4 * Copyright (C) 2008 Red Hat Inc, Steven Rostedt <srostedt@redhat.com>
6 * - Added format output of fields of the trace point.
7 * This was based off of work by Tom Zanussi <tzanussi@gmail.com>.
11 #define pr_fmt(fmt) fmt
13 #include <linux/workqueue.h>
14 #include <linux/spinlock.h>
15 #include <linux/kthread.h>
16 #include <linux/tracefs.h>
17 #include <linux/uaccess.h>
18 #include <linux/bsearch.h>
19 #include <linux/module.h>
20 #include <linux/ctype.h>
21 #include <linux/sort.h>
22 #include <linux/slab.h>
23 #include <linux/delay.h>
25 #include <trace/events/sched.h>
27 #include <asm/setup.h>
29 #include "trace_output.h"
32 #define TRACE_SYSTEM "TRACE_SYSTEM"
34 DEFINE_MUTEX(event_mutex);
36 LIST_HEAD(ftrace_events);
37 static LIST_HEAD(ftrace_generic_fields);
38 static LIST_HEAD(ftrace_common_fields);
40 #define GFP_TRACE (GFP_KERNEL | __GFP_ZERO)
42 static struct kmem_cache *field_cachep;
43 static struct kmem_cache *file_cachep;
45 static inline int system_refcount(struct event_subsystem *system)
47 return system->ref_count;
50 static int system_refcount_inc(struct event_subsystem *system)
52 return system->ref_count++;
55 static int system_refcount_dec(struct event_subsystem *system)
57 return --system->ref_count;
60 /* Double loops, do not use break, only goto's work */
61 #define do_for_each_event_file(tr, file) \
62 list_for_each_entry(tr, &ftrace_trace_arrays, list) { \
63 list_for_each_entry(file, &tr->events, list)
65 #define do_for_each_event_file_safe(tr, file) \
66 list_for_each_entry(tr, &ftrace_trace_arrays, list) { \
67 struct trace_event_file *___n; \
68 list_for_each_entry_safe(file, ___n, &tr->events, list)
70 #define while_for_each_event_file() \
73 static struct list_head *
74 trace_get_fields(struct trace_event_call *event_call)
76 if (!event_call->class->get_fields)
77 return &event_call->class->fields;
78 return event_call->class->get_fields(event_call);
81 static struct ftrace_event_field *
82 __find_event_field(struct list_head *head, char *name)
84 struct ftrace_event_field *field;
86 list_for_each_entry(field, head, link) {
87 if (!strcmp(field->name, name))
94 struct ftrace_event_field *
95 trace_find_event_field(struct trace_event_call *call, char *name)
97 struct ftrace_event_field *field;
98 struct list_head *head;
100 head = trace_get_fields(call);
101 field = __find_event_field(head, name);
105 field = __find_event_field(&ftrace_generic_fields, name);
109 return __find_event_field(&ftrace_common_fields, name);
112 static int __trace_define_field(struct list_head *head, const char *type,
113 const char *name, int offset, int size,
114 int is_signed, int filter_type)
116 struct ftrace_event_field *field;
118 field = kmem_cache_alloc(field_cachep, GFP_TRACE);
125 if (filter_type == FILTER_OTHER)
126 field->filter_type = filter_assign_type(type);
128 field->filter_type = filter_type;
130 field->offset = offset;
132 field->is_signed = is_signed;
134 list_add(&field->link, head);
139 int trace_define_field(struct trace_event_call *call, const char *type,
140 const char *name, int offset, int size, int is_signed,
143 struct list_head *head;
145 if (WARN_ON(!call->class))
148 head = trace_get_fields(call);
149 return __trace_define_field(head, type, name, offset, size,
150 is_signed, filter_type);
152 EXPORT_SYMBOL_GPL(trace_define_field);
154 #define __generic_field(type, item, filter_type) \
155 ret = __trace_define_field(&ftrace_generic_fields, #type, \
156 #item, 0, 0, is_signed_type(type), \
161 #define __common_field(type, item) \
162 ret = __trace_define_field(&ftrace_common_fields, #type, \
164 offsetof(typeof(ent), item), \
166 is_signed_type(type), FILTER_OTHER); \
170 static int trace_define_generic_fields(void)
174 __generic_field(int, CPU, FILTER_CPU);
175 __generic_field(int, cpu, FILTER_CPU);
176 __generic_field(char *, COMM, FILTER_COMM);
177 __generic_field(char *, comm, FILTER_COMM);
182 static int trace_define_common_fields(void)
185 struct trace_entry ent;
187 __common_field(unsigned short, type);
188 __common_field(unsigned char, flags);
189 __common_field(unsigned char, preempt_count);
190 __common_field(int, pid);
195 static void trace_destroy_fields(struct trace_event_call *call)
197 struct ftrace_event_field *field, *next;
198 struct list_head *head;
200 head = trace_get_fields(call);
201 list_for_each_entry_safe(field, next, head, link) {
202 list_del(&field->link);
203 kmem_cache_free(field_cachep, field);
207 int trace_event_raw_init(struct trace_event_call *call)
211 id = register_trace_event(&call->event);
217 EXPORT_SYMBOL_GPL(trace_event_raw_init);
219 bool trace_event_ignore_this_pid(struct trace_event_file *trace_file)
221 struct trace_array *tr = trace_file->tr;
222 struct trace_array_cpu *data;
223 struct trace_pid_list *pid_list;
225 pid_list = rcu_dereference_sched(tr->filtered_pids);
229 data = this_cpu_ptr(tr->trace_buffer.data);
231 return data->ignore_pid;
233 EXPORT_SYMBOL_GPL(trace_event_ignore_this_pid);
235 void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer,
236 struct trace_event_file *trace_file,
239 struct trace_event_call *event_call = trace_file->event_call;
241 if ((trace_file->flags & EVENT_FILE_FL_PID_FILTER) &&
242 trace_event_ignore_this_pid(trace_file))
245 local_save_flags(fbuffer->flags);
246 fbuffer->pc = preempt_count();
247 fbuffer->trace_file = trace_file;
250 trace_event_buffer_lock_reserve(&fbuffer->buffer, trace_file,
251 event_call->event.type, len,
252 fbuffer->flags, fbuffer->pc);
256 fbuffer->entry = ring_buffer_event_data(fbuffer->event);
257 return fbuffer->entry;
259 EXPORT_SYMBOL_GPL(trace_event_buffer_reserve);
261 static DEFINE_SPINLOCK(tracepoint_iter_lock);
263 static void output_printk(struct trace_event_buffer *fbuffer)
265 struct trace_event_call *event_call;
266 struct trace_event *event;
268 struct trace_iterator *iter = tracepoint_print_iter;
273 event_call = fbuffer->trace_file->event_call;
274 if (!event_call || !event_call->event.funcs ||
275 !event_call->event.funcs->trace)
278 event = &fbuffer->trace_file->event_call->event;
280 spin_lock_irqsave(&tracepoint_iter_lock, flags);
281 trace_seq_init(&iter->seq);
282 iter->ent = fbuffer->entry;
283 event_call->event.funcs->trace(iter, 0, event);
284 trace_seq_putc(&iter->seq, 0);
285 printk("%s", iter->seq.buffer);
287 spin_unlock_irqrestore(&tracepoint_iter_lock, flags);
290 void trace_event_buffer_commit(struct trace_event_buffer *fbuffer)
292 if (tracepoint_printk)
293 output_printk(fbuffer);
295 event_trigger_unlock_commit(fbuffer->trace_file, fbuffer->buffer,
296 fbuffer->event, fbuffer->entry,
297 fbuffer->flags, fbuffer->pc);
299 EXPORT_SYMBOL_GPL(trace_event_buffer_commit);
301 int trace_event_reg(struct trace_event_call *call,
302 enum trace_reg type, void *data)
304 struct trace_event_file *file = data;
306 WARN_ON(!(call->flags & TRACE_EVENT_FL_TRACEPOINT));
308 case TRACE_REG_REGISTER:
309 return tracepoint_probe_register(call->tp,
312 case TRACE_REG_UNREGISTER:
313 tracepoint_probe_unregister(call->tp,
318 #ifdef CONFIG_PERF_EVENTS
319 case TRACE_REG_PERF_REGISTER:
320 return tracepoint_probe_register(call->tp,
321 call->class->perf_probe,
323 case TRACE_REG_PERF_UNREGISTER:
324 tracepoint_probe_unregister(call->tp,
325 call->class->perf_probe,
328 case TRACE_REG_PERF_OPEN:
329 case TRACE_REG_PERF_CLOSE:
330 case TRACE_REG_PERF_ADD:
331 case TRACE_REG_PERF_DEL:
337 EXPORT_SYMBOL_GPL(trace_event_reg);
339 void trace_event_enable_cmd_record(bool enable)
341 struct trace_event_file *file;
342 struct trace_array *tr;
344 mutex_lock(&event_mutex);
345 do_for_each_event_file(tr, file) {
347 if (!(file->flags & EVENT_FILE_FL_ENABLED))
351 tracing_start_cmdline_record();
352 set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
354 tracing_stop_cmdline_record();
355 clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
357 } while_for_each_event_file();
358 mutex_unlock(&event_mutex);
361 static int __ftrace_event_enable_disable(struct trace_event_file *file,
362 int enable, int soft_disable)
364 struct trace_event_call *call = file->event_call;
365 struct trace_array *tr = file->tr;
372 * When soft_disable is set and enable is cleared, the sm_ref
373 * reference counter is decremented. If it reaches 0, we want
374 * to clear the SOFT_DISABLED flag but leave the event in the
375 * state that it was. That is, if the event was enabled and
376 * SOFT_DISABLED isn't set, then do nothing. But if SOFT_DISABLED
377 * is set we do not want the event to be enabled before we
380 * When soft_disable is not set but the SOFT_MODE flag is,
381 * we do nothing. Do not disable the tracepoint, otherwise
382 * "soft enable"s (clearing the SOFT_DISABLED bit) wont work.
