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(dir->ei);
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_dir(file->ei);
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(tr);
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);
1426 mutex_lock(&event_mutex);
1427 file = event_file_data(filp);
1428 if (likely(file && !(file->flags & EVENT_FILE_FL_FREED))) {
1429 ret = tracing_update_buffers(file->tr);
1431 mutex_unlock(&event_mutex);
1434 ret = ftrace_event_enable_disable(file, val);
1436 mutex_unlock(&event_mutex);
1445 return ret ? ret : cnt;
1449 system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1452 const char set_to_char[4] = { '?', '0', '1', 'X' };
1453 struct trace_subsystem_dir *dir = filp->private_data;
1454 struct event_subsystem *system = dir->subsystem;
1455 struct trace_event_call *call;
1456 struct trace_event_file *file;
1457 struct trace_array *tr = dir->tr;
1462 mutex_lock(&event_mutex);
1463 list_for_each_entry(file, &tr->events, list) {
1464 call = file->event_call;
1465 if ((call->flags & TRACE_EVENT_FL_IGNORE_ENABLE) ||
1466 !trace_event_name(call) || !call->class || !call->class->reg)
1469 if (system && strcmp(call->class->system, system->name) != 0)
1473 * We need to find out if all the events are set
1474 * or if all events or cleared, or if we have
1477 set |= (1 << !!(file->flags & EVENT_FILE_FL_ENABLED));
1480 * If we have a mixture, no need to look further.
1485 mutex_unlock(&event_mutex);
1487 buf[0] = set_to_char[set];
1490 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
1496 system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1499 struct trace_subsystem_dir *dir = filp->private_data;
1500 struct event_subsystem *system = dir->subsystem;
1501 const char *name = NULL;
1505 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1509 ret = tracing_update_buffers(dir->tr);
1513 if (val != 0 && val != 1)
1517 * Opening of "enable" adds a ref count to system,
1518 * so the name is safe to use.
1521 name = system->name;
1523 ret = __ftrace_set_clr_event(dir->tr, NULL, name, NULL, val);
1537 FORMAT_FIELD_SEPERATOR = 2,
1538 FORMAT_PRINTFMT = 3,
1541 static void *f_next(struct seq_file *m, void *v, loff_t *pos)
1543 struct trace_event_call *call = event_file_data(m->private);
1544 struct list_head *common_head = &ftrace_common_fields;
1545 struct list_head *head = trace_get_fields(call);
1546 struct list_head *node = v;
1550 switch ((unsigned long)v) {
1555 case FORMAT_FIELD_SEPERATOR:
1559 case FORMAT_PRINTFMT:
1565 if (node == common_head)
1566 return (void *)FORMAT_FIELD_SEPERATOR;
1567 else if (node == head)
1568 return (void *)FORMAT_PRINTFMT;
1573 static int f_show(struct seq_file *m, void *v)
1575 struct trace_event_call *call = event_file_data(m->private);
1576 struct ftrace_event_field *field;
1577 const char *array_descriptor;
1579 switch ((unsigned long)v) {
1581 seq_printf(m, "name: %s\n", trace_event_name(call));
1582 seq_printf(m, "ID: %d\n", call->event.type);
1583 seq_puts(m, "format:\n");
1586 case FORMAT_FIELD_SEPERATOR:
1590 case FORMAT_PRINTFMT:
1591 seq_printf(m, "\nprint fmt: %s\n",
1596 field = list_entry(v, struct ftrace_event_field, link);
1598 * Smartly shows the array type(except dynamic array).
1601 * If TYPE := TYPE[LEN], it is shown:
1602 * field:TYPE VAR[LEN]
1604 array_descriptor = strchr(field->type, '[');
1606 if (str_has_prefix(field->type, "__data_loc"))
1607 array_descriptor = NULL;
1609 if (!array_descriptor)
1610 seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1611 field->type, field->name, field->offset,
1612 field->size, !!field->is_signed);
1613 else if (field->len)
1614 seq_printf(m, "\tfield:%.*s %s[%d];\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1615 (int)(array_descriptor - field->type),
1616 field->type, field->name,
1617 field->len, field->offset,
1618 field->size, !!field->is_signed);
1620 seq_printf(m, "\tfield:%.*s %s[];\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1621 (int)(array_descriptor - field->type),
1622 field->type, field->name,
1623 field->offset, field->size, !!field->is_signed);
1628 static void *f_start(struct seq_file *m, loff_t *pos)
1630 void *p = (void *)FORMAT_HEADER;
1633 /* ->stop() is called even if ->start() fails */
1634 mutex_lock(&event_mutex);
1635 if (!event_file_data(m->private))
1636 return ERR_PTR(-ENODEV);
1638 while (l < *pos && p)
1639 p = f_next(m, p, &l);
1644 static void f_stop(struct seq_file *m, void *p)
1646 mutex_unlock(&event_mutex);
1649 static const struct seq_operations trace_format_seq_ops = {
1656 static int trace_format_open(struct inode *inode, struct file *file)
1661 /* Do we want to hide event format files on tracefs lockdown? */
1663 ret = seq_open(file, &trace_format_seq_ops);
1667 m = file->private_data;
1673 #ifdef CONFIG_PERF_EVENTS
1675 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1677 int id = (long)event_file_data(filp);
1684 len = sprintf(buf, "%d\n", id);
1686 return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
1691 event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1694 struct trace_event_file *file;
1695 struct trace_seq *s;
1701 s = kmalloc(sizeof(*s), GFP_KERNEL);
1708 mutex_lock(&event_mutex);
1709 file = event_file_data(filp);
1710 if (file && !(file->flags & EVENT_FILE_FL_FREED))
1711 print_event_filter(file, s);
1712 mutex_unlock(&event_mutex);
1715 r = simple_read_from_buffer(ubuf, cnt, ppos,
1716 s->buffer, trace_seq_used(s));
1724 event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1727 struct trace_event_file *file;
1731 if (cnt >= PAGE_SIZE)
1734 buf = memdup_user_nul(ubuf, cnt);
1736 return PTR_ERR(buf);
1738 mutex_lock(&event_mutex);
1739 file = event_file_data(filp);
1741 err = apply_event_filter(file, buf);
1742 mutex_unlock(&event_mutex);
1753 static LIST_HEAD(event_subsystems);
1755 static int subsystem_open(struct inode *inode, struct file *filp)
1757 struct trace_subsystem_dir *dir = NULL, *iter_dir;
1758 struct trace_array *tr = NULL, *iter_tr;
1759 struct event_subsystem *system = NULL;
1762 if (tracing_is_disabled())
1765 /* Make sure the system still exists */
1766 mutex_lock(&event_mutex);
1767 mutex_lock(&trace_types_lock);
1768 list_for_each_entry(iter_tr, &ftrace_trace_arrays, list) {
1769 list_for_each_entry(iter_dir, &iter_tr->systems, list) {
1770 if (iter_dir == inode->i_private) {
1771 /* Don't open systems with no events */
1774 if (dir->nr_events) {
1775 __get_system_dir(dir);
1776 system = dir->subsystem;
1783 mutex_unlock(&trace_types_lock);
1784 mutex_unlock(&event_mutex);
1789 /* Still need to increment the ref count of the system */
1790 if (trace_array_get(tr) < 0) {
1795 ret = tracing_open_generic(inode, filp);
1797 trace_array_put(tr);
1804 static int system_tr_open(struct inode *inode, struct file *filp)
1806 struct trace_subsystem_dir *dir;
1807 struct trace_array *tr = inode->i_private;
1810 /* Make a temporary dir that has no system but points to tr */
1811 dir = kzalloc(sizeof(*dir), GFP_KERNEL);
1815 ret = tracing_open_generic_tr(inode, filp);
1821 filp->private_data = dir;
1826 static int subsystem_release(struct inode *inode, struct file *file)
1828 struct trace_subsystem_dir *dir = file->private_data;
1830 trace_array_put(dir->tr);
1833 * If dir->subsystem is NULL, then this is a temporary
1834 * descriptor that was made for a trace_array to enable
1846 subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1849 struct trace_subsystem_dir *dir = filp->private_data;
1850 struct event_subsystem *system = dir->subsystem;
1851 struct trace_seq *s;
1857 s = kmalloc(sizeof(*s), GFP_KERNEL);
1863 print_subsystem_event_filter(system, s);
1864 r = simple_read_from_buffer(ubuf, cnt, ppos,
1865 s->buffer, trace_seq_used(s));
1873 subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1876 struct trace_subsystem_dir *dir = filp->private_data;
1880 if (cnt >= PAGE_SIZE)
1883 buf = memdup_user_nul(ubuf, cnt);
1885 return PTR_ERR(buf);
1887 err = apply_subsystem_event_filter(dir, buf);
1898 show_header_page_file(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1900 struct trace_array *tr = filp->private_data;
1901 struct trace_seq *s;
1907 s = kmalloc(sizeof(*s), GFP_KERNEL);
1913 ring_buffer_print_page_header(tr->array_buffer.buffer, s);
1914 r = simple_read_from_buffer(ubuf, cnt, ppos,
1915 s->buffer, trace_seq_used(s));
1923 show_header_event_file(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1925 struct trace_seq *s;
1931 s = kmalloc(sizeof(*s), GFP_KERNEL);
1937 ring_buffer_print_entry_header(s);
1938 r = simple_read_from_buffer(ubuf, cnt, ppos,
1939 s->buffer, trace_seq_used(s));
1946 static void ignore_task_cpu(void *data)
1948 struct trace_array *tr = data;
1949 struct trace_pid_list *pid_list;
1950 struct trace_pid_list *no_pid_list;
1953 * This function is called by on_each_cpu() while the
1954 * event_mutex is held.
