1 // SPDX-License-Identifier: GPL-2.0
5 * Copyright (C) 2008 Red Hat Inc, Steven Rostedt <srostedt@redhat.com>
7 * - Added format output of fields of the trace point.
8 * This was based off of work by Tom Zanussi <tzanussi@gmail.com>.
12 #define pr_fmt(fmt) fmt
14 #include <linux/workqueue.h>
15 #include <linux/security.h>
16 #include <linux/spinlock.h>
17 #include <linux/kthread.h>
18 #include <linux/tracefs.h>
19 #include <linux/uaccess.h>
20 #include <linux/module.h>
21 #include <linux/ctype.h>
22 #include <linux/sort.h>
23 #include <linux/slab.h>
24 #include <linux/delay.h>
26 #include <trace/events/sched.h>
27 #include <trace/syscall.h>
29 #include <asm/setup.h>
31 #include "trace_output.h"
34 #define TRACE_SYSTEM "TRACE_SYSTEM"
36 DEFINE_MUTEX(event_mutex);
38 LIST_HEAD(ftrace_events);
39 static LIST_HEAD(ftrace_generic_fields);
40 static LIST_HEAD(ftrace_common_fields);
41 static bool eventdir_initialized;
43 static LIST_HEAD(module_strings);
45 struct module_string {
46 struct list_head next;
47 struct module *module;
51 #define GFP_TRACE (GFP_KERNEL | __GFP_ZERO)
53 static struct kmem_cache *field_cachep;
54 static struct kmem_cache *file_cachep;
56 static inline int system_refcount(struct event_subsystem *system)
58 return system->ref_count;
61 static int system_refcount_inc(struct event_subsystem *system)
63 return system->ref_count++;
66 static int system_refcount_dec(struct event_subsystem *system)
68 return --system->ref_count;
71 /* Double loops, do not use break, only goto's work */
72 #define do_for_each_event_file(tr, file) \
73 list_for_each_entry(tr, &ftrace_trace_arrays, list) { \
74 list_for_each_entry(file, &tr->events, list)
76 #define do_for_each_event_file_safe(tr, file) \
77 list_for_each_entry(tr, &ftrace_trace_arrays, list) { \
78 struct trace_event_file *___n; \
79 list_for_each_entry_safe(file, ___n, &tr->events, list)
81 #define while_for_each_event_file() \
84 static struct ftrace_event_field *
85 __find_event_field(struct list_head *head, char *name)
87 struct ftrace_event_field *field;
89 list_for_each_entry(field, head, link) {
90 if (!strcmp(field->name, name))
97 struct ftrace_event_field *
98 trace_find_event_field(struct trace_event_call *call, char *name)
100 struct ftrace_event_field *field;
101 struct list_head *head;
103 head = trace_get_fields(call);
104 field = __find_event_field(head, name);
108 field = __find_event_field(&ftrace_generic_fields, name);
112 return __find_event_field(&ftrace_common_fields, name);
115 static int __trace_define_field(struct list_head *head, const char *type,
116 const char *name, int offset, int size,
117 int is_signed, int filter_type, int len)
119 struct ftrace_event_field *field;
121 field = kmem_cache_alloc(field_cachep, GFP_TRACE);
128 if (filter_type == FILTER_OTHER)
129 field->filter_type = filter_assign_type(type);
131 field->filter_type = filter_type;
133 field->offset = offset;
135 field->is_signed = is_signed;
138 list_add(&field->link, head);
143 int trace_define_field(struct trace_event_call *call, const char *type,
144 const char *name, int offset, int size, int is_signed,
147 struct list_head *head;
149 if (WARN_ON(!call->class))
152 head = trace_get_fields(call);
153 return __trace_define_field(head, type, name, offset, size,
154 is_signed, filter_type, 0);
156 EXPORT_SYMBOL_GPL(trace_define_field);
158 static int trace_define_field_ext(struct trace_event_call *call, const char *type,
159 const char *name, int offset, int size, int is_signed,
160 int filter_type, int len)
162 struct list_head *head;
164 if (WARN_ON(!call->class))
167 head = trace_get_fields(call);
168 return __trace_define_field(head, type, name, offset, size,
169 is_signed, filter_type, len);
172 #define __generic_field(type, item, filter_type) \
173 ret = __trace_define_field(&ftrace_generic_fields, #type, \
174 #item, 0, 0, is_signed_type(type), \
179 #define __common_field(type, item) \
180 ret = __trace_define_field(&ftrace_common_fields, #type, \
182 offsetof(typeof(ent), item), \
184 is_signed_type(type), FILTER_OTHER, 0); \
188 static int trace_define_generic_fields(void)
192 __generic_field(int, CPU, FILTER_CPU);
193 __generic_field(int, cpu, FILTER_CPU);
194 __generic_field(int, common_cpu, FILTER_CPU);
195 __generic_field(char *, COMM, FILTER_COMM);
196 __generic_field(char *, comm, FILTER_COMM);
197 __generic_field(char *, stacktrace, FILTER_STACKTRACE);
198 __generic_field(char *, STACKTRACE, FILTER_STACKTRACE);
203 static int trace_define_common_fields(void)
206 struct trace_entry ent;
208 __common_field(unsigned short, type);
209 __common_field(unsigned char, flags);
210 /* Holds both preempt_count and migrate_disable */
211 __common_field(unsigned char, preempt_count);
212 __common_field(int, pid);
217 static void trace_destroy_fields(struct trace_event_call *call)
219 struct ftrace_event_field *field, *next;
220 struct list_head *head;
222 head = trace_get_fields(call);
223 list_for_each_entry_safe(field, next, head, link) {
224 list_del(&field->link);
225 kmem_cache_free(field_cachep, field);
230 * run-time version of trace_event_get_offsets_<call>() that returns the last
231 * accessible offset of trace fields excluding __dynamic_array bytes
233 int trace_event_get_offsets(struct trace_event_call *call)
235 struct ftrace_event_field *tail;
236 struct list_head *head;
238 head = trace_get_fields(call);
240 * head->next points to the last field with the largest offset,
241 * since it was added last by trace_define_field()
243 tail = list_first_entry(head, struct ftrace_event_field, link);
244 return tail->offset + tail->size;
248 * Check if the referenced field is an array and return true,
249 * as arrays are OK to dereference.
251 static bool test_field(const char *fmt, struct trace_event_call *call)
253 struct trace_event_fields *field = call->class->fields_array;
254 const char *array_descriptor;
258 if (!(len = str_has_prefix(fmt, "REC->")))
261 for (p = fmt; *p; p++) {
262 if (!isalnum(*p) && *p != '_')
267 for (; field->type; field++) {
268 if (strncmp(field->name, fmt, len) ||
271 array_descriptor = strchr(field->type, '[');
272 /* This is an array and is OK to dereference. */
273 return array_descriptor != NULL;
279 * Examine the print fmt of the event looking for unsafe dereference
280 * pointers using %p* that could be recorded in the trace event and
281 * much later referenced after the pointer was freed. Dereferencing
282 * pointers are OK, if it is dereferenced into the event itself.
284 static void test_event_printk(struct trace_event_call *call)
286 u64 dereference_flags = 0;
288 const char *fmt, *c, *r, *a;
295 fmt = call->print_fmt;
300 for (i = 0; fmt[i]; i++) {
310 * The print fmt starts with a string that
311 * is processed first to find %p* usage,
312 * then after the first string, the print fmt
313 * contains arguments that are used to check
314 * if the dereferenced %p* usage is safe.
323 * If there was no %p* uses
326 if (!dereference_flags)
331 if (in_quote == fmt[i])
338 if (!first || !in_quote)
347 /* Find dereferencing fields */
348 switch (fmt[i + 1]) {
349 case 'B': case 'R': case 'r':
350 case 'b': case 'M': case 'm':
351 case 'I': case 'i': case 'E':
352 case 'U': case 'V': case 'N':
353 case 'a': case 'd': case 'D':
354 case 'g': case 't': case 'C':
356 if (WARN_ONCE(arg == 63,
357 "Too many args for event: %s",
358 trace_event_name(call)))
360 dereference_flags |= 1ULL << arg;
368 /* Increment arg if %*s exists. */
369 for (j = 0; fmt[i + j]; j++) {
370 if (isdigit(fmt[i + j]) ||
373 if (fmt[i + j] == '*') {
377 if ((fmt[i + j] == 's') && star)
396 if (WARN_ONCE(parens < 0,
397 "Paren mismatch for event: %s\narg='%s'\n%*s",
398 trace_event_name(call),
400 (i - start_arg) + 5, "^"))
404 if (in_quote || parens)
407 while (isspace(fmt[i]))
410 if (!(dereference_flags & (1ULL << arg)))
413 /* Find the REC-> in the argument */
414 c = strchr(fmt + i, ',');
415 r = strstr(fmt + i, "REC->");
416 if (r && (!c || r < c)) {
418 * Addresses of events on the buffer,
419 * or an array on the buffer is
421 * There's ways to fool this, but
422 * this is to catch common mistakes,
423 * not malicious code.
425 a = strchr(fmt + i, '&');
426 if ((a && (a < r)) || test_field(r, call))
427 dereference_flags &= ~(1ULL << arg);
428 } else if ((r = strstr(fmt + i, "__get_dynamic_array(")) &&
430 dereference_flags &= ~(1ULL << arg);
431 } else if ((r = strstr(fmt + i, "__get_sockaddr(")) &&
433 dereference_flags &= ~(1ULL << arg);
443 * If you triggered the below warning, the trace event reported
444 * uses an unsafe dereference pointer %p*. As the data stored
445 * at the trace event time may no longer exist when the trace
446 * event is printed, dereferencing to the original source is
447 * unsafe. The source of the dereference must be copied into the
448 * event itself, and the dereference must access the copy instead.
450 if (WARN_ON_ONCE(dereference_flags)) {
452 while (!(dereference_flags & 1)) {
453 dereference_flags >>= 1;
456 pr_warn("event %s has unsafe dereference of argument %d\n",
457 trace_event_name(call), arg);
458 pr_warn("print_fmt: %s\n", fmt);
462 int trace_event_raw_init(struct trace_event_call *call)
466 id = register_trace_event(&call->event);
470 test_event_printk(call);
474 EXPORT_SYMBOL_GPL(trace_event_raw_init);
476 bool trace_event_ignore_this_pid(struct trace_event_file *trace_file)
478 struct trace_array *tr = trace_file->tr;
479 struct trace_array_cpu *data;
480 struct trace_pid_list *no_pid_list;
481 struct trace_pid_list *pid_list;
483 pid_list = rcu_dereference_raw(tr->filtered_pids);
484 no_pid_list = rcu_dereference_raw(tr->filtered_no_pids);
486 if (!pid_list && !no_pid_list)
489 data = this_cpu_ptr(tr->array_buffer.data);
491 return data->ignore_pid;
493 EXPORT_SYMBOL_GPL(trace_event_ignore_this_pid);
495 void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer,
496 struct trace_event_file *trace_file,
499 struct trace_event_call *event_call = trace_file->event_call;
501 if ((trace_file->flags & EVENT_FILE_FL_PID_FILTER) &&
502 trace_event_ignore_this_pid(trace_file))
506 * If CONFIG_PREEMPTION is enabled, then the tracepoint itself disables
507 * preemption (adding one to the preempt_count). Since we are
508 * interested in the preempt_count at the time the tracepoint was
509 * hit, we need to subtract one to offset the increment.
