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/module.h>
19 #include <linux/ctype.h>
20 #include <linux/sort.h>
21 #include <linux/slab.h>
22 #include <linux/delay.h>
24 #include <trace/events/sched.h>
26 #include <asm/setup.h>
28 #include "trace_output.h"
31 #define TRACE_SYSTEM "TRACE_SYSTEM"
33 DEFINE_MUTEX(event_mutex);
35 LIST_HEAD(ftrace_events);
36 static LIST_HEAD(ftrace_generic_fields);
37 static LIST_HEAD(ftrace_common_fields);
39 #define GFP_TRACE (GFP_KERNEL | __GFP_ZERO)
41 static struct kmem_cache *field_cachep;
42 static struct kmem_cache *file_cachep;
44 static inline int system_refcount(struct event_subsystem *system)
46 return system->ref_count;
49 static int system_refcount_inc(struct event_subsystem *system)
51 return system->ref_count++;
54 static int system_refcount_dec(struct event_subsystem *system)
56 return --system->ref_count;
59 /* Double loops, do not use break, only goto's work */
60 #define do_for_each_event_file(tr, file) \
61 list_for_each_entry(tr, &ftrace_trace_arrays, list) { \
62 list_for_each_entry(file, &tr->events, list)
64 #define do_for_each_event_file_safe(tr, file) \
65 list_for_each_entry(tr, &ftrace_trace_arrays, list) { \
66 struct trace_event_file *___n; \
67 list_for_each_entry_safe(file, ___n, &tr->events, list)
69 #define while_for_each_event_file() \
72 static struct list_head *
73 trace_get_fields(struct trace_event_call *event_call)
75 if (!event_call->class->get_fields)
76 return &event_call->class->fields;
77 return event_call->class->get_fields(event_call);
80 static struct ftrace_event_field *
81 __find_event_field(struct list_head *head, char *name)
83 struct ftrace_event_field *field;
85 list_for_each_entry(field, head, link) {
86 if (!strcmp(field->name, name))
93 struct ftrace_event_field *
94 trace_find_event_field(struct trace_event_call *call, char *name)
96 struct ftrace_event_field *field;
97 struct list_head *head;
99 head = trace_get_fields(call);
100 field = __find_event_field(head, name);
104 field = __find_event_field(&ftrace_generic_fields, name);
108 return __find_event_field(&ftrace_common_fields, name);
111 static int __trace_define_field(struct list_head *head, const char *type,
112 const char *name, int offset, int size,
113 int is_signed, int filter_type)
115 struct ftrace_event_field *field;
117 field = kmem_cache_alloc(field_cachep, GFP_TRACE);
124 if (filter_type == FILTER_OTHER)
125 field->filter_type = filter_assign_type(type);
127 field->filter_type = filter_type;
129 field->offset = offset;
131 field->is_signed = is_signed;
133 list_add(&field->link, head);
138 int trace_define_field(struct trace_event_call *call, const char *type,
139 const char *name, int offset, int size, int is_signed,
142 struct list_head *head;
144 if (WARN_ON(!call->class))
147 head = trace_get_fields(call);
148 return __trace_define_field(head, type, name, offset, size,
149 is_signed, filter_type);
151 EXPORT_SYMBOL_GPL(trace_define_field);
153 #define __generic_field(type, item, filter_type) \
154 ret = __trace_define_field(&ftrace_generic_fields, #type, \
155 #item, 0, 0, is_signed_type(type), \
160 #define __common_field(type, item) \
161 ret = __trace_define_field(&ftrace_common_fields, #type, \
163 offsetof(typeof(ent), item), \
165 is_signed_type(type), FILTER_OTHER); \
169 static int trace_define_generic_fields(void)
173 __generic_field(int, CPU, FILTER_CPU);
174 __generic_field(int, cpu, FILTER_CPU);
175 __generic_field(char *, COMM, FILTER_COMM);
176 __generic_field(char *, comm, FILTER_COMM);
181 static int trace_define_common_fields(void)
184 struct trace_entry ent;
186 __common_field(unsigned short, type);
187 __common_field(unsigned char, flags);
188 __common_field(unsigned char, preempt_count);
189 __common_field(int, pid);
194 static void trace_destroy_fields(struct trace_event_call *call)
196 struct ftrace_event_field *field, *next;
197 struct list_head *head;
199 head = trace_get_fields(call);
200 list_for_each_entry_safe(field, next, head, link) {
201 list_del(&field->link);
202 kmem_cache_free(field_cachep, field);
207 * run-time version of trace_event_get_offsets_<call>() that returns the last
208 * accessible offset of trace fields excluding __dynamic_array bytes
210 int trace_event_get_offsets(struct trace_event_call *call)
212 struct ftrace_event_field *tail;
213 struct list_head *head;
215 head = trace_get_fields(call);
217 * head->next points to the last field with the largest offset,
218 * since it was added last by trace_define_field()
220 tail = list_first_entry(head, struct ftrace_event_field, link);
221 return tail->offset + tail->size;
224 int trace_event_raw_init(struct trace_event_call *call)
228 id = register_trace_event(&call->event);
234 EXPORT_SYMBOL_GPL(trace_event_raw_init);
236 bool trace_event_ignore_this_pid(struct trace_event_file *trace_file)
238 struct trace_array *tr = trace_file->tr;
239 struct trace_array_cpu *data;
240 struct trace_pid_list *pid_list;
242 pid_list = rcu_dereference_sched(tr->filtered_pids);
246 data = this_cpu_ptr(tr->trace_buffer.data);
248 return data->ignore_pid;
250 EXPORT_SYMBOL_GPL(trace_event_ignore_this_pid);
252 void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer,
253 struct trace_event_file *trace_file,
256 struct trace_event_call *event_call = trace_file->event_call;
258 if ((trace_file->flags & EVENT_FILE_FL_PID_FILTER) &&
259 trace_event_ignore_this_pid(trace_file))
262 local_save_flags(fbuffer->flags);
263 fbuffer->pc = preempt_count();
265 * If CONFIG_PREEMPT is enabled, then the tracepoint itself disables
266 * preemption (adding one to the preempt_count). Since we are
267 * interested in the preempt_count at the time the tracepoint was
268 * hit, we need to subtract one to offset the increment.
270 if (IS_ENABLED(CONFIG_PREEMPT))
272 fbuffer->trace_file = trace_file;
275 trace_event_buffer_lock_reserve(&fbuffer->buffer, trace_file,
276 event_call->event.type, len,
277 fbuffer->flags, fbuffer->pc);
281 fbuffer->entry = ring_buffer_event_data(fbuffer->event);
282 return fbuffer->entry;
284 EXPORT_SYMBOL_GPL(trace_event_buffer_reserve);
286 int trace_event_reg(struct trace_event_call *call,
287 enum trace_reg type, void *data)
289 struct trace_event_file *file = data;
291 WARN_ON(!(call->flags & TRACE_EVENT_FL_TRACEPOINT));
293 case TRACE_REG_REGISTER:
294 return tracepoint_probe_register(call->tp,
297 case TRACE_REG_UNREGISTER:
298 tracepoint_probe_unregister(call->tp,
303 #ifdef CONFIG_PERF_EVENTS
304 case TRACE_REG_PERF_REGISTER:
305 return tracepoint_probe_register(call->tp,
306 call->class->perf_probe,
308 case TRACE_REG_PERF_UNREGISTER:
309 tracepoint_probe_unregister(call->tp,
310 call->class->perf_probe,
313 case TRACE_REG_PERF_OPEN:
314 case TRACE_REG_PERF_CLOSE:
315 case TRACE_REG_PERF_ADD:
316 case TRACE_REG_PERF_DEL:
322 EXPORT_SYMBOL_GPL(trace_event_reg);
324 void trace_event_enable_cmd_record(bool enable)
326 struct trace_event_file *file;
327 struct trace_array *tr;
329 mutex_lock(&event_mutex);
330 do_for_each_event_file(tr, file) {
332 if (!(file->flags & EVENT_FILE_FL_ENABLED))
336 tracing_start_cmdline_record();
337 set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
339 tracing_stop_cmdline_record();
340 clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
342 } while_for_each_event_file();
343 mutex_unlock(&event_mutex);
346 void trace_event_enable_tgid_record(bool enable)
348 struct trace_event_file *file;
349 struct trace_array *tr;
351 mutex_lock(&event_mutex);
352 do_for_each_event_file(tr, file) {
353 if (!(file->flags & EVENT_FILE_FL_ENABLED))
357 tracing_start_tgid_record();
358 set_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
360 tracing_stop_tgid_record();
361 clear_bit(EVENT_FILE_FL_RECORDED_TGID_BIT,
364 } while_for_each_event_file();
365 mutex_unlock(&event_mutex);
368 static int __ftrace_event_enable_disable(struct trace_event_file *file,
369 int enable, int soft_disable)
371 struct trace_event_call *call = file->event_call;
372 struct trace_array *tr = file->tr;
373 unsigned long file_flags = file->flags;
380 * When soft_disable is set and enable is cleared, the sm_ref
381 * reference counter is decremented. If it reaches 0, we want
382 * to clear the SOFT_DISABLED flag but leave the event in the
383 * state that it was. That is, if the event was enabled and
384 * SOFT_DISABLED isn't set, then do nothing. But if SOFT_DISABLED
385 * is set we do not want the event to be enabled before we
388 * When soft_disable is not set but the SOFT_MODE flag is,
389 * we do nothing. Do not disable the tracepoint, otherwise
390 * "soft enable"s (clearing the SOFT_DISABLED bit) wont work.
