xref: /linux/kernel/trace/trace_events.c (revision 4a65896f94fa82370041823837cd75aac1186b54)
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/tracefs.h>
17 #include <linux/uaccess.h>
18 #include <linux/bsearch.h>
19 #include <linux/module.h>
20 #include <linux/ctype.h>
21 #include <linux/sort.h>
22 #include <linux/slab.h>
23 #include <linux/delay.h>
24 
25 #include <trace/events/sched.h>
26 
27 #include <asm/setup.h>
28 
29 #include "trace_output.h"
30 
31 #undef TRACE_SYSTEM
32 #define TRACE_SYSTEM "TRACE_SYSTEM"
33 
34 DEFINE_MUTEX(event_mutex);
35 
36 LIST_HEAD(ftrace_events);
37 static LIST_HEAD(ftrace_generic_fields);
38 static LIST_HEAD(ftrace_common_fields);
39 
40 #define GFP_TRACE (GFP_KERNEL | __GFP_ZERO)
41 
42 static struct kmem_cache *field_cachep;
43 static struct kmem_cache *file_cachep;
44 
45 static inline int system_refcount(struct event_subsystem *system)
46 {
47 	return system->ref_count;
48 }
49 
50 static int system_refcount_inc(struct event_subsystem *system)
51 {
52 	return system->ref_count++;
53 }
54 
55 static int system_refcount_dec(struct event_subsystem *system)
56 {
57 	return --system->ref_count;
58 }
59 
60 /* Double loops, do not use break, only goto's work */
61 #define do_for_each_event_file(tr, file)			\
62 	list_for_each_entry(tr, &ftrace_trace_arrays, list) {	\
63 		list_for_each_entry(file, &tr->events, list)
64 
65 #define do_for_each_event_file_safe(tr, file)			\
66 	list_for_each_entry(tr, &ftrace_trace_arrays, list) {	\
67 		struct trace_event_file *___n;				\
68 		list_for_each_entry_safe(file, ___n, &tr->events, list)
69 
70 #define while_for_each_event_file()		\
71 	}
72 
73 static struct list_head *
74 trace_get_fields(struct trace_event_call *event_call)
75 {
76 	if (!event_call->class->get_fields)
77 		return &event_call->class->fields;
78 	return event_call->class->get_fields(event_call);
79 }
80 
81 static struct ftrace_event_field *
82 __find_event_field(struct list_head *head, char *name)
83 {
84 	struct ftrace_event_field *field;
85 
86 	list_for_each_entry(field, head, link) {
87 		if (!strcmp(field->name, name))
88 			return field;
89 	}
90 
91 	return NULL;
92 }
93 
94 struct ftrace_event_field *
95 trace_find_event_field(struct trace_event_call *call, char *name)
96 {
97 	struct ftrace_event_field *field;
98 	struct list_head *head;
99 
100 	head = trace_get_fields(call);
101 	field = __find_event_field(head, name);
102 	if (field)
103 		return field;
104 
105 	field = __find_event_field(&ftrace_generic_fields, name);
106 	if (field)
107 		return field;
108 
109 	return __find_event_field(&ftrace_common_fields, name);
110 }
111 
112 static int __trace_define_field(struct list_head *head, const char *type,
113 				const char *name, int offset, int size,
114 				int is_signed, int filter_type)
115 {
116 	struct ftrace_event_field *field;
117 
118 	field = kmem_cache_alloc(field_cachep, GFP_TRACE);
119 	if (!field)
120 		return -ENOMEM;
121 
122 	field->name = name;
123 	field->type = type;
124 
125 	if (filter_type == FILTER_OTHER)
126 		field->filter_type = filter_assign_type(type);
127 	else
128 		field->filter_type = filter_type;
129 
130 	field->offset = offset;
131 	field->size = size;
132 	field->is_signed = is_signed;
133 
134 	list_add(&field->link, head);
135 
136 	return 0;
137 }
138 
139 int trace_define_field(struct trace_event_call *call, const char *type,
140 		       const char *name, int offset, int size, int is_signed,
141 		       int filter_type)
142 {
143 	struct list_head *head;
144 
145 	if (WARN_ON(!call->class))
146 		return 0;
147 
148 	head = trace_get_fields(call);
149 	return __trace_define_field(head, type, name, offset, size,
150 				    is_signed, filter_type);
151 }
152 EXPORT_SYMBOL_GPL(trace_define_field);
153 
154 #define __generic_field(type, item, filter_type)			\
155 	ret = __trace_define_field(&ftrace_generic_fields, #type,	\
156 				   #item, 0, 0, is_signed_type(type),	\
157 				   filter_type);			\
158 	if (ret)							\
159 		return ret;
160 
161 #define __common_field(type, item)					\
162 	ret = __trace_define_field(&ftrace_common_fields, #type,	\
163 				   "common_" #item,			\
164 				   offsetof(typeof(ent), item),		\
165 				   sizeof(ent.item),			\
166 				   is_signed_type(type), FILTER_OTHER);	\
167 	if (ret)							\
168 		return ret;
169 
170 static int trace_define_generic_fields(void)
171 {
172 	int ret;
173 
174 	__generic_field(int, CPU, FILTER_CPU);
175 	__generic_field(int, cpu, FILTER_CPU);
176 	__generic_field(char *, COMM, FILTER_COMM);
177 	__generic_field(char *, comm, FILTER_COMM);
178 
179 	return ret;
180 }
181 
182 static int trace_define_common_fields(void)
183 {
184 	int ret;
185 	struct trace_entry ent;
186 
187 	__common_field(unsigned short, type);
188 	__common_field(unsigned char, flags);
189 	__common_field(unsigned char, preempt_count);
190 	__common_field(int, pid);
191 
192 	return ret;
193 }
194 
195 static void trace_destroy_fields(struct trace_event_call *call)
196 {
197 	struct ftrace_event_field *field, *next;
198 	struct list_head *head;
199 
200 	head = trace_get_fields(call);
201 	list_for_each_entry_safe(field, next, head, link) {
202 		list_del(&field->link);
203 		kmem_cache_free(field_cachep, field);
204 	}
205 }
206 
207 /*
208  * run-time version of trace_event_get_offsets_<call>() that returns the last
209  * accessible offset of trace fields excluding __dynamic_array bytes
210  */
211 int trace_event_get_offsets(struct trace_event_call *call)
212 {
213 	struct ftrace_event_field *tail;
214 	struct list_head *head;
215 
216 	head = trace_get_fields(call);
217 	/*
218 	 * head->next points to the last field with the largest offset,
219 	 * since it was added last by trace_define_field()
220 	 */
221 	tail = list_first_entry(head, struct ftrace_event_field, link);
222 	return tail->offset + tail->size;
223 }
224 
225 int trace_event_raw_init(struct trace_event_call *call)
226 {
227 	int id;
228 
229 	id = register_trace_event(&call->event);
230 	if (!id)
231 		return -ENODEV;
232 
233 	return 0;
234 }
235 EXPORT_SYMBOL_GPL(trace_event_raw_init);
236 
237 bool trace_event_ignore_this_pid(struct trace_event_file *trace_file)
238 {
239 	struct trace_array *tr = trace_file->tr;
240 	struct trace_array_cpu *data;
241 	struct trace_pid_list *pid_list;
242 
243 	pid_list = rcu_dereference_sched(tr->filtered_pids);
244 	if (!pid_list)
245 		return false;
246 
247 	data = this_cpu_ptr(tr->trace_buffer.data);
248 
249 	return data->ignore_pid;
250 }
251 EXPORT_SYMBOL_GPL(trace_event_ignore_this_pid);
252 
253 void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer,
254 				 struct trace_event_file *trace_file,
255 				 unsigned long len)
256 {
257 	struct trace_event_call *event_call = trace_file->event_call;
258 
259 	if ((trace_file->flags & EVENT_FILE_FL_PID_FILTER) &&
260 	    trace_event_ignore_this_pid(trace_file))
261 		return NULL;
262 
263 	local_save_flags(fbuffer->flags);
264 	fbuffer->pc = preempt_count();
265 	fbuffer->trace_file = trace_file;
266 
267 	fbuffer->event =
268 		trace_event_buffer_lock_reserve(&fbuffer->buffer, trace_file,
269 						event_call->event.type, len,
270 						fbuffer->flags, fbuffer->pc);
271 	if (!fbuffer->event)
272 		return NULL;
273 
274 	fbuffer->entry = ring_buffer_event_data(fbuffer->event);
275 	return fbuffer->entry;
276 }
277 EXPORT_SYMBOL_GPL(trace_event_buffer_reserve);
278 
279 static DEFINE_SPINLOCK(tracepoint_iter_lock);
280 
281 static void output_printk(struct trace_event_buffer *fbuffer)
282 {
283 	struct trace_event_call *event_call;
284 	struct trace_event *event;
285 	unsigned long flags;
286 	struct trace_iterator *iter = tracepoint_print_iter;
287 
288 	if (!iter)
289 		return;
290 
291 	event_call = fbuffer->trace_file->event_call;
292 	if (!event_call || !event_call->event.funcs ||
293 	    !event_call->event.funcs->trace)
294 		return;
295 
296 	event = &fbuffer->trace_file->event_call->event;
297 
298 	spin_lock_irqsave(&tracepoint_iter_lock, flags);
299 	trace_seq_init(&iter->seq);
300 	iter->ent = fbuffer->entry;
301 	event_call->event.funcs->trace(iter, 0, event);
302 	trace_seq_putc(&iter->seq, 0);
303 	printk("%s", iter->seq.buffer);
304 
305 	spin_unlock_irqrestore(&tracepoint_iter_lock, flags);
306 }
307 
308 void trace_event_buffer_commit(struct trace_event_buffer *fbuffer)
309 {
310 	if (tracepoint_printk)
311 		output_printk(fbuffer);
312 
313 	event_trigger_unlock_commit(fbuffer->trace_file, fbuffer->buffer,
314 				    fbuffer->event, fbuffer->entry,
315 				    fbuffer->flags, fbuffer->pc);
316 }
317 EXPORT_SYMBOL_GPL(trace_event_buffer_commit);
318 
319 int trace_event_reg(struct trace_event_call *call,
320 		    enum trace_reg type, void *data)
321 {
322 	struct trace_event_file *file = data;
323 
324 	WARN_ON(!(call->flags & TRACE_EVENT_FL_TRACEPOINT));
325 	switch (type) {
326 	case TRACE_REG_REGISTER:
327 		return tracepoint_probe_register(call->tp,
328 						 call->class->probe,
329 						 file);
330 	case TRACE_REG_UNREGISTER:
331 		tracepoint_probe_unregister(call->tp,
332 					    call->class->probe,
333 					    file);
334 		return 0;
335 
336 #ifdef CONFIG_PERF_EVENTS
337 	case TRACE_REG_PERF_REGISTER:
338 		return tracepoint_probe_register(call->tp,
339 						 call->class->perf_probe,
340 						 call);
341 	case TRACE_REG_PERF_UNREGISTER:
342 		tracepoint_probe_unregister(call->tp,
343 					    call->class->perf_probe,
344 					    call);
345 		return 0;
346 	case TRACE_REG_PERF_OPEN:
347 	case TRACE_REG_PERF_CLOSE:
348 	case TRACE_REG_PERF_ADD:
349 	case TRACE_REG_PERF_DEL:
350 		return 0;
351 #endif
352 	}
353 	return 0;
354 }
355 EXPORT_SYMBOL_GPL(trace_event_reg);
356 
357 void trace_event_enable_cmd_record(bool enable)
358 {
359 	struct trace_event_file *file;
360 	struct trace_array *tr;
361 
362 	mutex_lock(&event_mutex);
363 	do_for_each_event_file(tr, file) {
364 
365 		if (!(file->flags & EVENT_FILE_FL_ENABLED))
366 			continue;
367 
368 		if (enable) {
369 			tracing_start_cmdline_record();
370 			set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
371 		} else {
372 			tracing_stop_cmdline_record();
373 			clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
374 		}
375 	} while_for_each_event_file();
376 	mutex_unlock(&event_mutex);
377 }
378 
379 static int __ftrace_event_enable_disable(struct trace_event_file *file,
380 					 int enable, int soft_disable)
381 {
382 	struct trace_event_call *call = file->event_call;
383 	struct trace_array *tr = file->tr;
384 	int ret = 0;
385 	int disable;
386 
387 	switch (enable) {
388 	case 0:
389 		/*
390 		 * When soft_disable is set and enable is cleared, the sm_ref
391 		 * reference counter is decremented. If it reaches 0, we want
392 		 * to clear the SOFT_DISABLED flag but leave the event in the
393 		 * state that it was. That is, if the event was enabled and
394 		 * SOFT_DISABLED isn't set, then do nothing. But if SOFT_DISABLED
395 		 * is set we do not want the event to be enabled before we
396 		 * clear the bit.
397 		 *
398 		 * When soft_disable is not set but the SOFT_MODE flag is,
399 		 * we do nothing. Do not disable the tracepoint, otherwise
400 		 * "soft enable"s (clearing the SOFT_DISABLED bit) wont work.
401 		 */
402 		if (soft_disable) {
403 			if (atomic_dec_return(&file->sm_ref) > 0)
404 				break;
405 			disable = file->flags & EVENT_FILE_FL_SOFT_DISABLED;
406 			clear_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
407 		} else
408 			disable = !(file->flags & EVENT_FILE_FL_SOFT_MODE);
409 
410 		if (disable && (file->flags & EVENT_FILE_FL_ENABLED)) {
411 			clear_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
412 			if (file->flags & EVENT_FILE_FL_RECORDED_CMD) {
413 				tracing_stop_cmdline_record();
414 				clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
415 			}
416 			call->class->reg(call, TRACE_REG_UNREGISTER, file);
417 		}
418 		/* If in SOFT_MODE, just set the SOFT_DISABLE_BIT, else clear it */
419 		if (file->flags & EVENT_FILE_FL_SOFT_MODE)
420 			set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
421 		else
422 			clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
423 		break;
424 	case 1:
425 		/*
426 		 * When soft_disable is set and enable is set, we want to
427 		 * register the tracepoint for the event, but leave the event
428 		 * as is. That means, if the event was already enabled, we do
429 		 * nothing (but set SOFT_MODE). If the event is disabled, we
430 		 * set SOFT_DISABLED before enabling the event tracepoint, so
431 		 * it still seems to be disabled.
