1/*
2 * event tracer
3 *
4 * Copyright (C) 2008 Red Hat Inc, Steven Rostedt <srostedt@redhat.com>
5 *
6 *  - Added format output of fields of the trace point.
7 *    This was based off of work by Tom Zanussi <tzanussi@gmail.com>.
8 *
9 */
10
11#define pr_fmt(fmt) fmt
12
13#include <linux/workqueue.h>
14#include <linux/spinlock.h>
15#include <linux/kthread.h>
16#include <linux/debugfs.h>
17#include <linux/uaccess.h>
18#include <linux/module.h>
19#include <linux/ctype.h>
20#include <linux/slab.h>
21#include <linux/delay.h>
22
23#include <asm/setup.h>
24
25#include "trace_output.h"
26
27#undef TRACE_SYSTEM
28#define TRACE_SYSTEM "TRACE_SYSTEM"
29
30DEFINE_MUTEX(event_mutex);
31
32LIST_HEAD(ftrace_events);
33static LIST_HEAD(ftrace_common_fields);
34
35#define GFP_TRACE (GFP_KERNEL | __GFP_ZERO)
36
37static struct kmem_cache *field_cachep;
38static struct kmem_cache *file_cachep;
39
40#define SYSTEM_FL_FREE_NAME		(1 << 31)
41
42static inline int system_refcount(struct event_subsystem *system)
43{
44	return system->ref_count & ~SYSTEM_FL_FREE_NAME;
45}
46
47static int system_refcount_inc(struct event_subsystem *system)
48{
49	return (system->ref_count++) & ~SYSTEM_FL_FREE_NAME;
50}
51
52static int system_refcount_dec(struct event_subsystem *system)
53{
54	return (--system->ref_count) & ~SYSTEM_FL_FREE_NAME;
55}
56
57/* Double loops, do not use break, only goto's work */
58#define do_for_each_event_file(tr, file)			\
59	list_for_each_entry(tr, &ftrace_trace_arrays, list) {	\
60		list_for_each_entry(file, &tr->events, list)
61
62#define do_for_each_event_file_safe(tr, file)			\
63	list_for_each_entry(tr, &ftrace_trace_arrays, list) {	\
64		struct ftrace_event_file *___n;				\
65		list_for_each_entry_safe(file, ___n, &tr->events, list)
66
67#define while_for_each_event_file()		\
68	}
69
70static struct list_head *
71trace_get_fields(struct ftrace_event_call *event_call)
72{
73	if (!event_call->class->get_fields)
74		return &event_call->class->fields;
75	return event_call->class->get_fields(event_call);
76}
77
78static struct ftrace_event_field *
79__find_event_field(struct list_head *head, char *name)
80{
81	struct ftrace_event_field *field;
82
83	list_for_each_entry(field, head, link) {
84		if (!strcmp(field->name, name))
85			return field;
86	}
87
88	return NULL;
89}
90
91struct ftrace_event_field *
92trace_find_event_field(struct ftrace_event_call *call, char *name)
93{
94	struct ftrace_event_field *field;
95	struct list_head *head;
96
97	field = __find_event_field(&ftrace_common_fields, name);
98	if (field)
99		return field;
100
101	head = trace_get_fields(call);
102	return __find_event_field(head, name);
103}
104
105static int __trace_define_field(struct list_head *head, const char *type,
106				const char *name, int offset, int size,
107				int is_signed, int filter_type)
108{
109	struct ftrace_event_field *field;
110
111	field = kmem_cache_alloc(field_cachep, GFP_TRACE);
112	if (!field)
113		return -ENOMEM;
114
115	field->name = name;
116	field->type = type;
117
118	if (filter_type == FILTER_OTHER)
119		field->filter_type = filter_assign_type(type);
120	else
121		field->filter_type = filter_type;
122
123	field->offset = offset;
124	field->size = size;
125	field->is_signed = is_signed;
126
127	list_add(&field->link, head);
128
129	return 0;
130}
131
132int trace_define_field(struct ftrace_event_call *call, const char *type,
133		       const char *name, int offset, int size, int is_signed,
134		       int filter_type)
135{
136	struct list_head *head;
137
138	if (WARN_ON(!call->class))
139		return 0;
140
141	head = trace_get_fields(call);
142	return __trace_define_field(head, type, name, offset, size,
143				    is_signed, filter_type);
144}
145EXPORT_SYMBOL_GPL(trace_define_field);
146
147#define __common_field(type, item)					\
148	ret = __trace_define_field(&ftrace_common_fields, #type,	\
149				   "common_" #item,			\
150				   offsetof(typeof(ent), item),		\
151				   sizeof(ent.item),			\
152				   is_signed_type(type), FILTER_OTHER);	\
153	if (ret)							\
154		return ret;
155
156static int trace_define_common_fields(void)
157{
158	int ret;
159	struct trace_entry ent;
160
161	__common_field(unsigned short, type);
162	__common_field(unsigned char, flags);
163	__common_field(unsigned char, preempt_count);
164	__common_field(int, pid);
165
166	return ret;
167}
168
169static void trace_destroy_fields(struct ftrace_event_call *call)
170{
171	struct ftrace_event_field *field, *next;
172	struct list_head *head;
173
174	head = trace_get_fields(call);
175	list_for_each_entry_safe(field, next, head, link) {
176		list_del(&field->link);
177		kmem_cache_free(field_cachep, field);
178	}
179}
180
181int trace_event_raw_init(struct ftrace_event_call *call)
182{
183	int id;
184
185	id = register_ftrace_event(&call->event);
186	if (!id)
187		return -ENODEV;
188
189	return 0;
190}
191EXPORT_SYMBOL_GPL(trace_event_raw_init);
192
193void *ftrace_event_buffer_reserve(struct ftrace_event_buffer *fbuffer,
194				  struct ftrace_event_file *ftrace_file,
195				  unsigned long len)
196{
197	struct ftrace_event_call *event_call = ftrace_file->event_call;
198
199	local_save_flags(fbuffer->flags);
200	fbuffer->pc = preempt_count();
201	fbuffer->ftrace_file = ftrace_file;
202
203	fbuffer->event =
204		trace_event_buffer_lock_reserve(&fbuffer->buffer, ftrace_file,
205						event_call->event.type, len,
206						fbuffer->flags, fbuffer->pc);
207	if (!fbuffer->event)
208		return NULL;
209
210	fbuffer->entry = ring_buffer_event_data(fbuffer->event);
211	return fbuffer->entry;
212}
213EXPORT_SYMBOL_GPL(ftrace_event_buffer_reserve);
214
215void ftrace_event_buffer_commit(struct ftrace_event_buffer *fbuffer)
216{
217	event_trigger_unlock_commit(fbuffer->ftrace_file, fbuffer->buffer,
218				    fbuffer->event, fbuffer->entry,
219				    fbuffer->flags, fbuffer->pc);
220}
221EXPORT_SYMBOL_GPL(ftrace_event_buffer_commit);
222
223int ftrace_event_reg(struct ftrace_event_call *call,
224		     enum trace_reg type, void *data)
225{
226	struct ftrace_event_file *file = data;
227
228	WARN_ON(!(call->flags & TRACE_EVENT_FL_TRACEPOINT));
229	switch (type) {
230	case TRACE_REG_REGISTER:
231		return tracepoint_probe_register(call->tp,
232						 call->class->probe,
233						 file);
234	case TRACE_REG_UNREGISTER:
235		tracepoint_probe_unregister(call->tp,
236					    call->class->probe,
237					    file);
238		return 0;
239
240#ifdef CONFIG_PERF_EVENTS
241	case TRACE_REG_PERF_REGISTER:
242		return tracepoint_probe_register(call->tp,
243						 call->class->perf_probe,
244						 call);
245	case TRACE_REG_PERF_UNREGISTER:
246		tracepoint_probe_unregister(call->tp,
247					    call->class->perf_probe,
248					    call);
249		return 0;
250	case TRACE_REG_PERF_OPEN:
251	case TRACE_REG_PERF_CLOSE:
252	case TRACE_REG_PERF_ADD:
253	case TRACE_REG_PERF_DEL:
254		return 0;
255#endif
256	}
257	return 0;
258}
259EXPORT_SYMBOL_GPL(ftrace_event_reg);
260
261void trace_event_enable_cmd_record(bool enable)
262{
263	struct ftrace_event_file *file;
264	struct trace_array *tr;
265
266	mutex_lock(&event_mutex);
267	do_for_each_event_file(tr, file) {
268
269		if (!(file->flags & FTRACE_EVENT_FL_ENABLED))
270			continue;
271
272		if (enable) {
273			tracing_start_cmdline_record();
274			set_bit(FTRACE_EVENT_FL_RECORDED_CMD_BIT, &file->flags);
275		} else {
276			tracing_stop_cmdline_record();
277			clear_bit(FTRACE_EVENT_FL_RECORDED_CMD_BIT, &file->flags);
278		}
279	} while_for_each_event_file();
280	mutex_unlock(&event_mutex);
281}
282
283static int __ftrace_event_enable_disable(struct ftrace_event_file *file,
284					 int enable, int soft_disable)
285{
286	struct ftrace_event_call *call = file->event_call;
287	int ret = 0;
288	int disable;
289
290	switch (enable) {
291	case 0:
292		/*
293		 * When soft_disable is set and enable is cleared, the sm_ref
294		 * reference counter is decremented. If it reaches 0, we want
295		 * to clear the SOFT_DISABLED flag but leave the event in the
296		 * state that it was. That is, if the event was enabled and
297		 * SOFT_DISABLED isn't set, then do nothing. But if SOFT_DISABLED
298		 * is set we do not want the event to be enabled before we
299		 * clear the bit.
300		 *
301		 * When soft_disable is not set but the SOFT_MODE flag is,
302		 * we do nothing. Do not disable the tracepoint, otherwise
303		 * "soft enable"s (clearing the SOFT_DISABLED bit) wont work.
