4 * Copyright (C) 2008 Red Hat Inc, Steven Rostedt <srostedt@redhat.com>
6 * - Added format output of fields of the trace point.
7 * This was based off of work by Tom Zanussi <tzanussi@gmail.com>.
11 #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/delay.h>
20 #include "trace_output.h"
22 #define TRACE_SYSTEM "TRACE_SYSTEM"
24 static DEFINE_MUTEX(event_mutex);
26 LIST_HEAD(ftrace_events);
28 int trace_define_field(struct ftrace_event_call *call, char *type,
29 char *name, int offset, int size, int is_signed)
31 struct ftrace_event_field *field;
33 field = kzalloc(sizeof(*field), GFP_KERNEL);
37 field->name = kstrdup(name, GFP_KERNEL);
41 field->type = kstrdup(type, GFP_KERNEL);
45 field->offset = offset;
47 field->is_signed = is_signed;
48 list_add(&field->link, &call->fields);
61 EXPORT_SYMBOL_GPL(trace_define_field);
65 static void trace_destroy_fields(struct ftrace_event_call *call)
67 struct ftrace_event_field *field, *next;
69 list_for_each_entry_safe(field, next, &call->fields, link) {
70 list_del(&field->link);
77 #endif /* CONFIG_MODULES */
79 static void ftrace_clear_events(void)
81 struct ftrace_event_call *call;
83 list_for_each_entry(call, &ftrace_events, list) {
92 static void ftrace_event_enable_disable(struct ftrace_event_call *call,
104 if (!call->enabled) {
112 static int ftrace_set_clr_event(char *buf, int set)
114 struct ftrace_event_call *call;
115 char *event = NULL, *sub = NULL, *match;
119 * The buf format can be <subsystem>:<event-name>
120 * *:<event-name> means any event by that name.
121 * :<event-name> is the same.
123 * <subsystem>:* means all events in that subsystem
124 * <subsystem>: means the same.
126 * <name> (no ':') means all events in a subsystem with
127 * the name <name> or any event that matches <name>
130 match = strsep(&buf, ":");
136 if (!strlen(sub) || strcmp(sub, "*") == 0)
138 if (!strlen(event) || strcmp(event, "*") == 0)
142 mutex_lock(&event_mutex);
143 list_for_each_entry(call, &ftrace_events, list) {
145 if (!call->name || !call->regfunc)
149 strcmp(match, call->name) != 0 &&
150 strcmp(match, call->system) != 0)
153 if (sub && strcmp(sub, call->system) != 0)
156 if (event && strcmp(event, call->name) != 0)
159 ftrace_event_enable_disable(call, set);
163 mutex_unlock(&event_mutex);
168 /* 128 should be much more than enough */
169 #define EVENT_BUF_SIZE 127
172 ftrace_event_write(struct file *file, const char __user *ubuf,
173 size_t cnt, loff_t *ppos)
184 ret = tracing_update_buffers();
188 ret = get_user(ch, ubuf++);
194 /* skip white space */
195 while (cnt && isspace(ch)) {
196 ret = get_user(ch, ubuf++);
203 /* Only white space found? */
210 buf = kmalloc(EVENT_BUF_SIZE+1, GFP_KERNEL);
214 if (cnt > EVENT_BUF_SIZE)
215 cnt = EVENT_BUF_SIZE;
218 while (cnt && !isspace(ch)) {
224 ret = get_user(ch, ubuf++);
234 ret = ftrace_set_clr_event(buf, set);
247 t_next(struct seq_file *m, void *v, loff_t *pos)
249 struct list_head *list = m->private;
250 struct ftrace_event_call *call;
255 if (list == &ftrace_events)
258 call = list_entry(list, struct ftrace_event_call, list);
261 * The ftrace subsystem is for showing formats only.
262 * They can not be enabled or disabled via the event files.
