sfc: Do not initialise buffer in efx_alloc_special_buffer()
[linux-2.6/btrfs-unstable.git] / kernel / trace / trace_events.c
blobd608d09d08c0de8ba8bab4a3e0b1edd6df30c033
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 int
495 ftrace_event_seq_open(struct inode *inode, struct file *file)
497 const struct seq_operations *seq_ops;
499 if ((file->f_mode & FMODE_WRITE) &&
500 (file->f_flags & O_TRUNC))
501 ftrace_clear_events();
503 seq_ops = inode->i_private;
504 return seq_open(file, seq_ops);
507 static ssize_t
508 event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
509 loff_t *ppos)
511 struct ftrace_event_call *call = filp->private_data;
512 char *buf;
514 if (call->flags & TRACE_EVENT_FL_ENABLED)
515 buf = "1\n";
516 else
517 buf = "0\n";
519 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
522 static ssize_t
523 event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
524 loff_t *ppos)
526 struct ftrace_event_call *call = filp->private_data;
527 unsigned long val;
528 int ret;
530 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
531 if (ret)
532 return ret;
534 ret = tracing_update_buffers();
535 if (ret < 0)
536 return ret;
538 switch (val) {
539 case 0:
540 case 1:
541 mutex_lock(&event_mutex);
542 ret = ftrace_event_enable_disable(call, val);
543 mutex_unlock(&event_mutex);
544 break;
546 default:
547 return -EINVAL;
550 *ppos += cnt;
552 return ret ? ret : cnt;
555 static ssize_t
556 system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
557 loff_t *ppos)
559 const char set_to_char[4] = { '?', '0', '1', 'X' };
560 struct event_subsystem *system = filp->private_data;
561 struct ftrace_event_call *call;
562 char buf[2];
563 int set = 0;
564 int ret;
566 mutex_lock(&event_mutex);
567 list_for_each_entry(call, &ftrace_events, list) {
568 if (!call->name || !call->class || !call->class->reg)
569 continue;
571 if (system && strcmp(call->class->system, system->name) != 0)
572 continue;
575 * We need to find out if all the events are set
576 * or if all events or cleared, or if we have
577 * a mixture.
579 set |= (1 << !!(call->flags & TRACE_EVENT_FL_ENABLED));
582 * If we have a mixture, no need to look further.
584 if (set == 3)
585 break;
587 mutex_unlock(&event_mutex);
589 buf[0] = set_to_char[set];
590 buf[1] = '\n';
592 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
594 return ret;
597 static ssize_t
598 system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
599 loff_t *ppos)
601 struct event_subsystem *system = filp->private_data;
602 const char *name = NULL;
603 unsigned long val;
604 ssize_t ret;
606 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
607 if (ret)
608 return ret;
610 ret = tracing_update_buffers();
611 if (ret < 0)
612 return ret;
614 if (val != 0 && val != 1)
615 return -EINVAL;
618 * Opening of "enable" adds a ref count to system,
619 * so the name is safe to use.