304		 */
305		if (soft_disable) {
306			if (atomic_dec_return(&file->sm_ref) > 0)
307				break;
308			disable = file->flags & FTRACE_EVENT_FL_SOFT_DISABLED;
309			clear_bit(FTRACE_EVENT_FL_SOFT_MODE_BIT, &file->flags);
310		} else
311			disable = !(file->flags & FTRACE_EVENT_FL_SOFT_MODE);
312
313		if (disable && (file->flags & FTRACE_EVENT_FL_ENABLED)) {
314			clear_bit(FTRACE_EVENT_FL_ENABLED_BIT, &file->flags);
315			if (file->flags & FTRACE_EVENT_FL_RECORDED_CMD) {
316				tracing_stop_cmdline_record();
317				clear_bit(FTRACE_EVENT_FL_RECORDED_CMD_BIT, &file->flags);
318			}
319			call->class->reg(call, TRACE_REG_UNREGISTER, file);
320		}
321		/* If in SOFT_MODE, just set the SOFT_DISABLE_BIT, else clear it */
322		if (file->flags & FTRACE_EVENT_FL_SOFT_MODE)
323			set_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &file->flags);
324		else
325			clear_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &file->flags);
326		break;
327	case 1:
328		/*
329		 * When soft_disable is set and enable is set, we want to
330		 * register the tracepoint for the event, but leave the event
331		 * as is. That means, if the event was already enabled, we do
332		 * nothing (but set SOFT_MODE). If the event is disabled, we
333		 * set SOFT_DISABLED before enabling the event tracepoint, so
334		 * it still seems to be disabled.
335		 */
336		if (!soft_disable)
337			clear_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &file->flags);
338		else {
339			if (atomic_inc_return(&file->sm_ref) > 1)
340				break;
341			set_bit(FTRACE_EVENT_FL_SOFT_MODE_BIT, &file->flags);
342		}
343
344		if (!(file->flags & FTRACE_EVENT_FL_ENABLED)) {
345
346			/* Keep the event disabled, when going to SOFT_MODE. */
347			if (soft_disable)
348				set_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &file->flags);
349
350			if (trace_flags & TRACE_ITER_RECORD_CMD) {
351				tracing_start_cmdline_record();
352				set_bit(FTRACE_EVENT_FL_RECORDED_CMD_BIT, &file->flags);
353			}
354			ret = call->class->reg(call, TRACE_REG_REGISTER, file);
355			if (ret) {
356				tracing_stop_cmdline_record();
357				pr_info("event trace: Could not enable event "
358					"%s\n", ftrace_event_name(call));
359				break;
360			}
361			set_bit(FTRACE_EVENT_FL_ENABLED_BIT, &file->flags);
362
363			/* WAS_ENABLED gets set but never cleared. */
364			call->flags |= TRACE_EVENT_FL_WAS_ENABLED;
365		}
366		break;
367	}
368
369	return ret;
370}
371
372int trace_event_enable_disable(struct ftrace_event_file *file,
373			       int enable, int soft_disable)
374{
375	return __ftrace_event_enable_disable(file, enable, soft_disable);
376}
377
378static int ftrace_event_enable_disable(struct ftrace_event_file *file,
379				       int enable)
380{
381	return __ftrace_event_enable_disable(file, enable, 0);
382}
383
384static void ftrace_clear_events(struct trace_array *tr)
385{
386	struct ftrace_event_file *file;
387
388	mutex_lock(&event_mutex);
389	list_for_each_entry(file, &tr->events, list) {
390		ftrace_event_enable_disable(file, 0);
391	}
392	mutex_unlock(&event_mutex);
393}
394
395static void __put_system(struct event_subsystem *system)
396{
397	struct event_filter *filter = system->filter;
398
399	WARN_ON_ONCE(system_refcount(system) == 0);
400	if (system_refcount_dec(system))
401		return;
402
403	list_del(&system->list);
404
405	if (filter) {
406		kfree(filter->filter_string);
407		kfree(filter);
408	}
409	if (system->ref_count & SYSTEM_FL_FREE_NAME)
410		kfree(system->name);
411	kfree(system);
412}
413
414static void __get_system(struct event_subsystem *system)
415{
416	WARN_ON_ONCE(system_refcount(system) == 0);
417	system_refcount_inc(system);
418}
419
420static void __get_system_dir(struct ftrace_subsystem_dir *dir)
421{
422	WARN_ON_ONCE(dir->ref_count == 0);
423	dir->ref_count++;
424	__get_system(dir->subsystem);
425}
426
427static void __put_system_dir(struct ftrace_subsystem_dir *dir)
428{
429	WARN_ON_ONCE(dir->ref_count == 0);
430	/* If the subsystem is about to be freed, the dir must be too */
431	WARN_ON_ONCE(system_refcount(dir->subsystem) == 1 && dir->ref_count != 1);
432
433	__put_system(dir->subsystem);
434	if (!--dir->ref_count)
435		kfree(dir);
436}
437
438static void put_system(struct ftrace_subsystem_dir *dir)
439{
440	mutex_lock(&event_mutex);
441	__put_system_dir(dir);
442	mutex_unlock(&event_mutex);
443}
444
445static void remove_subsystem(struct ftrace_subsystem_dir *dir)
446{
447	if (!dir)
448		return;
449
450	if (!--dir->nr_events) {
451		debugfs_remove_recursive(dir->entry);
452		list_del(&dir->list);
453		__put_system_dir(dir);
454	}
455}
456
457static void remove_event_file_dir(struct ftrace_event_file *file)
458{
459	struct dentry *dir = file->dir;
460	struct dentry *child;
461
462	if (dir) {
463		spin_lock(&dir->d_lock);	/* probably unneeded */
464		list_for_each_entry(child, &dir->d_subdirs, d_u.d_child) {
465			if (child->d_inode)	/* probably unneeded */
466				child->d_inode->i_private = NULL;
467		}
468		spin_unlock(&dir->d_lock);
469
470		debugfs_remove_recursive(dir);
471	}
472
473	list_del(&file->list);
474	remove_subsystem(file->system);
475	free_event_filter(file->filter);
476	kmem_cache_free(file_cachep, file);
477}
478
479/*
480 * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
481 */
482static int
483__ftrace_set_clr_event_nolock(struct trace_array *tr, const char *match,
484			      const char *sub, const char *event, int set)
485{
486	struct ftrace_event_file *file;
487	struct ftrace_event_call *call;
488	const char *name;
489	int ret = -EINVAL;
490
491	list_for_each_entry(file, &tr->events, list) {
492
493		call = file->event_call;
494		name = ftrace_event_name(call);
495
496		if (!name || !call->class || !call->class->reg)
497			continue;
498
499		if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
500			continue;
501
502		if (match &&
503		    strcmp(match, name) != 0 &&
504		    strcmp(match, call->class->system) != 0)
505			continue;
506
507		if (sub && strcmp(sub, call->class->system) != 0)
508			continue;
509
510		if (event && strcmp(event, name) != 0)
511			continue;
512
513		ftrace_event_enable_disable(file, set);
514
515		ret = 0;
516	}
517
518	return ret;
519}
520
521static int __ftrace_set_clr_event(struct trace_array *tr, const char *match,
522				  const char *sub, const char *event, int set)
523{
524	int ret;
525
526	mutex_lock(&event_mutex);
527	ret = __ftrace_set_clr_event_nolock(tr, match, sub, event, set);
528	mutex_unlock(&event_mutex);
529
530	return ret;
531}
532
533static int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set)
534{
535	char *event = NULL, *sub = NULL, *match;
536
537	/*
538	 * The buf format can be <subsystem>:<event-name>
539	 *  *:<event-name> means any event by that name.
540	 *  :<event-name> is the same.
541	 *
542	 *  <subsystem>:* means all events in that subsystem
543	 *  <subsystem>: means the same.
544	 *
545	 *  <name> (no ':') means all events in a subsystem with
546	 *  the name <name> or any event that matches <name>
547	 */
548
549	match = strsep(&buf, ":");
550	if (buf) {
551		sub = match;
552		event = buf;
553		match = NULL;
554
555		if (!strlen(sub) || strcmp(sub, "*") == 0)
556			sub = NULL;
557		if (!strlen(event) || strcmp(event, "*") == 0)
558			event = NULL;
559	}
560
561	return __ftrace_set_clr_event(tr, match, sub, event, set);
562}
563
564/**
565 * trace_set_clr_event - enable or disable an event
566 * @system: system name to match (NULL for any system)
567 * @event: event name to match (NULL for all events, within system)
568 * @set: 1 to enable, 0 to disable
569 *
570 * This is a way for other parts of the kernel to enable or disable
571 * event recording.
572 *
573 * Returns 0 on success, -EINVAL if the parameters do not match any
574 * registered events.
575 */
576int trace_set_clr_event(const char *system, const char *event, int set)
577{
578	struct trace_array *tr = top_trace_array();
579
580	if (!tr)
581		return -ENODEV;
582
583	return __ftrace_set_clr_event(tr, NULL, system, event, set);
584}
585EXPORT_SYMBOL_GPL(trace_set_clr_event);
586
587/* 128 should be much more than enough */
588#define EVENT_BUF_SIZE		127
589
590static ssize_t
591ftrace_event_write(struct file *file, const char __user *ubuf,
592		   size_t cnt, loff_t *ppos)
593{
594	struct trace_parser parser;
595	struct seq_file *m = file->private_data;
596	struct trace_array *tr = m->private;
597	ssize_t read, ret;
598
599	if (!cnt)
600		return 0;
601
602	ret = tracing_update_buffers();
603	if (ret < 0)
604		return ret;
605
606	if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
607		return -ENOMEM;
608
609	read = trace_get_user(&parser, ubuf, cnt, ppos);
610
611	if (read >= 0 && trace_parser_loaded((&parser))) {
612		int set = 1;
613
614		if (*parser.buffer == '!')
615			set = 0;
616
617		parser.buffer[parser.idx] = 0;
618
619		ret = ftrace_set_clr_event(tr, parser.buffer + !set, set);
620		if (ret)
621			goto out_put;
622	}
623
624	ret = read;
625
626 out_put:
627	trace_parser_put(&parser);
628
629	return ret;
630}
631
632static void *
633t_next(struct seq_file *m, void *v, loff_t *pos)
634{
635	struct ftrace_event_file *file = v;
636	struct ftrace_event_call *call;
637	struct trace_array *tr = m->private;
638
639	(*pos)++;
640
641	list_for_each_entry_continue(file, &tr->events, list) {
642		call = file->event_call;
643		/*
644		 * The ftrace subsystem is for showing formats only.
