tracing: Separate open function from set_event and available_events
[linux-2.6.git] / kernel / trace / trace_events.c
blobcb2df3b70f7f1e4cc14ec217e2b658446b6a222d
1 /*
2 * event tracer
4 * Copyright (C) 2008 Red Hat Inc, Steven Rostedt <srostedt@redhat.com>
6 * - Added format output of fields of the trace point.
7 * This was based off of work by Tom Zanussi <tzanussi@gmail.com>.
9 */
11 #include <linux/workqueue.h>
12 #include <linux/spinlock.h>
13 #include <linux/kthread.h>
14 #include <linux/debugfs.h>
15 #include <linux/uaccess.h>
16 #include <linux/module.h>
17 #include <linux/ctype.h>
18 #include <linux/slab.h>
19 #include <linux/delay.h>
21 #include <asm/setup.h>
23 #include "trace_output.h"
25 #undef TRACE_SYSTEM
26 #define TRACE_SYSTEM "TRACE_SYSTEM"
28 DEFINE_MUTEX(event_mutex);
30 DEFINE_MUTEX(event_storage_mutex);
31 EXPORT_SYMBOL_GPL(event_storage_mutex);
33 char event_storage[EVENT_STORAGE_SIZE];
34 EXPORT_SYMBOL_GPL(event_storage);
36 LIST_HEAD(ftrace_events);
37 LIST_HEAD(ftrace_common_fields);
39 struct list_head *
40 trace_get_fields(struct ftrace_event_call *event_call)
42 if (!event_call->class->get_fields)
43 return &event_call->class->fields;
44 return event_call->class->get_fields(event_call);
47 static int __trace_define_field(struct list_head *head, const char *type,
48 const char *name, int offset, int size,
49 int is_signed, int filter_type)
51 struct ftrace_event_field *field;
53 field = kzalloc(sizeof(*field), GFP_KERNEL);
54 if (!field)
55 goto err;
57 field->name = kstrdup(name, GFP_KERNEL);
58 if (!field->name)
59 goto err;
61 field->type = kstrdup(type, GFP_KERNEL);
62 if (!field->type)
63 goto err;
65 if (filter_type == FILTER_OTHER)
66 field->filter_type = filter_assign_type(type);
67 else
68 field->filter_type = filter_type;
70 field->offset = offset;
71 field->size = size;
72 field->is_signed = is_signed;
74 list_add(&field->link, head);
76 return 0;
78 err:
79 if (field)
80 kfree(field->name);
81 kfree(field);
83 return -ENOMEM;
86 int trace_define_field(struct ftrace_event_call *call, const char *type,
87 const char *name, int offset, int size, int is_signed,
88 int filter_type)
90 struct list_head *head;
92 if (WARN_ON(!call->class))
93 return 0;
95 head = trace_get_fields(call);
96 return __trace_define_field(head, type, name, offset, size,
97 is_signed, filter_type);
99 EXPORT_SYMBOL_GPL(trace_define_field);
101 #define __common_field(type, item) \
102 ret = __trace_define_field(&ftrace_common_fields, #type, \
103 "common_" #item, \
104 offsetof(typeof(ent), item), \
105 sizeof(ent.item), \
106 is_signed_type(type), FILTER_OTHER); \
107 if (ret) \
108 return ret;
110 static int trace_define_common_fields(void)
112 int ret;
113 struct trace_entry ent;
115 __common_field(unsigned short, type);
116 __common_field(unsigned char, flags);
117 __common_field(unsigned char, preempt_count);
118 __common_field(int, pid);
119 __common_field(int, padding);
121 return ret;
124 void trace_destroy_fields(struct ftrace_event_call *call)
126 struct ftrace_event_field *field, *next;
127 struct list_head *head;
129 head = trace_get_fields(call);
130 list_for_each_entry_safe(field, next, head, link) {
131 list_del(&field->link);
132 kfree(field->type);
133 kfree(field->name);
134 kfree(field);
138 int trace_event_raw_init(struct ftrace_event_call *call)
140 int id;
142 id = register_ftrace_event(&call->event);
143 if (!id)
144 return -ENODEV;
146 return 0;
148 EXPORT_SYMBOL_GPL(trace_event_raw_init);
150 int ftrace_event_reg(struct ftrace_event_call *call,
151 enum trace_reg type, void *data)
153 switch (type) {
154 case TRACE_REG_REGISTER:
155 return tracepoint_probe_register(call->name,
156 call->class->probe,
157 call);
158 case TRACE_REG_UNREGISTER:
159 tracepoint_probe_unregister(call->name,
160 call->class->probe,
161 call);
162 return 0;
164 #ifdef CONFIG_PERF_EVENTS
165 case TRACE_REG_PERF_REGISTER:
166 return tracepoint_probe_register(call->name,
167 call->class->perf_probe,
168 call);
169 case TRACE_REG_PERF_UNREGISTER:
170 tracepoint_probe_unregister(call->name,
171 call->class->perf_probe,
172 call);
173 return 0;
174 case TRACE_REG_PERF_OPEN:
175 case TRACE_REG_PERF_CLOSE:
176 case TRACE_REG_PERF_ADD:
177 case TRACE_REG_PERF_DEL:
178 return 0;
179 #endif
181 return 0;
183 EXPORT_SYMBOL_GPL(ftrace_event_reg);
185 void trace_event_enable_cmd_record(bool enable)
187 struct ftrace_event_call *call;
189 mutex_lock(&event_mutex);
190 list_for_each_entry(call, &ftrace_events, list) {
191 if (!(call->flags & TRACE_EVENT_FL_ENABLED))
192 continue;
194 if (enable) {
195 tracing_start_cmdline_record();
196 call->flags |= TRACE_EVENT_FL_RECORDED_CMD;
197 } else {
198 tracing_stop_cmdline_record();
199 call->flags &= ~TRACE_EVENT_FL_RECORDED_CMD;
202 mutex_unlock(&event_mutex);
205 static int ftrace_event_enable_disable(struct ftrace_event_call *call,