621 if (system)
622 name = system->name;
624 ret = __ftrace_set_clr_event(NULL, name, NULL, val);
625 if (ret)
626 goto out;
628 ret = cnt;
630 out:
631 *ppos += cnt;
633 return ret;
636 enum {
637 FORMAT_HEADER = 1,
638 FORMAT_FIELD_SEPERATOR = 2,
639 FORMAT_PRINTFMT = 3,
642 static void *f_next(struct seq_file *m, void *v, loff_t *pos)
644 struct ftrace_event_call *call = m->private;
645 struct ftrace_event_field *field;
646 struct list_head *common_head = &ftrace_common_fields;
647 struct list_head *head = trace_get_fields(call);
649 (*pos)++;
651 switch ((unsigned long)v) {
652 case FORMAT_HEADER:
653 if (unlikely(list_empty(common_head)))
654 return NULL;
656 field = list_entry(common_head->prev,
657 struct ftrace_event_field, link);
658 return field;
660 case FORMAT_FIELD_SEPERATOR:
661 if (unlikely(list_empty(head)))
662 return NULL;
664 field = list_entry(head->prev, struct ftrace_event_field, link);
665 return field;
667 case FORMAT_PRINTFMT:
668 /* all done */
669 return NULL;
672 field = v;
673 if (field->link.prev == common_head)
674 return (void *)FORMAT_FIELD_SEPERATOR;
675 else if (field->link.prev == head)
676 return (void *)FORMAT_PRINTFMT;
678 field = list_entry(field->link.prev, struct ftrace_event_field, link);
680 return field;
683 static void *f_start(struct seq_file *m, loff_t *pos)
685 loff_t l = 0;
686 void *p;
688 /* Start by showing the header */
689 if (!*pos)
690 return (void *)FORMAT_HEADER;
692 p = (void *)FORMAT_HEADER;
693 do {
694 p = f_next(m, p, &l);
695 } while (p && l < *pos);
697 return p;
700 static int f_show(struct seq_file *m, void *v)
702 struct ftrace_event_call *call = m->private;
703 struct ftrace_event_field *field;
704 const char *array_descriptor;
706 switch ((unsigned long)v) {
707 case FORMAT_HEADER:
708 seq_printf(m, "name: %s\n", call->name);
709 seq_printf(m, "ID: %d\n", call->event.type);
710 seq_printf(m, "format:\n");
711 return 0;
713 case FORMAT_FIELD_SEPERATOR:
714 seq_putc(m, '\n');
715 return 0;
717 case FORMAT_PRINTFMT:
718 seq_printf(m, "\nprint fmt: %s\n",
719 call->print_fmt);
720 return 0;
723 field = v;
726 * Smartly shows the array type(except dynamic array).
727 * Normal:
728 * field:TYPE VAR
729 * If TYPE := TYPE[LEN], it is shown:
730 * field:TYPE VAR[LEN]
732 array_descriptor = strchr(field->type, '[');
734 if (!strncmp(field->type, "__data_loc", 10))
735 array_descriptor = NULL;
737 if (!array_descriptor)
738 seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
739 field->type, field->name, field->offset,
740 field->size, !!field->is_signed);
741 else
742 seq_printf(m, "\tfield:%.*s %s%s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
743 (int)(array_descriptor - field->type),
744 field->type, field->name,
745 array_descriptor, field->offset,
746 field->size, !!field->is_signed);
748 return 0;
751 static void f_stop(struct seq_file *m, void *p)
755 static const struct seq_operations trace_format_seq_ops = {
756 .start = f_start,
757 .next = f_next,
758 .stop = f_stop,
759 .show = f_show,
762 static int trace_format_open(struct inode *inode, struct file *file)
764 struct ftrace_event_call *call = inode->i_private;
765 struct seq_file *m;
766 int ret;
768 ret = seq_open(file, &trace_format_seq_ops);
769 if (ret < 0)
770 return ret;
772 m = file->private_data;
773 m->private = call;
775 return 0;
778 static ssize_t
779 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
781 struct ftrace_event_call *call = filp->private_data;
782 struct trace_seq *s;
783 int r;
785 if (*ppos)
786 return 0;
788 s = kmalloc(sizeof(*s), GFP_KERNEL);
789 if (!s)
790 return -ENOMEM;
792 trace_seq_init(s);
793 trace_seq_printf(s, "%d\n", call->event.type);
795 r = simple_read_from_buffer(ubuf, cnt, ppos,
796 s->buffer, s->len);
797 kfree(s);
798 return r;
801 static ssize_t
802 event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
803 loff_t *ppos)
805 struct ftrace_event_call *call = filp->private_data;
806 struct trace_seq *s;
807 int r;
809 if (*ppos)
810 return 0;
812 s = kmalloc(sizeof(*s), GFP_KERNEL);
813 if (!s)
814 return -ENOMEM;
816 trace_seq_init(s);
818 print_event_filter(call, s);