511 fbuffer->trace_ctx = tracing_gen_ctx_dec();
512 fbuffer->trace_file = trace_file;
515 trace_event_buffer_lock_reserve(&fbuffer->buffer, trace_file,
516 event_call->event.type, len,
521 fbuffer->regs = NULL;
522 fbuffer->entry = ring_buffer_event_data(fbuffer->event);
523 return fbuffer->entry;
525 EXPORT_SYMBOL_GPL(trace_event_buffer_reserve);
527 int trace_event_reg(struct trace_event_call *call,
528 enum trace_reg type, void *data)
530 struct trace_event_file *file = data;
532 WARN_ON(!(call->flags & TRACE_EVENT_FL_TRACEPOINT));
534 case TRACE_REG_REGISTER:
535 return tracepoint_probe_register(call->tp,
538 case TRACE_REG_UNREGISTER:
539 tracepoint_probe_unregister(call->tp,
544 #ifdef CONFIG_PERF_EVENTS
545 case TRACE_REG_PERF_REGISTER:
546 return tracepoint_probe_register(call->tp,
547 call->class->perf_probe,
549 case TRACE_REG_PERF_UNREGISTER:
550 tracepoint_probe_unregister(call->tp,
551 call->class->perf_probe,
554 case TRACE_REG_PERF_OPEN:
555 case TRACE_REG_PERF_CLOSE:
556 case TRACE_REG_PERF_ADD:
557 case TRACE_REG_PERF_DEL:
563 EXPORT_SYMBOL_GPL(trace_event_reg);
565 void trace_event_enable_cmd_record(bool enable)
567 struct trace_event_file *file;
568 struct trace_array *tr;
570 lockdep_assert_held(&event_mutex);
572 do_for_each_event_file(tr, file) {
574 if (!(file->flags & EVENT_FILE_FL_ENABLED))
578 tracing_start_cmdline_record();
579 set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
581 tracing_stop_cmdline_record();
582 clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
584 } while_for_each_event_file();
587 void trace_event_enable_tgid_record(bool enable)
589 struct trace_event_file *file;
590 struct trace_array *tr;
592 lockdep_assert_held(&event_mutex);
594 do_for_each_event_file(tr, file) {
595 if (!(file->flags & EVENT_FILE_FL_ENABLED))
599 tracing_start_tgid_record();
600 set_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
602 tracing_stop_tgid_record();
603 clear_bit(EVENT_FILE_FL_RECORDED_TGID_BIT,
606 } while_for_each_event_file();
609 static int __ftrace_event_enable_disable(struct trace_event_file *file,
610 int enable, int soft_disable)
612 struct trace_event_call *call = file->event_call;
613 struct trace_array *tr = file->tr;
620 * When soft_disable is set and enable is cleared, the sm_ref
621 * reference counter is decremented. If it reaches 0, we want
622 * to clear the SOFT_DISABLED flag but leave the event in the
623 * state that it was. That is, if the event was enabled and
624 * SOFT_DISABLED isn't set, then do nothing. But if SOFT_DISABLED
625 * is set we do not want the event to be enabled before we
628 * When soft_disable is not set but the SOFT_MODE flag is,
629 * we do nothing. Do not disable the tracepoint, otherwise
630 * "soft enable"s (clearing the SOFT_DISABLED bit) wont work.
633 if (atomic_dec_return(&file->sm_ref) > 0)
635 disable = file->flags & EVENT_FILE_FL_SOFT_DISABLED;
636 clear_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
637 /* Disable use of trace_buffered_event */
638 trace_buffered_event_disable();
640 disable = !(file->flags & EVENT_FILE_FL_SOFT_MODE);
642 if (disable && (file->flags & EVENT_FILE_FL_ENABLED)) {
643 clear_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
644 if (file->flags & EVENT_FILE_FL_RECORDED_CMD) {
645 tracing_stop_cmdline_record();
646 clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
649 if (file->flags & EVENT_FILE_FL_RECORDED_TGID) {
650 tracing_stop_tgid_record();
651 clear_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
654 call->class->reg(call, TRACE_REG_UNREGISTER, file);
656 /* If in SOFT_MODE, just set the SOFT_DISABLE_BIT, else clear it */
657 if (file->flags & EVENT_FILE_FL_SOFT_MODE)
658 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
660 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
664 * When soft_disable is set and enable is set, we want to
665 * register the tracepoint for the event, but leave the event
666 * as is. That means, if the event was already enabled, we do
667 * nothing (but set SOFT_MODE). If the event is disabled, we
668 * set SOFT_DISABLED before enabling the event tracepoint, so
669 * it still seems to be disabled.
672 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
674 if (atomic_inc_return(&file->sm_ref) > 1)
676 set_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
677 /* Enable use of trace_buffered_event */
678 trace_buffered_event_enable();
681 if (!(file->flags & EVENT_FILE_FL_ENABLED)) {
682 bool cmd = false, tgid = false;
684 /* Keep the event disabled, when going to SOFT_MODE. */
686 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
688 if (tr->trace_flags & TRACE_ITER_RECORD_CMD) {
690 tracing_start_cmdline_record();
691 set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
694 if (tr->trace_flags & TRACE_ITER_RECORD_TGID) {
696 tracing_start_tgid_record();
697 set_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
700 ret = call->class->reg(call, TRACE_REG_REGISTER, file);
703 tracing_stop_cmdline_record();
705 tracing_stop_tgid_record();
706 pr_info("event trace: Could not enable event "
707 "%s\n", trace_event_name(call));
710 set_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
712 /* WAS_ENABLED gets set but never cleared. */
713 set_bit(EVENT_FILE_FL_WAS_ENABLED_BIT, &file->flags);
721 int trace_event_enable_disable(struct trace_event_file *file,
722 int enable, int soft_disable)
724 return __ftrace_event_enable_disable(file, enable, soft_disable);
727 static int ftrace_event_enable_disable(struct trace_event_file *file,
730 return __ftrace_event_enable_disable(file, enable, 0);
733 static void ftrace_clear_events(struct trace_array *tr)
735 struct trace_event_file *file;
737 mutex_lock(&event_mutex);
738 list_for_each_entry(file, &tr->events, list) {
739 ftrace_event_enable_disable(file, 0);
741 mutex_unlock(&event_mutex);
745 event_filter_pid_sched_process_exit(void *data, struct task_struct *task)
747 struct trace_pid_list *pid_list;
748 struct trace_array *tr = data;
750 pid_list = rcu_dereference_raw(tr->filtered_pids);
751 trace_filter_add_remove_task(pid_list, NULL, task);
753 pid_list = rcu_dereference_raw(tr->filtered_no_pids);
754 trace_filter_add_remove_task(pid_list, NULL, task);
758 event_filter_pid_sched_process_fork(void *data,
759 struct task_struct *self,
760 struct task_struct *task)
762 struct trace_pid_list *pid_list;
763 struct trace_array *tr = data;
765 pid_list = rcu_dereference_sched(tr->filtered_pids);
766 trace_filter_add_remove_task(pid_list, self, task);
768 pid_list = rcu_dereference_sched(tr->filtered_no_pids);
769 trace_filter_add_remove_task(pid_list, self, task);
772 void trace_event_follow_fork(struct trace_array *tr, bool enable)
775 register_trace_prio_sched_process_fork(event_filter_pid_sched_process_fork,
777 register_trace_prio_sched_process_free(event_filter_pid_sched_process_exit,
780 unregister_trace_sched_process_fork(event_filter_pid_sched_process_fork,
782 unregister_trace_sched_process_free(event_filter_pid_sched_process_exit,
788 event_filter_pid_sched_switch_probe_pre(void *data, bool preempt,
789 struct task_struct *prev,
790 struct task_struct *next,
791 unsigned int prev_state)
793 struct trace_array *tr = data;
794 struct trace_pid_list *no_pid_list;
795 struct trace_pid_list *pid_list;
798 pid_list = rcu_dereference_sched(tr->filtered_pids);
799 no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
802 * Sched switch is funny, as we only want to ignore it
803 * in the notrace case if both prev and next should be ignored.
805 ret = trace_ignore_this_task(NULL, no_pid_list, prev) &&
806 trace_ignore_this_task(NULL, no_pid_list, next);
808 this_cpu_write(tr->array_buffer.data->ignore_pid, ret ||
809 (trace_ignore_this_task(pid_list, NULL, prev) &&
810 trace_ignore_this_task(pid_list, NULL, next)));
814 event_filter_pid_sched_switch_probe_post(void *data, bool preempt,
815 struct task_struct *prev,
816 struct task_struct *next,
817 unsigned int prev_state)
819 struct trace_array *tr = data;
820 struct trace_pid_list *no_pid_list;
821 struct trace_pid_list *pid_list;
823 pid_list = rcu_dereference_sched(tr->filtered_pids);
824 no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
826 this_cpu_write(tr->array_buffer.data->ignore_pid,
827 trace_ignore_this_task(pid_list, no_pid_list, next));
831 event_filter_pid_sched_wakeup_probe_pre(void *data, struct task_struct *task)
833 struct trace_array *tr = data;
834 struct trace_pid_list *no_pid_list;
835 struct trace_pid_list *pid_list;
837 /* Nothing to do if we are already tracing */
838 if (!this_cpu_read(tr->array_buffer.data->ignore_pid))
841 pid_list = rcu_dereference_sched(tr->filtered_pids);
842 no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
844 this_cpu_write(tr->array_buffer.data->ignore_pid,
845 trace_ignore_this_task(pid_list, no_pid_list, task));
849 event_filter_pid_sched_wakeup_probe_post(void *data, struct task_struct *task)
851 struct trace_array *tr = data;
852 struct trace_pid_list *no_pid_list;
853 struct trace_pid_list *pid_list;
855 /* Nothing to do if we are not tracing */
856 if (this_cpu_read(tr->array_buffer.data->ignore_pid))
859 pid_list = rcu_dereference_sched(tr->filtered_pids);
860 no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
862 /* Set tracing if current is enabled */
863 this_cpu_write(tr->array_buffer.data->ignore_pid,
864 trace_ignore_this_task(pid_list, no_pid_list, current));
867 static void unregister_pid_events(struct trace_array *tr)
869 unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_pre, tr);
870 unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_post, tr);
872 unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre, tr);
873 unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_post, tr);
875 unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre, tr);
876 unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post, tr);
878 unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_pre, tr);
879 unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_post, tr);
882 static void __ftrace_clear_event_pids(struct trace_array *tr, int type)
884 struct trace_pid_list *pid_list;
885 struct trace_pid_list *no_pid_list;
886 struct trace_event_file *file;
889 pid_list = rcu_dereference_protected(tr->filtered_pids,
890 lockdep_is_held(&event_mutex));
891 no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
892 lockdep_is_held(&event_mutex));
894 /* Make sure there's something to do */
895 if (!pid_type_enabled(type, pid_list, no_pid_list))
898 if (!still_need_pid_events(type, pid_list, no_pid_list)) {
899 unregister_pid_events(tr);
901 list_for_each_entry(file, &tr->events, list) {
902 clear_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
905 for_each_possible_cpu(cpu)
906 per_cpu_ptr(tr->array_buffer.data, cpu)->ignore_pid = false;
909 if (type & TRACE_PIDS)
910 rcu_assign_pointer(tr->filtered_pids, NULL);
912 if (type & TRACE_NO_PIDS)
913 rcu_assign_pointer(tr->filtered_no_pids, NULL);
915 /* Wait till all users are no longer using pid filtering */
916 tracepoint_synchronize_unregister();
918 if ((type & TRACE_PIDS) && pid_list)
919 trace_pid_list_free(pid_list);
921 if ((type & TRACE_NO_PIDS) && no_pid_list)
922 trace_pid_list_free(no_pid_list);
925 static void ftrace_clear_event_pids(struct trace_array *tr, int type)
927 mutex_lock(&event_mutex);
928 __ftrace_clear_event_pids(tr, type);
929 mutex_unlock(&event_mutex);
932 static void __put_system(struct event_subsystem *system)
934 struct event_filter *filter = system->filter;
936 WARN_ON_ONCE(system_refcount(system) == 0);
937 if (system_refcount_dec(system))
940 list_del(&system->list);
943 kfree(filter->filter_string);
946 kfree_const(system->name);
950 static void __get_system(struct event_subsystem *system)
952 WARN_ON_ONCE(system_refcount(system) == 0);
953 system_refcount_inc(system);
956 static void __get_system_dir(struct trace_subsystem_dir *dir)
958 WARN_ON_ONCE(dir->ref_count == 0);
960 __get_system(dir->subsystem);
963 static void __put_system_dir(struct trace_subsystem_dir *dir)
965 WARN_ON_ONCE(dir->ref_count == 0);
966 /* If the subsystem is about to be freed, the dir must be too */
967 WARN_ON_ONCE(system_refcount(dir->subsystem) == 1 && dir->ref_count != 1);
969 __put_system(dir->subsystem);
970 if (!--dir->ref_count)
974 static void put_system(struct trace_subsystem_dir *dir)
976 mutex_lock(&event_mutex);
977 __put_system_dir(dir);
978 mutex_unlock(&event_mutex);
981 static void remove_subsystem(struct trace_subsystem_dir *dir)
986 if (!--dir->nr_events) {
987 eventfs_remove(dir->ef);
988 list_del(&dir->list);
989 __put_system_dir(dir);
993 void event_file_get(struct trace_event_file *file)
995 atomic_inc(&file->ref);
998 void event_file_put(struct trace_event_file *file)
1000 if (WARN_ON_ONCE(!atomic_read(&file->ref))) {
1001 if (file->flags & EVENT_FILE_FL_FREED)
1002 kmem_cache_free(file_cachep, file);
1006 if (atomic_dec_and_test(&file->ref)) {
1007 /* Count should only go to zero when it is freed */
1008 if (WARN_ON_ONCE(!(file->flags & EVENT_FILE_FL_FREED)))
1010 kmem_cache_free(file_cachep, file);
1014 static void remove_event_file_dir(struct trace_event_file *file)
1016 eventfs_remove(file->ef);
1017 list_del(&file->list);
1018 remove_subsystem(file->system);
1019 free_event_filter(file->filter);
1020 file->flags |= EVENT_FILE_FL_FREED;
1021 event_file_put(file);
1025 * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
1028 __ftrace_set_clr_event_nolock(struct trace_array *tr, const char *match,
1029 const char *sub, const char *event, int set)
1031 struct trace_event_file *file;
1032 struct trace_event_call *call;
1037 list_for_each_entry(file, &tr->events, list) {
1039 call = file->event_call;
1040 name = trace_event_name(call);
1042 if (!name || !call->class || !call->class->reg)
1045 if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
1049 strcmp(match, name) != 0 &&
1050 strcmp(match, call->class->system) != 0)
1053 if (sub && strcmp(sub, call->class->system) != 0)
1056 if (event && strcmp(event, name) != 0)
1059 ret = ftrace_event_enable_disable(file, set);
1062 * Save the first error and return that. Some events
1063 * may still have been enabled, but let the user
1064 * know that something went wrong.