270 m->private = list->next;
275 static void *t_start(struct seq_file *m, loff_t *pos)
277 return t_next(m, NULL, pos);
281 s_next(struct seq_file *m, void *v, loff_t *pos)
283 struct list_head *list = m->private;
284 struct ftrace_event_call *call;
289 if (list == &ftrace_events)
292 call = list_entry(list, struct ftrace_event_call, list);
294 if (!call->enabled) {
299 m->private = list->next;
304 static void *s_start(struct seq_file *m, loff_t *pos)
306 return s_next(m, NULL, pos);
309 static int t_show(struct seq_file *m, void *v)
311 struct ftrace_event_call *call = v;
313 if (strcmp(call->system, TRACE_SYSTEM) != 0)
314 seq_printf(m, "%s:", call->system);
315 seq_printf(m, "%s\n", call->name);
320 static void t_stop(struct seq_file *m, void *p)
325 ftrace_event_seq_open(struct inode *inode, struct file *file)
328 const struct seq_operations *seq_ops;
330 if ((file->f_mode & FMODE_WRITE) &&
331 !(file->f_flags & O_APPEND))
332 ftrace_clear_events();
334 seq_ops = inode->i_private;
335 ret = seq_open(file, seq_ops);
337 struct seq_file *m = file->private_data;
339 m->private = ftrace_events.next;
345 event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
348 struct ftrace_event_call *call = filp->private_data;
356 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
360 event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
363 struct ftrace_event_call *call = filp->private_data;
368 if (cnt >= sizeof(buf))
371 if (copy_from_user(&buf, ubuf, cnt))
376 ret = strict_strtoul(buf, 10, &val);
380 ret = tracing_update_buffers();
387 mutex_lock(&event_mutex);
388 ftrace_event_enable_disable(call, val);
389 mutex_unlock(&event_mutex);
401 extern char *__bad_type_size(void);
404 #define FIELD(type, name) \
405 sizeof(type) != sizeof(field.name) ? __bad_type_size() : \
406 #type, "common_" #name, offsetof(typeof(field), name), \
409 static int trace_write_header(struct trace_seq *s)
411 struct trace_entry field;
413 /* struct trace_entry */
414 return trace_seq_printf(s,
415 "\tfield:%s %s;\toffset:%zu;\tsize:%zu;\n"
416 "\tfield:%s %s;\toffset:%zu;\tsize:%zu;\n"
417 "\tfield:%s %s;\toffset:%zu;\tsize:%zu;\n"
418 "\tfield:%s %s;\toffset:%zu;\tsize:%zu;\n"
419 "\tfield:%s %s;\toffset:%zu;\tsize:%zu;\n"
421 FIELD(unsigned short, type),
422 FIELD(unsigned char, flags),
423 FIELD(unsigned char, preempt_count),
429 event_format_read(struct file *filp, char __user *ubuf, size_t cnt,
432 struct ftrace_event_call *call = filp->private_data;
440 s = kmalloc(sizeof(*s), GFP_KERNEL);
446 /* If any of the first writes fail, so will the show_format. */
448 trace_seq_printf(s, "name: %s\n", call->name);
449 trace_seq_printf(s, "ID: %d\n", call->id);
450 trace_seq_printf(s, "format:\n");
451 trace_write_header(s);
453 r = call->show_format(s);
456 * ug! The format output is bigger than a PAGE!!
