projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
tracing: add TRACE_EVENT_MAP
[deliverable/linux.git] / kernel / trace / trace_events.c
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
33 DEFINE_MUTEX(event_mutex);
34
35 LIST_HEAD(ftrace_events);
36 static LIST_HEAD(ftrace_generic_fields);
37 static LIST_HEAD(ftrace_common_fields);
38
39 #define GFP_TRACE (GFP_KERNEL | __GFP_ZERO)
40
41 static struct kmem_cache *field_cachep;
42 static struct kmem_cache *file_cachep;
43
44 static inline int system_refcount(struct event_subsystem *system)
45 {
46 return system->ref_count;
47 }
48
49 static int system_refcount_inc(struct event_subsystem *system)
50 {
51 return system->ref_count++;
52 }
53
54 static 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
72 static struct list_head *
73 trace_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
80 static 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
93 struct ftrace_event_field *
94 trace_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
111 static 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
138 int 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 }
151 EXPORT_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
169 static 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
181 static 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
194 static 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 */
210 int 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
224 int 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 }
234 EXPORT_SYMBOL_GPL(trace_event_raw_init);
235
236 bool 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 }
250 EXPORT_SYMBOL_GPL(trace_event_ignore_this_pid);
251
252 void *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 }
284 EXPORT_SYMBOL_GPL(trace_event_buffer_reserve);
285
286 static DEFINE_SPINLOCK(tracepoint_iter_lock);
287
288 static void output_printk(struct trace_event_buffer *fbuffer)
289 {
290 struct trace_event_call *event_call;
291 struct trace_event *event;
292 unsigned long flags;
293 struct trace_iterator *iter = tracepoint_print_iter;
294
295 if (!iter)
296 return;
297
298 event_call = fbuffer->trace_file->event_call;
299 if (!event_call || !event_call->event.funcs ||
300 !event_call->event.funcs->trace)
301 return;
302
303 event = &fbuffer->trace_file->event_call->event;
304
305 spin_lock_irqsave(&tracepoint_iter_lock, flags);
306 trace_seq_init(&iter->seq);
307 iter->ent = fbuffer->entry;
308 event_call->event.funcs->trace(iter, 0, event);
309 trace_seq_putc(&iter->seq, 0);
310 printk("%s", iter->seq.buffer);
311
312 spin_unlock_irqrestore(&tracepoint_iter_lock, flags);
313 }
314
315 void trace_event_buffer_commit(struct trace_event_buffer *fbuffer)
316 {
317 if (tracepoint_printk)
318 output_printk(fbuffer);
319
320 event_trigger_unlock_commit(fbuffer->trace_file, fbuffer->buffer,
321 fbuffer->event, fbuffer->entry,
322 fbuffer->flags, fbuffer->pc);
323 }
324 EXPORT_SYMBOL_GPL(trace_event_buffer_commit);
325
326 int trace_event_reg(struct trace_event_call *call,
327 enum trace_reg type, void *data)
328 {
329 struct trace_event_file *file = data;
330 struct tracepoint *tp;
331
332 WARN_ON(!(call->flags & TRACE_EVENT_FL_TRACEPOINT));
333
334 if (call->flags & TRACE_EVENT_FL_MAP)
335 tp = call->map->tp;
336 else
337 tp = call->tp;
338
339 switch (type) {
340 case TRACE_REG_REGISTER:
341 return tracepoint_probe_register(tp,
342 call->class->probe,
343 file);
344 case TRACE_REG_UNREGISTER:
345 tracepoint_probe_unregister(tp,
346 call->class->probe,
347 file);
348 return 0;
349
350 #ifdef CONFIG_PERF_EVENTS
351 case TRACE_REG_PERF_REGISTER:
352 return tracepoint_probe_register(tp,
353 call->class->perf_probe,
354 call);
355 case TRACE_REG_PERF_UNREGISTER:
356 tracepoint_probe_unregister(tp,
357 call->class->perf_probe,
358 call);
359 return 0;
360 case TRACE_REG_PERF_OPEN:
361 case TRACE_REG_PERF_CLOSE:
362 case TRACE_REG_PERF_ADD:
363 case TRACE_REG_PERF_DEL:
364 return 0;
365 #endif
366 }
367 return 0;
368 }
369 EXPORT_SYMBOL_GPL(trace_event_reg);
370
371 void trace_event_enable_cmd_record(bool enable)
372 {
373 struct trace_event_file *file;
374 struct trace_array *tr;
375
376 mutex_lock(&event_mutex);
377 do_for_each_event_file(tr, file) {
378
379 if (!(file->flags & EVENT_FILE_FL_ENABLED))
380 continue;
381
382 if (enable) {
383 tracing_start_cmdline_record();
384 set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
385 } else {
386 tracing_stop_cmdline_record();
387 clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
388 }
389 } while_for_each_event_file();
390 mutex_unlock(&event_mutex);
391 }
392
393 static int __ftrace_event_enable_disable(struct trace_event_file *file,
394 int enable, int soft_disable)
395 {
396 struct trace_event_call *call = file->event_call;
397 struct trace_array *tr = file->tr;
398 unsigned long file_flags = file->flags;
399 int ret = 0;
400 int disable;
401
402 switch (enable) {
403 case 0:
404 /*
405 * When soft_disable is set and enable is cleared, the sm_ref
406 * reference counter is decremented. If it reaches 0, we want
407 * to clear the SOFT_DISABLED flag but leave the event in the
408 * state that it was. That is, if the event was enabled and
409 * SOFT_DISABLED isn't set, then do nothing. But if SOFT_DISABLED
410 * is set we do not want the event to be enabled before we
411 * clear the bit.
412 *
413 * When soft_disable is not set but the SOFT_MODE flag is,
414 * we do nothing. Do not disable the tracepoint, otherwise
415 * "soft enable"s (clearing the SOFT_DISABLED bit) wont work.
416 */
417 if (soft_disable) {
418 if (atomic_dec_return(&file->sm_ref) > 0)
419 break;
420 disable = file->flags & EVENT_FILE_FL_SOFT_DISABLED;
421 clear_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
422 } else
423 disable = !(file->flags & EVENT_FILE_FL_SOFT_MODE);
424
425 if (disable && (file->flags & EVENT_FILE_FL_ENABLED)) {
426 clear_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
427 if (file->flags & EVENT_FILE_FL_RECORDED_CMD) {
428 tracing_stop_cmdline_record();
429 clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
430 }
431 call->class->reg(call, TRACE_REG_UNREGISTER, file);
432 }
433 /* If in SOFT_MODE, just set the SOFT_DISABLE_BIT, else clear it */
434 if (file->flags & EVENT_FILE_FL_SOFT_MODE)
435 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
436 else
437 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
438 break;
439 case 1:
440 /*
441 * When soft_disable is set and enable is set, we want to
442 * register the tracepoint for the event, but leave the event
443 * as is. That means, if the event was already enabled, we do
444 * nothing (but set SOFT_MODE). If the event is disabled, we
445 * set SOFT_DISABLED before enabling the event tracepoint, so
446 * it still seems to be disabled.
447 */
448 if (!soft_disable)
449 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
450 else {
451 if (atomic_inc_return(&file->sm_ref) > 1)
452 break;
453 set_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
454 }
455
456 if (!(file->flags & EVENT_FILE_FL_ENABLED)) {
457
458 /* Keep the event disabled, when going to SOFT_MODE. */
459 if (soft_disable)
460 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
461
462 if (tr->trace_flags & TRACE_ITER_RECORD_CMD) {
463 tracing_start_cmdline_record();
464 set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
465 }
466 ret = call->class->reg(call, TRACE_REG_REGISTER, file);
467 if (ret) {
468 tracing_stop_cmdline_record();
469 pr_info("event trace: Could not enable event "
470 "%s\n", trace_event_name(call));
471 break;
472 }
473 set_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
474
475 /* WAS_ENABLED gets set but never cleared. */
476 call->flags |= TRACE_EVENT_FL_WAS_ENABLED;
477 }
478 break;
479 }
480
481 /* Enable or disable use of trace_buffered_event */
482 if ((file_flags & EVENT_FILE_FL_SOFT_DISABLED) !=
483 (file->flags & EVENT_FILE_FL_SOFT_DISABLED)) {
484 if (file->flags & EVENT_FILE_FL_SOFT_DISABLED)
485 trace_buffered_event_enable();
486 else
487 trace_buffered_event_disable();
488 }
489
490 return ret;
491 }
492
493 int trace_event_enable_disable(struct trace_event_file *file,
494 int enable, int soft_disable)
495 {
496 return __ftrace_event_enable_disable(file, enable, soft_disable);
497 }
498
499 static int ftrace_event_enable_disable(struct trace_event_file *file,
500 int enable)
501 {
502 return __ftrace_event_enable_disable(file, enable, 0);
503 }
504
505 static void ftrace_clear_events(struct trace_array *tr)
506 {
507 struct trace_event_file *file;
508
509 mutex_lock(&event_mutex);
510 list_for_each_entry(file, &tr->events, list) {
511 ftrace_event_enable_disable(file, 0);
512 }
513 mutex_unlock(&event_mutex);
514 }
515
516 static void
517 event_filter_pid_sched_process_exit(void *data, struct task_struct *task)
518 {
519 struct trace_pid_list *pid_list;
520 struct trace_array *tr = data;
521
522 pid_list = rcu_dereference_sched(tr->filtered_pids);
523 trace_filter_add_remove_task(pid_list, NULL, task);
524 }
525
526 static void
527 event_filter_pid_sched_process_fork(void *data,
528 struct task_struct *self,
529 struct task_struct *task)
530 {
531 struct trace_pid_list *pid_list;
532 struct trace_array *tr = data;
533
534 pid_list = rcu_dereference_sched(tr->filtered_pids);
535 trace_filter_add_remove_task(pid_list, self, task);
536 }
537
538 void trace_event_follow_fork(struct trace_array *tr, bool enable)
539 {
540 if (enable) {
541 register_trace_prio_sched_process_fork(event_filter_pid_sched_process_fork,
542 tr, INT_MIN);
543 register_trace_prio_sched_process_exit(event_filter_pid_sched_process_exit,
544 tr, INT_MAX);
545 } else {
546 unregister_trace_sched_process_fork(event_filter_pid_sched_process_fork,
547 tr);
548 unregister_trace_sched_process_exit(event_filter_pid_sched_process_exit,
549 tr);
550 }
551 }
552
553 static void
554 event_filter_pid_sched_switch_probe_pre(void *data, bool preempt,
555 struct task_struct *prev, struct task_struct *next)
556 {
557 struct trace_array *tr = data;
558 struct trace_pid_list *pid_list;
559
560 pid_list = rcu_dereference_sched(tr->filtered_pids);
561
562 this_cpu_write(tr->trace_buffer.data->ignore_pid,
563 trace_ignore_this_task(pid_list, prev) &&
564 trace_ignore_this_task(pid_list, next));
565 }
566
567 static void
568 event_filter_pid_sched_switch_probe_post(void *data, bool preempt,
569 struct task_struct *prev, struct task_struct *next)
570 {
571 struct trace_array *tr = data;
572 struct trace_pid_list *pid_list;
573
574 pid_list = rcu_dereference_sched(tr->filtered_pids);
575
576 this_cpu_write(tr->trace_buffer.data->ignore_pid,
577 trace_ignore_this_task(pid_list, next));
578 }
579
580 static void
581 event_filter_pid_sched_wakeup_probe_pre(void *data, struct task_struct *task)
582 {
583 struct trace_array *tr = data;
584 struct trace_pid_list *pid_list;
585