1075 static int __ftrace_set_clr_event(struct trace_array *tr, const char *match,
1076 const char *sub, const char *event, int set)
1080 mutex_lock(&event_mutex);
1081 ret = __ftrace_set_clr_event_nolock(tr, match, sub, event, set);
1082 mutex_unlock(&event_mutex);
1087 int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set)
1089 char *event = NULL, *sub = NULL, *match;
1095 * The buf format can be <subsystem>:<event-name>
1096 * *:<event-name> means any event by that name.
1097 * :<event-name> is the same.
1099 * <subsystem>:* means all events in that subsystem
1100 * <subsystem>: means the same.
1102 * <name> (no ':') means all events in a subsystem with
1103 * the name <name> or any event that matches <name>
1106 match = strsep(&buf, ":");
1112 if (!strlen(sub) || strcmp(sub, "*") == 0)
1114 if (!strlen(event) || strcmp(event, "*") == 0)
1118 ret = __ftrace_set_clr_event(tr, match, sub, event, set);
1120 /* Put back the colon to allow this to be called again */
1128 * trace_set_clr_event - enable or disable an event
1129 * @system: system name to match (NULL for any system)
1130 * @event: event name to match (NULL for all events, within system)
1131 * @set: 1 to enable, 0 to disable
1133 * This is a way for other parts of the kernel to enable or disable
1136 * Returns 0 on success, -EINVAL if the parameters do not match any
1137 * registered events.
1139 int trace_set_clr_event(const char *system, const char *event, int set)
1141 struct trace_array *tr = top_trace_array();
1146 return __ftrace_set_clr_event(tr, NULL, system, event, set);
1148 EXPORT_SYMBOL_GPL(trace_set_clr_event);
1151 * trace_array_set_clr_event - enable or disable an event for a trace array.
1152 * @tr: concerned trace array.
1153 * @system: system name to match (NULL for any system)
1154 * @event: event name to match (NULL for all events, within system)
1155 * @enable: true to enable, false to disable
1157 * This is a way for other parts of the kernel to enable or disable
1160 * Returns 0 on success, -EINVAL if the parameters do not match any
1161 * registered events.
1163 int trace_array_set_clr_event(struct trace_array *tr, const char *system,
1164 const char *event, bool enable)
1171 set = (enable == true) ? 1 : 0;
1172 return __ftrace_set_clr_event(tr, NULL, system, event, set);
1174 EXPORT_SYMBOL_GPL(trace_array_set_clr_event);
1176 /* 128 should be much more than enough */
1177 #define EVENT_BUF_SIZE 127
1180 ftrace_event_write(struct file *file, const char __user *ubuf,
1181 size_t cnt, loff_t *ppos)
1183 struct trace_parser parser;
1184 struct seq_file *m = file->private_data;
1185 struct trace_array *tr = m->private;
1191 ret = tracing_update_buffers();
1195 if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
1198 read = trace_get_user(&parser, ubuf, cnt, ppos);
1200 if (read >= 0 && trace_parser_loaded((&parser))) {
1203 if (*parser.buffer == '!')
1206 ret = ftrace_set_clr_event(tr, parser.buffer + !set, set);
1214 trace_parser_put(&parser);
1220 t_next(struct seq_file *m, void *v, loff_t *pos)
1222 struct trace_event_file *file = v;
1223 struct trace_event_call *call;
1224 struct trace_array *tr = m->private;
1228 list_for_each_entry_continue(file, &tr->events, list) {
1229 call = file->event_call;
1231 * The ftrace subsystem is for showing formats only.
1232 * They can not be enabled or disabled via the event files.
1234 if (call->class && call->class->reg &&
1235 !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
1242 static void *t_start(struct seq_file *m, loff_t *pos)
1244 struct trace_event_file *file;
1245 struct trace_array *tr = m->private;
1248 mutex_lock(&event_mutex);
1250 file = list_entry(&tr->events, struct trace_event_file, list);
1251 for (l = 0; l <= *pos; ) {
1252 file = t_next(m, file, &l);
1260 s_next(struct seq_file *m, void *v, loff_t *pos)
1262 struct trace_event_file *file = v;
1263 struct trace_array *tr = m->private;
1267 list_for_each_entry_continue(file, &tr->events, list) {
1268 if (file->flags & EVENT_FILE_FL_ENABLED)
1275 static void *s_start(struct seq_file *m, loff_t *pos)
1277 struct trace_event_file *file;
1278 struct trace_array *tr = m->private;
1281 mutex_lock(&event_mutex);
1283 file = list_entry(&tr->events, struct trace_event_file, list);
1284 for (l = 0; l <= *pos; ) {
1285 file = s_next(m, file, &l);
1292 static int t_show(struct seq_file *m, void *v)
1294 struct trace_event_file *file = v;
1295 struct trace_event_call *call = file->event_call;
1297 if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
1298 seq_printf(m, "%s:", call->class->system);
1299 seq_printf(m, "%s\n", trace_event_name(call));
1304 static void t_stop(struct seq_file *m, void *p)
1306 mutex_unlock(&event_mutex);
1310 __next(struct seq_file *m, void *v, loff_t *pos, int type)
1312 struct trace_array *tr = m->private;
1313 struct trace_pid_list *pid_list;
1315 if (type == TRACE_PIDS)
1316 pid_list = rcu_dereference_sched(tr->filtered_pids);
1318 pid_list = rcu_dereference_sched(tr->filtered_no_pids);
1320 return trace_pid_next(pid_list, v, pos);
1324 p_next(struct seq_file *m, void *v, loff_t *pos)
1326 return __next(m, v, pos, TRACE_PIDS);
1330 np_next(struct seq_file *m, void *v, loff_t *pos)
1332 return __next(m, v, pos, TRACE_NO_PIDS);
1335 static void *__start(struct seq_file *m, loff_t *pos, int type)
1338 struct trace_pid_list *pid_list;
1339 struct trace_array *tr = m->private;
1342 * Grab the mutex, to keep calls to p_next() having the same
1343 * tr->filtered_pids as p_start() has.
1344 * If we just passed the tr->filtered_pids around, then RCU would
1345 * have been enough, but doing that makes things more complex.
1347 mutex_lock(&event_mutex);
1348 rcu_read_lock_sched();
1350 if (type == TRACE_PIDS)
1351 pid_list = rcu_dereference_sched(tr->filtered_pids);
1353 pid_list = rcu_dereference_sched(tr->filtered_no_pids);
1358 return trace_pid_start(pid_list, pos);
1361 static void *p_start(struct seq_file *m, loff_t *pos)
1364 return __start(m, pos, TRACE_PIDS);
1367 static void *np_start(struct seq_file *m, loff_t *pos)
1370 return __start(m, pos, TRACE_NO_PIDS);
1373 static void p_stop(struct seq_file *m, void *p)
1376 rcu_read_unlock_sched();
1377 mutex_unlock(&event_mutex);
1381 event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1384 struct trace_event_file *file;
1385 unsigned long flags;
1388 mutex_lock(&event_mutex);
1389 file = event_file_data(filp);
1391 flags = file->flags;
1392 mutex_unlock(&event_mutex);
1394 if (!file || flags & EVENT_FILE_FL_FREED)
1397 if (flags & EVENT_FILE_FL_ENABLED &&
1398 !(flags & EVENT_FILE_FL_SOFT_DISABLED))
1401 if (flags & EVENT_FILE_FL_SOFT_DISABLED ||
1402 flags & EVENT_FILE_FL_SOFT_MODE)
1407 return simple_read_from_buffer(ubuf, cnt, ppos, buf, strlen(buf));
1411 event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1414 struct trace_event_file *file;
1418 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1422 ret = tracing_update_buffers();
1430 mutex_lock(&event_mutex);
1431 file = event_file_data(filp);
1432 if (likely(file && !(file->flags & EVENT_FILE_FL_FREED)))
1433 ret = ftrace_event_enable_disable(file, val);
1434 mutex_unlock(&event_mutex);
1443 return ret ? ret : cnt;
1447 system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1450 const char set_to_char[4] = { '?', '0', '1', 'X' };
1451 struct trace_subsystem_dir *dir = filp->private_data;
1452 struct event_subsystem *system = dir->subsystem;
1453 struct trace_event_call *call;
1454 struct trace_event_file *file;
1455 struct trace_array *tr = dir->tr;
1460 mutex_lock(&event_mutex);
1461 list_for_each_entry(file, &tr->events, list) {
1462 call = file->event_call;
1463 if ((call->flags & TRACE_EVENT_FL_IGNORE_ENABLE) ||
1464 !trace_event_name(call) || !call->class || !call->class->reg)
1467 if (system && strcmp(call->class->system, system->name) != 0)
1471 * We need to find out if all the events are set
1472 * or if all events or cleared, or if we have
1475 set |= (1 << !!(file->flags & EVENT_FILE_FL_ENABLED));
1478 * If we have a mixture, no need to look further.
1483 mutex_unlock(&event_mutex);
1485 buf[0] = set_to_char[set];
1488 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
1494 system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1497 struct trace_subsystem_dir *dir = filp->private_data;
1498 struct event_subsystem *system = dir->subsystem;
1499 const char *name = NULL;
1503 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1507 ret = tracing_update_buffers();
1511 if (val != 0 && val != 1)
1515 * Opening of "enable" adds a ref count to system,
1516 * so the name is safe to use.