645		 * They can not be enabled or disabled via the event files.
646		 */
647		if (call->class && call->class->reg)
648			return file;
649	}
650
651	return NULL;
652}
653
654static void *t_start(struct seq_file *m, loff_t *pos)
655{
656	struct ftrace_event_file *file;
657	struct trace_array *tr = m->private;
658	loff_t l;
659
660	mutex_lock(&event_mutex);
661
662	file = list_entry(&tr->events, struct ftrace_event_file, list);
663	for (l = 0; l <= *pos; ) {
664		file = t_next(m, file, &l);
665		if (!file)
666			break;
667	}
668	return file;
669}
670
671static void *
672s_next(struct seq_file *m, void *v, loff_t *pos)
673{
674	struct ftrace_event_file *file = v;
675	struct trace_array *tr = m->private;
676
677	(*pos)++;
678
679	list_for_each_entry_continue(file, &tr->events, list) {
680		if (file->flags & FTRACE_EVENT_FL_ENABLED)
681			return file;
682	}
683
684	return NULL;
685}
686
687static void *s_start(struct seq_file *m, loff_t *pos)
688{
689	struct ftrace_event_file *file;
690	struct trace_array *tr = m->private;
691	loff_t l;
692
693	mutex_lock(&event_mutex);
694
695	file = list_entry(&tr->events, struct ftrace_event_file, list);
696	for (l = 0; l <= *pos; ) {
697		file = s_next(m, file, &l);
698		if (!file)
699			break;
700	}
701	return file;
702}
703
704static int t_show(struct seq_file *m, void *v)
705{
706	struct ftrace_event_file *file = v;
707	struct ftrace_event_call *call = file->event_call;
708
709	if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
710		seq_printf(m, "%s:", call->class->system);
711	seq_printf(m, "%s\n", ftrace_event_name(call));
712
713	return 0;
714}
715
716static void t_stop(struct seq_file *m, void *p)
717{
718	mutex_unlock(&event_mutex);
719}
720
721static ssize_t
722event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
723		  loff_t *ppos)
724{
725	struct ftrace_event_file *file;
726	unsigned long flags;
727	char buf[4] = "0";
728
729	mutex_lock(&event_mutex);
730	file = event_file_data(filp);
731	if (likely(file))
732		flags = file->flags;
733	mutex_unlock(&event_mutex);
734
735	if (!file)
736		return -ENODEV;
737
738	if (flags & FTRACE_EVENT_FL_ENABLED &&
739	    !(flags & FTRACE_EVENT_FL_SOFT_DISABLED))
740		strcpy(buf, "1");
741
742	if (flags & FTRACE_EVENT_FL_SOFT_DISABLED ||
743	    flags & FTRACE_EVENT_FL_SOFT_MODE)
744		strcat(buf, "*");
745
746	strcat(buf, "\n");
747
748	return simple_read_from_buffer(ubuf, cnt, ppos, buf, strlen(buf));
749}
750
751static ssize_t
752event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
753		   loff_t *ppos)
754{
755	struct ftrace_event_file *file;
756	unsigned long val;
757	int ret;
758
759	ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
760	if (ret)
761		return ret;
762
763	ret = tracing_update_buffers();
764	if (ret < 0)
765		return ret;
766
767	switch (val) {
768	case 0:
769	case 1:
770		ret = -ENODEV;
771		mutex_lock(&event_mutex);
772		file = event_file_data(filp);
773		if (likely(file))
774			ret = ftrace_event_enable_disable(file, val);
775		mutex_unlock(&event_mutex);
776		break;
777
778	default:
779		return -EINVAL;
780	}
781
782	*ppos += cnt;
783
784	return ret ? ret : cnt;
785}
786
787static ssize_t
788system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
789		   loff_t *ppos)
790{
791	const char set_to_char[4] = { '?', '0', '1', 'X' };
792	struct ftrace_subsystem_dir *dir = filp->private_data;
793	struct event_subsystem *system = dir->subsystem;
794	struct ftrace_event_call *call;
795	struct ftrace_event_file *file;
796	struct trace_array *tr = dir->tr;
797	char buf[2];
798	int set = 0;
799	int ret;
800
801	mutex_lock(&event_mutex);
802	list_for_each_entry(file, &tr->events, list) {
803		call = file->event_call;
804		if (!ftrace_event_name(call) || !call->class || !call->class->reg)
805			continue;
806
807		if (system && strcmp(call->class->system, system->name) != 0)
808			continue;
809
810		/*
811		 * We need to find out if all the events are set
812		 * or if all events or cleared, or if we have
813		 * a mixture.
814		 */
815		set |= (1 << !!(file->flags & FTRACE_EVENT_FL_ENABLED));
816
817		/*
818		 * If we have a mixture, no need to look further.
819		 */
820		if (set == 3)
821			break;
822	}
823	mutex_unlock(&event_mutex);
824
825	buf[0] = set_to_char[set];
826	buf[1] = '\n';
827
828	ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
829
830	return ret;
831}
832
833static ssize_t
834system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
835		    loff_t *ppos)
836{
837	struct ftrace_subsystem_dir *dir = filp->private_data;
838	struct event_subsystem *system = dir->subsystem;
839	const char *name = NULL;
840	unsigned long val;
841	ssize_t ret;
842
843	ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
844	if (ret)
845		return ret;
846
847	ret = tracing_update_buffers();
848	if (ret < 0)
849		return ret;
850
851	if (val != 0 && val != 1)
852		return -EINVAL;
853
854	/*
855	 * Opening of "enable" adds a ref count to system,
856	 * so the name is safe to use.
857	 */
858	if (system)
859		name = system->name;
860
861	ret = __ftrace_set_clr_event(dir->tr, NULL, name, NULL, val);
862	if (ret)
863		goto out;
864
865	ret = cnt;
866
867out:
868	*ppos += cnt;
869
870	return ret;
871}
872
873enum {
874	FORMAT_HEADER		= 1,
875	FORMAT_FIELD_SEPERATOR	= 2,
876	FORMAT_PRINTFMT		= 3,
877};
878
879static void *f_next(struct seq_file *m, void *v, loff_t *pos)
880{
881	struct ftrace_event_call *call = event_file_data(m->private);
882	struct list_head *common_head = &ftrace_common_fields;
883	struct list_head *head = trace_get_fields(call);
884	struct list_head *node = v;
885
886	(*pos)++;
887
888	switch ((unsigned long)v) {
889	case FORMAT_HEADER:
890		node = common_head;
891		break;
892
893	case FORMAT_FIELD_SEPERATOR:
894		node = head;
895		break;
896
897	case FORMAT_PRINTFMT:
898		/* all done */
899		return NULL;
900	}
901
902	node = node->prev;
903	if (node == common_head)
904		return (void *)FORMAT_FIELD_SEPERATOR;
905	else if (node == head)
906		return (void *)FORMAT_PRINTFMT;
907	else
908		return node;
909}
910
911static int f_show(struct seq_file *m, void *v)
912{
913	struct ftrace_event_call *call = event_file_data(m->private);
914	struct ftrace_event_field *field;
915	const char *array_descriptor;
916
917	switch ((unsigned long)v) {
918	case FORMAT_HEADER:
919		seq_printf(m, "name: %s\n", ftrace_event_name(call));
920		seq_printf(m, "ID: %d\n", call->event.type);
921		seq_printf(m, "format:\n");
922		return 0;
923
924	case FORMAT_FIELD_SEPERATOR:
925		seq_putc(m, '\n');
926		return 0;
927
928	case FORMAT_PRINTFMT:
929		seq_printf(m, "\nprint fmt: %s\n",
930			   call->print_fmt);
931		return 0;
932	}
933
934	field = list_entry(v, struct ftrace_event_field, link);
935	/*
936	 * Smartly shows the array type(except dynamic array).