1956 pid_list = rcu_dereference_protected(tr->filtered_pids,
1957 mutex_is_locked(&event_mutex));
1958 no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
1959 mutex_is_locked(&event_mutex));
1961 this_cpu_write(tr->array_buffer.data->ignore_pid,
1962 trace_ignore_this_task(pid_list, no_pid_list, current));
1965 static void register_pid_events(struct trace_array *tr)
1968 * Register a probe that is called before all other probes
1969 * to set ignore_pid if next or prev do not match.
1970 * Register a probe this is called after all other probes
1971 * to only keep ignore_pid set if next pid matches.
1973 register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_pre,
1975 register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_post,
1978 register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre,
1980 register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_post,
1983 register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre,
1985 register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post,
1988 register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_pre,
1990 register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_post,
1995 event_pid_write(struct file *filp, const char __user *ubuf,
1996 size_t cnt, loff_t *ppos, int type)
1998 struct seq_file *m = filp->private_data;
1999 struct trace_array *tr = m->private;
2000 struct trace_pid_list *filtered_pids = NULL;
2001 struct trace_pid_list *other_pids = NULL;
2002 struct trace_pid_list *pid_list;
2003 struct trace_event_file *file;
2009 ret = tracing_update_buffers(tr);
2013 mutex_lock(&event_mutex);
2015 if (type == TRACE_PIDS) {
2016 filtered_pids = rcu_dereference_protected(tr->filtered_pids,
2017 lockdep_is_held(&event_mutex));
2018 other_pids = rcu_dereference_protected(tr->filtered_no_pids,
2019 lockdep_is_held(&event_mutex));
2021 filtered_pids = rcu_dereference_protected(tr->filtered_no_pids,
2022 lockdep_is_held(&event_mutex));
2023 other_pids = rcu_dereference_protected(tr->filtered_pids,
2024 lockdep_is_held(&event_mutex));
2027 ret = trace_pid_write(filtered_pids, &pid_list, ubuf, cnt);
2031 if (type == TRACE_PIDS)
2032 rcu_assign_pointer(tr->filtered_pids, pid_list);
2034 rcu_assign_pointer(tr->filtered_no_pids, pid_list);
2036 list_for_each_entry(file, &tr->events, list) {
2037 set_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
2040 if (filtered_pids) {
2041 tracepoint_synchronize_unregister();
2042 trace_pid_list_free(filtered_pids);
2043 } else if (pid_list && !other_pids) {
2044 register_pid_events(tr);
2048 * Ignoring of pids is done at task switch. But we have to
2049 * check for those tasks that are currently running.
2050 * Always do this in case a pid was appended or removed.
2052 on_each_cpu(ignore_task_cpu, tr, 1);
2055 mutex_unlock(&event_mutex);
2064 ftrace_event_pid_write(struct file *filp, const char __user *ubuf,
2065 size_t cnt, loff_t *ppos)
2067 return event_pid_write(filp, ubuf, cnt, ppos, TRACE_PIDS);
2071 ftrace_event_npid_write(struct file *filp, const char __user *ubuf,
2072 size_t cnt, loff_t *ppos)
2074 return event_pid_write(filp, ubuf, cnt, ppos, TRACE_NO_PIDS);
2077 static int ftrace_event_avail_open(struct inode *inode, struct file *file);
2078 static int ftrace_event_set_open(struct inode *inode, struct file *file);
2079 static int ftrace_event_set_pid_open(struct inode *inode, struct file *file);
2080 static int ftrace_event_set_npid_open(struct inode *inode, struct file *file);
2081 static int ftrace_event_release(struct inode *inode, struct file *file);
2083 static const struct seq_operations show_event_seq_ops = {
2090 static const struct seq_operations show_set_event_seq_ops = {
2097 static const struct seq_operations show_set_pid_seq_ops = {
2100 .show = trace_pid_show,
2104 static const struct seq_operations show_set_no_pid_seq_ops = {
2107 .show = trace_pid_show,
2111 static const struct file_operations ftrace_avail_fops = {
2112 .open = ftrace_event_avail_open,
2114 .llseek = seq_lseek,
2115 .release = seq_release,
2118 static const struct file_operations ftrace_set_event_fops = {
2119 .open = ftrace_event_set_open,
2121 .write = ftrace_event_write,
2122 .llseek = seq_lseek,
2123 .release = ftrace_event_release,
2126 static const struct file_operations ftrace_set_event_pid_fops = {
2127 .open = ftrace_event_set_pid_open,
2129 .write = ftrace_event_pid_write,
2130 .llseek = seq_lseek,
2131 .release = ftrace_event_release,
2134 static const struct file_operations ftrace_set_event_notrace_pid_fops = {
2135 .open = ftrace_event_set_npid_open,
2137 .write = ftrace_event_npid_write,
2138 .llseek = seq_lseek,
2139 .release = ftrace_event_release,
2142 static const struct file_operations ftrace_enable_fops = {
2143 .open = tracing_open_file_tr,
2144 .read = event_enable_read,
2145 .write = event_enable_write,
2146 .release = tracing_release_file_tr,
2147 .llseek = default_llseek,
2150 static const struct file_operations ftrace_event_format_fops = {
2151 .open = trace_format_open,
2153 .llseek = seq_lseek,
2154 .release = seq_release,
2157 #ifdef CONFIG_PERF_EVENTS
2158 static const struct file_operations ftrace_event_id_fops = {
2159 .read = event_id_read,
2160 .llseek = default_llseek,
2164 static const struct file_operations ftrace_event_filter_fops = {
2165 .open = tracing_open_file_tr,
2166 .read = event_filter_read,
2167 .write = event_filter_write,
2168 .release = tracing_release_file_tr,
2169 .llseek = default_llseek,
2172 static const struct file_operations ftrace_subsystem_filter_fops = {
2173 .open = subsystem_open,
2174 .read = subsystem_filter_read,
2175 .write = subsystem_filter_write,
2176 .llseek = default_llseek,
2177 .release = subsystem_release,
2180 static const struct file_operations ftrace_system_enable_fops = {
2181 .open = subsystem_open,
2182 .read = system_enable_read,
2183 .write = system_enable_write,
2184 .llseek = default_llseek,
2185 .release = subsystem_release,
2188 static const struct file_operations ftrace_tr_enable_fops = {
2189 .open = system_tr_open,