432 		 */
433 		if (!soft_disable)
434 			clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
435 		else {
436 			if (atomic_inc_return(&file->sm_ref) > 1)
437 				break;
438 			set_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
439 		}
440 
441 		if (!(file->flags & EVENT_FILE_FL_ENABLED)) {
442 
443 			/* Keep the event disabled, when going to SOFT_MODE. */
444 			if (soft_disable)
445 				set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
446 
447 			if (tr->trace_flags & TRACE_ITER_RECORD_CMD) {
448 				tracing_start_cmdline_record();
449 				set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
450 			}
451 			ret = call->class->reg(call, TRACE_REG_REGISTER, file);
452 			if (ret) {
453 				tracing_stop_cmdline_record();
454 				pr_info("event trace: Could not enable event "
455 					"%s\n", trace_event_name(call));
456 				break;
457 			}
458 			set_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
459 
460 			/* WAS_ENABLED gets set but never cleared. */
461 			call->flags |= TRACE_EVENT_FL_WAS_ENABLED;
462 		}
463 		break;
464 	}
465 
466 	return ret;
467 }
468 
469 int trace_event_enable_disable(struct trace_event_file *file,
470 			       int enable, int soft_disable)
471 {
472 	return __ftrace_event_enable_disable(file, enable, soft_disable);
473 }
474 
475 static int ftrace_event_enable_disable(struct trace_event_file *file,
476 				       int enable)
477 {
478 	return __ftrace_event_enable_disable(file, enable, 0);
479 }
480 
481 static void ftrace_clear_events(struct trace_array *tr)
482 {
483 	struct trace_event_file *file;
484 
485 	mutex_lock(&event_mutex);
486 	list_for_each_entry(file, &tr->events, list) {
487 		ftrace_event_enable_disable(file, 0);
488 	}
489 	mutex_unlock(&event_mutex);
490 }
491 
492 static int cmp_pid(const void *key, const void *elt)
493 {
494 	const pid_t *search_pid = key;
495 	const pid_t *pid = elt;
496 
497 	if (*search_pid == *pid)
498 		return 0;
499 	if (*search_pid < *pid)
500 		return -1;
501 	return 1;
502 }
503 
504 static bool
505 check_ignore_pid(struct trace_pid_list *filtered_pids, struct task_struct *task)
506 {
507 	pid_t search_pid;
508 	pid_t *pid;
509 
510 	/*
511 	 * Return false, because if filtered_pids does not exist,
512 	 * all pids are good to trace.
513 	 */
514 	if (!filtered_pids)
515 		return false;
516 
517 	search_pid = task->pid;
518 
519 	pid = bsearch(&search_pid, filtered_pids->pids,
520 		      filtered_pids->nr_pids, sizeof(pid_t),
521 		      cmp_pid);
522 	if (!pid)
523 		return true;
524 
525 	return false;
526 }
527 
528 static void
529 event_filter_pid_sched_switch_probe_pre(void *data, bool preempt,
530 		    struct task_struct *prev, struct task_struct *next)
531 {
532 	struct trace_array *tr = data;
533 	struct trace_pid_list *pid_list;
534 
535 	pid_list = rcu_dereference_sched(tr->filtered_pids);
536 
537 	this_cpu_write(tr->trace_buffer.data->ignore_pid,
538 		       check_ignore_pid(pid_list, prev) &&
539 		       check_ignore_pid(pid_list, next));
540 }
541 
542 static void
543 event_filter_pid_sched_switch_probe_post(void *data, bool preempt,
544 		    struct task_struct *prev, struct task_struct *next)
545 {
546 	struct trace_array *tr = data;
547 	struct trace_pid_list *pid_list;
548 
549 	pid_list = rcu_dereference_sched(tr->filtered_pids);
550 
551 	this_cpu_write(tr->trace_buffer.data->ignore_pid,
552 		       check_ignore_pid(pid_list, next));
553 }
554 
555 static void
556 event_filter_pid_sched_wakeup_probe_pre(void *data, struct task_struct *task)
557 {
558 	struct trace_array *tr = data;
559 	struct trace_pid_list *pid_list;
560 
561 	/* Nothing to do if we are already tracing */
562 	if (!this_cpu_read(tr->trace_buffer.data->ignore_pid))
563 		return;
564 
565 	pid_list = rcu_dereference_sched(tr->filtered_pids);
566 
567 	this_cpu_write(tr->trace_buffer.data->ignore_pid,
568 		       check_ignore_pid(pid_list, task));
569 }
570 
571 static void
572 event_filter_pid_sched_wakeup_probe_post(void *data, struct task_struct *task)
573 {
574 	struct trace_array *tr = data;
575 	struct trace_pid_list *pid_list;
576 
577 	/* Nothing to do if we are not tracing */
578 	if (this_cpu_read(tr->trace_buffer.data->ignore_pid))
579 		return;
580 
581 	pid_list = rcu_dereference_sched(tr->filtered_pids);
582 
583 	/* Set tracing if current is enabled */
584 	this_cpu_write(tr->trace_buffer.data->ignore_pid,
585 		       check_ignore_pid(pid_list, current));
586 }
587 
588 static void __ftrace_clear_event_pids(struct trace_array *tr)
589 {
590 	struct trace_pid_list *pid_list;
591 	struct trace_event_file *file;
592 	int cpu;
593 
594 	pid_list = rcu_dereference_protected(tr->filtered_pids,
595 					     lockdep_is_held(&event_mutex));
596 	if (!pid_list)
597 		return;
598 
599 	unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_pre, tr);
600 	unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_post, tr);
601 
602 	unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre, tr);
603 	unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_post, tr);
604 
605 	unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre, tr);
606 	unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post, tr);
607 
608 	unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_pre, tr);
609 	unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_post, tr);
610 
611 	list_for_each_entry(file, &tr->events, list) {
612 		clear_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
613 	}
614 
615 	for_each_possible_cpu(cpu)
616 		per_cpu_ptr(tr->trace_buffer.data, cpu)->ignore_pid = false;
617 
618 	rcu_assign_pointer(tr->filtered_pids, NULL);
619 
620 	/* Wait till all users are no longer using pid filtering */
621 	synchronize_sched();
622 
623 	free_pages((unsigned long)pid_list->pids, pid_list->order);
624 	kfree(pid_list);
625 }
626 
627 static void ftrace_clear_event_pids(struct trace_array *tr)
628 {
629 	mutex_lock(&event_mutex);
630 	__ftrace_clear_event_pids(tr);
631 	mutex_unlock(&event_mutex);
632 }
633 
634 static void __put_system(struct event_subsystem *system)
635 {
636 	struct event_filter *filter = system->filter;
637 
638 	WARN_ON_ONCE(system_refcount(system) == 0);
639 	if (system_refcount_dec(system))
640 		return;
641 
642 	list_del(&system->list);
643 
644 	if (filter) {
645 		kfree(filter->filter_string);
646 		kfree(filter);
647 	}
648 	kfree_const(system->name);
649 	kfree(system);
650 }
651 
652 static void __get_system(struct event_subsystem *system)
653 {
654 	WARN_ON_ONCE(system_refcount(system) == 0);
655 	system_refcount_inc(system);
656 }
657 
658 static void __get_system_dir(struct trace_subsystem_dir *dir)
659 {
660 	WARN_ON_ONCE(dir->ref_count == 0);
661 	dir->ref_count++;
662 	__get_system(dir->subsystem);
663 }
664 
665 static void __put_system_dir(struct trace_subsystem_dir *dir)
666 {
667 	WARN_ON_ONCE(dir->ref_count == 0);
668 	/* If the subsystem is about to be freed, the dir must be too */
669 	WARN_ON_ONCE(system_refcount(dir->subsystem) == 1 && dir->ref_count != 1);
670 
671 	__put_system(dir->subsystem);
672 	if (!--dir->ref_count)
673 		kfree(dir);
674 }
675 
676 static void put_system(struct trace_subsystem_dir *dir)
677 {
678 	mutex_lock(&event_mutex);
679 	__put_system_dir(dir);
680 	mutex_unlock(&event_mutex);
681 }
682 
683 static void remove_subsystem(struct trace_subsystem_dir *dir)
684 {
685 	if (!dir)
686 		return;
687 
688 	if (!--dir->nr_events) {
689 		tracefs_remove_recursive(dir->entry);
690 		list_del(&dir->list);
691 		__put_system_dir(dir);
692 	}
693 }
694 
695 static void remove_event_file_dir(struct trace_event_file *file)
696 {
697 	struct dentry *dir = file->dir;
698 	struct dentry *child;
699 
700 	if (dir) {
701 		spin_lock(&dir->d_lock);	/* probably unneeded */
702 		list_for_each_entry(child, &dir->d_subdirs, d_child) {
703 			if (d_really_is_positive(child))	/* probably unneeded */
704 				d_inode(child)->i_private = NULL;
705 		}
706 		spin_unlock(&dir->d_lock);
707 
708 		tracefs_remove_recursive(dir);
709 	}
710 
711 	list_del(&file->list);
712 	remove_subsystem(file->system);
713 	free_event_filter(file->filter);
714 	kmem_cache_free(file_cachep, file);
715 }
716 
717 /*
718  * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
719  */
720 static int
721 __ftrace_set_clr_event_nolock(struct trace_array *tr, const char *match,
722 			      const char *sub, const char *event, int set)
723 {
724 	struct trace_event_file *file;
725 	struct trace_event_call *call;
726 	const char *name;
727 	int ret = -EINVAL;
728 
729 	list_for_each_entry(file, &tr->events, list) {
730 
731 		call = file->event_call;
732 		name = trace_event_name(call);
733 
734 		if (!name || !call->class || !call->class->reg)
735 			continue;
736 
737 		if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
738 			continue;
739 
740 		if (match &&
741 		    strcmp(match, name) != 0 &&
742 		    strcmp(match, call->class->system) != 0)
743 			continue;
744 
745 		if (sub && strcmp(sub, call->class->system) != 0)
746 			continue;
747 
748 		if (event && strcmp(event, name) != 0)
749 			continue;
750 
751 		ftrace_event_enable_disable(file, set);
752 
753 		ret = 0;
754 	}
755 
756 	return ret;
757 }
758 
759 static int __ftrace_set_clr_event(struct trace_array *tr, const char *match,
760 				  const char *sub, const char *event, int set)
761 {
762 	int ret;
763 
764 	mutex_lock(&event_mutex);
765 	ret = __ftrace_set_clr_event_nolock(tr, match, sub, event, set);
766 	mutex_unlock(&event_mutex);
767 
768 	return ret;
769 }
770 
771 static int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set)
772 {
773 	char *event = NULL, *sub = NULL, *match;
774 	int ret;
775 
776 	/*
777 	 * The buf format can be <subsystem>:<event-name>
778 	 *  *:<event-name> means any event by that name.
779 	 *  :<event-name> is the same.
780 	 *
781 	 *  <subsystem>:* means all events in that subsystem
782 	 *  <subsystem>: means the same.
783 	 *
784 	 *  <name> (no ':') means all events in a subsystem with
785 	 *  the name <name> or any event that matches <name>
786 	 */
787 
788 	match = strsep(&buf, ":");
789 	if (buf) {
790 		sub = match;
791 		event = buf;
792 		match = NULL;
793 
794 		if (!strlen(sub) || strcmp(sub, "*") == 0)
795 			sub = NULL;
796 		if (!strlen(event) || strcmp(event, "*") == 0)
797 			event = NULL;
798 	}
799 
800 	ret = __ftrace_set_clr_event(tr, match, sub, event, set);
801 
802 	/* Put back the colon to allow this to be called again */
803 	if (buf)
804 		*(buf - 1) = ':';
805 
806 	return ret;
807 }
808 
809 /**
810  * trace_set_clr_event - enable or disable an event
811  * @system: system name to match (NULL for any system)
812  * @event: event name to match (NULL for all events, within system)
813  * @set: 1 to enable, 0 to disable
814  *
815  * This is a way for other parts of the kernel to enable or disable
816  * event recording.
817  *
818  * Returns 0 on success, -EINVAL if the parameters do not match any
819  * registered events.
820  */
821 int trace_set_clr_event(const char *system, const char *event, int set)
822 {
823 	struct trace_array *tr = top_trace_array();
824 
825 	if (!tr)
826 		return -ENODEV;
827 
828 	return __ftrace_set_clr_event(tr, NULL, system, event, set);
829 }
830 EXPORT_SYMBOL_GPL(trace_set_clr_event);
831 
832 /* 128 should be much more than enough */
833 #define EVENT_BUF_SIZE		127
834 
835 static ssize_t
836 ftrace_event_write(struct file *file, const char __user *ubuf,
837 		   size_t cnt, loff_t *ppos)
838 {
839 	struct trace_parser parser;
840 	struct seq_file *m = file->private_data;
841 	struct trace_array *tr = m->private;
842 	ssize_t read, ret;
843 
844 	if (!cnt)
845 		return 0;
846 
847 	ret = tracing_update_buffers();
848 	if (ret < 0)
849 		return ret;
850 
851 	if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
852 		return -ENOMEM;
853 
854 	read = trace_get_user(&parser, ubuf, cnt, ppos);
855 
856 	if (read >= 0 && trace_parser_loaded((&parser))) {
857 		int set = 1;
858 
859 		if (*parser.buffer == '!')