819 r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
821 kfree(s);
823 return r;
826 static ssize_t
827 event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
828 loff_t *ppos)
830 struct ftrace_event_call *call = filp->private_data;
831 char *buf;
832 int err;
834 if (cnt >= PAGE_SIZE)
835 return -EINVAL;
837 buf = (char *)__get_free_page(GFP_TEMPORARY);
838 if (!buf)
839 return -ENOMEM;
841 if (copy_from_user(buf, ubuf, cnt)) {
842 free_page((unsigned long) buf);
843 return -EFAULT;
845 buf[cnt] = '\0';
847 err = apply_event_filter(call, buf);
848 free_page((unsigned long) buf);
849 if (err < 0)
850 return err;
852 *ppos += cnt;
854 return cnt;
857 static LIST_HEAD(event_subsystems);
859 static int subsystem_open(struct inode *inode, struct file *filp)
861 struct event_subsystem *system = NULL;
862 int ret;
864 if (!inode->i_private)
865 goto skip_search;
867 /* Make sure the system still exists */
868 mutex_lock(&event_mutex);
869 list_for_each_entry(system, &event_subsystems, list) {
870 if (system == inode->i_private) {
871 /* Don't open systems with no events */
872 if (!system->nr_events) {
873 system = NULL;
874 break;
876 __get_system(system);
877 break;
880 mutex_unlock(&event_mutex);
882 if (system != inode->i_private)
883 return -ENODEV;
885 skip_search:
886 ret = tracing_open_generic(inode, filp);
887 if (ret < 0 && system)
888 put_system(system);
890 return ret;
893 static int subsystem_release(struct inode *inode, struct file *file)
895 struct event_subsystem *system = inode->i_private;
897 if (system)
898 put_system(system);
900 return 0;
903 static ssize_t
904 subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
905 loff_t *ppos)
907 struct event_subsystem *system = filp->private_data;
908 struct trace_seq *s;
909 int r;
911 if (*ppos)
912 return 0;
914 s = kmalloc(sizeof(*s), GFP_KERNEL);
915 if (!s)
916 return -ENOMEM;
918 trace_seq_init(s);
920 print_subsystem_event_filter(system, s);
921 r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
923 kfree(s);
925 return r;
928 static ssize_t
929 subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
930 loff_t *ppos)
932 struct event_subsystem *system = filp->private_data;
933 char *buf;
934 int err;
936 if (cnt >= PAGE_SIZE)
937 return -EINVAL;
939 buf = (char *)__get_free_page(GFP_TEMPORARY);
940 if (!buf)
941 return -ENOMEM;
943 if (copy_from_user(buf, ubuf, cnt)) {
944 free_page((unsigned long) buf);
945 return -EFAULT;
947 buf[cnt] = '\0';
949 err = apply_subsystem_event_filter(system, buf);
950 free_page((unsigned long) buf);
951 if (err < 0)
952 return err;
954 *ppos += cnt;
956 return cnt;
959 static ssize_t
960 show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
962 int (*func)(struct trace_seq *s) = filp->private_data;
963 struct trace_seq *s;
964 int r;
966 if (*ppos)
967 return 0;
969 s = kmalloc(sizeof(*s), GFP_KERNEL);
970 if (!s)
971 return -ENOMEM;
973 trace_seq_init(s);
975 func(s);
976 r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
978 kfree(s);
980 return r;
983 static const struct seq_operations show_event_seq_ops = {
984 .start = t_start,
985 .next = t_next,
986 .show = t_show,
987 .stop = t_stop,
990 static const struct seq_operations show_set_event_seq_ops = {
991 .start = s_start,
992 .next = s_next,
993 .show = t_show,
994 .stop = t_stop,
997 static const struct file_operations ftrace_avail_fops = {
998 .open = ftrace_event_seq_open,
999 .read = seq_read,
1000 .llseek = seq_lseek,
1001 .release = seq_release,
1004 static const struct file_operations ftrace_set_event_fops = {
1005 .open = ftrace_event_seq_open,
1006 .read = seq_read,
1007 .write = ftrace_event_write,
1008 .llseek = seq_lseek,
1009 .release = seq_release,
1012 static const struct file_operations ftrace_enable_fops = {
1013 .open = tracing_open_generic,
1014 .read = event_enable_read,
1015 .write = event_enable_write,
1016 .llseek = default_llseek,
1019 static const struct file_operations ftrace_event_format_fops = {
1020 .open = trace_format_open,
1021 .read = seq_read,
1022 .llseek = seq_lseek,
1023 .release = seq_release,
1026 static const struct file_operations ftrace_event_id_fops = {