586 /* Nothing to do if we are already tracing */
587 if (!this_cpu_read(tr->trace_buffer.data->ignore_pid))
588 return;
589
590 pid_list = rcu_dereference_sched(tr->filtered_pids);
591
592 this_cpu_write(tr->trace_buffer.data->ignore_pid,
593 trace_ignore_this_task(pid_list, task));
594 }
595
596 static void
597 event_filter_pid_sched_wakeup_probe_post(void *data, struct task_struct *task)
598 {
599 struct trace_array *tr = data;
600 struct trace_pid_list *pid_list;
601
602 /* Nothing to do if we are not tracing */
603 if (this_cpu_read(tr->trace_buffer.data->ignore_pid))
604 return;
605
606 pid_list = rcu_dereference_sched(tr->filtered_pids);
607
608 /* Set tracing if current is enabled */
609 this_cpu_write(tr->trace_buffer.data->ignore_pid,
610 trace_ignore_this_task(pid_list, current));
611 }
612
613 static void __ftrace_clear_event_pids(struct trace_array *tr)
614 {
615 struct trace_pid_list *pid_list;
616 struct trace_event_file *file;
617 int cpu;
618
619 pid_list = rcu_dereference_protected(tr->filtered_pids,
620 lockdep_is_held(&event_mutex));
621 if (!pid_list)
622 return;
623
624 unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_pre, tr);
625 unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_post, tr);
626
627 unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre, tr);
628 unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_post, tr);
629
630 unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre, tr);
631 unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post, tr);
632
633 unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_pre, tr);
634 unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_post, tr);
635
636 list_for_each_entry(file, &tr->events, list) {
637 clear_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
638 }
639
640 for_each_possible_cpu(cpu)
641 per_cpu_ptr(tr->trace_buffer.data, cpu)->ignore_pid = false;
642
643 rcu_assign_pointer(tr->filtered_pids, NULL);
644
645 /* Wait till all users are no longer using pid filtering */
646 synchronize_sched();
647
648 trace_free_pid_list(pid_list);
649 }
650
651 static void ftrace_clear_event_pids(struct trace_array *tr)
652 {
653 mutex_lock(&event_mutex);
654 __ftrace_clear_event_pids(tr);
655 mutex_unlock(&event_mutex);
656 }
657
658 static void __put_system(struct event_subsystem *system)
659 {
660 struct event_filter *filter = system->filter;
661
662 WARN_ON_ONCE(system_refcount(system) == 0);
663 if (system_refcount_dec(system))
664 return;
665
666 list_del(&system->list);
667
668 if (filter) {
669 kfree(filter->filter_string);
670 kfree(filter);
671 }
672 kfree_const(system->name);
673 kfree(system);
674 }
675
676 static void __get_system(struct event_subsystem *system)
677 {
678 WARN_ON_ONCE(system_refcount(system) == 0);
679 system_refcount_inc(system);
680 }
681
682 static void __get_system_dir(struct trace_subsystem_dir *dir)
683 {
684 WARN_ON_ONCE(dir->ref_count == 0);
685 dir->ref_count++;
686 __get_system(dir->subsystem);
687 }
688
689 static void __put_system_dir(struct trace_subsystem_dir *dir)
690 {
691 WARN_ON_ONCE(dir->ref_count == 0);
692 /* If the subsystem is about to be freed, the dir must be too */
693 WARN_ON_ONCE(system_refcount(dir->subsystem) == 1 && dir->ref_count != 1);
694
695 __put_system(dir->subsystem);
696 if (!--dir->ref_count)
697 kfree(dir);
698 }
699
700 static void put_system(struct trace_subsystem_dir *dir)
701 {
702 mutex_lock(&event_mutex);
703 __put_system_dir(dir);
704 mutex_unlock(&event_mutex);
705 }
706
707 static void remove_subsystem(struct trace_subsystem_dir *dir)
708 {
709 if (!dir)
710 return;
711
712 if (!--dir->nr_events) {
713 tracefs_remove_recursive(dir->entry);
714 list_del(&dir->list);
715 __put_system_dir(dir);
716 }
717 }
718
719 static void remove_event_file_dir(struct trace_event_file *file)
720 {
721 struct dentry *dir = file->dir;
722 struct dentry *child;
723
724 if (dir) {
725 spin_lock(&dir->d_lock); /* probably unneeded */
726 list_for_each_entry(child, &dir->d_subdirs, d_child) {
727 if (d_really_is_positive(child)) /* probably unneeded */
728 d_inode(child)->i_private = NULL;
729 }
730 spin_unlock(&dir->d_lock);
731
732 tracefs_remove_recursive(dir);
733 }
734
735 list_del(&file->list);
736 remove_subsystem(file->system);
737 free_event_filter(file->filter);
738 kmem_cache_free(file_cachep, file);
739 }
740
741 /*
742 * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
743 */
744 static int
745 __ftrace_set_clr_event_nolock(struct trace_array *tr, const char *match,
746 const char *sub, const char *event, int set)
747 {
748 struct trace_event_file *file;
749 struct trace_event_call *call;
750 const char *name;
751 int ret = -EINVAL;
752
753 list_for_each_entry(file, &tr->events, list) {
754
755 call = file->event_call;
756 name = trace_event_name(call);
757
758 if (!name || !call->class || !call->class->reg)
759 continue;
760
761 if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
762 continue;
763
764 if (match &&
765 strcmp(match, name) != 0 &&
766 strcmp(match, call->class->system) != 0)
767 continue;
768
769 if (sub && strcmp(sub, call->class->system) != 0)
770 continue;
771
772 if (event && strcmp(event, name) != 0)
773 continue;
774
775 ftrace_event_enable_disable(file, set);
776
777 ret = 0;
778 }
779
780 return ret;
781 }
782
783 static int __ftrace_set_clr_event(struct trace_array *tr, const char *match,
784 const char *sub, const char *event, int set)
785 {
786 int ret;
787
788 mutex_lock(&event_mutex);
789 ret = __ftrace_set_clr_event_nolock(tr, match, sub, event, set);
790 mutex_unlock(&event_mutex);
791
792 return ret;
793 }
794
795 static int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set)
796 {
797 char *event = NULL, *sub = NULL, *match;
798 int ret;
799
800 /*
801 * The buf format can be <subsystem>:<event-name>
802 * *:<event-name> means any event by that name.
803 * :<event-name> is the same.
804 *
805 * <subsystem>:* means all events in that subsystem
806 * <subsystem>: means the same.
807 *
808 * <name> (no ':') means all events in a subsystem with
809 * the name <name> or any event that matches <name>
810 */
811
812 match = strsep(&buf, ":");
813 if (buf) {
814 sub = match;
815 event = buf;
816 match = NULL;
817
818 if (!strlen(sub) || strcmp(sub, "*") == 0)
819 sub = NULL;
820 if (!strlen(event) || strcmp(event, "*") == 0)
821 event = NULL;
822 }
823
824 ret = __ftrace_set_clr_event(tr, match, sub, event, set);
825
826 /* Put back the colon to allow this to be called again */
827 if (buf)
828 *(buf - 1) = ':';
829
830 return ret;
831 }
832
833 /**
834 * trace_set_clr_event - enable or disable an event
835 * @system: system name to match (NULL for any system)
836 * @event: event name to match (NULL for all events, within system)
837 * @set: 1 to enable, 0 to disable
838 *
839 * This is a way for other parts of the kernel to enable or disable
840 * event recording.
841 *
842 * Returns 0 on success, -EINVAL if the parameters do not match any
843 * registered events.
844 */
845 int trace_set_clr_event(const char *system, const char *event, int set)
846 {
847 struct trace_array *tr = top_trace_array();
848
849 if (!tr)
850 return -ENODEV;
851
852 return __ftrace_set_clr_event(tr, NULL, system, event, set);
853 }
854 EXPORT_SYMBOL_GPL(trace_set_clr_event);
855
856 /* 128 should be much more than enough */
857 #define EVENT_BUF_SIZE 127
858
859 static ssize_t
860 ftrace_event_write(struct file *file, const char __user *ubuf,
861 size_t cnt, loff_t *ppos)
862 {
863 struct trace_parser parser;
864 struct seq_file *m = file->private_data;
865 struct trace_array *tr = m->private;
866 ssize_t read, ret;
867
868 if (!cnt)
869 return 0;
870
871 ret = tracing_update_buffers();
872 if (ret < 0)
873 return ret;
874
875 if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
876 return -ENOMEM;
877
878 read = trace_get_user(&parser, ubuf, cnt, ppos);
879
880 if (read >= 0 && trace_parser_loaded((&parser))) {
881 int set = 1;
882
883 if (*parser.buffer == '!')
884 set = 0;
885
886 parser.buffer[parser.idx] = 0;
887
888 ret = ftrace_set_clr_event(tr, parser.buffer + !set, set);
889 if (ret)
890 goto out_put;
891 }
892
893 ret = read;
894
895 out_put:
896 trace_parser_put(&parser);
897
898 return ret;
899 }
900
901 static void *
902 t_next(struct seq_file *m, void *v, loff_t *pos)
903 {
904 struct trace_event_file *file = v;
905 struct trace_event_call *call;
906 struct trace_array *tr = m->private;
907
908 (*pos)++;
909
910 list_for_each_entry_continue(file, &tr->events, list) {
911 call = file->event_call;
912 /*
913 * The ftrace subsystem is for showing formats only.
914 * They can not be enabled or disabled via the event files.
915 */
916 if (call->class && call->class->reg &&
917 !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
918 return file;
919 }
920
921 return NULL;
922 }
923
924 static void *t_start(struct seq_file *m, loff_t *pos)
925 {
926 struct trace_event_file *file;
927 struct trace_array *tr = m->private;
928 loff_t l;
929
930 mutex_lock(&event_mutex);
931
932 file = list_entry(&tr->events, struct trace_event_file, list);
933 for (l = 0; l <= *pos; ) {
934 file = t_next(m, file, &l);
935 if (!file)
936 break;
937 }
938 return file;
939 }
940
941 static void *
942 s_next(struct seq_file *m, void *v, loff_t *pos)
943 {
944 struct trace_event_file *file = v;
945 struct trace_array *tr = m->private;
946
947 (*pos)++;
948
949 list_for_each_entry_continue(file, &tr->events, list) {
950 if (file->flags & EVENT_FILE_FL_ENABLED)
951 return file;
952 }
953
954 return NULL;
955 }
956
957 static void *s_start(struct seq_file *m, loff_t *pos)
958 {
959 struct trace_event_file *file;
960 struct trace_array *tr = m->private;
961 loff_t l;
962
963 mutex_lock(&event_mutex);
964
965 file = list_entry(&tr->events, struct trace_event_file, list);
966 for (l = 0; l <= *pos; ) {
967 file = s_next(m, file, &l);
968 if (!file)
969 break;
970 }
971 return file;
972 }
973
974 static int t_show(struct seq_file *m, void *v)
975 {
976 struct trace_event_file *file = v;
977 struct trace_event_call *call = file->event_call;
978
979 if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
980 seq_printf(m, "%s:", call->class->system);
981 seq_printf(m, "%s\n", trace_event_name(call));
982
983 return 0;
984 }
985
986 static void t_stop(struct seq_file *m, void *p)
987 {
988 mutex_unlock(&event_mutex);
989 }
990
991 static void *
992 p_next(struct seq_file *m, void *v, loff_t *pos)
993 {
994 struct trace_array *tr = m->private;
995 struct trace_pid_list *pid_list = rcu_dereference_sched(tr->filtered_pids);
996
997 return trace_pid_next(pid_list, v, pos);
998 }
999
1000 static void *p_start(struct seq_file *m, loff_t *pos)
1001 __acquires(RCU)
1002 {
1003 struct trace_pid_list *pid_list;
1004 struct trace_array *tr = m->private;
1005
1006 /*
1007 * Grab the mutex, to keep calls to p_next() having the same
1008 * tr->filtered_pids as p_start() has.