860 			set = 0;
861 
862 		parser.buffer[parser.idx] = 0;
863 
864 		ret = ftrace_set_clr_event(tr, parser.buffer + !set, set);
865 		if (ret)
866 			goto out_put;
867 	}
868 
869 	ret = read;
870 
871  out_put:
872 	trace_parser_put(&parser);
873 
874 	return ret;
875 }
876 
877 static void *
878 t_next(struct seq_file *m, void *v, loff_t *pos)
879 {
880 	struct trace_event_file *file = v;
881 	struct trace_event_call *call;
882 	struct trace_array *tr = m->private;
883 
884 	(*pos)++;
885 
886 	list_for_each_entry_continue(file, &tr->events, list) {
887 		call = file->event_call;
888 		/*
889 		 * The ftrace subsystem is for showing formats only.
890 		 * They can not be enabled or disabled via the event files.
891 		 */
892 		if (call->class && call->class->reg &&
893 		    !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
894 			return file;
895 	}
896 
897 	return NULL;
898 }
899 
900 static void *t_start(struct seq_file *m, loff_t *pos)
901 {
902 	struct trace_event_file *file;
903 	struct trace_array *tr = m->private;
904 	loff_t l;
905 
906 	mutex_lock(&event_mutex);
907 
908 	file = list_entry(&tr->events, struct trace_event_file, list);
909 	for (l = 0; l <= *pos; ) {
910 		file = t_next(m, file, &l);
911 		if (!file)
912 			break;
913 	}
914 	return file;
915 }
916 
917 static void *
918 s_next(struct seq_file *m, void *v, loff_t *pos)
919 {
920 	struct trace_event_file *file = v;
921 	struct trace_array *tr = m->private;
922 
923 	(*pos)++;
924 
925 	list_for_each_entry_continue(file, &tr->events, list) {
926 		if (file->flags & EVENT_FILE_FL_ENABLED)
927 			return file;
928 	}
929 
930 	return NULL;
931 }
932 
933 static void *s_start(struct seq_file *m, loff_t *pos)
934 {
935 	struct trace_event_file *file;
936 	struct trace_array *tr = m->private;
937 	loff_t l;
938 
939 	mutex_lock(&event_mutex);
940 
941 	file = list_entry(&tr->events, struct trace_event_file, list);
942 	for (l = 0; l <= *pos; ) {
943 		file = s_next(m, file, &l);
944 		if (!file)
945 			break;
946 	}
947 	return file;
948 }
949 
950 static int t_show(struct seq_file *m, void *v)
951 {
952 	struct trace_event_file *file = v;
953 	struct trace_event_call *call = file->event_call;
954 
955 	if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
956 		seq_printf(m, "%s:", call->class->system);
957 	seq_printf(m, "%s\n", trace_event_name(call));
958 
959 	return 0;
960 }
961 
962 static void t_stop(struct seq_file *m, void *p)
963 {
964 	mutex_unlock(&event_mutex);
965 }
966 
967 static void *p_start(struct seq_file *m, loff_t *pos)
968 	__acquires(RCU)
969 {
970 	struct trace_pid_list *pid_list;
971 	struct trace_array *tr = m->private;
972 
973 	/*
974 	 * Grab the mutex, to keep calls to p_next() having the same
975 	 * tr->filtered_pids as p_start() has.
976 	 * If we just passed the tr->filtered_pids around, then RCU would
977 	 * have been enough, but doing that makes things more complex.
978 	 */
979 	mutex_lock(&event_mutex);
980 	rcu_read_lock_sched();
981 
982 	pid_list = rcu_dereference_sched(tr->filtered_pids);
983 
984 	if (!pid_list || *pos >= pid_list->nr_pids)
985 		return NULL;
986 
987 	return (void *)&pid_list->pids[*pos];
988 }
989 
990 static void p_stop(struct seq_file *m, void *p)
991 	__releases(RCU)
992 {
993 	rcu_read_unlock_sched();
994 	mutex_unlock(&event_mutex);
995 }
996 
997 static void *
998 p_next(struct seq_file *m, void *v, loff_t *pos)
999 {
1000 	struct trace_array *tr = m->private;
1001 	struct trace_pid_list *pid_list = rcu_dereference_sched(tr->filtered_pids);
1002 
1003 	(*pos)++;
1004 
1005 	if (*pos >= pid_list->nr_pids)
1006 		return NULL;
1007 
1008 	return (void *)&pid_list->pids[*pos];
1009 }
1010 
1011 static int p_show(struct seq_file *m, void *v)
1012 {
1013 	pid_t *pid = v;
1014 
1015 	seq_printf(m, "%d\n", *pid);
1016 	return 0;
1017 }
1018 
1019 static ssize_t
1020 event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1021 		  loff_t *ppos)
1022 {
1023 	struct trace_event_file *file;
1024 	unsigned long flags;
1025 	char buf[4] = "0";
1026 
1027 	mutex_lock(&event_mutex);
1028 	file = event_file_data(filp);
1029 	if (likely(file))
1030 		flags = file->flags;
1031 	mutex_unlock(&event_mutex);
1032 
1033 	if (!file)
1034 		return -ENODEV;
1035 
1036 	if (flags & EVENT_FILE_FL_ENABLED &&
1037 	    !(flags & EVENT_FILE_FL_SOFT_DISABLED))
1038 		strcpy(buf, "1");
1039 
1040 	if (flags & EVENT_FILE_FL_SOFT_DISABLED ||
1041 	    flags & EVENT_FILE_FL_SOFT_MODE)
1042 		strcat(buf, "*");
1043 
1044 	strcat(buf, "\n");
1045 
1046 	return simple_read_from_buffer(ubuf, cnt, ppos, buf, strlen(buf));
1047 }
1048 
1049 static ssize_t
1050 event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1051 		   loff_t *ppos)
1052 {
1053 	struct trace_event_file *file;
1054 	unsigned long val;
1055 	int ret;
1056 
1057 	ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1058 	if (ret)
1059 		return ret;
1060 
1061 	ret = tracing_update_buffers();
1062 	if (ret < 0)
1063 		return ret;
1064 
1065 	switch (val) {
1066 	case 0:
1067 	case 1:
1068 		ret = -ENODEV;
1069 		mutex_lock(&event_mutex);
1070 		file = event_file_data(filp);
1071 		if (likely(file))
1072 			ret = ftrace_event_enable_disable(file, val);
1073 		mutex_unlock(&event_mutex);
1074 		break;
1075 
1076 	default:
1077 		return -EINVAL;
1078 	}
1079 
1080 	*ppos += cnt;
1081 
1082 	return ret ? ret : cnt;
1083 }
1084 
1085 static ssize_t
1086 system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1087 		   loff_t *ppos)
1088 {
1089 	const char set_to_char[4] = { '?', '0', '1', 'X' };
1090 	struct trace_subsystem_dir *dir = filp->private_data;
1091 	struct event_subsystem *system = dir->subsystem;
1092 	struct trace_event_call *call;
1093 	struct trace_event_file *file;
1094 	struct trace_array *tr = dir->tr;
1095 	char buf[2];
1096 	int set = 0;
1097 	int ret;
1098 
1099 	mutex_lock(&event_mutex);
1100 	list_for_each_entry(file, &tr->events, list) {
1101 		call = file->event_call;
1102 		if (!trace_event_name(call) || !call->class || !call->class->reg)
1103 			continue;
1104 
1105 		if (system && strcmp(call->class->system, system->name) != 0)
1106 			continue;
1107 
1108 		/*
1109 		 * We need to find out if all the events are set
1110 		 * or if all events or cleared, or if we have
1111 		 * a mixture.
1112 		 */
1113 		set |= (1 << !!(file->flags & EVENT_FILE_FL_ENABLED));
1114 
1115 		/*
1116 		 * If we have a mixture, no need to look further.
1117 		 */
1118 		if (set == 3)
1119 			break;
1120 	}
1121 	mutex_unlock(&event_mutex);
1122 
1123 	buf[0] = set_to_char[set];
1124 	buf[1] = '\n';
1125 
1126 	ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
1127 
1128 	return ret;
1129 }
1130 
1131 static ssize_t
1132 system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1133 		    loff_t *ppos)
1134 {
1135 	struct trace_subsystem_dir *dir = filp->private_data;
1136 	struct event_subsystem *system = dir->subsystem;
1137 	const char *name = NULL;
1138 	unsigned long val;
1139 	ssize_t ret;
1140 
1141 	ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1142 	if (ret)
1143 		return ret;
1144 
1145 	ret = tracing_update_buffers();
1146 	if (ret < 0)
1147 		return ret;
1148 
1149 	if (val != 0 && val != 1)
1150 		return -EINVAL;
1151 
1152 	/*
1153 	 * Opening of "enable" adds a ref count to system,
1154 	 * so the name is safe to use.
1155 	 */
1156 	if (system)
1157 		name = system->name;
1158 
1159 	ret = __ftrace_set_clr_event(dir->tr, NULL, name, NULL, val);
1160 	if (ret)
1161 		goto out;
1162 
1163 	ret = cnt;
1164 
1165 out:
1166 	*ppos += cnt;
1167 
1168 	return ret;
1169 }
1170 
1171 enum {
1172 	FORMAT_HEADER		= 1,
1173 	FORMAT_FIELD_SEPERATOR	= 2,
1174 	FORMAT_PRINTFMT		= 3,
1175 };
1176 
1177 static void *f_next(struct seq_file *m, void *v, loff_t *pos)
1178 {
1179 	struct trace_event_call *call = event_file_data(m->private);
1180 	struct list_head *common_head = &ftrace_common_fields;
1181 	struct list_head *head = trace_get_fields(call);
1182 	struct list_head *node = v;
1183 
1184 	(*pos)++;
1185 
1186 	switch ((unsigned long)v) {
1187 	case FORMAT_HEADER:
1188 		node = common_head;
1189 		break;
1190 
1191 	case FORMAT_FIELD_SEPERATOR:
1192 		node = head;
1193 		break;
1194 
1195 	case FORMAT_PRINTFMT:
1196 		/* all done */
1197 		return NULL;
1198 	}
1199 
1200 	node = node->prev;
1201 	if (node == common_head)
1202 		return (void *)FORMAT_FIELD_SEPERATOR;
1203 	else if (node == head)
1204 		return (void *)FORMAT_PRINTFMT;
1205 	else
1206 		return node;
1207 }
1208 
1209 static int f_show(struct seq_file *m, void *v)
1210 {
1211 	struct trace_event_call *call = event_file_data(m->private);
1212 	struct ftrace_event_field *field;
1213 	const char *array_descriptor;
1214 
1215 	switch ((unsigned long)v) {
1216 	case FORMAT_HEADER:
1217 		seq_printf(m, "name: %s\n", trace_event_name(call));
1218 		seq_printf(m, "ID: %d\n", call->event.type);
1219 		seq_puts(m, "format:\n");
1220 		return 0;
1221 
1222 	case FORMAT_FIELD_SEPERATOR:
1223 		seq_putc(m, '\n');
1224 		return 0;
1225 
1226 	case FORMAT_PRINTFMT:
1227 		seq_printf(m, "\nprint fmt: %s\n",
1228 			   call->print_fmt);
1229 		return 0;
1230 	}
1231 
1232 	field = list_entry(v, struct ftrace_event_field, link);
1233 	/*
1234 	 * Smartly shows the array type(except dynamic array).