2190 .read = system_enable_read,
2191 .write = system_enable_write,
2192 .llseek = default_llseek,
2193 .release = subsystem_release,
2196 static const struct file_operations ftrace_show_header_page_fops = {
2197 .open = tracing_open_generic_tr,
2198 .read = show_header_page_file,
2199 .llseek = default_llseek,
2200 .release = tracing_release_generic_tr,
2203 static const struct file_operations ftrace_show_header_event_fops = {
2204 .open = tracing_open_generic_tr,
2205 .read = show_header_event_file,
2206 .llseek = default_llseek,
2207 .release = tracing_release_generic_tr,
2211 ftrace_event_open(struct inode *inode, struct file *file,
2212 const struct seq_operations *seq_ops)
2217 ret = security_locked_down(LOCKDOWN_TRACEFS);
2221 ret = seq_open(file, seq_ops);
2224 m = file->private_data;
2225 /* copy tr over to seq ops */
2226 m->private = inode->i_private;
2231 static int ftrace_event_release(struct inode *inode, struct file *file)
2233 struct trace_array *tr = inode->i_private;
2235 trace_array_put(tr);
2237 return seq_release(inode, file);
2241 ftrace_event_avail_open(struct inode *inode, struct file *file)
2243 const struct seq_operations *seq_ops = &show_event_seq_ops;
2245 /* Checks for tracefs lockdown */
2246 return ftrace_event_open(inode, file, seq_ops);
2250 ftrace_event_set_open(struct inode *inode, struct file *file)
2252 const struct seq_operations *seq_ops = &show_set_event_seq_ops;
2253 struct trace_array *tr = inode->i_private;
2256 ret = tracing_check_open_get_tr(tr);
2260 if ((file->f_mode & FMODE_WRITE) &&
2261 (file->f_flags & O_TRUNC))
2262 ftrace_clear_events(tr);
2264 ret = ftrace_event_open(inode, file, seq_ops);
2266 trace_array_put(tr);
2271 ftrace_event_set_pid_open(struct inode *inode, struct file *file)
2273 const struct seq_operations *seq_ops = &show_set_pid_seq_ops;
2274 struct trace_array *tr = inode->i_private;
2277 ret = tracing_check_open_get_tr(tr);
2281 if ((file->f_mode & FMODE_WRITE) &&
2282 (file->f_flags & O_TRUNC))
2283 ftrace_clear_event_pids(tr, TRACE_PIDS);
2285 ret = ftrace_event_open(inode, file, seq_ops);
2287 trace_array_put(tr);
2292 ftrace_event_set_npid_open(struct inode *inode, struct file *file)
2294 const struct seq_operations *seq_ops = &show_set_no_pid_seq_ops;
2295 struct trace_array *tr = inode->i_private;
2298 ret = tracing_check_open_get_tr(tr);
2302 if ((file->f_mode & FMODE_WRITE) &&
2303 (file->f_flags & O_TRUNC))
2304 ftrace_clear_event_pids(tr, TRACE_NO_PIDS);
2306 ret = ftrace_event_open(inode, file, seq_ops);
2308 trace_array_put(tr);
2312 static struct event_subsystem *
2313 create_new_subsystem(const char *name)
2315 struct event_subsystem *system;
2317 /* need to create new entry */
2318 system = kmalloc(sizeof(*system), GFP_KERNEL);
2322 system->ref_count = 1;
2324 /* Only allocate if dynamic (kprobes and modules) */
2325 system->name = kstrdup_const(name, GFP_KERNEL);
2329 system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
2330 if (!system->filter)
2333 list_add(&system->list, &event_subsystems);
2338 kfree_const(system->name);
2343 static int system_callback(const char *name, umode_t *mode, void **data,
2344 const struct file_operations **fops)
2346 if (strcmp(name, "filter") == 0)
2347 *fops = &ftrace_subsystem_filter_fops;
2349 else if (strcmp(name, "enable") == 0)
2350 *fops = &ftrace_system_enable_fops;
2355 *mode = TRACE_MODE_WRITE;
2359 static struct eventfs_inode *
2360 event_subsystem_dir(struct trace_array *tr, const char *name,
2361 struct trace_event_file *file, struct eventfs_inode *parent)
2363 struct event_subsystem *system, *iter;
2364 struct trace_subsystem_dir *dir;
2365 struct eventfs_inode *ei;
2367 static struct eventfs_entry system_entries[] = {
2370 .callback = system_callback,
2374 .callback = system_callback,
2378 /* First see if we did not already create this dir */
2379 list_for_each_entry(dir, &tr->systems, list) {
2380 system = dir->subsystem;
2381 if (strcmp(system->name, name) == 0) {
2388 /* Now see if the system itself exists. */
2390 list_for_each_entry(iter, &event_subsystems, list) {
2391 if (strcmp(iter->name, name) == 0) {
2397 dir = kmalloc(sizeof(*dir), GFP_KERNEL);
2402 system = create_new_subsystem(name);
2406 __get_system(system);
2408 /* ftrace only has directories no files */
2409 if (strcmp(name, "ftrace") == 0)
2412 nr_entries = ARRAY_SIZE(system_entries);
2414 ei = eventfs_create_dir(name, parent, system_entries, nr_entries, dir);
2416 pr_warn("Failed to create system directory %s\n", name);
2417 __put_system(system);
2425 dir->subsystem = system;
2428 list_add(&dir->list, &tr->systems);
2435 /* Only print this message if failed on memory allocation */
2436 if (!dir || !system)
2437 pr_warn("No memory to create event subsystem %s\n", name);
2442 event_define_fields(struct trace_event_call *call)
2444 struct list_head *head;
2448 * Other events may have the same class. Only update
2449 * the fields if they are not already defined.
2451 head = trace_get_fields(call);
2452 if (list_empty(head)) {
2453 struct trace_event_fields *field = call->class->fields_array;
2454 unsigned int offset = sizeof(struct trace_entry);
2456 for (; field->type; field++) {
2457 if (field->type == TRACE_FUNCTION_TYPE) {
2458 field->define_fields(call);
2462 offset = ALIGN(offset, field->align);
2463 ret = trace_define_field_ext(call, field->type, field->name,
2464 offset, field->size,
2465 field->is_signed, field->filter_type,
2467 if (WARN_ON_ONCE(ret)) {
2468 pr_err("error code is %d\n", ret);
2472 offset += field->size;
2479 static int event_callback(const char *name, umode_t *mode, void **data,
2480 const struct file_operations **fops)
2482 struct trace_event_file *file = *data;
2483 struct trace_event_call *call = file->event_call;
2485 if (strcmp(name, "format") == 0) {
2486 *mode = TRACE_MODE_READ;
2487 *fops = &ftrace_event_format_fops;
2493 * Only event directories that can be enabled should have
2494 * triggers or filters, with the exception of the "print"
2495 * event that can have a "trigger" file.