1009 * If we just passed the tr->filtered_pids around, then RCU would
1010 * have been enough, but doing that makes things more complex.
1011 */
1012 mutex_lock(&event_mutex);
1013 rcu_read_lock_sched();
1014
1015 pid_list = rcu_dereference_sched(tr->filtered_pids);
1016
1017 if (!pid_list)
1018 return NULL;
1019
1020 return trace_pid_start(pid_list, pos);
1021 }
1022
1023 static void p_stop(struct seq_file *m, void *p)
1024 __releases(RCU)
1025 {
1026 rcu_read_unlock_sched();
1027 mutex_unlock(&event_mutex);
1028 }
1029
1030 static ssize_t
1031 event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1032 loff_t *ppos)
1033 {
1034 struct trace_event_file *file;
1035 unsigned long flags;
1036 char buf[4] = "0";
1037
1038 mutex_lock(&event_mutex);
1039 file = event_file_data(filp);
1040 if (likely(file))
1041 flags = file->flags;
1042 mutex_unlock(&event_mutex);
1043
1044 if (!file)
1045 return -ENODEV;
1046
1047 if (flags & EVENT_FILE_FL_ENABLED &&
1048 !(flags & EVENT_FILE_FL_SOFT_DISABLED))
1049 strcpy(buf, "1");
1050
1051 if (flags & EVENT_FILE_FL_SOFT_DISABLED ||
1052 flags & EVENT_FILE_FL_SOFT_MODE)
1053 strcat(buf, "*");
1054
1055 strcat(buf, "\n");
1056
1057 return simple_read_from_buffer(ubuf, cnt, ppos, buf, strlen(buf));
1058 }
1059
1060 static ssize_t
1061 event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1062 loff_t *ppos)
1063 {
1064 struct trace_event_file *file;
1065 unsigned long val;
1066 int ret;
1067
1068 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1069 if (ret)
1070 return ret;
1071
1072 ret = tracing_update_buffers();
1073 if (ret < 0)
1074 return ret;
1075
1076 switch (val) {
1077 case 0:
1078 case 1:
1079 ret = -ENODEV;
1080 mutex_lock(&event_mutex);
1081 file = event_file_data(filp);
1082 if (likely(file))
1083 ret = ftrace_event_enable_disable(file, val);
1084 mutex_unlock(&event_mutex);
1085 break;
1086
1087 default:
1088 return -EINVAL;
1089 }
1090
1091 *ppos += cnt;
1092
1093 return ret ? ret : cnt;
1094 }
1095
1096 static ssize_t
1097 system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1098 loff_t *ppos)
1099 {
1100 const char set_to_char[4] = { '?', '0', '1', 'X' };
1101 struct trace_subsystem_dir *dir = filp->private_data;
1102 struct event_subsystem *system = dir->subsystem;
1103 struct trace_event_call *call;
1104 struct trace_event_file *file;
1105 struct trace_array *tr = dir->tr;
1106 char buf[2];
1107 int set = 0;
1108 int ret;
1109
1110 mutex_lock(&event_mutex);
1111 list_for_each_entry(file, &tr->events, list) {
1112 call = file->event_call;
1113 if (!trace_event_name(call) || !call->class || !call->class->reg)
1114 continue;
1115
1116 if (system && strcmp(call->class->system, system->name) != 0)
1117 continue;
1118
1119 /*
1120 * We need to find out if all the events are set
1121 * or if all events or cleared, or if we have
1122 * a mixture.
1123 */
1124 set |= (1 << !!(file->flags & EVENT_FILE_FL_ENABLED));
1125
1126 /*
1127 * If we have a mixture, no need to look further.
1128 */
1129 if (set == 3)
1130 break;
1131 }
1132 mutex_unlock(&event_mutex);
1133
1134 buf[0] = set_to_char[set];
1135 buf[1] = '\n';
1136
1137 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
1138
1139 return ret;
1140 }
1141
1142 static ssize_t
1143 system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1144 loff_t *ppos)
1145 {
1146 struct trace_subsystem_dir *dir = filp->private_data;
1147 struct event_subsystem *system = dir->subsystem;
1148 const char *name = NULL;
1149 unsigned long val;
1150 ssize_t ret;
1151
1152 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1153 if (ret)
1154 return ret;
1155
1156 ret = tracing_update_buffers();
1157 if (ret < 0)
1158 return ret;
1159
1160 if (val != 0 && val != 1)
1161 return -EINVAL;
1162
1163 /*
1164 * Opening of "enable" adds a ref count to system,
1165 * so the name is safe to use.
1166 */
1167 if (system)
1168 name = system->name;
1169
1170 ret = __ftrace_set_clr_event(dir->tr, NULL, name, NULL, val);
1171 if (ret)
1172 goto out;
1173
1174 ret = cnt;
1175
1176 out:
1177 *ppos += cnt;
1178
1179 return ret;
1180 }
1181
1182 enum {
1183 FORMAT_HEADER = 1,
1184 FORMAT_FIELD_SEPERATOR = 2,
1185 FORMAT_PRINTFMT = 3,
1186 };
1187
1188 static void *f_next(struct seq_file *m, void *v, loff_t *pos)
1189 {
1190 struct trace_event_call *call = event_file_data(m->private);
1191 struct list_head *common_head = &ftrace_common_fields;
1192 struct list_head *head = trace_get_fields(call);
1193 struct list_head *node = v;
1194
1195 (*pos)++;
1196
1197 switch ((unsigned long)v) {
1198 case FORMAT_HEADER:
1199 node = common_head;
1200 break;
1201
1202 case FORMAT_FIELD_SEPERATOR:
1203 node = head;
1204 break;
1205
1206 case FORMAT_PRINTFMT:
1207 /* all done */
1208 return NULL;
1209 }
1210
1211 node = node->prev;
1212 if (node == common_head)
1213 return (void *)FORMAT_FIELD_SEPERATOR;
1214 else if (node == head)
1215 return (void *)FORMAT_PRINTFMT;
1216 else
1217 return node;
1218 }
1219
1220 static int f_show(struct seq_file *m, void *v)
1221 {
1222 struct trace_event_call *call = event_file_data(m->private);
1223 struct ftrace_event_field *field;
1224 const char *array_descriptor;
1225
1226 switch ((unsigned long)v) {
1227 case FORMAT_HEADER:
1228 seq_printf(m, "name: %s\n", trace_event_name(call));
1229 seq_printf(m, "ID: %d\n", call->event.type);
1230 seq_puts(m, "format:\n");
1231 return 0;
1232
1233 case FORMAT_FIELD_SEPERATOR:
1234 seq_putc(m, '\n');
1235 return 0;
1236
1237 case FORMAT_PRINTFMT:
1238 seq_printf(m, "\nprint fmt: %s\n",
1239 call->print_fmt);
1240 return 0;
1241 }
1242
1243 field = list_entry(v, struct ftrace_event_field, link);
1244 /*
1245 * Smartly shows the array type(except dynamic array).
1246 * Normal:
1247 * field:TYPE VAR
1248 * If TYPE := TYPE[LEN], it is shown:
1249 * field:TYPE VAR[LEN]
1250 */
1251 array_descriptor = strchr(field->type, '[');
1252
1253 if (!strncmp(field->type, "__data_loc", 10))
1254 array_descriptor = NULL;
1255
1256 if (!array_descriptor)
1257 seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1258 field->type, field->name, field->offset,
1259 field->size, !!field->is_signed);
1260 else
1261 seq_printf(m, "\tfield:%.*s %s%s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1262 (int)(array_descriptor - field->type),
1263 field->type, field->name,
1264 array_descriptor, field->offset,
1265 field->size, !!field->is_signed);
1266
1267 return 0;
1268 }
1269
1270 static void *f_start(struct seq_file *m, loff_t *pos)
1271 {
1272 void *p = (void *)FORMAT_HEADER;
1273 loff_t l = 0;
1274
1275 /* ->stop() is called even if ->start() fails */
1276 mutex_lock(&event_mutex);
1277 if (!event_file_data(m->private))
1278 return ERR_PTR(-ENODEV);
1279
1280 while (l < *pos && p)
1281 p = f_next(m, p, &l);
1282
1283 return p;
1284 }
1285
1286 static void f_stop(struct seq_file *m, void *p)
1287 {
1288 mutex_unlock(&event_mutex);
1289 }
1290
1291 static const struct seq_operations trace_format_seq_ops = {
1292 .start = f_start,
1293 .next = f_next,
1294 .stop = f_stop,
1295 .show = f_show,
1296 };
1297
1298 static int trace_format_open(struct inode *inode, struct file *file)
1299 {
1300 struct seq_file *m;
1301 int ret;
1302
1303 ret = seq_open(file, &trace_format_seq_ops);
1304 if (ret < 0)
1305 return ret;
1306
1307 m = file->private_data;
1308 m->private = file;
1309
1310 return 0;
1311 }
1312
1313 static ssize_t
1314 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1315 {
1316 int id = (long)event_file_data(filp);
1317 char buf[32];
1318 int len;
1319
1320 if (*ppos)
1321 return 0;
1322
1323 if (unlikely(!id))
1324 return -ENODEV;
1325
1326 len = sprintf(buf, "%d\n", id);
1327
1328 return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
1329 }
1330
1331 static ssize_t
1332 event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1333 loff_t *ppos)
1334 {
1335 struct trace_event_file *file;
1336 struct trace_seq *s;
1337 int r = -ENODEV;
1338
1339 if (*ppos)
1340 return 0;
1341
1342 s = kmalloc(sizeof(*s), GFP_KERNEL);
1343
1344 if (!s)
1345 return -ENOMEM;
1346
1347 trace_seq_init(s);
1348
1349 mutex_lock(&event_mutex);
1350 file = event_file_data(filp);
1351 if (file)
1352 print_event_filter(file, s);
1353 mutex_unlock(&event_mutex);
1354
1355 if (file)
1356 r = simple_read_from_buffer(ubuf, cnt, ppos,
1357 s->buffer, trace_seq_used(s));
1358
1359 kfree(s);
1360
1361 return r;
1362 }
1363
1364 static ssize_t
1365 event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1366 loff_t *ppos)
1367 {
1368 struct trace_event_file *file;
1369 char *buf;
1370 int err = -ENODEV;
1371
1372 if (cnt >= PAGE_SIZE)
1373 return -EINVAL;
1374
1375 buf = memdup_user_nul(ubuf, cnt);
1376 if (IS_ERR(buf))
1377 return PTR_ERR(buf);
1378
1379 mutex_lock(&event_mutex);
1380 file = event_file_data(filp);
1381 if (file)
1382 err = apply_event_filter(file, buf);
1383 mutex_unlock(&event_mutex);
1384
1385 kfree(buf);
1386 if (err < 0)
1387 return err;
1388
1389 *ppos += cnt;
1390
1391 return cnt;
1392 }
1393
1394 static LIST_HEAD(event_subsystems);
1395
1396 static int subsystem_open(struct inode *inode, struct file *filp)
1397 {
1398 struct event_subsystem *system = NULL;
1399 struct trace_subsystem_dir *dir = NULL; /* Initialize for gcc */
1400 struct trace_array *tr;
1401 int ret;
1402
1403 if (tracing_is_disabled())
1404 return -ENODEV;
1405
1406 /* Make sure the system still exists */
1407 mutex_lock(&trace_types_lock);
1408 mutex_lock(&event_mutex);
1409 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
1410 list_for_each_entry(dir, &tr->systems, list) {
1411 if (dir == inode->i_private) {
1412 /* Don't open systems with no events */
1413 if (dir->nr_events) {
1414 __get_system_dir(dir);
1415 system = dir->subsystem;
1416 }
1417 goto exit_loop;
1418 }
1419 }
1420 }
1421 exit_loop:
1422 mutex_unlock(&event_mutex);
1423 mutex_unlock(&trace_types_lock);
1424
1425 if (!system)
1426 return -ENODEV;
1427
1428 /* Some versions of gcc think dir can be uninitialized here */
1429 WARN_ON(!dir);
1430
1431 /* Still need to increment the ref count of the system */
1432 if (trace_array_get(tr) < 0) {
1433 put_system(dir);
1434 return -ENODEV;
1435 }
1436
1437 ret = tracing_open_generic(inode, filp);
1438 if (ret < 0) {
1439 trace_array_put(tr);
1440 put_system(dir);
1441 }
1442
1443 return ret;
1444 }
1445
1446 static int system_tr_open(struct inode *inode, struct file *filp)
1447 {
1448 struct trace_subsystem_dir *dir;
1449 struct trace_array *tr = inode->i_private;
1450 int ret;
1451
1452 if (tracing_is_disabled())
1453 return -ENODEV;
1454
1455 if (trace_array_get(tr) < 0)
1456 return -ENODEV;
1457
1458 /* Make a temporary dir that has no system but points to tr */
1459 dir = kzalloc(sizeof(*dir), GFP_KERNEL);
1460 if (!dir) {
1461 trace_array_put(tr);
1462 return -ENOMEM;
1463 }
1464
1465 dir->tr = tr;
1466
1467 ret = tracing_open_generic(inode, filp);
1468 if (ret < 0) {
1469 trace_array_put(tr);
1470 kfree(dir);
1471 return ret;
1472 }
1473
1474 filp->private_data = dir;
1475
1476 return 0;
1477 }
1478
1479 static int subsystem_release(struct inode *inode, struct file *file)
1480 {
1481 struct trace_subsystem_dir *dir = file->private_data;
1482
1483 trace_array_put(dir->tr);
1484
1485 /*
1486 * If dir->subsystem is NULL, then this is a temporary
1487 * descriptor that was made for a trace_array to enable
1488 * all subsystems.
