nr_migrate", 0644, debugfs_sched, &sysctl_sched_nr_migrate);
302
303 mutex_lock(&sched_domains_mutex);
304 update_sched_domain_debugfs();
305 mutex_unlock(&sched_domains_mutex);
302#endif306#endif
303 307
304#ifdef CONFIG_NUMA_BALANCING308#ifdef CONFIG_NUMA_BALANCING
@@ -316,229 +320,88 @@ late_initcall(sched_init_debug); · 11 unmodified lines
316 320
317#ifdef CONFIG_SMP321#ifdef CONFIG_SMP
318 322
319#ifdef CONFIG_SYSCTL323static cpumask_var_t sd_sysctl_cpus;
320 324static struct dentry *sd_dentry;
321static struct ctl_table sd_ctl_dir[] = {
322 {
323 .procname = "sched_domain",
324 .mode = 0555,
325 },
326 {}
327};
328
329static struct ctl_table sd_ctl_root[] = {
330 {
331 .procname = "kernel",
332 .mode = 0555,
333 .child = sd_ctl_dir,
334 },
335 {}
336};
337
338static struct ctl_table *sd_alloc_ctl_entry(int n)
339{
340 struct ctl_table *entry =
341 kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL);
342
343 return entry;
344}
345
346static void sd_free_ctl_entry(struct ctl_table **tablep)
347{
348 struct ctl_table *entry;
349
350 /*
351 * In the intermediate directories, both the child directory and
352 * procname are dynamically allocated and could fail but the mode
353 * will always be set. In the lowest directory the names are
354 * static strings and all have proc handlers.
355 */
356 for (entry = *tablep; entry->mode; entry++) {
357 if (entry->child)
358 sd_free_ctl_entry(&entry->child);
359 if (entry->proc_handler == NULL)
360 kfree(entry->procname);
361 }
362
363 kfree(*tablep);
364 *tablep = NULL;
365}
366
367static void
368set_table_entry(struct ctl_table *entry,
369 const char *procname, void *data, int maxlen,
370 umode_t mode, proc_handler *proc_handler)
371{
372 entry->procname = procname;
373 entry->data = data;
374 entry->maxlen = maxlen;
375 entry->mode = mode;
376 entry->proc_handler = proc_handler;
377}
378 325
379static int sd_ctl_doflags(struct ctl_table *table, int write,326static int sd_flags_show(struct seq_file *m, void *v)
380 void *buffer, size_t *lenp, loff_t *ppos)
381{327{
382 unsigned long flags = *(unsigned long *)table->data;328 unsigned long flags = *(unsigned int *)m->private;
383 size_t data_size = 0;
384 size_t len = 0;
385 char *tmp, *buf;
386 int idx;329 int idx;
387 330
388 if (write)
389 return 0;
390
391 for_each_set_bit(idx, &flags, __SD_FLAG_CNT) {
392 char *name = sd_flag_debug[idx].name;
393
394 /* Name plus whitespace */
395 data_size += strlen(name) + 1;
396 }
397
398 if (*ppos > data_size) {
399 *lenp = 0;
400 return 0;
401 }
402
403 buf = kcalloc(data_size + 1, sizeof(*buf), GFP_KERNEL);
404 if (!buf)
405 return -ENOMEM;
406
407 for_each_set_bit(idx, &flags, __SD_FLAG_CNT) {331 for_each_set_bit(idx, &flags, __SD_FLAG_CNT) {
408 char *name = sd_flag_debug[idx].name;332 seq_puts(m, sd_flag_debug[idx].name);
409 333 seq_puts(m, " ");
410 len += snprintf(buf + len, strlen(name) + 2, "%s ", name);
411 }
412
413 tmp = buf + *ppos;
414 len -= *ppos;
415
416 if (len > *lenp)
417 len = *lenp;
418 if (len)
419 memcpy(buffer, tmp, len);
420 if (len < *lenp) {
421 ((char *)buffer)[len] = '\n';
422 len++;
423 }334 }
424 335 seq_puts(m, "\n");
425 *lenp = len;
426 *ppos += len;
427
428 kfree(buf);
429 336
430 return 0;337 return 0;
431}338}
432 339
433static struct ctl_table *340static int sd_flags_open(struct inode *inode, struct file *file)
434sd_alloc_ctl_domain_table(struct sched_domain *sd)
435{
436 struct ctl_table *table = sd_alloc_ctl_entry(9);
437
438 if (table == NULL)
439 return NULL;
440
441 set_table_entry(&table[0], "min_interval", &sd->min_interval, sizeof(long), 0644, proc_doulongvec_minmax);
442 set_table_entry(&table[1], "max_interval", &sd->max_interval, sizeof(long), 0644, proc_doulongvec_minmax);
443 set_table_entry(&table[2], "busy_factor", &sd->busy_factor, sizeof(int), 0644, proc_dointvec_minmax);