393 if (atomic_dec_return(&file->sm_ref) > 0)
395 disable = file->flags & EVENT_FILE_FL_SOFT_DISABLED;
396 clear_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
398 disable = !(file->flags & EVENT_FILE_FL_SOFT_MODE);
400 if (disable && (file->flags & EVENT_FILE_FL_ENABLED)) {
401 clear_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
402 if (file->flags & EVENT_FILE_FL_RECORDED_CMD) {
403 tracing_stop_cmdline_record();
404 clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
407 if (file->flags & EVENT_FILE_FL_RECORDED_TGID) {
408 tracing_stop_tgid_record();
409 clear_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
412 call->class->reg(call, TRACE_REG_UNREGISTER, file);
414 /* If in SOFT_MODE, just set the SOFT_DISABLE_BIT, else clear it */
415 if (file->flags & EVENT_FILE_FL_SOFT_MODE)
416 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
418 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
422 * When soft_disable is set and enable is set, we want to
423 * register the tracepoint for the event, but leave the event
424 * as is. That means, if the event was already enabled, we do
425 * nothing (but set SOFT_MODE). If the event is disabled, we
426 * set SOFT_DISABLED before enabling the event tracepoint, so
427 * it still seems to be disabled.
430 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
432 if (atomic_inc_return(&file->sm_ref) > 1)
434 set_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
437 if (!(file->flags & EVENT_FILE_FL_ENABLED)) {
438 bool cmd = false, tgid = false;
440 /* Keep the event disabled, when going to SOFT_MODE. */
442 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
444 if (tr->trace_flags & TRACE_ITER_RECORD_CMD) {
446 tracing_start_cmdline_record();
447 set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
450 if (tr->trace_flags & TRACE_ITER_RECORD_TGID) {
452 tracing_start_tgid_record();
453 set_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
456 ret = call->class->reg(call, TRACE_REG_REGISTER, file);
459 tracing_stop_cmdline_record();
461 tracing_stop_tgid_record();
462 pr_info("event trace: Could not enable event "
463 "%s\n", trace_event_name(call));
466 set_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
468 /* WAS_ENABLED gets set but never cleared. */
469 set_bit(EVENT_FILE_FL_WAS_ENABLED_BIT, &file->flags);
474 /* Enable or disable use of trace_buffered_event */
475 if ((file_flags & EVENT_FILE_FL_SOFT_DISABLED) !=
476 (file->flags & EVENT_FILE_FL_SOFT_DISABLED)) {
477 if (file->flags & EVENT_FILE_FL_SOFT_DISABLED)
478 trace_buffered_event_enable();
480 trace_buffered_event_disable();
486 int trace_event_enable_disable(struct trace_event_file *file,
487 int enable, int soft_disable)
489 return __ftrace_event_enable_disable(file, enable, soft_disable);
492 static int ftrace_event_enable_disable(struct trace_event_file *file,
495 return __ftrace_event_enable_disable(file, enable, 0);
498 static void ftrace_clear_events(struct trace_array *tr)
500 struct trace_event_file *file;
502 mutex_lock(&event_mutex);
503 list_for_each_entry(file, &tr->events, list) {
504 ftrace_event_enable_disable(file, 0);
506 mutex_unlock(&event_mutex);
510 event_filter_pid_sched_process_exit(void *data, struct task_struct *task)
512 struct trace_pid_list *pid_list;
513 struct trace_array *tr = data;
515 pid_list = rcu_dereference_sched(tr->filtered_pids);
516 trace_filter_add_remove_task(pid_list, NULL, task);
520 event_filter_pid_sched_process_fork(void *data,
521 struct task_struct *self,
522 struct task_struct *task)
524 struct trace_pid_list *pid_list;
525 struct trace_array *tr = data;
527 pid_list = rcu_dereference_sched(tr->filtered_pids);
528 trace_filter_add_remove_task(pid_list, self, task);
531 void trace_event_follow_fork(struct trace_array *tr, bool enable)
534 register_trace_prio_sched_process_fork(event_filter_pid_sched_process_fork,
536 register_trace_prio_sched_process_exit(event_filter_pid_sched_process_exit,
539 unregister_trace_sched_process_fork(event_filter_pid_sched_process_fork,
541 unregister_trace_sched_process_exit(event_filter_pid_sched_process_exit,
547 event_filter_pid_sched_switch_probe_pre(void *data, bool preempt,
548 struct task_struct *prev, struct task_struct *next)
550 struct trace_array *tr = data;
551 struct trace_pid_list *pid_list;
553 pid_list = rcu_dereference_sched(tr->filtered_pids);
555 this_cpu_write(tr->trace_buffer.data->ignore_pid,
556 trace_ignore_this_task(pid_list, prev) &&
557 trace_ignore_this_task(pid_list, next));
561 event_filter_pid_sched_switch_probe_post(void *data, bool preempt,
562 struct task_struct *prev, struct task_struct *next)
564 struct trace_array *tr = data;
565 struct trace_pid_list *pid_list;
567 pid_list = rcu_dereference_sched(tr->filtered_pids);
569 this_cpu_write(tr->trace_buffer.data->ignore_pid,
570 trace_ignore_this_task(pid_list, next));
574 event_filter_pid_sched_wakeup_probe_pre(void *data, struct task_struct *task)
576 struct trace_array *tr = data;
577 struct trace_pid_list *pid_list;
579 /* Nothing to do if we are already tracing */
580 if (!this_cpu_read(tr->trace_buffer.data->ignore_pid))
583 pid_list = rcu_dereference_sched(tr->filtered_pids);
585 this_cpu_write(tr->trace_buffer.data->ignore_pid,
586 trace_ignore_this_task(pid_list, task));
590 event_filter_pid_sched_wakeup_probe_post(void *data, struct task_struct *task)
592 struct trace_array *tr = data;
593 struct trace_pid_list *pid_list;
595 /* Nothing to do if we are not tracing */
596 if (this_cpu_read(tr->trace_buffer.data->ignore_pid))
599 pid_list = rcu_dereference_sched(tr->filtered_pids);
601 /* Set tracing if current is enabled */
602 this_cpu_write(tr->trace_buffer.data->ignore_pid,
603 trace_ignore_this_task(pid_list, current));
606 static void __ftrace_clear_event_pids(struct trace_array *tr)
608 struct trace_pid_list *pid_list;
609 struct trace_event_file *file;
612 pid_list = rcu_dereference_protected(tr->filtered_pids,
613 lockdep_is_held(&event_mutex));
617 unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_pre, tr);
618 unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_post, tr);
620 unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre, tr);
621 unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_post, tr);
623 unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre, tr);
624 unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post, tr);
626 unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_pre, tr);
627 unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_post, tr);
629 list_for_each_entry(file, &tr->events, list) {
630 clear_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
633 for_each_possible_cpu(cpu)
634 per_cpu_ptr(tr->trace_buffer.data, cpu)->ignore_pid = false;
636 rcu_assign_pointer(tr->filtered_pids, NULL);
638 /* Wait till all users are no longer using pid filtering */
641 trace_free_pid_list(pid_list);
644 static void ftrace_clear_event_pids(struct trace_array *tr)
646 mutex_lock(&event_mutex);
647 __ftrace_clear_event_pids(tr);
648 mutex_unlock(&event_mutex);
651 static void __put_system(struct event_subsystem *system)
653 struct event_filter *filter = system->filter;
655 WARN_ON_ONCE(system_refcount(system) == 0);
656 if (system_refcount_dec(system))
659 list_del(&system->list);
662 kfree(filter->filter_string);
665 kfree_const(system->name);
669 static void __get_system(struct event_subsystem *system)
671 WARN_ON_ONCE(system_refcount(system) == 0);
672 system_refcount_inc(system);
675 static void __get_system_dir(struct trace_subsystem_dir *dir)
677 WARN_ON_ONCE(dir->ref_count == 0);
679 __get_system(dir->subsystem);
682 static void __put_system_dir(struct trace_subsystem_dir *dir)
684 WARN_ON_ONCE(dir->ref_count == 0);
685 /* If the subsystem is about to be freed, the dir must be too */
686 WARN_ON_ONCE(system_refcount(dir->subsystem) == 1 && dir->ref_count != 1);
688 __put_system(dir->subsystem);
689 if (!--dir->ref_count)
693 static void put_system(struct trace_subsystem_dir *dir)
695 mutex_lock(&event_mutex);
696 __put_system_dir(dir);
697 mutex_unlock(&event_mutex);
700 static void remove_subsystem(struct trace_subsystem_dir *dir)
705 if (!--dir->nr_events) {
706 tracefs_remove_recursive(dir->entry);
707 list_del(&dir->list);
708 __put_system_dir(dir);
712 static void remove_event_file_dir(struct trace_event_file *file)
714 struct dentry *dir = file->dir;
715 struct dentry *child;
718 spin_lock(&dir->d_lock); /* probably unneeded */
719 list_for_each_entry(child, &dir->d_subdirs, d_child) {
720 if (d_really_is_positive(child)) /* probably unneeded */
721 d_inode(child)->i_private = NULL;
723 spin_unlock(&dir->d_lock);
725 tracefs_remove_recursive(dir);
728 list_del(&file->list);
729 remove_subsystem(file->system);
730 free_event_filter(file->filter);
731 kmem_cache_free(file_cachep, file);
735 * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
738 __ftrace_set_clr_event_nolock(struct trace_array *tr, const char *match,
739 const char *sub, const char *event, int set)
741 struct trace_event_file *file;
742 struct trace_event_call *call;
747 list_for_each_entry(file, &tr->events, list) {
749 call = file->event_call;
750 name = trace_event_name(call);
752 if (!name || !call->class || !call->class->reg)
755 if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
759 strcmp(match, name) != 0 &&
760 strcmp(match, call->class->system) != 0)
763 if (sub && strcmp(sub, call->class->system) != 0)
766 if (event && strcmp(event, name) != 0)
769 ret = ftrace_event_enable_disable(file, set);
772 * Save the first error and return that. Some events
773 * may still have been enabled, but let the user
774 * know that something went wrong.