1235 	 * Normal:
1236 	 *	field:TYPE VAR
1237 	 * If TYPE := TYPE[LEN], it is shown:
1238 	 *	field:TYPE VAR[LEN]
1239 	 */
1240 	array_descriptor = strchr(field->type, '[');
1241 
1242 	if (!strncmp(field->type, "__data_loc", 10))
1243 		array_descriptor = NULL;
1244 
1245 	if (!array_descriptor)
1246 		seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1247 			   field->type, field->name, field->offset,
1248 			   field->size, !!field->is_signed);
1249 	else
1250 		seq_printf(m, "\tfield:%.*s %s%s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1251 			   (int)(array_descriptor - field->type),
1252 			   field->type, field->name,
1253 			   array_descriptor, field->offset,
1254 			   field->size, !!field->is_signed);
1255 
1256 	return 0;
1257 }
1258 
1259 static void *f_start(struct seq_file *m, loff_t *pos)
1260 {
1261 	void *p = (void *)FORMAT_HEADER;
1262 	loff_t l = 0;
1263 
1264 	/* ->stop() is called even if ->start() fails */
1265 	mutex_lock(&event_mutex);
1266 	if (!event_file_data(m->private))
1267 		return ERR_PTR(-ENODEV);
1268 
1269 	while (l < *pos && p)
1270 		p = f_next(m, p, &l);
1271 
1272 	return p;
1273 }
1274 
1275 static void f_stop(struct seq_file *m, void *p)
1276 {
1277 	mutex_unlock(&event_mutex);
1278 }
1279 
1280 static const struct seq_operations trace_format_seq_ops = {
1281 	.start		= f_start,
1282 	.next		= f_next,
1283 	.stop		= f_stop,
1284 	.show		= f_show,
1285 };
1286 
1287 static int trace_format_open(struct inode *inode, struct file *file)
1288 {
1289 	struct seq_file *m;
1290 	int ret;
1291 
1292 	ret = seq_open(file, &trace_format_seq_ops);
1293 	if (ret < 0)
1294 		return ret;
1295 
1296 	m = file->private_data;
1297 	m->private = file;
1298 
1299 	return 0;
1300 }
1301 
1302 static ssize_t
1303 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1304 {
1305 	int id = (long)event_file_data(filp);
1306 	char buf[32];
1307 	int len;
1308 
1309 	if (*ppos)
1310 		return 0;
1311 
1312 	if (unlikely(!id))
1313 		return -ENODEV;
1314 
1315 	len = sprintf(buf, "%d\n", id);
1316 
1317 	return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
1318 }
1319 
1320 static ssize_t
1321 event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1322 		  loff_t *ppos)
1323 {
1324 	struct trace_event_file *file;
1325 	struct trace_seq *s;
1326 	int r = -ENODEV;
1327 
1328 	if (*ppos)
1329 		return 0;
1330 
1331 	s = kmalloc(sizeof(*s), GFP_KERNEL);
1332 
1333 	if (!s)
1334 		return -ENOMEM;
1335 
1336 	trace_seq_init(s);
1337 
1338 	mutex_lock(&event_mutex);
1339 	file = event_file_data(filp);
1340 	if (file)
1341 		print_event_filter(file, s);
1342 	mutex_unlock(&event_mutex);
1343 
1344 	if (file)
1345 		r = simple_read_from_buffer(ubuf, cnt, ppos,
1346 					    s->buffer, trace_seq_used(s));
1347 
1348 	kfree(s);
1349 
1350 	return r;
1351 }
1352 
1353 static ssize_t
1354 event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1355 		   loff_t *ppos)
1356 {
1357 	struct trace_event_file *file;
1358 	char *buf;
1359 	int err = -ENODEV;
1360 
1361 	if (cnt >= PAGE_SIZE)
1362 		return -EINVAL;
1363 
1364 	buf = memdup_user_nul(ubuf, cnt);
1365 	if (IS_ERR(buf))
1366 		return PTR_ERR(buf);
1367 
1368 	mutex_lock(&event_mutex);
1369 	file = event_file_data(filp);
1370 	if (file)
1371 		err = apply_event_filter(file, buf);
1372 	mutex_unlock(&event_mutex);
1373 
1374 	kfree(buf);
1375 	if (err < 0)
1376 		return err;
1377 
1378 	*ppos += cnt;
1379 
1380 	return cnt;
1381 }
1382 
1383 static LIST_HEAD(event_subsystems);
1384 
1385 static int subsystem_open(struct inode *inode, struct file *filp)
1386 {
1387 	struct event_subsystem *system = NULL;
1388 	struct trace_subsystem_dir *dir = NULL; /* Initialize for gcc */
1389 	struct trace_array *tr;
1390 	int ret;
1391 
1392 	if (tracing_is_disabled())
1393 		return -ENODEV;
1394 
1395 	/* Make sure the system still exists */
1396 	mutex_lock(&trace_types_lock);
1397 	mutex_lock(&event_mutex);
1398 	list_for_each_entry(tr, &ftrace_trace_arrays, list) {
1399 		list_for_each_entry(dir, &tr->systems, list) {
1400 			if (dir == inode->i_private) {
1401 				/* Don't open systems with no events */
1402 				if (dir->nr_events) {
1403 					__get_system_dir(dir);
1404 					system = dir->subsystem;
1405 				}
1406 				goto exit_loop;
1407 			}
1408 		}
1409 	}
1410  exit_loop:
1411 	mutex_unlock(&event_mutex);
1412 	mutex_unlock(&trace_types_lock);
1413 
1414 	if (!system)
1415 		return -ENODEV;
1416 
1417 	/* Some versions of gcc think dir can be uninitialized here */
1418 	WARN_ON(!dir);
1419 
1420 	/* Still need to increment the ref count of the system */
1421 	if (trace_array_get(tr) < 0) {
1422 		put_system(dir);
1423 		return -ENODEV;
1424 	}
1425 
1426 	ret = tracing_open_generic(inode, filp);
1427 	if (ret < 0) {
1428 		trace_array_put(tr);
1429 		put_system(dir);
1430 	}
1431 
1432 	return ret;
1433 }
1434 
1435 static int system_tr_open(struct inode *inode, struct file *filp)
1436 {
1437 	struct trace_subsystem_dir *dir;
1438 	struct trace_array *tr = inode->i_private;
1439 	int ret;
1440 
1441 	if (tracing_is_disabled())
1442 		return -ENODEV;
1443 
1444 	if (trace_array_get(tr) < 0)
1445 		return -ENODEV;
1446 
1447 	/* Make a temporary dir that has no system but points to tr */
1448 	dir = kzalloc(sizeof(*dir), GFP_KERNEL);
1449 	if (!dir) {
1450 		trace_array_put(tr);
1451 		return -ENOMEM;
1452 	}
1453 
1454 	dir->tr = tr;
1455 
1456 	ret = tracing_open_generic(inode, filp);
1457 	if (ret < 0) {
1458 		trace_array_put(tr);
1459 		kfree(dir);
1460 		return ret;
1461 	}
1462 
1463 	filp->private_data = dir;
1464 
1465 	return 0;
1466 }
1467 
1468 static int subsystem_release(struct inode *inode, struct file *file)
1469 {
1470 	struct trace_subsystem_dir *dir = file->private_data;
1471 
1472 	trace_array_put(dir->tr);
1473 
1474 	/*
1475 	 * If dir->subsystem is NULL, then this is a temporary
1476 	 * descriptor that was made for a trace_array to enable
1477 	 * all subsystems.
1478 	 */
1479 	if (dir->subsystem)
1480 		put_system(dir);
1481 	else
1482 		kfree(dir);
1483 
1484 	return 0;
1485 }
1486 
1487 static ssize_t
1488 subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1489 		      loff_t *ppos)
1490 {
1491 	struct trace_subsystem_dir *dir = filp->private_data;
1492 	struct event_subsystem *system = dir->subsystem;
1493 	struct trace_seq *s;
1494 	int r;
1495 
1496 	if (*ppos)
1497 		return 0;
1498 
1499 	s = kmalloc(sizeof(*s), GFP_KERNEL);
1500 	if (!s)
1501 		return -ENOMEM;
1502 
1503 	trace_seq_init(s);
1504 
1505 	print_subsystem_event_filter(system, s);
1506 	r = simple_read_from_buffer(ubuf, cnt, ppos,
1507 				    s->buffer, trace_seq_used(s));
1508 
1509 	kfree(s);
1510 
1511 	return r;
1512 }
1513 
1514 static ssize_t
1515 subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1516 		       loff_t *ppos)
1517 {
1518 	struct trace_subsystem_dir *dir = filp->private_data;
1519 	char *buf;
1520 	int err;
1521 
1522 	if (cnt >= PAGE_SIZE)
1523 		return -EINVAL;
1524 
1525 	buf = memdup_user_nul(ubuf, cnt);
1526 	if (IS_ERR(buf))
1527 		return PTR_ERR(buf);
1528 
1529 	err = apply_subsystem_event_filter(dir, buf);
1530 	kfree(buf);
1531 	if (err < 0)
1532 		return err;
1533 
1534 	*ppos += cnt;
1535 
1536 	return cnt;
1537 }
1538 
1539 static ssize_t
1540 show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1541 {
1542 	int (*func)(struct trace_seq *s) = filp->private_data;
1543 	struct trace_seq *s;
1544 	int r;
1545 
1546 	if (*ppos)
1547 		return 0;
1548 
1549 	s = kmalloc(sizeof(*s), GFP_KERNEL);
1550 	if (!s)
1551 		return -ENOMEM;
1552 
1553 	trace_seq_init(s);
1554 
1555 	func(s);
1556 	r = simple_read_from_buffer(ubuf, cnt, ppos,
1557 				    s->buffer, trace_seq_used(s));
1558 
1559 	kfree(s);
1560 
1561 	return r;
1562 }
1563 
1564 static int max_pids(struct trace_pid_list *pid_list)
1565 {
1566 	return (PAGE_SIZE << pid_list->order) / sizeof(pid_t);
1567 }
1568 
1569 static void ignore_task_cpu(void *data)
1570 {
1571 	struct trace_array *tr = data;
1572 	struct trace_pid_list *pid_list;
1573 
1574 	/*
1575 	 * This function is called by on_each_cpu() while the
1576 	 * event_mutex is held.
1577 	 */
1578 	pid_list = rcu_dereference_protected(tr->filtered_pids,
1579 					     mutex_is_locked(&event_mutex));
1580 
1581 	this_cpu_write(tr->trace_buffer.data->ignore_pid,
1582 		       check_ignore_pid(pid_list, current));
1583 }
1584 
1585 static ssize_t
1586 ftrace_event_pid_write(struct file *filp, const char __user *ubuf,
1587 		       size_t cnt, loff_t *ppos)
1588 {
1589 	struct seq_file *m = filp->private_data;
1590 	struct trace_array *tr = m->private;
1591 	struct trace_pid_list *filtered_pids = NULL;
1592 	struct trace_pid_list *pid_list = NULL;
1593 	struct trace_event_file *file;
1594 	struct trace_parser parser;
1595 	unsigned long val;
1596 	loff_t this_pos;
1597 	ssize_t read = 0;
1598 	ssize_t ret = 0;
1599 	pid_t pid;
1600 	int i;
1601 
1602 	if (!cnt)
1603 		return 0;
1604 
1605 	ret = tracing_update_buffers();
1606 	if (ret < 0)
1607 		return ret;
1608 
1609 	if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
1610 		return -ENOMEM;
1611 
1612 	mutex_lock(&event_mutex);
1613 	/*
1614 	 * Load as many pids into the array before doing a
1615 	 * swap from the tr->filtered_pids to the new list.
1616 	 */
1617 	while (cnt > 0) {
1618 
1619 		this_pos = 0;
1620 
1621 		ret = trace_get_user(&parser, ubuf, cnt, &this_pos);
1622 		if (ret < 0 || !trace_parser_loaded(&parser))
1623 			break;
1624 
1625 		read += ret;
1626 		ubuf += ret;
1627 		cnt -= ret;
1628 
1629 		parser.buffer[parser.idx] = 0;
1630 
1631 		ret = -EINVAL;
1632 		if (kstrtoul(parser.buffer, 0, &val))
1633 			break;
1634 		if (val > INT_MAX)
1635 			break;
1636 
1637 		pid = (pid_t)val;
1638 
1639 		ret = -ENOMEM;
1640 		if (!pid_list) {
1641 			pid_list = kmalloc(sizeof(*pid_list), GFP_KERNEL);
1642 			if (!pid_list)
1643 				break;
1644 
1645 			filtered_pids = rcu_dereference_protected(tr->filtered_pids,
1646 							lockdep_is_held(&event_mutex));
1647 			if (filtered_pids)
1648 				pid_list->order = filtered_pids->order;
1649 			else
1650 				pid_list->order = 0;
1651 
1652 			pid_list->pids = (void *)__get_free_pages(GFP_KERNEL,
1653 								  pid_list->order);
1654 			if (!pid_list->pids)
1655 				break;
1656 
1657 			if (filtered_pids) {
1658 				pid_list->nr_pids = filtered_pids->nr_pids;
1659 				memcpy(pid_list->pids, filtered_pids->pids,
1660 				       pid_list->nr_pids * sizeof(pid_t));
1661 			} else
1662 				pid_list->nr_pids = 0;
1663 		}
1664 
1665 		if (pid_list->nr_pids >= max_pids(pid_list)) {
1666 			pid_t *pid_page;
1667 
1668 			pid_page = (void *)__get_free_pages(GFP_KERNEL,
1669 							    pid_list->order + 1);
1670 			if (!pid_page)
1671 				break;
1672 			memcpy(pid_page, pid_list->pids,
1673 			       pid_list->nr_pids * sizeof(pid_t));
1674 			free_pages((unsigned long)pid_list->pids, pid_list->order);
1675 
1676 			pid_list->order++;
1677 			pid_list->pids = pid_page;
1678 		}
1679 
1680 		pid_list->pids[pid_list->nr_pids++] = pid;
1681 		trace_parser_clear(&parser);
1682 		ret = 0;
1683 	}
1684 	trace_parser_put(&parser);
1685 
1686 	if (ret < 0) {
1687 		if (pid_list)
1688 			free_pages((unsigned long)pid_list->pids, pid_list->order);
1689 		kfree(pid_list);
1690 		mutex_unlock(&event_mutex);
1691 		return ret;
1692 	}
1693 
1694 	if (!pid_list) {
1695 		mutex_unlock(&event_mutex);
1696 		return ret;
1697 	}
1698 
1699 	sort(pid_list->pids, pid_list->nr_pids, sizeof(pid_t), cmp_pid, NULL);
1700 
1701 	/* Remove duplicates */
1702 	for (i = 1; i < pid_list->nr_pids; i++) {
1703 		int start = i;
1704 
1705 		while (i < pid_list->nr_pids &&
1706 		       pid_list->pids[i - 1] == pid_list->pids[i])
1707 			i++;
1708 
1709 		if (start != i) {
1710 			if (i < pid_list->nr_pids) {
1711 				memmove(&pid_list->pids[start], &pid_list->pids[i],
1712 					(pid_list->nr_pids - i) * sizeof(pid_t));
1713 				pid_list->nr_pids -= i - start;
1714 				i = start;
1715 			} else
1716 				pid_list->nr_pids = start;
1717 		}
1718 	}
1719 
1720 	rcu_assign_pointer(tr->filtered_pids, pid_list);
1721 
1722 	list_for_each_entry(file, &tr->events, list) {
1723 		set_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
1724 	}
1725 
1726 	if (filtered_pids) {
1727 		synchronize_sched();
1728 
1729 		free_pages((unsigned long)filtered_pids->pids, filtered_pids->order);
1730 		kfree(filtered_pids);
1731 	} else {
1732 		/*
1733 		 * Register a probe that is called before all other probes
1734 		 * to set ignore_pid if next or prev do not match.
1735 		 * Register a probe this is called after all other probes
1736 		 * to only keep ignore_pid set if next pid matches.
1737 		 */
1738 		register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_pre,
1739 						 tr, INT_MAX);
1740 		register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_post,
1741 						 tr, 0);
1742 
1743 		register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre,
1744 						 tr, INT_MAX);
1745 		register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_post,
1746 						 tr, 0);
1747 
1748 		register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre,
1749 						     tr, INT_MAX);
1750 		register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post,
1751 						     tr, 0);
1752 
1753 		register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_pre,
1754 						 tr, INT_MAX);
1755 		register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_post,
1756 						 tr, 0);
1757 	}
1758 
1759 	/*
1760 	 * Ignoring of pids is done at task switch. But we have to
1761 	 * check for those tasks that are currently running.