444 set_table_entry(&table[3], "imbalance_pct", &sd->imbalance_pct, sizeof(int), 0644, proc_dointvec_minmax);
445 set_table_entry(&table[4], "cache_nice_tries", &sd->cache_nice_tries, sizeof(int), 0644, proc_dointvec_minmax);
446 set_table_entry(&table[5], "flags", &sd->flags, sizeof(int), 0444, sd_ctl_doflags);
447 set_table_entry(&table[6], "max_newidle_lb_cost", &sd->max_newidle_lb_cost, sizeof(long), 0644, proc_doulongvec_minmax);
448 set_table_entry(&table[7], "name", sd->name, CORENAME_MAX_SIZE, 0444, proc_dostring);
449 /* &table[8] is terminator */
450
451 return table;
452}
453
454static struct ctl_table *sd_alloc_ctl_cpu_table(int cpu)
455{341{
456 struct ctl_table *entry, *table;342 return single_open(file, sd_flags_show, inode->i_private);
457 struct sched_domain *sd;
458 int domain_num = 0, i;
459 char buf[32];
460
461 for_each_domain(cpu, sd)
462 domain_num++;
463 entry = table = sd_alloc_ctl_entry(domain_num + 1);
464 if (table == NULL)
465 return NULL;
466
467 i = 0;
468 for_each_domain(cpu, sd) {
469 snprintf(buf, 32, "domain%d", i);
470 entry->procname = kstrdup(buf, GFP_KERNEL);
471 entry->mode = 0555;
472 entry->child = sd_alloc_ctl_domain_table(sd);
473 entry++;
474 i++;
475 }
476 return table;
477}343}
478 344
479static cpumask_var_t sd_sysctl_cpus;345static const struct file_operations sd_flags_fops = {
480static struct ctl_table_header *sd_sysctl_header;346 .open = sd_flags_open,
347 .read = seq_read,
348 .llseek = seq_lseek,
349 .release = single_release,
350};
481 351
482void register_sched_domain_sysctl(void)352static void register_sd(struct sched_domain *sd, struct dentry *parent)
483{353{
484 static struct ctl_table *cpu_entries;354#define SDM(type, mode, member) \
485 static struct ctl_table **cpu_idx;355 debugfs_create_##type(#member, mode, parent, &sd->member)
486 static bool init_done = false;
487 char buf[32];
488 int i;
489 356
490 if (!cpu_entries) {357 SDM(ulong, 0644, min_interval);
491 cpu_entries = sd_alloc_ctl_entry(num_possible_cpus() + 1);358 SDM(ulong, 0644, max_interval);
492 if (!cpu_entries)359 SDM(u64, 0644, max_newidle_lb_cost);
493 return;360 SDM(u32, 0644, busy_factor);
494 361 SDM(u32, 0644, imbalance_pct);
495 WARN_ON(sd_ctl_dir[0].child);362 SDM(u32, 0644, cache_nice_tries);
496 sd_ctl_dir[0].child = cpu_entries;363 SDM(str, 0444, name);
497 }
498 364
499 if (!cpu_idx) {365#undef SDM
500 struct ctl_table *e = cpu_entries;
501 366
502 cpu_idx = kcalloc(nr_cpu_ids, sizeof(struct ctl_table*), GFP_KERNEL);367 debugfs_create_file("flags", 0444, parent, &sd->flags, &sd_flags_fops);
503 if (!cpu_idx)368}
504 return;
505 369
506 /* deal with sparse possible map */370void update_sched_domain_debugfs(void)
507 for_each_possible_cpu(i) {371{
508 cpu_idx[i] = e;372 int cpu, i;
509 e++;
510 }
511 }
512 373
513 if (!cpumask_available(sd_sysctl_cpus)) {374 if (!cpumask_available(sd_sysctl_cpus)) {
514 if (!alloc_cpumask_var(&sd_sysctl_cpus, GFP_KERNEL))375 if (!alloc_cpumask_var(&sd_sysctl_cpus, GFP_KERNEL))
515 return;376 return;
516 }
517
518 if (!init_done) {
519 init_done = true;
520 /* init to possible to not have holes in @cpu_entries */
521 cpumask_copy(sd_sysctl_cpus, cpu_possible_mask);377 cpumask_copy(sd_sysctl_cpus, cpu_possible_mask);
522 }378 }
523 379
524 for_each_cpu(i, sd_sysctl_cpus) {380 if (!sd_dentry)
525 struct ctl_table *e = cpu_idx[i];381 sd_dentry = debugfs_create_dir("domains", debugfs_sched);
382
383 for_each_cpu(cpu, sd_sysctl_cpus) {
384 struct sched_domain *sd;
385 struct dentry *d_cpu;
386 char buf[32];
387
388 snprintf(buf, sizeof(buf), "cpu%d", cpu);
389 debugfs_remove(debugfs_lookup(buf, sd_dentry));
390 d_cpu = debugfs_create_dir(buf, sd_dentry);
526 391
527 if (e->child)392 i = 0;
528 sd_free_ctl_entry(&e->child);393 for_each_domain(cpu, sd) {
394 struct dentry *d_sd;
529 395
530 if (!e->procname) {396 snprintf(buf, sizeof(buf), "domain%d", i);