785 static int __ftrace_set_clr_event(struct trace_array *tr, const char *match,
786 const char *sub, const char *event, int set)
790 mutex_lock(&event_mutex);
791 ret = __ftrace_set_clr_event_nolock(tr, match, sub, event, set);
792 mutex_unlock(&event_mutex);
797 static int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set)
799 char *event = NULL, *sub = NULL, *match;
803 * The buf format can be <subsystem>:<event-name>
804 * *:<event-name> means any event by that name.
805 * :<event-name> is the same.
807 * <subsystem>:* means all events in that subsystem
808 * <subsystem>: means the same.
810 * <name> (no ':') means all events in a subsystem with
811 * the name <name> or any event that matches <name>
814 match = strsep(&buf, ":");
820 if (!strlen(sub) || strcmp(sub, "*") == 0)
822 if (!strlen(event) || strcmp(event, "*") == 0)
826 ret = __ftrace_set_clr_event(tr, match, sub, event, set);
828 /* Put back the colon to allow this to be called again */
836 * trace_set_clr_event - enable or disable an event
837 * @system: system name to match (NULL for any system)
838 * @event: event name to match (NULL for all events, within system)
839 * @set: 1 to enable, 0 to disable
841 * This is a way for other parts of the kernel to enable or disable
844 * Returns 0 on success, -EINVAL if the parameters do not match any
847 int trace_set_clr_event(const char *system, const char *event, int set)
849 struct trace_array *tr = top_trace_array();
854 return __ftrace_set_clr_event(tr, NULL, system, event, set);
856 EXPORT_SYMBOL_GPL(trace_set_clr_event);
858 /* 128 should be much more than enough */
859 #define EVENT_BUF_SIZE 127
862 ftrace_event_write(struct file *file, const char __user *ubuf,
863 size_t cnt, loff_t *ppos)
865 struct trace_parser parser;
866 struct seq_file *m = file->private_data;
867 struct trace_array *tr = m->private;
873 ret = tracing_update_buffers();
877 if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
880 read = trace_get_user(&parser, ubuf, cnt, ppos);
882 if (read >= 0 && trace_parser_loaded((&parser))) {
885 if (*parser.buffer == '!')
888 ret = ftrace_set_clr_event(tr, parser.buffer + !set, set);
896 trace_parser_put(&parser);
902 t_next(struct seq_file *m, void *v, loff_t *pos)
904 struct trace_event_file *file = v;
905 struct trace_event_call *call;
906 struct trace_array *tr = m->private;
910 list_for_each_entry_continue(file, &tr->events, list) {
911 call = file->event_call;
913 * The ftrace subsystem is for showing formats only.
914 * They can not be enabled or disabled via the event files.
916 if (call->class && call->class->reg &&
917 !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
924 static void *t_start(struct seq_file *m, loff_t *pos)
926 struct trace_event_file *file;
927 struct trace_array *tr = m->private;
930 mutex_lock(&event_mutex);
932 file = list_entry(&tr->events, struct trace_event_file, list);
933 for (l = 0; l <= *pos; ) {
934 file = t_next(m, file, &l);
942 s_next(struct seq_file *m, void *v, loff_t *pos)
944 struct trace_event_file *file = v;
945 struct trace_array *tr = m->private;
949 list_for_each_entry_continue(file, &tr->events, list) {
950 if (file->flags & EVENT_FILE_FL_ENABLED)
957 static void *s_start(struct seq_file *m, loff_t *pos)
959 struct trace_event_file *file;
960 struct trace_array *tr = m->private;
963 mutex_lock(&event_mutex);
965 file = list_entry(&tr->events, struct trace_event_file, list);
966 for (l = 0; l <= *pos; ) {
967 file = s_next(m, file, &l);
974 static int t_show(struct seq_file *m, void *v)
976 struct trace_event_file *file = v;
977 struct trace_event_call *call = file->event_call;
979 if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
980 seq_printf(m, "%s:", call->class->system);
981 seq_printf(m, "%s\n", trace_event_name(call));
986 static void t_stop(struct seq_file *m, void *p)
988 mutex_unlock(&event_mutex);
992 p_next(struct seq_file *m, void *v, loff_t *pos)
994 struct trace_array *tr = m->private;
995 struct trace_pid_list *pid_list = rcu_dereference_sched(tr->filtered_pids);
997 return trace_pid_next(pid_list, v, pos);
1000 static void *p_start(struct seq_file *m, loff_t *pos)
1003 struct trace_pid_list *pid_list;
1004 struct trace_array *tr = m->private;
1007 * Grab the mutex, to keep calls to p_next() having the same
1008 * tr->filtered_pids as p_start() has.
1009 * If we just passed the tr->filtered_pids around, then RCU would
1010 * have been enough, but doing that makes things more complex.
1012 mutex_lock(&event_mutex);
1013 rcu_read_lock_sched();
1015 pid_list = rcu_dereference_sched(tr->filtered_pids);
1020 return trace_pid_start(pid_list, pos);
1023 static void p_stop(struct seq_file *m, void *p)
1026 rcu_read_unlock_sched();
1027 mutex_unlock(&event_mutex);
1031 event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1034 struct trace_event_file *file;
1035 unsigned long flags;
1038 mutex_lock(&event_mutex);
1039 file = event_file_data(filp);
1041 flags = file->flags;
1042 mutex_unlock(&event_mutex);
1047 if (flags & EVENT_FILE_FL_ENABLED &&
1048 !(flags & EVENT_FILE_FL_SOFT_DISABLED))
1051 if (flags & EVENT_FILE_FL_SOFT_DISABLED ||
1052 flags & EVENT_FILE_FL_SOFT_MODE)
1057 return simple_read_from_buffer(ubuf, cnt, ppos, buf, strlen(buf));
1061 event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1064 struct trace_event_file *file;
1068 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1072 ret = tracing_update_buffers();
1080 mutex_lock(&event_mutex);
1081 file = event_file_data(filp);
1083 ret = ftrace_event_enable_disable(file, val);
1084 mutex_unlock(&event_mutex);
1093 return ret ? ret : cnt;
1097 system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1100 const char set_to_char[4] = { '?', '0', '1', 'X' };
1101 struct trace_subsystem_dir *dir = filp->private_data;
1102 struct event_subsystem *system = dir->subsystem;
1103 struct trace_event_call *call;
1104 struct trace_event_file *file;
1105 struct trace_array *tr = dir->tr;
1110 mutex_lock(&event_mutex);
1111 list_for_each_entry(file, &tr->events, list) {
1112 call = file->event_call;
1113 if (!trace_event_name(call) || !call->class || !call->class->reg)
1116 if (system && strcmp(call->class->system, system->name) != 0)
1120 * We need to find out if all the events are set
1121 * or if all events or cleared, or if we have
1124 set |= (1 << !!(file->flags & EVENT_FILE_FL_ENABLED));
1127 * If we have a mixture, no need to look further.
1132 mutex_unlock(&event_mutex);
1134 buf[0] = set_to_char[set];
1137 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
1143 system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1146 struct trace_subsystem_dir *dir = filp->private_data;
1147 struct event_subsystem *system = dir->subsystem;
1148 const char *name = NULL;
1152 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1156 ret = tracing_update_buffers();
1160 if (val != 0 && val != 1)
1164 * Opening of "enable" adds a ref count to system,
1165 * so the name is safe to use.
1168 name = system->name;
1170 ret = __ftrace_set_clr_event(dir->tr, NULL, name, NULL, val);
1184 FORMAT_FIELD_SEPERATOR = 2,
1185 FORMAT_PRINTFMT = 3,
1188 static void *f_next(struct seq_file *m, void *v, loff_t *pos)
1190 struct trace_event_call *call = event_file_data(m->private);
1191 struct list_head *common_head = &ftrace_common_fields;
1192 struct list_head *head = trace_get_fields(call);
1193 struct list_head *node = v;
1197 switch ((unsigned long)v) {
1202 case FORMAT_FIELD_SEPERATOR:
1206 case FORMAT_PRINTFMT:
1212 if (node == common_head)
1213 return (void *)FORMAT_FIELD_SEPERATOR;
1214 else if (node == head)
1215 return (void *)FORMAT_PRINTFMT;
1220 static int f_show(struct seq_file *m, void *v)
1222 struct trace_event_call *call = event_file_data(m->private);
1223 struct ftrace_event_field *field;
1224 const char *array_descriptor;
1226 switch ((unsigned long)v) {
1228 seq_printf(m, "name: %s\n", trace_event_name(call));
1229 seq_printf(m, "ID: %d\n", call->event.type);
1230 seq_puts(m, "format:\n");
1233 case FORMAT_FIELD_SEPERATOR:
1237 case FORMAT_PRINTFMT:
1238 seq_printf(m, "\nprint fmt: %s\n",
1243 field = list_entry(v, struct ftrace_event_field, link);
1245 * Smartly shows the array type(except dynamic array).