2497 if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)) {
2498 if (call->class->reg && strcmp(name, "enable") == 0) {
2499 *mode = TRACE_MODE_WRITE;
2500 *fops = &ftrace_enable_fops;
2504 if (strcmp(name, "filter") == 0) {
2505 *mode = TRACE_MODE_WRITE;
2506 *fops = &ftrace_event_filter_fops;
2511 if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE) ||
2512 strcmp(trace_event_name(call), "print") == 0) {
2513 if (strcmp(name, "trigger") == 0) {
2514 *mode = TRACE_MODE_WRITE;
2515 *fops = &event_trigger_fops;
2520 #ifdef CONFIG_PERF_EVENTS
2521 if (call->event.type && call->class->reg &&
2522 strcmp(name, "id") == 0) {
2523 *mode = TRACE_MODE_READ;
2524 *data = (void *)(long)call->event.type;
2525 *fops = &ftrace_event_id_fops;
2530 #ifdef CONFIG_HIST_TRIGGERS
2531 if (strcmp(name, "hist") == 0) {
2532 *mode = TRACE_MODE_READ;
2533 *fops = &event_hist_fops;
2537 #ifdef CONFIG_HIST_TRIGGERS_DEBUG
2538 if (strcmp(name, "hist_debug") == 0) {
2539 *mode = TRACE_MODE_READ;
2540 *fops = &event_hist_debug_fops;
2544 #ifdef CONFIG_TRACE_EVENT_INJECT
2545 if (call->event.type && call->class->reg &&
2546 strcmp(name, "inject") == 0) {
2548 *fops = &event_inject_fops;
2555 /* The file is incremented on creation and freeing the enable file decrements it */
2556 static void event_release(const char *name, void *data)
2558 struct trace_event_file *file = data;
2560 event_file_put(file);
2564 event_create_dir(struct eventfs_inode *parent, struct trace_event_file *file)
2566 struct trace_event_call *call = file->event_call;
2567 struct trace_array *tr = file->tr;
2568 struct eventfs_inode *e_events;
2569 struct eventfs_inode *ei;
2573 static struct eventfs_entry event_entries[] = {
2576 .callback = event_callback,
2577 .release = event_release,
2581 .callback = event_callback,
2585 .callback = event_callback,
2589 .callback = event_callback,
2591 #ifdef CONFIG_PERF_EVENTS
2594 .callback = event_callback,
2597 #ifdef CONFIG_HIST_TRIGGERS
2600 .callback = event_callback,
2603 #ifdef CONFIG_HIST_TRIGGERS_DEBUG
2605 .name = "hist_debug",
2606 .callback = event_callback,
2609 #ifdef CONFIG_TRACE_EVENT_INJECT
2612 .callback = event_callback,
2618 * If the trace point header did not define TRACE_SYSTEM
2619 * then the system would be called "TRACE_SYSTEM". This should
2622 if (WARN_ON_ONCE(strcmp(call->class->system, TRACE_SYSTEM) == 0))
2625 e_events = event_subsystem_dir(tr, call->class->system, file, parent);
2629 nr_entries = ARRAY_SIZE(event_entries);
2631 name = trace_event_name(call);
2632 ei = eventfs_create_dir(name, e_events, event_entries, nr_entries, file);
2634 pr_warn("Could not create tracefs '%s' directory\n", name);
2640 ret = event_define_fields(call);
2642 pr_warn("Could not initialize trace point events/%s\n", name);
2646 /* Gets decremented on freeing of the "enable" file */
2647 event_file_get(file);
2652 static void remove_event_from_tracers(struct trace_event_call *call)
2654 struct trace_event_file *file;
2655 struct trace_array *tr;
2657 do_for_each_event_file_safe(tr, file) {
2658 if (file->event_call != call)
2661 remove_event_file_dir(file);
2663 * The do_for_each_event_file_safe() is
2664 * a double loop. After finding the call for this
2665 * trace_array, we use break to jump to the next
2669 } while_for_each_event_file();
2672 static void event_remove(struct trace_event_call *call)
2674 struct trace_array *tr;
2675 struct trace_event_file *file;
2677 do_for_each_event_file(tr, file) {
2678 if (file->event_call != call)
2681 if (file->flags & EVENT_FILE_FL_WAS_ENABLED)
2682 tr->clear_trace = true;
2684 ftrace_event_enable_disable(file, 0);
2686 * The do_for_each_event_file() is
2687 * a double loop. After finding the call for this
2688 * trace_array, we use break to jump to the next
2692 } while_for_each_event_file();
2694 if (call->event.funcs)
2695 __unregister_trace_event(&call->event);
2696 remove_event_from_tracers(call);
2697 list_del(&call->list);
2700 static int event_init(struct trace_event_call *call)
2705 name = trace_event_name(call);
2709 if (call->class->raw_init) {
2710 ret = call->class->raw_init(call);
2711 if (ret < 0 && ret != -ENOSYS)
2712 pr_warn("Could not initialize trace events/%s\n", name);
2719 __register_event(struct trace_event_call *call, struct module *mod)
2723 ret = event_init(call);
2727 list_add(&call->list, &ftrace_events);
2728 if (call->flags & TRACE_EVENT_FL_DYNAMIC)
2729 atomic_set(&call->refcnt, 0);
2736 static char *eval_replace(char *ptr, struct trace_eval_map *map, int len)
2741 /* Find the length of the eval value as a string */
2742 elen = snprintf(ptr, 0, "%ld", map->eval_value);
2743 /* Make sure there's enough room to replace the string with the value */
2747 snprintf(ptr, elen + 1, "%ld", map->eval_value);
2749 /* Get the rest of the string of ptr */
2750 rlen = strlen(ptr + len);
2751 memmove(ptr + elen, ptr + len, rlen);
2752 /* Make sure we end the new string */
2753 ptr[elen + rlen] = 0;
2758 static void update_event_printk(struct trace_event_call *call,
2759 struct trace_eval_map *map)
2763 int len = strlen(map->eval_string);
2765 for (ptr = call->print_fmt; *ptr; ptr++) {
2779 if (isdigit(*ptr)) {
2783 /* Check for alpha chars like ULL */
2784 } while (isalnum(*ptr));
2788 * A number must have some kind of delimiter after
2789 * it, and we can ignore that too.
2793 if (isalpha(*ptr) || *ptr == '_') {
2794 if (strncmp(map->eval_string, ptr, len) == 0 &&
2795 !isalnum(ptr[len]) && ptr[len] != '_') {
2796 ptr = eval_replace(ptr, map, len);
2797 /* enum/sizeof string smaller than value */
2798 if (WARN_ON_ONCE(!ptr))
2801 * No need to decrement here, as eval_replace()
2802 * returns the pointer to the character passed
2803 * the eval, and two evals can not be placed
2804 * back to back without something in between.
2805 * We can skip that something in between.
2812 } while (isalnum(*ptr) || *ptr == '_');
2816 * If what comes after this variable is a '.' or
2817 * '->' then we can continue to ignore that string.
2819 if (*ptr == '.' || (ptr[0] == '-' && ptr[1] == '>')) {
2820 ptr += *ptr == '.' ? 1 : 2;
2826 * Once again, we can skip the delimiter that came
2834 static void add_str_to_module(struct module *module, char *str)
2836 struct module_string *modstr;
2838 modstr = kmalloc(sizeof(*modstr), GFP_KERNEL);
2841 * If we failed to allocate memory here, then we'll just
2842 * let the str memory leak when the module is removed.
2843 * If this fails to allocate, there's worse problems than
2844 * a leaked string on module removal.
2846 if (WARN_ON_ONCE(!modstr))
2849 modstr->module = module;
2852 list_add(&modstr->next, &module_strings);
2855 static void update_event_fields(struct trace_event_call *call,
2856 struct trace_eval_map *map)
2858 struct ftrace_event_field *field;
2859 struct list_head *head;
2862 int len = strlen(map->eval_string);
2864 /* Dynamic events should never have field maps */
2865 if (WARN_ON_ONCE(call->flags & TRACE_EVENT_FL_DYNAMIC))
2868 head = trace_get_fields(call);
2869 list_for_each_entry(field, head, link) {
2870 ptr = strchr(field->type, '[');
2875 if (!isalpha(*ptr) && *ptr != '_')
2878 if (strncmp(map->eval_string, ptr, len) != 0)
2881 str = kstrdup(field->type, GFP_KERNEL);
2882 if (WARN_ON_ONCE(!str))
2884 ptr = str + (ptr - field->type);
2885 ptr = eval_replace(ptr, map, len);
2886 /* enum/sizeof string smaller than value */
2887 if (WARN_ON_ONCE(!ptr)) {
2893 * If the event is part of a module, then we need to free the string
2894 * when the module is removed. Otherwise, it will stay allocated
2898 add_str_to_module(call->module, str);
2904 void trace_event_eval_update(struct trace_eval_map **map, int len)
2906 struct trace_event_call *call, *p;
2907 const char *last_system = NULL;
2912 down_write(&trace_event_sem);
2913 list_for_each_entry_safe(call, p, &ftrace_events, list) {
2914 /* events are usually grouped together with systems */
2915 if (!last_system || call->class->system != last_system) {
2918 last_system = call->class->system;
2922 * Since calls are grouped by systems, the likelihood that the
2923 * next call in the iteration belongs to the same system as the
2924 * previous call is high. As an optimization, we skip searching
2925 * for a map[] that matches the call's system if the last call
2926 * was from the same system. That's what last_i is for. If the
2927 * call has the same system as the previous call, then last_i
2928 * will be the index of the first map[] that has a matching
2931 for (i = last_i; i < len; i++) {
2932 if (call->class->system == map[i]->system) {
2933 /* Save the first system if need be */
2938 update_event_printk(call, map[i]);
2939 update_event_fields(call, map[i]);
2944 up_write(&trace_event_sem);
2947 static bool event_in_systems(struct trace_event_call *call,
2948 const char *systems)
2956 system = call->class->system;
2957 p = strstr(systems, system);
2961 if (p != systems && !isspace(*(p - 1)) && *(p - 1) != ',')
2964 p += strlen(system);
2965 return !*p || isspace(*p) || *p == ',';
2968 static struct trace_event_file *
2969 trace_create_new_event(struct trace_event_call *call,
2970 struct trace_array *tr)
2972 struct trace_pid_list *no_pid_list;
2973 struct trace_pid_list *pid_list;
2974 struct trace_event_file *file;
2977 if (!event_in_systems(call, tr->system_names))
2980 file = kmem_cache_alloc(file_cachep, GFP_TRACE);
2982 return ERR_PTR(-ENOMEM);
2984 pid_list = rcu_dereference_protected(tr->filtered_pids,
2985 lockdep_is_held(&event_mutex));
2986 no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
2987 lockdep_is_held(&event_mutex));
2989 if (!trace_pid_list_first(pid_list, &first) ||
2990 !trace_pid_list_first(no_pid_list, &first))
2991 file->flags |= EVENT_FILE_FL_PID_FILTER;
2993 file->event_call = call;
2995 atomic_set(&file->sm_ref, 0);
2996 atomic_set(&file->tm_ref, 0);
2997 INIT_LIST_HEAD(&file->triggers);
2998 list_add(&file->list, &tr->events);
2999 event_file_get(file);
3004 #define MAX_BOOT_TRIGGERS 32
3006 static struct boot_triggers {
3009 } bootup_triggers[MAX_BOOT_TRIGGERS];
3011 static char bootup_trigger_buf[COMMAND_LINE_SIZE];
3012 static int nr_boot_triggers;
3014 static __init int setup_trace_triggers(char *str)
3020 strscpy(bootup_trigger_buf, str, COMMAND_LINE_SIZE);
3021 trace_set_ring_buffer_expanded(NULL);
3022 disable_tracing_selftest("running event triggers");
3024 buf = bootup_trigger_buf;
3025 for (i = 0; i < MAX_BOOT_TRIGGERS; i++) {
3026 trigger = strsep(&buf, ",");
3029 bootup_triggers[i].event = strsep(&trigger, ".");
3030 bootup_triggers[i].trigger = trigger;
3031 if (!bootup_triggers[i].trigger)
3035 nr_boot_triggers = i;
3038 __setup("trace_trigger=", setup_trace_triggers);
3040 /* Add an event to a trace directory */
3042 __trace_add_new_event(struct trace_event_call *call, struct trace_array *tr)
3044 struct trace_event_file *file;
3046 file = trace_create_new_event(call, tr);
3048 * trace_create_new_event() returns ERR_PTR(-ENOMEM) if failed
3049 * allocation, or NULL if the event is not part of the tr->system_names.