458 buf = "FORMAT TOO BIG\n";
459 r = simple_read_from_buffer(ubuf, cnt, ppos,
464 r = simple_read_from_buffer(ubuf, cnt, ppos,
472 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
474 struct ftrace_event_call *call = filp->private_data;
481 s = kmalloc(sizeof(*s), GFP_KERNEL);
486 trace_seq_printf(s, "%d\n", call->id);
488 r = simple_read_from_buffer(ubuf, cnt, ppos,
495 event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
498 struct ftrace_event_call *call = filp->private_data;
505 s = kmalloc(sizeof(*s), GFP_KERNEL);
511 print_event_filter(call, s);
512 r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
520 event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
523 struct ftrace_event_call *call = filp->private_data;
527 if (cnt >= PAGE_SIZE)
530 buf = (char *)__get_free_page(GFP_TEMPORARY);
534 if (copy_from_user(buf, ubuf, cnt)) {
535 free_page((unsigned long) buf);
540 err = apply_event_filter(call, buf);
541 free_page((unsigned long) buf);
551 subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
554 struct event_subsystem *system = filp->private_data;
561 s = kmalloc(sizeof(*s), GFP_KERNEL);
567 print_subsystem_event_filter(system, s);
568 r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
576 subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
579 struct event_subsystem *system = filp->private_data;
583 if (cnt >= PAGE_SIZE)
586 buf = (char *)__get_free_page(GFP_TEMPORARY);
590 if (copy_from_user(buf, ubuf, cnt)) {
591 free_page((unsigned long) buf);
596 err = apply_subsystem_event_filter(system, buf);
597 free_page((unsigned long) buf);
607 show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
609 int (*func)(struct trace_seq *s) = filp->private_data;
616 s = kmalloc(sizeof(*s), GFP_KERNEL);
623 r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
630 static const struct seq_operations show_event_seq_ops = {
637 static const struct seq_operations show_set_event_seq_ops = {
644 static const struct file_operations ftrace_avail_fops = {
645 .open = ftrace_event_seq_open,
648 .release = seq_release,
651 static const struct file_operations ftrace_set_event_fops = {
652 .open = ftrace_event_seq_open,
654 .write = ftrace_event_write,
656 .release = seq_release,
659 static const struct file_operations ftrace_enable_fops = {
660 .open = tracing_open_generic,
661 .read = event_enable_read,
662 .write = event_enable_write,
665 static const struct file_operations ftrace_event_format_fops = {
666 .open = tracing_open_generic,
667 .read = event_format_read,
670 static const struct file_operations ftrace_event_id_fops = {
671 .open = tracing_open_generic,
672 .read = event_id_read,
675 static const struct file_operations ftrace_event_filter_fops = {
676 .open = tracing_open_generic,
677 .read = event_filter_read,
678 .write = event_filter_write,
681 static const struct file_operations ftrace_subsystem_filter_fops = {
682 .open = tracing_open_generic,
683 .read = subsystem_filter_read,
684 .write = subsystem_filter_write,
687 static const struct file_operations ftrace_show_header_fops = {
688 .open = tracing_open_generic,
692 static struct dentry *event_trace_events_dir(void)
694 static struct dentry *d_tracer;
695 static struct dentry *d_events;
700 d_tracer = tracing_init_dentry();
704 d_events = debugfs_create_dir("events", d_tracer);
706 pr_warning("Could not create debugfs "
707 "'events' directory\n");
712 static LIST_HEAD(event_subsystems);
714 static struct dentry *
715 event_subsystem_dir(const char *name, struct dentry *d_events)
717 struct event_subsystem *system;
718 struct dentry *entry;
720 /* First see if we did not already create this dir */
721 list_for_each_entry(system, &event_subsystems, list) {
722 if (strcmp(system->name, name) == 0)
723 return system->entry;
726 /* need to create new entry */
727 system = kmalloc(sizeof(*system), GFP_KERNEL);
729 pr_warning("No memory to create event subsystem %s\n",
734 system->entry = debugfs_create_dir(name, d_events);
735 if (!system->entry) {
736 pr_warning("Could not create event subsystem %s\n",
742 system->name = kstrdup(name, GFP_KERNEL);
744 debugfs_remove(system->entry);
749 list_add(&system->list, &event_subsystems);
751 system->filter = NULL;
753 system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
754 if (!system->filter) {
755 pr_warning("Could not allocate filter for subsystem "
757 return system->entry;
760 entry = debugfs_create_file("filter", 0644, system->entry, system,
761 &ftrace_subsystem_filter_fops);
763 kfree(system->filter);
764 system->filter = NULL;
765 pr_warning("Could not create debugfs "
766 "'%s/filter' entry\n", name);
769 return system->entry;
773 event_create_dir(struct ftrace_event_call *call, struct dentry *d_events,
774 const struct file_operations *id,
775 const struct file_operations *enable,
776 const struct file_operations *filter,
777 const struct file_operations *format)
779 struct dentry *entry;
783 * If the trace point header did not define TRACE_SYSTEM
784 * then the system would be called "TRACE_SYSTEM".