937	 * Normal:
938	 *	field:TYPE VAR
939	 * If TYPE := TYPE[LEN], it is shown:
940	 *	field:TYPE VAR[LEN]
941	 */
942	array_descriptor = strchr(field->type, '[');
943
944	if (!strncmp(field->type, "__data_loc", 10))
945		array_descriptor = NULL;
946
947	if (!array_descriptor)
948		seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
949			   field->type, field->name, field->offset,
950			   field->size, !!field->is_signed);
951	else
952		seq_printf(m, "\tfield:%.*s %s%s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
953			   (int)(array_descriptor - field->type),
954			   field->type, field->name,
955			   array_descriptor, field->offset,
956			   field->size, !!field->is_signed);
957
958	return 0;
959}
960
961static void *f_start(struct seq_file *m, loff_t *pos)
962{
963	void *p = (void *)FORMAT_HEADER;
964	loff_t l = 0;
965
966	/* ->stop() is called even if ->start() fails */
967	mutex_lock(&event_mutex);
968	if (!event_file_data(m->private))
969		return ERR_PTR(-ENODEV);
970
971	while (l < *pos && p)
972		p = f_next(m, p, &l);
973
974	return p;
975}
976
977static void f_stop(struct seq_file *m, void *p)
978{
979	mutex_unlock(&event_mutex);
980}
981
982static const struct seq_operations trace_format_seq_ops = {
983	.start		= f_start,
984	.next		= f_next,
985	.stop		= f_stop,
986	.show		= f_show,
987};
988
989static int trace_format_open(struct inode *inode, struct file *file)
990{
991	struct seq_file *m;
992	int ret;
993
994	ret = seq_open(file, &trace_format_seq_ops);
995	if (ret < 0)
996		return ret;
997
998	m = file->private_data;
999	m->private = file;
1000
1001	return 0;
1002}
1003
1004static ssize_t
1005event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1006{
1007	int id = (long)event_file_data(filp);
1008	char buf[32];
1009	int len;
1010
1011	if (*ppos)
1012		return 0;
1013
1014	if (unlikely(!id))
1015		return -ENODEV;
1016
1017	len = sprintf(buf, "%d\n", id);
1018
1019	return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
1020}
1021
1022static ssize_t
1023event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1024		  loff_t *ppos)
1025{
1026	struct ftrace_event_file *file;
1027	struct trace_seq *s;
1028	int r = -ENODEV;
1029
1030	if (*ppos)
1031		return 0;
1032
1033	s = kmalloc(sizeof(*s), GFP_KERNEL);
1034
1035	if (!s)
1036		return -ENOMEM;
1037
1038	trace_seq_init(s);
1039
1040	mutex_lock(&event_mutex);
1041	file = event_file_data(filp);
1042	if (file)
1043		print_event_filter(file, s);
1044	mutex_unlock(&event_mutex);
1045
1046	if (file)
1047		r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
1048
1049	kfree(s);
1050
1051	return r;
1052}
1053
1054static ssize_t
1055event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1056		   loff_t *ppos)
1057{
1058	struct ftrace_event_file *file;
1059	char *buf;
1060	int err = -ENODEV;
1061
1062	if (cnt >= PAGE_SIZE)
1063		return -EINVAL;
1064
1065	buf = (char *)__get_free_page(GFP_TEMPORARY);
1066	if (!buf)
1067		return -ENOMEM;
1068
1069	if (copy_from_user(buf, ubuf, cnt)) {
1070		free_page((unsigned long) buf);
1071		return -EFAULT;
1072	}
1073	buf[cnt] = '\0';
1074
1075	mutex_lock(&event_mutex);
1076	file = event_file_data(filp);
1077	if (file)
1078		err = apply_event_filter(file, buf);
1079	mutex_unlock(&event_mutex);
1080
1081	free_page((unsigned long) buf);
1082	if (err < 0)
1083		return err;
1084
1085	*ppos += cnt;
1086
1087	return cnt;
1088}
1089
1090static LIST_HEAD(event_subsystems);
1091
1092static int subsystem_open(struct inode *inode, struct file *filp)
1093{
1094	struct event_subsystem *system = NULL;
1095	struct ftrace_subsystem_dir *dir = NULL; /* Initialize for gcc */
1096	struct trace_array *tr;
1097	int ret;
1098
1099	if (tracing_is_disabled())
1100		return -ENODEV;
1101
1102	/* Make sure the system still exists */
1103	mutex_lock(&trace_types_lock);
1104	mutex_lock(&event_mutex);
1105	list_for_each_entry(tr, &ftrace_trace_arrays, list) {
1106		list_for_each_entry(dir, &tr->systems, list) {
1107			if (dir == inode->i_private) {
1108				/* Don't open systems with no events */
1109				if (dir->nr_events) {
1110					__get_system_dir(dir);
1111					system = dir->subsystem;
1112				}
1113				goto exit_loop;
1114			}
1115		}
1116	}
1117 exit_loop:
1118	mutex_unlock(&event_mutex);
1119	mutex_unlock(&trace_types_lock);
1120
1121	if (!system)
1122		return -ENODEV;
1123
1124	/* Some versions of gcc think dir can be uninitialized here */
1125	WARN_ON(!dir);
1126
1127	/* Still need to increment the ref count of the system */
1128	if (trace_array_get(tr) < 0) {
1129		put_system(dir);
1130		return -ENODEV;
1131	}
1132
1133	ret = tracing_open_generic(inode, filp);
1134	if (ret < 0) {
1135		trace_array_put(tr);
1136		put_system(dir);
1137	}
1138
1139	return ret;
1140}
1141
1142static int system_tr_open(struct inode *inode, struct file *filp)
1143{
1144	struct ftrace_subsystem_dir *dir;
1145	struct trace_array *tr = inode->i_private;
1146	int ret;
1147
1148	if (tracing_is_disabled())
1149		return -ENODEV;
1150
1151	if (trace_array_get(tr) < 0)
1152		return -ENODEV;
1153
1154	/* Make a temporary dir that has no system but points to tr */
1155	dir = kzalloc(sizeof(*dir), GFP_KERNEL);
1156	if (!dir) {
1157		trace_array_put(tr);
1158		return -ENOMEM;
1159	}
1160
1161	dir->tr = tr;
1162
1163	ret = tracing_open_generic(inode, filp);
1164	if (ret < 0) {
1165		trace_array_put(tr);
1166		kfree(dir);
1167		return ret;
1168	}
1169
1170	filp->private_data = dir;
1171
1172	return 0;
1173}
1174
1175static int subsystem_release(struct inode *inode, struct file *file)
1176{
1177	struct ftrace_subsystem_dir *dir = file->private_data;
1178
1179	trace_array_put(dir->tr);
1180
1181	/*
1182	 * If dir->subsystem is NULL, then this is a temporary
1183	 * descriptor that was made for a trace_array to enable
1184	 * all subsystems.
1185	 */
1186	if (dir->subsystem)
1187		put_system(dir);
1188	else
1189		kfree(dir);
1190
1191	return 0;
1192}
1193
1194static ssize_t
1195subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1196		      loff_t *ppos)
1197{
1198	struct ftrace_subsystem_dir *dir = filp->private_data;
1199	struct event_subsystem *system = dir->subsystem;
1200	struct trace_seq *s;
1201	int r;
1202
1203	if (*ppos)
1204		return 0;
1205
1206	s = kmalloc(sizeof(*s), GFP_KERNEL);
1207	if (!s)
1208		return -ENOMEM;
1209
1210	trace_seq_init(s);
1211
1212	print_subsystem_event_filter(system, s);
1213	r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
1214
1215	kfree(s);
1216
1217	return r;
1218}
1219
1220static ssize_t
1221subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1222		       loff_t *ppos)
1223{
1224	struct ftrace_subsystem_dir *dir = filp->private_data;
1225	char *buf;
1226	int err;
1227
1228	if (cnt >= PAGE_SIZE)
1229		return -EINVAL;
1230
1231	buf = (char *)__get_free_page(GFP_TEMPORARY);
1232	if (!buf)
1233		return -ENOMEM;
1234
1235	if (copy_from_user(buf, ubuf, cnt)) {
1236		free_page((unsigned long) buf);
1237		return -EFAULT;
1238	}
1239	buf[cnt] = '\0';
1240
1241	err = apply_subsystem_event_filter(dir, buf);
1242	free_page((unsigned long) buf);
1243	if (err < 0)
1244		return err;
1245
1246	*ppos += cnt;
1247
1248	return cnt;
1249}
1250
1251static ssize_t
1252show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1253{
1254	int (*func)(struct trace_seq *s) = filp->private_data;
1255	struct trace_seq *s;
1256	int r;
1257
1258	if (*ppos)
1259		return 0;
1260
1261	s = kmalloc(sizeof(*s), GFP_KERNEL);
1262	if (!s)
1263		return -ENOMEM;
1264
1265	trace_seq_init(s);
1266
1267	func(s);
1268	r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
1269
1270	kfree(s);
1271
1272	return r;
1273}
1274
1275static int ftrace_event_avail_open(struct inode *inode, struct file *file);
1276static int ftrace_event_set_open(struct inode *inode, struct file *file);
1277static int ftrace_event_release(struct inode *inode, struct file *file);
1278
1279static const struct seq_operations show_event_seq_ops = {
1280	.start = t_start,
1281	.next = t_next,
1282	.show = t_show,
1283	.stop = t_stop,
1284};
1285
1286static const struct seq_operations show_set_event_seq_ops = {
1287	.start = s_start,
1288	.next = s_next,
1289	.show = t_show,
1290	.stop = t_stop,
1291};
1292
1293static const struct file_operations ftrace_avail_fops = {
1294	.open = ftrace_event_avail_open,
1295	.read = seq_read,
1296	.llseek = seq_lseek,
1297	.release = seq_release,
1298};
1299
1300static const struct file_operations ftrace_set_event_fops = {
1301	.open = ftrace_event_set_open,
1302	.read = seq_read,
1303	.write = ftrace_event_write,
1304	.llseek = seq_lseek,
1305	.release = ftrace_event_release,
1306};
1307
1308static const struct file_operations ftrace_enable_fops = {
1309	.open = tracing_open_generic,
1310	.read = event_enable_read,
1311	.write = event_enable_write,
1312	.llseek = default_llseek,
1313};
1314
1315static const struct file_operations ftrace_event_format_fops = {
1316	.open = trace_format_open,
1317	.read = seq_read,
1318	.llseek = seq_lseek,
1319	.release = seq_release,
1320};
1321
1322static const struct file_operations ftrace_event_id_fops = {
1323	.read = event_id_read,
1324	.llseek = default_llseek,
1325};
1326
1327static const struct file_operations ftrace_event_filter_fops = {
1328	.open = tracing_open_generic,
1329	.read = event_filter_read,
1330	.write = event_filter_write,
1331	.llseek = default_llseek,
1332};
1333
1334static const struct file_operations ftrace_subsystem_filter_fops = {
1335	.open = subsystem_open,
1336	.read = subsystem_filter_read,
1337	.write = subsystem_filter_write,
1338	.llseek = default_llseek,
1339	.release = subsystem_release,
1340};
1341
1342static const struct file_operations ftrace_system_enable_fops = {
1343	.open = subsystem_open,
1344	.read = system_enable_read,
1345	.write = system_enable_write,