3050 * When the event is not part of the tr->system_names, return zero, not
3057 return PTR_ERR(file);
3059 if (eventdir_initialized)
3060 return event_create_dir(tr->event_dir, file);
3062 return event_define_fields(call);
3065 static void trace_early_triggers(struct trace_event_file *file, const char *name)
3070 for (i = 0; i < nr_boot_triggers; i++) {
3071 if (strcmp(name, bootup_triggers[i].event))
3073 mutex_lock(&event_mutex);
3074 ret = trigger_process_regex(file, bootup_triggers[i].trigger);
3075 mutex_unlock(&event_mutex);
3077 pr_err("Failed to register trigger '%s' on event %s\n",
3078 bootup_triggers[i].trigger,
3079 bootup_triggers[i].event);
3084 * Just create a descriptor for early init. A descriptor is required
3085 * for enabling events at boot. We want to enable events before
3086 * the filesystem is initialized.
3089 __trace_early_add_new_event(struct trace_event_call *call,
3090 struct trace_array *tr)
3092 struct trace_event_file *file;
3095 file = trace_create_new_event(call, tr);
3097 * trace_create_new_event() returns ERR_PTR(-ENOMEM) if failed
3098 * allocation, or NULL if the event is not part of the tr->system_names.
3099 * When the event is not part of the tr->system_names, return zero, not
3106 return PTR_ERR(file);
3108 ret = event_define_fields(call);
3112 trace_early_triggers(file, trace_event_name(call));
3117 struct ftrace_module_file_ops;
3118 static void __add_event_to_tracers(struct trace_event_call *call);
3120 /* Add an additional event_call dynamically */
3121 int trace_add_event_call(struct trace_event_call *call)
3124 lockdep_assert_held(&event_mutex);
3126 mutex_lock(&trace_types_lock);
3128 ret = __register_event(call, NULL);
3130 __add_event_to_tracers(call);
3132 mutex_unlock(&trace_types_lock);
3135 EXPORT_SYMBOL_GPL(trace_add_event_call);
3138 * Must be called under locking of trace_types_lock, event_mutex and
3141 static void __trace_remove_event_call(struct trace_event_call *call)
3144 trace_destroy_fields(call);
3145 free_event_filter(call->filter);
3146 call->filter = NULL;
3149 static int probe_remove_event_call(struct trace_event_call *call)
3151 struct trace_array *tr;
3152 struct trace_event_file *file;
3154 #ifdef CONFIG_PERF_EVENTS
3155 if (call->perf_refcount)
3158 do_for_each_event_file(tr, file) {
3159 if (file->event_call != call)
3162 * We can't rely on ftrace_event_enable_disable(enable => 0)
3163 * we are going to do, EVENT_FILE_FL_SOFT_MODE can suppress
3164 * TRACE_REG_UNREGISTER.
3166 if (file->flags & EVENT_FILE_FL_ENABLED)
3169 if (file->flags & EVENT_FILE_FL_WAS_ENABLED)
3170 tr->clear_trace = true;
3172 * The do_for_each_event_file_safe() is
3173 * a double loop. After finding the call for this
3174 * trace_array, we use break to jump to the next
3178 } while_for_each_event_file();
3180 __trace_remove_event_call(call);
3184 /* No need to clear the trace now */
3185 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
3186 tr->clear_trace = false;
3191 /* Remove an event_call */
3192 int trace_remove_event_call(struct trace_event_call *call)
3196 lockdep_assert_held(&event_mutex);
3198 mutex_lock(&trace_types_lock);
3199 down_write(&trace_event_sem);
3200 ret = probe_remove_event_call(call);
3201 up_write(&trace_event_sem);
3202 mutex_unlock(&trace_types_lock);
3206 EXPORT_SYMBOL_GPL(trace_remove_event_call);
3208 #define for_each_event(event, start, end) \
3209 for (event = start; \
3210 (unsigned long)event < (unsigned long)end; \
3213 #ifdef CONFIG_MODULES
3215 static void trace_module_add_events(struct module *mod)
3217 struct trace_event_call **call, **start, **end;
3219 if (!mod->num_trace_events)
3222 /* Don't add infrastructure for mods without tracepoints */
3223 if (trace_module_has_bad_taint(mod)) {
3224 pr_err("%s: module has bad taint, not creating trace events\n",
3229 start = mod->trace_events;
3230 end = mod->trace_events + mod->num_trace_events;
3232 for_each_event(call, start, end) {
3233 __register_event(*call, mod);
3234 __add_event_to_tracers(*call);
3238 static void trace_module_remove_events(struct module *mod)
3240 struct trace_event_call *call, *p;
3241 struct module_string *modstr, *m;
3243 down_write(&trace_event_sem);
3244 list_for_each_entry_safe(call, p, &ftrace_events, list) {
3245 if ((call->flags & TRACE_EVENT_FL_DYNAMIC) || !call->module)
3247 if (call->module == mod)
3248 __trace_remove_event_call(call);
3250 /* Check for any strings allocade for this module */
3251 list_for_each_entry_safe(modstr, m, &module_strings, next) {
3252 if (modstr->module != mod)
3254 list_del(&modstr->next);
3258 up_write(&trace_event_sem);
3261 * It is safest to reset the ring buffer if the module being unloaded
3262 * registered any events that were used. The only worry is if
3263 * a new module gets loaded, and takes on the same id as the events
3264 * of this module. When printing out the buffer, traced events left
3265 * over from this module may be passed to the new module events and
3266 * unexpected results may occur.