1762 	 * Always do this in case a pid was appended or removed.
1763 	 */
1764 	on_each_cpu(ignore_task_cpu, tr, 1);
1765 
1766 	mutex_unlock(&event_mutex);
1767 
1768 	ret = read;
1769 	*ppos += read;
1770 
1771 	return ret;
1772 }
1773 
1774 static int ftrace_event_avail_open(struct inode *inode, struct file *file);
1775 static int ftrace_event_set_open(struct inode *inode, struct file *file);
1776 static int ftrace_event_set_pid_open(struct inode *inode, struct file *file);
1777 static int ftrace_event_release(struct inode *inode, struct file *file);
1778 
1779 static const struct seq_operations show_event_seq_ops = {
1780 	.start = t_start,
1781 	.next = t_next,
1782 	.show = t_show,
1783 	.stop = t_stop,
1784 };
1785 
1786 static const struct seq_operations show_set_event_seq_ops = {
1787 	.start = s_start,
1788 	.next = s_next,
1789 	.show = t_show,
1790 	.stop = t_stop,
1791 };
1792 
1793 static const struct seq_operations show_set_pid_seq_ops = {
1794 	.start = p_start,
1795 	.next = p_next,
1796 	.show = p_show,
1797 	.stop = p_stop,
1798 };
1799 
1800 static const struct file_operations ftrace_avail_fops = {
1801 	.open = ftrace_event_avail_open,
1802 	.read = seq_read,
1803 	.llseek = seq_lseek,
1804 	.release = seq_release,
1805 };
1806 
1807 static const struct file_operations ftrace_set_event_fops = {
1808 	.open = ftrace_event_set_open,
1809 	.read = seq_read,
1810 	.write = ftrace_event_write,
1811 	.llseek = seq_lseek,
1812 	.release = ftrace_event_release,
1813 };
1814 
1815 static const struct file_operations ftrace_set_event_pid_fops = {
1816 	.open = ftrace_event_set_pid_open,
1817 	.read = seq_read,
1818 	.write = ftrace_event_pid_write,
1819 	.llseek = seq_lseek,
1820 	.release = ftrace_event_release,
1821 };
1822 
1823 static const struct file_operations ftrace_enable_fops = {
1824 	.open = tracing_open_generic,
1825 	.read = event_enable_read,
1826 	.write = event_enable_write,
1827 	.llseek = default_llseek,
1828 };
1829 
1830 static const struct file_operations ftrace_event_format_fops = {
1831 	.open = trace_format_open,
1832 	.read = seq_read,
1833 	.llseek = seq_lseek,
1834 	.release = seq_release,
1835 };
1836 
1837 static const struct file_operations ftrace_event_id_fops = {
1838 	.read = event_id_read,
1839 	.llseek = default_llseek,
1840 };
1841 
1842 static const struct file_operations ftrace_event_filter_fops = {
1843 	.open = tracing_open_generic,
1844 	.read = event_filter_read,
1845 	.write = event_filter_write,
1846 	.llseek = default_llseek,
1847 };
1848 
1849 static const struct file_operations ftrace_subsystem_filter_fops = {
1850 	.open = subsystem_open,
1851 	.read = subsystem_filter_read,
1852 	.write = subsystem_filter_write,
1853 	.llseek = default_llseek,
1854 	.release = subsystem_release,
1855 };
1856 
1857 static const struct file_operations ftrace_system_enable_fops = {
1858 	.open = subsystem_open,
1859 	.read = system_enable_read,
1860 	.write = system_enable_write,
1861 	.llseek = default_llseek,
1862 	.release = subsystem_release,
1863 };
1864 
1865 static const struct file_operations ftrace_tr_enable_fops = {
1866 	.open = system_tr_open,
1867 	.read = system_enable_read,
1868 	.write = system_enable_write,
1869 	.llseek = default_llseek,
1870 	.release = subsystem_release,
1871 };
1872 
1873 static const struct file_operations ftrace_show_header_fops = {
1874 	.open = tracing_open_generic,
1875 	.read = show_header,
1876 	.llseek = default_llseek,
1877 };
1878 
1879 static int
1880 ftrace_event_open(struct inode *inode, struct file *file,
1881 		  const struct seq_operations *seq_ops)
1882 {
1883 	struct seq_file *m;
1884 	int ret;
1885 
1886 	ret = seq_open(file, seq_ops);
1887 	if (ret < 0)
1888 		return ret;
1889 	m = file->private_data;
1890 	/* copy tr over to seq ops */
1891 	m->private = inode->i_private;
1892 
1893 	return ret;
1894 }
1895 
1896 static int ftrace_event_release(struct inode *inode, struct file *file)
1897 {
1898 	struct trace_array *tr = inode->i_private;
1899 
1900 	trace_array_put(tr);
1901 
1902 	return seq_release(inode, file);
1903 }
1904 
1905 static int
1906 ftrace_event_avail_open(struct inode *inode, struct file *file)
1907 {
1908 	const struct seq_operations *seq_ops = &show_event_seq_ops;
1909 
1910 	return ftrace_event_open(inode, file, seq_ops);
1911 }
1912 
1913 static int
1914 ftrace_event_set_open(struct inode *inode, struct file *file)
1915 {
1916 	const struct seq_operations *seq_ops = &show_set_event_seq_ops;
1917 	struct trace_array *tr = inode->i_private;
1918 	int ret;
1919 
1920 	if (trace_array_get(tr) < 0)
1921 		return -ENODEV;
1922 
1923 	if ((file->f_mode & FMODE_WRITE) &&
1924 	    (file->f_flags & O_TRUNC))
1925 		ftrace_clear_events(tr);
1926 
1927 	ret = ftrace_event_open(inode, file, seq_ops);
1928 	if (ret < 0)
1929 		trace_array_put(tr);
1930 	return ret;
1931 }
1932 
1933 static int
1934 ftrace_event_set_pid_open(struct inode *inode, struct file *file)
1935 {
1936 	const struct seq_operations *seq_ops = &show_set_pid_seq_ops;
1937 	struct trace_array *tr = inode->i_private;
1938 	int ret;
1939 
1940 	if (trace_array_get(tr) < 0)
1941 		return -ENODEV;
1942 
1943 	if ((file->f_mode & FMODE_WRITE) &&
1944 	    (file->f_flags & O_TRUNC))
1945 		ftrace_clear_event_pids(tr);
1946 
1947 	ret = ftrace_event_open(inode, file, seq_ops);
1948 	if (ret < 0)
1949 		trace_array_put(tr);
1950 	return ret;
1951 }
1952 
1953 static struct event_subsystem *
1954 create_new_subsystem(const char *name)
1955 {
1956 	struct event_subsystem *system;
1957 
1958 	/* need to create new entry */
1959 	system = kmalloc(sizeof(*system), GFP_KERNEL);
1960 	if (!system)
1961 		return NULL;
1962 
1963 	system->ref_count = 1;
1964 
1965 	/* Only allocate if dynamic (kprobes and modules) */
1966 	system->name = kstrdup_const(name, GFP_KERNEL);
1967 	if (!system->name)
1968 		goto out_free;
1969 
1970 	system->filter = NULL;
1971 
1972 	system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
1973 	if (!system->filter)
1974 		goto out_free;
1975 
1976 	list_add(&system->list, &event_subsystems);
1977 
1978 	return system;
1979 
1980  out_free:
1981 	kfree_const(system->name);
1982 	kfree(system);
1983 	return NULL;
1984 }
1985 
1986 static struct dentry *
1987 event_subsystem_dir(struct trace_array *tr, const char *name,
1988 		    struct trace_event_file *file, struct dentry *parent)
1989 {
1990 	struct trace_subsystem_dir *dir;
1991 	struct event_subsystem *system;
1992 	struct dentry *entry;
1993 
1994 	/* First see if we did not already create this dir */
1995 	list_for_each_entry(dir, &tr->systems, list) {
1996 		system = dir->subsystem;
1997 		if (strcmp(system->name, name) == 0) {
1998 			dir->nr_events++;
1999 			file->system = dir;
2000 			return dir->entry;
2001 		}
2002 	}
2003 
2004 	/* Now see if the system itself exists. */
2005 	list_for_each_entry(system, &event_subsystems, list) {
2006 		if (strcmp(system->name, name) == 0)
2007 			break;
2008 	}
2009 	/* Reset system variable when not found */
2010 	if (&system->list == &event_subsystems)
2011 		system = NULL;
2012 
2013 	dir = kmalloc(sizeof(*dir), GFP_KERNEL);
2014 	if (!dir)
2015 		goto out_fail;
2016 
2017 	if (!system) {
2018 		system = create_new_subsystem(name);
2019 		if (!system)
2020 			goto out_free;
2021 	} else
2022 		__get_system(system);
2023 
2024 	dir->entry = tracefs_create_dir(name, parent);
2025 	if (!dir->entry) {
2026 		pr_warn("Failed to create system directory %s\n", name);
2027 		__put_system(system);
2028 		goto out_free;
2029 	}
2030 
2031 	dir->tr = tr;
2032 	dir->ref_count = 1;
2033 	dir->nr_events = 1;
2034 	dir->subsystem = system;
2035 	file->system = dir;
2036 
2037 	entry = tracefs_create_file("filter", 0644, dir->entry, dir,
2038 				    &ftrace_subsystem_filter_fops);
2039 	if (!entry) {
2040 		kfree(system->filter);
2041 		system->filter = NULL;
2042 		pr_warn("Could not create tracefs '%s/filter' entry\n", name);
2043 	}
2044 
2045 	trace_create_file("enable", 0644, dir->entry, dir,
2046 			  &ftrace_system_enable_fops);
2047 
2048 	list_add(&dir->list, &tr->systems);
2049 
2050 	return dir->entry;
2051 
2052  out_free:
2053 	kfree(dir);
2054  out_fail:
2055 	/* Only print this message if failed on memory allocation */
2056 	if (!dir || !system)
2057 		pr_warn("No memory to create event subsystem %s\n", name);
2058 	return NULL;
2059 }
2060 
2061 static int
2062 event_create_dir(struct dentry *parent, struct trace_event_file *file)
2063 {
2064 	struct trace_event_call *call = file->event_call;
2065 	struct trace_array *tr = file->tr;
2066 	struct list_head *head;
2067 	struct dentry *d_events;
2068 	const char *name;
2069 	int ret;
2070 
2071 	/*
2072 	 * If the trace point header did not define TRACE_SYSTEM
2073 	 * then the system would be called "TRACE_SYSTEM".
2074 	 */
2075 	if (strcmp(call->class->system, TRACE_SYSTEM) != 0) {
2076 		d_events = event_subsystem_dir(tr, call->class->system, file, parent);
2077 		if (!d_events)
2078 			return -ENOMEM;
2079 	} else
2080 		d_events = parent;
2081 
2082 	name = trace_event_name(call);
2083 	file->dir = tracefs_create_dir(name, d_events);
2084 	if (!file->dir) {
2085 		pr_warn("Could not create tracefs '%s' directory\n", name);
2086 		return -1;
2087 	}
2088 
2089 	if (call->class->reg && !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
2090 		trace_create_file("enable", 0644, file->dir, file,
2091 				  &ftrace_enable_fops);
2092 
2093 #ifdef CONFIG_PERF_EVENTS
2094 	if (call->event.type && call->class->reg)
2095 		trace_create_file("id", 0444, file->dir,
2096 				  (void *)(long)call->event.type,
2097 				  &ftrace_event_id_fops);
2098 #endif
2099 
2100 	/*
2101 	 * Other events may have the same class. Only update
2102 	 * the fields if they are not already defined.
2103 	 */
2104 	head = trace_get_fields(call);
2105 	if (list_empty(head)) {
2106 		ret = call->class->define_fields(call);
2107 		if (ret < 0) {
2108 			pr_warn("Could not initialize trace point events/%s\n",
2109 				name);
2110 			return -1;
2111 		}
2112 	}
2113 	trace_create_file("filter", 0644, file->dir, file,
2114 			  &ftrace_event_filter_fops);
2115 
2116 	/*
2117 	 * Only event directories that can be enabled should have
2118 	 * triggers.
2119 	 */
2120 	if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
2121 		trace_create_file("trigger", 0644, file->dir, file,
2122 				  &event_trigger_fops);
2123 
2124 	trace_create_file("format", 0444, file->dir, call,
2125 			  &ftrace_event_format_fops);
2126 
2127 	return 0;
2128 }
2129 
2130 static void remove_event_from_tracers(struct trace_event_call *call)
2131 {
2132 	struct trace_event_file *file;
2133 	struct trace_array *tr;
2134 
2135 	do_for_each_event_file_safe(tr, file) {
2136 		if (file->event_call != call)
2137 			continue;
2138 
2139 		remove_event_file_dir(file);
2140 		/*
2141 		 * The do_for_each_event_file_safe() is
2142 		 * a double loop. After finding the call for this
2143 		 * trace_array, we use break to jump to the next
2144 		 * trace_array.
2145 		 */
2146 		break;
2147 	} while_for_each_event_file();
2148 }
2149 
2150 static void event_remove(struct trace_event_call *call)
2151 {
2152 	struct trace_array *tr;
2153 	struct trace_event_file *file;
2154 
2155 	do_for_each_event_file(tr, file) {
2156 		if (file->event_call != call)
2157 			continue;
2158 		ftrace_event_enable_disable(file, 0);
2159 		/*
2160 		 * The do_for_each_event_file() is
2161 		 * a double loop. After finding the call for this
2162 		 * trace_array, we use break to jump to the next
2163 		 * trace_array.