1519 name = system->name;
1521 ret = __ftrace_set_clr_event(dir->tr, NULL, name, NULL, val);
1535 FORMAT_FIELD_SEPERATOR = 2,
1536 FORMAT_PRINTFMT = 3,
1539 static void *f_next(struct seq_file *m, void *v, loff_t *pos)
1541 struct trace_event_call *call = event_file_data(m->private);
1542 struct list_head *common_head = &ftrace_common_fields;
1543 struct list_head *head = trace_get_fields(call);
1544 struct list_head *node = v;
1548 switch ((unsigned long)v) {
1553 case FORMAT_FIELD_SEPERATOR:
1557 case FORMAT_PRINTFMT:
1563 if (node == common_head)
1564 return (void *)FORMAT_FIELD_SEPERATOR;
1565 else if (node == head)
1566 return (void *)FORMAT_PRINTFMT;
1571 static int f_show(struct seq_file *m, void *v)
1573 struct trace_event_call *call = event_file_data(m->private);
1574 struct ftrace_event_field *field;
1575 const char *array_descriptor;
1577 switch ((unsigned long)v) {
1579 seq_printf(m, "name: %s\n", trace_event_name(call));
1580 seq_printf(m, "ID: %d\n", call->event.type);
1581 seq_puts(m, "format:\n");
1584 case FORMAT_FIELD_SEPERATOR:
1588 case FORMAT_PRINTFMT:
1589 seq_printf(m, "\nprint fmt: %s\n",
1594 field = list_entry(v, struct ftrace_event_field, link);
1596 * Smartly shows the array type(except dynamic array).
1599 * If TYPE := TYPE[LEN], it is shown:
1600 * field:TYPE VAR[LEN]
1602 array_descriptor = strchr(field->type, '[');
1604 if (str_has_prefix(field->type, "__data_loc"))
1605 array_descriptor = NULL;
1607 if (!array_descriptor)
1608 seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1609 field->type, field->name, field->offset,
1610 field->size, !!field->is_signed);
1611 else if (field->len)
1612 seq_printf(m, "\tfield:%.*s %s[%d];\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1613 (int)(array_descriptor - field->type),
1614 field->type, field->name,
1615 field->len, field->offset,
1616 field->size, !!field->is_signed);
1618 seq_printf(m, "\tfield:%.*s %s[];\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1619 (int)(array_descriptor - field->type),
1620 field->type, field->name,
1621 field->offset, field->size, !!field->is_signed);
1626 static void *f_start(struct seq_file *m, loff_t *pos)
1628 void *p = (void *)FORMAT_HEADER;
1631 /* ->stop() is called even if ->start() fails */
1632 mutex_lock(&event_mutex);
1633 if (!event_file_data(m->private))
1634 return ERR_PTR(-ENODEV);
1636 while (l < *pos && p)
1637 p = f_next(m, p, &l);
1642 static void f_stop(struct seq_file *m, void *p)
1644 mutex_unlock(&event_mutex);
1647 static const struct seq_operations trace_format_seq_ops = {
1654 static int trace_format_open(struct inode *inode, struct file *file)
1659 /* Do we want to hide event format files on tracefs lockdown? */
1661 ret = seq_open(file, &trace_format_seq_ops);
1665 m = file->private_data;
1672 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1674 int id = (long)event_file_data(filp);
1681 len = sprintf(buf, "%d\n", id);
1683 return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
1687 event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1690 struct trace_event_file *file;
1691 struct trace_seq *s;
1697 s = kmalloc(sizeof(*s), GFP_KERNEL);
1704 mutex_lock(&event_mutex);
1705 file = event_file_data(filp);
1706 if (file && !(file->flags & EVENT_FILE_FL_FREED))
1707 print_event_filter(file, s);
1708 mutex_unlock(&event_mutex);
1711 r = simple_read_from_buffer(ubuf, cnt, ppos,
1712 s->buffer, trace_seq_used(s));
1720 event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1723 struct trace_event_file *file;
1727 if (cnt >= PAGE_SIZE)
1730 buf = memdup_user_nul(ubuf, cnt);
1732 return PTR_ERR(buf);
1734 mutex_lock(&event_mutex);
1735 file = event_file_data(filp);
1737 err = apply_event_filter(file, buf);
1738 mutex_unlock(&event_mutex);
1749 static LIST_HEAD(event_subsystems);
1751 static int subsystem_open(struct inode *inode, struct file *filp)
1753 struct trace_subsystem_dir *dir = NULL, *iter_dir;
1754 struct trace_array *tr = NULL, *iter_tr;
1755 struct event_subsystem *system = NULL;
1758 if (tracing_is_disabled())
1761 /* Make sure the system still exists */
1762 mutex_lock(&event_mutex);
1763 mutex_lock(&trace_types_lock);
1764 list_for_each_entry(iter_tr, &ftrace_trace_arrays, list) {
1765 list_for_each_entry(iter_dir, &iter_tr->systems, list) {
1766 if (iter_dir == inode->i_private) {
1767 /* Don't open systems with no events */
1770 if (dir->nr_events) {
1771 __get_system_dir(dir);
1772 system = dir->subsystem;
1779 mutex_unlock(&trace_types_lock);
1780 mutex_unlock(&event_mutex);
1785 /* Still need to increment the ref count of the system */
1786 if (trace_array_get(tr) < 0) {
1791 ret = tracing_open_generic(inode, filp);
1793 trace_array_put(tr);
1800 static int system_tr_open(struct inode *inode, struct file *filp)
1802 struct trace_subsystem_dir *dir;
1803 struct trace_array *tr = inode->i_private;
1806 /* Make a temporary dir that has no system but points to tr */
1807 dir = kzalloc(sizeof(*dir), GFP_KERNEL);
1811 ret = tracing_open_generic_tr(inode, filp);
1817 filp->private_data = dir;
1822 static int subsystem_release(struct inode *inode, struct file *file)
1824 struct trace_subsystem_dir *dir = file->private_data;
1826 trace_array_put(dir->tr);
1829 * If dir->subsystem is NULL, then this is a temporary
1830 * descriptor that was made for a trace_array to enable
1842 subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1845 struct trace_subsystem_dir *dir = filp->private_data;
1846 struct event_subsystem *system = dir->subsystem;
1847 struct trace_seq *s;
1853 s = kmalloc(sizeof(*s), GFP_KERNEL);
1859 print_subsystem_event_filter(system, s);
1860 r = simple_read_from_buffer(ubuf, cnt, ppos,
1861 s->buffer, trace_seq_used(s));
1869 subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1872 struct trace_subsystem_dir *dir = filp->private_data;
1876 if (cnt >= PAGE_SIZE)
1879 buf = memdup_user_nul(ubuf, cnt);
1881 return PTR_ERR(buf);
1883 err = apply_subsystem_event_filter(dir, buf);
1894 show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1896 int (*func)(struct trace_seq *s) = filp->private_data;
1897 struct trace_seq *s;
1903 s = kmalloc(sizeof(*s), GFP_KERNEL);
1910 r = simple_read_from_buffer(ubuf, cnt, ppos,
1911 s->buffer, trace_seq_used(s));
1918 static void ignore_task_cpu(void *data)
1920 struct trace_array *tr = data;
1921 struct trace_pid_list *pid_list;
1922 struct trace_pid_list *no_pid_list;
1925 * This function is called by on_each_cpu() while the
1926 * event_mutex is held.
1928 pid_list = rcu_dereference_protected(tr->filtered_pids,
1929 mutex_is_locked(&event_mutex));
1930 no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
1931 mutex_is_locked(&event_mutex));
1933 this_cpu_write(tr->array_buffer.data->ignore_pid,
1934 trace_ignore_this_task(pid_list, no_pid_list, current));
1937 static void register_pid_events(struct trace_array *tr)
1940 * Register a probe that is called before all other probes
1941 * to set ignore_pid if next or prev do not match.
1942 * Register a probe this is called after all other probes
1943 * to only keep ignore_pid set if next pid matches.
1945 register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_pre,
1947 register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_post,
1950 register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre,
1952 register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_post,
1955 register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre,
1957 register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post,
1960 register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_pre,
1962 register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_post,
1967 event_pid_write(struct file *filp, const char __user *ubuf,
1968 size_t cnt, loff_t *ppos, int type)
1970 struct seq_file *m = filp->private_data;
1971 struct trace_array *tr = m->private;
1972 struct trace_pid_list *filtered_pids = NULL;
1973 struct trace_pid_list *other_pids = NULL;
1974 struct trace_pid_list *pid_list;
1975 struct trace_event_file *file;
1981 ret = tracing_update_buffers();
1985 mutex_lock(&event_mutex);
1987 if (type == TRACE_PIDS) {
1988 filtered_pids = rcu_dereference_protected(tr->filtered_pids,
1989 lockdep_is_held(&event_mutex));
1990 other_pids = rcu_dereference_protected(tr->filtered_no_pids,
1991 lockdep_is_held(&event_mutex));
1993 filtered_pids = rcu_dereference_protected(tr->filtered_no_pids,
1994 lockdep_is_held(&event_mutex));
1995 other_pids = rcu_dereference_protected(tr->filtered_pids,
1996 lockdep_is_held(&event_mutex));
1999 ret = trace_pid_write(filtered_pids, &pid_list, ubuf, cnt);
2003 if (type == TRACE_PIDS)
2004 rcu_assign_pointer(tr->filtered_pids, pid_list);
2006 rcu_assign_pointer(tr->filtered_no_pids, pid_list);
2008 list_for_each_entry(file, &tr->events, list) {
2009 set_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
2012 if (filtered_pids) {
2013 tracepoint_synchronize_unregister();
2014 trace_pid_list_free(filtered_pids);
2015 } else if (pid_list && !other_pids) {
2016 register_pid_events(tr);
2020 * Ignoring of pids is done at task switch. But we have to
2021 * check for those tasks that are currently running.
2022 * Always do this in case a pid was appended or removed.