1346	.llseek = default_llseek,
1347	.release = subsystem_release,
1348};
1349
1350static const struct file_operations ftrace_tr_enable_fops = {
1351	.open = system_tr_open,
1352	.read = system_enable_read,
1353	.write = system_enable_write,
1354	.llseek = default_llseek,
1355	.release = subsystem_release,
1356};
1357
1358static const struct file_operations ftrace_show_header_fops = {
1359	.open = tracing_open_generic,
1360	.read = show_header,
1361	.llseek = default_llseek,
1362};
1363
1364static int
1365ftrace_event_open(struct inode *inode, struct file *file,
1366		  const struct seq_operations *seq_ops)
1367{
1368	struct seq_file *m;
1369	int ret;
1370
1371	ret = seq_open(file, seq_ops);
1372	if (ret < 0)
1373		return ret;
1374	m = file->private_data;
1375	/* copy tr over to seq ops */
1376	m->private = inode->i_private;
1377
1378	return ret;
1379}
1380
1381static int ftrace_event_release(struct inode *inode, struct file *file)
1382{
1383	struct trace_array *tr = inode->i_private;
1384
1385	trace_array_put(tr);
1386
1387	return seq_release(inode, file);
1388}
1389
1390static int
1391ftrace_event_avail_open(struct inode *inode, struct file *file)
1392{
1393	const struct seq_operations *seq_ops = &show_event_seq_ops;
1394
1395	return ftrace_event_open(inode, file, seq_ops);
1396}
1397
1398static int
1399ftrace_event_set_open(struct inode *inode, struct file *file)
1400{
1401	const struct seq_operations *seq_ops = &show_set_event_seq_ops;
1402	struct trace_array *tr = inode->i_private;
1403	int ret;
1404
1405	if (trace_array_get(tr) < 0)
1406		return -ENODEV;
1407
1408	if ((file->f_mode & FMODE_WRITE) &&
1409	    (file->f_flags & O_TRUNC))
1410		ftrace_clear_events(tr);
1411
1412	ret = ftrace_event_open(inode, file, seq_ops);
1413	if (ret < 0)
1414		trace_array_put(tr);
1415	return ret;
1416}
1417
1418static struct event_subsystem *
1419create_new_subsystem(const char *name)
1420{
1421	struct event_subsystem *system;
1422
1423	/* need to create new entry */
1424	system = kmalloc(sizeof(*system), GFP_KERNEL);
1425	if (!system)
1426		return NULL;
1427
1428	system->ref_count = 1;
1429
1430	/* Only allocate if dynamic (kprobes and modules) */
1431	if (!core_kernel_data((unsigned long)name)) {
1432		system->ref_count |= SYSTEM_FL_FREE_NAME;
1433		system->name = kstrdup(name, GFP_KERNEL);
1434		if (!system->name)
1435			goto out_free;
1436	} else
1437		system->name = name;
1438
1439	system->filter = NULL;
1440
1441	system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
1442	if (!system->filter)
1443		goto out_free;
1444
1445	list_add(&system->list, &event_subsystems);
1446
1447	return system;
1448
1449 out_free:
1450	if (system->ref_count & SYSTEM_FL_FREE_NAME)
1451		kfree(system->name);
1452	kfree(system);
1453	return NULL;
1454}
1455
1456static struct dentry *
1457event_subsystem_dir(struct trace_array *tr, const char *name,
1458		    struct ftrace_event_file *file, struct dentry *parent)
1459{
1460	struct ftrace_subsystem_dir *dir;
1461	struct event_subsystem *system;
1462	struct dentry *entry;
1463
1464	/* First see if we did not already create this dir */
1465	list_for_each_entry(dir, &tr->systems, list) {
1466		system = dir->subsystem;
1467		if (strcmp(system->name, name) == 0) {
1468			dir->nr_events++;
1469			file->system = dir;
1470			return dir->entry;
1471		}
1472	}
1473
1474	/* Now see if the system itself exists. */
1475	list_for_each_entry(system, &event_subsystems, list) {
1476		if (strcmp(system->name, name) == 0)
1477			break;
1478	}
1479	/* Reset system variable when not found */
1480	if (&system->list == &event_subsystems)
1481		system = NULL;
1482
1483	dir = kmalloc(sizeof(*dir), GFP_KERNEL);
1484	if (!dir)
1485		goto out_fail;
1486
1487	if (!system) {
1488		system = create_new_subsystem(name);
1489		if (!system)
1490			goto out_free;
1491	} else
1492		__get_system(system);
1493
1494	dir->entry = debugfs_create_dir(name, parent);
1495	if (!dir->entry) {
1496		pr_warn("Failed to create system directory %s\n", name);
1497		__put_system(system);
1498		goto out_free;
1499	}
1500
1501	dir->tr = tr;
1502	dir->ref_count = 1;
1503	dir->nr_events = 1;
1504	dir->subsystem = system;
1505	file->system = dir;
1506
1507	entry = debugfs_create_file("filter", 0644, dir->entry, dir,
1508				    &ftrace_subsystem_filter_fops);
1509	if (!entry) {
1510		kfree(system->filter);
1511		system->filter = NULL;
1512		pr_warn("Could not create debugfs '%s/filter' entry\n", name);
1513	}
1514
1515	trace_create_file("enable", 0644, dir->entry, dir,
1516			  &ftrace_system_enable_fops);
1517
1518	list_add(&dir->list, &tr->systems);
1519
1520	return dir->entry;
1521
1522 out_free:
1523	kfree(dir);
1524 out_fail:
1525	/* Only print this message if failed on memory allocation */
1526	if (!dir || !system)
1527		pr_warn("No memory to create event subsystem %s\n", name);
1528	return NULL;
1529}
1530
1531static int
1532event_create_dir(struct dentry *parent, struct ftrace_event_file *file)
1533{
1534	struct ftrace_event_call *call = file->event_call;
1535	struct trace_array *tr = file->tr;
1536	struct list_head *head;
1537	struct dentry *d_events;
1538	const char *name;
1539	int ret;
1540
1541	/*
1542	 * If the trace point header did not define TRACE_SYSTEM
1543	 * then the system would be called "TRACE_SYSTEM".
1544	 */
1545	if (strcmp(call->class->system, TRACE_SYSTEM) != 0) {
1546		d_events = event_subsystem_dir(tr, call->class->system, file, parent);
1547		if (!d_events)
1548			return -ENOMEM;
1549	} else
1550		d_events = parent;
1551
1552	name = ftrace_event_name(call);
1553	file->dir = debugfs_create_dir(name, d_events);
1554	if (!file->dir) {
1555		pr_warn("Could not create debugfs '%s' directory\n", name);
1556		return -1;
1557	}
1558
1559	if (call->class->reg && !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
1560		trace_create_file("enable", 0644, file->dir, file,
1561				  &ftrace_enable_fops);
1562
1563#ifdef CONFIG_PERF_EVENTS
1564	if (call->event.type && call->class->reg)
1565		trace_create_file("id", 0444, file->dir,
1566				  (void *)(long)call->event.type,
1567				  &ftrace_event_id_fops);
1568#endif
1569
1570	/*
1571	 * Other events may have the same class. Only update
1572	 * the fields if they are not already defined.
1573	 */
1574	head = trace_get_fields(call);
1575	if (list_empty(head)) {
1576		ret = call->class->define_fields(call);
1577		if (ret < 0) {
1578			pr_warn("Could not initialize trace point events/%s\n",
1579				name);
1580			return -1;
1581		}
1582	}
1583	trace_create_file("filter", 0644, file->dir, file,
1584			  &ftrace_event_filter_fops);
1585
1586	trace_create_file("trigger", 0644, file->dir, file,
1587			  &event_trigger_fops);
1588
1589	trace_create_file("format", 0444, file->dir, call,
1590			  &ftrace_event_format_fops);
1591
1592	return 0;
1593}
1594
1595static void remove_event_from_tracers(struct ftrace_event_call *call)
1596{
1597	struct ftrace_event_file *file;
1598	struct trace_array *tr;
1599
1600	do_for_each_event_file_safe(tr, file) {
1601		if (file->event_call != call)
1602			continue;
1603
1604		remove_event_file_dir(file);
1605		/*
1606		 * The do_for_each_event_file_safe() is
1607		 * a double loop. After finding the call for this
1608		 * trace_array, we use break to jump to the next
1609		 * trace_array.
1610		 */
1611		break;
1612	} while_for_each_event_file();
1613}
1614
1615static void event_remove(struct ftrace_event_call *call)
1616{
1617	struct trace_array *tr;
1618	struct ftrace_event_file *file;
1619
1620	do_for_each_event_file(tr, file) {
1621		if (file->event_call != call)
1622			continue;
1623		ftrace_event_enable_disable(file, 0);
1624		/*
1625		 * The do_for_each_event_file() is
1626		 * a double loop. After finding the call for this
1627		 * trace_array, we use break to jump to the next
1628		 * trace_array.
1629		 */
1630		break;
1631	} while_for_each_event_file();
1632
1633	if (call->event.funcs)
1634		__unregister_ftrace_event(&call->event);
1635	remove_event_from_tracers(call);
1636	list_del(&call->list);
1637}
1638
1639static int event_init(struct ftrace_event_call *call)
1640{
1641	int ret = 0;
1642	const char *name;
1643
1644	name = ftrace_event_name(call);
1645	if (WARN_ON(!name))
1646		return -EINVAL;
1647
1648	if (call->class->raw_init) {
1649		ret = call->class->raw_init(call);
1650		if (ret < 0 && ret != -ENOSYS)
1651			pr_warn("Could not initialize trace events/%s\n", name);
1652	}
1653
1654	return ret;
1655}
1656
1657static int
1658__register_event(struct ftrace_event_call *call, struct module *mod)
1659{
1660	int ret;
1661
1662	ret = event_init(call);
1663	if (ret < 0)
1664		return ret;
1665
1666	list_add(&call->list, &ftrace_events);
1667	call->mod = mod;
1668
1669	return 0;
1670}
1671
1672static struct ftrace_event_file *
1673trace_create_new_event(struct ftrace_event_call *call,
1674		       struct trace_array *tr)
1675{
1676	struct ftrace_event_file *file;
1677
1678	file = kmem_cache_alloc(file_cachep, GFP_TRACE);
1679	if (!file)
1680		return NULL;
1681
1682	file->event_call = call;
1683	file->tr = tr;
1684	atomic_set(&file->sm_ref, 0);
1685	atomic_set(&file->tm_ref, 0);
1686	INIT_LIST_HEAD(&file->triggers);
1687	list_add(&file->list, &tr->events);
1688
1689	return file;
1690}
1691
1692/* Add an event to a trace directory */
1693static int
1694__trace_add_new_event(struct ftrace_event_call *call, struct trace_array *tr)
1695{
1696	struct ftrace_event_file *file;
1697
1698	file = trace_create_new_event(call, tr);
1699	if (!file)
1700		return -ENOMEM;
1701
1702	return event_create_dir(tr->event_dir, file);
1703}
1704
1705/*
1706 * Just create a decriptor for early init. A descriptor is required
1707 * for enabling events at boot. We want to enable events before
1708 * the filesystem is initialized.