3268 tracing_reset_all_online_cpus_unlocked();
3271 static int trace_module_notify(struct notifier_block *self,
3272 unsigned long val, void *data)
3274 struct module *mod = data;
3276 mutex_lock(&event_mutex);
3277 mutex_lock(&trace_types_lock);
3279 case MODULE_STATE_COMING:
3280 trace_module_add_events(mod);
3282 case MODULE_STATE_GOING:
3283 trace_module_remove_events(mod);
3286 mutex_unlock(&trace_types_lock);
3287 mutex_unlock(&event_mutex);
3292 static struct notifier_block trace_module_nb = {
3293 .notifier_call = trace_module_notify,
3294 .priority = 1, /* higher than trace.c module notify */
3296 #endif /* CONFIG_MODULES */
3298 /* Create a new event directory structure for a trace directory. */
3300 __trace_add_event_dirs(struct trace_array *tr)
3302 struct trace_event_call *call;
3305 list_for_each_entry(call, &ftrace_events, list) {
3306 ret = __trace_add_new_event(call, tr);
3308 pr_warn("Could not create directory for event %s\n",
3309 trace_event_name(call));
3313 /* Returns any file that matches the system and event */
3314 struct trace_event_file *
3315 __find_event_file(struct trace_array *tr, const char *system, const char *event)
3317 struct trace_event_file *file;
3318 struct trace_event_call *call;
3321 list_for_each_entry(file, &tr->events, list) {
3323 call = file->event_call;
3324 name = trace_event_name(call);
3326 if (!name || !call->class)
3329 if (strcmp(event, name) == 0 &&
3330 strcmp(system, call->class->system) == 0)
3336 /* Returns valid trace event files that match system and event */
3337 struct trace_event_file *
3338 find_event_file(struct trace_array *tr, const char *system, const char *event)
3340 struct trace_event_file *file;
3342 file = __find_event_file(tr, system, event);
3343 if (!file || !file->event_call->class->reg ||
3344 file->event_call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
3351 * trace_get_event_file - Find and return a trace event file
3352 * @instance: The name of the trace instance containing the event
3353 * @system: The name of the system containing the event
3354 * @event: The name of the event
3356 * Return a trace event file given the trace instance name, trace
3357 * system, and trace event name. If the instance name is NULL, it
3358 * refers to the top-level trace array.
3360 * This function will look it up and return it if found, after calling
3361 * trace_array_get() to prevent the instance from going away, and
3362 * increment the event's module refcount to prevent it from being
3365 * To release the file, call trace_put_event_file(), which will call
3366 * trace_array_put() and decrement the event's module refcount.
3368 * Return: The trace event on success, ERR_PTR otherwise.
3370 struct trace_event_file *trace_get_event_file(const char *instance,
3374 struct trace_array *tr = top_trace_array();
3375 struct trace_event_file *file = NULL;
3379 tr = trace_array_find_get(instance);
3381 return ERR_PTR(-ENOENT);
3383 ret = trace_array_get(tr);
3385 return ERR_PTR(ret);
3388 mutex_lock(&event_mutex);
3390 file = find_event_file(tr, system, event);
3392 trace_array_put(tr);
3397 /* Don't let event modules unload while in use */
3398 ret = trace_event_try_get_ref(file->event_call);
3400 trace_array_put(tr);
3407 mutex_unlock(&event_mutex);
3410 file = ERR_PTR(ret);
3414 EXPORT_SYMBOL_GPL(trace_get_event_file);
3417 * trace_put_event_file - Release a file from trace_get_event_file()
3418 * @file: The trace event file
3420 * If a file was retrieved using trace_get_event_file(), this should
3421 * be called when it's no longer needed. It will cancel the previous
3422 * trace_array_get() called by that function, and decrement the
3423 * event's module refcount.
3425 void trace_put_event_file(struct trace_event_file *file)
3427 mutex_lock(&event_mutex);
3428 trace_event_put_ref(file->event_call);
3429 mutex_unlock(&event_mutex);
3431 trace_array_put(file->tr);
3433 EXPORT_SYMBOL_GPL(trace_put_event_file);
3435 #ifdef CONFIG_DYNAMIC_FTRACE
3438 #define ENABLE_EVENT_STR "enable_event"
3439 #define DISABLE_EVENT_STR "disable_event"
3441 struct event_probe_data {
3442 struct trace_event_file *file;
3443 unsigned long count;
3448 static void update_event_probe(struct event_probe_data *data)
3451 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
3453 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
3457 event_enable_probe(unsigned long ip, unsigned long parent_ip,
3458 struct trace_array *tr, struct ftrace_probe_ops *ops,
3461 struct ftrace_func_mapper *mapper = data;
3462 struct event_probe_data *edata;
3465 pdata = ftrace_func_mapper_find_ip(mapper, ip);
3466 if (!pdata || !*pdata)
3470 update_event_probe(edata);
3474 event_enable_count_probe(unsigned long ip, unsigned long parent_ip,
3475 struct trace_array *tr, struct ftrace_probe_ops *ops,
3478 struct ftrace_func_mapper *mapper = data;
3479 struct event_probe_data *edata;
3482 pdata = ftrace_func_mapper_find_ip(mapper, ip);
3483 if (!pdata || !*pdata)
3491 /* Skip if the event is in a state we want to switch to */
3492 if (edata->enable == !(edata->file->flags & EVENT_FILE_FL_SOFT_DISABLED))
3495 if (edata->count != -1)
3498 update_event_probe(edata);
3502 event_enable_print(struct seq_file *m, unsigned long ip,
3503 struct ftrace_probe_ops *ops, void *data)
3505 struct ftrace_func_mapper *mapper = data;
3506 struct event_probe_data *edata;
3509 pdata = ftrace_func_mapper_find_ip(mapper, ip);
3511 if (WARN_ON_ONCE(!pdata || !*pdata))
3516 seq_printf(m, "%ps:", (void *)ip);
3518 seq_printf(m, "%s:%s:%s",
3519 edata->enable ? ENABLE_EVENT_STR : DISABLE_EVENT_STR,
3520 edata->file->event_call->class->system,
3521 trace_event_name(edata->file->event_call));
3523 if (edata->count == -1)
3524 seq_puts(m, ":unlimited\n");
3526 seq_printf(m, ":count=%ld\n", edata->count);
3532 event_enable_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
3533 unsigned long ip, void *init_data, void **data)
3535 struct ftrace_func_mapper *mapper = *data;
3536 struct event_probe_data *edata = init_data;
3540 mapper = allocate_ftrace_func_mapper();
3546 ret = ftrace_func_mapper_add_ip(mapper, ip, edata);
3555 static int free_probe_data(void *data)
3557 struct event_probe_data *edata = data;
3561 /* Remove the SOFT_MODE flag */
3562 __ftrace_event_enable_disable(edata->file, 0, 1);
3563 trace_event_put_ref(edata->file->event_call);
3570 event_enable_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
3571 unsigned long ip, void *data)
3573 struct ftrace_func_mapper *mapper = data;
3574 struct event_probe_data *edata;
3579 free_ftrace_func_mapper(mapper, free_probe_data);
3583 edata = ftrace_func_mapper_remove_ip(mapper, ip);
3585 if (WARN_ON_ONCE(!edata))
3588 if (WARN_ON_ONCE(edata->ref <= 0))
3591 free_probe_data(edata);
3594 static struct ftrace_probe_ops event_enable_probe_ops = {
3595 .func = event_enable_probe,
3596 .print = event_enable_print,
3597 .init = event_enable_init,
3598 .free = event_enable_free,
3601 static struct ftrace_probe_ops event_enable_count_probe_ops = {
3602 .func = event_enable_count_probe,
3603 .print = event_enable_print,
3604 .init = event_enable_init,
3605 .free = event_enable_free,
3608 static struct ftrace_probe_ops event_disable_probe_ops = {
3609 .func = event_enable_probe,
3610 .print = event_enable_print,
3611 .init = event_enable_init,
3612 .free = event_enable_free,
3615 static struct ftrace_probe_ops event_disable_count_probe_ops = {
3616 .func = event_enable_count_probe,
3617 .print = event_enable_print,
3618 .init = event_enable_init,
3619 .free = event_enable_free,
3623 event_enable_func(struct trace_array *tr, struct ftrace_hash *hash,
3624 char *glob, char *cmd, char *param, int enabled)
3626 struct trace_event_file *file;
3627 struct ftrace_probe_ops *ops;
3628 struct event_probe_data *data;
3638 /* hash funcs only work with set_ftrace_filter */
3639 if (!enabled || !param)
3642 system = strsep(¶m, ":");
3646 event = strsep(¶m, ":");
3648 mutex_lock(&event_mutex);
3651 file = find_event_file(tr, system, event);
3655 enable = strcmp(cmd, ENABLE_EVENT_STR) == 0;
3658 ops = param ? &event_enable_count_probe_ops : &event_enable_probe_ops;
3660 ops = param ? &event_disable_count_probe_ops : &event_disable_probe_ops;
3662 if (glob[0] == '!') {
3663 ret = unregister_ftrace_function_probe_func(glob+1, tr, ops);
3669 data = kzalloc(sizeof(*data), GFP_KERNEL);
3673 data->enable = enable;
3680 number = strsep(¶m, ":");
3683 if (!strlen(number))
3687 * We use the callback data field (which is a pointer)
3690 ret = kstrtoul(number, 0, &data->count);
3695 /* Don't let event modules unload while probe registered */
3696 ret = trace_event_try_get_ref(file->event_call);
3702 ret = __ftrace_event_enable_disable(file, 1, 1);
3706 ret = register_ftrace_function_probe(glob, tr, ops, data);
3708 * The above returns on success the # of functions enabled,
3709 * but if it didn't find any functions it returns zero.