1248 * If TYPE := TYPE[LEN], it is shown:
1249 * field:TYPE VAR[LEN]
1251 array_descriptor = strchr(field->type, '[');
1253 if (!strncmp(field->type, "__data_loc", 10))
1254 array_descriptor = NULL;
1256 if (!array_descriptor)
1257 seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1258 field->type, field->name, field->offset,
1259 field->size, !!field->is_signed);
1261 seq_printf(m, "\tfield:%.*s %s%s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1262 (int)(array_descriptor - field->type),
1263 field->type, field->name,
1264 array_descriptor, field->offset,
1265 field->size, !!field->is_signed);
1270 static void *f_start(struct seq_file *m, loff_t *pos)
1272 void *p = (void *)FORMAT_HEADER;
1275 /* ->stop() is called even if ->start() fails */
1276 mutex_lock(&event_mutex);
1277 if (!event_file_data(m->private))
1278 return ERR_PTR(-ENODEV);
1280 while (l < *pos && p)
1281 p = f_next(m, p, &l);
1286 static void f_stop(struct seq_file *m, void *p)
1288 mutex_unlock(&event_mutex);
1291 static const struct seq_operations trace_format_seq_ops = {
1298 static int trace_format_open(struct inode *inode, struct file *file)
1303 ret = seq_open(file, &trace_format_seq_ops);
1307 m = file->private_data;
1314 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1316 int id = (long)event_file_data(filp);
1326 len = sprintf(buf, "%d\n", id);
1328 return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
1332 event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1335 struct trace_event_file *file;
1336 struct trace_seq *s;
1342 s = kmalloc(sizeof(*s), GFP_KERNEL);
1349 mutex_lock(&event_mutex);
1350 file = event_file_data(filp);
1352 print_event_filter(file, s);
1353 mutex_unlock(&event_mutex);
1356 r = simple_read_from_buffer(ubuf, cnt, ppos,
1357 s->buffer, trace_seq_used(s));
1365 event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1368 struct trace_event_file *file;
1372 if (cnt >= PAGE_SIZE)
1375 buf = memdup_user_nul(ubuf, cnt);
1377 return PTR_ERR(buf);
1379 mutex_lock(&event_mutex);
1380 file = event_file_data(filp);
1382 err = apply_event_filter(file, buf);
1383 mutex_unlock(&event_mutex);
1394 static LIST_HEAD(event_subsystems);
1396 static int subsystem_open(struct inode *inode, struct file *filp)
1398 struct event_subsystem *system = NULL;
1399 struct trace_subsystem_dir *dir = NULL; /* Initialize for gcc */
1400 struct trace_array *tr;
1403 if (tracing_is_disabled())
1406 /* Make sure the system still exists */
1407 mutex_lock(&event_mutex);
1408 mutex_lock(&trace_types_lock);
1409 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
1410 list_for_each_entry(dir, &tr->systems, list) {
1411 if (dir == inode->i_private) {
1412 /* Don't open systems with no events */
1413 if (dir->nr_events) {
1414 __get_system_dir(dir);
1415 system = dir->subsystem;
1422 mutex_unlock(&trace_types_lock);
1423 mutex_unlock(&event_mutex);
1428 /* Some versions of gcc think dir can be uninitialized here */
1431 /* Still need to increment the ref count of the system */
1432 if (trace_array_get(tr) < 0) {
1437 ret = tracing_open_generic(inode, filp);
1439 trace_array_put(tr);
1446 static int system_tr_open(struct inode *inode, struct file *filp)
1448 struct trace_subsystem_dir *dir;
1449 struct trace_array *tr = inode->i_private;
1452 if (tracing_is_disabled())
1455 if (trace_array_get(tr) < 0)
1458 /* Make a temporary dir that has no system but points to tr */
1459 dir = kzalloc(sizeof(*dir), GFP_KERNEL);
1461 trace_array_put(tr);
1467 ret = tracing_open_generic(inode, filp);
1469 trace_array_put(tr);
1474 filp->private_data = dir;
1479 static int subsystem_release(struct inode *inode, struct file *file)
1481 struct trace_subsystem_dir *dir = file->private_data;
1483 trace_array_put(dir->tr);
1486 * If dir->subsystem is NULL, then this is a temporary
1487 * descriptor that was made for a trace_array to enable
1499 subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1502 struct trace_subsystem_dir *dir = filp->private_data;
1503 struct event_subsystem *system = dir->subsystem;
1504 struct trace_seq *s;
1510 s = kmalloc(sizeof(*s), GFP_KERNEL);
1516 print_subsystem_event_filter(system, s);
1517 r = simple_read_from_buffer(ubuf, cnt, ppos,
1518 s->buffer, trace_seq_used(s));
1526 subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1529 struct trace_subsystem_dir *dir = filp->private_data;
1533 if (cnt >= PAGE_SIZE)
1536 buf = memdup_user_nul(ubuf, cnt);
1538 return PTR_ERR(buf);
1540 err = apply_subsystem_event_filter(dir, buf);
1551 show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1553 int (*func)(struct trace_seq *s) = filp->private_data;
1554 struct trace_seq *s;
1560 s = kmalloc(sizeof(*s), GFP_KERNEL);
1567 r = simple_read_from_buffer(ubuf, cnt, ppos,
1568 s->buffer, trace_seq_used(s));
1575 static void ignore_task_cpu(void *data)
1577 struct trace_array *tr = data;
1578 struct trace_pid_list *pid_list;
1581 * This function is called by on_each_cpu() while the
1582 * event_mutex is held.
1584 pid_list = rcu_dereference_protected(tr->filtered_pids,
1585 mutex_is_locked(&event_mutex));
1587 this_cpu_write(tr->trace_buffer.data->ignore_pid,
1588 trace_ignore_this_task(pid_list, current));
1592 ftrace_event_pid_write(struct file *filp, const char __user *ubuf,
1593 size_t cnt, loff_t *ppos)
1595 struct seq_file *m = filp->private_data;
1596 struct trace_array *tr = m->private;
1597 struct trace_pid_list *filtered_pids = NULL;
1598 struct trace_pid_list *pid_list;
1599 struct trace_event_file *file;
1605 ret = tracing_update_buffers();
1609 mutex_lock(&event_mutex);
1611 filtered_pids = rcu_dereference_protected(tr->filtered_pids,
1612 lockdep_is_held(&event_mutex));
1614 ret = trace_pid_write(filtered_pids, &pid_list, ubuf, cnt);
1618 rcu_assign_pointer(tr->filtered_pids, pid_list);
1620 list_for_each_entry(file, &tr->events, list) {
1621 set_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
1624 if (filtered_pids) {
1625 synchronize_sched();
1626 trace_free_pid_list(filtered_pids);
1627 } else if (pid_list) {
1629 * Register a probe that is called before all other probes
1630 * to set ignore_pid if next or prev do not match.
1631 * Register a probe this is called after all other probes
1632 * to only keep ignore_pid set if next pid matches.
1634 register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_pre,
1636 register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_post,
1639 register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre,
1641 register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_post,
1644 register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre,
1646 register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post,
1649 register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_pre,
1651 register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_post,
1656 * Ignoring of pids is done at task switch. But we have to
1657 * check for those tasks that are currently running.
1658 * Always do this in case a pid was appended or removed.
