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