3710 * Consider no functions a failure too.
3717 /* Just return zero, not the number of enabled functions */
3720 mutex_unlock(&event_mutex);
3724 __ftrace_event_enable_disable(file, 0, 1);
3726 trace_event_put_ref(file->event_call);
3732 static struct ftrace_func_command event_enable_cmd = {
3733 .name = ENABLE_EVENT_STR,
3734 .func = event_enable_func,
3737 static struct ftrace_func_command event_disable_cmd = {
3738 .name = DISABLE_EVENT_STR,
3739 .func = event_enable_func,
3742 static __init int register_event_cmds(void)
3746 ret = register_ftrace_command(&event_enable_cmd);
3747 if (WARN_ON(ret < 0))
3749 ret = register_ftrace_command(&event_disable_cmd);
3750 if (WARN_ON(ret < 0))
3751 unregister_ftrace_command(&event_enable_cmd);
3755 static inline int register_event_cmds(void) { return 0; }
3756 #endif /* CONFIG_DYNAMIC_FTRACE */
3759 * The top level array and trace arrays created by boot-time tracing
3760 * have already had its trace_event_file descriptors created in order
3761 * to allow for early events to be recorded.
3762 * This function is called after the tracefs has been initialized,
3763 * and we now have to create the files associated to the events.
3765 static void __trace_early_add_event_dirs(struct trace_array *tr)
3767 struct trace_event_file *file;
3771 list_for_each_entry(file, &tr->events, list) {
3772 ret = event_create_dir(tr->event_dir, file);
3774 pr_warn("Could not create directory for event %s\n",
3775 trace_event_name(file->event_call));
3780 * For early boot up, the top trace array and the trace arrays created
3781 * by boot-time tracing require to have a list of events that can be
3782 * enabled. This must be done before the filesystem is set up in order
3783 * to allow events to be traced early.
3785 void __trace_early_add_events(struct trace_array *tr)
3787 struct trace_event_call *call;
3790 list_for_each_entry(call, &ftrace_events, list) {
3791 /* Early boot up should not have any modules loaded */
3792 if (!(call->flags & TRACE_EVENT_FL_DYNAMIC) &&
3793 WARN_ON_ONCE(call->module))
3796 ret = __trace_early_add_new_event(call, tr);
3798 pr_warn("Could not create early event %s\n",
3799 trace_event_name(call));
3803 /* Remove the event directory structure for a trace directory. */
3805 __trace_remove_event_dirs(struct trace_array *tr)
3807 struct trace_event_file *file, *next;
3809 list_for_each_entry_safe(file, next, &tr->events, list)
3810 remove_event_file_dir(file);
3813 static void __add_event_to_tracers(struct trace_event_call *call)
3815 struct trace_array *tr;
3817 list_for_each_entry(tr, &ftrace_trace_arrays, list)
3818 __trace_add_new_event(call, tr);
3821 extern struct trace_event_call *__start_ftrace_events[];
3822 extern struct trace_event_call *__stop_ftrace_events[];
3824 static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
3826 static __init int setup_trace_event(char *str)
3828 strscpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
3829 trace_set_ring_buffer_expanded(NULL);
3830 disable_tracing_selftest("running event tracing");
3834 __setup("trace_event=", setup_trace_event);
3836 static int events_callback(const char *name, umode_t *mode, void **data,
3837 const struct file_operations **fops)
3839 if (strcmp(name, "enable") == 0) {
3840 *mode = TRACE_MODE_WRITE;
3841 *fops = &ftrace_tr_enable_fops;
3845 if (strcmp(name, "header_page") == 0) {
3846 *mode = TRACE_MODE_READ;
3847 *fops = &ftrace_show_header_page_fops;
3849 } else if (strcmp(name, "header_event") == 0) {
3850 *mode = TRACE_MODE_READ;
3851 *fops = &ftrace_show_header_event_fops;
3858 /* Expects to have event_mutex held when called */
3860 create_event_toplevel_files(struct dentry *parent, struct trace_array *tr)
3862 struct eventfs_inode *e_events;
3863 struct dentry *entry;
3865 static struct eventfs_entry events_entries[] = {
3868 .callback = events_callback,
3871 .name = "header_page",
3872 .callback = events_callback,
3875 .name = "header_event",
3876 .callback = events_callback,
3880 entry = trace_create_file("set_event", TRACE_MODE_WRITE, parent,
3881 tr, &ftrace_set_event_fops);
3885 nr_entries = ARRAY_SIZE(events_entries);
3887 e_events = eventfs_create_events_dir("events", parent, events_entries,
3889 if (IS_ERR(e_events)) {
3890 pr_warn("Could not create tracefs 'events' directory\n");
3894 /* There are not as crucial, just warn if they are not created */
3896 trace_create_file("set_event_pid", TRACE_MODE_WRITE, parent,
3897 tr, &ftrace_set_event_pid_fops);
3899 trace_create_file("set_event_notrace_pid",
3900 TRACE_MODE_WRITE, parent, tr,
3901 &ftrace_set_event_notrace_pid_fops);
3903 tr->event_dir = e_events;
3909 * event_trace_add_tracer - add a instance of a trace_array to events
3910 * @parent: The parent dentry to place the files/directories for events in
3911 * @tr: The trace array associated with these events
3913 * When a new instance is created, it needs to set up its events
3914 * directory, as well as other files associated with events. It also
3915 * creates the event hierarchy in the @parent/events directory.
3917 * Returns 0 on success.
3919 * Must be called with event_mutex held.
3921 int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr)
3925 lockdep_assert_held(&event_mutex);
3927 ret = create_event_toplevel_files(parent, tr);
3931 down_write(&trace_event_sem);
3932 /* If tr already has the event list, it is initialized in early boot. */
3933 if (unlikely(!list_empty(&tr->events)))
3934 __trace_early_add_event_dirs(tr);
3936 __trace_add_event_dirs(tr);
3937 up_write(&trace_event_sem);
3944 * The top trace array already had its file descriptors created.
3945 * Now the files themselves need to be created.
3948 early_event_add_tracer(struct dentry *parent, struct trace_array *tr)
3952 mutex_lock(&event_mutex);
3954 ret = create_event_toplevel_files(parent, tr);
3958 down_write(&trace_event_sem);
3959 __trace_early_add_event_dirs(tr);
3960 up_write(&trace_event_sem);
3963 mutex_unlock(&event_mutex);
3968 /* Must be called with event_mutex held */
3969 int event_trace_del_tracer(struct trace_array *tr)
3971 lockdep_assert_held(&event_mutex);
3973 /* Disable any event triggers and associated soft-disabled events */
3974 clear_event_triggers(tr);
3976 /* Clear the pid list */
3977 __ftrace_clear_event_pids(tr, TRACE_PIDS | TRACE_NO_PIDS);
3979 /* Disable any running events */
3980 __ftrace_set_clr_event_nolock(tr, NULL, NULL, NULL, 0);
3982 /* Make sure no more events are being executed */
3983 tracepoint_synchronize_unregister();
3985 down_write(&trace_event_sem);
3986 __trace_remove_event_dirs(tr);
3987 eventfs_remove_events_dir(tr->event_dir);
3988 up_write(&trace_event_sem);
3990 tr->event_dir = NULL;
3995 static __init int event_trace_memsetup(void)
3997 field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC);
3998 file_cachep = KMEM_CACHE(trace_event_file, SLAB_PANIC);
4003 early_enable_events(struct trace_array *tr, char *buf, bool disable_first)
4009 token = strsep(&buf, ",");
4015 /* Restarting syscalls requires that we stop them first */
4017 ftrace_set_clr_event(tr, token, 0);
4019 ret = ftrace_set_clr_event(tr, token, 1);
4021 pr_warn("Failed to enable trace event: %s\n", token);
4024 /* Put back the comma to allow this to be called again */
4030 static __init int event_trace_enable(void)
4032 struct trace_array *tr = top_trace_array();
4033 struct trace_event_call **iter, *call;
4039 for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
4042 ret = event_init(call);
4044 list_add(&call->list, &ftrace_events);
4047 register_trigger_cmds();
4050 * We need the top trace array to have a working set of trace
4051 * points at early init, before the debug files and directories
4052 * are created. Create the file entries now, and attach them
4053 * to the actual file dentries later.