1660 on_each_cpu(ignore_task_cpu, tr, 1);
1663 mutex_unlock(&event_mutex);
1671 static int ftrace_event_avail_open(struct inode *inode, struct file *file);
1672 static int ftrace_event_set_open(struct inode *inode, struct file *file);
1673 static int ftrace_event_set_pid_open(struct inode *inode, struct file *file);
1674 static int ftrace_event_release(struct inode *inode, struct file *file);
1676 static const struct seq_operations show_event_seq_ops = {
1683 static const struct seq_operations show_set_event_seq_ops = {
1690 static const struct seq_operations show_set_pid_seq_ops = {
1693 .show = trace_pid_show,
1697 static const struct file_operations ftrace_avail_fops = {
1698 .open = ftrace_event_avail_open,
1700 .llseek = seq_lseek,
1701 .release = seq_release,
1704 static const struct file_operations ftrace_set_event_fops = {
1705 .open = ftrace_event_set_open,
1707 .write = ftrace_event_write,
1708 .llseek = seq_lseek,
1709 .release = ftrace_event_release,
1712 static const struct file_operations ftrace_set_event_pid_fops = {
1713 .open = ftrace_event_set_pid_open,
1715 .write = ftrace_event_pid_write,
1716 .llseek = seq_lseek,
1717 .release = ftrace_event_release,
1720 static const struct file_operations ftrace_enable_fops = {
1721 .open = tracing_open_generic,
1722 .read = event_enable_read,
1723 .write = event_enable_write,
1724 .llseek = default_llseek,
1727 static const struct file_operations ftrace_event_format_fops = {
1728 .open = trace_format_open,
1730 .llseek = seq_lseek,
1731 .release = seq_release,
1734 static const struct file_operations ftrace_event_id_fops = {
1735 .read = event_id_read,
1736 .llseek = default_llseek,
1739 static const struct file_operations ftrace_event_filter_fops = {
1740 .open = tracing_open_generic,
1741 .read = event_filter_read,
1742 .write = event_filter_write,
1743 .llseek = default_llseek,
1746 static const struct file_operations ftrace_subsystem_filter_fops = {
1747 .open = subsystem_open,
1748 .read = subsystem_filter_read,
1749 .write = subsystem_filter_write,
1750 .llseek = default_llseek,
1751 .release = subsystem_release,
1754 static const struct file_operations ftrace_system_enable_fops = {
1755 .open = subsystem_open,
1756 .read = system_enable_read,
1757 .write = system_enable_write,
1758 .llseek = default_llseek,
1759 .release = subsystem_release,
1762 static const struct file_operations ftrace_tr_enable_fops = {
1763 .open = system_tr_open,
1764 .read = system_enable_read,
1765 .write = system_enable_write,
1766 .llseek = default_llseek,
1767 .release = subsystem_release,
1770 static const struct file_operations ftrace_show_header_fops = {
1771 .open = tracing_open_generic,
1772 .read = show_header,
1773 .llseek = default_llseek,
1777 ftrace_event_open(struct inode *inode, struct file *file,
1778 const struct seq_operations *seq_ops)
1783 ret = seq_open(file, seq_ops);
1786 m = file->private_data;
1787 /* copy tr over to seq ops */
1788 m->private = inode->i_private;
1793 static int ftrace_event_release(struct inode *inode, struct file *file)
1795 struct trace_array *tr = inode->i_private;
1797 trace_array_put(tr);
1799 return seq_release(inode, file);
1803 ftrace_event_avail_open(struct inode *inode, struct file *file)
1805 const struct seq_operations *seq_ops = &show_event_seq_ops;
1807 return ftrace_event_open(inode, file, seq_ops);
1811 ftrace_event_set_open(struct inode *inode, struct file *file)
1813 const struct seq_operations *seq_ops = &show_set_event_seq_ops;
1814 struct trace_array *tr = inode->i_private;
1817 if (trace_array_get(tr) < 0)
1820 if ((file->f_mode & FMODE_WRITE) &&
1821 (file->f_flags & O_TRUNC))
1822 ftrace_clear_events(tr);
1824 ret = ftrace_event_open(inode, file, seq_ops);
1826 trace_array_put(tr);
1831 ftrace_event_set_pid_open(struct inode *inode, struct file *file)
1833 const struct seq_operations *seq_ops = &show_set_pid_seq_ops;
1834 struct trace_array *tr = inode->i_private;
1837 if (trace_array_get(tr) < 0)
1840 if ((file->f_mode & FMODE_WRITE) &&
1841 (file->f_flags & O_TRUNC))
1842 ftrace_clear_event_pids(tr);
1844 ret = ftrace_event_open(inode, file, seq_ops);
1846 trace_array_put(tr);
1850 static struct event_subsystem *
1851 create_new_subsystem(const char *name)
1853 struct event_subsystem *system;
1855 /* need to create new entry */
1856 system = kmalloc(sizeof(*system), GFP_KERNEL);
1860 system->ref_count = 1;
1862 /* Only allocate if dynamic (kprobes and modules) */
1863 system->name = kstrdup_const(name, GFP_KERNEL);
1867 system->filter = NULL;
1869 system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
1870 if (!system->filter)
1873 list_add(&system->list, &event_subsystems);
1878 kfree_const(system->name);
1883 static struct dentry *
1884 event_subsystem_dir(struct trace_array *tr, const char *name,
1885 struct trace_event_file *file, struct dentry *parent)
1887 struct trace_subsystem_dir *dir;
1888 struct event_subsystem *system;
1889 struct dentry *entry;
1891 /* First see if we did not already create this dir */
1892 list_for_each_entry(dir, &tr->systems, list) {
1893 system = dir->subsystem;
1894 if (strcmp(system->name, name) == 0) {
1901 /* Now see if the system itself exists. */
1902 list_for_each_entry(system, &event_subsystems, list) {
1903 if (strcmp(system->name, name) == 0)
1906 /* Reset system variable when not found */
1907 if (&system->list == &event_subsystems)
1910 dir = kmalloc(sizeof(*dir), GFP_KERNEL);
1915 system = create_new_subsystem(name);
1919 __get_system(system);
1921 dir->entry = tracefs_create_dir(name, parent);
1923 pr_warn("Failed to create system directory %s\n", name);
1924 __put_system(system);
1931 dir->subsystem = system;
1934 entry = tracefs_create_file("filter", 0644, dir->entry, dir,
1935 &ftrace_subsystem_filter_fops);
1937 kfree(system->filter);
1938 system->filter = NULL;
1939 pr_warn("Could not create tracefs '%s/filter' entry\n", name);
1942 trace_create_file("enable", 0644, dir->entry, dir,
1943 &ftrace_system_enable_fops);
1945 list_add(&dir->list, &tr->systems);
1952 /* Only print this message if failed on memory allocation */
1953 if (!dir || !system)
1954 pr_warn("No memory to create event subsystem %s\n", name);
1959 event_create_dir(struct dentry *parent, struct trace_event_file *file)
1961 struct trace_event_call *call = file->event_call;
1962 struct trace_array *tr = file->tr;
1963 struct list_head *head;
1964 struct dentry *d_events;
1969 * If the trace point header did not define TRACE_SYSTEM
1970 * then the system would be called "TRACE_SYSTEM".
1972 if (strcmp(call->class->system, TRACE_SYSTEM) != 0) {
1973 d_events = event_subsystem_dir(tr, call->class->system, file, parent);
1979 name = trace_event_name(call);
1980 file->dir = tracefs_create_dir(name, d_events);
1982 pr_warn("Could not create tracefs '%s' directory\n", name);
1986 if (call->class->reg && !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
1987 trace_create_file("enable", 0644, file->dir, file,
1988 &ftrace_enable_fops);
1990 #ifdef CONFIG_PERF_EVENTS
1991 if (call->event.type && call->class->reg)
1992 trace_create_file("id", 0444, file->dir,
1993 (void *)(long)call->event.type,
1994 &ftrace_event_id_fops);
1998 * Other events may have the same class. Only update
1999 * the fields if they are not already defined.
2001 head = trace_get_fields(call);
2002 if (list_empty(head)) {
2003 ret = call->class->define_fields(call);
2005 pr_warn("Could not initialize trace point events/%s\n",
2012 * Only event directories that can be enabled should have
2013 * triggers or filters.
2015 if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)) {
2016 trace_create_file("filter", 0644, file->dir, file,
2017 &ftrace_event_filter_fops);
2019 trace_create_file("trigger", 0644, file->dir, file,
2020 &event_trigger_fops);
2023 #ifdef CONFIG_HIST_TRIGGERS
2024 trace_create_file("hist", 0444, file->dir, file,
2027 trace_create_file("format", 0444, file->dir, call,
2028 &ftrace_event_format_fops);
2033 static void remove_event_from_tracers(struct trace_event_call *call)
2035 struct trace_event_file *file;
2036 struct trace_array *tr;
2038 do_for_each_event_file_safe(tr, file) {
2039 if (file->event_call != call)
2042 remove_event_file_dir(file);
2044 * The do_for_each_event_file_safe() is
2045 * a double loop. After finding the call for this
2046 * trace_array, we use break to jump to the next
2050 } while_for_each_event_file();
2053 static void event_remove(struct trace_event_call *call)
2055 struct trace_array *tr;
2056 struct trace_event_file *file;
2058 do_for_each_event_file(tr, file) {
2059 if (file->event_call != call)
2062 if (file->flags & EVENT_FILE_FL_WAS_ENABLED)
2063 tr->clear_trace = true;
2065 ftrace_event_enable_disable(file, 0);
2067 * The do_for_each_event_file() is
2068 * a double loop. After finding the call for this
2069 * trace_array, we use break to jump to the next
2073 } while_for_each_event_file();
2075 if (call->event.funcs)
2076 __unregister_trace_event(&call->event);
2077 remove_event_from_tracers(call);
2078 list_del(&call->list);
2081 static int event_init(struct trace_event_call *call)
2086 name = trace_event_name(call);
2090 if (call->class->raw_init) {
2091 ret = call->class->raw_init(call);
2092 if (ret < 0 && ret != -ENOSYS)
2093 pr_warn("Could not initialize trace events/%s\n", name);
2100 __register_event(struct trace_event_call *call, struct module *mod)
2104 ret = event_init(call);
2108 list_add(&call->list, &ftrace_events);
2114 static char *eval_replace(char *ptr, struct trace_eval_map *map, int len)
2119 /* Find the length of the eval value as a string */
2120 elen = snprintf(ptr, 0, "%ld", map->eval_value);
2121 /* Make sure there's enough room to replace the string with the value */
2125 snprintf(ptr, elen + 1, "%ld", map->eval_value);
2127 /* Get the rest of the string of ptr */
2128 rlen = strlen(ptr + len);
2129 memmove(ptr + elen, ptr + len, rlen);
2130 /* Make sure we end the new string */
2131 ptr[elen + rlen] = 0;
2136 static void update_event_printk(struct trace_event_call *call,
2137 struct trace_eval_map *map)
2141 int len = strlen(map->eval_string);
2143 for (ptr = call->print_fmt; *ptr; ptr++) {
2157 if (isdigit(*ptr)) {
2161 /* Check for alpha chars like ULL */
2162 } while (isalnum(*ptr));
2166 * A number must have some kind of delimiter after
2167 * it, and we can ignore that too.
2171 if (isalpha(*ptr) || *ptr == '_') {
2172 if (strncmp(map->eval_string, ptr, len) == 0 &&
2173 !isalnum(ptr[len]) && ptr[len] != '_') {
2174 ptr = eval_replace(ptr, map, len);
2175 /* enum/sizeof string smaller than value */
2176 if (WARN_ON_ONCE(!ptr))
2179 * No need to decrement here, as eval_replace()
2180 * returns the pointer to the character passed
2181 * the eval, and two evals can not be placed
2182 * back to back without something in between.