2024 on_each_cpu(ignore_task_cpu, tr, 1);
2027 mutex_unlock(&event_mutex);
2036 ftrace_event_pid_write(struct file *filp, const char __user *ubuf,
2037 size_t cnt, loff_t *ppos)
2039 return event_pid_write(filp, ubuf, cnt, ppos, TRACE_PIDS);
2043 ftrace_event_npid_write(struct file *filp, const char __user *ubuf,
2044 size_t cnt, loff_t *ppos)
2046 return event_pid_write(filp, ubuf, cnt, ppos, TRACE_NO_PIDS);
2049 static int ftrace_event_avail_open(struct inode *inode, struct file *file);
2050 static int ftrace_event_set_open(struct inode *inode, struct file *file);
2051 static int ftrace_event_set_pid_open(struct inode *inode, struct file *file);
2052 static int ftrace_event_set_npid_open(struct inode *inode, struct file *file);
2053 static int ftrace_event_release(struct inode *inode, struct file *file);
2055 static const struct seq_operations show_event_seq_ops = {
2062 static const struct seq_operations show_set_event_seq_ops = {
2069 static const struct seq_operations show_set_pid_seq_ops = {
2072 .show = trace_pid_show,
2076 static const struct seq_operations show_set_no_pid_seq_ops = {
2079 .show = trace_pid_show,
2083 static const struct file_operations ftrace_avail_fops = {
2084 .open = ftrace_event_avail_open,
2086 .llseek = seq_lseek,
2087 .release = seq_release,
2090 static const struct file_operations ftrace_set_event_fops = {
2091 .open = ftrace_event_set_open,
2093 .write = ftrace_event_write,
2094 .llseek = seq_lseek,
2095 .release = ftrace_event_release,
2098 static const struct file_operations ftrace_set_event_pid_fops = {
2099 .open = ftrace_event_set_pid_open,
2101 .write = ftrace_event_pid_write,
2102 .llseek = seq_lseek,
2103 .release = ftrace_event_release,
2106 static const struct file_operations ftrace_set_event_notrace_pid_fops = {
2107 .open = ftrace_event_set_npid_open,
2109 .write = ftrace_event_npid_write,
2110 .llseek = seq_lseek,
2111 .release = ftrace_event_release,
2114 static const struct file_operations ftrace_enable_fops = {
2115 .open = tracing_open_file_tr,
2116 .read = event_enable_read,
2117 .write = event_enable_write,
2118 .release = tracing_release_file_tr,
2119 .llseek = default_llseek,
2122 static const struct file_operations ftrace_event_format_fops = {
2123 .open = trace_format_open,
2125 .llseek = seq_lseek,
2126 .release = seq_release,
2129 static const struct file_operations ftrace_event_id_fops = {
2130 .read = event_id_read,
2131 .llseek = default_llseek,
2134 static const struct file_operations ftrace_event_filter_fops = {
2135 .open = tracing_open_file_tr,
2136 .read = event_filter_read,
2137 .write = event_filter_write,
2138 .release = tracing_release_file_tr,
2139 .llseek = default_llseek,
2142 static const struct file_operations ftrace_subsystem_filter_fops = {
2143 .open = subsystem_open,
2144 .read = subsystem_filter_read,
2145 .write = subsystem_filter_write,
2146 .llseek = default_llseek,
2147 .release = subsystem_release,
2150 static const struct file_operations ftrace_system_enable_fops = {
2151 .open = subsystem_open,
2152 .read = system_enable_read,
2153 .write = system_enable_write,
2154 .llseek = default_llseek,
2155 .release = subsystem_release,
2158 static const struct file_operations ftrace_tr_enable_fops = {
2159 .open = system_tr_open,
2160 .read = system_enable_read,
2161 .write = system_enable_write,
2162 .llseek = default_llseek,
2163 .release = subsystem_release,
2166 static const struct file_operations ftrace_show_header_fops = {
2167 .open = tracing_open_generic,
2168 .read = show_header,
2169 .llseek = default_llseek,
2173 ftrace_event_open(struct inode *inode, struct file *file,
2174 const struct seq_operations *seq_ops)
2179 ret = security_locked_down(LOCKDOWN_TRACEFS);
2183 ret = seq_open(file, seq_ops);
2186 m = file->private_data;
2187 /* copy tr over to seq ops */
2188 m->private = inode->i_private;
2193 static int ftrace_event_release(struct inode *inode, struct file *file)
2195 struct trace_array *tr = inode->i_private;
2197 trace_array_put(tr);
2199 return seq_release(inode, file);
2203 ftrace_event_avail_open(struct inode *inode, struct file *file)
2205 const struct seq_operations *seq_ops = &show_event_seq_ops;
2207 /* Checks for tracefs lockdown */
2208 return ftrace_event_open(inode, file, seq_ops);
2212 ftrace_event_set_open(struct inode *inode, struct file *file)
2214 const struct seq_operations *seq_ops = &show_set_event_seq_ops;
2215 struct trace_array *tr = inode->i_private;
2218 ret = tracing_check_open_get_tr(tr);
2222 if ((file->f_mode & FMODE_WRITE) &&
2223 (file->f_flags & O_TRUNC))
2224 ftrace_clear_events(tr);
2226 ret = ftrace_event_open(inode, file, seq_ops);
2228 trace_array_put(tr);
2233 ftrace_event_set_pid_open(struct inode *inode, struct file *file)
2235 const struct seq_operations *seq_ops = &show_set_pid_seq_ops;
2236 struct trace_array *tr = inode->i_private;
2239 ret = tracing_check_open_get_tr(tr);
2243 if ((file->f_mode & FMODE_WRITE) &&
2244 (file->f_flags & O_TRUNC))
2245 ftrace_clear_event_pids(tr, TRACE_PIDS);
2247 ret = ftrace_event_open(inode, file, seq_ops);
2249 trace_array_put(tr);
2254 ftrace_event_set_npid_open(struct inode *inode, struct file *file)
2256 const struct seq_operations *seq_ops = &show_set_no_pid_seq_ops;
2257 struct trace_array *tr = inode->i_private;
2260 ret = tracing_check_open_get_tr(tr);
2264 if ((file->f_mode & FMODE_WRITE) &&
2265 (file->f_flags & O_TRUNC))
2266 ftrace_clear_event_pids(tr, TRACE_NO_PIDS);
2268 ret = ftrace_event_open(inode, file, seq_ops);
2270 trace_array_put(tr);
2274 static struct event_subsystem *
2275 create_new_subsystem(const char *name)
2277 struct event_subsystem *system;
2279 /* need to create new entry */
2280 system = kmalloc(sizeof(*system), GFP_KERNEL);
2284 system->ref_count = 1;
2286 /* Only allocate if dynamic (kprobes and modules) */
2287 system->name = kstrdup_const(name, GFP_KERNEL);
2291 system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
2292 if (!system->filter)
2295 list_add(&system->list, &event_subsystems);
2300 kfree_const(system->name);
2305 static struct eventfs_file *
2306 event_subsystem_dir(struct trace_array *tr, const char *name,
2307 struct trace_event_file *file, struct dentry *parent)
2309 struct event_subsystem *system, *iter;
2310 struct trace_subsystem_dir *dir;
2311 struct eventfs_file *ef;
2314 /* First see if we did not already create this dir */
2315 list_for_each_entry(dir, &tr->systems, list) {
2316 system = dir->subsystem;
2317 if (strcmp(system->name, name) == 0) {
2324 /* Now see if the system itself exists. */
2326 list_for_each_entry(iter, &event_subsystems, list) {
2327 if (strcmp(iter->name, name) == 0) {
2333 dir = kmalloc(sizeof(*dir), GFP_KERNEL);
2338 system = create_new_subsystem(name);
2342 __get_system(system);
2344 ef = eventfs_add_subsystem_dir(name, parent);
2346 pr_warn("Failed to create system directory %s\n", name);
2347 __put_system(system);
2355 dir->subsystem = system;
2358 /* the ftrace system is special, do not create enable or filter files */
2359 if (strcmp(name, "ftrace") != 0) {
2361 res = eventfs_add_file("filter", TRACE_MODE_WRITE,
2363 &ftrace_subsystem_filter_fops);
2365 kfree(system->filter);
2366 system->filter = NULL;
2367 pr_warn("Could not create tracefs '%s/filter' entry\n", name);
2370 eventfs_add_file("enable", TRACE_MODE_WRITE, dir->ef, dir,
2371 &ftrace_system_enable_fops);
2374 list_add(&dir->list, &tr->systems);
2381 /* Only print this message if failed on memory allocation */
2382 if (!dir || !system)
2383 pr_warn("No memory to create event subsystem %s\n", name);
2388 event_define_fields(struct trace_event_call *call)
2390 struct list_head *head;
2394 * Other events may have the same class. Only update
2395 * the fields if they are not already defined.
2397 head = trace_get_fields(call);
2398 if (list_empty(head)) {
2399 struct trace_event_fields *field = call->class->fields_array;
2400 unsigned int offset = sizeof(struct trace_entry);
2402 for (; field->type; field++) {
2403 if (field->type == TRACE_FUNCTION_TYPE) {
2404 field->define_fields(call);
2408 offset = ALIGN(offset, field->align);
2409 ret = trace_define_field_ext(call, field->type, field->name,
2410 offset, field->size,
2411 field->is_signed, field->filter_type,
2413 if (WARN_ON_ONCE(ret)) {
2414 pr_err("error code is %d\n", ret);
2418 offset += field->size;
2426 event_create_dir(struct dentry *parent, struct trace_event_file *file)
2428 struct trace_event_call *call = file->event_call;
2429 struct eventfs_file *ef_subsystem = NULL;
2430 struct trace_array *tr = file->tr;
2431 struct eventfs_file *ef;
2436 * If the trace point header did not define TRACE_SYSTEM
2437 * then the system would be called "TRACE_SYSTEM". This should
2440 if (WARN_ON_ONCE(strcmp(call->class->system, TRACE_SYSTEM) == 0))
2443 ef_subsystem = event_subsystem_dir(tr, call->class->system, file, parent);
2447 name = trace_event_name(call);
2448 ef = eventfs_add_dir(name, ef_subsystem);
2450 pr_warn("Could not create tracefs '%s' directory\n", name);
2456 if (call->class->reg && !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
2457 eventfs_add_file("enable", TRACE_MODE_WRITE, file->ef, file,
2458 &ftrace_enable_fops);
2460 #ifdef CONFIG_PERF_EVENTS
2461 if (call->event.type && call->class->reg)
2462 eventfs_add_file("id", TRACE_MODE_READ, file->ef,
2463 (void *)(long)call->event.type,
2464 &ftrace_event_id_fops);
2467 ret = event_define_fields(call);
2469 pr_warn("Could not initialize trace point events/%s\n", name);
2474 * Only event directories that can be enabled should have
2475 * triggers or filters.
2477 if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)) {
2478 eventfs_add_file("filter", TRACE_MODE_WRITE, file->ef,
2479 file, &ftrace_event_filter_fops);
2481 eventfs_add_file("trigger", TRACE_MODE_WRITE, file->ef,
2482 file, &event_trigger_fops);
2485 #ifdef CONFIG_HIST_TRIGGERS
2486 eventfs_add_file("hist", TRACE_MODE_READ, file->ef, file,
2489 #ifdef CONFIG_HIST_TRIGGERS_DEBUG
2490 eventfs_add_file("hist_debug", TRACE_MODE_READ, file->ef, file,
2491 &event_hist_debug_fops);
2493 eventfs_add_file("format", TRACE_MODE_READ, file->ef, call,
2494 &ftrace_event_format_fops);
2496 #ifdef CONFIG_TRACE_EVENT_INJECT
2497 if (call->event.type && call->class->reg)
2498 eventfs_add_file("inject", 0200, file->ef, file,
2499 &event_inject_fops);
2505 static void remove_event_from_tracers(struct trace_event_call *call)
2507 struct trace_event_file *file;
2508 struct trace_array *tr;
2510 do_for_each_event_file_safe(tr, file) {
2511 if (file->event_call != call)
2514 remove_event_file_dir(file);
2516 * The do_for_each_event_file_safe() is
2517 * a double loop. After finding the call for this
2518 * trace_array, we use break to jump to the next
2522 } while_for_each_event_file();
2525 static void event_remove(struct trace_event_call *call)
2527 struct trace_array *tr;
2528 struct trace_event_file *file;
2530 do_for_each_event_file(tr, file) {
2531 if (file->event_call != call)
2534 if (file->flags & EVENT_FILE_FL_WAS_ENABLED)
2535 tr->clear_trace = true;
2537 ftrace_event_enable_disable(file, 0);
2539 * The do_for_each_event_file() is
2540 * a double loop. After finding the call for this
2541 * trace_array, we use break to jump to the next
2545 } while_for_each_event_file();
2547 if (call->event.funcs)
2548 __unregister_trace_event(&call->event);
2549 remove_event_from_tracers(call);
2550 list_del(&call->list);
2553 static int event_init(struct trace_event_call *call)
2558 name = trace_event_name(call);
2562 if (call->class->raw_init) {
2563 ret = call->class->raw_init(call);
2564 if (ret < 0 && ret != -ENOSYS)
2565 pr_warn("Could not initialize trace events/%s\n", name);
2572 __register_event(struct trace_event_call *call, struct module *mod)
2576 ret = event_init(call);
2580 list_add(&call->list, &ftrace_events);
2581 if (call->flags & TRACE_EVENT_FL_DYNAMIC)
2582 atomic_set(&call->refcnt, 0);
2589 static char *eval_replace(char *ptr, struct trace_eval_map *map, int len)
2594 /* Find the length of the eval value as a string */
2595 elen = snprintf(ptr, 0, "%ld", map->eval_value);
2596 /* Make sure there's enough room to replace the string with the value */
2600 snprintf(ptr, elen + 1, "%ld", map->eval_value);
2602 /* Get the rest of the string of ptr */
2603 rlen = strlen(ptr + len);
2604 memmove(ptr + elen, ptr + len, rlen);
2605 /* Make sure we end the new string */
2606 ptr[elen + rlen] = 0;
2611 static void update_event_printk(struct trace_event_call *call,
2612 struct trace_eval_map *map)
2616 int len = strlen(map->eval_string);
2618 for (ptr = call->print_fmt; *ptr; ptr++) {
2632 if (isdigit(*ptr)) {
2636 /* Check for alpha chars like ULL */
2637 } while (isalnum(*ptr));
2641 * A number must have some kind of delimiter after
2642 * it, and we can ignore that too.