385 if (atomic_dec_return(&file->sm_ref) > 0)
387 disable = file->flags & EVENT_FILE_FL_SOFT_DISABLED;
388 clear_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
390 disable = !(file->flags & EVENT_FILE_FL_SOFT_MODE);
392 if (disable && (file->flags & EVENT_FILE_FL_ENABLED)) {
393 clear_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
394 if (file->flags & EVENT_FILE_FL_RECORDED_CMD) {
395 tracing_stop_cmdline_record();
396 clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
398 call->class->reg(call, TRACE_REG_UNREGISTER, file);
400 /* If in SOFT_MODE, just set the SOFT_DISABLE_BIT, else clear it */
401 if (file->flags & EVENT_FILE_FL_SOFT_MODE)
402 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
404 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
408 * When soft_disable is set and enable is set, we want to
409 * register the tracepoint for the event, but leave the event
410 * as is. That means, if the event was already enabled, we do
411 * nothing (but set SOFT_MODE). If the event is disabled, we
412 * set SOFT_DISABLED before enabling the event tracepoint, so
413 * it still seems to be disabled.
416 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
418 if (atomic_inc_return(&file->sm_ref) > 1)
420 set_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
423 if (!(file->flags & EVENT_FILE_FL_ENABLED)) {
425 /* Keep the event disabled, when going to SOFT_MODE. */
427 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
429 if (tr->trace_flags & TRACE_ITER_RECORD_CMD) {
430 tracing_start_cmdline_record();
431 set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
433 ret = call->class->reg(call, TRACE_REG_REGISTER, file);
435 tracing_stop_cmdline_record();
436 pr_info("event trace: Could not enable event "
437 "%s\n", trace_event_name(call));
440 set_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
442 /* WAS_ENABLED gets set but never cleared. */
443 call->flags |= TRACE_EVENT_FL_WAS_ENABLED;
451 int trace_event_enable_disable(struct trace_event_file *file,
452 int enable, int soft_disable)
454 return __ftrace_event_enable_disable(file, enable, soft_disable);
457 static int ftrace_event_enable_disable(struct trace_event_file *file,
460 return __ftrace_event_enable_disable(file, enable, 0);
463 static void ftrace_clear_events(struct trace_array *tr)
465 struct trace_event_file *file;
467 mutex_lock(&event_mutex);
468 list_for_each_entry(file, &tr->events, list) {
469 ftrace_event_enable_disable(file, 0);
471 mutex_unlock(&event_mutex);
474 static int cmp_pid(const void *key, const void *elt)
476 const pid_t *search_pid = key;
477 const pid_t *pid = elt;
479 if (*search_pid == *pid)
481 if (*search_pid < *pid)
487 check_ignore_pid(struct trace_pid_list *filtered_pids, struct task_struct *task)
493 * Return false, because if filtered_pids does not exist,
494 * all pids are good to trace.
499 search_pid = task->pid;
501 pid = bsearch(&search_pid, filtered_pids->pids,
502 filtered_pids->nr_pids, sizeof(pid_t),
511 event_filter_pid_sched_switch_probe_pre(void *data, bool preempt,
512 struct task_struct *prev, struct task_struct *next)
514 struct trace_array *tr = data;
515 struct trace_pid_list *pid_list;
517 pid_list = rcu_dereference_sched(tr->filtered_pids);
519 this_cpu_write(tr->trace_buffer.data->ignore_pid,
520 check_ignore_pid(pid_list, prev) &&
521 check_ignore_pid(pid_list, next));
525 event_filter_pid_sched_switch_probe_post(void *data, bool preempt,
526 struct task_struct *prev, struct task_struct *next)
528 struct trace_array *tr = data;
529 struct trace_pid_list *pid_list;
531 pid_list = rcu_dereference_sched(tr->filtered_pids);
533 this_cpu_write(tr->trace_buffer.data->ignore_pid,
534 check_ignore_pid(pid_list, next));
538 event_filter_pid_sched_wakeup_probe_pre(void *data, struct task_struct *task)
540 struct trace_array *tr = data;
541 struct trace_pid_list *pid_list;
543 /* Nothing to do if we are already tracing */
544 if (!this_cpu_read(tr->trace_buffer.data->ignore_pid))
547 pid_list = rcu_dereference_sched(tr->filtered_pids);
549 this_cpu_write(tr->trace_buffer.data->ignore_pid,
550 check_ignore_pid(pid_list, task));
554 event_filter_pid_sched_wakeup_probe_post(void *data, struct task_struct *task)
556 struct trace_array *tr = data;
557 struct trace_pid_list *pid_list;
559 /* Nothing to do if we are not tracing */
560 if (this_cpu_read(tr->trace_buffer.data->ignore_pid))
563 pid_list = rcu_dereference_sched(tr->filtered_pids);
565 /* Set tracing if current is enabled */
566 this_cpu_write(tr->trace_buffer.data->ignore_pid,
567 check_ignore_pid(pid_list, current));
570 static void __ftrace_clear_event_pids(struct trace_array *tr)
572 struct trace_pid_list *pid_list;
573 struct trace_event_file *file;
576 pid_list = rcu_dereference_protected(tr->filtered_pids,
577 lockdep_is_held(&event_mutex));
581 unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_pre, tr);
582 unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_post, tr);
584 unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre, tr);
585 unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_post, tr);
587 unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre, tr);
588 unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post, tr);
590 unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_pre, tr);
591 unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_post, tr);
593 list_for_each_entry(file, &tr->events, list) {
594 clear_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
597 for_each_possible_cpu(cpu)
598 per_cpu_ptr(tr->trace_buffer.data, cpu)->ignore_pid = false;
600 rcu_assign_pointer(tr->filtered_pids, NULL);
602 /* Wait till all users are no longer using pid filtering */
605 free_pages((unsigned long)pid_list->pids, pid_list->order);
609 static void ftrace_clear_event_pids(struct trace_array *tr)
611 mutex_lock(&event_mutex);
612 __ftrace_clear_event_pids(tr);
613 mutex_unlock(&event_mutex);
616 static void __put_system(struct event_subsystem *system)
618 struct event_filter *filter = system->filter;
620 WARN_ON_ONCE(system_refcount(system) == 0);
621 if (system_refcount_dec(system))
624 list_del(&system->list);
627 kfree(filter->filter_string);
630 kfree_const(system->name);
634 static void __get_system(struct event_subsystem *system)
636 WARN_ON_ONCE(system_refcount(system) == 0);
637 system_refcount_inc(system);
640 static void __get_system_dir(struct trace_subsystem_dir *dir)
642 WARN_ON_ONCE(dir->ref_count == 0);
644 __get_system(dir->subsystem);
647 static void __put_system_dir(struct trace_subsystem_dir *dir)
649 WARN_ON_ONCE(dir->ref_count == 0);
650 /* If the subsystem is about to be freed, the dir must be too */
651 WARN_ON_ONCE(system_refcount(dir->subsystem) == 1 && dir->ref_count != 1);
653 __put_system(dir->subsystem);
654 if (!--dir->ref_count)
658 static void put_system(struct trace_subsystem_dir *dir)
660 mutex_lock(&event_mutex);
661 __put_system_dir(dir);
662 mutex_unlock(&event_mutex);
665 static void remove_subsystem(struct trace_subsystem_dir *dir)
670 if (!--dir->nr_events) {
671 tracefs_remove_recursive(dir->entry);
672 list_del(&dir->list);
673 __put_system_dir(dir);
677 static void remove_event_file_dir(struct trace_event_file *file)
679 struct dentry *dir = file->dir;
680 struct dentry *child;
683 spin_lock(&dir->d_lock); /* probably unneeded */
684 list_for_each_entry(child, &dir->d_subdirs, d_child) {
685 if (d_really_is_positive(child)) /* probably unneeded */
686 d_inode(child)->i_private = NULL;
688 spin_unlock(&dir->d_lock);
690 tracefs_remove_recursive(dir);
693 list_del(&file->list);
694 remove_subsystem(file->system);
695 free_event_filter(file->filter);
696 kmem_cache_free(file_cachep, file);
700 * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
703 __ftrace_set_clr_event_nolock(struct trace_array *tr, const char *match,
704 const char *sub, const char *event, int set)
706 struct trace_event_file *file;
707 struct trace_event_call *call;
711 list_for_each_entry(file, &tr->events, list) {
713 call = file->event_call;
714 name = trace_event_name(call);
716 if (!name || !call->class || !call->class->reg)
719 if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
723 strcmp(match, name) != 0 &&
724 strcmp(match, call->class->system) != 0)
727 if (sub && strcmp(sub, call->class->system) != 0)
730 if (event && strcmp(event, name) != 0)
733 ftrace_event_enable_disable(file, set);
741 static int __ftrace_set_clr_event(struct trace_array *tr, const char *match,
742 const char *sub, const char *event, int set)
746 mutex_lock(&event_mutex);
747 ret = __ftrace_set_clr_event_nolock(tr, match, sub, event, set);
748 mutex_unlock(&event_mutex);
753 static int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set)
755 char *event = NULL, *sub = NULL, *match;
761 * The buf format can be <subsystem>:<event-name>
762 * *:<event-name> means any event by that name.