1489 */
1490 if (dir->subsystem)
1491 put_system(dir);
1492 else
1493 kfree(dir);
1494
1495 return 0;
1496 }
1497
1498 static ssize_t
1499 subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1500 loff_t *ppos)
1501 {
1502 struct trace_subsystem_dir *dir = filp->private_data;
1503 struct event_subsystem *system = dir->subsystem;
1504 struct trace_seq *s;
1505 int r;
1506
1507 if (*ppos)
1508 return 0;
1509
1510 s = kmalloc(sizeof(*s), GFP_KERNEL);
1511 if (!s)
1512 return -ENOMEM;
1513
1514 trace_seq_init(s);
1515
1516 print_subsystem_event_filter(system, s);
1517 r = simple_read_from_buffer(ubuf, cnt, ppos,
1518 s->buffer, trace_seq_used(s));
1519
1520 kfree(s);
1521
1522 return r;
1523 }
1524
1525 static ssize_t
1526 subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1527 loff_t *ppos)
1528 {
1529 struct trace_subsystem_dir *dir = filp->private_data;
1530 char *buf;
1531 int err;
1532
1533 if (cnt >= PAGE_SIZE)
1534 return -EINVAL;
1535
1536 buf = memdup_user_nul(ubuf, cnt);
1537 if (IS_ERR(buf))
1538 return PTR_ERR(buf);
1539
1540 err = apply_subsystem_event_filter(dir, buf);
1541 kfree(buf);
1542 if (err < 0)
1543 return err;
1544
1545 *ppos += cnt;
1546
1547 return cnt;
1548 }
1549
1550 static ssize_t
1551 show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1552 {
1553 int (*func)(struct trace_seq *s) = filp->private_data;
1554 struct trace_seq *s;
1555 int r;
1556
1557 if (*ppos)
1558 return 0;
1559
1560 s = kmalloc(sizeof(*s), GFP_KERNEL);
1561 if (!s)
1562 return -ENOMEM;
1563
1564 trace_seq_init(s);
1565
1566 func(s);
1567 r = simple_read_from_buffer(ubuf, cnt, ppos,
1568 s->buffer, trace_seq_used(s));
1569
1570 kfree(s);
1571
1572 return r;
1573 }
1574
1575 static void ignore_task_cpu(void *data)
1576 {
1577 struct trace_array *tr = data;
1578 struct trace_pid_list *pid_list;
1579
1580 /*
1581 * This function is called by on_each_cpu() while the
1582 * event_mutex is held.
1583 */
1584 pid_list = rcu_dereference_protected(tr->filtered_pids,
1585 mutex_is_locked(&event_mutex));
1586
1587 this_cpu_write(tr->trace_buffer.data->ignore_pid,
1588 trace_ignore_this_task(pid_list, current));
1589 }
1590
1591 static ssize_t
1592 ftrace_event_pid_write(struct file *filp, const char __user *ubuf,
1593 size_t cnt, loff_t *ppos)
1594 {
1595 struct seq_file *m = filp->private_data;
1596 struct trace_array *tr = m->private;
1597 struct trace_pid_list *filtered_pids = NULL;
1598 struct trace_pid_list *pid_list;
1599 struct trace_event_file *file;
1600 ssize_t ret;
1601
1602 if (!cnt)
1603 return 0;
1604
1605 ret = tracing_update_buffers();
1606 if (ret < 0)
1607 return ret;
1608
1609 mutex_lock(&event_mutex);
1610
1611 filtered_pids = rcu_dereference_protected(tr->filtered_pids,
1612 lockdep_is_held(&event_mutex));
1613
1614 ret = trace_pid_write(filtered_pids, &pid_list, ubuf, cnt);
1615 if (ret < 0)
1616 goto out;
1617
1618 rcu_assign_pointer(tr->filtered_pids, pid_list);
1619
1620 list_for_each_entry(file, &tr->events, list) {
1621 set_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
1622 }
1623
1624 if (filtered_pids) {
1625 synchronize_sched();
1626 trace_free_pid_list(filtered_pids);
1627 } else if (pid_list) {
1628 /*
1629 * Register a probe that is called before all other probes
1630 * to set ignore_pid if next or prev do not match.
1631 * Register a probe this is called after all other probes
1632 * to only keep ignore_pid set if next pid matches.
1633 */
1634 register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_pre,
1635 tr, INT_MAX);
1636 register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_post,
1637 tr, 0);
1638
1639 register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre,
1640 tr, INT_MAX);
1641 register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_post,
1642 tr, 0);
1643
1644 register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre,
1645 tr, INT_MAX);
1646 register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post,
1647 tr, 0);
1648
1649 register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_pre,
1650 tr, INT_MAX);
1651 register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_post,
1652 tr, 0);
1653 }
1654
1655 /*
1656 * Ignoring of pids is done at task switch. But we have to
1657 * check for those tasks that are currently running.
1658 * Always do this in case a pid was appended or removed.
1659 */
1660 on_each_cpu(ignore_task_cpu, tr, 1);
1661
1662 out:
1663 mutex_unlock(&event_mutex);
1664
1665 if (ret > 0)
1666 *ppos += ret;
1667
1668 return ret;
1669 }
1670
1671 static int ftrace_event_avail_open(struct inode *inode, struct file *file);
1672 static int ftrace_event_set_open(struct inode *inode, struct file *file);
1673 static int ftrace_event_set_pid_open(struct inode *inode, struct file *file);
1674 static int ftrace_event_release(struct inode *inode, struct file *file);
1675
1676 static const struct seq_operations show_event_seq_ops = {
1677 .start = t_start,
1678 .next = t_next,
1679 .show = t_show,
1680 .stop = t_stop,
1681 };
1682
1683 static const struct seq_operations show_set_event_seq_ops = {
1684 .start = s_start,
1685 .next = s_next,
1686 .show = t_show,
1687 .stop = t_stop,
1688 };
1689
1690 static const struct seq_operations show_set_pid_seq_ops = {
1691 .start = p_start,
1692 .next = p_next,
1693 .show = trace_pid_show,
1694 .stop = p_stop,
1695 };
1696
1697 static const struct file_operations ftrace_avail_fops = {
1698 .open = ftrace_event_avail_open,
1699 .read = seq_read,
1700 .llseek = seq_lseek,
1701 .release = seq_release,
1702 };
1703
1704 static const struct file_operations ftrace_set_event_fops = {
1705 .open = ftrace_event_set_open,
1706 .read = seq_read,
1707 .write = ftrace_event_write,
1708 .llseek = seq_lseek,
1709 .release = ftrace_event_release,
1710 };
1711
1712 static const struct file_operations ftrace_set_event_pid_fops = {
1713 .open = ftrace_event_set_pid_open,
1714 .read = seq_read,
1715 .write = ftrace_event_pid_write,
1716 .llseek = seq_lseek,
1717 .release = ftrace_event_release,
1718 };
1719
1720 static const struct file_operations ftrace_enable_fops = {
1721 .open = tracing_open_generic,
1722 .read = event_enable_read,
1723 .write = event_enable_write,
1724 .llseek = default_llseek,
1725 };
1726
1727 static const struct file_operations ftrace_event_format_fops = {
1728 .open = trace_format_open,
1729 .read = seq_read,
1730 .llseek = seq_lseek,
1731 .release = seq_release,
1732 };
1733
1734 static const struct file_operations ftrace_event_id_fops = {
1735 .read = event_id_read,
1736 .llseek = default_llseek,
1737 };
1738
1739 static const struct file_operations ftrace_event_filter_fops = {
1740 .open = tracing_open_generic,
1741 .read = event_filter_read,
1742 .write = event_filter_write,
1743 .llseek = default_llseek,
1744 };
1745
1746 static const struct file_operations ftrace_subsystem_filter_fops = {
1747 .open = subsystem_open,
1748 .read = subsystem_filter_read,
1749 .write = subsystem_filter_write,
1750 .llseek = default_llseek,
1751 .release = subsystem_release,
1752 };
1753
1754 static const struct file_operations ftrace_system_enable_fops = {
1755 .open = subsystem_open,
1756 .read = system_enable_read,
1757 .write = system_enable_write,
1758 .llseek = default_llseek,
1759 .release = subsystem_release,
1760 };
1761
1762 static const struct file_operations ftrace_tr_enable_fops = {
1763 .open = system_tr_open,
1764 .read = system_enable_read,
1765 .write = system_enable_write,
1766 .llseek = default_llseek,
1767 .release = subsystem_release,
1768 };
1769
1770 static const struct file_operations ftrace_show_header_fops = {
1771 .open = tracing_open_generic,
1772 .read = show_header,
1773 .llseek = default_llseek,
1774 };
1775
1776 static int
1777 ftrace_event_open(struct inode *inode, struct file *file,
1778 const struct seq_operations *seq_ops)
1779 {
1780 struct seq_file *m;
1781 int ret;
1782
1783 ret = seq_open(file, seq_ops);
1784 if (ret < 0)
1785 return ret;
1786 m = file->private_data;
1787 /* copy tr over to seq ops */
1788 m->private = inode->i_private;
1789
1790 return ret;
1791 }
1792
1793 static int ftrace_event_release(struct inode *inode, struct file *file)
1794 {
1795 struct trace_array *tr = inode->i_private;
1796
1797 trace_array_put(tr);
1798
1799 return seq_release(inode, file);
1800 }
1801
1802 static int
1803 ftrace_event_avail_open(struct inode *inode, struct file *file)
1804 {
1805 const struct seq_operations *seq_ops = &show_event_seq_ops;
1806
1807 return ftrace_event_open(inode, file, seq_ops);
1808 }
1809
1810 static int
1811 ftrace_event_set_open(struct inode *inode, struct file *file)
1812 {
1813 const struct seq_operations *seq_ops = &show_set_event_seq_ops;
1814 struct trace_array *tr = inode->i_private;
1815 int ret;
1816
1817 if (trace_array_get(tr) < 0)
1818 return -ENODEV;
1819
1820 if ((file->f_mode & FMODE_WRITE) &&
1821 (file->f_flags & O_TRUNC))