2646 if (isalpha(*ptr) || *ptr == '_') {
2647 if (strncmp(map->eval_string, ptr, len) == 0 &&
2648 !isalnum(ptr[len]) && ptr[len] != '_') {
2649 ptr = eval_replace(ptr, map, len);
2650 /* enum/sizeof string smaller than value */
2651 if (WARN_ON_ONCE(!ptr))
2654 * No need to decrement here, as eval_replace()
2655 * returns the pointer to the character passed
2656 * the eval, and two evals can not be placed
2657 * back to back without something in between.
2658 * We can skip that something in between.
2665 } while (isalnum(*ptr) || *ptr == '_');
2669 * If what comes after this variable is a '.' or
2670 * '->' then we can continue to ignore that string.
2672 if (*ptr == '.' || (ptr[0] == '-' && ptr[1] == '>')) {
2673 ptr += *ptr == '.' ? 1 : 2;
2679 * Once again, we can skip the delimiter that came
2687 static void add_str_to_module(struct module *module, char *str)
2689 struct module_string *modstr;
2691 modstr = kmalloc(sizeof(*modstr), GFP_KERNEL);
2694 * If we failed to allocate memory here, then we'll just
2695 * let the str memory leak when the module is removed.
2696 * If this fails to allocate, there's worse problems than
2697 * a leaked string on module removal.
2699 if (WARN_ON_ONCE(!modstr))
2702 modstr->module = module;
2705 list_add(&modstr->next, &module_strings);
2708 static void update_event_fields(struct trace_event_call *call,
2709 struct trace_eval_map *map)
2711 struct ftrace_event_field *field;
2712 struct list_head *head;
2715 int len = strlen(map->eval_string);
2717 /* Dynamic events should never have field maps */
2718 if (WARN_ON_ONCE(call->flags & TRACE_EVENT_FL_DYNAMIC))
2721 head = trace_get_fields(call);
2722 list_for_each_entry(field, head, link) {
2723 ptr = strchr(field->type, '[');
2728 if (!isalpha(*ptr) && *ptr != '_')
2731 if (strncmp(map->eval_string, ptr, len) != 0)
2734 str = kstrdup(field->type, GFP_KERNEL);
2735 if (WARN_ON_ONCE(!str))
2737 ptr = str + (ptr - field->type);
2738 ptr = eval_replace(ptr, map, len);
2739 /* enum/sizeof string smaller than value */
2740 if (WARN_ON_ONCE(!ptr)) {
2746 * If the event is part of a module, then we need to free the string
2747 * when the module is removed. Otherwise, it will stay allocated
2751 add_str_to_module(call->module, str);
2757 void trace_event_eval_update(struct trace_eval_map **map, int len)
2759 struct trace_event_call *call, *p;
2760 const char *last_system = NULL;
2765 down_write(&trace_event_sem);
2766 list_for_each_entry_safe(call, p, &ftrace_events, list) {
2767 /* events are usually grouped together with systems */
2768 if (!last_system || call->class->system != last_system) {
2771 last_system = call->class->system;
2775 * Since calls are grouped by systems, the likelihood that the
2776 * next call in the iteration belongs to the same system as the
2777 * previous call is high. As an optimization, we skip searching
2778 * for a map[] that matches the call's system if the last call
2779 * was from the same system. That's what last_i is for. If the
2780 * call has the same system as the previous call, then last_i
2781 * will be the index of the first map[] that has a matching
2784 for (i = last_i; i < len; i++) {
2785 if (call->class->system == map[i]->system) {
2786 /* Save the first system if need be */
2791 update_event_printk(call, map[i]);
2792 update_event_fields(call, map[i]);
2797 up_write(&trace_event_sem);
2800 static struct trace_event_file *
2801 trace_create_new_event(struct trace_event_call *call,
2802 struct trace_array *tr)
2804 struct trace_pid_list *no_pid_list;
2805 struct trace_pid_list *pid_list;
2806 struct trace_event_file *file;
2809 file = kmem_cache_alloc(file_cachep, GFP_TRACE);
2813 pid_list = rcu_dereference_protected(tr->filtered_pids,
2814 lockdep_is_held(&event_mutex));
2815 no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
2816 lockdep_is_held(&event_mutex));
2818 if (!trace_pid_list_first(pid_list, &first) ||
2819 !trace_pid_list_first(no_pid_list, &first))
2820 file->flags |= EVENT_FILE_FL_PID_FILTER;
2822 file->event_call = call;
2824 atomic_set(&file->sm_ref, 0);
2825 atomic_set(&file->tm_ref, 0);
2826 INIT_LIST_HEAD(&file->triggers);
2827 list_add(&file->list, &tr->events);
2828 event_file_get(file);
2833 #define MAX_BOOT_TRIGGERS 32
2835 static struct boot_triggers {
2838 } bootup_triggers[MAX_BOOT_TRIGGERS];
2840 static char bootup_trigger_buf[COMMAND_LINE_SIZE];
2841 static int nr_boot_triggers;
2843 static __init int setup_trace_triggers(char *str)
2849 strscpy(bootup_trigger_buf, str, COMMAND_LINE_SIZE);
2850 ring_buffer_expanded = true;
2851 disable_tracing_selftest("running event triggers");
2853 buf = bootup_trigger_buf;
2854 for (i = 0; i < MAX_BOOT_TRIGGERS; i++) {
2855 trigger = strsep(&buf, ",");
2858 bootup_triggers[i].event = strsep(&trigger, ".");
2859 bootup_triggers[i].trigger = trigger;
2860 if (!bootup_triggers[i].trigger)
2864 nr_boot_triggers = i;
2867 __setup("trace_trigger=", setup_trace_triggers);
2869 /* Add an event to a trace directory */
2871 __trace_add_new_event(struct trace_event_call *call, struct trace_array *tr)
2873 struct trace_event_file *file;
2875 file = trace_create_new_event(call, tr);
2879 if (eventdir_initialized)
2880 return event_create_dir(tr->event_dir, file);
2882 return event_define_fields(call);
2885 static void trace_early_triggers(struct trace_event_file *file, const char *name)
2890 for (i = 0; i < nr_boot_triggers; i++) {
2891 if (strcmp(name, bootup_triggers[i].event))
2893 mutex_lock(&event_mutex);
2894 ret = trigger_process_regex(file, bootup_triggers[i].trigger);
2895 mutex_unlock(&event_mutex);
2897 pr_err("Failed to register trigger '%s' on event %s\n",
2898 bootup_triggers[i].trigger,
2899 bootup_triggers[i].event);
2904 * Just create a descriptor for early init. A descriptor is required
2905 * for enabling events at boot. We want to enable events before
2906 * the filesystem is initialized.
2909 __trace_early_add_new_event(struct trace_event_call *call,
2910 struct trace_array *tr)
2912 struct trace_event_file *file;
2915 file = trace_create_new_event(call, tr);
2919 ret = event_define_fields(call);
2923 trace_early_triggers(file, trace_event_name(call));
2928 struct ftrace_module_file_ops;
2929 static void __add_event_to_tracers(struct trace_event_call *call);
2931 /* Add an additional event_call dynamically */
2932 int trace_add_event_call(struct trace_event_call *call)
2935 lockdep_assert_held(&event_mutex);
2937 mutex_lock(&trace_types_lock);
2939 ret = __register_event(call, NULL);
2941 __add_event_to_tracers(call);
2943 mutex_unlock(&trace_types_lock);
2946 EXPORT_SYMBOL_GPL(trace_add_event_call);
2949 * Must be called under locking of trace_types_lock, event_mutex and
2952 static void __trace_remove_event_call(struct trace_event_call *call)
2955 trace_destroy_fields(call);
2956 free_event_filter(call->filter);
2957 call->filter = NULL;
2960 static int probe_remove_event_call(struct trace_event_call *call)
2962 struct trace_array *tr;
2963 struct trace_event_file *file;
2965 #ifdef CONFIG_PERF_EVENTS
2966 if (call->perf_refcount)
2969 do_for_each_event_file(tr, file) {
2970 if (file->event_call != call)
2973 * We can't rely on ftrace_event_enable_disable(enable => 0)
2974 * we are going to do, EVENT_FILE_FL_SOFT_MODE can suppress
2975 * TRACE_REG_UNREGISTER.
2977 if (file->flags & EVENT_FILE_FL_ENABLED)
2980 if (file->flags & EVENT_FILE_FL_WAS_ENABLED)
2981 tr->clear_trace = true;
2983 * The do_for_each_event_file_safe() is
2984 * a double loop. After finding the call for this
2985 * trace_array, we use break to jump to the next
2989 } while_for_each_event_file();
2991 __trace_remove_event_call(call);
2995 /* No need to clear the trace now */
2996 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
2997 tr->clear_trace = false;
3002 /* Remove an event_call */
3003 int trace_remove_event_call(struct trace_event_call *call)
3007 lockdep_assert_held(&event_mutex);
3009 mutex_lock(&trace_types_lock);
3010 down_write(&trace_event_sem);
3011 ret = probe_remove_event_call(call);
3012 up_write(&trace_event_sem);
3013 mutex_unlock(&trace_types_lock);
3017 EXPORT_SYMBOL_GPL(trace_remove_event_call);
3019 #define for_each_event(event, start, end) \
3020 for (event = start; \
3021 (unsigned long)event < (unsigned long)end; \
3024 #ifdef CONFIG_MODULES
3026 static void trace_module_add_events(struct module *mod)
3028 struct trace_event_call **call, **start, **end;
3030 if (!mod->num_trace_events)
3033 /* Don't add infrastructure for mods without tracepoints */
3034 if (trace_module_has_bad_taint(mod)) {
3035 pr_err("%s: module has bad taint, not creating trace events\n",
3040 start = mod->trace_events;
3041 end = mod->trace_events + mod->num_trace_events;
3043 for_each_event(call, start, end) {
3044 __register_event(*call, mod);
3045 __add_event_to_tracers(*call);
3049 static void trace_module_remove_events(struct module *mod)
3051 struct trace_event_call *call, *p;
3052 struct module_string *modstr, *m;
3054 down_write(&trace_event_sem);
3055 list_for_each_entry_safe(call, p, &ftrace_events, list) {
3056 if ((call->flags & TRACE_EVENT_FL_DYNAMIC) || !call->module)
3058 if (call->module == mod)
3059 __trace_remove_event_call(call);
3061 /* Check for any strings allocade for this module */
3062 list_for_each_entry_safe(modstr, m, &module_strings, next) {
3063 if (modstr->module != mod)
3065 list_del(&modstr->next);
3069 up_write(&trace_event_sem);
3072 * It is safest to reset the ring buffer if the module being unloaded
3073 * registered any events that were used. The only worry is if
3074 * a new module gets loaded, and takes on the same id as the events
3075 * of this module. When printing out the buffer, traced events left
3076 * over from this module may be passed to the new module events and
3077 * unexpected results may occur.