763 * :<event-name> is the same.
765 * <subsystem>:* means all events in that subsystem
766 * <subsystem>: means the same.
768 * <name> (no ':') means all events in a subsystem with
769 * the name <name> or any event that matches <name>
772 match = strsep(&buf, ":");
778 if (!strlen(sub) || strcmp(sub, "*") == 0)
780 if (!strlen(event) || strcmp(event, "*") == 0)
784 ret = __ftrace_set_clr_event(tr, match, sub, event, set);
786 /* Put back the colon to allow this to be called again */
794 * trace_set_clr_event - enable or disable an event
795 * @system: system name to match (NULL for any system)
796 * @event: event name to match (NULL for all events, within system)
797 * @set: 1 to enable, 0 to disable
799 * This is a way for other parts of the kernel to enable or disable
802 * Returns 0 on success, -EINVAL if the parameters do not match any
805 int trace_set_clr_event(const char *system, const char *event, int set)
807 struct trace_array *tr = top_trace_array();
812 return __ftrace_set_clr_event(tr, NULL, system, event, set);
814 EXPORT_SYMBOL_GPL(trace_set_clr_event);
816 /* 128 should be much more than enough */
817 #define EVENT_BUF_SIZE 127
820 ftrace_event_write(struct file *file, const char __user *ubuf,
821 size_t cnt, loff_t *ppos)
823 struct trace_parser parser;
824 struct seq_file *m = file->private_data;
825 struct trace_array *tr = m->private;
831 ret = tracing_update_buffers();
835 if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
838 read = trace_get_user(&parser, ubuf, cnt, ppos);
840 if (read >= 0 && trace_parser_loaded((&parser))) {
843 if (*parser.buffer == '!')
846 parser.buffer[parser.idx] = 0;
848 ret = ftrace_set_clr_event(tr, parser.buffer + !set, set);
856 trace_parser_put(&parser);
862 t_next(struct seq_file *m, void *v, loff_t *pos)
864 struct trace_event_file *file = v;
865 struct trace_event_call *call;
866 struct trace_array *tr = m->private;
870 list_for_each_entry_continue(file, &tr->events, list) {
871 call = file->event_call;
873 * The ftrace subsystem is for showing formats only.
874 * They can not be enabled or disabled via the event files.
876 if (call->class && call->class->reg &&
877 !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
884 static void *t_start(struct seq_file *m, loff_t *pos)
886 struct trace_event_file *file;
887 struct trace_array *tr = m->private;
890 mutex_lock(&event_mutex);
892 file = list_entry(&tr->events, struct trace_event_file, list);
893 for (l = 0; l <= *pos; ) {
894 file = t_next(m, file, &l);
902 s_next(struct seq_file *m, void *v, loff_t *pos)
904 struct trace_event_file *file = v;
905 struct trace_array *tr = m->private;
909 list_for_each_entry_continue(file, &tr->events, list) {
910 if (file->flags & EVENT_FILE_FL_ENABLED)
917 static void *s_start(struct seq_file *m, loff_t *pos)
919 struct trace_event_file *file;
920 struct trace_array *tr = m->private;
923 mutex_lock(&event_mutex);
925 file = list_entry(&tr->events, struct trace_event_file, list);
926 for (l = 0; l <= *pos; ) {
927 file = s_next(m, file, &l);
934 static int t_show(struct seq_file *m, void *v)
936 struct trace_event_file *file = v;
937 struct trace_event_call *call = file->event_call;
939 if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
940 seq_printf(m, "%s:", call->class->system);
941 seq_printf(m, "%s\n", trace_event_name(call));
946 static void t_stop(struct seq_file *m, void *p)
948 mutex_unlock(&event_mutex);
951 static void *p_start(struct seq_file *m, loff_t *pos)
954 struct trace_pid_list *pid_list;
955 struct trace_array *tr = m->private;
958 * Grab the mutex, to keep calls to p_next() having the same
959 * tr->filtered_pids as p_start() has.
960 * If we just passed the tr->filtered_pids around, then RCU would
961 * have been enough, but doing that makes things more complex.
963 mutex_lock(&event_mutex);
964 rcu_read_lock_sched();
966 pid_list = rcu_dereference_sched(tr->filtered_pids);
968 if (!pid_list || *pos >= pid_list->nr_pids)
971 return (void *)&pid_list->pids[*pos];
974 static void p_stop(struct seq_file *m, void *p)
977 rcu_read_unlock_sched();
978 mutex_unlock(&event_mutex);
982 p_next(struct seq_file *m, void *v, loff_t *pos)
984 struct trace_array *tr = m->private;
985 struct trace_pid_list *pid_list = rcu_dereference_sched(tr->filtered_pids);
989 if (*pos >= pid_list->nr_pids)
992 return (void *)&pid_list->pids[*pos];
995 static int p_show(struct seq_file *m, void *v)
999 seq_printf(m, "%d\n", *pid);
1004 event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1007 struct trace_event_file *file;
1008 unsigned long flags;
1011 mutex_lock(&event_mutex);
1012 file = event_file_data(filp);
1014 flags = file->flags;
1015 mutex_unlock(&event_mutex);
1020 if (flags & EVENT_FILE_FL_ENABLED &&
1021 !(flags & EVENT_FILE_FL_SOFT_DISABLED))
1024 if (flags & EVENT_FILE_FL_SOFT_DISABLED ||
1025 flags & EVENT_FILE_FL_SOFT_MODE)
1030 return simple_read_from_buffer(ubuf, cnt, ppos, buf, strlen(buf));
1034 event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1037 struct trace_event_file *file;
1041 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1045 ret = tracing_update_buffers();
1053 mutex_lock(&event_mutex);
1054 file = event_file_data(filp);
1056 ret = ftrace_event_enable_disable(file, val);
1057 mutex_unlock(&event_mutex);
1066 return ret ? ret : cnt;
1070 system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1073 const char set_to_char[4] = { '?', '0', '1', 'X' };
1074 struct trace_subsystem_dir *dir = filp->private_data;
1075 struct event_subsystem *system = dir->subsystem;
1076 struct trace_event_call *call;
1077 struct trace_event_file *file;
1078 struct trace_array *tr = dir->tr;
1083 mutex_lock(&event_mutex);
1084 list_for_each_entry(file, &tr->events, list) {
1085 call = file->event_call;
1086 if ((call->flags & TRACE_EVENT_FL_IGNORE_ENABLE) ||
1087 !trace_event_name(call) || !call->class || !call->class->reg)
1090 if (system && strcmp(call->class->system, system->name) != 0)
1094 * We need to find out if all the events are set
1095 * or if all events or cleared, or if we have
1098 set |= (1 << !!(file->flags & EVENT_FILE_FL_ENABLED));
1101 * If we have a mixture, no need to look further.
1106 mutex_unlock(&event_mutex);
1108 buf[0] = set_to_char[set];
1111 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
1117 system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1120 struct trace_subsystem_dir *dir = filp->private_data;
1121 struct event_subsystem *system = dir->subsystem;
1122 const char *name = NULL;
1126 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1130 ret = tracing_update_buffers();
1134 if (val != 0 && val != 1)
1138 * Opening of "enable" adds a ref count to system,
1139 * so the name is safe to use.
1142 name = system->name;
1144 ret = __ftrace_set_clr_event(dir->tr, NULL, name, NULL, val);
1158 FORMAT_FIELD_SEPERATOR = 2,
1159 FORMAT_PRINTFMT = 3,
1162 static void *f_next(struct seq_file *m, void *v, loff_t *pos)
1164 struct trace_event_call *call = event_file_data(m->private);
1165 struct list_head *common_head = &ftrace_common_fields;
1166 struct list_head *head = trace_get_fields(call);
1167 struct list_head *node = v;
1171 switch ((unsigned long)v) {
1176 case FORMAT_FIELD_SEPERATOR:
1180 case FORMAT_PRINTFMT:
1186 if (node == common_head)
1187 return (void *)FORMAT_FIELD_SEPERATOR;
1188 else if (node == head)
1189 return (void *)FORMAT_PRINTFMT;
1194 static int f_show(struct seq_file *m, void *v)
1196 struct trace_event_call *call = event_file_data(m->private);
1197 struct ftrace_event_field *field;
1198 const char *array_descriptor;
1200 switch ((unsigned long)v) {
1202 seq_printf(m, "name: %s\n", trace_event_name(call));
1203 seq_printf(m, "ID: %d\n", call->event.type);
1204 seq_puts(m, "format:\n");
1207 case FORMAT_FIELD_SEPERATOR:
1211 case FORMAT_PRINTFMT:
1212 seq_printf(m, "\nprint fmt: %s\n",
1217 field = list_entry(v, struct ftrace_event_field, link);
1219 * Smartly shows the array type(except dynamic array).