1822 ftrace_clear_events(tr);
1823
1824 ret = ftrace_event_open(inode, file, seq_ops);
1825 if (ret < 0)
1826 trace_array_put(tr);
1827 return ret;
1828 }
1829
1830 static int
1831 ftrace_event_set_pid_open(struct inode *inode, struct file *file)
1832 {
1833 const struct seq_operations *seq_ops = &show_set_pid_seq_ops;
1834 struct trace_array *tr = inode->i_private;
1835 int ret;
1836
1837 if (trace_array_get(tr) < 0)
1838 return -ENODEV;
1839
1840 if ((file->f_mode & FMODE_WRITE) &&
1841 (file->f_flags & O_TRUNC))
1842 ftrace_clear_event_pids(tr);
1843
1844 ret = ftrace_event_open(inode, file, seq_ops);
1845 if (ret < 0)
1846 trace_array_put(tr);
1847 return ret;
1848 }
1849
1850 static struct event_subsystem *
1851 create_new_subsystem(const char *name)
1852 {
1853 struct event_subsystem *system;
1854
1855 /* need to create new entry */
1856 system = kmalloc(sizeof(*system), GFP_KERNEL);
1857 if (!system)
1858 return NULL;
1859
1860 system->ref_count = 1;
1861
1862 /* Only allocate if dynamic (kprobes and modules) */
1863 system->name = kstrdup_const(name, GFP_KERNEL);
1864 if (!system->name)
1865 goto out_free;
1866
1867 system->filter = NULL;
1868
1869 system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
1870 if (!system->filter)
1871 goto out_free;
1872
1873 list_add(&system->list, &event_subsystems);
1874
1875 return system;
1876
1877 out_free:
1878 kfree_const(system->name);
1879 kfree(system);
1880 return NULL;
1881 }
1882
1883 static struct dentry *
1884 event_subsystem_dir(struct trace_array *tr, const char *name,
1885 struct trace_event_file *file, struct dentry *parent)
1886 {
1887 struct trace_subsystem_dir *dir;
1888 struct event_subsystem *system;
1889 struct dentry *entry;
1890
1891 /* First see if we did not already create this dir */
1892 list_for_each_entry(dir, &tr->systems, list) {
1893 system = dir->subsystem;
1894 if (strcmp(system->name, name) == 0) {
1895 dir->nr_events++;
1896 file->system = dir;
1897 return dir->entry;
1898 }
1899 }
1900
1901 /* Now see if the system itself exists. */
1902 list_for_each_entry(system, &event_subsystems, list) {
1903 if (strcmp(system->name, name) == 0)
1904 break;
1905 }
1906 /* Reset system variable when not found */
1907 if (&system->list == &event_subsystems)
1908 system = NULL;
1909
1910 dir = kmalloc(sizeof(*dir), GFP_KERNEL);
1911 if (!dir)
1912 goto out_fail;
1913
1914 if (!system) {
1915 system = create_new_subsystem(name);
1916 if (!system)
1917 goto out_free;
1918 } else
1919 __get_system(system);
1920
1921 dir->entry = tracefs_create_dir(name, parent);
1922 if (!dir->entry) {
1923 pr_warn("Failed to create system directory %s\n", name);
1924 __put_system(system);
1925 goto out_free;
1926 }
1927
1928 dir->tr = tr;
1929 dir->ref_count = 1;
1930 dir->nr_events = 1;
1931 dir->subsystem = system;
1932 file->system = dir;
1933
1934 entry = tracefs_create_file("filter", 0644, dir->entry, dir,
1935 &ftrace_subsystem_filter_fops);
1936 if (!entry) {
1937 kfree(system->filter);
1938 system->filter = NULL;
1939 pr_warn("Could not create tracefs '%s/filter' entry\n", name);
1940 }
1941
1942 trace_create_file("enable", 0644, dir->entry, dir,
1943 &ftrace_system_enable_fops);
1944
1945 list_add(&dir->list, &tr->systems);
1946
1947 return dir->entry;
1948
1949 out_free:
1950 kfree(dir);
1951 out_fail:
1952 /* Only print this message if failed on memory allocation */
1953 if (!dir || !system)
1954 pr_warn("No memory to create event subsystem %s\n", name);
1955 return NULL;
1956 }
1957
1958 static int
1959 event_create_dir(struct dentry *parent, struct trace_event_file *file)
1960 {
1961 struct trace_event_call *call = file->event_call;
1962 struct trace_array *tr = file->tr;
1963 struct list_head *head;
1964 struct dentry *d_events;
1965 const char *name;
1966 int ret;
1967
1968 /*
1969 * If the trace point header did not define TRACE_SYSTEM
1970 * then the system would be called "TRACE_SYSTEM".
1971 */
1972 if (strcmp(call->class->system, TRACE_SYSTEM) != 0) {
1973 d_events = event_subsystem_dir(tr, call->class->system, file, parent);
1974 if (!d_events)
1975 return -ENOMEM;
1976 } else
1977 d_events = parent;
1978
1979 name = trace_event_name(call);
1980 file->dir = tracefs_create_dir(name, d_events);
1981 if (!file->dir) {
1982 pr_warn("Could not create tracefs '%s' directory\n", name);
1983 return -1;
1984 }
1985
1986 if (call->class->reg && !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
1987 trace_create_file("enable", 0644, file->dir, file,
1988 &ftrace_enable_fops);
1989
1990 #ifdef CONFIG_PERF_EVENTS
1991 if (call->event.type && call->class->reg)
1992 trace_create_file("id", 0444, file->dir,
1993 (void *)(long)call->event.type,
1994 &ftrace_event_id_fops);
1995 #endif
1996
1997 /*
1998 * Other events may have the same class. Only update
1999 * the fields if they are not already defined.
2000 */
2001 head = trace_get_fields(call);
2002 if (list_empty(head)) {
2003 ret = call->class->define_fields(call);
2004 if (ret < 0) {
2005 pr_warn("Could not initialize trace point events/%s\n",
2006 name);
2007 return -1;
2008 }
2009 }
2010 trace_create_file("filter", 0644, file->dir, file,
2011 &ftrace_event_filter_fops);
2012
2013 /*
2014 * Only event directories that can be enabled should have
2015 * triggers.
2016 */
2017 if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
2018 trace_create_file("trigger", 0644, file->dir, file,
2019 &event_trigger_fops);
2020
2021 #ifdef CONFIG_HIST_TRIGGERS
2022 trace_create_file("hist", 0444, file->dir, file,
2023 &event_hist_fops);
2024 #endif
2025 trace_create_file("format", 0444, file->dir, call,
2026 &ftrace_event_format_fops);
2027
2028 return 0;
2029 }
2030
2031 static void remove_event_from_tracers(struct trace_event_call *call)
2032 {
2033 struct trace_event_file *file;
2034 struct trace_array *tr;
2035
2036 do_for_each_event_file_safe(tr, file) {
2037 if (file->event_call != call)
2038 continue;
2039
2040 remove_event_file_dir(file);
2041 /*
2042 * The do_for_each_event_file_safe() is
2043 * a double loop. After finding the call for this
2044 * trace_array, we use break to jump to the next
2045 * trace_array.
2046 */
2047 break;
2048 } while_for_each_event_file();
2049 }
2050
2051 static void event_remove(struct trace_event_call *call)
2052 {
2053 struct trace_array *tr;
2054 struct trace_event_file *file;
2055
2056 do_for_each_event_file(tr, file) {
2057 if (file->event_call != call)
2058 continue;
2059 ftrace_event_enable_disable(file, 0);
2060 /*
2061 * The do_for_each_event_file() is
2062 * a double loop. After finding the call for this
2063 * trace_array, we use break to jump to the next
2064 * trace_array.
2065 */
2066 break;
2067 } while_for_each_event_file();
2068
2069 if (call->event.funcs)
2070 __unregister_trace_event(&call->event);
2071 remove_event_from_tracers(call);
2072 list_del(&call->list);
2073 }
2074
2075 static int event_init(struct trace_event_call *call)
2076 {
2077 int ret = 0;
2078 const char *name;
2079
2080 name = trace_event_name(call);
2081 if (WARN_ON(!name))
2082 return -EINVAL;
2083
2084 if (call->class->raw_init) {
2085 ret = call->class->raw_init(call);
2086 if (ret < 0 && ret != -ENOSYS)
2087 pr_warn("Could not initialize trace events/%s\n", name);
2088 }
2089
2090 return ret;
2091 }
2092
2093 static int
2094 __register_event(struct trace_event_call *call, struct module *mod)
2095 {
2096 int ret;
2097
2098 ret = event_init(call);
2099 if (ret < 0)
2100 return ret;
2101
2102 list_add(&call->list, &ftrace_events);
2103 call->mod = mod;
2104
2105 return 0;
2106 }
2107
2108 static char *enum_replace(char *ptr, struct trace_enum_map *map, int len)
2109 {
2110 int rlen;
2111 int elen;
2112
2113 /* Find the length of the enum value as a string */
2114 elen = snprintf(ptr, 0, "%ld", map->enum_value);
2115 /* Make sure there's enough room to replace the string with the value */
2116 if (len < elen)
2117 return NULL;
2118
2119 snprintf(ptr, elen + 1, "%ld", map->enum_value);
2120
2121 /* Get the rest of the string of ptr */
2122 rlen = strlen(ptr + len);
2123 memmove(ptr + elen, ptr + len, rlen);
2124 /* Make sure we end the new string */
2125 ptr[elen + rlen] = 0;
2126
2127 return ptr + elen;
2128 }
2129
2130 static void update_event_printk(struct trace_event_call *call,
2131 struct trace_enum_map *map)
2132 {
2133 char *ptr;
2134 int quote = 0;
2135 int len = strlen(map->enum_string);
2136
2137 for (ptr = call->print_fmt; *ptr; ptr++) {
2138 if (*ptr == '\\') {
2139 ptr++;
2140 /* paranoid */
2141 if (!*ptr)
2142 break;
2143 continue;
2144 }
2145 if (*ptr == '"') {
2146 quote ^= 1;
2147 continue;
2148 }
2149 if (quote)
2150 continue;
2151 if (isdigit(*ptr)) {
2152 /* skip numbers */
2153 do {
2154 ptr++;
2155 /* Check for alpha chars like ULL */
2156 } while (isalnum(*ptr));
2157 if (!*ptr)
2158 break;
2159 /*
2160 * A number must have some kind of delimiter after
2161 * it, and we can ignore that too.