1027 .open = tracing_open_generic,
1028 .read = event_id_read,
1029 .llseek = default_llseek,
1032 static const struct file_operations ftrace_event_filter_fops = {
1033 .open = tracing_open_generic,
1034 .read = event_filter_read,
1035 .write = event_filter_write,
1036 .llseek = default_llseek,
1039 static const struct file_operations ftrace_subsystem_filter_fops = {
1040 .open = subsystem_open,
1041 .read = subsystem_filter_read,
1042 .write = subsystem_filter_write,
1043 .llseek = default_llseek,
1044 .release = subsystem_release,
1047 static const struct file_operations ftrace_system_enable_fops = {
1048 .open = subsystem_open,
1049 .read = system_enable_read,
1050 .write = system_enable_write,
1051 .llseek = default_llseek,
1052 .release = subsystem_release,
1055 static const struct file_operations ftrace_show_header_fops = {
1056 .open = tracing_open_generic,
1057 .read = show_header,
1058 .llseek = default_llseek,
1061 static struct dentry *event_trace_events_dir(void)
1063 static struct dentry *d_tracer;
1064 static struct dentry *d_events;
1066 if (d_events)
1067 return d_events;
1069 d_tracer = tracing_init_dentry();
1070 if (!d_tracer)
1071 return NULL;
1073 d_events = debugfs_create_dir("events", d_tracer);
1074 if (!d_events)
1075 pr_warning("Could not create debugfs "
1076 "'events' directory\n");
1078 return d_events;
1081 static struct dentry *
1082 event_subsystem_dir(const char *name, struct dentry *d_events)
1084 struct event_subsystem *system;
1085 struct dentry *entry;
1087 /* First see if we did not already create this dir */
1088 list_for_each_entry(system, &event_subsystems, list) {
1089 if (strcmp(system->name, name) == 0) {
1090 system->nr_events++;
1091 return system->entry;
1095 /* need to create new entry */
1096 system = kmalloc(sizeof(*system), GFP_KERNEL);
1097 if (!system) {
1098 pr_warning("No memory to create event subsystem %s\n",
1099 name);
1100 return d_events;
1103 system->entry = debugfs_create_dir(name, d_events);
1104 if (!system->entry) {
1105 pr_warning("Could not create event subsystem %s\n",
1106 name);
1107 kfree(system);
1108 return d_events;
1111 system->nr_events = 1;
1112 system->ref_count = 1;
1113 system->name = kstrdup(name, GFP_KERNEL);
1114 if (!system->name) {
1115 debugfs_remove(system->entry);
1116 kfree(system);
1117 return d_events;
1120 list_add(&system->list, &event_subsystems);
1122 system->filter = NULL;
1124 system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
1125 if (!system->filter) {
1126 pr_warning("Could not allocate filter for subsystem "
1127 "'%s'\n", name);
1128 return system->entry;
1131 entry = debugfs_create_file("filter", 0644, system->entry, system,
1132 &ftrace_subsystem_filter_fops);
1133 if (!entry) {
1134 kfree(system->filter);
1135 system->filter = NULL;
1136 pr_warning("Could not create debugfs "
1137 "'%s/filter' entry\n", name);
1140 trace_create_file("enable", 0644, system->entry, system,
1141 &ftrace_system_enable_fops);
1143 return system->entry;
1146 static int
1147 event_create_dir(struct ftrace_event_call *call, struct dentry *d_events,
1148 const struct file_operations *id,
1149 const struct file_operations *enable,
1150 const struct file_operations *filter,
1151 const struct file_operations *format)
1153 struct list_head *head;
1154 int ret;
1157 * If the trace point header did not define TRACE_SYSTEM
1158 * then the system would be called "TRACE_SYSTEM".
1160 if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
1161 d_events = event_subsystem_dir(call->class->system, d_events);
1163 call->dir = debugfs_create_dir(call->name, d_events);
1164 if (!call->dir) {
1165 pr_warning("Could not create debugfs "
1166 "'%s' directory\n", call->name);
1167 return -1;
1170 if (call->class->reg && !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
1171 trace_create_file("enable", 0644, call->dir, call,
1172 enable);
1174 #ifdef CONFIG_PERF_EVENTS
1175 if (call->event.type && call->class->reg)
1176 trace_create_file("id", 0444, call->dir, call,
1177 id);
1178 #endif
1181 * Other events may have the same class. Only update
1182 * the fields if they are not already defined.