1709 */
1710static __init int
1711__trace_early_add_new_event(struct ftrace_event_call *call,
1712			    struct trace_array *tr)
1713{
1714	struct ftrace_event_file *file;
1715
1716	file = trace_create_new_event(call, tr);
1717	if (!file)
1718		return -ENOMEM;
1719
1720	return 0;
1721}
1722
1723struct ftrace_module_file_ops;
1724static void __add_event_to_tracers(struct ftrace_event_call *call);
1725
1726/* Add an additional event_call dynamically */
1727int trace_add_event_call(struct ftrace_event_call *call)
1728{
1729	int ret;
1730	mutex_lock(&trace_types_lock);
1731	mutex_lock(&event_mutex);
1732
1733	ret = __register_event(call, NULL);
1734	if (ret >= 0)
1735		__add_event_to_tracers(call);
1736
1737	mutex_unlock(&event_mutex);
1738	mutex_unlock(&trace_types_lock);
1739	return ret;
1740}
1741
1742/*
1743 * Must be called under locking of trace_types_lock, event_mutex and
1744 * trace_event_sem.
1745 */
1746static void __trace_remove_event_call(struct ftrace_event_call *call)
1747{
1748	event_remove(call);
1749	trace_destroy_fields(call);
1750	free_event_filter(call->filter);
1751	call->filter = NULL;
1752}
1753
1754static int probe_remove_event_call(struct ftrace_event_call *call)
1755{
1756	struct trace_array *tr;
1757	struct ftrace_event_file *file;
1758
1759#ifdef CONFIG_PERF_EVENTS
1760	if (call->perf_refcount)
1761		return -EBUSY;
1762#endif
1763	do_for_each_event_file(tr, file) {
1764		if (file->event_call != call)
1765			continue;
1766		/*
1767		 * We can't rely on ftrace_event_enable_disable(enable => 0)
1768		 * we are going to do, FTRACE_EVENT_FL_SOFT_MODE can suppress
1769		 * TRACE_REG_UNREGISTER.
1770		 */
1771		if (file->flags & FTRACE_EVENT_FL_ENABLED)
1772			return -EBUSY;
1773		/*
1774		 * The do_for_each_event_file_safe() is
1775		 * a double loop. After finding the call for this
1776		 * trace_array, we use break to jump to the next
1777		 * trace_array.
1778		 */
1779		break;
1780	} while_for_each_event_file();
1781
1782	__trace_remove_event_call(call);
1783
1784	return 0;
1785}
1786
1787/* Remove an event_call */
1788int trace_remove_event_call(struct ftrace_event_call *call)
1789{
1790	int ret;
1791
1792	mutex_lock(&trace_types_lock);
1793	mutex_lock(&event_mutex);
1794	down_write(&trace_event_sem);
1795	ret = probe_remove_event_call(call);
1796	up_write(&trace_event_sem);
1797	mutex_unlock(&event_mutex);
1798	mutex_unlock(&trace_types_lock);
1799
1800	return ret;
1801}
1802
1803#define for_each_event(event, start, end)			\
1804	for (event = start;					\
1805	     (unsigned long)event < (unsigned long)end;		\
1806	     event++)
1807
1808#ifdef CONFIG_MODULES
1809
1810static void trace_module_add_events(struct module *mod)
1811{
1812	struct ftrace_event_call **call, **start, **end;
1813
1814	if (!mod->num_trace_events)
1815		return;
1816
1817	/* Don't add infrastructure for mods without tracepoints */
1818	if (trace_module_has_bad_taint(mod)) {
1819		pr_err("%s: module has bad taint, not creating trace events\n",
1820		       mod->name);
1821		return;
1822	}
1823
1824	start = mod->trace_events;
1825	end = mod->trace_events + mod->num_trace_events;
1826
1827	for_each_event(call, start, end) {
1828		__register_event(*call, mod);
1829		__add_event_to_tracers(*call);
1830	}
1831}
1832
1833static void trace_module_remove_events(struct module *mod)
1834{
1835	struct ftrace_event_call *call, *p;
1836	bool clear_trace = false;
1837
1838	down_write(&trace_event_sem);
1839	list_for_each_entry_safe(call, p, &ftrace_events, list) {
1840		if (call->mod == mod) {
1841			if (call->flags & TRACE_EVENT_FL_WAS_ENABLED)
1842				clear_trace = true;
1843			__trace_remove_event_call(call);
1844		}
1845	}
1846	up_write(&trace_event_sem);
1847
1848	/*
1849	 * It is safest to reset the ring buffer if the module being unloaded
1850	 * registered any events that were used. The only worry is if
1851	 * a new module gets loaded, and takes on the same id as the events
1852	 * of this module. When printing out the buffer, traced events left
1853	 * over from this module may be passed to the new module events and
1854	 * unexpected results may occur.
1855	 */
1856	if (clear_trace)
1857		tracing_reset_all_online_cpus();
1858}
1859
1860static int trace_module_notify(struct notifier_block *self,
1861			       unsigned long val, void *data)
1862{
1863	struct module *mod = data;
1864
1865	mutex_lock(&trace_types_lock);
1866	mutex_lock(&event_mutex);
1867	switch (val) {
1868	case MODULE_STATE_COMING:
1869		trace_module_add_events(mod);
1870		break;
1871	case MODULE_STATE_GOING:
1872		trace_module_remove_events(mod);
1873		break;
1874	}
1875	mutex_unlock(&event_mutex);
1876	mutex_unlock(&trace_types_lock);
1877
1878	return 0;
1879}
1880
1881static struct notifier_block trace_module_nb = {
1882	.notifier_call = trace_module_notify,
1883	.priority = 0,
1884};
1885#endif /* CONFIG_MODULES */
1886
1887/* Create a new event directory structure for a trace directory. */
1888static void
1889__trace_add_event_dirs(struct trace_array *tr)
1890{
1891	struct ftrace_event_call *call;
1892	int ret;
1893
1894	list_for_each_entry(call, &ftrace_events, list) {
1895		ret = __trace_add_new_event(call, tr);
1896		if (ret < 0)
1897			pr_warn("Could not create directory for event %s\n",
1898				ftrace_event_name(call));
1899	}
1900}
1901
1902struct ftrace_event_file *
1903find_event_file(struct trace_array *tr, const char *system,  const char *event)
1904{
1905	struct ftrace_event_file *file;
1906	struct ftrace_event_call *call;
1907	const char *name;
1908
1909	list_for_each_entry(file, &tr->events, list) {
1910
1911		call = file->event_call;
1912		name = ftrace_event_name(call);
1913
1914		if (!name || !call->class || !call->class->reg)
1915			continue;
1916
1917		if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
1918			continue;
1919
1920		if (strcmp(event, name) == 0 &&
1921		    strcmp(system, call->class->system) == 0)
1922			return file;
1923	}
1924	return NULL;
1925}
1926
1927#ifdef CONFIG_DYNAMIC_FTRACE
1928
1929/* Avoid typos */
1930#define ENABLE_EVENT_STR	"enable_event"
1931#define DISABLE_EVENT_STR	"disable_event"
1932
1933struct event_probe_data {
1934	struct ftrace_event_file	*file;
1935	unsigned long			count;
1936	int				ref;
1937	bool				enable;
1938};
1939
1940static void
1941event_enable_probe(unsigned long ip, unsigned long parent_ip, void **_data)
1942{
1943	struct event_probe_data **pdata = (struct event_probe_data **)_data;
1944	struct event_probe_data *data = *pdata;
1945
1946	if (!data)
1947		return;
1948
1949	if (data->enable)
1950		clear_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &data->file->flags);
1951	else
1952		set_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &data->file->flags);
1953}
1954
1955static void
1956event_enable_count_probe(unsigned long ip, unsigned long parent_ip, void **_data)
1957{
1958	struct event_probe_data **pdata = (struct event_probe_data **)_data;
1959	struct event_probe_data *data = *pdata;
1960
1961	if (!data)
1962		return;
1963
1964	if (!data->count)
1965		return;
1966
1967	/* Skip if the event is in a state we want to switch to */
1968	if (data->enable == !(data->file->flags & FTRACE_EVENT_FL_SOFT_DISABLED))
1969		return;
1970
1971	if (data->count != -1)
1972		(data->count)--;
1973
1974	event_enable_probe(ip, parent_ip, _data);
1975}
1976
1977static int
1978event_enable_print(struct seq_file *m, unsigned long ip,
1979		      struct ftrace_probe_ops *ops, void *_data)
1980{
1981	struct event_probe_data *data = _data;
1982
1983	seq_printf(m, "%ps:", (void *)ip);
1984
1985	seq_printf(m, "%s:%s:%s",
1986		   data->enable ? ENABLE_EVENT_STR : DISABLE_EVENT_STR,
1987		   data->file->event_call->class->system,
1988		   ftrace_event_name(data->file->event_call));
1989
1990	if (data->count == -1)
1991		seq_printf(m, ":unlimited\n");
1992	else
1993		seq_printf(m, ":count=%ld\n", data->count);
1994
1995	return 0;
1996}
1997
1998static int
1999event_enable_init(struct ftrace_probe_ops *ops, unsigned long ip,
2000		  void **_data)
2001{
2002	struct event_probe_data **pdata = (struct event_probe_data **)_data;
2003	struct event_probe_data *data = *pdata;
2004
2005	data->ref++;
2006	return 0;
2007}
2008
2009static void
2010event_enable_free(struct ftrace_probe_ops *ops, unsigned long ip,
2011		  void **_data)
2012{
2013	struct event_probe_data **pdata = (struct event_probe_data **)_data;
2014	struct event_probe_data *data = *pdata;
2015
2016	if (WARN_ON_ONCE(data->ref <= 0))
2017		return;
2018
2019	data->ref--;
2020	if (!data->ref) {
2021		/* Remove the SOFT_MODE flag */
2022		__ftrace_event_enable_disable(data->file, 0, 1);
2023		module_put(data->file->event_call->mod);
2024		kfree(data);
2025	}
2026	*pdata = NULL;
2027}
2028
2029static struct ftrace_probe_ops event_enable_probe_ops = {
2030	.func			= event_enable_probe,
2031	.print			= event_enable_print,
2032	.init			= event_enable_init,
2033	.free			= event_enable_free,
2034};
2035
2036static struct ftrace_probe_ops event_enable_count_probe_ops = {
2037	.func			= event_enable_count_probe,
2038	.print			= event_enable_print,
2039	.init			= event_enable_init,
2040	.free			= event_enable_free,
2041};
2042
2043static struct ftrace_probe_ops event_disable_probe_ops = {
2044	.func			= event_enable_probe,
2045	.print			= event_enable_print,
2046	.init			= event_enable_init,
2047	.free			= event_enable_free,
2048};
2049
2050static struct ftrace_probe_ops event_disable_count_probe_ops = {
2051	.func			= event_enable_count_probe,
2052	.print			= event_enable_print,
2053	.init			= event_enable_init,
2054	.free			= event_enable_free,
2055};
2056
2057static int
2058event_enable_func(struct ftrace_hash *hash,
2059		  char *glob, char *cmd, char *param, int enabled)
2060{
2061	struct trace_array *tr = top_trace_array();
2062	struct ftrace_event_file *file;
2063	struct ftrace_probe_ops *ops;
2064	struct event_probe_data *data;
2065	const char *system;
2066	const char *event;
2067	char *number;
2068	bool enable;
2069	int ret;
2070
2071	if (!tr)
2072		return -ENODEV;
2073
2074	/* hash funcs only work with set_ftrace_filter */
2075	if (!enabled || !param)
2076		return -EINVAL;
2077
2078	system = strsep(&param, ":");
2079	if (!param)
2080		return -EINVAL;
2081
2082	event = strsep(&param, ":");
2083
2084	mutex_lock(&event_mutex);
2085
2086	ret = -EINVAL;
2087	file = find_event_file(tr, system, event);
2088	if (!file)
2089		goto out;
2090
2091	enable = strcmp(cmd, ENABLE_EVENT_STR) == 0;
2092
2093	if (enable)
2094		ops = param ? &event_enable_count_probe_ops : &event_enable_probe_ops;
2095	else
2096		ops = param ? &event_disable_count_probe_ops : &event_disable_probe_ops;
2097
2098	if (glob[0] == '!') {
2099		unregister_ftrace_function_probe_func(glob+1, ops);
2100		ret = 0;
2101		goto out;
2102	}
2103
2104	ret = -ENOMEM;
2105	data = kzalloc(sizeof(*data), GFP_KERNEL);
2106	if (!data)
2107		goto out;
2108
2109	data->enable = enable;
2110	data->count = -1;
2111	data->file = file;
2112
2113	if (!param)
2114		goto out_reg;
2115
2116	number = strsep(&param, ":");
2117
2118	ret = -EINVAL;
2119	if (!strlen(number))
2120		goto out_free;
2121
2122	/*
2123	 * We use the callback data field (which is a pointer)
2124	 * as our counter.