1222 * If TYPE := TYPE[LEN], it is shown:
1223 * field:TYPE VAR[LEN]
1225 array_descriptor = strchr(field->type, '[');
1227 if (!strncmp(field->type, "__data_loc", 10))
1228 array_descriptor = NULL;
1230 if (!array_descriptor)
1231 seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1232 field->type, field->name, field->offset,
1233 field->size, !!field->is_signed);
1235 seq_printf(m, "\tfield:%.*s %s%s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1236 (int)(array_descriptor - field->type),
1237 field->type, field->name,
1238 array_descriptor, field->offset,
1239 field->size, !!field->is_signed);
1244 static void *f_start(struct seq_file *m, loff_t *pos)
1246 void *p = (void *)FORMAT_HEADER;
1249 /* ->stop() is called even if ->start() fails */
1250 mutex_lock(&event_mutex);
1251 if (!event_file_data(m->private))
1252 return ERR_PTR(-ENODEV);
1254 while (l < *pos && p)
1255 p = f_next(m, p, &l);
1260 static void f_stop(struct seq_file *m, void *p)
1262 mutex_unlock(&event_mutex);
1265 static const struct seq_operations trace_format_seq_ops = {
1272 static int trace_format_open(struct inode *inode, struct file *file)
1277 ret = seq_open(file, &trace_format_seq_ops);
1281 m = file->private_data;
1288 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1290 int id = (long)event_file_data(filp);
1297 len = sprintf(buf, "%d\n", id);
1299 return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
1303 event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1306 struct trace_event_file *file;
1307 struct trace_seq *s;
1313 s = kmalloc(sizeof(*s), GFP_KERNEL);
1320 mutex_lock(&event_mutex);
1321 file = event_file_data(filp);
1323 print_event_filter(file, s);
1324 mutex_unlock(&event_mutex);
1327 r = simple_read_from_buffer(ubuf, cnt, ppos,
1328 s->buffer, trace_seq_used(s));
1336 event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1339 struct trace_event_file *file;
1343 if (cnt >= PAGE_SIZE)
1346 buf = (char *)__get_free_page(GFP_TEMPORARY);
1350 if (copy_from_user(buf, ubuf, cnt)) {
1351 free_page((unsigned long) buf);
1356 mutex_lock(&event_mutex);
1357 file = event_file_data(filp);
1359 err = apply_event_filter(file, buf);
1360 mutex_unlock(&event_mutex);
1362 free_page((unsigned long) buf);
1371 static LIST_HEAD(event_subsystems);
1373 static int subsystem_open(struct inode *inode, struct file *filp)
1375 struct event_subsystem *system = NULL;
1376 struct trace_subsystem_dir *dir = NULL; /* Initialize for gcc */
1377 struct trace_array *tr;
1380 if (tracing_is_disabled())
1383 /* Make sure the system still exists */
1384 mutex_lock(&trace_types_lock);
1385 mutex_lock(&event_mutex);
1386 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
1387 list_for_each_entry(dir, &tr->systems, list) {
1388 if (dir == inode->i_private) {
1389 /* Don't open systems with no events */
1390 if (dir->nr_events) {
1391 __get_system_dir(dir);
1392 system = dir->subsystem;
1399 mutex_unlock(&event_mutex);
1400 mutex_unlock(&trace_types_lock);
1405 /* Some versions of gcc think dir can be uninitialized here */
1408 /* Still need to increment the ref count of the system */
1409 if (trace_array_get(tr) < 0) {
1414 ret = tracing_open_generic(inode, filp);
1416 trace_array_put(tr);
1423 static int system_tr_open(struct inode *inode, struct file *filp)
1425 struct trace_subsystem_dir *dir;
1426 struct trace_array *tr = inode->i_private;
1429 if (tracing_is_disabled())
1432 if (trace_array_get(tr) < 0)
1435 /* Make a temporary dir that has no system but points to tr */
1436 dir = kzalloc(sizeof(*dir), GFP_KERNEL);
1438 trace_array_put(tr);
1444 ret = tracing_open_generic(inode, filp);
1446 trace_array_put(tr);
1451 filp->private_data = dir;
1456 static int subsystem_release(struct inode *inode, struct file *file)
1458 struct trace_subsystem_dir *dir = file->private_data;
1460 trace_array_put(dir->tr);
1463 * If dir->subsystem is NULL, then this is a temporary
1464 * descriptor that was made for a trace_array to enable
1476 subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1479 struct trace_subsystem_dir *dir = filp->private_data;
1480 struct event_subsystem *system = dir->subsystem;
1481 struct trace_seq *s;
1487 s = kmalloc(sizeof(*s), GFP_KERNEL);
1493 print_subsystem_event_filter(system, s);
1494 r = simple_read_from_buffer(ubuf, cnt, ppos,
1495 s->buffer, trace_seq_used(s));
1503 subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1506 struct trace_subsystem_dir *dir = filp->private_data;
1510 if (cnt >= PAGE_SIZE)
1513 buf = (char *)__get_free_page(GFP_TEMPORARY);
1517 if (copy_from_user(buf, ubuf, cnt)) {
1518 free_page((unsigned long) buf);
1523 err = apply_subsystem_event_filter(dir, buf);
1524 free_page((unsigned long) buf);
1534 show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1536 int (*func)(struct trace_seq *s) = filp->private_data;
1537 struct trace_seq *s;
1543 s = kmalloc(sizeof(*s), GFP_KERNEL);
1550 r = simple_read_from_buffer(ubuf, cnt, ppos,
1551 s->buffer, trace_seq_used(s));
1558 static int max_pids(struct trace_pid_list *pid_list)
1560 return (PAGE_SIZE << pid_list->order) / sizeof(pid_t);
1563 static void ignore_task_cpu(void *data)
1565 struct trace_array *tr = data;
1566 struct trace_pid_list *pid_list;
1569 * This function is called by on_each_cpu() while the
1570 * event_mutex is held.
1572 pid_list = rcu_dereference_protected(tr->filtered_pids,
1573 mutex_is_locked(&event_mutex));
1575 this_cpu_write(tr->trace_buffer.data->ignore_pid,
1576 check_ignore_pid(pid_list, current));
1580 ftrace_event_pid_write(struct file *filp, const char __user *ubuf,
1581 size_t cnt, loff_t *ppos)
1583 struct seq_file *m = filp->private_data;
1584 struct trace_array *tr = m->private;
1585 struct trace_pid_list *filtered_pids = NULL;
1586 struct trace_pid_list *pid_list = NULL;
1587 struct trace_event_file *file;
1588 struct trace_parser parser;
1599 ret = tracing_update_buffers();
1603 if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
1606 mutex_lock(&event_mutex);
1608 * Load as many pids into the array before doing a
1609 * swap from the tr->filtered_pids to the new list.
1615 ret = trace_get_user(&parser, ubuf, cnt, &this_pos);
1616 if (ret < 0 || !trace_parser_loaded(&parser))
1623 parser.buffer[parser.idx] = 0;
1626 if (kstrtoul(parser.buffer, 0, &val))
1635 pid_list = kmalloc(sizeof(*pid_list), GFP_KERNEL);
1639 filtered_pids = rcu_dereference_protected(tr->filtered_pids,
1640 lockdep_is_held(&event_mutex));
1642 pid_list->order = filtered_pids->order;
1644 pid_list->order = 0;
1646 pid_list->pids = (void *)__get_free_pages(GFP_KERNEL,
1648 if (!pid_list->pids)
1651 if (filtered_pids) {
1652 pid_list->nr_pids = filtered_pids->nr_pids;
1653 memcpy(pid_list->pids, filtered_pids->pids,
1654 pid_list->nr_pids * sizeof(pid_t));
1656 pid_list->nr_pids = 0;
1659 if (pid_list->nr_pids >= max_pids(pid_list)) {
1662 pid_page = (void *)__get_free_pages(GFP_KERNEL,
1663 pid_list->order + 1);
1666 memcpy(pid_page, pid_list->pids,
1667 pid_list->nr_pids * sizeof(pid_t));
1668 free_pages((unsigned long)pid_list->pids, pid_list->order);
1671 pid_list->pids = pid_page;
1674 pid_list->pids[pid_list->nr_pids++] = pid;
1675 trace_parser_clear(&parser);
1678 trace_parser_put(&parser);
1682 free_pages((unsigned long)pid_list->pids, pid_list->order);
1684 mutex_unlock(&event_mutex);
1689 mutex_unlock(&event_mutex);
1693 sort(pid_list->pids, pid_list->nr_pids, sizeof(pid_t), cmp_pid, NULL);
1695 /* Remove duplicates */
1696 for (i = 1; i < pid_list->nr_pids; i++) {
1699 while (i < pid_list->nr_pids &&
1700 pid_list->pids[i - 1] == pid_list->pids[i])
1704 if (i < pid_list->nr_pids) {
1705 memmove(&pid_list->pids[start], &pid_list->pids[i],
1706 (pid_list->nr_pids - i) * sizeof(pid_t));
1707 pid_list->nr_pids -= i - start;
1710 pid_list->nr_pids = start;
1714 rcu_assign_pointer(tr->filtered_pids, pid_list);
1716 list_for_each_entry(file, &tr->events, list) {
1717 set_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
1720 if (filtered_pids) {
1721 synchronize_sched();
1723 free_pages((unsigned long)filtered_pids->pids, filtered_pids->order);
1724 kfree(filtered_pids);
1727 * Register a probe that is called before all other probes
1728 * to set ignore_pid if next or prev do not match.
1729 * Register a probe this is called after all other probes
1730 * to only keep ignore_pid set if next pid matches.
1732 register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_pre,
1734 register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_post,
1737 register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre,
1739 register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_post,
1742 register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre,
1744 register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post,
1747 register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_pre,
1749 register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_post,
1754 * Ignoring of pids is done at task switch. But we have to
1755 * check for those tasks that are currently running.
1756 * Always do this in case a pid was appended or removed.