1184 head = trace_get_fields(call);
1185 if (list_empty(head)) {
1186 ret = call->class->define_fields(call);
1187 if (ret < 0) {
1188 pr_warning("Could not initialize trace point"
1189 " events/%s\n", call->name);
1190 return ret;
1193 trace_create_file("filter", 0644, call->dir, call,
1194 filter);
1196 trace_create_file("format", 0444, call->dir, call,
1197 format);
1199 return 0;
1202 static void event_remove(struct ftrace_event_call *call)
1204 ftrace_event_enable_disable(call, 0);
1205 if (call->event.funcs)
1206 __unregister_ftrace_event(&call->event);
1207 list_del(&call->list);
1210 static int event_init(struct ftrace_event_call *call)
1212 int ret = 0;
1214 if (WARN_ON(!call->name))
1215 return -EINVAL;
1217 if (call->class->raw_init) {
1218 ret = call->class->raw_init(call);
1219 if (ret < 0 && ret != -ENOSYS)
1220 pr_warn("Could not initialize trace events/%s\n",
1221 call->name);
1224 return ret;
1227 static int
1228 __trace_add_event_call(struct ftrace_event_call *call, struct module *mod,
1229 const struct file_operations *id,
1230 const struct file_operations *enable,
1231 const struct file_operations *filter,
1232 const struct file_operations *format)
1234 struct dentry *d_events;
1235 int ret;
1237 ret = event_init(call);
1238 if (ret < 0)
1239 return ret;
1241 d_events = event_trace_events_dir();
1242 if (!d_events)
1243 return -ENOENT;
1245 ret = event_create_dir(call, d_events, id, enable, filter, format);
1246 if (!ret)
1247 list_add(&call->list, &ftrace_events);
1248 call->mod = mod;
1250 return ret;
1253 /* Add an additional event_call dynamically */
1254 int trace_add_event_call(struct ftrace_event_call *call)
1256 int ret;
1257 mutex_lock(&event_mutex);
1258 ret = __trace_add_event_call(call, NULL, &ftrace_event_id_fops,
1259 &ftrace_enable_fops,
1260 &ftrace_event_filter_fops,
1261 &ftrace_event_format_fops);
1262 mutex_unlock(&event_mutex);
1263 return ret;
1266 static void remove_subsystem_dir(const char *name)
1268 struct event_subsystem *system;
1270 if (strcmp(name, TRACE_SYSTEM) == 0)
1271 return;
1273 list_for_each_entry(system, &event_subsystems, list) {
1274 if (strcmp(system->name, name) == 0) {
1275 if (!--system->nr_events) {
1276 debugfs_remove_recursive(system->entry);
1277 list_del(&system->list);
1278 __put_system(system);
1280 break;
1286 * Must be called under locking both of event_mutex and trace_event_mutex.
1288 static void __trace_remove_event_call(struct ftrace_event_call *call)
1290 event_remove(call);
1291 trace_destroy_fields(call);
1292 destroy_preds(call);
1293 debugfs_remove_recursive(call->dir);
1294 remove_subsystem_dir(call->class->system);
1297 /* Remove an event_call */
1298 void trace_remove_event_call(struct ftrace_event_call *call)
1300 mutex_lock(&event_mutex);
1301 down_write(&trace_event_mutex);
1302 __trace_remove_event_call(call);
1303 up_write(&trace_event_mutex);
1304 mutex_unlock(&event_mutex);
1307 #define for_each_event(event, start, end) \
1308 for (event = start; \
1309 (unsigned long)event < (unsigned long)end; \
1310 event++)
1312 #ifdef CONFIG_MODULES
1314 static LIST_HEAD(ftrace_module_file_list);
1317 * Modules must own their file_operations to keep up with
1318 * reference counting.
1320 struct ftrace_module_file_ops {
1321 struct list_head list;
1322 struct module *mod;
1323 struct file_operations id;
1324 struct file_operations enable;
1325 struct file_operations format;
1326 struct file_operations filter;
1329 static struct ftrace_module_file_ops *
1330 trace_create_file_ops(struct module *mod)
1332 struct ftrace_module_file_ops *file_ops;
1335 * This is a bit of a PITA. To allow for correct reference
1336 * counting, modules must "own" their file_operations.
1337 * To do this, we allocate the file operations that will be
1338 * used in the event directory.