2183 * We can skip that something in between.
2190 } while (isalnum(*ptr) || *ptr == '_');
2194 * If what comes after this variable is a '.' or
2195 * '->' then we can continue to ignore that string.
2197 if (*ptr == '.' || (ptr[0] == '-' && ptr[1] == '>')) {
2198 ptr += *ptr == '.' ? 1 : 2;
2204 * Once again, we can skip the delimiter that came
2212 void trace_event_eval_update(struct trace_eval_map **map, int len)
2214 struct trace_event_call *call, *p;
2215 const char *last_system = NULL;
2220 down_write(&trace_event_sem);
2221 list_for_each_entry_safe(call, p, &ftrace_events, list) {
2222 /* events are usually grouped together with systems */
2223 if (!last_system || call->class->system != last_system) {
2226 last_system = call->class->system;
2230 * Since calls are grouped by systems, the likelyhood that the
2231 * next call in the iteration belongs to the same system as the
2232 * previous call is high. As an optimization, we skip seaching
2233 * for a map[] that matches the call's system if the last call
2234 * was from the same system. That's what last_i is for. If the
2235 * call has the same system as the previous call, then last_i
2236 * will be the index of the first map[] that has a matching
2239 for (i = last_i; i < len; i++) {
2240 if (call->class->system == map[i]->system) {
2241 /* Save the first system if need be */
2246 update_event_printk(call, map[i]);
2250 up_write(&trace_event_sem);
2253 static struct trace_event_file *
2254 trace_create_new_event(struct trace_event_call *call,
2255 struct trace_array *tr)
2257 struct trace_event_file *file;
2259 file = kmem_cache_alloc(file_cachep, GFP_TRACE);
2263 file->event_call = call;
2265 atomic_set(&file->sm_ref, 0);
2266 atomic_set(&file->tm_ref, 0);
2267 INIT_LIST_HEAD(&file->triggers);
2268 list_add(&file->list, &tr->events);
2273 /* Add an event to a trace directory */
2275 __trace_add_new_event(struct trace_event_call *call, struct trace_array *tr)
2277 struct trace_event_file *file;
2279 file = trace_create_new_event(call, tr);
2283 return event_create_dir(tr->event_dir, file);
2287 * Just create a decriptor for early init. A descriptor is required
2288 * for enabling events at boot. We want to enable events before
2289 * the filesystem is initialized.
2292 __trace_early_add_new_event(struct trace_event_call *call,
2293 struct trace_array *tr)
2295 struct trace_event_file *file;
2297 file = trace_create_new_event(call, tr);
2304 struct ftrace_module_file_ops;
2305 static void __add_event_to_tracers(struct trace_event_call *call);
2307 /* Add an additional event_call dynamically */
2308 int trace_add_event_call(struct trace_event_call *call)
2311 mutex_lock(&event_mutex);
2312 mutex_lock(&trace_types_lock);
2314 ret = __register_event(call, NULL);
2316 __add_event_to_tracers(call);
2318 mutex_unlock(&trace_types_lock);
2319 mutex_unlock(&event_mutex);
2324 * Must be called under locking of trace_types_lock, event_mutex and
2327 static void __trace_remove_event_call(struct trace_event_call *call)
2330 trace_destroy_fields(call);
2331 free_event_filter(call->filter);
2332 call->filter = NULL;
2335 static int probe_remove_event_call(struct trace_event_call *call)
2337 struct trace_array *tr;
2338 struct trace_event_file *file;
2340 #ifdef CONFIG_PERF_EVENTS
2341 if (call->perf_refcount)
2344 do_for_each_event_file(tr, file) {
2345 if (file->event_call != call)
2348 * We can't rely on ftrace_event_enable_disable(enable => 0)
2349 * we are going to do, EVENT_FILE_FL_SOFT_MODE can suppress
2350 * TRACE_REG_UNREGISTER.
2352 if (file->flags & EVENT_FILE_FL_ENABLED)
2355 * The do_for_each_event_file_safe() is
2356 * a double loop. After finding the call for this
2357 * trace_array, we use break to jump to the next
2361 } while_for_each_event_file();
2363 __trace_remove_event_call(call);
2368 /* Remove an event_call */
2369 int trace_remove_event_call(struct trace_event_call *call)
2373 mutex_lock(&event_mutex);
2374 mutex_lock(&trace_types_lock);
2375 down_write(&trace_event_sem);
2376 ret = probe_remove_event_call(call);
2377 up_write(&trace_event_sem);
2378 mutex_unlock(&trace_types_lock);
2379 mutex_unlock(&event_mutex);
2384 #define for_each_event(event, start, end) \
2385 for (event = start; \
2386 (unsigned long)event < (unsigned long)end; \
2389 #ifdef CONFIG_MODULES
2391 static void trace_module_add_events(struct module *mod)
2393 struct trace_event_call **call, **start, **end;
2395 if (!mod->num_trace_events)
2398 /* Don't add infrastructure for mods without tracepoints */
2399 if (trace_module_has_bad_taint(mod)) {
2400 pr_err("%s: module has bad taint, not creating trace events\n",
2405 start = mod->trace_events;
2406 end = mod->trace_events + mod->num_trace_events;
2408 for_each_event(call, start, end) {
2409 __register_event(*call, mod);
2410 __add_event_to_tracers(*call);
2414 static void trace_module_remove_events(struct module *mod)
2416 struct trace_event_call *call, *p;
2418 down_write(&trace_event_sem);
2419 list_for_each_entry_safe(call, p, &ftrace_events, list) {
2420 if (call->mod == mod)
2421 __trace_remove_event_call(call);
2423 up_write(&trace_event_sem);
2426 * It is safest to reset the ring buffer if the module being unloaded
2427 * registered any events that were used. The only worry is if
2428 * a new module gets loaded, and takes on the same id as the events
2429 * of this module. When printing out the buffer, traced events left
2430 * over from this module may be passed to the new module events and
2431 * unexpected results may occur.
2433 tracing_reset_all_online_cpus();
2436 static int trace_module_notify(struct notifier_block *self,
2437 unsigned long val, void *data)
2439 struct module *mod = data;
2441 mutex_lock(&event_mutex);
2442 mutex_lock(&trace_types_lock);
2444 case MODULE_STATE_COMING:
2445 trace_module_add_events(mod);
2447 case MODULE_STATE_GOING:
2448 trace_module_remove_events(mod);
2451 mutex_unlock(&trace_types_lock);
2452 mutex_unlock(&event_mutex);
2457 static struct notifier_block trace_module_nb = {
2458 .notifier_call = trace_module_notify,
2459 .priority = 1, /* higher than trace.c module notify */
2461 #endif /* CONFIG_MODULES */
2463 /* Create a new event directory structure for a trace directory. */
2465 __trace_add_event_dirs(struct trace_array *tr)
2467 struct trace_event_call *call;
2470 list_for_each_entry(call, &ftrace_events, list) {
2471 ret = __trace_add_new_event(call, tr);
2473 pr_warn("Could not create directory for event %s\n",
2474 trace_event_name(call));
2478 /* Returns any file that matches the system and event */
2479 struct trace_event_file *
2480 __find_event_file(struct trace_array *tr, const char *system, const char *event)
2482 struct trace_event_file *file;
2483 struct trace_event_call *call;
2486 list_for_each_entry(file, &tr->events, list) {
2488 call = file->event_call;
2489 name = trace_event_name(call);
2491 if (!name || !call->class)
2494 if (strcmp(event, name) == 0 &&
2495 strcmp(system, call->class->system) == 0)
2501 /* Returns valid trace event files that match system and event */
2502 struct trace_event_file *
2503 find_event_file(struct trace_array *tr, const char *system, const char *event)
2505 struct trace_event_file *file;
2507 file = __find_event_file(tr, system, event);
2508 if (!file || !file->event_call->class->reg ||
2509 file->event_call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
2515 #ifdef CONFIG_DYNAMIC_FTRACE
2518 #define ENABLE_EVENT_STR "enable_event"
2519 #define DISABLE_EVENT_STR "disable_event"
2521 struct event_probe_data {
2522 struct trace_event_file *file;
2523 unsigned long count;
2528 static void update_event_probe(struct event_probe_data *data)
2531 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
2533 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
2537 event_enable_probe(unsigned long ip, unsigned long parent_ip,
2538 struct trace_array *tr, struct ftrace_probe_ops *ops,
2541 struct ftrace_func_mapper *mapper = data;
2542 struct event_probe_data *edata;
2545 pdata = ftrace_func_mapper_find_ip(mapper, ip);
2546 if (!pdata || !*pdata)
2550 update_event_probe(edata);
2554 event_enable_count_probe(unsigned long ip, unsigned long parent_ip,
2555 struct trace_array *tr, struct ftrace_probe_ops *ops,
2558 struct ftrace_func_mapper *mapper = data;
2559 struct event_probe_data *edata;
2562 pdata = ftrace_func_mapper_find_ip(mapper, ip);
2563 if (!pdata || !*pdata)
2571 /* Skip if the event is in a state we want to switch to */
2572 if (edata->enable == !(edata->file->flags & EVENT_FILE_FL_SOFT_DISABLED))
2575 if (edata->count != -1)
2578 update_event_probe(edata);
2582 event_enable_print(struct seq_file *m, unsigned long ip,