2125	 */
2126	ret = kstrtoul(number, 0, &data->count);
2127	if (ret)
2128		goto out_free;
2129
2130 out_reg:
2131	/* Don't let event modules unload while probe registered */
2132	ret = try_module_get(file->event_call->mod);
2133	if (!ret) {
2134		ret = -EBUSY;
2135		goto out_free;
2136	}
2137
2138	ret = __ftrace_event_enable_disable(file, 1, 1);
2139	if (ret < 0)
2140		goto out_put;
2141	ret = register_ftrace_function_probe(glob, ops, data);
2142	/*
2143	 * The above returns on success the # of functions enabled,
2144	 * but if it didn't find any functions it returns zero.
2145	 * Consider no functions a failure too.
2146	 */
2147	if (!ret) {
2148		ret = -ENOENT;
2149		goto out_disable;
2150	} else if (ret < 0)
2151		goto out_disable;
2152	/* Just return zero, not the number of enabled functions */
2153	ret = 0;
2154 out:
2155	mutex_unlock(&event_mutex);
2156	return ret;
2157
2158 out_disable:
2159	__ftrace_event_enable_disable(file, 0, 1);
2160 out_put:
2161	module_put(file->event_call->mod);
2162 out_free:
2163	kfree(data);
2164	goto out;
2165}
2166
2167static struct ftrace_func_command event_enable_cmd = {
2168	.name			= ENABLE_EVENT_STR,
2169	.func			= event_enable_func,
2170};
2171
2172static struct ftrace_func_command event_disable_cmd = {
2173	.name			= DISABLE_EVENT_STR,
2174	.func			= event_enable_func,
2175};
2176
2177static __init int register_event_cmds(void)
2178{
2179	int ret;
2180
2181	ret = register_ftrace_command(&event_enable_cmd);
2182	if (WARN_ON(ret < 0))
2183		return ret;
2184	ret = register_ftrace_command(&event_disable_cmd);
2185	if (WARN_ON(ret < 0))
2186		unregister_ftrace_command(&event_enable_cmd);
2187	return ret;
2188}
2189#else
2190static inline int register_event_cmds(void) { return 0; }
2191#endif /* CONFIG_DYNAMIC_FTRACE */
2192
2193/*
2194 * The top level array has already had its ftrace_event_file
2195 * descriptors created in order to allow for early events to
2196 * be recorded. This function is called after the debugfs has been
2197 * initialized, and we now have to create the files associated
2198 * to the events.
2199 */
2200static __init void
2201__trace_early_add_event_dirs(struct trace_array *tr)
2202{
2203	struct ftrace_event_file *file;
2204	int ret;
2205
2206
2207	list_for_each_entry(file, &tr->events, list) {
2208		ret = event_create_dir(tr->event_dir, file);
2209		if (ret < 0)
2210			pr_warn("Could not create directory for event %s\n",
2211				ftrace_event_name(file->event_call));
2212	}
2213}
2214
2215/*
2216 * For early boot up, the top trace array requires to have
2217 * a list of events that can be enabled. This must be done before
2218 * the filesystem is set up in order to allow events to be traced
2219 * early.
2220 */
2221static __init void
2222__trace_early_add_events(struct trace_array *tr)
2223{
2224	struct ftrace_event_call *call;
2225	int ret;
2226
2227	list_for_each_entry(call, &ftrace_events, list) {
2228		/* Early boot up should not have any modules loaded */
2229		if (WARN_ON_ONCE(call->mod))
2230			continue;
2231
2232		ret = __trace_early_add_new_event(call, tr);
2233		if (ret < 0)
2234			pr_warn("Could not create early event %s\n",
2235				ftrace_event_name(call));
2236	}
2237}
2238
2239/* Remove the event directory structure for a trace directory. */
2240static void
2241__trace_remove_event_dirs(struct trace_array *tr)
2242{
2243	struct ftrace_event_file *file, *next;
2244
2245	list_for_each_entry_safe(file, next, &tr->events, list)
2246		remove_event_file_dir(file);
2247}
2248
2249static void __add_event_to_tracers(struct ftrace_event_call *call)
2250{
2251	struct trace_array *tr;
2252
2253	list_for_each_entry(tr, &ftrace_trace_arrays, list)
2254		__trace_add_new_event(call, tr);
2255}
2256
2257extern struct ftrace_event_call *__start_ftrace_events[];
2258extern struct ftrace_event_call *__stop_ftrace_events[];
2259
2260static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
2261
2262static __init int setup_trace_event(char *str)
2263{
2264	strlcpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
2265	ring_buffer_expanded = true;
2266	tracing_selftest_disabled = true;
2267
2268	return 1;
2269}
2270__setup("trace_event=", setup_trace_event);
2271
2272/* Expects to have event_mutex held when called */
2273static int
2274create_event_toplevel_files(struct dentry *parent, struct trace_array *tr)
2275{
2276	struct dentry *d_events;
2277	struct dentry *entry;
2278
2279	entry = debugfs_create_file("set_event", 0644, parent,
2280				    tr, &ftrace_set_event_fops);
2281	if (!entry) {
2282		pr_warn("Could not create debugfs 'set_event' entry\n");
2283		return -ENOMEM;
2284	}
2285
2286	d_events = debugfs_create_dir("events", parent);
2287	if (!d_events) {
2288		pr_warn("Could not create debugfs 'events' directory\n");
2289		return -ENOMEM;
2290	}
2291
2292	/* ring buffer internal formats */
2293	trace_create_file("header_page", 0444, d_events,
2294			  ring_buffer_print_page_header,
2295			  &ftrace_show_header_fops);
2296
2297	trace_create_file("header_event", 0444, d_events,
2298			  ring_buffer_print_entry_header,
2299			  &ftrace_show_header_fops);
2300
2301	trace_create_file("enable", 0644, d_events,
2302			  tr, &ftrace_tr_enable_fops);
2303
2304	tr->event_dir = d_events;
2305
2306	return 0;
2307}
2308
2309/**
2310 * event_trace_add_tracer - add a instance of a trace_array to events
2311 * @parent: The parent dentry to place the files/directories for events in
2312 * @tr: The trace array associated with these events
2313 *
2314 * When a new instance is created, it needs to set up its events
2315 * directory, as well as other files associated with events. It also
2316 * creates the event hierachry in the @parent/events directory.
2317 *
2318 * Returns 0 on success.
2319 */
2320int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr)
2321{
2322	int ret;
2323
2324	mutex_lock(&event_mutex);
2325
2326	ret = create_event_toplevel_files(parent, tr);
2327	if (ret)
2328		goto out_unlock;
2329
2330	down_write(&trace_event_sem);
2331	__trace_add_event_dirs(tr);
2332	up_write(&trace_event_sem);
2333
2334 out_unlock:
2335	mutex_unlock(&event_mutex);
2336
2337	return ret;
2338}
2339
2340/*
2341 * The top trace array already had its file descriptors created.
2342 * Now the files themselves need to be created.