206 int enable)
208 int ret = 0;
210 switch (enable) {
211 case 0:
212 if (call->flags & TRACE_EVENT_FL_ENABLED) {
213 call->flags &= ~TRACE_EVENT_FL_ENABLED;
214 if (call->flags & TRACE_EVENT_FL_RECORDED_CMD) {
215 tracing_stop_cmdline_record();
216 call->flags &= ~TRACE_EVENT_FL_RECORDED_CMD;
218 call->class->reg(call, TRACE_REG_UNREGISTER, NULL);
220 break;
221 case 1:
222 if (!(call->flags & TRACE_EVENT_FL_ENABLED)) {
223 if (trace_flags & TRACE_ITER_RECORD_CMD) {
224 tracing_start_cmdline_record();
225 call->flags |= TRACE_EVENT_FL_RECORDED_CMD;
227 ret = call->class->reg(call, TRACE_REG_REGISTER, NULL);
228 if (ret) {
229 tracing_stop_cmdline_record();
230 pr_info("event trace: Could not enable event "
231 "%s\n", call->name);
232 break;
234 call->flags |= TRACE_EVENT_FL_ENABLED;
236 break;
239 return ret;
242 static void ftrace_clear_events(void)
244 struct ftrace_event_call *call;
246 mutex_lock(&event_mutex);
247 list_for_each_entry(call, &ftrace_events, list) {
248 ftrace_event_enable_disable(call, 0);
250 mutex_unlock(&event_mutex);
253 static void __put_system(struct event_subsystem *system)
255 struct event_filter *filter = system->filter;
257 WARN_ON_ONCE(system->ref_count == 0);
258 if (--system->ref_count)
259 return;
261 if (filter) {
262 kfree(filter->filter_string);
263 kfree(filter);
265 kfree(system->name);
266 kfree(system);
269 static void __get_system(struct event_subsystem *system)
271 WARN_ON_ONCE(system->ref_count == 0);
272 system->ref_count++;
275 static void put_system(struct event_subsystem *system)
277 mutex_lock(&event_mutex);
278 __put_system(system);
279 mutex_unlock(&event_mutex);
283 * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
285 static int __ftrace_set_clr_event(const char *match, const char *sub,
286 const char *event, int set)
288 struct ftrace_event_call *call;
289 int ret = -EINVAL;
291 mutex_lock(&event_mutex);
292 list_for_each_entry(call, &ftrace_events, list) {
294 if (!call->name || !call->class || !call->class->reg)
295 continue;
297 if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
298 continue;
300 if (match &&
301 strcmp(match, call->name) != 0 &&
302 strcmp(match, call->class->system) != 0)
303 continue;
305 if (sub && strcmp(sub, call->class->system) != 0)
306 continue;
308 if (event && strcmp(event, call->name) != 0)
309 continue;
311 ftrace_event_enable_disable(call, set);
313 ret = 0;
315 mutex_unlock(&event_mutex);
317 return ret;
320 static int ftrace_set_clr_event(char *buf, int set)
322 char *event = NULL, *sub = NULL, *match;
325 * The buf format can be <subsystem>:<event-name>
326 * *:<event-name> means any event by that name.
327 * :<event-name> is the same.
329 * <subsystem>:* means all events in that subsystem
330 * <subsystem>: means the same.
332 * <name> (no ':') means all events in a subsystem with
333 * the name <name> or any event that matches <name>
336 match = strsep(&buf, ":");
337 if (buf) {
338 sub = match;
339 event = buf;
340 match = NULL;
342 if (!strlen(sub) || strcmp(sub, "*") == 0)
343 sub = NULL;
344 if (!strlen(event) || strcmp(event, "*") == 0)
345 event = NULL;
348 return __ftrace_set_clr_event(match, sub, event, set);
352 * trace_set_clr_event - enable or disable an event
353 * @system: system name to match (NULL for any system)
354 * @event: event name to match (NULL for all events, within system)
355 * @set: 1 to enable, 0 to disable
357 * This is a way for other parts of the kernel to enable or disable
358 * event recording.
360 * Returns 0 on success, -EINVAL if the parameters do not match any
361 * registered events.
363 int trace_set_clr_event(const char *system, const char *event, int set)
365 return __ftrace_set_clr_event(NULL, system, event, set);
367 EXPORT_SYMBOL_GPL(trace_set_clr_event);
369 /* 128 should be much more than enough */
370 #define EVENT_BUF_SIZE 127
372 static ssize_t
373 ftrace_event_write(struct file *file, const char __user *ubuf,
374 size_t cnt, loff_t *ppos)
376 struct trace_parser parser;
377 ssize_t read, ret;
379 if (!cnt)
380 return 0;
382 ret = tracing_update_buffers();
383 if (ret < 0)
384 return ret;
386 if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
387 return -ENOMEM;
389 read = trace_get_user(&parser, ubuf, cnt, ppos);
391 if (read >= 0 && trace_parser_loaded((&parser))) {
392 int set = 1;
394 if (*parser.buffer == '!')
395 set = 0;
397 parser.buffer[parser.idx] = 0;
399 ret = ftrace_set_clr_event(parser.buffer + !set, set);
400 if (ret)
401 goto out_put;
404 ret = read;
406 out_put:
407 trace_parser_put(&parser);
409 return ret;
412 static void *
413 t_next(struct seq_file *m, void *v, loff_t *pos)
415 struct ftrace_event_call *call = v;
417 (*pos)++;
419 list_for_each_entry_continue(call, &ftrace_events, list) {
421 * The ftrace subsystem is for showing formats only.
422 * They can not be enabled or disabled via the event files.