1758 on_each_cpu(ignore_task_cpu, tr, 1);
1760 mutex_unlock(&event_mutex);
1768 static int ftrace_event_avail_open(struct inode *inode, struct file *file);
1769 static int ftrace_event_set_open(struct inode *inode, struct file *file);
1770 static int ftrace_event_set_pid_open(struct inode *inode, struct file *file);
1771 static int ftrace_event_release(struct inode *inode, struct file *file);
1773 static const struct seq_operations show_event_seq_ops = {
1780 static const struct seq_operations show_set_event_seq_ops = {
1787 static const struct seq_operations show_set_pid_seq_ops = {
1794 static const struct file_operations ftrace_avail_fops = {
1795 .open = ftrace_event_avail_open,
1797 .llseek = seq_lseek,
1798 .release = seq_release,
1801 static const struct file_operations ftrace_set_event_fops = {
1802 .open = ftrace_event_set_open,
1804 .write = ftrace_event_write,
1805 .llseek = seq_lseek,
1806 .release = ftrace_event_release,
1809 static const struct file_operations ftrace_set_event_pid_fops = {
1810 .open = ftrace_event_set_pid_open,
1812 .write = ftrace_event_pid_write,
1813 .llseek = seq_lseek,
1814 .release = ftrace_event_release,
1817 static const struct file_operations ftrace_enable_fops = {
1818 .open = tracing_open_generic,
1819 .read = event_enable_read,
1820 .write = event_enable_write,
1821 .llseek = default_llseek,
1824 static const struct file_operations ftrace_event_format_fops = {
1825 .open = trace_format_open,
1827 .llseek = seq_lseek,
1828 .release = seq_release,
1831 static const struct file_operations ftrace_event_id_fops = {
1832 .read = event_id_read,
1833 .llseek = default_llseek,
1836 static const struct file_operations ftrace_event_filter_fops = {
1837 .open = tracing_open_generic,
1838 .read = event_filter_read,
1839 .write = event_filter_write,
1840 .llseek = default_llseek,
1843 static const struct file_operations ftrace_subsystem_filter_fops = {
1844 .open = subsystem_open,
1845 .read = subsystem_filter_read,
1846 .write = subsystem_filter_write,
1847 .llseek = default_llseek,
1848 .release = subsystem_release,
1851 static const struct file_operations ftrace_system_enable_fops = {
1852 .open = subsystem_open,
1853 .read = system_enable_read,
1854 .write = system_enable_write,
1855 .llseek = default_llseek,
1856 .release = subsystem_release,
1859 static const struct file_operations ftrace_tr_enable_fops = {
1860 .open = system_tr_open,
1861 .read = system_enable_read,
1862 .write = system_enable_write,
1863 .llseek = default_llseek,
1864 .release = subsystem_release,
1867 static const struct file_operations ftrace_show_header_fops = {
1868 .open = tracing_open_generic,
1869 .read = show_header,
1870 .llseek = default_llseek,
1874 ftrace_event_open(struct inode *inode, struct file *file,
1875 const struct seq_operations *seq_ops)
1880 ret = seq_open(file, seq_ops);
1883 m = file->private_data;
1884 /* copy tr over to seq ops */
1885 m->private = inode->i_private;
1890 static int ftrace_event_release(struct inode *inode, struct file *file)
1892 struct trace_array *tr = inode->i_private;
1894 trace_array_put(tr);
1896 return seq_release(inode, file);
1900 ftrace_event_avail_open(struct inode *inode, struct file *file)
1902 const struct seq_operations *seq_ops = &show_event_seq_ops;
1904 return ftrace_event_open(inode, file, seq_ops);
1908 ftrace_event_set_open(struct inode *inode, struct file *file)
1910 const struct seq_operations *seq_ops = &show_set_event_seq_ops;
1911 struct trace_array *tr = inode->i_private;
1914 if (trace_array_get(tr) < 0)
1917 if ((file->f_mode & FMODE_WRITE) &&
1918 (file->f_flags & O_TRUNC))
1919 ftrace_clear_events(tr);
1921 ret = ftrace_event_open(inode, file, seq_ops);
1923 trace_array_put(tr);
1928 ftrace_event_set_pid_open(struct inode *inode, struct file *file)
1930 const struct seq_operations *seq_ops = &show_set_pid_seq_ops;
1931 struct trace_array *tr = inode->i_private;
1934 if (trace_array_get(tr) < 0)
1937 if ((file->f_mode & FMODE_WRITE) &&
1938 (file->f_flags & O_TRUNC))
1939 ftrace_clear_event_pids(tr);
1941 ret = ftrace_event_open(inode, file, seq_ops);
1943 trace_array_put(tr);
1947 static struct event_subsystem *
1948 create_new_subsystem(const char *name)
1950 struct event_subsystem *system;
1952 /* need to create new entry */
1953 system = kmalloc(sizeof(*system), GFP_KERNEL);
1957 system->ref_count = 1;
1959 /* Only allocate if dynamic (kprobes and modules) */
1960 system->name = kstrdup_const(name, GFP_KERNEL);
1964 system->filter = NULL;
1966 system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
1967 if (!system->filter)
1970 list_add(&system->list, &event_subsystems);
1975 kfree_const(system->name);
1980 static struct dentry *
1981 event_subsystem_dir(struct trace_array *tr, const char *name,
1982 struct trace_event_file *file, struct dentry *parent)
1984 struct trace_subsystem_dir *dir;
1985 struct event_subsystem *system;
1986 struct dentry *entry;
1988 /* First see if we did not already create this dir */
1989 list_for_each_entry(dir, &tr->systems, list) {
1990 system = dir->subsystem;
1991 if (strcmp(system->name, name) == 0) {
1998 /* Now see if the system itself exists. */
1999 list_for_each_entry(system, &event_subsystems, list) {
2000 if (strcmp(system->name, name) == 0)
2003 /* Reset system variable when not found */
2004 if (&system->list == &event_subsystems)
2007 dir = kmalloc(sizeof(*dir), GFP_KERNEL);
2012 system = create_new_subsystem(name);
2016 __get_system(system);
2018 dir->entry = tracefs_create_dir(name, parent);
2020 pr_warn("Failed to create system directory %s\n", name);
2021 __put_system(system);
2028 dir->subsystem = system;
2031 entry = tracefs_create_file("filter", 0644, dir->entry, dir,
2032 &ftrace_subsystem_filter_fops);
2034 kfree(system->filter);
2035 system->filter = NULL;
2036 pr_warn("Could not create tracefs '%s/filter' entry\n", name);
2039 trace_create_file("enable", 0644, dir->entry, dir,
2040 &ftrace_system_enable_fops);
2042 list_add(&dir->list, &tr->systems);
2049 /* Only print this message if failed on memory allocation */
2050 if (!dir || !system)
2051 pr_warn("No memory to create event subsystem %s\n", name);
2056 event_create_dir(struct dentry *parent, struct trace_event_file *file)
2058 struct trace_event_call *call = file->event_call;
2059 struct trace_array *tr = file->tr;
2060 struct list_head *head;
2061 struct dentry *d_events;
2066 * If the trace point header did not define TRACE_SYSTEM
2067 * then the system would be called "TRACE_SYSTEM".
2069 if (strcmp(call->class->system, TRACE_SYSTEM) != 0) {
2070 d_events = event_subsystem_dir(tr, call->class->system, file, parent);
2076 name = trace_event_name(call);
2077 file->dir = tracefs_create_dir(name, d_events);
2079 pr_warn("Could not create tracefs '%s' directory\n", name);
2083 if (call->class->reg && !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
2084 trace_create_file("enable", 0644, file->dir, file,
2085 &ftrace_enable_fops);
2087 #ifdef CONFIG_PERF_EVENTS
2088 if (call->event.type && call->class->reg)
2089 trace_create_file("id", 0444, file->dir,
2090 (void *)(long)call->event.type,
2091 &ftrace_event_id_fops);
2095 * Other events may have the same class. Only update
2096 * the fields if they are not already defined.
2098 head = trace_get_fields(call);
2099 if (list_empty(head)) {
2100 ret = call->class->define_fields(call);
2102 pr_warn("Could not initialize trace point events/%s\n",
2107 trace_create_file("filter", 0644, file->dir, file,
2108 &ftrace_event_filter_fops);
2111 * Only event directories that can be enabled should have
2114 if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
2115 trace_create_file("trigger", 0644, file->dir, file,
2116 &event_trigger_fops);
2118 trace_create_file("format", 0444, file->dir, call,
2119 &ftrace_event_format_fops);
2124 static void remove_event_from_tracers(struct trace_event_call *call)
2126 struct trace_event_file *file;
2127 struct trace_array *tr;
2129 do_for_each_event_file_safe(tr, file) {
2130 if (file->event_call != call)
2133 remove_event_file_dir(file);
2135 * The do_for_each_event_file_safe() is
2136 * a double loop. After finding the call for this
2137 * trace_array, we use break to jump to the next
2141 } while_for_each_event_file();
2144 static void event_remove(struct trace_event_call *call)
2146 struct trace_array *tr;
2147 struct trace_event_file *file;
2149 do_for_each_event_file(tr, file) {
2150 if (file->event_call != call)
2152 ftrace_event_enable_disable(file, 0);
2154 * The do_for_each_event_file() is
2155 * a double loop. After finding the call for this
2156 * trace_array, we use break to jump to the next
2160 } while_for_each_event_file();
2162 if (call->event.funcs)
2163 __unregister_trace_event(&call->event);
2164 remove_event_from_tracers(call);
2165 list_del(&call->list);
2168 static int event_init(struct trace_event_call *call)
2173 name = trace_event_name(call);
2177 if (call->class->raw_init) {
2178 ret = call->class->raw_init(call);
2179 if (ret < 0 && ret != -ENOSYS)
2180 pr_warn("Could not initialize trace events/%s\n", name);
2187 __register_event(struct trace_event_call *call, struct module *mod)
2191 ret = event_init(call);
2195 list_add(&call->list, &ftrace_events);
2201 static char *enum_replace(char *ptr, struct trace_enum_map *map, int len)
2206 /* Find the length of the enum value as a string */
2207 elen = snprintf(ptr, 0, "%ld", map->enum_value);
2208 /* Make sure there's enough room to replace the string with the value */
2212 snprintf(ptr, elen + 1, "%ld", map->enum_value);
2214 /* Get the rest of the string of ptr */
2215 rlen = strlen(ptr + len);
2216 memmove(ptr + elen, ptr + len, rlen);
2217 /* Make sure we end the new string */
2218 ptr[elen + rlen] = 0;
2223 static void update_event_printk(struct trace_event_call *call,
2224 struct trace_enum_map *map)
2228 int len = strlen(map->enum_string);
2230 for (ptr = call->print_fmt; *ptr; ptr++) {
2244 if (isdigit(*ptr)) {
2248 /* Check for alpha chars like ULL */
2249 } while (isalnum(*ptr));
2253 * A number must have some kind of delimiter after
2254 * it, and we can ignore that too.