2164 		 */
2165 		break;
2166 	} while_for_each_event_file();
2167 
2168 	if (call->event.funcs)
2169 		__unregister_trace_event(&call->event);
2170 	remove_event_from_tracers(call);
2171 	list_del(&call->list);
2172 }
2173 
2174 static int event_init(struct trace_event_call *call)
2175 {
2176 	int ret = 0;
2177 	const char *name;
2178 
2179 	name = trace_event_name(call);
2180 	if (WARN_ON(!name))
2181 		return -EINVAL;
2182 
2183 	if (call->class->raw_init) {
2184 		ret = call->class->raw_init(call);
2185 		if (ret < 0 && ret != -ENOSYS)
2186 			pr_warn("Could not initialize trace events/%s\n", name);
2187 	}
2188 
2189 	return ret;
2190 }
2191 
2192 static int
2193 __register_event(struct trace_event_call *call, struct module *mod)
2194 {
2195 	int ret;
2196 
2197 	ret = event_init(call);
2198 	if (ret < 0)
2199 		return ret;
2200 
2201 	list_add(&call->list, &ftrace_events);
2202 	call->mod = mod;
2203 
2204 	return 0;
2205 }
2206 
2207 static char *enum_replace(char *ptr, struct trace_enum_map *map, int len)
2208 {
2209 	int rlen;
2210 	int elen;
2211 
2212 	/* Find the length of the enum value as a string */
2213 	elen = snprintf(ptr, 0, "%ld", map->enum_value);
2214 	/* Make sure there's enough room to replace the string with the value */
2215 	if (len < elen)
2216 		return NULL;
2217 
2218 	snprintf(ptr, elen + 1, "%ld", map->enum_value);
2219 
2220 	/* Get the rest of the string of ptr */
2221 	rlen = strlen(ptr + len);
2222 	memmove(ptr + elen, ptr + len, rlen);
2223 	/* Make sure we end the new string */
2224 	ptr[elen + rlen] = 0;
2225 
2226 	return ptr + elen;
2227 }
2228 
2229 static void update_event_printk(struct trace_event_call *call,
2230 				struct trace_enum_map *map)
2231 {
2232 	char *ptr;
2233 	int quote = 0;
2234 	int len = strlen(map->enum_string);
2235 
2236 	for (ptr = call->print_fmt; *ptr; ptr++) {
2237 		if (*ptr == '\\') {
2238 			ptr++;
2239 			/* paranoid */
2240 			if (!*ptr)
2241 				break;
2242 			continue;
2243 		}
2244 		if (*ptr == '"') {
2245 			quote ^= 1;
2246 			continue;
2247 		}
2248 		if (quote)
2249 			continue;
2250 		if (isdigit(*ptr)) {
2251 			/* skip numbers */
2252 			do {
2253 				ptr++;
2254 				/* Check for alpha chars like ULL */
2255 			} while (isalnum(*ptr));
2256 			if (!*ptr)
2257 				break;
2258 			/*
2259 			 * A number must have some kind of delimiter after
2260 			 * it, and we can ignore that too.
2261 			 */
2262 			continue;
2263 		}
2264 		if (isalpha(*ptr) || *ptr == '_') {
2265 			if (strncmp(map->enum_string, ptr, len) == 0 &&
2266 			    !isalnum(ptr[len]) && ptr[len] != '_') {
2267 				ptr = enum_replace(ptr, map, len);
2268 				/* Hmm, enum string smaller than value */
2269 				if (WARN_ON_ONCE(!ptr))
2270 					return;
2271 				/*
2272 				 * No need to decrement here, as enum_replace()
2273 				 * returns the pointer to the character passed
2274 				 * the enum, and two enums can not be placed
2275 				 * back to back without something in between.
2276 				 * We can skip that something in between.
2277 				 */
2278 				continue;
2279 			}
2280 		skip_more:
2281 			do {
2282 				ptr++;
2283 			} while (isalnum(*ptr) || *ptr == '_');
2284 			if (!*ptr)
2285 				break;
2286 			/*
2287 			 * If what comes after this variable is a '.' or
2288 			 * '->' then we can continue to ignore that string.
2289 			 */
2290 			if (*ptr == '.' || (ptr[0] == '-' && ptr[1] == '>')) {
2291 				ptr += *ptr == '.' ? 1 : 2;
2292 				if (!*ptr)
2293 					break;
2294 				goto skip_more;
2295 			}
2296 			/*
2297 			 * Once again, we can skip the delimiter that came
2298 			 * after the string.
2299 			 */
2300 			continue;
2301 		}
2302 	}
2303 }
2304 
2305 void trace_event_enum_update(struct trace_enum_map **map, int len)
2306 {
2307 	struct trace_event_call *call, *p;
2308 	const char *last_system = NULL;
2309 	int last_i;
2310 	int i;
2311 
2312 	down_write(&trace_event_sem);
2313 	list_for_each_entry_safe(call, p, &ftrace_events, list) {
2314 		/* events are usually grouped together with systems */
2315 		if (!last_system || call->class->system != last_system) {
2316 			last_i = 0;
2317 			last_system = call->class->system;
2318 		}
2319 
2320 		for (i = last_i; i < len; i++) {
2321 			if (call->class->system == map[i]->system) {
2322 				/* Save the first system if need be */
2323 				if (!last_i)
2324 					last_i = i;
2325 				update_event_printk(call, map[i]);
2326 			}
2327 		}
2328 	}
2329 	up_write(&trace_event_sem);
2330 }
2331 
2332 static struct trace_event_file *
2333 trace_create_new_event(struct trace_event_call *call,
2334 		       struct trace_array *tr)
2335 {
2336 	struct trace_event_file *file;
2337 
2338 	file = kmem_cache_alloc(file_cachep, GFP_TRACE);
2339 	if (!file)
2340 		return NULL;
2341 
2342 	file->event_call = call;
2343 	file->tr = tr;
2344 	atomic_set(&file->sm_ref, 0);
2345 	atomic_set(&file->tm_ref, 0);
2346 	INIT_LIST_HEAD(&file->triggers);
2347 	list_add(&file->list, &tr->events);
2348 
2349 	return file;
2350 }
2351 
2352 /* Add an event to a trace directory */
2353 static int
2354 __trace_add_new_event(struct trace_event_call *call, struct trace_array *tr)
2355 {
2356 	struct trace_event_file *file;
2357 
2358 	file = trace_create_new_event(call, tr);
2359 	if (!file)
2360 		return -ENOMEM;
2361 
2362 	return event_create_dir(tr->event_dir, file);
2363 }
2364 
2365 /*
2366  * Just create a decriptor for early init. A descriptor is required
2367  * for enabling events at boot. We want to enable events before
2368  * the filesystem is initialized.
2369  */
2370 static __init int
2371 __trace_early_add_new_event(struct trace_event_call *call,
2372 			    struct trace_array *tr)
2373 {
2374 	struct trace_event_file *file;
2375 
2376 	file = trace_create_new_event(call, tr);
2377 	if (!file)
2378 		return -ENOMEM;
2379 
2380 	return 0;
2381 }
2382 
2383 struct ftrace_module_file_ops;
2384 static void __add_event_to_tracers(struct trace_event_call *call);
2385 
2386 /* Add an additional event_call dynamically */
2387 int trace_add_event_call(struct trace_event_call *call)
2388 {
2389 	int ret;
2390 	mutex_lock(&trace_types_lock);
2391 	mutex_lock(&event_mutex);
2392 
2393 	ret = __register_event(call, NULL);
2394 	if (ret >= 0)
2395 		__add_event_to_tracers(call);
2396 
2397 	mutex_unlock(&event_mutex);
2398 	mutex_unlock(&trace_types_lock);
2399 	return ret;
2400 }
2401 
2402 /*
2403  * Must be called under locking of trace_types_lock, event_mutex and
2404  * trace_event_sem.
2405  */
2406 static void __trace_remove_event_call(struct trace_event_call *call)
2407 {
2408 	event_remove(call);
2409 	trace_destroy_fields(call);
2410 	free_event_filter(call->filter);
2411 	call->filter = NULL;
2412 }
2413 
2414 static int probe_remove_event_call(struct trace_event_call *call)
2415 {
2416 	struct trace_array *tr;
2417 	struct trace_event_file *file;
2418 
2419 #ifdef CONFIG_PERF_EVENTS
2420 	if (call->perf_refcount)
2421 		return -EBUSY;
2422 #endif
2423 	do_for_each_event_file(tr, file) {
2424 		if (file->event_call != call)
2425 			continue;
2426 		/*
2427 		 * We can't rely on ftrace_event_enable_disable(enable => 0)
2428 		 * we are going to do, EVENT_FILE_FL_SOFT_MODE can suppress
2429 		 * TRACE_REG_UNREGISTER.
2430 		 */
2431 		if (file->flags & EVENT_FILE_FL_ENABLED)
2432 			return -EBUSY;
2433 		/*
2434 		 * The do_for_each_event_file_safe() is
2435 		 * a double loop. After finding the call for this
2436 		 * trace_array, we use break to jump to the next
2437 		 * trace_array.
2438 		 */
2439 		break;
2440 	} while_for_each_event_file();
2441 
2442 	__trace_remove_event_call(call);
2443 
2444 	return 0;
2445 }
2446 
2447 /* Remove an event_call */
2448 int trace_remove_event_call(struct trace_event_call *call)
2449 {
2450 	int ret;
2451 
2452 	mutex_lock(&trace_types_lock);
2453 	mutex_lock(&event_mutex);
2454 	down_write(&trace_event_sem);
2455 	ret = probe_remove_event_call(call);
2456 	up_write(&trace_event_sem);
2457 	mutex_unlock(&event_mutex);
2458 	mutex_unlock(&trace_types_lock);
2459 
2460 	return ret;
2461 }
2462 
2463 #define for_each_event(event, start, end)			\
2464 	for (event = start;					\
2465 	     (unsigned long)event < (unsigned long)end;		\
2466 	     event++)
2467 
2468 #ifdef CONFIG_MODULES
2469 
2470 static void trace_module_add_events(struct module *mod)
2471 {
2472 	struct trace_event_call **call, **start, **end;
2473 
2474 	if (!mod->num_trace_events)
2475 		return;
2476 
2477 	/* Don't add infrastructure for mods without tracepoints */
2478 	if (trace_module_has_bad_taint(mod)) {
2479 		pr_err("%s: module has bad taint, not creating trace events\n",
2480 		       mod->name);
2481 		return;
2482 	}
2483 
2484 	start = mod->trace_events;
2485 	end = mod->trace_events + mod->num_trace_events;
2486 
2487 	for_each_event(call, start, end) {
2488 		__register_event(*call, mod);
2489 		__add_event_to_tracers(*call);
2490 	}
2491 }
2492 
2493 static void trace_module_remove_events(struct module *mod)
2494 {
2495 	struct trace_event_call *call, *p;
2496 	bool clear_trace = false;
2497 
2498 	down_write(&trace_event_sem);
2499 	list_for_each_entry_safe(call, p, &ftrace_events, list) {
2500 		if (call->mod == mod) {
2501 			if (call->flags & TRACE_EVENT_FL_WAS_ENABLED)
2502 				clear_trace = true;
2503 			__trace_remove_event_call(call);
2504 		}
2505 	}
2506 	up_write(&trace_event_sem);
2507 
2508 	/*
2509 	 * It is safest to reset the ring buffer if the module being unloaded
2510 	 * registered any events that were used. The only worry is if
2511 	 * a new module gets loaded, and takes on the same id as the events
2512 	 * of this module. When printing out the buffer, traced events left
2513 	 * over from this module may be passed to the new module events and
2514 	 * unexpected results may occur.