424 if (call->class && call->class->reg)
425 return call;
428 return NULL;
431 static void *t_start(struct seq_file *m, loff_t *pos)
433 struct ftrace_event_call *call;
434 loff_t l;
436 mutex_lock(&event_mutex);
438 call = list_entry(&ftrace_events, struct ftrace_event_call, list);
439 for (l = 0; l <= *pos; ) {
440 call = t_next(m, call, &l);
441 if (!call)
442 break;
444 return call;
447 static void *
448 s_next(struct seq_file *m, void *v, loff_t *pos)
450 struct ftrace_event_call *call = v;
452 (*pos)++;
454 list_for_each_entry_continue(call, &ftrace_events, list) {
455 if (call->flags & TRACE_EVENT_FL_ENABLED)
456 return call;
459 return NULL;
462 static void *s_start(struct seq_file *m, loff_t *pos)
464 struct ftrace_event_call *call;
465 loff_t l;
467 mutex_lock(&event_mutex);
469 call = list_entry(&ftrace_events, struct ftrace_event_call, list);
470 for (l = 0; l <= *pos; ) {
471 call = s_next(m, call, &l);
472 if (!call)
473 break;
475 return call;
478 static int t_show(struct seq_file *m, void *v)
480 struct ftrace_event_call *call = v;
482 if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
483 seq_printf(m, "%s:", call->class->system);
484 seq_printf(m, "%s\n", call->name);
486 return 0;
489 static void t_stop(struct seq_file *m, void *p)
491 mutex_unlock(&event_mutex);
494 static ssize_t
495 event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
496 loff_t *ppos)
498 struct ftrace_event_call *call = filp->private_data;
499 char *buf;
501 if (call->flags & TRACE_EVENT_FL_ENABLED)
502 buf = "1\n";
503 else
504 buf = "0\n";
506 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
509 static ssize_t
510 event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
511 loff_t *ppos)
513 struct ftrace_event_call *call = filp->private_data;
514 unsigned long val;
515 int ret;
517 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
518 if (ret)
519 return ret;
521 ret = tracing_update_buffers();
522 if (ret < 0)
523 return ret;
525 switch (val) {
526 case 0:
527 case 1:
528 mutex_lock(&event_mutex);
529 ret = ftrace_event_enable_disable(call, val);
530 mutex_unlock(&event_mutex);
531 break;
533 default:
534 return -EINVAL;
537 *ppos += cnt;
539 return ret ? ret : cnt;
542 static ssize_t
543 system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
544 loff_t *ppos)
546 const char set_to_char[4] = { '?', '0', '1', 'X' };
547 struct event_subsystem *system = filp->private_data;
548 struct ftrace_event_call *call;
549 char buf[2];
550 int set = 0;
551 int ret;
553 mutex_lock(&event_mutex);
554 list_for_each_entry(call, &ftrace_events, list) {
555 if (!call->name || !call->class || !call->class->reg)
556 continue;
558 if (system && strcmp(call->class->system, system->name) != 0)
559 continue;
562 * We need to find out if all the events are set
563 * or if all events or cleared, or if we have
564 * a mixture.
566 set |= (1 << !!(call->flags & TRACE_EVENT_FL_ENABLED));
569 * If we have a mixture, no need to look further.
571 if (set == 3)
572 break;
574 mutex_unlock(&event_mutex);
576 buf[0] = set_to_char[set];
577 buf[1] = '\n';
579 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
581 return ret;
584 static ssize_t
585 system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
586 loff_t *ppos)
588 struct event_subsystem *system = filp->private_data;
589 const char *name = NULL;
590 unsigned long val;
591 ssize_t ret;
593 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
594 if (ret)
595 return ret;
597 ret = tracing_update_buffers();
598 if (ret < 0)
599 return ret;
601 if (val != 0 && val != 1)
602 return -EINVAL;
605 * Opening of "enable" adds a ref count to system,
606 * so the name is safe to use.
608 if (system)
609 name = system->name;
611 ret = __ftrace_set_clr_event(NULL, name, NULL, val);
612 if (ret)
613 goto out;
615 ret = cnt;
617 out:
618 *ppos += cnt;
620 return ret;
623 enum {
624 FORMAT_HEADER = 1,
625 FORMAT_FIELD_SEPERATOR = 2,
626 FORMAT_PRINTFMT = 3,
629 static void *f_next(struct seq_file *m, void *v, loff_t *pos)
631 struct ftrace_event_call *call = m->private;
632 struct ftrace_event_field *field;
633 struct list_head *common_head = &ftrace_common_fields;
634 struct list_head *head = trace_get_fields(call);
636 (*pos)++;
638 switch ((unsigned long)v) {
639 case FORMAT_HEADER:
640 if (unlikely(list_empty(common_head)))
641 return NULL;
643 field = list_entry(common_head->prev,
644 struct ftrace_event_field, link);
645 return field;
647 case FORMAT_FIELD_SEPERATOR:
648 if (unlikely(list_empty(head)))
649 return NULL;
651 field = list_entry(head->prev, struct ftrace_event_field, link);
652 return field;
654 case FORMAT_PRINTFMT:
655 /* all done */
656 return NULL;
659 field = v;
660 if (field->link.prev == common_head)
661 return (void *)FORMAT_FIELD_SEPERATOR;
662 else if (field->link.prev == head)
663 return (void *)FORMAT_PRINTFMT;
665 field = list_entry(field->link.prev, struct ftrace_event_field, link);
667 return field;
670 static void *f_start(struct seq_file *m, loff_t *pos)
672 loff_t l = 0;
673 void *p;
675 /* Start by showing the header */
676 if (!*pos)
677 return (void *)FORMAT_HEADER;
679 p = (void *)FORMAT_HEADER;
680 do {
681 p = f_next(m, p, &l);
682 } while (p && l < *pos);
684 return p;
687 static int f_show(struct seq_file *m, void *v)
689 struct ftrace_event_call *call = m->private;
690 struct ftrace_event_field *field;
691 const char *array_descriptor;
693 switch ((unsigned long)v) {
694 case FORMAT_HEADER:
695 seq_printf(m, "name: %s\n", call->name);
696 seq_printf(m, "ID: %d\n", call->event.type);
697 seq_printf(m, "format:\n");
698 return 0;
700 case FORMAT_FIELD_SEPERATOR:
701 seq_putc(m, '\n');
702 return 0;
704 case FORMAT_PRINTFMT:
705 seq_printf(m, "\nprint fmt: %s\n",
706 call->print_fmt);
707 return 0;
710 field = v;
713 * Smartly shows the array type(except dynamic array).