2583 struct ftrace_probe_ops *ops, void *data)
2585 struct ftrace_func_mapper *mapper = data;
2586 struct event_probe_data *edata;
2589 pdata = ftrace_func_mapper_find_ip(mapper, ip);
2591 if (WARN_ON_ONCE(!pdata || !*pdata))
2596 seq_printf(m, "%ps:", (void *)ip);
2598 seq_printf(m, "%s:%s:%s",
2599 edata->enable ? ENABLE_EVENT_STR : DISABLE_EVENT_STR,
2600 edata->file->event_call->class->system,
2601 trace_event_name(edata->file->event_call));
2603 if (edata->count == -1)
2604 seq_puts(m, ":unlimited\n");
2606 seq_printf(m, ":count=%ld\n", edata->count);
2612 event_enable_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
2613 unsigned long ip, void *init_data, void **data)
2615 struct ftrace_func_mapper *mapper = *data;
2616 struct event_probe_data *edata = init_data;
2620 mapper = allocate_ftrace_func_mapper();
2626 ret = ftrace_func_mapper_add_ip(mapper, ip, edata);
2635 static int free_probe_data(void *data)
2637 struct event_probe_data *edata = data;
2641 /* Remove the SOFT_MODE flag */
2642 __ftrace_event_enable_disable(edata->file, 0, 1);
2643 module_put(edata->file->event_call->mod);
2650 event_enable_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
2651 unsigned long ip, void *data)
2653 struct ftrace_func_mapper *mapper = data;
2654 struct event_probe_data *edata;
2659 free_ftrace_func_mapper(mapper, free_probe_data);
2663 edata = ftrace_func_mapper_remove_ip(mapper, ip);
2665 if (WARN_ON_ONCE(!edata))
2668 if (WARN_ON_ONCE(edata->ref <= 0))
2671 free_probe_data(edata);
2674 static struct ftrace_probe_ops event_enable_probe_ops = {
2675 .func = event_enable_probe,
2676 .print = event_enable_print,
2677 .init = event_enable_init,
2678 .free = event_enable_free,
2681 static struct ftrace_probe_ops event_enable_count_probe_ops = {
2682 .func = event_enable_count_probe,
2683 .print = event_enable_print,
2684 .init = event_enable_init,
2685 .free = event_enable_free,
2688 static struct ftrace_probe_ops event_disable_probe_ops = {
2689 .func = event_enable_probe,
2690 .print = event_enable_print,
2691 .init = event_enable_init,
2692 .free = event_enable_free,
2695 static struct ftrace_probe_ops event_disable_count_probe_ops = {
2696 .func = event_enable_count_probe,
2697 .print = event_enable_print,
2698 .init = event_enable_init,
2699 .free = event_enable_free,
2703 event_enable_func(struct trace_array *tr, struct ftrace_hash *hash,
2704 char *glob, char *cmd, char *param, int enabled)
2706 struct trace_event_file *file;
2707 struct ftrace_probe_ops *ops;
2708 struct event_probe_data *data;
2718 /* hash funcs only work with set_ftrace_filter */
2719 if (!enabled || !param)
2722 system = strsep(¶m, ":");
2726 event = strsep(¶m, ":");
2728 mutex_lock(&event_mutex);
2731 file = find_event_file(tr, system, event);
2735 enable = strcmp(cmd, ENABLE_EVENT_STR) == 0;
2738 ops = param ? &event_enable_count_probe_ops : &event_enable_probe_ops;
2740 ops = param ? &event_disable_count_probe_ops : &event_disable_probe_ops;
2742 if (glob[0] == '!') {
2743 ret = unregister_ftrace_function_probe_func(glob+1, tr, ops);
2749 data = kzalloc(sizeof(*data), GFP_KERNEL);
2753 data->enable = enable;
2760 number = strsep(¶m, ":");
2763 if (!strlen(number))
2767 * We use the callback data field (which is a pointer)
2770 ret = kstrtoul(number, 0, &data->count);
2775 /* Don't let event modules unload while probe registered */
2776 ret = try_module_get(file->event_call->mod);
2782 ret = __ftrace_event_enable_disable(file, 1, 1);
2786 ret = register_ftrace_function_probe(glob, tr, ops, data);
2788 * The above returns on success the # of functions enabled,
2789 * but if it didn't find any functions it returns zero.
2790 * Consider no functions a failure too.
2797 /* Just return zero, not the number of enabled functions */
2800 mutex_unlock(&event_mutex);
2804 __ftrace_event_enable_disable(file, 0, 1);
2806 module_put(file->event_call->mod);
2812 static struct ftrace_func_command event_enable_cmd = {
2813 .name = ENABLE_EVENT_STR,
2814 .func = event_enable_func,
2817 static struct ftrace_func_command event_disable_cmd = {
2818 .name = DISABLE_EVENT_STR,
2819 .func = event_enable_func,
2822 static __init int register_event_cmds(void)
2826 ret = register_ftrace_command(&event_enable_cmd);
2827 if (WARN_ON(ret < 0))
2829 ret = register_ftrace_command(&event_disable_cmd);
2830 if (WARN_ON(ret < 0))
2831 unregister_ftrace_command(&event_enable_cmd);
2835 static inline int register_event_cmds(void) { return 0; }
2836 #endif /* CONFIG_DYNAMIC_FTRACE */
2839 * The top level array has already had its trace_event_file
2840 * descriptors created in order to allow for early events to
2841 * be recorded. This function is called after the tracefs has been
2842 * initialized, and we now have to create the files associated
2846 __trace_early_add_event_dirs(struct trace_array *tr)
2848 struct trace_event_file *file;
2852 list_for_each_entry(file, &tr->events, list) {
2853 ret = event_create_dir(tr->event_dir, file);
2855 pr_warn("Could not create directory for event %s\n",
2856 trace_event_name(file->event_call));
2861 * For early boot up, the top trace array requires to have
2862 * a list of events that can be enabled. This must be done before
2863 * the filesystem is set up in order to allow events to be traced
2867 __trace_early_add_events(struct trace_array *tr)
2869 struct trace_event_call *call;
2872 list_for_each_entry(call, &ftrace_events, list) {
2873 /* Early boot up should not have any modules loaded */
2874 if (WARN_ON_ONCE(call->mod))
2877 ret = __trace_early_add_new_event(call, tr);
2879 pr_warn("Could not create early event %s\n",
2880 trace_event_name(call));
2884 /* Remove the event directory structure for a trace directory. */
2886 __trace_remove_event_dirs(struct trace_array *tr)
2888 struct trace_event_file *file, *next;
2890 list_for_each_entry_safe(file, next, &tr->events, list)
2891 remove_event_file_dir(file);
2894 static void __add_event_to_tracers(struct trace_event_call *call)
2896 struct trace_array *tr;
2898 list_for_each_entry(tr, &ftrace_trace_arrays, list)
2899 __trace_add_new_event(call, tr);
2902 extern struct trace_event_call *__start_ftrace_events[];
2903 extern struct trace_event_call *__stop_ftrace_events[];
2905 static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
2907 static __init int setup_trace_event(char *str)
2909 strlcpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
2910 ring_buffer_expanded = true;
2911 tracing_selftest_disabled = true;
2915 __setup("trace_event=", setup_trace_event);
2917 /* Expects to have event_mutex held when called */
2919 create_event_toplevel_files(struct dentry *parent, struct trace_array *tr)
2921 struct dentry *d_events;
2922 struct dentry *entry;
2924 entry = tracefs_create_file("set_event", 0644, parent,
2925 tr, &ftrace_set_event_fops);
2927 pr_warn("Could not create tracefs 'set_event' entry\n");
2931 d_events = tracefs_create_dir("events", parent);
2933 pr_warn("Could not create tracefs 'events' directory\n");
2937 entry = trace_create_file("enable", 0644, d_events,
2938 tr, &ftrace_tr_enable_fops);
2940 pr_warn("Could not create tracefs 'enable' entry\n");
2944 /* There are not as crucial, just warn if they are not created */
2946 entry = tracefs_create_file("set_event_pid", 0644, parent,
2947 tr, &ftrace_set_event_pid_fops);
2949 pr_warn("Could not create tracefs 'set_event_pid' entry\n");
2951 /* ring buffer internal formats */
2952 entry = trace_create_file("header_page", 0444, d_events,
2953 ring_buffer_print_page_header,
2954 &ftrace_show_header_fops);
2956 pr_warn("Could not create tracefs 'header_page' entry\n");
2958 entry = trace_create_file("header_event", 0444, d_events,
2959 ring_buffer_print_entry_header,
2960 &ftrace_show_header_fops);
2962 pr_warn("Could not create tracefs 'header_event' entry\n");
2964 tr->event_dir = d_events;
2970 * event_trace_add_tracer - add a instance of a trace_array to events
2971 * @parent: The parent dentry to place the files/directories for events in
2972 * @tr: The trace array associated with these events
2974 * When a new instance is created, it needs to set up its events
2975 * directory, as well as other files associated with events. It also
2976 * creates the event hierachry in the @parent/events directory.
2978 * Returns 0 on success.
2980 * Must be called with event_mutex held.
2982 int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr)
2986 lockdep_assert_held(&event_mutex);
2988 ret = create_event_toplevel_files(parent, tr);
2992 down_write(&trace_event_sem);
2993 __trace_add_event_dirs(tr);
2994 up_write(&trace_event_sem);
3001 * The top trace array already had its file descriptors created.
3002 * Now the files themselves need to be created.