1341 file_ops = kmalloc(sizeof(*file_ops), GFP_KERNEL);
1342 if (!file_ops)
1343 return NULL;
1345 file_ops->mod = mod;
1347 file_ops->id = ftrace_event_id_fops;
1348 file_ops->id.owner = mod;
1350 file_ops->enable = ftrace_enable_fops;
1351 file_ops->enable.owner = mod;
1353 file_ops->filter = ftrace_event_filter_fops;
1354 file_ops->filter.owner = mod;
1356 file_ops->format = ftrace_event_format_fops;
1357 file_ops->format.owner = mod;
1359 list_add(&file_ops->list, &ftrace_module_file_list);
1361 return file_ops;
1364 static void trace_module_add_events(struct module *mod)
1366 struct ftrace_module_file_ops *file_ops = NULL;
1367 struct ftrace_event_call **call, **start, **end;
1369 start = mod->trace_events;
1370 end = mod->trace_events + mod->num_trace_events;
1372 if (start == end)
1373 return;
1375 file_ops = trace_create_file_ops(mod);
1376 if (!file_ops)
1377 return;
1379 for_each_event(call, start, end) {
1380 __trace_add_event_call(*call, mod,
1381 &file_ops->id, &file_ops->enable,
1382 &file_ops->filter, &file_ops->format);
1386 static void trace_module_remove_events(struct module *mod)
1388 struct ftrace_module_file_ops *file_ops;
1389 struct ftrace_event_call *call, *p;
1390 bool found = false;
1392 down_write(&trace_event_mutex);
1393 list_for_each_entry_safe(call, p, &ftrace_events, list) {
1394 if (call->mod == mod) {
1395 found = true;
1396 __trace_remove_event_call(call);
1400 /* Now free the file_operations */
1401 list_for_each_entry(file_ops, &ftrace_module_file_list, list) {
1402 if (file_ops->mod == mod)
1403 break;
1405 if (&file_ops->list != &ftrace_module_file_list) {
1406 list_del(&file_ops->list);
1407 kfree(file_ops);
1411 * It is safest to reset the ring buffer if the module being unloaded
1412 * registered any events.
1414 if (found)
1415 tracing_reset_current_online_cpus();
1416 up_write(&trace_event_mutex);
1419 static int trace_module_notify(struct notifier_block *self,
1420 unsigned long val, void *data)
1422 struct module *mod = data;
1424 mutex_lock(&event_mutex);
1425 switch (val) {
1426 case MODULE_STATE_COMING:
1427 trace_module_add_events(mod);
1428 break;
1429 case MODULE_STATE_GOING:
1430 trace_module_remove_events(mod);
1431 break;
1433 mutex_unlock(&event_mutex);
1435 return 0;
1437 #else
1438 static int trace_module_notify(struct notifier_block *self,
1439 unsigned long val, void *data)
1441 return 0;
1443 #endif /* CONFIG_MODULES */
1445 static struct notifier_block trace_module_nb = {
1446 .notifier_call = trace_module_notify,
1447 .priority = 0,
1450 extern struct ftrace_event_call *__start_ftrace_events[];
1451 extern struct ftrace_event_call *__stop_ftrace_events[];
1453 static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
1455 static __init int setup_trace_event(char *str)
1457 strlcpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
1458 ring_buffer_expanded = 1;
1459 tracing_selftest_disabled = 1;
1461 return 1;
1463 __setup("trace_event=", setup_trace_event);
1465 static __init int event_trace_enable(void)
1467 struct ftrace_event_call **iter, *call;
1468 char *buf = bootup_event_buf;
1469 char *token;
1470 int ret;
1472 for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
1474 call = *iter;
1475 ret = event_init(call);
1476 if (!ret)
1477 list_add(&call->list, &ftrace_events);
1480 while (true) {
1481 token = strsep(&buf, ",");
1483 if (!token)
1484 break;
1485 if (!*token)
1486 continue;
1488 ret = ftrace_set_clr_event(token, 1);
1489 if (ret)
1490 pr_warn("Failed to enable trace event: %s\n", token);
1492 return 0;
1495 static __init int event_trace_init(void)