714 * Normal:
715 * field:TYPE VAR
716 * If TYPE := TYPE[LEN], it is shown:
717 * field:TYPE VAR[LEN]
719 array_descriptor = strchr(field->type, '[');
721 if (!strncmp(field->type, "__data_loc", 10))
722 array_descriptor = NULL;
724 if (!array_descriptor)
725 seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
726 field->type, field->name, field->offset,
727 field->size, !!field->is_signed);
728 else
729 seq_printf(m, "\tfield:%.*s %s%s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
730 (int)(array_descriptor - field->type),
731 field->type, field->name,
732 array_descriptor, field->offset,
733 field->size, !!field->is_signed);
735 return 0;
738 static void f_stop(struct seq_file *m, void *p)
742 static const struct seq_operations trace_format_seq_ops = {
743 .start = f_start,
744 .next = f_next,
745 .stop = f_stop,
746 .show = f_show,
749 static int trace_format_open(struct inode *inode, struct file *file)
751 struct ftrace_event_call *call = inode->i_private;
752 struct seq_file *m;
753 int ret;
755 ret = seq_open(file, &trace_format_seq_ops);
756 if (ret < 0)
757 return ret;
759 m = file->private_data;
760 m->private = call;
762 return 0;
765 static ssize_t
766 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
768 struct ftrace_event_call *call = filp->private_data;
769 struct trace_seq *s;
770 int r;
772 if (*ppos)
773 return 0;
775 s = kmalloc(sizeof(*s), GFP_KERNEL);
776 if (!s)
777 return -ENOMEM;
779 trace_seq_init(s);
780 trace_seq_printf(s, "%d\n", call->event.type);
782 r = simple_read_from_buffer(ubuf, cnt, ppos,
783 s->buffer, s->len);
784 kfree(s);
785 return r;
788 static ssize_t
789 event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
790 loff_t *ppos)
792 struct ftrace_event_call *call = filp->private_data;
793 struct trace_seq *s;
794 int r;
796 if (*ppos)
797 return 0;
799 s = kmalloc(sizeof(*s), GFP_KERNEL);
800 if (!s)
801 return -ENOMEM;
803 trace_seq_init(s);
805 print_event_filter(call, s);
806 r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
808 kfree(s);
810 return r;
813 static ssize_t
814 event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
815 loff_t *ppos)
817 struct ftrace_event_call *call = filp->private_data;
818 char *buf;
819 int err;
821 if (cnt >= PAGE_SIZE)
822 return -EINVAL;
824 buf = (char *)__get_free_page(GFP_TEMPORARY);
825 if (!buf)
826 return -ENOMEM;
828 if (copy_from_user(buf, ubuf, cnt)) {
829 free_page((unsigned long) buf);
830 return -EFAULT;
832 buf[cnt] = '\0';
834 err = apply_event_filter(call, buf);
835 free_page((unsigned long) buf);
836 if (err < 0)
837 return err;
839 *ppos += cnt;
841 return cnt;
844 static LIST_HEAD(event_subsystems);
846 static int subsystem_open(struct inode *inode, struct file *filp)
848 struct event_subsystem *system = NULL;
849 int ret;
851 if (!inode->i_private)
852 goto skip_search;
854 /* Make sure the system still exists */
855 mutex_lock(&event_mutex);
856 list_for_each_entry(system, &event_subsystems, list) {
857 if (system == inode->i_private) {
858 /* Don't open systems with no events */
859 if (!system->nr_events) {
860 system = NULL;
861 break;
863 __get_system(system);
864 break;
867 mutex_unlock(&event_mutex);
869 if (system != inode->i_private)
870 return -ENODEV;
872 skip_search:
873 ret = tracing_open_generic(inode, filp);
874 if (ret < 0 && system)
875 put_system(system);
877 return ret;
880 static int subsystem_release(struct inode *inode, struct file *file)
882 struct event_subsystem *system = inode->i_private;
884 if (system)
885 put_system(system);
887 return 0;
890 static ssize_t
891 subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
892 loff_t *ppos)
894 struct event_subsystem *system = filp->private_data;
895 struct trace_seq *s;
896 int r;
898 if (*ppos)
899 return 0;
901 s = kmalloc(sizeof(*s), GFP_KERNEL);
902 if (!s)
903 return -ENOMEM;
905 trace_seq_init(s);
907 print_subsystem_event_filter(system, s);
908 r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
910 kfree(s);
912 return r;
915 static ssize_t
916 subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
917 loff_t *ppos)
919 struct event_subsystem *system = filp->private_data;
920 char *buf;
921 int err;
923 if (cnt >= PAGE_SIZE)
924 return -EINVAL;
926 buf = (char *)__get_free_page(GFP_TEMPORARY);
927 if (!buf)
928 return -ENOMEM;
930 if (copy_from_user(buf, ubuf, cnt)) {
931 free_page((unsigned long) buf);
932 return -EFAULT;
934 buf[cnt] = '\0';
936 err = apply_subsystem_event_filter(system, buf);
937 free_page((unsigned long) buf);
938 if (err < 0)
939 return err;
941 *ppos += cnt;
943 return cnt;
946 static ssize_t
947 show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
949 int (*func)(struct trace_seq *s) = filp->private_data;
950 struct trace_seq *s;
951 int r;
953 if (*ppos)
954 return 0;
956 s = kmalloc(sizeof(*s), GFP_KERNEL);
957 if (!s)
958 return -ENOMEM;
960 trace_seq_init(s);
962 func(s);
963 r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
965 kfree(s);
967 return r;
970 static int ftrace_event_avail_open(struct inode *inode, struct file *file);