4055 __trace_early_add_events(tr);
4057 early_enable_events(tr, bootup_event_buf, false);
4059 trace_printk_start_comm();
4061 register_event_cmds();
4068 * event_trace_enable() is called from trace_event_init() first to
4069 * initialize events and perhaps start any events that are on the
4070 * command line. Unfortunately, there are some events that will not
4071 * start this early, like the system call tracepoints that need
4072 * to set the %SYSCALL_WORK_SYSCALL_TRACEPOINT flag of pid 1. But
4073 * event_trace_enable() is called before pid 1 starts, and this flag
4074 * is never set, making the syscall tracepoint never get reached, but
4075 * the event is enabled regardless (and not doing anything).
4077 static __init int event_trace_enable_again(void)
4079 struct trace_array *tr;
4081 tr = top_trace_array();
4085 early_enable_events(tr, bootup_event_buf, true);
4090 early_initcall(event_trace_enable_again);
4092 /* Init fields which doesn't related to the tracefs */
4093 static __init int event_trace_init_fields(void)
4095 if (trace_define_generic_fields())
4096 pr_warn("tracing: Failed to allocated generic fields");
4098 if (trace_define_common_fields())
4099 pr_warn("tracing: Failed to allocate common fields");
4104 __init int event_trace_init(void)
4106 struct trace_array *tr;
4109 tr = top_trace_array();
4113 trace_create_file("available_events", TRACE_MODE_READ,
4114 NULL, tr, &ftrace_avail_fops);
4116 ret = early_event_add_tracer(NULL, tr);
4120 #ifdef CONFIG_MODULES
4121 ret = register_module_notifier(&trace_module_nb);
4123 pr_warn("Failed to register trace events module notifier\n");
4126 eventdir_initialized = true;
4131 void __init trace_event_init(void)
4133 event_trace_memsetup();
4134 init_ftrace_syscalls();
4135 event_trace_enable();
4136 event_trace_init_fields();
4139 #ifdef CONFIG_EVENT_TRACE_STARTUP_TEST
4141 static DEFINE_SPINLOCK(test_spinlock);
4142 static DEFINE_SPINLOCK(test_spinlock_irq);
4143 static DEFINE_MUTEX(test_mutex);
4145 static __init void test_work(struct work_struct *dummy)
4147 spin_lock(&test_spinlock);
4148 spin_lock_irq(&test_spinlock_irq);
4150 spin_unlock_irq(&test_spinlock_irq);
4151 spin_unlock(&test_spinlock);
4153 mutex_lock(&test_mutex);
4155 mutex_unlock(&test_mutex);
4158 static __init int event_test_thread(void *unused)
4162 test_malloc = kmalloc(1234, GFP_KERNEL);
4164 pr_info("failed to kmalloc\n");
4166 schedule_on_each_cpu(test_work);
4170 set_current_state(TASK_INTERRUPTIBLE);
4171 while (!kthread_should_stop()) {
4173 set_current_state(TASK_INTERRUPTIBLE);
4175 __set_current_state(TASK_RUNNING);
4181 * Do various things that may trigger events.
4183 static __init void event_test_stuff(void)
4185 struct task_struct *test_thread;
4187 test_thread = kthread_run(event_test_thread, NULL, "test-events");
4189 kthread_stop(test_thread);
4193 * For every trace event defined, we will test each trace point separately,
4194 * and then by groups, and finally all trace points.
4196 static __init void event_trace_self_tests(void)
4198 struct trace_subsystem_dir *dir;
4199 struct trace_event_file *file;
4200 struct trace_event_call *call;
4201 struct event_subsystem *system;
4202 struct trace_array *tr;
4205 tr = top_trace_array();
4209 pr_info("Running tests on trace events:\n");
4211 list_for_each_entry(file, &tr->events, list) {
4213 call = file->event_call;
4215 /* Only test those that have a probe */
4216 if (!call->class || !call->class->probe)
4220 * Testing syscall events here is pretty useless, but
4221 * we still do it if configured. But this is time consuming.
4222 * What we really need is a user thread to perform the
4223 * syscalls as we test.
4225 #ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
4226 if (call->class->system &&
4227 strcmp(call->class->system, "syscalls") == 0)
4231 pr_info("Testing event %s: ", trace_event_name(call));
4234 * If an event is already enabled, someone is using
4235 * it and the self test should not be on.
4237 if (file->flags & EVENT_FILE_FL_ENABLED) {
4238 pr_warn("Enabled event during self test!\n");
4243 ftrace_event_enable_disable(file, 1);
4245 ftrace_event_enable_disable(file, 0);
4250 /* Now test at the sub system level */
4252 pr_info("Running tests on trace event systems:\n");
4254 list_for_each_entry(dir, &tr->systems, list) {
4256 system = dir->subsystem;
4258 /* the ftrace system is special, skip it */
4259 if (strcmp(system->name, "ftrace") == 0)
4262 pr_info("Testing event system %s: ", system->name);
4264 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1);
4265 if (WARN_ON_ONCE(ret)) {
4266 pr_warn("error enabling system %s\n",
4273 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0);
4274 if (WARN_ON_ONCE(ret)) {
4275 pr_warn("error disabling system %s\n",
4283 /* Test with all events enabled */
4285 pr_info("Running tests on all trace events:\n");
4286 pr_info("Testing all events: ");
4288 ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1);
4289 if (WARN_ON_ONCE(ret)) {
4290 pr_warn("error enabling all events\n");
4297 ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
4298 if (WARN_ON_ONCE(ret)) {
4299 pr_warn("error disabling all events\n");
4306 #ifdef CONFIG_FUNCTION_TRACER
4308 static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
4310 static struct trace_event_file event_trace_file __initdata;
4313 function_test_events_call(unsigned long ip, unsigned long parent_ip,
4314 struct ftrace_ops *op, struct ftrace_regs *regs)
4316 struct trace_buffer *buffer;
4317 struct ring_buffer_event *event;
4318 struct ftrace_entry *entry;
4319 unsigned int trace_ctx;
4323 trace_ctx = tracing_gen_ctx();
4324 preempt_disable_notrace();
4325 cpu = raw_smp_processor_id();
4326 disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
4331 event = trace_event_buffer_lock_reserve(&buffer, &event_trace_file,
4332 TRACE_FN, sizeof(*entry),
4336 entry = ring_buffer_event_data(event);
4338 entry->parent_ip = parent_ip;
4340 event_trigger_unlock_commit(&event_trace_file, buffer, event,
4343 atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
4344 preempt_enable_notrace();
4347 static struct ftrace_ops trace_ops __initdata =
4349 .func = function_test_events_call,
4352 static __init void event_trace_self_test_with_function(void)
4356 event_trace_file.tr = top_trace_array();
4357 if (WARN_ON(!event_trace_file.tr))
4360 ret = register_ftrace_function(&trace_ops);
4361 if (WARN_ON(ret < 0)) {
4362 pr_info("Failed to enable function tracer for event tests\n");
4365 pr_info("Running tests again, along with the function tracer\n");
4366 event_trace_self_tests();
4367 unregister_ftrace_function(&trace_ops);
4370 static __init void event_trace_self_test_with_function(void)
4375 static __init int event_trace_self_tests_init(void)
4377 if (!tracing_selftest_disabled) {
4378 event_trace_self_tests();
4379 event_trace_self_test_with_function();
4385 late_initcall(event_trace_self_tests_init);