3079 tracing_reset_all_online_cpus_unlocked();
3082 static int trace_module_notify(struct notifier_block *self,
3083 unsigned long val, void *data)
3085 struct module *mod = data;
3087 mutex_lock(&event_mutex);
3088 mutex_lock(&trace_types_lock);
3090 case MODULE_STATE_COMING:
3091 trace_module_add_events(mod);
3093 case MODULE_STATE_GOING:
3094 trace_module_remove_events(mod);
3097 mutex_unlock(&trace_types_lock);
3098 mutex_unlock(&event_mutex);
3103 static struct notifier_block trace_module_nb = {
3104 .notifier_call = trace_module_notify,
3105 .priority = 1, /* higher than trace.c module notify */
3107 #endif /* CONFIG_MODULES */
3109 /* Create a new event directory structure for a trace directory. */
3111 __trace_add_event_dirs(struct trace_array *tr)
3113 struct trace_event_call *call;
3116 list_for_each_entry(call, &ftrace_events, list) {
3117 ret = __trace_add_new_event(call, tr);
3119 pr_warn("Could not create directory for event %s\n",
3120 trace_event_name(call));
3124 /* Returns any file that matches the system and event */
3125 struct trace_event_file *
3126 __find_event_file(struct trace_array *tr, const char *system, const char *event)
3128 struct trace_event_file *file;
3129 struct trace_event_call *call;
3132 list_for_each_entry(file, &tr->events, list) {
3134 call = file->event_call;
3135 name = trace_event_name(call);
3137 if (!name || !call->class)
3140 if (strcmp(event, name) == 0 &&
3141 strcmp(system, call->class->system) == 0)
3147 /* Returns valid trace event files that match system and event */
3148 struct trace_event_file *
3149 find_event_file(struct trace_array *tr, const char *system, const char *event)
3151 struct trace_event_file *file;
3153 file = __find_event_file(tr, system, event);
3154 if (!file || !file->event_call->class->reg ||
3155 file->event_call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
3162 * trace_get_event_file - Find and return a trace event file
3163 * @instance: The name of the trace instance containing the event
3164 * @system: The name of the system containing the event
3165 * @event: The name of the event
3167 * Return a trace event file given the trace instance name, trace
3168 * system, and trace event name. If the instance name is NULL, it
3169 * refers to the top-level trace array.
3171 * This function will look it up and return it if found, after calling
3172 * trace_array_get() to prevent the instance from going away, and
3173 * increment the event's module refcount to prevent it from being
3176 * To release the file, call trace_put_event_file(), which will call
3177 * trace_array_put() and decrement the event's module refcount.
3179 * Return: The trace event on success, ERR_PTR otherwise.
3181 struct trace_event_file *trace_get_event_file(const char *instance,
3185 struct trace_array *tr = top_trace_array();
3186 struct trace_event_file *file = NULL;
3190 tr = trace_array_find_get(instance);
3192 return ERR_PTR(-ENOENT);
3194 ret = trace_array_get(tr);
3196 return ERR_PTR(ret);
3199 mutex_lock(&event_mutex);
3201 file = find_event_file(tr, system, event);
3203 trace_array_put(tr);
3208 /* Don't let event modules unload while in use */
3209 ret = trace_event_try_get_ref(file->event_call);
3211 trace_array_put(tr);
3218 mutex_unlock(&event_mutex);
3221 file = ERR_PTR(ret);
3225 EXPORT_SYMBOL_GPL(trace_get_event_file);
3228 * trace_put_event_file - Release a file from trace_get_event_file()
3229 * @file: The trace event file
3231 * If a file was retrieved using trace_get_event_file(), this should
3232 * be called when it's no longer needed. It will cancel the previous
3233 * trace_array_get() called by that function, and decrement the
3234 * event's module refcount.
3236 void trace_put_event_file(struct trace_event_file *file)
3238 mutex_lock(&event_mutex);
3239 trace_event_put_ref(file->event_call);
3240 mutex_unlock(&event_mutex);
3242 trace_array_put(file->tr);
3244 EXPORT_SYMBOL_GPL(trace_put_event_file);
3246 #ifdef CONFIG_DYNAMIC_FTRACE
3249 #define ENABLE_EVENT_STR "enable_event"
3250 #define DISABLE_EVENT_STR "disable_event"
3252 struct event_probe_data {
3253 struct trace_event_file *file;
3254 unsigned long count;
3259 static void update_event_probe(struct event_probe_data *data)
3262 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
3264 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
3268 event_enable_probe(unsigned long ip, unsigned long parent_ip,
3269 struct trace_array *tr, struct ftrace_probe_ops *ops,
3272 struct ftrace_func_mapper *mapper = data;
3273 struct event_probe_data *edata;
3276 pdata = ftrace_func_mapper_find_ip(mapper, ip);
3277 if (!pdata || !*pdata)
3281 update_event_probe(edata);
3285 event_enable_count_probe(unsigned long ip, unsigned long parent_ip,
3286 struct trace_array *tr, struct ftrace_probe_ops *ops,
3289 struct ftrace_func_mapper *mapper = data;
3290 struct event_probe_data *edata;
3293 pdata = ftrace_func_mapper_find_ip(mapper, ip);
3294 if (!pdata || !*pdata)
3302 /* Skip if the event is in a state we want to switch to */
3303 if (edata->enable == !(edata->file->flags & EVENT_FILE_FL_SOFT_DISABLED))
3306 if (edata->count != -1)
3309 update_event_probe(edata);
3313 event_enable_print(struct seq_file *m, unsigned long ip,
3314 struct ftrace_probe_ops *ops, void *data)
3316 struct ftrace_func_mapper *mapper = data;
3317 struct event_probe_data *edata;
3320 pdata = ftrace_func_mapper_find_ip(mapper, ip);
3322 if (WARN_ON_ONCE(!pdata || !*pdata))
3327 seq_printf(m, "%ps:", (void *)ip);
3329 seq_printf(m, "%s:%s:%s",
3330 edata->enable ? ENABLE_EVENT_STR : DISABLE_EVENT_STR,
3331 edata->file->event_call->class->system,
3332 trace_event_name(edata->file->event_call));
3334 if (edata->count == -1)
3335 seq_puts(m, ":unlimited\n");
3337 seq_printf(m, ":count=%ld\n", edata->count);
3343 event_enable_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
3344 unsigned long ip, void *init_data, void **data)
3346 struct ftrace_func_mapper *mapper = *data;
3347 struct event_probe_data *edata = init_data;
3351 mapper = allocate_ftrace_func_mapper();
3357 ret = ftrace_func_mapper_add_ip(mapper, ip, edata);
3366 static int free_probe_data(void *data)
3368 struct event_probe_data *edata = data;
3372 /* Remove the SOFT_MODE flag */
3373 __ftrace_event_enable_disable(edata->file, 0, 1);
3374 trace_event_put_ref(edata->file->event_call);
3381 event_enable_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
3382 unsigned long ip, void *data)
3384 struct ftrace_func_mapper *mapper = data;
3385 struct event_probe_data *edata;
3390 free_ftrace_func_mapper(mapper, free_probe_data);
3394 edata = ftrace_func_mapper_remove_ip(mapper, ip);
3396 if (WARN_ON_ONCE(!edata))
3399 if (WARN_ON_ONCE(edata->ref <= 0))
3402 free_probe_data(edata);
3405 static struct ftrace_probe_ops event_enable_probe_ops = {
3406 .func = event_enable_probe,
3407 .print = event_enable_print,
3408 .init = event_enable_init,
3409 .free = event_enable_free,
3412 static struct ftrace_probe_ops event_enable_count_probe_ops = {
3413 .func = event_enable_count_probe,
3414 .print = event_enable_print,
3415 .init = event_enable_init,
3416 .free = event_enable_free,
3419 static struct ftrace_probe_ops event_disable_probe_ops = {
3420 .func = event_enable_probe,
3421 .print = event_enable_print,
3422 .init = event_enable_init,
3423 .free = event_enable_free,
3426 static struct ftrace_probe_ops event_disable_count_probe_ops = {
3427 .func = event_enable_count_probe,
3428 .print = event_enable_print,
3429 .init = event_enable_init,
3430 .free = event_enable_free,
3434 event_enable_func(struct trace_array *tr, struct ftrace_hash *hash,
3435 char *glob, char *cmd, char *param, int enabled)
3437 struct trace_event_file *file;
3438 struct ftrace_probe_ops *ops;
3439 struct event_probe_data *data;
3449 /* hash funcs only work with set_ftrace_filter */
3450 if (!enabled || !param)
3453 system = strsep(¶m, ":");
3457 event = strsep(¶m, ":");
3459 mutex_lock(&event_mutex);
3462 file = find_event_file(tr, system, event);
3466 enable = strcmp(cmd, ENABLE_EVENT_STR) == 0;
3469 ops = param ? &event_enable_count_probe_ops : &event_enable_probe_ops;
3471 ops = param ? &event_disable_count_probe_ops : &event_disable_probe_ops;
3473 if (glob[0] == '!') {
3474 ret = unregister_ftrace_function_probe_func(glob+1, tr, ops);
3480 data = kzalloc(sizeof(*data), GFP_KERNEL);
3484 data->enable = enable;
3491 number = strsep(¶m, ":");
3494 if (!strlen(number))
3498 * We use the callback data field (which is a pointer)
3501 ret = kstrtoul(number, 0, &data->count);
3506 /* Don't let event modules unload while probe registered */
3507 ret = trace_event_try_get_ref(file->event_call);
3513 ret = __ftrace_event_enable_disable(file, 1, 1);
3517 ret = register_ftrace_function_probe(glob, tr, ops, data);
3519 * The above returns on success the # of functions enabled,
3520 * but if it didn't find any functions it returns zero.
3521 * Consider no functions a failure too.
3528 /* Just return zero, not the number of enabled functions */
3531 mutex_unlock(&event_mutex);
3535 __ftrace_event_enable_disable(file, 0, 1);
3537 trace_event_put_ref(file->event_call);
3543 static struct ftrace_func_command event_enable_cmd = {
3544 .name = ENABLE_EVENT_STR,
3545 .func = event_enable_func,
3548 static struct ftrace_func_command event_disable_cmd = {
3549 .name = DISABLE_EVENT_STR,
3550 .func = event_enable_func,
3553 static __init int register_event_cmds(void)
3557 ret = register_ftrace_command(&event_enable_cmd);
3558 if (WARN_ON(ret < 0))
3560 ret = register_ftrace_command(&event_disable_cmd);
3561 if (WARN_ON(ret < 0))
3562 unregister_ftrace_command(&event_enable_cmd);
3566 static inline int register_event_cmds(void) { return 0; }
3567 #endif /* CONFIG_DYNAMIC_FTRACE */
3570 * The top level array and trace arrays created by boot-time tracing
3571 * have already had its trace_event_file descriptors created in order
3572 * to allow for early events to be recorded.
3573 * This function is called after the tracefs has been initialized,
3574 * and we now have to create the files associated to the events.
3576 static void __trace_early_add_event_dirs(struct trace_array *tr)
3578 struct trace_event_file *file;
3582 list_for_each_entry(file, &tr->events, list) {
3583 ret = event_create_dir(tr->event_dir, file);
3585 pr_warn("Could not create directory for event %s\n",
3586 trace_event_name(file->event_call));
3591 * For early boot up, the top trace array and the trace arrays created
3592 * by boot-time tracing require to have a list of events that can be
3593 * enabled. This must be done before the filesystem is set up in order
3594 * to allow events to be traced early.