971 static int ftrace_event_set_open(struct inode *inode, struct file *file);
973 static const struct seq_operations show_event_seq_ops = {
974 .start = t_start,
975 .next = t_next,
976 .show = t_show,
977 .stop = t_stop,
980 static const struct seq_operations show_set_event_seq_ops = {
981 .start = s_start,
982 .next = s_next,
983 .show = t_show,
984 .stop = t_stop,
987 static const struct file_operations ftrace_avail_fops = {
988 .open = ftrace_event_avail_open,
989 .read = seq_read,
990 .llseek = seq_lseek,
991 .release = seq_release,
994 static const struct file_operations ftrace_set_event_fops = {
995 .open = ftrace_event_set_open,
996 .read = seq_read,
997 .write = ftrace_event_write,
998 .llseek = seq_lseek,
999 .release = seq_release,
1002 static const struct file_operations ftrace_enable_fops = {
1003 .open = tracing_open_generic,
1004 .read = event_enable_read,
1005 .write = event_enable_write,
1006 .llseek = default_llseek,
1009 static const struct file_operations ftrace_event_format_fops = {
1010 .open = trace_format_open,
1011 .read = seq_read,
1012 .llseek = seq_lseek,
1013 .release = seq_release,
1016 static const struct file_operations ftrace_event_id_fops = {
1017 .open = tracing_open_generic,
1018 .read = event_id_read,
1019 .llseek = default_llseek,
1022 static const struct file_operations ftrace_event_filter_fops = {
1023 .open = tracing_open_generic,
1024 .read = event_filter_read,
1025 .write = event_filter_write,
1026 .llseek = default_llseek,
1029 static const struct file_operations ftrace_subsystem_filter_fops = {
1030 .open = subsystem_open,
1031 .read = subsystem_filter_read,
1032 .write = subsystem_filter_write,
1033 .llseek = default_llseek,
1034 .release = subsystem_release,
1037 static const struct file_operations ftrace_system_enable_fops = {
1038 .open = subsystem_open,
1039 .read = system_enable_read,
1040 .write = system_enable_write,
1041 .llseek = default_llseek,
1042 .release = subsystem_release,
1045 static const struct file_operations ftrace_show_header_fops = {
1046 .open = tracing_open_generic,
1047 .read = show_header,
1048 .llseek = default_llseek,
1051 static struct dentry *event_trace_events_dir(void)
1053 static struct dentry *d_tracer;
1054 static struct dentry *d_events;
1056 if (d_events)
1057 return d_events;
1059 d_tracer = tracing_init_dentry();
1060 if (!d_tracer)
1061 return NULL;
1063 d_events = debugfs_create_dir("events", d_tracer);
1064 if (!d_events)
1065 pr_warning("Could not create debugfs "
1066 "'events' directory\n");
1068 return d_events;
1071 static int
1072 ftrace_event_avail_open(struct inode *inode, struct file *file)
1074 const struct seq_operations *seq_ops = &show_event_seq_ops;
1076 return seq_open(file, seq_ops);
1079 static int
1080 ftrace_event_set_open(struct inode *inode, struct file *file)
1082 const struct seq_operations *seq_ops = &show_set_event_seq_ops;
1084 if ((file->f_mode & FMODE_WRITE) &&
1085 (file->f_flags & O_TRUNC))
1086 ftrace_clear_events();
1088 return seq_open(file, seq_ops);
1091 static struct dentry *
1092 event_subsystem_dir(const char *name, struct dentry *d_events)
1094 struct event_subsystem *system;
1095 struct dentry *entry;
1097 /* First see if we did not already create this dir */
1098 list_for_each_entry(system, &event_subsystems, list) {
1099 if (strcmp(system->name, name) == 0) {
1100 system->nr_events++;
1101 return system->entry;
1105 /* need to create new entry */
1106 system = kmalloc(sizeof(*system), GFP_KERNEL);
1107 if (!system) {
1108 pr_warning("No memory to create event subsystem %s\n",
1109 name);
1110 return d_events;
1113 system->entry = debugfs_create_dir(name, d_events);
1114 if (!system->entry) {
1115 pr_warning("Could not create event subsystem %s\n",
1116 name);
1117 kfree(system);
1118 return d_events;
1121 system->nr_events = 1;
1122 system->ref_count = 1;
1123 system->name = kstrdup(name, GFP_KERNEL);
1124 if (!system->name) {
1125 debugfs_remove(system->entry);
1126 kfree(system);
1127 return d_events;
1130 list_add(&system->list, &event_subsystems);
1132 system->filter = NULL;
1134 system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
1135 if (!system->filter) {
1136 pr_warning("Could not allocate filter for subsystem "
1137 "'%s'\n", name);
1138 return system->entry;
1141 entry = debugfs_create_file("filter", 0644, system->entry, system,
1142 &ftrace_subsystem_filter_fops);
1143 if (!entry) {
1144 kfree(system->filter);
1145 system->filter = NULL;
1146 pr_warning("Could not create debugfs "
1147 "'%s/filter' entry\n", name);
1150 trace_create_file("enable", 0644, system->entry, system,
1151 &ftrace_system_enable_fops);
1153 return system->entry;
1156 static int
1157 event_create_dir(struct ftrace_event_call *call, struct dentry *d_events,
1158 const struct file_operations *id,
1159 const struct file_operations *enable,
1160 const struct file_operations *filter,
1161 const struct file_operations *format)
1163 struct list_head *head;
1164 int ret;
1167 * If the trace point header did not define TRACE_SYSTEM
1168 * then the system would be called "TRACE_SYSTEM".