2258 if (isalpha(*ptr) || *ptr == '_') {
2259 if (strncmp(map->enum_string, ptr, len) == 0 &&
2260 !isalnum(ptr[len]) && ptr[len] != '_') {
2261 ptr = enum_replace(ptr, map, len);
2262 /* Hmm, enum string smaller than value */
2263 if (WARN_ON_ONCE(!ptr))
2266 * No need to decrement here, as enum_replace()
2267 * returns the pointer to the character passed
2268 * the enum, and two enums can not be placed
2269 * back to back without something in between.
2270 * We can skip that something in between.
2277 } while (isalnum(*ptr) || *ptr == '_');
2281 * If what comes after this variable is a '.' or
2282 * '->' then we can continue to ignore that string.
2284 if (*ptr == '.' || (ptr[0] == '-' && ptr[1] == '>')) {
2285 ptr += *ptr == '.' ? 1 : 2;
2291 * Once again, we can skip the delimiter that came
2299 void trace_event_enum_update(struct trace_enum_map **map, int len)
2301 struct trace_event_call *call, *p;
2302 const char *last_system = NULL;
2307 down_write(&trace_event_sem);
2308 list_for_each_entry_safe(call, p, &ftrace_events, list) {
2309 /* events are usually grouped together with systems */
2310 if (!last_system || call->class->system != last_system) {
2313 last_system = call->class->system;
2317 * Since calls are grouped by systems, the likelyhood that the
2318 * next call in the iteration belongs to the same system as the
2319 * previous call is high. As an optimization, we skip seaching
2320 * for a map[] that matches the call's system if the last call
2321 * was from the same system. That's what last_i is for. If the
2322 * call has the same system as the previous call, then last_i
2323 * will be the index of the first map[] that has a matching
2326 for (i = last_i; i < len; i++) {
2327 if (call->class->system == map[i]->system) {
2328 /* Save the first system if need be */
2333 update_event_printk(call, map[i]);
2337 up_write(&trace_event_sem);
2340 static struct trace_event_file *
2341 trace_create_new_event(struct trace_event_call *call,
2342 struct trace_array *tr)
2344 struct trace_pid_list *pid_list;
2345 struct trace_event_file *file;
2347 file = kmem_cache_alloc(file_cachep, GFP_TRACE);
2351 pid_list = rcu_dereference_protected(tr->filtered_pids,
2352 lockdep_is_held(&event_mutex));
2355 file->flags |= EVENT_FILE_FL_PID_FILTER;
2357 file->event_call = call;
2359 atomic_set(&file->sm_ref, 0);
2360 atomic_set(&file->tm_ref, 0);
2361 INIT_LIST_HEAD(&file->triggers);
2362 list_add(&file->list, &tr->events);
2367 /* Add an event to a trace directory */
2369 __trace_add_new_event(struct trace_event_call *call, struct trace_array *tr)
2371 struct trace_event_file *file;
2373 file = trace_create_new_event(call, tr);
2377 return event_create_dir(tr->event_dir, file);
2381 * Just create a decriptor for early init. A descriptor is required
2382 * for enabling events at boot. We want to enable events before
2383 * the filesystem is initialized.
2386 __trace_early_add_new_event(struct trace_event_call *call,
2387 struct trace_array *tr)
2389 struct trace_event_file *file;
2391 file = trace_create_new_event(call, tr);
2398 struct ftrace_module_file_ops;
2399 static void __add_event_to_tracers(struct trace_event_call *call);
2401 /* Add an additional event_call dynamically */
2402 int trace_add_event_call(struct trace_event_call *call)
2405 mutex_lock(&trace_types_lock);
2406 mutex_lock(&event_mutex);
2408 ret = __register_event(call, NULL);
2410 __add_event_to_tracers(call);
2412 mutex_unlock(&event_mutex);
2413 mutex_unlock(&trace_types_lock);
2418 * Must be called under locking of trace_types_lock, event_mutex and
2421 static void __trace_remove_event_call(struct trace_event_call *call)
2424 trace_destroy_fields(call);
2425 free_event_filter(call->filter);
2426 call->filter = NULL;
2429 static int probe_remove_event_call(struct trace_event_call *call)
2431 struct trace_array *tr;
2432 struct trace_event_file *file;
2434 #ifdef CONFIG_PERF_EVENTS
2435 if (call->perf_refcount)
2438 do_for_each_event_file(tr, file) {
2439 if (file->event_call != call)
2442 * We can't rely on ftrace_event_enable_disable(enable => 0)
2443 * we are going to do, EVENT_FILE_FL_SOFT_MODE can suppress
2444 * TRACE_REG_UNREGISTER.
2446 if (file->flags & EVENT_FILE_FL_ENABLED)
2449 * The do_for_each_event_file_safe() is
2450 * a double loop. After finding the call for this
2451 * trace_array, we use break to jump to the next
2455 } while_for_each_event_file();
2457 __trace_remove_event_call(call);
2462 /* Remove an event_call */
2463 int trace_remove_event_call(struct trace_event_call *call)
2467 mutex_lock(&trace_types_lock);
2468 mutex_lock(&event_mutex);
2469 down_write(&trace_event_sem);
2470 ret = probe_remove_event_call(call);
2471 up_write(&trace_event_sem);
2472 mutex_unlock(&event_mutex);
2473 mutex_unlock(&trace_types_lock);
2478 #define for_each_event(event, start, end) \
2479 for (event = start; \
2480 (unsigned long)event < (unsigned long)end; \
2483 #ifdef CONFIG_MODULES
2485 static void trace_module_add_events(struct module *mod)
2487 struct trace_event_call **call, **start, **end;
2489 if (!mod->num_trace_events)
2492 /* Don't add infrastructure for mods without tracepoints */
2493 if (trace_module_has_bad_taint(mod)) {
2494 pr_err("%s: module has bad taint, not creating trace events\n",
2499 start = mod->trace_events;
2500 end = mod->trace_events + mod->num_trace_events;
2502 for_each_event(call, start, end) {
2503 __register_event(*call, mod);
2504 __add_event_to_tracers(*call);
2508 static void trace_module_remove_events(struct module *mod)
2510 struct trace_event_call *call, *p;
2511 bool clear_trace = false;
2513 down_write(&trace_event_sem);
2514 list_for_each_entry_safe(call, p, &ftrace_events, list) {
2515 if (call->mod == mod) {
2516 if (call->flags & TRACE_EVENT_FL_WAS_ENABLED)
2518 __trace_remove_event_call(call);
2521 up_write(&trace_event_sem);
2524 * It is safest to reset the ring buffer if the module being unloaded
2525 * registered any events that were used. The only worry is if
2526 * a new module gets loaded, and takes on the same id as the events
2527 * of this module. When printing out the buffer, traced events left
2528 * over from this module may be passed to the new module events and
2529 * unexpected results may occur.