2515 	 */
2516 	if (clear_trace)
2517 		tracing_reset_all_online_cpus();
2518 }
2519 
2520 static int trace_module_notify(struct notifier_block *self,
2521 			       unsigned long val, void *data)
2522 {
2523 	struct module *mod = data;
2524 
2525 	mutex_lock(&trace_types_lock);
2526 	mutex_lock(&event_mutex);
2527 	switch (val) {
2528 	case MODULE_STATE_COMING:
2529 		trace_module_add_events(mod);
2530 		break;
2531 	case MODULE_STATE_GOING:
2532 		trace_module_remove_events(mod);
2533 		break;
2534 	}
2535 	mutex_unlock(&event_mutex);
2536 	mutex_unlock(&trace_types_lock);
2537 
2538 	return 0;
2539 }
2540 
2541 static struct notifier_block trace_module_nb = {
2542 	.notifier_call = trace_module_notify,
2543 	.priority = 1, /* higher than trace.c module notify */
2544 };
2545 #endif /* CONFIG_MODULES */
2546 
2547 /* Create a new event directory structure for a trace directory. */
2548 static void
2549 __trace_add_event_dirs(struct trace_array *tr)
2550 {
2551 	struct trace_event_call *call;
2552 	int ret;
2553 
2554 	list_for_each_entry(call, &ftrace_events, list) {
2555 		ret = __trace_add_new_event(call, tr);
2556 		if (ret < 0)
2557 			pr_warn("Could not create directory for event %s\n",
2558 				trace_event_name(call));
2559 	}
2560 }
2561 
2562 struct trace_event_file *
2563 find_event_file(struct trace_array *tr, const char *system,  const char *event)
2564 {
2565 	struct trace_event_file *file;
2566 	struct trace_event_call *call;
2567 	const char *name;
2568 
2569 	list_for_each_entry(file, &tr->events, list) {
2570 
2571 		call = file->event_call;
2572 		name = trace_event_name(call);
2573 
2574 		if (!name || !call->class || !call->class->reg)
2575 			continue;
2576 
2577 		if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
2578 			continue;
2579 
2580 		if (strcmp(event, name) == 0 &&
2581 		    strcmp(system, call->class->system) == 0)
2582 			return file;
2583 	}
2584 	return NULL;
2585 }
2586 
2587 #ifdef CONFIG_DYNAMIC_FTRACE
2588 
2589 /* Avoid typos */
2590 #define ENABLE_EVENT_STR	"enable_event"
2591 #define DISABLE_EVENT_STR	"disable_event"
2592 
2593 struct event_probe_data {
2594 	struct trace_event_file	*file;
2595 	unsigned long			count;
2596 	int				ref;
2597 	bool				enable;
2598 };
2599 
2600 static void
2601 event_enable_probe(unsigned long ip, unsigned long parent_ip, void **_data)
2602 {
2603 	struct event_probe_data **pdata = (struct event_probe_data **)_data;
2604 	struct event_probe_data *data = *pdata;
2605 
2606 	if (!data)
2607 		return;
2608 
2609 	if (data->enable)
2610 		clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
2611 	else
2612 		set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
2613 }
2614 
2615 static void
2616 event_enable_count_probe(unsigned long ip, unsigned long parent_ip, void **_data)
2617 {
2618 	struct event_probe_data **pdata = (struct event_probe_data **)_data;
2619 	struct event_probe_data *data = *pdata;
2620 
2621 	if (!data)
2622 		return;
2623 
2624 	if (!data->count)
2625 		return;
2626 
2627 	/* Skip if the event is in a state we want to switch to */
2628 	if (data->enable == !(data->file->flags & EVENT_FILE_FL_SOFT_DISABLED))
2629 		return;
2630 
2631 	if (data->count != -1)
2632 		(data->count)--;
2633 
2634 	event_enable_probe(ip, parent_ip, _data);
2635 }
2636 
2637 static int
2638 event_enable_print(struct seq_file *m, unsigned long ip,
2639 		      struct ftrace_probe_ops *ops, void *_data)
2640 {
2641 	struct event_probe_data *data = _data;
2642 
2643 	seq_printf(m, "%ps:", (void *)ip);
2644 
2645 	seq_printf(m, "%s:%s:%s",
2646 		   data->enable ? ENABLE_EVENT_STR : DISABLE_EVENT_STR,
2647 		   data->file->event_call->class->system,
2648 		   trace_event_name(data->file->event_call));
2649 
2650 	if (data->count == -1)
2651 		seq_puts(m, ":unlimited\n");
2652 	else
2653 		seq_printf(m, ":count=%ld\n", data->count);
2654 
2655 	return 0;
2656 }
2657 
2658 static int
2659 event_enable_init(struct ftrace_probe_ops *ops, unsigned long ip,
2660 		  void **_data)
2661 {
2662 	struct event_probe_data **pdata = (struct event_probe_data **)_data;
2663 	struct event_probe_data *data = *pdata;
2664 
2665 	data->ref++;
2666 	return 0;
2667 }
2668 
2669 static void
2670 event_enable_free(struct ftrace_probe_ops *ops, unsigned long ip,
2671 		  void **_data)
2672 {
2673 	struct event_probe_data **pdata = (struct event_probe_data **)_data;
2674 	struct event_probe_data *data = *pdata;
2675 
2676 	if (WARN_ON_ONCE(data->ref <= 0))
2677 		return;
2678 
2679 	data->ref--;
2680 	if (!data->ref) {
2681 		/* Remove the SOFT_MODE flag */
2682 		__ftrace_event_enable_disable(data->file, 0, 1);
2683 		module_put(data->file->event_call->mod);
2684 		kfree(data);
2685 	}
2686 	*pdata = NULL;
2687 }
2688 
2689 static struct ftrace_probe_ops event_enable_probe_ops = {
2690 	.func			= event_enable_probe,
2691 	.print			= event_enable_print,
2692 	.init			= event_enable_init,
2693 	.free			= event_enable_free,
2694 };
2695 
2696 static struct ftrace_probe_ops event_enable_count_probe_ops = {
2697 	.func			= event_enable_count_probe,
2698 	.print			= event_enable_print,
2699 	.init			= event_enable_init,
2700 	.free			= event_enable_free,
2701 };
2702 
2703 static struct ftrace_probe_ops event_disable_probe_ops = {
2704 	.func			= event_enable_probe,
2705 	.print			= event_enable_print,
2706 	.init			= event_enable_init,
2707 	.free			= event_enable_free,
2708 };
2709 
2710 static struct ftrace_probe_ops event_disable_count_probe_ops = {
2711 	.func			= event_enable_count_probe,
2712 	.print			= event_enable_print,
2713 	.init			= event_enable_init,
2714 	.free			= event_enable_free,
2715 };
2716 
2717 static int
2718 event_enable_func(struct ftrace_hash *hash,
2719 		  char *glob, char *cmd, char *param, int enabled)
2720 {
2721 	struct trace_array *tr = top_trace_array();
2722 	struct trace_event_file *file;
2723 	struct ftrace_probe_ops *ops;
2724 	struct event_probe_data *data;
2725 	const char *system;
2726 	const char *event;
2727 	char *number;
2728 	bool enable;
2729 	int ret;
2730 
2731 	if (!tr)
2732 		return -ENODEV;
2733 
2734 	/* hash funcs only work with set_ftrace_filter */
2735 	if (!enabled || !param)
2736 		return -EINVAL;
2737 
2738 	system = strsep(&param, ":");
2739 	if (!param)
2740 		return -EINVAL;
2741 
2742 	event = strsep(&param, ":");
2743 
2744 	mutex_lock(&event_mutex);
2745 
2746 	ret = -EINVAL;
2747 	file = find_event_file(tr, system, event);
2748 	if (!file)
2749 		goto out;
2750 
2751 	enable = strcmp(cmd, ENABLE_EVENT_STR) == 0;
2752 
2753 	if (enable)
2754 		ops = param ? &event_enable_count_probe_ops : &event_enable_probe_ops;
2755 	else
2756 		ops = param ? &event_disable_count_probe_ops : &event_disable_probe_ops;
2757 
2758 	if (glob[0] == '!') {
2759 		unregister_ftrace_function_probe_func(glob+1, ops);
2760 		ret = 0;
2761 		goto out;
2762 	}
2763 
2764 	ret = -ENOMEM;
2765 	data = kzalloc(sizeof(*data), GFP_KERNEL);
2766 	if (!data)
2767 		goto out;
2768 
2769 	data->enable = enable;
2770 	data->count = -1;
2771 	data->file = file;
2772 
2773 	if (!param)
2774 		goto out_reg;
2775 
2776 	number = strsep(&param, ":");
2777 
2778 	ret = -EINVAL;
2779 	if (!strlen(number))
2780 		goto out_free;
2781 
2782 	/*
2783 	 * We use the callback data field (which is a pointer)
2784 	 * as our counter.
2785 	 */
2786 	ret = kstrtoul(number, 0, &data->count);
2787 	if (ret)
2788 		goto out_free;
2789 
2790  out_reg:
2791 	/* Don't let event modules unload while probe registered */
2792 	ret = try_module_get(file->event_call->mod);
2793 	if (!ret) {
2794 		ret = -EBUSY;
2795 		goto out_free;
2796 	}
2797 
2798 	ret = __ftrace_event_enable_disable(file, 1, 1);
2799 	if (ret < 0)
2800 		goto out_put;
2801 	ret = register_ftrace_function_probe(glob, ops, data);
2802 	/*
2803 	 * The above returns on success the # of functions enabled,
2804 	 * but if it didn't find any functions it returns zero.
2805 	 * Consider no functions a failure too.
2806 	 */
2807 	if (!ret) {
2808 		ret = -ENOENT;
2809 		goto out_disable;
2810 	} else if (ret < 0)
2811 		goto out_disable;
2812 	/* Just return zero, not the number of enabled functions */
2813 	ret = 0;
2814  out:
2815 	mutex_unlock(&event_mutex);
2816 	return ret;
2817 
2818  out_disable:
2819 	__ftrace_event_enable_disable(file, 0, 1);
2820  out_put:
2821 	module_put(file->event_call->mod);
2822  out_free:
2823 	kfree(data);
2824 	goto out;
2825 }
2826 
2827 static struct ftrace_func_command event_enable_cmd = {
2828 	.name			= ENABLE_EVENT_STR,
2829 	.func			= event_enable_func,
2830 };
2831 
2832 static struct ftrace_func_command event_disable_cmd = {
2833 	.name			= DISABLE_EVENT_STR,
2834 	.func			= event_enable_func,
2835 };
2836 
2837 static __init int register_event_cmds(void)
2838 {
2839 	int ret;
2840 
2841 	ret = register_ftrace_command(&event_enable_cmd);
2842 	if (WARN_ON(ret < 0))
2843 		return ret;
2844 	ret = register_ftrace_command(&event_disable_cmd);
2845 	if (WARN_ON(ret < 0))
2846 		unregister_ftrace_command(&event_enable_cmd);
2847 	return ret;
2848 }
2849 #else
2850 static inline int register_event_cmds(void) { return 0; }
2851 #endif /* CONFIG_DYNAMIC_FTRACE */
2852 
2853 /*
2854  * The top level array has already had its trace_event_file
2855  * descriptors created in order to allow for early events to
2856  * be recorded. This function is called after the tracefs has been
2857  * initialized, and we now have to create the files associated
2858  * to the events.
2859  */
2860 static __init void
2861 __trace_early_add_event_dirs(struct trace_array *tr)
2862 {
2863 	struct trace_event_file *file;
2864 	int ret;
2865 
2866 
2867 	list_for_each_entry(file, &tr->events, list) {
2868 		ret = event_create_dir(tr->event_dir, file);
2869 		if (ret < 0)
2870 			pr_warn("Could not create directory for event %s\n",
2871 				trace_event_name(file->event_call));
2872 	}
2873 }
2874 
2875 /*
2876  * For early boot up, the top trace array requires to have
2877  * a list of events that can be enabled. This must be done before
2878  * the filesystem is set up in order to allow events to be traced
2879  * early.
2880  */
2881 static __init void
2882 __trace_early_add_events(struct trace_array *tr)
2883 {
2884 	struct trace_event_call *call;
2885 	int ret;
2886 
2887 	list_for_each_entry(call, &ftrace_events, list) {
2888 		/* Early boot up should not have any modules loaded */
2889 		if (WARN_ON_ONCE(call->mod))
2890 			continue;
2891 
2892 		ret = __trace_early_add_new_event(call, tr);
2893 		if (ret < 0)
2894 			pr_warn("Could not create early event %s\n",
2895 				trace_event_name(call));
2896 	}
2897 }
2898 
2899 /* Remove the event directory structure for a trace directory. */
2900 static void
2901 __trace_remove_event_dirs(struct trace_array *tr)
2902 {
2903 	struct trace_event_file *file, *next;
2904 
2905 	list_for_each_entry_safe(file, next, &tr->events, list)
2906 		remove_event_file_dir(file);
2907 }
2908 
2909 static void __add_event_to_tracers(struct trace_event_call *call)
2910 {
2911 	struct trace_array *tr;
2912 
2913 	list_for_each_entry(tr, &ftrace_trace_arrays, list)
2914 		__trace_add_new_event(call, tr);
2915 }
2916 
2917 extern struct trace_event_call *__start_ftrace_events[];
2918 extern struct trace_event_call *__stop_ftrace_events[];
2919 
2920 static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
2921 
2922 static __init int setup_trace_event(char *str)
2923 {
2924 	strlcpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
2925 	ring_buffer_expanded = true;
2926 	tracing_selftest_disabled = true;
2927 
2928 	return 1;
2929 }
2930 __setup("trace_event=", setup_trace_event);
2931 
2932 /* Expects to have event_mutex held when called */
2933 static int
2934 create_event_toplevel_files(struct dentry *parent, struct trace_array *tr)
2935 {
2936 	struct dentry *d_events;
2937 	struct dentry *entry;
2938 
2939 	entry = tracefs_create_file("set_event", 0644, parent,
2940 				    tr, &ftrace_set_event_fops);
2941 	if (!entry) {
2942 		pr_warn("Could not create tracefs 'set_event' entry\n");
2943 		return -ENOMEM;
2944 	}
2945 
2946 	d_events = tracefs_create_dir("events", parent);
2947 	if (!d_events) {
2948 		pr_warn("Could not create tracefs 'events' directory\n");
2949 		return -ENOMEM;
2950 	}
2951 
2952 	entry = tracefs_create_file("set_event_pid", 0644, parent,
2953 				    tr, &ftrace_set_event_pid_fops);
2954 
2955 	/* ring buffer internal formats */
2956 	trace_create_file("header_page", 0444, d_events,
2957 			  ring_buffer_print_page_header,
2958 			  &ftrace_show_header_fops);
2959 
2960 	trace_create_file("header_event", 0444, d_events,
2961 			  ring_buffer_print_entry_header,
2962 			  &ftrace_show_header_fops);
2963 
2964 	trace_create_file("enable", 0644, d_events,
2965 			  tr, &ftrace_tr_enable_fops);
2966 
2967 	tr->event_dir = d_events;
2968 
2969 	return 0;
2970 }
2971 
2972 /**
2973  * event_trace_add_tracer - add a instance of a trace_array to events
2974  * @parent: The parent dentry to place the files/directories for events in
2975  * @tr: The trace array associated with these events
2976  *
2977  * When a new instance is created, it needs to set up its events
2978  * directory, as well as other files associated with events. It also
2979  * creates the event hierachry in the @parent/events directory.
2980  *
2981  * Returns 0 on success.
2982  */
2983 int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr)
2984 {
2985 	int ret;
2986 
2987 	mutex_lock(&event_mutex);
2988 
2989 	ret = create_event_toplevel_files(parent, tr);
2990 	if (ret)
2991 		goto out_unlock;
2992 
2993 	down_write(&trace_event_sem);
2994 	__trace_add_event_dirs(tr);
2995 	up_write(&trace_event_sem);
2996 
2997  out_unlock:
2998 	mutex_unlock(&event_mutex);
2999 
3000 	return ret;
3001 }
3002 
3003 /*
3004  * The top trace array already had its file descriptors created.
3005  * Now the files themselves need to be created.