2162 */
2163 continue;
2164 }
2165 if (isalpha(*ptr) || *ptr == '_') {
2166 if (strncmp(map->enum_string, ptr, len) == 0 &&
2167 !isalnum(ptr[len]) && ptr[len] != '_') {
2168 ptr = enum_replace(ptr, map, len);
2169 /* Hmm, enum string smaller than value */
2170 if (WARN_ON_ONCE(!ptr))
2171 return;
2172 /*
2173 * No need to decrement here, as enum_replace()
2174 * returns the pointer to the character passed
2175 * the enum, and two enums can not be placed
2176 * back to back without something in between.
2177 * We can skip that something in between.
2178 */
2179 continue;
2180 }
2181 skip_more:
2182 do {
2183 ptr++;
2184 } while (isalnum(*ptr) || *ptr == '_');
2185 if (!*ptr)
2186 break;
2187 /*
2188 * If what comes after this variable is a '.' or
2189 * '->' then we can continue to ignore that string.
2190 */
2191 if (*ptr == '.' || (ptr[0] == '-' && ptr[1] == '>')) {
2192 ptr += *ptr == '.' ? 1 : 2;
2193 if (!*ptr)
2194 break;
2195 goto skip_more;
2196 }
2197 /*
2198 * Once again, we can skip the delimiter that came
2199 * after the string.
2200 */
2201 continue;
2202 }
2203 }
2204 }
2205
2206 void trace_event_enum_update(struct trace_enum_map **map, int len)
2207 {
2208 struct trace_event_call *call, *p;
2209 const char *last_system = NULL;
2210 int last_i;
2211 int i;
2212
2213 down_write(&trace_event_sem);
2214 list_for_each_entry_safe(call, p, &ftrace_events, list) {
2215 /* events are usually grouped together with systems */
2216 if (!last_system || call->class->system != last_system) {
2217 last_i = 0;
2218 last_system = call->class->system;
2219 }
2220
2221 for (i = last_i; i < len; i++) {
2222 if (call->class->system == map[i]->system) {
2223 /* Save the first system if need be */
2224 if (!last_i)
2225 last_i = i;
2226 update_event_printk(call, map[i]);
2227 }
2228 }
2229 }
2230 up_write(&trace_event_sem);
2231 }
2232
2233 static struct trace_event_file *
2234 trace_create_new_event(struct trace_event_call *call,
2235 struct trace_array *tr)
2236 {
2237 struct trace_event_file *file;
2238
2239 file = kmem_cache_alloc(file_cachep, GFP_TRACE);
2240 if (!file)
2241 return NULL;
2242
2243 file->event_call = call;
2244 file->tr = tr;
2245 atomic_set(&file->sm_ref, 0);
2246 atomic_set(&file->tm_ref, 0);
2247 INIT_LIST_HEAD(&file->triggers);
2248 list_add(&file->list, &tr->events);
2249
2250 return file;
2251 }
2252
2253 /* Add an event to a trace directory */
2254 static int
2255 __trace_add_new_event(struct trace_event_call *call, struct trace_array *tr)
2256 {
2257 struct trace_event_file *file;
2258
2259 file = trace_create_new_event(call, tr);
2260 if (!file)
2261 return -ENOMEM;
2262
2263 return event_create_dir(tr->event_dir, file);
2264 }
2265
2266 /*
2267 * Just create a decriptor for early init. A descriptor is required
2268 * for enabling events at boot. We want to enable events before
2269 * the filesystem is initialized.
2270 */
2271 static __init int
2272 __trace_early_add_new_event(struct trace_event_call *call,
2273 struct trace_array *tr)
2274 {
2275 struct trace_event_file *file;
2276
2277 file = trace_create_new_event(call, tr);
2278 if (!file)
2279 return -ENOMEM;
2280
2281 return 0;
2282 }
2283
2284 struct ftrace_module_file_ops;
2285 static void __add_event_to_tracers(struct trace_event_call *call);
2286
2287 /* Add an additional event_call dynamically */
2288 int trace_add_event_call(struct trace_event_call *call)
2289 {
2290 int ret;
2291 mutex_lock(&trace_types_lock);
2292 mutex_lock(&event_mutex);
2293
2294 ret = __register_event(call, NULL);
2295 if (ret >= 0)
2296 __add_event_to_tracers(call);
2297
2298 mutex_unlock(&event_mutex);
2299 mutex_unlock(&trace_types_lock);
2300 return ret;
2301 }
2302
2303 /*
2304 * Must be called under locking of trace_types_lock, event_mutex and
2305 * trace_event_sem.
2306 */
2307 static void __trace_remove_event_call(struct trace_event_call *call)
2308 {
2309 event_remove(call);
2310 trace_destroy_fields(call);
2311 free_event_filter(call->filter);
2312 call->filter = NULL;
2313 }
2314
2315 static int probe_remove_event_call(struct trace_event_call *call)
2316 {
2317 struct trace_array *tr;
2318 struct trace_event_file *file;
2319
2320 #ifdef CONFIG_PERF_EVENTS
2321 if (call->perf_refcount)
2322 return -EBUSY;
2323 #endif
2324 do_for_each_event_file(tr, file) {
2325 if (file->event_call != call)
2326 continue;
2327 /*
2328 * We can't rely on ftrace_event_enable_disable(enable => 0)
2329 * we are going to do, EVENT_FILE_FL_SOFT_MODE can suppress
2330 * TRACE_REG_UNREGISTER.
2331 */
2332 if (file->flags & EVENT_FILE_FL_ENABLED)
2333 return -EBUSY;
2334 /*
2335 * The do_for_each_event_file_safe() is
2336 * a double loop. After finding the call for this
2337 * trace_array, we use break to jump to the next
2338 * trace_array.
2339 */
2340 break;
2341 } while_for_each_event_file();
2342
2343 __trace_remove_event_call(call);
2344
2345 return 0;
2346 }
2347
2348 /* Remove an event_call */
2349 int trace_remove_event_call(struct trace_event_call *call)
2350 {
2351 int ret;
2352
2353 mutex_lock(&trace_types_lock);
2354 mutex_lock(&event_mutex);
2355 down_write(&trace_event_sem);
2356 ret = probe_remove_event_call(call);
2357 up_write(&trace_event_sem);
2358 mutex_unlock(&event_mutex);
2359 mutex_unlock(&trace_types_lock);
2360
2361 return ret;
2362 }
2363
2364 #define for_each_event(event, start, end) \
2365 for (event = start; \
2366 (unsigned long)event < (unsigned long)end; \
2367 event++)
2368
2369 #ifdef CONFIG_MODULES
2370
2371 static void trace_module_add_events(struct module *mod)
2372 {
2373 struct trace_event_call **call, **start, **end;
2374
2375 if (!mod->num_trace_events)
2376 return;
2377
2378 /* Don't add infrastructure for mods without tracepoints */
2379 if (trace_module_has_bad_taint(mod)) {
2380 pr_err("%s: module has bad taint, not creating trace events\n",
2381 mod->name);
2382 return;
2383 }
2384
2385 start = mod->trace_events;
2386 end = mod->trace_events + mod->num_trace_events;
2387
2388 for_each_event(call, start, end) {
2389 __register_event(*call, mod);
2390 __add_event_to_tracers(*call);
2391 }
2392 }
2393
2394 static void trace_module_remove_events(struct module *mod)
2395 {
2396 struct trace_event_call *call, *p;
2397 bool clear_trace = false;
2398
2399 down_write(&trace_event_sem);
2400 list_for_each_entry_safe(call, p, &ftrace_events, list) {
2401 if (call->mod == mod) {
2402 if (call->flags & TRACE_EVENT_FL_WAS_ENABLED)
2403 clear_trace = true;
2404 __trace_remove_event_call(call);
2405 }
2406 }
2407 up_write(&trace_event_sem);
2408
2409 /*
2410 * It is safest to reset the ring buffer if the module being unloaded
2411 * registered any events that were used. The only worry is if
2412 * a new module gets loaded, and takes on the same id as the events
2413 * of this module. When printing out the buffer, traced events left
2414 * over from this module may be passed to the new module events and
2415 * unexpected results may occur.