1497 struct ftrace_event_call *call;
1498 struct dentry *d_tracer;
1499 struct dentry *entry;
1500 struct dentry *d_events;
1501 int ret;
1503 d_tracer = tracing_init_dentry();
1504 if (!d_tracer)
1505 return 0;
1507 entry = debugfs_create_file("available_events", 0444, d_tracer,
1508 (void *)&show_event_seq_ops,
1509 &ftrace_avail_fops);
1510 if (!entry)
1511 pr_warning("Could not create debugfs "
1512 "'available_events' entry\n");
1514 entry = debugfs_create_file("set_event", 0644, d_tracer,
1515 (void *)&show_set_event_seq_ops,
1516 &ftrace_set_event_fops);
1517 if (!entry)
1518 pr_warning("Could not create debugfs "
1519 "'set_event' entry\n");
1521 d_events = event_trace_events_dir();
1522 if (!d_events)
1523 return 0;
1525 /* ring buffer internal formats */
1526 trace_create_file("header_page", 0444, d_events,
1527 ring_buffer_print_page_header,
1528 &ftrace_show_header_fops);
1530 trace_create_file("header_event", 0444, d_events,
1531 ring_buffer_print_entry_header,
1532 &ftrace_show_header_fops);
1534 trace_create_file("enable", 0644, d_events,
1535 NULL, &ftrace_system_enable_fops);
1537 if (trace_define_common_fields())
1538 pr_warning("tracing: Failed to allocate common fields");
1541 * Early initialization already enabled ftrace event.
1542 * Now it's only necessary to create the event directory.
1544 list_for_each_entry(call, &ftrace_events, list) {
1546 ret = event_create_dir(call, d_events,
1547 &ftrace_event_id_fops,
1548 &ftrace_enable_fops,
1549 &ftrace_event_filter_fops,
1550 &ftrace_event_format_fops);
1551 if (ret < 0)
1552 event_remove(call);
1555 ret = register_module_notifier(&trace_module_nb);
1556 if (ret)
1557 pr_warning("Failed to register trace events module notifier\n");
1559 return 0;
1561 core_initcall(event_trace_enable);
1562 fs_initcall(event_trace_init);
1564 #ifdef CONFIG_FTRACE_STARTUP_TEST
1566 static DEFINE_SPINLOCK(test_spinlock);
1567 static DEFINE_SPINLOCK(test_spinlock_irq);
1568 static DEFINE_MUTEX(test_mutex);
1570 static __init void test_work(struct work_struct *dummy)
1572 spin_lock(&test_spinlock);
1573 spin_lock_irq(&test_spinlock_irq);
1574 udelay(1);
1575 spin_unlock_irq(&test_spinlock_irq);
1576 spin_unlock(&test_spinlock);
1578 mutex_lock(&test_mutex);
1579 msleep(1);
1580 mutex_unlock(&test_mutex);
1583 static __init int event_test_thread(void *unused)
1585 void *test_malloc;
1587 test_malloc = kmalloc(1234, GFP_KERNEL);
1588 if (!test_malloc)
1589 pr_info("failed to kmalloc\n");
1591 schedule_on_each_cpu(test_work);
1593 kfree(test_malloc);
1595 set_current_state(TASK_INTERRUPTIBLE);
1596 while (!kthread_should_stop())
1597 schedule();
1599 return 0;
1603 * Do various things that may trigger events.
1605 static __init void event_test_stuff(void)
1607 struct task_struct *test_thread;
1609 test_thread = kthread_run(event_test_thread, NULL, "test-events");
1610 msleep(1);
1611 kthread_stop(test_thread);
1615 * For every trace event defined, we will test each trace point separately,
1616 * and then by groups, and finally all trace points.
1618 static __init void event_trace_self_tests(void)
1620 struct ftrace_event_call *call;
1621 struct event_subsystem *system;
1622 int ret;
1624 pr_info("Running tests on trace events:\n");
1626 list_for_each_entry(call, &ftrace_events, list) {
1628 /* Only test those that have a probe */
1629 if (!call->class || !call->class->probe)
1630 continue;
1633 * Testing syscall events here is pretty useless, but
1634 * we still do it if configured. But this is time consuming.
1635 * What we really need is a user thread to perform the
1636 * syscalls as we test.