2532 tracing_reset_all_online_cpus();
2535 static int trace_module_notify(struct notifier_block *self,
2536 unsigned long val, void *data)
2538 struct module *mod = data;
2540 mutex_lock(&trace_types_lock);
2541 mutex_lock(&event_mutex);
2543 case MODULE_STATE_COMING:
2544 trace_module_add_events(mod);
2546 case MODULE_STATE_GOING:
2547 trace_module_remove_events(mod);
2550 mutex_unlock(&event_mutex);
2551 mutex_unlock(&trace_types_lock);
2556 static struct notifier_block trace_module_nb = {
2557 .notifier_call = trace_module_notify,
2558 .priority = 1, /* higher than trace.c module notify */
2560 #endif /* CONFIG_MODULES */
2562 /* Create a new event directory structure for a trace directory. */
2564 __trace_add_event_dirs(struct trace_array *tr)
2566 struct trace_event_call *call;
2569 list_for_each_entry(call, &ftrace_events, list) {
2570 ret = __trace_add_new_event(call, tr);
2572 pr_warn("Could not create directory for event %s\n",
2573 trace_event_name(call));
2577 struct trace_event_file *
2578 find_event_file(struct trace_array *tr, const char *system, const char *event)
2580 struct trace_event_file *file;
2581 struct trace_event_call *call;
2584 list_for_each_entry(file, &tr->events, list) {
2586 call = file->event_call;
2587 name = trace_event_name(call);
2589 if (!name || !call->class || !call->class->reg)
2592 if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
2595 if (strcmp(event, name) == 0 &&
2596 strcmp(system, call->class->system) == 0)
2602 #ifdef CONFIG_DYNAMIC_FTRACE
2605 #define ENABLE_EVENT_STR "enable_event"
2606 #define DISABLE_EVENT_STR "disable_event"
2608 struct event_probe_data {
2609 struct trace_event_file *file;
2610 unsigned long count;
2616 event_enable_probe(unsigned long ip, unsigned long parent_ip, void **_data)
2618 struct event_probe_data **pdata = (struct event_probe_data **)_data;
2619 struct event_probe_data *data = *pdata;
2625 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
2627 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
2631 event_enable_count_probe(unsigned long ip, unsigned long parent_ip, void **_data)
2633 struct event_probe_data **pdata = (struct event_probe_data **)_data;
2634 struct event_probe_data *data = *pdata;
2642 /* Skip if the event is in a state we want to switch to */
2643 if (data->enable == !(data->file->flags & EVENT_FILE_FL_SOFT_DISABLED))
2646 if (data->count != -1)
2649 event_enable_probe(ip, parent_ip, _data);
2653 event_enable_print(struct seq_file *m, unsigned long ip,
2654 struct ftrace_probe_ops *ops, void *_data)
2656 struct event_probe_data *data = _data;
2658 seq_printf(m, "%ps:", (void *)ip);
2660 seq_printf(m, "%s:%s:%s",
2661 data->enable ? ENABLE_EVENT_STR : DISABLE_EVENT_STR,
2662 data->file->event_call->class->system,
2663 trace_event_name(data->file->event_call));
2665 if (data->count == -1)
2666 seq_puts(m, ":unlimited\n");
2668 seq_printf(m, ":count=%ld\n", data->count);
2674 event_enable_init(struct ftrace_probe_ops *ops, unsigned long ip,
2677 struct event_probe_data **pdata = (struct event_probe_data **)_data;
2678 struct event_probe_data *data = *pdata;
2685 event_enable_free(struct ftrace_probe_ops *ops, unsigned long ip,
2688 struct event_probe_data **pdata = (struct event_probe_data **)_data;
2689 struct event_probe_data *data = *pdata;
2691 if (WARN_ON_ONCE(data->ref <= 0))
2696 /* Remove the SOFT_MODE flag */
2697 __ftrace_event_enable_disable(data->file, 0, 1);
2698 module_put(data->file->event_call->mod);
2704 static struct ftrace_probe_ops event_enable_probe_ops = {
2705 .func = event_enable_probe,
2706 .print = event_enable_print,
2707 .init = event_enable_init,
2708 .free = event_enable_free,
2711 static struct ftrace_probe_ops event_enable_count_probe_ops = {
2712 .func = event_enable_count_probe,
2713 .print = event_enable_print,
2714 .init = event_enable_init,
2715 .free = event_enable_free,
2718 static struct ftrace_probe_ops event_disable_probe_ops = {
2719 .func = event_enable_probe,
2720 .print = event_enable_print,
2721 .init = event_enable_init,
2722 .free = event_enable_free,
2725 static struct ftrace_probe_ops event_disable_count_probe_ops = {
2726 .func = event_enable_count_probe,
2727 .print = event_enable_print,
2728 .init = event_enable_init,
2729 .free = event_enable_free,
2733 event_enable_func(struct ftrace_hash *hash,
2734 char *glob, char *cmd, char *param, int enabled)
2736 struct trace_array *tr = top_trace_array();
2737 struct trace_event_file *file;
2738 struct ftrace_probe_ops *ops;
2739 struct event_probe_data *data;
2749 /* hash funcs only work with set_ftrace_filter */
2750 if (!enabled || !param)
2753 system = strsep(¶m, ":");
2757 event = strsep(¶m, ":");
2759 mutex_lock(&event_mutex);
2762 file = find_event_file(tr, system, event);
2766 enable = strcmp(cmd, ENABLE_EVENT_STR) == 0;
2769 ops = param ? &event_enable_count_probe_ops : &event_enable_probe_ops;
2771 ops = param ? &event_disable_count_probe_ops : &event_disable_probe_ops;
2773 if (glob[0] == '!') {
2774 unregister_ftrace_function_probe_func(glob+1, ops);
2780 data = kzalloc(sizeof(*data), GFP_KERNEL);
2784 data->enable = enable;
2791 number = strsep(¶m, ":");
2794 if (!strlen(number))
2798 * We use the callback data field (which is a pointer)
2801 ret = kstrtoul(number, 0, &data->count);
2806 /* Don't let event modules unload while probe registered */
2807 ret = try_module_get(file->event_call->mod);
2813 ret = __ftrace_event_enable_disable(file, 1, 1);
2816 ret = register_ftrace_function_probe(glob, ops, data);
2818 * The above returns on success the # of functions enabled,
2819 * but if it didn't find any functions it returns zero.
2820 * Consider no functions a failure too.
2827 /* Just return zero, not the number of enabled functions */
2830 mutex_unlock(&event_mutex);
2834 __ftrace_event_enable_disable(file, 0, 1);
2836 module_put(file->event_call->mod);
2842 static struct ftrace_func_command event_enable_cmd = {
2843 .name = ENABLE_EVENT_STR,
2844 .func = event_enable_func,
2847 static struct ftrace_func_command event_disable_cmd = {
2848 .name = DISABLE_EVENT_STR,
2849 .func = event_enable_func,
2852 static __init int register_event_cmds(void)
2856 ret = register_ftrace_command(&event_enable_cmd);
2857 if (WARN_ON(ret < 0))
2859 ret = register_ftrace_command(&event_disable_cmd);
2860 if (WARN_ON(ret < 0))
2861 unregister_ftrace_command(&event_enable_cmd);
2865 static inline int register_event_cmds(void) { return 0; }
2866 #endif /* CONFIG_DYNAMIC_FTRACE */
2869 * The top level array has already had its trace_event_file
2870 * descriptors created in order to allow for early events to
2871 * be recorded. This function is called after the tracefs has been
2872 * initialized, and we now have to create the files associated
2876 __trace_early_add_event_dirs(struct trace_array *tr)
2878 struct trace_event_file *file;
2882 list_for_each_entry(file, &tr->events, list) {
2883 ret = event_create_dir(tr->event_dir, file);
2885 pr_warn("Could not create directory for event %s\n",
2886 trace_event_name(file->event_call));
2891 * For early boot up, the top trace array requires to have
2892 * a list of events that can be enabled. This must be done before
2893 * the filesystem is set up in order to allow events to be traced
2897 __trace_early_add_events(struct trace_array *tr)
2899 struct trace_event_call *call;
2902 list_for_each_entry(call, &ftrace_events, list) {
2903 /* Early boot up should not have any modules loaded */
2904 if (WARN_ON_ONCE(call->mod))
2907 ret = __trace_early_add_new_event(call, tr);
2909 pr_warn("Could not create early event %s\n",
2910 trace_event_name(call));
2914 /* Remove the event directory structure for a trace directory. */
2916 __trace_remove_event_dirs(struct trace_array *tr)
2918 struct trace_event_file *file, *next;
2920 list_for_each_entry_safe(file, next, &tr->events, list)
2921 remove_event_file_dir(file);
2924 static void __add_event_to_tracers(struct trace_event_call *call)
2926 struct trace_array *tr;
2928 list_for_each_entry(tr, &ftrace_trace_arrays, list)
2929 __trace_add_new_event(call, tr);
2932 extern struct trace_event_call *__start_ftrace_events[];
2933 extern struct trace_event_call *__stop_ftrace_events[];
2935 static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
2937 static __init int setup_trace_event(char *str)
2939 strlcpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
2940 ring_buffer_expanded = true;
2941 tracing_selftest_disabled = true;
2945 __setup("trace_event=", setup_trace_event);
2947 /* Expects to have event_mutex held when called */
2949 create_event_toplevel_files(struct dentry *parent, struct trace_array *tr)
2951 struct dentry *d_events;
2952 struct dentry *entry;
2954 entry = tracefs_create_file("set_event", 0644, parent,
2955 tr, &ftrace_set_event_fops);
2957 pr_warn("Could not create tracefs 'set_event' entry\n");
2961 d_events = tracefs_create_dir("events", parent);
2963 pr_warn("Could not create tracefs 'events' directory\n");
2967 entry = tracefs_create_file("set_event_pid", 0644, parent,
2968 tr, &ftrace_set_event_pid_fops);
2970 /* ring buffer internal formats */
2971 trace_create_file("header_page", 0444, d_events,
2972 ring_buffer_print_page_header,
2973 &ftrace_show_header_fops);
2975 trace_create_file("header_event", 0444, d_events,
2976 ring_buffer_print_entry_header,
2977 &ftrace_show_header_fops);
2979 trace_create_file("enable", 0644, d_events,
2980 tr, &ftrace_tr_enable_fops);
2982 tr->event_dir = d_events;
2988 * event_trace_add_tracer - add a instance of a trace_array to events
2989 * @parent: The parent dentry to place the files/directories for events in
2990 * @tr: The trace array associated with these events
2992 * When a new instance is created, it needs to set up its events
2993 * directory, as well as other files associated with events. It also
2994 * creates the event hierachry in the @parent/events directory.
2996 * Returns 0 on success.
2998 int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr)
3002 mutex_lock(&event_mutex);
3004 ret = create_event_toplevel_files(parent, tr);
3008 down_write(&trace_event_sem);
3009 __trace_add_event_dirs(tr);
3010 up_write(&trace_event_sem);
3013 mutex_unlock(&event_mutex);
3019 * The top trace array already had its file descriptors created.
3020 * Now the files themselves need to be created.