1170 if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
1171 d_events = event_subsystem_dir(call->class->system, d_events);
1173 call->dir = debugfs_create_dir(call->name, d_events);
1174 if (!call->dir) {
1175 pr_warning("Could not create debugfs "
1176 "'%s' directory\n", call->name);
1177 return -1;
1180 if (call->class->reg && !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
1181 trace_create_file("enable", 0644, call->dir, call,
1182 enable);
1184 #ifdef CONFIG_PERF_EVENTS
1185 if (call->event.type && call->class->reg)
1186 trace_create_file("id", 0444, call->dir, call,
1187 id);
1188 #endif
1191 * Other events may have the same class. Only update
1192 * the fields if they are not already defined.
1194 head = trace_get_fields(call);
1195 if (list_empty(head)) {
1196 ret = call->class->define_fields(call);
1197 if (ret < 0) {
1198 pr_warning("Could not initialize trace point"
1199 " events/%s\n", call->name);
1200 return ret;
1203 trace_create_file("filter", 0644, call->dir, call,
1204 filter);
1206 trace_create_file("format", 0444, call->dir, call,
1207 format);
1209 return 0;
1212 static void event_remove(struct ftrace_event_call *call)
1214 ftrace_event_enable_disable(call, 0);
1215 if (call->event.funcs)
1216 __unregister_ftrace_event(&call->event);
1217 list_del(&call->list);
1220 static int event_init(struct ftrace_event_call *call)
1222 int ret = 0;
1224 if (WARN_ON(!call->name))
1225 return -EINVAL;
1227 if (call->class->raw_init) {
1228 ret = call->class->raw_init(call);
1229 if (ret < 0 && ret != -ENOSYS)
1230 pr_warn("Could not initialize trace events/%s\n",
1231 call->name);
1234 return ret;
1237 static int
1238 __trace_add_event_call(struct ftrace_event_call *call, struct module *mod,
1239 const struct file_operations *id,
1240 const struct file_operations *enable,
1241 const struct file_operations *filter,
1242 const struct file_operations *format)
1244 struct dentry *d_events;
1245 int ret;
1247 ret = event_init(call);
1248 if (ret < 0)
1249 return ret;
1251 d_events = event_trace_events_dir();
1252 if (!d_events)
1253 return -ENOENT;
1255 ret = event_create_dir(call, d_events, id, enable, filter, format);
1256 if (!ret)
1257 list_add(&call->list, &ftrace_events);
1258 call->mod = mod;
1260 return ret;
1263 /* Add an additional event_call dynamically */
1264 int trace_add_event_call(struct ftrace_event_call *call)
1266 int ret;
1267 mutex_lock(&event_mutex);
1268 ret = __trace_add_event_call(call, NULL, &ftrace_event_id_fops,
1269 &ftrace_enable_fops,
1270 &ftrace_event_filter_fops,
1271 &ftrace_event_format_fops);
1272 mutex_unlock(&event_mutex);
1273 return ret;
1276 static void remove_subsystem_dir(const char *name)
1278 struct event_subsystem *system;
1280 if (strcmp(name, TRACE_SYSTEM) == 0)
1281 return;
1283 list_for_each_entry(system, &event_subsystems, list) {
1284 if (strcmp(system->name, name) == 0) {
1285 if (!--system->nr_events) {
1286 debugfs_remove_recursive(system->entry);
1287 list_del(&system->list);
1288 __put_system(system);
1290 break;
1296 * Must be called under locking both of event_mutex and trace_event_mutex.
1298 static void __trace_remove_event_call(struct ftrace_event_call *call)
1300 event_remove(call);
1301 trace_destroy_fields(call);
1302 destroy_preds(call);
1303 debugfs_remove_recursive(call->dir);
1304 remove_subsystem_dir(call->class->system);
1307 /* Remove an event_call */
1308 void trace_remove_event_call(struct ftrace_event_call *call)
1310 mutex_lock(&event_mutex);
1311 down_write(&trace_event_mutex);
1312 __trace_remove_event_call(call);
1313 up_write(&trace_event_mutex);
1314 mutex_unlock(&event_mutex);
1317 #define for_each_event(event, start, end) \
1318 for (event = start; \
1319 (unsigned long)event < (unsigned long)end; \
1320 event++)
1322 #ifdef CONFIG_MODULES
1324 static LIST_HEAD(ftrace_module_file_list);
1327 * Modules must own their file_operations to keep up with
1328 * reference counting.
1330 struct ftrace_module_file_ops {
1331 struct list_head list;
1332 struct module *mod;
1333 struct file_operations id;
1334 struct file_operations enable;
1335 struct file_operations format;
1336 struct file_operations filter;
1339 static struct ftrace_module_file_ops *
1340 trace_create_file_ops(struct module *mod)
1342 struct ftrace_module_file_ops *file_ops;
1345 * This is a bit of a PITA. To allow for correct reference
1346 * counting, modules must "own" their file_operations.
1347 * To do this, we allocate the file operations that will be
1348 * used in the event directory.