3596 void __trace_early_add_events(struct trace_array *tr)
3598 struct trace_event_call *call;
3601 list_for_each_entry(call, &ftrace_events, list) {
3602 /* Early boot up should not have any modules loaded */
3603 if (!(call->flags & TRACE_EVENT_FL_DYNAMIC) &&
3604 WARN_ON_ONCE(call->module))
3607 ret = __trace_early_add_new_event(call, tr);
3609 pr_warn("Could not create early event %s\n",
3610 trace_event_name(call));
3614 /* Remove the event directory structure for a trace directory. */
3616 __trace_remove_event_dirs(struct trace_array *tr)
3618 struct trace_event_file *file, *next;
3620 list_for_each_entry_safe(file, next, &tr->events, list)
3621 remove_event_file_dir(file);
3624 static void __add_event_to_tracers(struct trace_event_call *call)
3626 struct trace_array *tr;
3628 list_for_each_entry(tr, &ftrace_trace_arrays, list)
3629 __trace_add_new_event(call, tr);
3632 extern struct trace_event_call *__start_ftrace_events[];
3633 extern struct trace_event_call *__stop_ftrace_events[];
3635 static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
3637 static __init int setup_trace_event(char *str)
3639 strscpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
3640 ring_buffer_expanded = true;
3641 disable_tracing_selftest("running event tracing");
3645 __setup("trace_event=", setup_trace_event);
3647 /* Expects to have event_mutex held when called */
3649 create_event_toplevel_files(struct dentry *parent, struct trace_array *tr)
3651 struct dentry *d_events;
3652 struct dentry *entry;
3655 entry = trace_create_file("set_event", TRACE_MODE_WRITE, parent,
3656 tr, &ftrace_set_event_fops);
3660 d_events = eventfs_create_events_dir("events", parent);
3661 if (IS_ERR(d_events)) {
3662 pr_warn("Could not create tracefs 'events' directory\n");
3666 error = eventfs_add_events_file("enable", TRACE_MODE_WRITE, d_events,
3667 tr, &ftrace_tr_enable_fops);
3671 /* There are not as crucial, just warn if they are not created */
3673 trace_create_file("set_event_pid", TRACE_MODE_WRITE, parent,
3674 tr, &ftrace_set_event_pid_fops);
3676 trace_create_file("set_event_notrace_pid",
3677 TRACE_MODE_WRITE, parent, tr,
3678 &ftrace_set_event_notrace_pid_fops);
3680 /* ring buffer internal formats */
3681 eventfs_add_events_file("header_page", TRACE_MODE_READ, d_events,
3682 ring_buffer_print_page_header,
3683 &ftrace_show_header_fops);
3685 eventfs_add_events_file("header_event", TRACE_MODE_READ, d_events,
3686 ring_buffer_print_entry_header,
3687 &ftrace_show_header_fops);
3689 tr->event_dir = d_events;
3695 * event_trace_add_tracer - add a instance of a trace_array to events
3696 * @parent: The parent dentry to place the files/directories for events in
3697 * @tr: The trace array associated with these events
3699 * When a new instance is created, it needs to set up its events
3700 * directory, as well as other files associated with events. It also
3701 * creates the event hierarchy in the @parent/events directory.
3703 * Returns 0 on success.
3705 * Must be called with event_mutex held.
3707 int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr)
3711 lockdep_assert_held(&event_mutex);
3713 ret = create_event_toplevel_files(parent, tr);
3717 down_write(&trace_event_sem);
3718 /* If tr already has the event list, it is initialized in early boot. */
3719 if (unlikely(!list_empty(&tr->events)))
3720 __trace_early_add_event_dirs(tr);
3722 __trace_add_event_dirs(tr);
3723 up_write(&trace_event_sem);
3730 * The top trace array already had its file descriptors created.
3731 * Now the files themselves need to be created.
3734 early_event_add_tracer(struct dentry *parent, struct trace_array *tr)
3738 mutex_lock(&event_mutex);
3740 ret = create_event_toplevel_files(parent, tr);
3744 down_write(&trace_event_sem);
3745 __trace_early_add_event_dirs(tr);
3746 up_write(&trace_event_sem);
3749 mutex_unlock(&event_mutex);
3754 /* Must be called with event_mutex held */
3755 int event_trace_del_tracer(struct trace_array *tr)
3757 lockdep_assert_held(&event_mutex);
3759 /* Disable any event triggers and associated soft-disabled events */
3760 clear_event_triggers(tr);
3762 /* Clear the pid list */
3763 __ftrace_clear_event_pids(tr, TRACE_PIDS | TRACE_NO_PIDS);
3765 /* Disable any running events */
3766 __ftrace_set_clr_event_nolock(tr, NULL, NULL, NULL, 0);
3768 /* Make sure no more events are being executed */
3769 tracepoint_synchronize_unregister();
3771 down_write(&trace_event_sem);
3772 __trace_remove_event_dirs(tr);
3773 eventfs_remove_events_dir(tr->event_dir);
3774 up_write(&trace_event_sem);
3776 tr->event_dir = NULL;
3781 static __init int event_trace_memsetup(void)
3783 field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC);
3784 file_cachep = KMEM_CACHE(trace_event_file, SLAB_PANIC);
3789 early_enable_events(struct trace_array *tr, char *buf, bool disable_first)
3795 token = strsep(&buf, ",");
3801 /* Restarting syscalls requires that we stop them first */
3803 ftrace_set_clr_event(tr, token, 0);
3805 ret = ftrace_set_clr_event(tr, token, 1);
3807 pr_warn("Failed to enable trace event: %s\n", token);
3810 /* Put back the comma to allow this to be called again */
3816 static __init int event_trace_enable(void)
3818 struct trace_array *tr = top_trace_array();
3819 struct trace_event_call **iter, *call;
3825 for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
3828 ret = event_init(call);
3830 list_add(&call->list, &ftrace_events);
3833 register_trigger_cmds();
3836 * We need the top trace array to have a working set of trace
3837 * points at early init, before the debug files and directories
3838 * are created. Create the file entries now, and attach them
3839 * to the actual file dentries later.
3841 __trace_early_add_events(tr);
3843 early_enable_events(tr, bootup_event_buf, false);
3845 trace_printk_start_comm();
3847 register_event_cmds();
3854 * event_trace_enable() is called from trace_event_init() first to
3855 * initialize events and perhaps start any events that are on the
3856 * command line. Unfortunately, there are some events that will not
3857 * start this early, like the system call tracepoints that need
3858 * to set the %SYSCALL_WORK_SYSCALL_TRACEPOINT flag of pid 1. But
3859 * event_trace_enable() is called before pid 1 starts, and this flag
3860 * is never set, making the syscall tracepoint never get reached, but
3861 * the event is enabled regardless (and not doing anything).
3863 static __init int event_trace_enable_again(void)
3865 struct trace_array *tr;
3867 tr = top_trace_array();
3871 early_enable_events(tr, bootup_event_buf, true);
3876 early_initcall(event_trace_enable_again);
3878 /* Init fields which doesn't related to the tracefs */
3879 static __init int event_trace_init_fields(void)
3881 if (trace_define_generic_fields())
3882 pr_warn("tracing: Failed to allocated generic fields");
3884 if (trace_define_common_fields())
3885 pr_warn("tracing: Failed to allocate common fields");
3890 __init int event_trace_init(void)
3892 struct trace_array *tr;
3895 tr = top_trace_array();
3899 trace_create_file("available_events", TRACE_MODE_READ,
3900 NULL, tr, &ftrace_avail_fops);
3902 ret = early_event_add_tracer(NULL, tr);
3906 #ifdef CONFIG_MODULES
3907 ret = register_module_notifier(&trace_module_nb);
3909 pr_warn("Failed to register trace events module notifier\n");
3912 eventdir_initialized = true;
3917 void __init trace_event_init(void)
3919 event_trace_memsetup();
3920 init_ftrace_syscalls();
3921 event_trace_enable();
3922 event_trace_init_fields();
3925 #ifdef CONFIG_EVENT_TRACE_STARTUP_TEST
3927 static DEFINE_SPINLOCK(test_spinlock);
3928 static DEFINE_SPINLOCK(test_spinlock_irq);
3929 static DEFINE_MUTEX(test_mutex);
3931 static __init void test_work(struct work_struct *dummy)
3933 spin_lock(&test_spinlock);
3934 spin_lock_irq(&test_spinlock_irq);
3936 spin_unlock_irq(&test_spinlock_irq);
3937 spin_unlock(&test_spinlock);
3939 mutex_lock(&test_mutex);
3941 mutex_unlock(&test_mutex);
3944 static __init int event_test_thread(void *unused)
3948 test_malloc = kmalloc(1234, GFP_KERNEL);
3950 pr_info("failed to kmalloc\n");
3952 schedule_on_each_cpu(test_work);
3956 set_current_state(TASK_INTERRUPTIBLE);
3957 while (!kthread_should_stop()) {
3959 set_current_state(TASK_INTERRUPTIBLE);
3961 __set_current_state(TASK_RUNNING);
3967 * Do various things that may trigger events.
3969 static __init void event_test_stuff(void)
3971 struct task_struct *test_thread;
3973 test_thread = kthread_run(event_test_thread, NULL, "test-events");
3975 kthread_stop(test_thread);
3979 * For every trace event defined, we will test each trace point separately,
3980 * and then by groups, and finally all trace points.
3982 static __init void event_trace_self_tests(void)
3984 struct trace_subsystem_dir *dir;
3985 struct trace_event_file *file;
3986 struct trace_event_call *call;
3987 struct event_subsystem *system;
3988 struct trace_array *tr;
3991 tr = top_trace_array();
3995 pr_info("Running tests on trace events:\n");
3997 list_for_each_entry(file, &tr->events, list) {
3999 call = file->event_call;
4001 /* Only test those that have a probe */
4002 if (!call->class || !call->class->probe)
4006 * Testing syscall events here is pretty useless, but
4007 * we still do it if configured. But this is time consuming.
4008 * What we really need is a user thread to perform the
4009 * syscalls as we test.
4011 #ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
4012 if (call->class->system &&
4013 strcmp(call->class->system, "syscalls") == 0)
4017 pr_info("Testing event %s: ", trace_event_name(call));
4020 * If an event is already enabled, someone is using
4021 * it and the self test should not be on.
4023 if (file->flags & EVENT_FILE_FL_ENABLED) {
4024 pr_warn("Enabled event during self test!\n");
4029 ftrace_event_enable_disable(file, 1);
4031 ftrace_event_enable_disable(file, 0);
4036 /* Now test at the sub system level */
4038 pr_info("Running tests on trace event systems:\n");
4040 list_for_each_entry(dir, &tr->systems, list) {
4042 system = dir->subsystem;
4044 /* the ftrace system is special, skip it */
4045 if (strcmp(system->name, "ftrace") == 0)
4048 pr_info("Testing event system %s: ", system->name);
4050 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1);
4051 if (WARN_ON_ONCE(ret)) {
4052 pr_warn("error enabling system %s\n",
4059 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0);
4060 if (WARN_ON_ONCE(ret)) {
4061 pr_warn("error disabling system %s\n",
4069 /* Test with all events enabled */
4071 pr_info("Running tests on all trace events:\n");
4072 pr_info("Testing all events: ");
4074 ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1);
4075 if (WARN_ON_ONCE(ret)) {
4076 pr_warn("error enabling all events\n");
4083 ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
4084 if (WARN_ON_ONCE(ret)) {
4085 pr_warn("error disabling all events\n");
4092 #ifdef CONFIG_FUNCTION_TRACER
4094 static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
4096 static struct trace_event_file event_trace_file __initdata;
4099 function_test_events_call(unsigned long ip, unsigned long parent_ip,
4100 struct ftrace_ops *op, struct ftrace_regs *regs)
4102 struct trace_buffer *buffer;
4103 struct ring_buffer_event *event;
4104 struct ftrace_entry *entry;
4105 unsigned int trace_ctx;
4109 trace_ctx = tracing_gen_ctx();
4110 preempt_disable_notrace();
4111 cpu = raw_smp_processor_id();
4112 disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
4117 event = trace_event_buffer_lock_reserve(&buffer, &event_trace_file,
4118 TRACE_FN, sizeof(*entry),
4122 entry = ring_buffer_event_data(event);
4124 entry->parent_ip = parent_ip;
4126 event_trigger_unlock_commit(&event_trace_file, buffer, event,
4129 atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
4130 preempt_enable_notrace();
4133 static struct ftrace_ops trace_ops __initdata =
4135 .func = function_test_events_call,
4138 static __init void event_trace_self_test_with_function(void)
4142 event_trace_file.tr = top_trace_array();
4143 if (WARN_ON(!event_trace_file.tr))
4146 ret = register_ftrace_function(&trace_ops);
4147 if (WARN_ON(ret < 0)) {
4148 pr_info("Failed to enable function tracer for event tests\n");
4151 pr_info("Running tests again, along with the function tracer\n");
4152 event_trace_self_tests();
4153 unregister_ftrace_function(&trace_ops);
4156 static __init void event_trace_self_test_with_function(void)
4161 static __init int event_trace_self_tests_init(void)
4163 if (!tracing_selftest_disabled) {
4164 event_trace_self_tests();
4165 event_trace_self_test_with_function();
4171 late_initcall(event_trace_self_tests_init);