786 if (strcmp(call->system, TRACE_SYSTEM) != 0)
787 d_events = event_subsystem_dir(call->system, d_events);
789 if (call->raw_init) {
790 ret = call->raw_init();
792 pr_warning("Could not initialize trace point"
793 " events/%s\n", call->name);
798 call->dir = debugfs_create_dir(call->name, d_events);
800 pr_warning("Could not create debugfs "
801 "'%s' directory\n", call->name);
806 entry = trace_create_file("enable", 0644, call->dir, call,
810 entry = trace_create_file("id", 0444, call->dir, call,
813 if (call->define_fields) {
814 ret = call->define_fields();
816 pr_warning("Could not initialize trace point"
817 " events/%s\n", call->name);
820 entry = trace_create_file("filter", 0644, call->dir, call,
824 /* A trace may not want to export its format */
825 if (!call->show_format)
828 entry = trace_create_file("format", 0444, call->dir, call,
834 #define for_each_event(event, start, end) \
835 for (event = start; \
836 (unsigned long)event < (unsigned long)end; \
839 #ifdef CONFIG_MODULES
841 static LIST_HEAD(ftrace_module_file_list);
844 * Modules must own their file_operations to keep up with
845 * reference counting.
847 struct ftrace_module_file_ops {
848 struct list_head list;
850 struct file_operations id;
851 struct file_operations enable;
852 struct file_operations format;
853 struct file_operations filter;
856 static struct ftrace_module_file_ops *
857 trace_create_file_ops(struct module *mod)
859 struct ftrace_module_file_ops *file_ops;
862 * This is a bit of a PITA. To allow for correct reference
863 * counting, modules must "own" their file_operations.
864 * To do this, we allocate the file operations that will be
865 * used in the event directory.
868 file_ops = kmalloc(sizeof(*file_ops), GFP_KERNEL);
874 file_ops->id = ftrace_event_id_fops;
875 file_ops->id.owner = mod;
877 file_ops->enable = ftrace_enable_fops;
878 file_ops->enable.owner = mod;
880 file_ops->filter = ftrace_event_filter_fops;
881 file_ops->filter.owner = mod;
883 file_ops->format = ftrace_event_format_fops;
884 file_ops->format.owner = mod;
886 list_add(&file_ops->list, &ftrace_module_file_list);
891 static void trace_module_add_events(struct module *mod)
893 struct ftrace_module_file_ops *file_ops = NULL;
894 struct ftrace_event_call *call, *start, *end;
895 struct dentry *d_events;
897 start = mod->trace_events;
898 end = mod->trace_events + mod->num_trace_events;
903 d_events = event_trace_events_dir();
907 for_each_event(call, start, end) {
908 /* The linker may leave blanks */
913 * This module has events, create file ops for this module
914 * if not already done.