3006  */
3007 static __init int
3008 early_event_add_tracer(struct dentry *parent, struct trace_array *tr)
3009 {
3010 	int ret;
3011 
3012 	mutex_lock(&event_mutex);
3013 
3014 	ret = create_event_toplevel_files(parent, tr);
3015 	if (ret)
3016 		goto out_unlock;
3017 
3018 	down_write(&trace_event_sem);
3019 	__trace_early_add_event_dirs(tr);
3020 	up_write(&trace_event_sem);
3021 
3022  out_unlock:
3023 	mutex_unlock(&event_mutex);
3024 
3025 	return ret;
3026 }
3027 
3028 int event_trace_del_tracer(struct trace_array *tr)
3029 {
3030 	mutex_lock(&event_mutex);
3031 
3032 	/* Disable any event triggers and associated soft-disabled events */
3033 	clear_event_triggers(tr);
3034 
3035 	/* Clear the pid list */
3036 	__ftrace_clear_event_pids(tr);
3037 
3038 	/* Disable any running events */
3039 	__ftrace_set_clr_event_nolock(tr, NULL, NULL, NULL, 0);
3040 
3041 	/* Access to events are within rcu_read_lock_sched() */
3042 	synchronize_sched();
3043 
3044 	down_write(&trace_event_sem);
3045 	__trace_remove_event_dirs(tr);
3046 	tracefs_remove_recursive(tr->event_dir);
3047 	up_write(&trace_event_sem);
3048 
3049 	tr->event_dir = NULL;
3050 
3051 	mutex_unlock(&event_mutex);
3052 
3053 	return 0;
3054 }
3055 
3056 static __init int event_trace_memsetup(void)
3057 {
3058 	field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC);
3059 	file_cachep = KMEM_CACHE(trace_event_file, SLAB_PANIC);
3060 	return 0;
3061 }
3062 
3063 static __init void
3064 early_enable_events(struct trace_array *tr, bool disable_first)
3065 {
3066 	char *buf = bootup_event_buf;
3067 	char *token;
3068 	int ret;
3069 
3070 	while (true) {
3071 		token = strsep(&buf, ",");
3072 
3073 		if (!token)
3074 			break;
3075 
3076 		if (*token) {
3077 			/* Restarting syscalls requires that we stop them first */
3078 			if (disable_first)
3079 				ftrace_set_clr_event(tr, token, 0);
3080 
3081 			ret = ftrace_set_clr_event(tr, token, 1);
3082 			if (ret)
3083 				pr_warn("Failed to enable trace event: %s\n", token);
3084 		}
3085 
3086 		/* Put back the comma to allow this to be called again */
3087 		if (buf)
3088 			*(buf - 1) = ',';
3089 	}
3090 }
3091 
3092 static __init int event_trace_enable(void)
3093 {
3094 	struct trace_array *tr = top_trace_array();
3095 	struct trace_event_call **iter, *call;
3096 	int ret;
3097 
3098 	if (!tr)
3099 		return -ENODEV;
3100 
3101 	for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
3102 
3103 		call = *iter;
3104 		ret = event_init(call);
3105 		if (!ret)
3106 			list_add(&call->list, &ftrace_events);
3107 	}
3108 
3109 	/*
3110 	 * We need the top trace array to have a working set of trace
3111 	 * points at early init, before the debug files and directories
3112 	 * are created. Create the file entries now, and attach them
3113 	 * to the actual file dentries later.
3114 	 */
3115 	__trace_early_add_events(tr);
3116 
3117 	early_enable_events(tr, false);
3118 
3119 	trace_printk_start_comm();
3120 
3121 	register_event_cmds();
3122 
3123 	register_trigger_cmds();
3124 
3125 	return 0;
3126 }
3127 
3128 /*
3129  * event_trace_enable() is called from trace_event_init() first to
3130  * initialize events and perhaps start any events that are on the
3131  * command line. Unfortunately, there are some events that will not
3132  * start this early, like the system call tracepoints that need
3133  * to set the TIF_SYSCALL_TRACEPOINT flag of pid 1. But event_trace_enable()
3134  * is called before pid 1 starts, and this flag is never set, making
3135  * the syscall tracepoint never get reached, but the event is enabled
3136  * regardless (and not doing anything).
3137  */
3138 static __init int event_trace_enable_again(void)
3139 {
3140 	struct trace_array *tr;
3141 
3142 	tr = top_trace_array();
3143 	if (!tr)
3144 		return -ENODEV;
3145 
3146 	early_enable_events(tr, true);
3147 
3148 	return 0;
3149 }
3150 
3151 early_initcall(event_trace_enable_again);
3152 
3153 static __init int event_trace_init(void)
3154 {
3155 	struct trace_array *tr;
3156 	struct dentry *d_tracer;
3157 	struct dentry *entry;
3158 	int ret;
3159 
3160 	tr = top_trace_array();
3161 	if (!tr)
3162 		return -ENODEV;
3163 
3164 	d_tracer = tracing_init_dentry();
3165 	if (IS_ERR(d_tracer))
3166 		return 0;
3167 
3168 	entry = tracefs_create_file("available_events", 0444, d_tracer,
3169 				    tr, &ftrace_avail_fops);
3170 	if (!entry)
3171 		pr_warn("Could not create tracefs 'available_events' entry\n");
3172 
3173 	if (trace_define_generic_fields())
3174 		pr_warn("tracing: Failed to allocated generic fields");
3175 
3176 	if (trace_define_common_fields())
3177 		pr_warn("tracing: Failed to allocate common fields");
3178 
3179 	ret = early_event_add_tracer(d_tracer, tr);
3180 	if (ret)
3181 		return ret;
3182 
3183 #ifdef CONFIG_MODULES
3184 	ret = register_module_notifier(&trace_module_nb);
3185 	if (ret)
3186 		pr_warn("Failed to register trace events module notifier\n");
3187 #endif
3188 	return 0;
3189 }
3190 
3191 void __init trace_event_init(void)
3192 {
3193 	event_trace_memsetup();
3194 	init_ftrace_syscalls();
3195 	event_trace_enable();
3196 }
3197 
3198 fs_initcall(event_trace_init);
3199 
3200 #ifdef CONFIG_FTRACE_STARTUP_TEST
3201 
3202 static DEFINE_SPINLOCK(test_spinlock);
3203 static DEFINE_SPINLOCK(test_spinlock_irq);
3204 static DEFINE_MUTEX(test_mutex);
3205 
3206 static __init void test_work(struct work_struct *dummy)
3207 {
3208 	spin_lock(&test_spinlock);
3209 	spin_lock_irq(&test_spinlock_irq);
3210 	udelay(1);
3211 	spin_unlock_irq(&test_spinlock_irq);
3212 	spin_unlock(&test_spinlock);
3213 
3214 	mutex_lock(&test_mutex);
3215 	msleep(1);
3216 	mutex_unlock(&test_mutex);
3217 }
3218 
3219 static __init int event_test_thread(void *unused)
3220 {
3221 	void *test_malloc;
3222 
3223 	test_malloc = kmalloc(1234, GFP_KERNEL);
3224 	if (!test_malloc)
3225 		pr_info("failed to kmalloc\n");
3226 
3227 	schedule_on_each_cpu(test_work);
3228 
3229 	kfree(test_malloc);
3230 
3231 	set_current_state(TASK_INTERRUPTIBLE);
3232 	while (!kthread_should_stop()) {
3233 		schedule();
3234 		set_current_state(TASK_INTERRUPTIBLE);
3235 	}
3236 	__set_current_state(TASK_RUNNING);
3237 
3238 	return 0;
3239 }
3240 
3241 /*
3242  * Do various things that may trigger events.
3243  */
3244 static __init void event_test_stuff(void)
3245 {
3246 	struct task_struct *test_thread;
3247 
3248 	test_thread = kthread_run(event_test_thread, NULL, "test-events");
3249 	msleep(1);
3250 	kthread_stop(test_thread);
3251 }
3252 
3253 /*
3254  * For every trace event defined, we will test each trace point separately,
3255  * and then by groups, and finally all trace points.
3256  */
3257 static __init void event_trace_self_tests(void)
3258 {
3259 	struct trace_subsystem_dir *dir;
3260 	struct trace_event_file *file;
3261 	struct trace_event_call *call;
3262 	struct event_subsystem *system;
3263 	struct trace_array *tr;
3264 	int ret;
3265 
3266 	tr = top_trace_array();
3267 	if (!tr)
3268 		return;
3269 
3270 	pr_info("Running tests on trace events:\n");
3271 
3272 	list_for_each_entry(file, &tr->events, list) {
3273 
3274 		call = file->event_call;
3275 
3276 		/* Only test those that have a probe */
3277 		if (!call->class || !call->class->probe)
3278 			continue;
3279 
3280 /*
3281  * Testing syscall events here is pretty useless, but
3282  * we still do it if configured. But this is time consuming.
3283  * What we really need is a user thread to perform the
3284  * syscalls as we test.
3285  */
3286 #ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
3287 		if (call->class->system &&
3288 		    strcmp(call->class->system, "syscalls") == 0)
3289 			continue;
3290 #endif
3291 
3292 		pr_info("Testing event %s: ", trace_event_name(call));
3293 
3294 		/*
3295 		 * If an event is already enabled, someone is using
3296 		 * it and the self test should not be on.
3297 		 */
3298 		if (file->flags & EVENT_FILE_FL_ENABLED) {
3299 			pr_warn("Enabled event during self test!\n");
3300 			WARN_ON_ONCE(1);
3301 			continue;
3302 		}
3303 
3304 		ftrace_event_enable_disable(file, 1);
3305 		event_test_stuff();
3306 		ftrace_event_enable_disable(file, 0);
3307 
3308 		pr_cont("OK\n");
3309 	}
3310 
3311 	/* Now test at the sub system level */
3312 
3313 	pr_info("Running tests on trace event systems:\n");
3314 
3315 	list_for_each_entry(dir, &tr->systems, list) {
3316 
3317 		system = dir->subsystem;
3318 
3319 		/* the ftrace system is special, skip it */
3320 		if (strcmp(system->name, "ftrace") == 0)
3321 			continue;
3322 
3323 		pr_info("Testing event system %s: ", system->name);
3324 
3325 		ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1);
3326 		if (WARN_ON_ONCE(ret)) {
3327 			pr_warn("error enabling system %s\n",
3328 				system->name);
3329 			continue;
3330 		}
3331 
3332 		event_test_stuff();
3333 
3334 		ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0);
3335 		if (WARN_ON_ONCE(ret)) {
3336 			pr_warn("error disabling system %s\n",
3337 				system->name);
3338 			continue;
3339 		}
3340 
3341 		pr_cont("OK\n");
3342 	}
3343 
3344 	/* Test with all events enabled */
3345 
3346 	pr_info("Running tests on all trace events:\n");
3347 	pr_info("Testing all events: ");
3348 
3349 	ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1);
3350 	if (WARN_ON_ONCE(ret)) {
3351 		pr_warn("error enabling all events\n");
3352 		return;
3353 	}
3354 
3355 	event_test_stuff();
3356 
3357 	/* reset sysname */
3358 	ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
3359 	if (WARN_ON_ONCE(ret)) {
3360 		pr_warn("error disabling all events\n");
3361 		return;
3362 	}
3363 
3364 	pr_cont("OK\n");
3365 }
3366 
3367 #ifdef CONFIG_FUNCTION_TRACER
3368 
3369 static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
3370 
3371 static struct trace_array *event_tr;
3372 
3373 static void __init
3374 function_test_events_call(unsigned long ip, unsigned long parent_ip,
3375 			  struct ftrace_ops *op, struct pt_regs *pt_regs)
3376 {
3377 	struct ring_buffer_event *event;
3378 	struct ring_buffer *buffer;
3379 	struct ftrace_entry *entry;
3380 	unsigned long flags;
3381 	long disabled;
3382 	int cpu;
3383 	int pc;
3384 
3385 	pc = preempt_count();
3386 	preempt_disable_notrace();
3387 	cpu = raw_smp_processor_id();
3388 	disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
3389 
3390 	if (disabled != 1)
3391 		goto out;
3392 
3393 	local_save_flags(flags);
3394 
3395 	event = trace_current_buffer_lock_reserve(&buffer,
3396 						  TRACE_FN, sizeof(*entry),
3397 						  flags, pc);
3398 	if (!event)
3399 		goto out;
3400 	entry	= ring_buffer_event_data(event);
3401 	entry->ip			= ip;
3402 	entry->parent_ip		= parent_ip;
3403 
3404 	trace_buffer_unlock_commit(event_tr, buffer, event, flags, pc);
3405 
3406  out:
3407 	atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
3408 	preempt_enable_notrace();
3409 }
3410 
3411 static struct ftrace_ops trace_ops __initdata  =
3412 {
3413 	.func = function_test_events_call,
3414 	.flags = FTRACE_OPS_FL_RECURSION_SAFE,
3415 };
3416 
3417 static __init void event_trace_self_test_with_function(void)
3418 {
3419 	int ret;
3420 	event_tr = top_trace_array();
3421 	if (WARN_ON(!event_tr))
3422 		return;
3423 	ret = register_ftrace_function(&trace_ops);
3424 	if (WARN_ON(ret < 0)) {
3425 		pr_info("Failed to enable function tracer for event tests\n");
3426 		return;
3427 	}
3428 	pr_info("Running tests again, along with the function tracer\n");
3429 	event_trace_self_tests();
3430 	unregister_ftrace_function(&trace_ops);
3431 }
3432 #else
3433 static __init void event_trace_self_test_with_function(void)
3434 {
3435 }
3436 #endif
3437 
3438 static __init int event_trace_self_tests_init(void)
3439 {
3440 	if (!tracing_selftest_disabled) {
3441 		event_trace_self_tests();
3442 		event_trace_self_test_with_function();
3443 	}
3444 
3445 	return 0;
3446 }
3447 
3448 late_initcall(event_trace_self_tests_init);
3449 
3450 #endif
3451