531 snprintf(buf, 32, "cpu%d", i);397 d_sd = debugfs_create_dir(buf, d_cpu);
532 e->procname = kstrdup(buf, GFP_KERNEL);398
399 register_sd(sd, d_sd);
400 i++;
533 }401 }
534 e->mode = 0555;
535 e->child = sd_alloc_ctl_cpu_table(i);
536 402
537 __cpumask_clear_cpu(i, sd_sysctl_cpus);403 __cpumask_clear_cpu(cpu, sd_sysctl_cpus);
538 }404 }
539
540 WARN_ON(sd_sysctl_header);
541 sd_sysctl_header = register_sysctl_table(sd_ctl_root);
542}405}
543 406
544void dirty_sched_domain_sysctl(int cpu)407void dirty_sched_domain_sysctl(int cpu)
@@ -547,13 +410,6 @@ void dirty_sched_domain_sysctl(int cpu) · 2 unmodified lines
547 __cpumask_set_cpu(cpu, sd_sysctl_cpus);410 __cpumask_set_cpu(cpu, sd_sysctl_cpus);
548}411}
549 412
550/* may be called multiple times per register */
551void unregister_sched_domain_sysctl(void)
552{
553 unregister_sysctl_table(sd_sysctl_header);
554 sd_sysctl_header = NULL;
555}
556#endif /* CONFIG_SYSCTL */
557#endif /* CONFIG_SMP */413#endif /* CONFIG_SMP */
558 414
559#ifdef CONFIG_FAIR_GROUP_SCHED415#ifdef CONFIG_FAIR_GROUP_SCHED
kernel/sched/sched.h+3−7
@@ -1552,20 +1552,16 @@ static inline unsigned int group_first_cpu(struct sched_group *group) · 1551 unmodified lines
1552 1552
1553extern int group_balance_cpu(struct sched_group *sg);1553extern int group_balance_cpu(struct sched_group *sg);
1554 1554
1555#if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL)1555#ifdef CONFIG_SCHED_DEBUG
1556void register_sched_domain_sysctl(void);1556void update_sched_domain_debugfs(void);
1557void dirty_sched_domain_sysctl(int cpu);1557void dirty_sched_domain_sysctl(int cpu);
1558void unregister_sched_domain_sysctl(void);
1559#else1558#else
1560static inline void register_sched_domain_sysctl(void)1559static inline void update_sched_domain_debugfs(void)
1561{1560{
1562}1561}
1563static inline void dirty_sched_domain_sysctl(int cpu)1562static inline void dirty_sched_domain_sysctl(int cpu)
1564{1563{
1565}1564}
1566static inline void unregister_sched_domain_sysctl(void)
1567{
1568}
1569#endif1565#endif
1570 1566
1571extern int sched_update_scaling(void);1567extern int sched_update_scaling(void);
kernel/sched/topology.c+1−5
@@ -2223,7 +2223,6 @@ int sched_init_domains(const struct cpumask *cpu_map) · 2222 unmodified lines
2223 doms_cur = &fallback_doms;2223 doms_cur = &fallback_doms;
2224 cpumask_and(doms_cur[0], cpu_map, housekeeping_cpumask(HK_FLAG_DOMAIN));2224 cpumask_and(doms_cur[0], cpu_map, housekeeping_cpumask(HK_FLAG_DOMAIN));
2225 err = build_sched_domains(doms_cur[0], NULL);2225 err = build_sched_domains(doms_cur[0], NULL);
2226 register_sched_domain_sysctl();
2227 2226
2228 return err;2227 return err;
2229}2228}
@@ -2298,9 +2297,6 @@ void partition_sched_domains_locked(int ndoms_new, cpumask_var_t doms_new[], · 68 unmodified lines
2298 2297
2299 lockdep_assert_held(&sched_domains_mutex);2298 lockdep_assert_held(&sched_domains_mutex);
2300 2299
2301 /* Always unregister in case we don't destroy any domains: */
2302 unregister_sched_domain_sysctl();
2303
2304 /* Let the architecture update CPU core mappings: */2300 /* Let the architecture update CPU core mappings: */
2305 new_topology = arch_update_cpu_topology();2301 new_topology = arch_update_cpu_topology();
2306 2302
@@ -2389,7 +2385,7 @@ match3: · 82 unmodified lines
2389 dattr_cur = dattr_new;2385 dattr_cur = dattr_new;
2390 ndoms_cur = ndoms_new;2386 ndoms_cur = ndoms_new;
2391 2387
2392 register_sched_domain_sysctl();2388 update_sched_domain_debugfs();
2393}2389}
2394 2390
2395/*2391/*

Keyboard shortcuts

General

Show keyboard shortcuts
?
Focus the current search
/

Issues and pull requests

Open filters
f

Issues, pull requests and Inbox

Select next item
j↓
Select previous item
k↑
Open selected item
Enter
Dismiss filters or clear selection
Esc

Repository code

Go to file
t