917 file_ops = trace_create_file_ops(mod);
922 list_add(&call->list, &ftrace_events);
923 event_create_dir(call, d_events,
924 &file_ops->id, &file_ops->enable,
925 &file_ops->filter, &file_ops->format);
929 static void trace_module_remove_events(struct module *mod)
931 struct ftrace_module_file_ops *file_ops;
932 struct ftrace_event_call *call, *p;
934 list_for_each_entry_safe(call, p, &ftrace_events, list) {
935 if (call->mod == mod) {
941 unregister_ftrace_event(call->event);
942 debugfs_remove_recursive(call->dir);
943 list_del(&call->list);
944 trace_destroy_fields(call);
949 /* Now free the file_operations */
950 list_for_each_entry(file_ops, &ftrace_module_file_list, list) {
951 if (file_ops->mod == mod)
954 if (&file_ops->list != &ftrace_module_file_list) {
955 list_del(&file_ops->list);
960 static int trace_module_notify(struct notifier_block *self,
961 unsigned long val, void *data)
963 struct module *mod = data;
965 mutex_lock(&event_mutex);
967 case MODULE_STATE_COMING:
968 trace_module_add_events(mod);
970 case MODULE_STATE_GOING:
971 trace_module_remove_events(mod);
974 mutex_unlock(&event_mutex);
979 static int trace_module_notify(struct notifier_block *self,
980 unsigned long val, void *data)
984 #endif /* CONFIG_MODULES */
986 struct notifier_block trace_module_nb = {
987 .notifier_call = trace_module_notify,
991 extern struct ftrace_event_call __start_ftrace_events[];
992 extern struct ftrace_event_call __stop_ftrace_events[];
994 static __init int event_trace_init(void)
996 struct ftrace_event_call *call;
997 struct dentry *d_tracer;
998 struct dentry *entry;
999 struct dentry *d_events;
1002 d_tracer = tracing_init_dentry();
1006 entry = debugfs_create_file("available_events", 0444, d_tracer,
1007 (void *)&show_event_seq_ops,
1008 &ftrace_avail_fops);
1010 pr_warning("Could not create debugfs "
1011 "'available_events' entry\n");
1013 entry = debugfs_create_file("set_event", 0644, d_tracer,
1014 (void *)&show_set_event_seq_ops,
1015 &ftrace_set_event_fops);
1017 pr_warning("Could not create debugfs "
1018 "'set_event' entry\n");
1020 d_events = event_trace_events_dir();
1024 /* ring buffer internal formats */
1025 trace_create_file("header_page", 0444, d_events,
1026 ring_buffer_print_page_header,
1027 &ftrace_show_header_fops);
1029 trace_create_file("header_event", 0444, d_events,
1030 ring_buffer_print_entry_header,
1031 &ftrace_show_header_fops);
1033 for_each_event(call, __start_ftrace_events, __stop_ftrace_events) {
1034 /* The linker may leave blanks */
1037 list_add(&call->list, &ftrace_events);
1038 event_create_dir(call, d_events, &ftrace_event_id_fops,
1039 &ftrace_enable_fops, &ftrace_event_filter_fops,
1040 &ftrace_event_format_fops);
1043 ret = register_module_notifier(&trace_module_nb);
1045 pr_warning("Failed to register trace events module notifier\n");
1049 fs_initcall(event_trace_init);
1051 #ifdef CONFIG_FTRACE_STARTUP_TEST
1053 static DEFINE_SPINLOCK(test_spinlock);
1054 static DEFINE_SPINLOCK(test_spinlock_irq);
1055 static DEFINE_MUTEX(test_mutex);
1057 static __init void test_work(struct work_struct *dummy)
1059 spin_lock(&test_spinlock);
1060 spin_lock_irq(&test_spinlock_irq);
1062 spin_unlock_irq(&test_spinlock_irq);
1063 spin_unlock(&test_spinlock);
1065 mutex_lock(&test_mutex);
1067 mutex_unlock(&test_mutex);
1070 static __init int event_test_thread(void *unused)
1074 test_malloc = kmalloc(1234, GFP_KERNEL);
1076 pr_info("failed to kmalloc\n");
1078 schedule_on_each_cpu(test_work);
1082 set_current_state(TASK_INTERRUPTIBLE);
1083 while (!kthread_should_stop())
1090 * Do various things that may trigger events.
1092 static __init void event_test_stuff(void)
1094 struct task_struct *test_thread;
1096 test_thread = kthread_run(event_test_thread, NULL, "test-events");
1098 kthread_stop(test_thread);
1102 * For every trace event defined, we will test each trace point separately,
1103 * and then by groups, and finally all trace points.