3005 early_event_add_tracer(struct dentry *parent, struct trace_array *tr)
3009 mutex_lock(&event_mutex);
3011 ret = create_event_toplevel_files(parent, tr);
3015 down_write(&trace_event_sem);
3016 __trace_early_add_event_dirs(tr);
3017 up_write(&trace_event_sem);
3020 mutex_unlock(&event_mutex);
3025 /* Must be called with event_mutex held */
3026 int event_trace_del_tracer(struct trace_array *tr)
3028 lockdep_assert_held(&event_mutex);
3030 /* Disable any event triggers and associated soft-disabled events */
3031 clear_event_triggers(tr);
3033 /* Clear the pid list */
3034 __ftrace_clear_event_pids(tr);
3036 /* Disable any running events */
3037 __ftrace_set_clr_event_nolock(tr, NULL, NULL, NULL, 0);
3039 /* Access to events are within rcu_read_lock_sched() */
3040 synchronize_sched();
3042 down_write(&trace_event_sem);
3043 __trace_remove_event_dirs(tr);
3044 tracefs_remove_recursive(tr->event_dir);
3045 up_write(&trace_event_sem);
3047 tr->event_dir = NULL;
3052 static __init int event_trace_memsetup(void)
3054 field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC);
3055 file_cachep = KMEM_CACHE(trace_event_file, SLAB_PANIC);
3060 early_enable_events(struct trace_array *tr, bool disable_first)
3062 char *buf = bootup_event_buf;
3067 token = strsep(&buf, ",");
3073 /* Restarting syscalls requires that we stop them first */
3075 ftrace_set_clr_event(tr, token, 0);
3077 ret = ftrace_set_clr_event(tr, token, 1);
3079 pr_warn("Failed to enable trace event: %s\n", token);
3082 /* Put back the comma to allow this to be called again */
3088 static __init int event_trace_enable(void)
3090 struct trace_array *tr = top_trace_array();
3091 struct trace_event_call **iter, *call;
3097 for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
3100 ret = event_init(call);
3102 list_add(&call->list, &ftrace_events);
3106 * We need the top trace array to have a working set of trace
3107 * points at early init, before the debug files and directories
3108 * are created. Create the file entries now, and attach them
3109 * to the actual file dentries later.
3111 __trace_early_add_events(tr);
3113 early_enable_events(tr, false);
3115 trace_printk_start_comm();
3117 register_event_cmds();
3119 register_trigger_cmds();
3125 * event_trace_enable() is called from trace_event_init() first to
3126 * initialize events and perhaps start any events that are on the
3127 * command line. Unfortunately, there are some events that will not
3128 * start this early, like the system call tracepoints that need
3129 * to set the TIF_SYSCALL_TRACEPOINT flag of pid 1. But event_trace_enable()
3130 * is called before pid 1 starts, and this flag is never set, making
3131 * the syscall tracepoint never get reached, but the event is enabled
3132 * regardless (and not doing anything).
3134 static __init int event_trace_enable_again(void)
3136 struct trace_array *tr;
3138 tr = top_trace_array();
3142 early_enable_events(tr, true);
3147 early_initcall(event_trace_enable_again);
3149 __init int event_trace_init(void)
3151 struct trace_array *tr;
3152 struct dentry *d_tracer;
3153 struct dentry *entry;
3156 tr = top_trace_array();
3160 d_tracer = tracing_init_dentry();
3161 if (IS_ERR(d_tracer))
3164 entry = tracefs_create_file("available_events", 0444, d_tracer,
3165 tr, &ftrace_avail_fops);
3167 pr_warn("Could not create tracefs 'available_events' entry\n");
3169 if (trace_define_generic_fields())
3170 pr_warn("tracing: Failed to allocated generic fields");
3172 if (trace_define_common_fields())
3173 pr_warn("tracing: Failed to allocate common fields");
3175 ret = early_event_add_tracer(d_tracer, tr);
3179 #ifdef CONFIG_MODULES
3180 ret = register_module_notifier(&trace_module_nb);
3182 pr_warn("Failed to register trace events module notifier\n");
3187 void __init trace_event_init(void)
3189 event_trace_memsetup();
3190 init_ftrace_syscalls();
3191 event_trace_enable();
3194 #ifdef CONFIG_FTRACE_STARTUP_TEST
3196 static DEFINE_SPINLOCK(test_spinlock);
3197 static DEFINE_SPINLOCK(test_spinlock_irq);
3198 static DEFINE_MUTEX(test_mutex);
3200 static __init void test_work(struct work_struct *dummy)
3202 spin_lock(&test_spinlock);
3203 spin_lock_irq(&test_spinlock_irq);
3205 spin_unlock_irq(&test_spinlock_irq);
3206 spin_unlock(&test_spinlock);
3208 mutex_lock(&test_mutex);
3210 mutex_unlock(&test_mutex);
3213 static __init int event_test_thread(void *unused)
3217 test_malloc = kmalloc(1234, GFP_KERNEL);
3219 pr_info("failed to kmalloc\n");
3221 schedule_on_each_cpu(test_work);
3225 set_current_state(TASK_INTERRUPTIBLE);
3226 while (!kthread_should_stop()) {
3228 set_current_state(TASK_INTERRUPTIBLE);
3230 __set_current_state(TASK_RUNNING);
3236 * Do various things that may trigger events.
3238 static __init void event_test_stuff(void)
3240 struct task_struct *test_thread;
3242 test_thread = kthread_run(event_test_thread, NULL, "test-events");
3244 kthread_stop(test_thread);
3248 * For every trace event defined, we will test each trace point separately,
3249 * and then by groups, and finally all trace points.
3251 static __init void event_trace_self_tests(void)
3253 struct trace_subsystem_dir *dir;
3254 struct trace_event_file *file;
3255 struct trace_event_call *call;
3256 struct event_subsystem *system;
3257 struct trace_array *tr;
3260 tr = top_trace_array();
3264 pr_info("Running tests on trace events:\n");
3266 list_for_each_entry(file, &tr->events, list) {
3268 call = file->event_call;
3270 /* Only test those that have a probe */
3271 if (!call->class || !call->class->probe)
3275 * Testing syscall events here is pretty useless, but
3276 * we still do it if configured. But this is time consuming.
3277 * What we really need is a user thread to perform the
3278 * syscalls as we test.
3280 #ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
3281 if (call->class->system &&
3282 strcmp(call->class->system, "syscalls") == 0)
3286 pr_info("Testing event %s: ", trace_event_name(call));
3289 * If an event is already enabled, someone is using
3290 * it and the self test should not be on.
3292 if (file->flags & EVENT_FILE_FL_ENABLED) {
3293 pr_warn("Enabled event during self test!\n");
3298 ftrace_event_enable_disable(file, 1);
3300 ftrace_event_enable_disable(file, 0);
3305 /* Now test at the sub system level */
3307 pr_info("Running tests on trace event systems:\n");
3309 list_for_each_entry(dir, &tr->systems, list) {
3311 system = dir->subsystem;
3313 /* the ftrace system is special, skip it */
3314 if (strcmp(system->name, "ftrace") == 0)
3317 pr_info("Testing event system %s: ", system->name);
3319 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1);
3320 if (WARN_ON_ONCE(ret)) {
3321 pr_warn("error enabling system %s\n",
3328 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0);
3329 if (WARN_ON_ONCE(ret)) {
3330 pr_warn("error disabling system %s\n",
3338 /* Test with all events enabled */
3340 pr_info("Running tests on all trace events:\n");
3341 pr_info("Testing all events: ");
3343 ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1);
3344 if (WARN_ON_ONCE(ret)) {
3345 pr_warn("error enabling all events\n");
3352 ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
3353 if (WARN_ON_ONCE(ret)) {
3354 pr_warn("error disabling all events\n");
3361 #ifdef CONFIG_FUNCTION_TRACER
3363 static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
3365 static struct trace_event_file event_trace_file __initdata;
3368 function_test_events_call(unsigned long ip, unsigned long parent_ip,
3369 struct ftrace_ops *op, struct pt_regs *pt_regs)
3371 struct ring_buffer_event *event;
3372 struct ring_buffer *buffer;
3373 struct ftrace_entry *entry;
3374 unsigned long flags;
3379 pc = preempt_count();
3380 preempt_disable_notrace();
3381 cpu = raw_smp_processor_id();
3382 disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
3387 local_save_flags(flags);
3389 event = trace_event_buffer_lock_reserve(&buffer, &event_trace_file,
3390 TRACE_FN, sizeof(*entry),
3394 entry = ring_buffer_event_data(event);
3396 entry->parent_ip = parent_ip;
3398 event_trigger_unlock_commit(&event_trace_file, buffer, event,
3401 atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
3402 preempt_enable_notrace();
3405 static struct ftrace_ops trace_ops __initdata =
3407 .func = function_test_events_call,
3408 .flags = FTRACE_OPS_FL_RECURSION_SAFE,
3411 static __init void event_trace_self_test_with_function(void)
3415 event_trace_file.tr = top_trace_array();
3416 if (WARN_ON(!event_trace_file.tr))
3419 ret = register_ftrace_function(&trace_ops);
3420 if (WARN_ON(ret < 0)) {
3421 pr_info("Failed to enable function tracer for event tests\n");
3424 pr_info("Running tests again, along with the function tracer\n");
3425 event_trace_self_tests();
3426 unregister_ftrace_function(&trace_ops);
3429 static __init void event_trace_self_test_with_function(void)
3434 static __init int event_trace_self_tests_init(void)
3436 if (!tracing_selftest_disabled) {
3437 event_trace_self_tests();
3438 event_trace_self_test_with_function();
3444 late_initcall(event_trace_self_tests_init);