3023 early_event_add_tracer(struct dentry *parent, struct trace_array *tr)
3027 mutex_lock(&event_mutex);
3029 ret = create_event_toplevel_files(parent, tr);
3033 down_write(&trace_event_sem);
3034 __trace_early_add_event_dirs(tr);
3035 up_write(&trace_event_sem);
3038 mutex_unlock(&event_mutex);
3043 int event_trace_del_tracer(struct trace_array *tr)
3045 mutex_lock(&event_mutex);
3047 /* Disable any event triggers and associated soft-disabled events */
3048 clear_event_triggers(tr);
3050 /* Clear the pid list */
3051 __ftrace_clear_event_pids(tr);
3053 /* Disable any running events */
3054 __ftrace_set_clr_event_nolock(tr, NULL, NULL, NULL, 0);
3056 /* Access to events are within rcu_read_lock_sched() */
3057 synchronize_sched();
3059 down_write(&trace_event_sem);
3060 __trace_remove_event_dirs(tr);
3061 tracefs_remove_recursive(tr->event_dir);
3062 up_write(&trace_event_sem);
3064 tr->event_dir = NULL;
3066 mutex_unlock(&event_mutex);
3071 static __init int event_trace_memsetup(void)
3073 field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC);
3074 file_cachep = KMEM_CACHE(trace_event_file, SLAB_PANIC);
3079 early_enable_events(struct trace_array *tr, bool disable_first)
3081 char *buf = bootup_event_buf;
3086 token = strsep(&buf, ",");
3092 /* Restarting syscalls requires that we stop them first */
3094 ftrace_set_clr_event(tr, token, 0);
3096 ret = ftrace_set_clr_event(tr, token, 1);
3098 pr_warn("Failed to enable trace event: %s\n", token);
3101 /* Put back the comma to allow this to be called again */
3107 static __init int event_trace_enable(void)
3109 struct trace_array *tr = top_trace_array();
3110 struct trace_event_call **iter, *call;
3116 for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
3119 ret = event_init(call);
3121 list_add(&call->list, &ftrace_events);
3125 * We need the top trace array to have a working set of trace
3126 * points at early init, before the debug files and directories
3127 * are created. Create the file entries now, and attach them
3128 * to the actual file dentries later.
3130 __trace_early_add_events(tr);
3132 early_enable_events(tr, false);
3134 trace_printk_start_comm();
3136 register_event_cmds();
3138 register_trigger_cmds();
3144 * event_trace_enable() is called from trace_event_init() first to
3145 * initialize events and perhaps start any events that are on the
3146 * command line. Unfortunately, there are some events that will not
3147 * start this early, like the system call tracepoints that need
3148 * to set the TIF_SYSCALL_TRACEPOINT flag of pid 1. But event_trace_enable()
3149 * is called before pid 1 starts, and this flag is never set, making
3150 * the syscall tracepoint never get reached, but the event is enabled
3151 * regardless (and not doing anything).
3153 static __init int event_trace_enable_again(void)
3155 struct trace_array *tr;
3157 tr = top_trace_array();
3161 early_enable_events(tr, true);
3166 early_initcall(event_trace_enable_again);
3168 static __init int event_trace_init(void)
3170 struct trace_array *tr;
3171 struct dentry *d_tracer;
3172 struct dentry *entry;
3175 tr = top_trace_array();
3179 d_tracer = tracing_init_dentry();
3180 if (IS_ERR(d_tracer))
3183 entry = tracefs_create_file("available_events", 0444, d_tracer,
3184 tr, &ftrace_avail_fops);
3186 pr_warn("Could not create tracefs 'available_events' entry\n");
3188 if (trace_define_generic_fields())
3189 pr_warn("tracing: Failed to allocated generic fields");
3191 if (trace_define_common_fields())
3192 pr_warn("tracing: Failed to allocate common fields");
3194 ret = early_event_add_tracer(d_tracer, tr);
3198 #ifdef CONFIG_MODULES
3199 ret = register_module_notifier(&trace_module_nb);
3201 pr_warn("Failed to register trace events module notifier\n");
3206 void __init trace_event_init(void)
3208 event_trace_memsetup();
3209 init_ftrace_syscalls();
3210 event_trace_enable();
3213 fs_initcall(event_trace_init);
3215 #ifdef CONFIG_FTRACE_STARTUP_TEST
3217 static DEFINE_SPINLOCK(test_spinlock);
3218 static DEFINE_SPINLOCK(test_spinlock_irq);
3219 static DEFINE_MUTEX(test_mutex);
3221 static __init void test_work(struct work_struct *dummy)
3223 spin_lock(&test_spinlock);
3224 spin_lock_irq(&test_spinlock_irq);
3226 spin_unlock_irq(&test_spinlock_irq);
3227 spin_unlock(&test_spinlock);
3229 mutex_lock(&test_mutex);
3231 mutex_unlock(&test_mutex);
3234 static __init int event_test_thread(void *unused)
3238 test_malloc = kmalloc(1234, GFP_KERNEL);
3240 pr_info("failed to kmalloc\n");
3242 schedule_on_each_cpu(test_work);
3246 set_current_state(TASK_INTERRUPTIBLE);
3247 while (!kthread_should_stop()) {
3249 set_current_state(TASK_INTERRUPTIBLE);
3251 __set_current_state(TASK_RUNNING);
3257 * Do various things that may trigger events.
3259 static __init void event_test_stuff(void)
3261 struct task_struct *test_thread;
3263 test_thread = kthread_run(event_test_thread, NULL, "test-events");
3265 kthread_stop(test_thread);
3269 * For every trace event defined, we will test each trace point separately,
3270 * and then by groups, and finally all trace points.
3272 static __init void event_trace_self_tests(void)
3274 struct trace_subsystem_dir *dir;
3275 struct trace_event_file *file;
3276 struct trace_event_call *call;
3277 struct event_subsystem *system;
3278 struct trace_array *tr;
3281 tr = top_trace_array();
3285 pr_info("Running tests on trace events:\n");
3287 list_for_each_entry(file, &tr->events, list) {
3289 call = file->event_call;
3291 /* Only test those that have a probe */
3292 if (!call->class || !call->class->probe)
3296 * Testing syscall events here is pretty useless, but
3297 * we still do it if configured. But this is time consuming.
3298 * What we really need is a user thread to perform the
3299 * syscalls as we test.
3301 #ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
3302 if (call->class->system &&
3303 strcmp(call->class->system, "syscalls") == 0)
3307 pr_info("Testing event %s: ", trace_event_name(call));
3310 * If an event is already enabled, someone is using
3311 * it and the self test should not be on.
3313 if (file->flags & EVENT_FILE_FL_ENABLED) {
3314 pr_warn("Enabled event during self test!\n");
3319 ftrace_event_enable_disable(file, 1);
3321 ftrace_event_enable_disable(file, 0);
3326 /* Now test at the sub system level */
3328 pr_info("Running tests on trace event systems:\n");
3330 list_for_each_entry(dir, &tr->systems, list) {
3332 system = dir->subsystem;
3334 /* the ftrace system is special, skip it */
3335 if (strcmp(system->name, "ftrace") == 0)
3338 pr_info("Testing event system %s: ", system->name);
3340 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1);
3341 if (WARN_ON_ONCE(ret)) {
3342 pr_warn("error enabling system %s\n",
3349 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0);
3350 if (WARN_ON_ONCE(ret)) {
3351 pr_warn("error disabling system %s\n",
3359 /* Test with all events enabled */
3361 pr_info("Running tests on all trace events:\n");
3362 pr_info("Testing all events: ");
3364 ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1);
3365 if (WARN_ON_ONCE(ret)) {
3366 pr_warn("error enabling all events\n");
3373 ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
3374 if (WARN_ON_ONCE(ret)) {
3375 pr_warn("error disabling all events\n");
3382 #ifdef CONFIG_FUNCTION_TRACER
3384 static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
3386 static struct trace_array *event_tr;
3389 function_test_events_call(unsigned long ip, unsigned long parent_ip,
3390 struct ftrace_ops *op, struct pt_regs *pt_regs)
3392 struct ring_buffer_event *event;
3393 struct ring_buffer *buffer;
3394 struct ftrace_entry *entry;
3395 unsigned long flags;
3400 pc = preempt_count();
3401 preempt_disable_notrace();
3402 cpu = raw_smp_processor_id();
3403 disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
3408 local_save_flags(flags);
3410 event = trace_current_buffer_lock_reserve(&buffer,
3411 TRACE_FN, sizeof(*entry),
3415 entry = ring_buffer_event_data(event);
3417 entry->parent_ip = parent_ip;
3419 trace_buffer_unlock_commit(event_tr, buffer, event, flags, pc);
3422 atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
3423 preempt_enable_notrace();
3426 static struct ftrace_ops trace_ops __initdata =
3428 .func = function_test_events_call,
3429 .flags = FTRACE_OPS_FL_RECURSION_SAFE,
3432 static __init void event_trace_self_test_with_function(void)
3435 event_tr = top_trace_array();
3436 if (WARN_ON(!event_tr))
3438 ret = register_ftrace_function(&trace_ops);
3439 if (WARN_ON(ret < 0)) {
3440 pr_info("Failed to enable function tracer for event tests\n");
3443 pr_info("Running tests again, along with the function tracer\n");
3444 event_trace_self_tests();
3445 unregister_ftrace_function(&trace_ops);
3448 static __init void event_trace_self_test_with_function(void)
3453 static __init int event_trace_self_tests_init(void)
3455 if (!tracing_selftest_disabled) {
3456 event_trace_self_tests();
3457 event_trace_self_test_with_function();
3463 late_initcall(event_trace_self_tests_init);