1105 static __init void event_trace_self_tests(void)
1107 struct ftrace_event_call *call;
1108 struct event_subsystem *system;
1112 pr_info("Running tests on trace events:\n");
1114 list_for_each_entry(call, &ftrace_events, list) {
1116 /* Only test those that have a regfunc */
1120 pr_info("Testing event %s: ", call->name);
1123 * If an event is already enabled, someone is using
1124 * it and the self test should not be on.
1126 if (call->enabled) {
1127 pr_warning("Enabled event during self test!\n");
1143 /* Now test at the sub system level */
1145 pr_info("Running tests on trace event systems:\n");
1147 list_for_each_entry(system, &event_subsystems, list) {
1149 /* the ftrace system is special, skip it */
1150 if (strcmp(system->name, "ftrace") == 0)
1153 pr_info("Testing event system %s: ", system->name);
1155 /* ftrace_set_clr_event can modify the name passed in. */
1156 sysname = kstrdup(system->name, GFP_KERNEL);
1157 if (WARN_ON(!sysname)) {
1158 pr_warning("Can't allocate memory, giving up!\n");
1161 ret = ftrace_set_clr_event(sysname, 1);
1163 if (WARN_ON_ONCE(ret)) {
1164 pr_warning("error enabling system %s\n",
1171 sysname = kstrdup(system->name, GFP_KERNEL);
1172 if (WARN_ON(!sysname)) {
1173 pr_warning("Can't allocate memory, giving up!\n");
1176 ret = ftrace_set_clr_event(sysname, 0);
1179 if (WARN_ON_ONCE(ret))
1180 pr_warning("error disabling system %s\n",
1186 /* Test with all events enabled */
1188 pr_info("Running tests on all trace events:\n");
1189 pr_info("Testing all events: ");
1191 sysname = kmalloc(4, GFP_KERNEL);
1192 if (WARN_ON(!sysname)) {
1193 pr_warning("Can't allocate memory, giving up!\n");
1196 memcpy(sysname, "*:*", 4);
1197 ret = ftrace_set_clr_event(sysname, 1);
1198 if (WARN_ON_ONCE(ret)) {
1200 pr_warning("error enabling all events\n");
1207 memcpy(sysname, "*:*", 4);
1208 ret = ftrace_set_clr_event(sysname, 0);
1211 if (WARN_ON_ONCE(ret)) {
1212 pr_warning("error disabling all events\n");
1219 #ifdef CONFIG_FUNCTION_TRACER
1221 static DEFINE_PER_CPU(atomic_t, test_event_disable);
1224 function_test_events_call(unsigned long ip, unsigned long parent_ip)
1226 struct ring_buffer_event *event;
1227 struct ftrace_entry *entry;
1228 unsigned long flags;
1234 pc = preempt_count();
1235 resched = ftrace_preempt_disable();
1236 cpu = raw_smp_processor_id();
1237 disabled = atomic_inc_return(&per_cpu(test_event_disable, cpu));
1242 local_save_flags(flags);
1244 event = trace_current_buffer_lock_reserve(TRACE_FN, sizeof(*entry),
1248 entry = ring_buffer_event_data(event);
1250 entry->parent_ip = parent_ip;
1252 trace_nowake_buffer_unlock_commit(event, flags, pc);
1255 atomic_dec(&per_cpu(test_event_disable, cpu));
1256 ftrace_preempt_enable(resched);
1259 static struct ftrace_ops trace_ops __initdata =
1261 .func = function_test_events_call,
1264 static __init void event_trace_self_test_with_function(void)
1266 register_ftrace_function(&trace_ops);
1267 pr_info("Running tests again, along with the function tracer\n");
1268 event_trace_self_tests();
1269 unregister_ftrace_function(&trace_ops);
1272 static __init void event_trace_self_test_with_function(void)
1277 static __init int event_trace_self_tests_init(void)
1280 event_trace_self_tests();
1282 event_trace_self_test_with_function();
1287 late_initcall(event_trace_self_tests_init);