1351 file_ops = kmalloc(sizeof(*file_ops), GFP_KERNEL);
1352 if (!file_ops)
1353 return NULL;
1355 file_ops->mod = mod;
1357 file_ops->id = ftrace_event_id_fops;
1358 file_ops->id.owner = mod;
1360 file_ops->enable = ftrace_enable_fops;
1361 file_ops->enable.owner = mod;
1363 file_ops->filter = ftrace_event_filter_fops;
1364 file_ops->filter.owner = mod;
1366 file_ops->format = ftrace_event_format_fops;
1367 file_ops->format.owner = mod;
1369 list_add(&file_ops->list, &ftrace_module_file_list);
1371 return file_ops;
1374 static void trace_module_add_events(struct module *mod)
1376 struct ftrace_module_file_ops *file_ops = NULL;
1377 struct ftrace_event_call **call, **start, **end;
1379 start = mod->trace_events;
1380 end = mod->trace_events + mod->num_trace_events;
1382 if (start == end)
1383 return;
1385 file_ops = trace_create_file_ops(mod);
1386 if (!file_ops)
1387 return;
1389 for_each_event(call, start, end) {
1390 __trace_add_event_call(*call, mod,
1391 &file_ops->id, &file_ops->enable,
1392 &file_ops->filter, &file_ops->format);
1396 static void trace_module_remove_events(struct module *mod)
1398 struct ftrace_module_file_ops *file_ops;
1399 struct ftrace_event_call *call, *p;
1400 bool found = false;
1402 down_write(&trace_event_mutex);
1403 list_for_each_entry_safe(call, p, &ftrace_events, list) {
1404 if (call->mod == mod) {
1405 found = true;
1406 __trace_remove_event_call(call);
1410 /* Now free the file_operations */
1411 list_for_each_entry(file_ops, &ftrace_module_file_list, list) {
1412 if (file_ops->mod == mod)
1413 break;
1415 if (&file_ops->list != &ftrace_module_file_list) {
1416 list_del(&file_ops->list);
1417 kfree(file_ops);
1421 * It is safest to reset the ring buffer if the module being unloaded
1422 * registered any events.
1424 if (found)
1425 tracing_reset_current_online_cpus();
1426 up_write(&trace_event_mutex);
1429 static int trace_module_notify(struct notifier_block *self,
1430 unsigned long val, void *data)
1432 struct module *mod = data;
1434 mutex_lock(&event_mutex);
1435 switch (val) {
1436 case MODULE_STATE_COMING:
1437 trace_module_add_events(mod);
1438 break;
1439 case MODULE_STATE_GOING:
1440 trace_module_remove_events(mod);
1441 break;
1443 mutex_unlock(&event_mutex);
1445 return 0;
1447 #else
1448 static int trace_module_notify(struct notifier_block *self,
1449 unsigned long val, void *data)
1451 return 0;
1453 #endif /* CONFIG_MODULES */
1455 static struct notifier_block trace_module_nb = {
1456 .notifier_call = trace_module_notify,
1457 .priority = 0,
1460 extern struct ftrace_event_call *__start_ftrace_events[];
1461 extern struct ftrace_event_call *__stop_ftrace_events[];
1463 static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
1465 static __init int setup_trace_event(char *str)
1467 strlcpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
1468 ring_buffer_expanded = 1;
1469 tracing_selftest_disabled = 1;
1471 return 1;
1473 __setup("trace_event=", setup_trace_event);
1475 static __init int event_trace_enable(void)
1477 struct ftrace_event_call **iter, *call;
1478 char *buf = bootup_event_buf;
1479 char *token;
1480 int ret;
1482 for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
1484 call = *iter;
1485 ret = event_init(call);
1486 if (!ret)
1487 list_add(&call->list, &ftrace_events);
1490 while (true) {
1491 token = strsep(&buf, ",");
1493 if (!token)
1494 break;
1495 if (!*token)
1496 continue;
1498 ret = ftrace_set_clr_event(token, 1);
1499 if (ret)
1500 pr_warn("Failed to enable trace event: %s\n", token);
1503 trace_printk_start_comm();
1505 return 0;
1508 static __init int event_trace_init(void)
1510 struct ftrace_event_call *call;
1511 struct dentry *d_tracer;
1512 struct dentry *entry;
1513 struct dentry *d_events;
1514 int ret;
1516 d_tracer = tracing_init_dentry();
1517 if (!d_tracer)
1518 return 0;
1520 entry = debugfs_create_file("available_events", 0444, d_tracer,
1521 NULL, &ftrace_avail_fops);
1522 if (!entry)
1523 pr_warning("Could not create debugfs "
1524 "'available_events' entry\n");
1526 entry = debugfs_create_file("set_event", 0644, d_tracer,
1527 NULL, &ftrace_set_event_fops);
1528 if (!entry)
1529 pr_warning("Could not create debugfs "
1530 "'set_event' entry\n");
1532 d_events = event_trace_events_dir();
1533 if (!d_events)
1534 return 0;
1536 /* ring buffer internal formats */
1537 trace_create_file("header_page", 0444, d_events,
1538 ring_buffer_print_page_header,
1539 &ftrace_show_header_fops);
1541 trace_create_file("header_event", 0444, d_events,
1542 ring_buffer_print_entry_header,
1543 &ftrace_show_header_fops);
1545 trace_create_file("enable", 0644, d_events,
1546 NULL, &ftrace_system_enable_fops);
1548 if (trace_define_common_fields())
1549 pr_warning("tracing: Failed to allocate common fields");
1552 * Early initialization already enabled ftrace event.
1553 * Now it's only necessary to create the event directory.
1555 list_for_each_entry(call, &ftrace_events, list) {
1557 ret = event_create_dir(call, d_events,
1558 &ftrace_event_id_fops,
1559 &ftrace_enable_fops,
1560 &ftrace_event_filter_fops,
1561 &ftrace_event_format_fops);
1562 if (ret < 0)
1563 event_remove(call);
1566 ret = register_module_notifier(&trace_module_nb);
1567 if (ret)
1568 pr_warning("Failed to register trace events module notifier\n");
1570 return 0;
1572 core_initcall(event_trace_enable);
1573 fs_initcall(event_trace_init);
1575 #ifdef CONFIG_FTRACE_STARTUP_TEST
1577 static DEFINE_SPINLOCK(test_spinlock);
1578 static DEFINE_SPINLOCK(test_spinlock_irq);
1579 static DEFINE_MUTEX(test_mutex);
1581 static __init void test_work(struct work_struct *dummy)
1583 spin_lock(&test_spinlock);
1584 spin_lock_irq(&test_spinlock_irq);
1585 udelay(1);
1586 spin_unlock_irq(&test_spinlock_irq);
1587 spin_unlock(&test_spinlock);
1589 mutex_lock(&test_mutex);
1590 msleep(1);
1591 mutex_unlock(&test_mutex);
1594 static __init int event_test_thread(void *unused)
1596 void *test_malloc;
1598 test_malloc = kmalloc(1234, GFP_KERNEL);
1599 if (!test_malloc)
1600 pr_info("failed to kmalloc\n");
1602 schedule_on_each_cpu(test_work);
1604 kfree(test_malloc);
1606 set_current_state(TASK_INTERRUPTIBLE);
1607 while (!kthread_should_stop())
1608 schedule();
1610 return 0;
1614 * Do various things that may trigger events.