2343 */
2344static __init int
2345early_event_add_tracer(struct dentry *parent, struct trace_array *tr)
2346{
2347	int ret;
2348
2349	mutex_lock(&event_mutex);
2350
2351	ret = create_event_toplevel_files(parent, tr);
2352	if (ret)
2353		goto out_unlock;
2354
2355	down_write(&trace_event_sem);
2356	__trace_early_add_event_dirs(tr);
2357	up_write(&trace_event_sem);
2358
2359 out_unlock:
2360	mutex_unlock(&event_mutex);
2361
2362	return ret;
2363}
2364
2365int event_trace_del_tracer(struct trace_array *tr)
2366{
2367	mutex_lock(&event_mutex);
2368
2369	/* Disable any event triggers and associated soft-disabled events */
2370	clear_event_triggers(tr);
2371
2372	/* Disable any running events */
2373	__ftrace_set_clr_event_nolock(tr, NULL, NULL, NULL, 0);
2374
2375	/* Access to events are within rcu_read_lock_sched() */
2376	synchronize_sched();
2377
2378	down_write(&trace_event_sem);
2379	__trace_remove_event_dirs(tr);
2380	debugfs_remove_recursive(tr->event_dir);
2381	up_write(&trace_event_sem);
2382
2383	tr->event_dir = NULL;
2384
2385	mutex_unlock(&event_mutex);
2386
2387	return 0;
2388}
2389
2390static __init int event_trace_memsetup(void)
2391{
2392	field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC);
2393	file_cachep = KMEM_CACHE(ftrace_event_file, SLAB_PANIC);
2394	return 0;
2395}
2396
2397static __init int event_trace_enable(void)
2398{
2399	struct trace_array *tr = top_trace_array();
2400	struct ftrace_event_call **iter, *call;
2401	char *buf = bootup_event_buf;
2402	char *token;
2403	int ret;
2404
2405	if (!tr)
2406		return -ENODEV;
2407
2408	for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
2409
2410		call = *iter;
2411		ret = event_init(call);
2412		if (!ret)
2413			list_add(&call->list, &ftrace_events);
2414	}
2415
2416	/*
2417	 * We need the top trace array to have a working set of trace
2418	 * points at early init, before the debug files and directories
2419	 * are created. Create the file entries now, and attach them
2420	 * to the actual file dentries later.
2421	 */
2422	__trace_early_add_events(tr);
2423
2424	while (true) {
2425		token = strsep(&buf, ",");
2426
2427		if (!token)
2428			break;
2429		if (!*token)
2430			continue;
2431
2432		ret = ftrace_set_clr_event(tr, token, 1);
2433		if (ret)
2434			pr_warn("Failed to enable trace event: %s\n", token);
2435	}
2436
2437	trace_printk_start_comm();
2438
2439	register_event_cmds();
2440
2441	register_trigger_cmds();
2442
2443	return 0;
2444}
2445
2446static __init int event_trace_init(void)
2447{
2448	struct trace_array *tr;
2449	struct dentry *d_tracer;
2450	struct dentry *entry;
2451	int ret;
2452
2453	tr = top_trace_array();
2454	if (!tr)
2455		return -ENODEV;
2456
2457	d_tracer = tracing_init_dentry();
2458	if (!d_tracer)
2459		return 0;
2460
2461	entry = debugfs_create_file("available_events", 0444, d_tracer,
2462				    tr, &ftrace_avail_fops);
2463	if (!entry)
2464		pr_warn("Could not create debugfs 'available_events' entry\n");
2465
2466	if (trace_define_common_fields())
2467		pr_warn("tracing: Failed to allocate common fields");
2468
2469	ret = early_event_add_tracer(d_tracer, tr);
2470	if (ret)
2471		return ret;
2472
2473#ifdef CONFIG_MODULES
2474	ret = register_module_notifier(&trace_module_nb);
2475	if (ret)
2476		pr_warn("Failed to register trace events module notifier\n");
2477#endif
2478	return 0;
2479}
2480early_initcall(event_trace_memsetup);
2481core_initcall(event_trace_enable);
2482fs_initcall(event_trace_init);
2483
2484#ifdef CONFIG_FTRACE_STARTUP_TEST
2485
2486static DEFINE_SPINLOCK(test_spinlock);
2487static DEFINE_SPINLOCK(test_spinlock_irq);
2488static DEFINE_MUTEX(test_mutex);
2489
2490static __init void test_work(struct work_struct *dummy)
2491{
2492	spin_lock(&test_spinlock);
2493	spin_lock_irq(&test_spinlock_irq);
2494	udelay(1);
2495	spin_unlock_irq(&test_spinlock_irq);
2496	spin_unlock(&test_spinlock);
2497
2498	mutex_lock(&test_mutex);
2499	msleep(1);
2500	mutex_unlock(&test_mutex);
2501}
2502
2503static __init int event_test_thread(void *unused)
2504{
2505	void *test_malloc;
2506
2507	test_malloc = kmalloc(1234, GFP_KERNEL);
2508	if (!test_malloc)
2509		pr_info("failed to kmalloc\n");
2510
2511	schedule_on_each_cpu(test_work);
2512
2513	kfree(test_malloc);
2514
2515	set_current_state(TASK_INTERRUPTIBLE);
2516	while (!kthread_should_stop()) {
2517		schedule();
2518		set_current_state(TASK_INTERRUPTIBLE);
2519	}
2520	__set_current_state(TASK_RUNNING);
2521
2522	return 0;
2523}
2524
2525/*
2526 * Do various things that may trigger events.
2527 */
2528static __init void event_test_stuff(void)
2529{
2530	struct task_struct *test_thread;
2531
2532	test_thread = kthread_run(event_test_thread, NULL, "test-events");
2533	msleep(1);
2534	kthread_stop(test_thread);
2535}
2536
2537/*
2538 * For every trace event defined, we will test each trace point separately,
2539 * and then by groups, and finally all trace points.
2540 */
2541static __init void event_trace_self_tests(void)
2542{
2543	struct ftrace_subsystem_dir *dir;
2544	struct ftrace_event_file *file;
2545	struct ftrace_event_call *call;
2546	struct event_subsystem *system;
2547	struct trace_array *tr;
2548	int ret;
2549
2550	tr = top_trace_array();
2551	if (!tr)
2552		return;
2553
2554	pr_info("Running tests on trace events:\n");
2555
2556	list_for_each_entry(file, &tr->events, list) {
2557
2558		call = file->event_call;
2559
2560		/* Only test those that have a probe */
2561		if (!call->class || !call->class->probe)
2562			continue;
2563
2564/*
2565 * Testing syscall events here is pretty useless, but
2566 * we still do it if configured. But this is time consuming.
2567 * What we really need is a user thread to perform the
2568 * syscalls as we test.
2569 */
2570#ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
2571		if (call->class->system &&
2572		    strcmp(call->class->system, "syscalls") == 0)
2573			continue;
2574#endif
2575
2576		pr_info("Testing event %s: ", ftrace_event_name(call));
2577
2578		/*
2579		 * If an event is already enabled, someone is using
2580		 * it and the self test should not be on.
2581		 */
2582		if (file->flags & FTRACE_EVENT_FL_ENABLED) {
2583			pr_warn("Enabled event during self test!\n");
2584			WARN_ON_ONCE(1);
2585			continue;
2586		}
2587
2588		ftrace_event_enable_disable(file, 1);
2589		event_test_stuff();
2590		ftrace_event_enable_disable(file, 0);
2591
2592		pr_cont("OK\n");
2593	}
2594
2595	/* Now test at the sub system level */
2596
2597	pr_info("Running tests on trace event systems:\n");
2598
2599	list_for_each_entry(dir, &tr->systems, list) {
2600
2601		system = dir->subsystem;
2602
2603		/* the ftrace system is special, skip it */
2604		if (strcmp(system->name, "ftrace") == 0)
2605			continue;
2606
2607		pr_info("Testing event system %s: ", system->name);
2608
2609		ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1);
2610		if (WARN_ON_ONCE(ret)) {
2611			pr_warn("error enabling system %s\n",
2612				system->name);
2613			continue;
2614		}
2615
2616		event_test_stuff();
2617
2618		ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0);
2619		if (WARN_ON_ONCE(ret)) {
2620			pr_warn("error disabling system %s\n",
2621				system->name);
2622			continue;
2623		}
2624
2625		pr_cont("OK\n");
2626	}
2627
2628	/* Test with all events enabled */
2629
2630	pr_info("Running tests on all trace events:\n");
2631	pr_info("Testing all events: ");
2632
2633	ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1);
2634	if (WARN_ON_ONCE(ret)) {
2635		pr_warn("error enabling all events\n");
2636		return;
2637	}
2638
2639	event_test_stuff();
2640
2641	/* reset sysname */
2642	ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
2643	if (WARN_ON_ONCE(ret)) {
2644		pr_warn("error disabling all events\n");
2645		return;
2646	}
2647
2648	pr_cont("OK\n");
2649}
2650
2651#ifdef CONFIG_FUNCTION_TRACER
2652
2653static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
2654
2655static void
2656function_test_events_call(unsigned long ip, unsigned long parent_ip,
2657			  struct ftrace_ops *op, struct pt_regs *pt_regs)
2658{
2659	struct ring_buffer_event *event;
2660	struct ring_buffer *buffer;
2661	struct ftrace_entry *entry;
2662	unsigned long flags;
2663	long disabled;
2664	int cpu;
2665	int pc;
2666
2667	pc = preempt_count();
2668	preempt_disable_notrace();
2669	cpu = raw_smp_processor_id();
2670	disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
2671
2672	if (disabled != 1)
2673		goto out;
2674
2675	local_save_flags(flags);
2676
2677	event = trace_current_buffer_lock_reserve(&buffer,
2678						  TRACE_FN, sizeof(*entry),
2679						  flags, pc);
2680	if (!event)
2681		goto out;
2682	entry	= ring_buffer_event_data(event);
2683	entry->ip			= ip;
2684	entry->parent_ip		= parent_ip;
2685
2686	trace_buffer_unlock_commit(buffer, event, flags, pc);
2687
2688 out:
2689	atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
2690	preempt_enable_notrace();
2691}
2692
2693static struct ftrace_ops trace_ops __initdata  =
2694{
2695	.func = function_test_events_call,
2696	.flags = FTRACE_OPS_FL_RECURSION_SAFE,
2697};
2698
2699static __init void event_trace_self_test_with_function(void)
2700{
2701	int ret;
2702	ret = register_ftrace_function(&trace_ops);
2703	if (WARN_ON(ret < 0)) {
2704		pr_info("Failed to enable function tracer for event tests\n");
2705		return;
2706	}
2707	pr_info("Running tests again, along with the function tracer\n");
2708	event_trace_self_tests();
2709	unregister_ftrace_function(&trace_ops);
2710}
2711#else
2712static __init void event_trace_self_test_with_function(void)
2713{
2714}
2715#endif
2716
2717static __init int event_trace_self_tests_init(void)
2718{
2719	if (!tracing_selftest_disabled) {
2720		event_trace_self_tests();
2721		event_trace_self_test_with_function();
2722	}
2723
2724	return 0;
2725}
2726
2727late_initcall(event_trace_self_tests_init);
2728
2729#endif
2730