2416 */
2417 if (clear_trace)
2418 tracing_reset_all_online_cpus();
2419 }
2420
2421 static int trace_module_notify(struct notifier_block *self,
2422 unsigned long val, void *data)
2423 {
2424 struct module *mod = data;
2425
2426 mutex_lock(&trace_types_lock);
2427 mutex_lock(&event_mutex);
2428 switch (val) {
2429 case MODULE_STATE_COMING:
2430 trace_module_add_events(mod);
2431 break;
2432 case MODULE_STATE_GOING:
2433 trace_module_remove_events(mod);
2434 break;
2435 }
2436 mutex_unlock(&event_mutex);
2437 mutex_unlock(&trace_types_lock);
2438
2439 return 0;
2440 }
2441
2442 static struct notifier_block trace_module_nb = {
2443 .notifier_call = trace_module_notify,
2444 .priority = 1, /* higher than trace.c module notify */
2445 };
2446 #endif /* CONFIG_MODULES */
2447
2448 /* Create a new event directory structure for a trace directory. */
2449 static void
2450 __trace_add_event_dirs(struct trace_array *tr)
2451 {
2452 struct trace_event_call *call;
2453 int ret;
2454
2455 list_for_each_entry(call, &ftrace_events, list) {
2456 ret = __trace_add_new_event(call, tr);
2457 if (ret < 0)
2458 pr_warn("Could not create directory for event %s\n",
2459 trace_event_name(call));
2460 }
2461 }
2462
2463 struct trace_event_file *
2464 find_event_file(struct trace_array *tr, const char *system, const char *event)
2465 {
2466 struct trace_event_file *file;
2467 struct trace_event_call *call;
2468 const char *name;
2469
2470 list_for_each_entry(file, &tr->events, list) {
2471
2472 call = file->event_call;
2473 name = trace_event_name(call);
2474
2475 if (!name || !call->class || !call->class->reg)
2476 continue;
2477
2478 if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
2479 continue;
2480
2481 if (strcmp(event, name) == 0 &&
2482 strcmp(system, call->class->system) == 0)
2483 return file;
2484 }
2485 return NULL;
2486 }
2487
2488 #ifdef CONFIG_DYNAMIC_FTRACE
2489
2490 /* Avoid typos */
2491 #define ENABLE_EVENT_STR "enable_event"
2492 #define DISABLE_EVENT_STR "disable_event"
2493
2494 struct event_probe_data {
2495 struct trace_event_file *file;
2496 unsigned long count;
2497 int ref;
2498 bool enable;
2499 };
2500
2501 static void
2502 event_enable_probe(unsigned long ip, unsigned long parent_ip, void **_data)
2503 {
2504 struct event_probe_data **pdata = (struct event_probe_data **)_data;
2505 struct event_probe_data *data = *pdata;
2506
2507 if (!data)
2508 return;
2509
2510 if (data->enable)
2511 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
2512 else
2513 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
2514 }
2515
2516 static void
2517 event_enable_count_probe(unsigned long ip, unsigned long parent_ip, void **_data)
2518 {
2519 struct event_probe_data **pdata = (struct event_probe_data **)_data;
2520 struct event_probe_data *data = *pdata;
2521
2522 if (!data)
2523 return;
2524
2525 if (!data->count)
2526 return;
2527
2528 /* Skip if the event is in a state we want to switch to */
2529 if (data->enable == !(data->file->flags & EVENT_FILE_FL_SOFT_DISABLED))
2530 return;
2531
2532 if (data->count != -1)
2533 (data->count)--;
2534
2535 event_enable_probe(ip, parent_ip, _data);
2536 }
2537
2538 static int
2539 event_enable_print(struct seq_file *m, unsigned long ip,
2540 struct ftrace_probe_ops *ops, void *_data)
2541 {
2542 struct event_probe_data *data = _data;
2543
2544 seq_printf(m, "%ps:", (void *)ip);
2545
2546 seq_printf(m, "%s:%s:%s",
2547 data->enable ? ENABLE_EVENT_STR : DISABLE_EVENT_STR,
2548 data->file->event_call->class->system,
2549 trace_event_name(data->file->event_call));
2550
2551 if (data->count == -1)
2552 seq_puts(m, ":unlimited\n");
2553 else
2554 seq_printf(m, ":count=%ld\n", data->count);
2555
2556 return 0;
2557 }
2558
2559 static int
2560 event_enable_init(struct ftrace_probe_ops *ops, unsigned long ip,
2561 void **_data)
2562 {
2563 struct event_probe_data **pdata = (struct event_probe_data **)_data;
2564 struct event_probe_data *data = *pdata;
2565
2566 data->ref++;
2567 return 0;
2568 }
2569
2570 static void
2571 event_enable_free(struct ftrace_probe_ops *ops, unsigned long ip,
2572 void **_data)
2573 {
2574 struct event_probe_data **pdata = (struct event_probe_data **)_data;
2575 struct event_probe_data *data = *pdata;
2576
2577 if (WARN_ON_ONCE(data->ref <= 0))
2578 return;
2579
2580 data->ref--;
2581 if (!data->ref) {
2582 /* Remove the SOFT_MODE flag */
2583 __ftrace_event_enable_disable(data->file, 0, 1);
2584 module_put(data->file->event_call->mod);
2585 kfree(data);
2586 }
2587 *pdata = NULL;
2588 }
2589
2590 static struct ftrace_probe_ops event_enable_probe_ops = {
2591 .func = event_enable_probe,
2592 .print = event_enable_print,
2593 .init = event_enable_init,
2594 .free = event_enable_free,
2595 };
2596
2597 static struct ftrace_probe_ops event_enable_count_probe_ops = {
2598 .func = event_enable_count_probe,
2599 .print = event_enable_print,
2600 .init = event_enable_init,
2601 .free = event_enable_free,
2602 };
2603
2604 static struct ftrace_probe_ops event_disable_probe_ops = {
2605 .func = event_enable_probe,
2606 .print = event_enable_print,
2607 .init = event_enable_init,
2608 .free = event_enable_free,
2609 };
2610
2611 static struct ftrace_probe_ops event_disable_count_probe_ops = {
2612 .func = event_enable_count_probe,
2613 .print = event_enable_print,
2614 .init = event_enable_init,
2615 .free = event_enable_free,
2616 };
2617
2618 static int
2619 event_enable_func(struct ftrace_hash *hash,
2620 char *glob, char *cmd, char *param, int enabled)
2621 {
2622 struct trace_array *tr = top_trace_array();
2623 struct trace_event_file *file;
2624 struct ftrace_probe_ops *ops;
2625 struct event_probe_data *data;
2626 const char *system;
2627 const char *event;
2628 char *number;
2629 bool enable;
2630 int ret;
2631
2632 if (!tr)
2633 return -ENODEV;
2634
2635 /* hash funcs only work with set_ftrace_filter */
2636 if (!enabled || !param)
2637 return -EINVAL;
2638
2639 system = strsep(&param, ":");
2640 if (!param)
2641 return -EINVAL;
2642
2643 event = strsep(&param, ":");
2644
2645 mutex_lock(&event_mutex);
2646
2647 ret = -EINVAL;
2648 file = find_event_file(tr, system, event);
2649 if (!file)
2650 goto out;
2651
2652 enable = strcmp(cmd, ENABLE_EVENT_STR) == 0;
2653
2654 if (enable)
2655 ops = param ? &event_enable_count_probe_ops : &event_enable_probe_ops;
2656 else
2657 ops = param ? &event_disable_count_probe_ops : &event_disable_probe_ops;
2658
2659 if (glob[0] == '!') {
2660 unregister_ftrace_function_probe_func(glob+1, ops);
2661 ret = 0;
2662 goto out;
2663 }
2664
2665 ret = -ENOMEM;
2666 data = kzalloc(sizeof(*data), GFP_KERNEL);
2667 if (!data)
2668 goto out;
2669
2670 data->enable = enable;
2671 data->count = -1;
2672 data->file = file;
2673
2674 if (!param)
2675 goto out_reg;
2676
2677 number = strsep(&param, ":");
2678
2679 ret = -EINVAL;
2680 if (!strlen(number))
2681 goto out_free;
2682
2683 /*
2684 * We use the callback data field (which is a pointer)
2685 * as our counter.
2686 */
2687 ret = kstrtoul(number, 0, &data->count);
2688 if (ret)
2689 goto out_free;
2690
2691 out_reg:
2692 /* Don't let event modules unload while probe registered */
2693 ret = try_module_get(file->event_call->mod);
2694 if (!ret) {
2695 ret = -EBUSY;
2696 goto out_free;
2697 }
2698
2699 ret = __ftrace_event_enable_disable(file, 1, 1);
2700 if (ret < 0)
2701 goto out_put;
2702 ret = register_ftrace_function_probe(glob, ops, data);
2703 /*
2704 * The above returns on success the # of functions enabled,
2705 * but if it didn't find any functions it returns zero.
2706 * Consider no functions a failure too.
2707 */
2708 if (!ret) {
2709 ret = -ENOENT;
2710 goto out_disable;
2711 } else if (ret < 0)
2712 goto out_disable;
2713 /* Just return zero, not the number of enabled functions */
2714 ret = 0;
2715 out:
2716 mutex_unlock(&event_mutex);
2717 return ret;
2718
2719 out_disable:
2720 __ftrace_event_enable_disable(file, 0, 1);
2721 out_put:
2722 module_put(file->event_call->mod);
2723 out_free:
2724 kfree(data);
2725 goto out;
2726 }
2727
2728 static struct ftrace_func_command event_enable_cmd = {
2729 .name = ENABLE_EVENT_STR,
2730 .func = event_enable_func,
2731 };
2732
2733 static struct ftrace_func_command event_disable_cmd = {
2734 .name = DISABLE_EVENT_STR,
2735 .func = event_enable_func,
2736 };
2737
2738 static __init int register_event_cmds(void)
2739 {
2740 int ret;
2741
2742 ret = register_ftrace_command(&event_enable_cmd);
2743 if (WARN_ON(ret < 0))
2744 return ret;
2745 ret = register_ftrace_command(&event_disable_cmd);
2746 if (WARN_ON(ret < 0))
2747 unregister_ftrace_command(&event_enable_cmd);
2748 return ret;
2749 }
2750 #else
2751 static inline int register_event_cmds(void) { return 0; }
2752 #endif /* CONFIG_DYNAMIC_FTRACE */
2753
2754 /*
2755 * The top level array has already had its trace_event_file
2756 * descriptors created in order to allow for early events to
2757 * be recorded. This function is called after the tracefs has been
2758 * initialized, and we now have to create the files associated
2759 * to the events.
2760 */
2761 static __init void
2762 __trace_early_add_event_dirs(struct trace_array *tr)
2763 {
2764 struct trace_event_file *file;
2765 int ret;
2766
2767
2768 list_for_each_entry(file, &tr->events, list) {
2769 ret = event_create_dir(tr->event_dir, file);
2770 if (ret < 0)
2771 pr_warn("Could not create directory for event %s\n",
2772 trace_event_name(file->event_call));
2773 }
2774 }
2775
2776 /*
2777 * For early boot up, the top trace array requires to have
2778 * a list of events that can be enabled. This must be done before
2779 * the filesystem is set up in order to allow events to be traced
2780 * early.
2781 */
2782 static __init void
2783 __trace_early_add_events(struct trace_array *tr)
2784 {
2785 struct trace_event_call *call;
2786 int ret;
2787
2788 list_for_each_entry(call, &ftrace_events, list) {
2789 /* Early boot up should not have any modules loaded */
2790 if (WARN_ON_ONCE(call->mod))
2791 continue;
2792
2793 ret = __trace_early_add_new_event(call, tr);
2794 if (ret < 0)
2795 pr_warn("Could not create early event %s\n",
2796 trace_event_name(call));
2797 }
2798 }
2799
2800 /* Remove the event directory structure for a trace directory. */
2801 static void
2802 __trace_remove_event_dirs(struct trace_array *tr)
2803 {
2804 struct trace_event_file *file, *next;
2805
2806 list_for_each_entry_safe(file, next, &tr->events, list)
2807 remove_event_file_dir(file);
2808 }
2809
2810 static void __add_event_to_tracers(struct trace_event_call *call)
2811 {
2812 struct trace_array *tr;
2813
2814 list_for_each_entry(tr, &ftrace_trace_arrays, list)
2815 __trace_add_new_event(call, tr);
2816 }
2817
2818 extern struct trace_event_call *__start_ftrace_events[];
2819 extern struct trace_event_call *__stop_ftrace_events[];
2820
2821 static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
2822
2823 static __init int setup_trace_event(char *str)
2824 {
2825 strlcpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
2826 ring_buffer_expanded = true;
2827 tracing_selftest_disabled = true;
2828
2829 return 1;
2830 }
2831 __setup("trace_event=", setup_trace_event);
2832
2833 /* Expects to have event_mutex held when called */
2834 static int
2835 create_event_toplevel_files(struct dentry *parent, struct trace_array *tr)
2836 {
2837 struct dentry *d_events;
2838 struct dentry *entry;
2839
2840 entry = tracefs_create_file("set_event", 0644, parent,
2841 tr, &ftrace_set_event_fops);
2842 if (!entry) {
2843 pr_warn("Could not create tracefs 'set_event' entry\n");
2844 return -ENOMEM;
2845 }
2846
2847 d_events = tracefs_create_dir("events", parent);
2848 if (!d_events) {
2849 pr_warn("Could not create tracefs 'events' directory\n");
2850 return -ENOMEM;
2851 }
2852
2853 entry = tracefs_create_file("set_event_pid", 0644, parent,
2854 tr, &ftrace_set_event_pid_fops);
2855
2856 /* ring buffer internal formats */
2857 trace_create_file("header_page", 0444, d_events,
2858 ring_buffer_print_page_header,
2859 &ftrace_show_header_fops);
2860
2861 trace_create_file("header_event", 0444, d_events,
2862 ring_buffer_print_entry_header,
2863 &ftrace_show_header_fops);
2864
2865 trace_create_file("enable", 0644, d_events,
2866 tr, &ftrace_tr_enable_fops);
2867
2868 tr->event_dir = d_events;
2869
2870 return 0;
2871 }
2872
2873 /**
2874 * event_trace_add_tracer - add a instance of a trace_array to events
2875 * @parent: The parent dentry to place the files/directories for events in
2876 * @tr: The trace array associated with these events
2877 *
2878 * When a new instance is created, it needs to set up its events
2879 * directory, as well as other files associated with events. It also
2880 * creates the event hierachry in the @parent/events directory.
2881 *
2882 * Returns 0 on success.