1616 static __init void event_test_stuff(void)
1618 struct task_struct *test_thread;
1620 test_thread = kthread_run(event_test_thread, NULL, "test-events");
1621 msleep(1);
1622 kthread_stop(test_thread);
1626 * For every trace event defined, we will test each trace point separately,
1627 * and then by groups, and finally all trace points.
1629 static __init void event_trace_self_tests(void)
1631 struct ftrace_event_call *call;
1632 struct event_subsystem *system;
1633 int ret;
1635 pr_info("Running tests on trace events:\n");
1637 list_for_each_entry(call, &ftrace_events, list) {
1639 /* Only test those that have a probe */
1640 if (!call->class || !call->class->probe)
1641 continue;
1644 * Testing syscall events here is pretty useless, but
1645 * we still do it if configured. But this is time consuming.
1646 * What we really need is a user thread to perform the
1647 * syscalls as we test.
1649 #ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
1650 if (call->class->system &&
1651 strcmp(call->class->system, "syscalls") == 0)
1652 continue;
1653 #endif
1655 pr_info("Testing event %s: ", call->name);
1658 * If an event is already enabled, someone is using
1659 * it and the self test should not be on.
1661 if (call->flags & TRACE_EVENT_FL_ENABLED) {
1662 pr_warning("Enabled event during self test!\n");
1663 WARN_ON_ONCE(1);
1664 continue;
1667 ftrace_event_enable_disable(call, 1);
1668 event_test_stuff();
1669 ftrace_event_enable_disable(call, 0);
1671 pr_cont("OK\n");
1674 /* Now test at the sub system level */
1676 pr_info("Running tests on trace event systems:\n");
1678 list_for_each_entry(system, &event_subsystems, list) {
1680 /* the ftrace system is special, skip it */
1681 if (strcmp(system->name, "ftrace") == 0)
1682 continue;
1684 pr_info("Testing event system %s: ", system->name);
1686 ret = __ftrace_set_clr_event(NULL, system->name, NULL, 1);
1687 if (WARN_ON_ONCE(ret)) {
1688 pr_warning("error enabling system %s\n",
1689 system->name);
1690 continue;
1693 event_test_stuff();
1695 ret = __ftrace_set_clr_event(NULL, system->name, NULL, 0);
1696 if (WARN_ON_ONCE(ret)) {
1697 pr_warning("error disabling system %s\n",
1698 system->name);
1699 continue;
1702 pr_cont("OK\n");
1705 /* Test with all events enabled */
1707 pr_info("Running tests on all trace events:\n");
1708 pr_info("Testing all events: ");
1710 ret = __ftrace_set_clr_event(NULL, NULL, NULL, 1);
1711 if (WARN_ON_ONCE(ret)) {
1712 pr_warning("error enabling all events\n");
1713 return;
1716 event_test_stuff();
1718 /* reset sysname */
1719 ret = __ftrace_set_clr_event(NULL, NULL, NULL, 0);
1720 if (WARN_ON_ONCE(ret)) {
1721 pr_warning("error disabling all events\n");
1722 return;
1725 pr_cont("OK\n");
1728 #ifdef CONFIG_FUNCTION_TRACER
1730 static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
1732 static void
1733 function_test_events_call(unsigned long ip, unsigned long parent_ip,
1734 struct ftrace_ops *op, struct pt_regs *pt_regs)
1736 struct ring_buffer_event *event;
1737 struct ring_buffer *buffer;
1738 struct ftrace_entry *entry;
1739 unsigned long flags;
1740 long disabled;
1741 int cpu;
1742 int pc;
1744 pc = preempt_count();
1745 preempt_disable_notrace();
1746 cpu = raw_smp_processor_id();
1747 disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
1749 if (disabled != 1)
1750 goto out;
1752 local_save_flags(flags);
1754 event = trace_current_buffer_lock_reserve(&buffer,
1755 TRACE_FN, sizeof(*entry),
1756 flags, pc);
1757 if (!event)
1758 goto out;
1759 entry = ring_buffer_event_data(event);
1760 entry->ip = ip;
1761 entry->parent_ip = parent_ip;
1763 trace_nowake_buffer_unlock_commit(buffer, event, flags, pc);
1765 out:
1766 atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
1767 preempt_enable_notrace();
1770 static struct ftrace_ops trace_ops __initdata =
1772 .func = function_test_events_call,
1773 .flags = FTRACE_OPS_FL_RECURSION_SAFE,
1776 static __init void event_trace_self_test_with_function(void)
1778 int ret;
1779 ret = register_ftrace_function(&trace_ops);
1780 if (WARN_ON(ret < 0)) {
1781 pr_info("Failed to enable function tracer for event tests\n");
1782 return;
1784 pr_info("Running tests again, along with the function tracer\n");
1785 event_trace_self_tests();
1786 unregister_ftrace_function(&trace_ops);
1788 #else
1789 static __init void event_trace_self_test_with_function(void)
1792 #endif
1794 static __init int event_trace_self_tests_init(void)
1796 if (!tracing_selftest_disabled) {
1797 event_trace_self_tests();
1798 event_trace_self_test_with_function();
1801 return 0;
1804 late_initcall(event_trace_self_tests_init);
1806 #endif