1638 #ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
1639 if (call->class->system &&
1640 strcmp(call->class->system, "syscalls") == 0)
1641 continue;
1642 #endif
1644 pr_info("Testing event %s: ", call->name);
1647 * If an event is already enabled, someone is using
1648 * it and the self test should not be on.
1650 if (call->flags & TRACE_EVENT_FL_ENABLED) {
1651 pr_warning("Enabled event during self test!\n");
1652 WARN_ON_ONCE(1);
1653 continue;
1656 ftrace_event_enable_disable(call, 1);
1657 event_test_stuff();
1658 ftrace_event_enable_disable(call, 0);
1660 pr_cont("OK\n");
1663 /* Now test at the sub system level */
1665 pr_info("Running tests on trace event systems:\n");
1667 list_for_each_entry(system, &event_subsystems, list) {
1669 /* the ftrace system is special, skip it */
1670 if (strcmp(system->name, "ftrace") == 0)
1671 continue;
1673 pr_info("Testing event system %s: ", system->name);
1675 ret = __ftrace_set_clr_event(NULL, system->name, NULL, 1);
1676 if (WARN_ON_ONCE(ret)) {
1677 pr_warning("error enabling system %s\n",
1678 system->name);
1679 continue;
1682 event_test_stuff();
1684 ret = __ftrace_set_clr_event(NULL, system->name, NULL, 0);
1685 if (WARN_ON_ONCE(ret)) {
1686 pr_warning("error disabling system %s\n",
1687 system->name);
1688 continue;
1691 pr_cont("OK\n");
1694 /* Test with all events enabled */
1696 pr_info("Running tests on all trace events:\n");
1697 pr_info("Testing all events: ");
1699 ret = __ftrace_set_clr_event(NULL, NULL, NULL, 1);
1700 if (WARN_ON_ONCE(ret)) {
1701 pr_warning("error enabling all events\n");
1702 return;
1705 event_test_stuff();
1707 /* reset sysname */
1708 ret = __ftrace_set_clr_event(NULL, NULL, NULL, 0);
1709 if (WARN_ON_ONCE(ret)) {
1710 pr_warning("error disabling all events\n");
1711 return;
1714 pr_cont("OK\n");
1717 #ifdef CONFIG_FUNCTION_TRACER
1719 static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
1721 static void
1722 function_test_events_call(unsigned long ip, unsigned long parent_ip,
1723 struct ftrace_ops *op, struct pt_regs *pt_regs)
1725 struct ring_buffer_event *event;
1726 struct ring_buffer *buffer;
1727 struct ftrace_entry *entry;
1728 unsigned long flags;
1729 long disabled;
1730 int cpu;
1731 int pc;
1733 pc = preempt_count();
1734 preempt_disable_notrace();
1735 cpu = raw_smp_processor_id();
1736 disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
1738 if (disabled != 1)
1739 goto out;
1741 local_save_flags(flags);
1743 event = trace_current_buffer_lock_reserve(&buffer,
1744 TRACE_FN, sizeof(*entry),
1745 flags, pc);
1746 if (!event)
1747 goto out;
1748 entry = ring_buffer_event_data(event);
1749 entry->ip = ip;
1750 entry->parent_ip = parent_ip;
1752 trace_nowake_buffer_unlock_commit(buffer, event, flags, pc);
1754 out:
1755 atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
1756 preempt_enable_notrace();
1759 static struct ftrace_ops trace_ops __initdata =
1761 .func = function_test_events_call,
1762 .flags = FTRACE_OPS_FL_RECURSION_SAFE,
1765 static __init void event_trace_self_test_with_function(void)
1767 int ret;
1768 ret = register_ftrace_function(&trace_ops);
1769 if (WARN_ON(ret < 0)) {
1770 pr_info("Failed to enable function tracer for event tests\n");
1771 return;
1773 pr_info("Running tests again, along with the function tracer\n");
1774 event_trace_self_tests();
1775 unregister_ftrace_function(&trace_ops);
1777 #else
1778 static __init void event_trace_self_test_with_function(void)
1781 #endif
1783 static __init int event_trace_self_tests_init(void)
1785 if (!tracing_selftest_disabled) {
1786 event_trace_self_tests();
1787 event_trace_self_test_with_function();
1790 return 0;
1793 late_initcall(event_trace_self_tests_init);
1795 #endif