2883 */
2884 int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr)
2885 {
2886 int ret;
2887
2888 mutex_lock(&event_mutex);
2889
2890 ret = create_event_toplevel_files(parent, tr);
2891 if (ret)
2892 goto out_unlock;
2893
2894 down_write(&trace_event_sem);
2895 __trace_add_event_dirs(tr);
2896 up_write(&trace_event_sem);
2897
2898 out_unlock:
2899 mutex_unlock(&event_mutex);
2900
2901 return ret;
2902 }
2903
2904 /*
2905 * The top trace array already had its file descriptors created.
2906 * Now the files themselves need to be created.
2907 */
2908 static __init int
2909 early_event_add_tracer(struct dentry *parent, struct trace_array *tr)
2910 {
2911 int ret;
2912
2913 mutex_lock(&event_mutex);
2914
2915 ret = create_event_toplevel_files(parent, tr);
2916 if (ret)
2917 goto out_unlock;
2918
2919 down_write(&trace_event_sem);
2920 __trace_early_add_event_dirs(tr);
2921 up_write(&trace_event_sem);
2922
2923 out_unlock:
2924 mutex_unlock(&event_mutex);
2925
2926 return ret;
2927 }
2928
2929 int event_trace_del_tracer(struct trace_array *tr)
2930 {
2931 mutex_lock(&event_mutex);
2932
2933 /* Disable any event triggers and associated soft-disabled events */
2934 clear_event_triggers(tr);
2935
2936 /* Clear the pid list */
2937 __ftrace_clear_event_pids(tr);
2938
2939 /* Disable any running events */
2940 __ftrace_set_clr_event_nolock(tr, NULL, NULL, NULL, 0);
2941
2942 /* Access to events are within rcu_read_lock_sched() */
2943 synchronize_sched();
2944
2945 down_write(&trace_event_sem);
2946 __trace_remove_event_dirs(tr);
2947 tracefs_remove_recursive(tr->event_dir);
2948 up_write(&trace_event_sem);
2949
2950 tr->event_dir = NULL;
2951
2952 mutex_unlock(&event_mutex);
2953
2954 return 0;
2955 }
2956
2957 static __init int event_trace_memsetup(void)
2958 {
2959 field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC);
2960 file_cachep = KMEM_CACHE(trace_event_file, SLAB_PANIC);
2961 return 0;
2962 }
2963
2964 static __init void
2965 early_enable_events(struct trace_array *tr, bool disable_first)
2966 {
2967 char *buf = bootup_event_buf;
2968 char *token;
2969 int ret;
2970
2971 while (true) {
2972 token = strsep(&buf, ",");
2973
2974 if (!token)
2975 break;
2976
2977 if (*token) {
2978 /* Restarting syscalls requires that we stop them first */
2979 if (disable_first)
2980 ftrace_set_clr_event(tr, token, 0);
2981
2982 ret = ftrace_set_clr_event(tr, token, 1);
2983 if (ret)
2984 pr_warn("Failed to enable trace event: %s\n", token);
2985 }
2986
2987 /* Put back the comma to allow this to be called again */
2988 if (buf)
2989 *(buf - 1) = ',';
2990 }
2991 }
2992
2993 static __init int event_trace_enable(void)
2994 {
2995 struct trace_array *tr = top_trace_array();
2996 struct trace_event_call **iter, *call;
2997 int ret;
2998
2999 if (!tr)
3000 return -ENODEV;
3001
3002 for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
3003
3004 call = *iter;
3005 ret = event_init(call);
3006 if (!ret)
3007 list_add(&call->list, &ftrace_events);
3008 }
3009
3010 /*
3011 * We need the top trace array to have a working set of trace
3012 * points at early init, before the debug files and directories
3013 * are created. Create the file entries now, and attach them
3014 * to the actual file dentries later.
3015 */
3016 __trace_early_add_events(tr);
3017
3018 early_enable_events(tr, false);
3019
3020 trace_printk_start_comm();
3021
3022 register_event_cmds();
3023
3024 register_trigger_cmds();
3025
3026 return 0;
3027 }
3028
3029 /*
3030 * event_trace_enable() is called from trace_event_init() first to
3031 * initialize events and perhaps start any events that are on the
3032 * command line. Unfortunately, there are some events that will not
3033 * start this early, like the system call tracepoints that need
3034 * to set the TIF_SYSCALL_TRACEPOINT flag of pid 1. But event_trace_enable()
3035 * is called before pid 1 starts, and this flag is never set, making
3036 * the syscall tracepoint never get reached, but the event is enabled
3037 * regardless (and not doing anything).
3038 */
3039 static __init int event_trace_enable_again(void)
3040 {
3041 struct trace_array *tr;
3042
3043 tr = top_trace_array();
3044 if (!tr)
3045 return -ENODEV;
3046
3047 early_enable_events(tr, true);
3048
3049 return 0;
3050 }
3051
3052 early_initcall(event_trace_enable_again);
3053
3054 static __init int event_trace_init(void)
3055 {
3056 struct trace_array *tr;
3057 struct dentry *d_tracer;
3058 struct dentry *entry;
3059 int ret;
3060
3061 tr = top_trace_array();
3062 if (!tr)
3063 return -ENODEV;
3064
3065 d_tracer = tracing_init_dentry();
3066 if (IS_ERR(d_tracer))
3067 return 0;
3068
3069 entry = tracefs_create_file("available_events", 0444, d_tracer,
3070 tr, &ftrace_avail_fops);
3071 if (!entry)
3072 pr_warn("Could not create tracefs 'available_events' entry\n");
3073
3074 if (trace_define_generic_fields())
3075 pr_warn("tracing: Failed to allocated generic fields");
3076
3077 if (trace_define_common_fields())
3078 pr_warn("tracing: Failed to allocate common fields");
3079
3080 ret = early_event_add_tracer(d_tracer, tr);
3081 if (ret)
3082 return ret;
3083
3084 #ifdef CONFIG_MODULES
3085 ret = register_module_notifier(&trace_module_nb);
3086 if (ret)
3087 pr_warn("Failed to register trace events module notifier\n");
3088 #endif
3089 return 0;
3090 }
3091
3092 void __init trace_event_init(void)
3093 {
3094 event_trace_memsetup();
3095 init_ftrace_syscalls();
3096 event_trace_enable();
3097 }
3098
3099 fs_initcall(event_trace_init);
3100
3101 #ifdef CONFIG_FTRACE_STARTUP_TEST
3102
3103 static DEFINE_SPINLOCK(test_spinlock);
3104 static DEFINE_SPINLOCK(test_spinlock_irq);
3105 static DEFINE_MUTEX(test_mutex);
3106
3107 static __init void test_work(struct work_struct *dummy)
3108 {
3109 spin_lock(&test_spinlock);
3110 spin_lock_irq(&test_spinlock_irq);
3111 udelay(1);
3112 spin_unlock_irq(&test_spinlock_irq);
3113 spin_unlock(&test_spinlock);
3114
3115 mutex_lock(&test_mutex);
3116 msleep(1);
3117 mutex_unlock(&test_mutex);
3118 }
3119
3120 static __init int event_test_thread(void *unused)
3121 {
3122 void *test_malloc;
3123
3124 test_malloc = kmalloc(1234, GFP_KERNEL);
3125 if (!test_malloc)
3126 pr_info("failed to kmalloc\n");
3127
3128 schedule_on_each_cpu(test_work);
3129
3130 kfree(test_malloc);
3131
3132 set_current_state(TASK_INTERRUPTIBLE);
3133 while (!kthread_should_stop()) {
3134 schedule();
3135 set_current_state(TASK_INTERRUPTIBLE);
3136 }
3137 __set_current_state(TASK_RUNNING);
3138
3139 return 0;
3140 }
3141
3142 /*
3143 * Do various things that may trigger events.
3144 */
3145 static __init void event_test_stuff(void)
3146 {
3147 struct task_struct *test_thread;
3148
3149 test_thread = kthread_run(event_test_thread, NULL, "test-events");
3150 msleep(1);
3151 kthread_stop(test_thread);
3152 }
3153
3154 /*
3155 * For every trace event defined, we will test each trace point separately,
3156 * and then by groups, and finally all trace points.
3157 */
3158 static __init void event_trace_self_tests(void)
3159 {
3160 struct trace_subsystem_dir *dir;
3161 struct trace_event_file *file;
3162 struct trace_event_call *call;
3163 struct event_subsystem *system;
3164 struct trace_array *tr;
3165 int ret;
3166
3167 tr = top_trace_array();
3168 if (!tr)
3169 return;
3170
3171 pr_info("Running tests on trace events:\n");
3172
3173 list_for_each_entry(file, &tr->events, list) {
3174
3175 call = file->event_call;
3176
3177 /* Only test those that have a probe */
3178 if (!call->class || !call->class->probe)
3179 continue;
3180
3181 /*
3182 * Testing syscall events here is pretty useless, but
3183 * we still do it if configured. But this is time consuming.
3184 * What we really need is a user thread to perform the
3185 * syscalls as we test.
3186 */
3187 #ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
3188 if (call->class->system &&
3189 strcmp(call->class->system, "syscalls") == 0)
3190 continue;
3191 #endif
3192
3193 pr_info("Testing event %s: ", trace_event_name(call));
3194
3195 /*
3196 * If an event is already enabled, someone is using
3197 * it and the self test should not be on.
3198 */
3199 if (file->flags & EVENT_FILE_FL_ENABLED) {
3200 pr_warn("Enabled event during self test!\n");
3201 WARN_ON_ONCE(1);
3202 continue;
3203 }
3204
3205 ftrace_event_enable_disable(file, 1);
3206 event_test_stuff();
3207 ftrace_event_enable_disable(file, 0);
3208
3209 pr_cont("OK\n");
3210 }
3211
3212 /* Now test at the sub system level */
3213
3214 pr_info("Running tests on trace event systems:\n");
3215
3216 list_for_each_entry(dir, &tr->systems, list) {
3217
3218 system = dir->subsystem;
3219
3220 /* the ftrace system is special, skip it */
3221 if (strcmp(system->name, "ftrace") == 0)
3222 continue;
3223
3224 pr_info("Testing event system %s: ", system->name);
3225
3226 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1);
3227 if (WARN_ON_ONCE(ret)) {
3228 pr_warn("error enabling system %s\n",
3229 system->name);
3230 continue;
3231 }
3232
3233 event_test_stuff();
3234
3235 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0);
3236 if (WARN_ON_ONCE(ret)) {
3237 pr_warn("error disabling system %s\n",
3238 system->name);
3239 continue;
3240 }
3241
3242 pr_cont("OK\n");
3243 }
3244
3245 /* Test with all events enabled */
3246
3247 pr_info("Running tests on all trace events:\n");
3248 pr_info("Testing all events: ");
3249
3250 ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1);
3251 if (WARN_ON_ONCE(ret)) {
3252 pr_warn("error enabling all events\n");
3253 return;
3254 }
3255
3256 event_test_stuff();
3257
3258 /* reset sysname */
3259 ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
3260 if (WARN_ON_ONCE(ret)) {
3261 pr_warn("error disabling all events\n");
3262 return;
3263 }
3264
3265 pr_cont("OK\n");
3266 }
3267
3268 #ifdef CONFIG_FUNCTION_TRACER
3269
3270 static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
3271
3272 static struct trace_event_file event_trace_file __initdata;
3273
3274 static void __init
3275 function_test_events_call(unsigned long ip, unsigned long parent_ip,
3276 struct ftrace_ops *op, struct pt_regs *pt_regs)
3277 {
3278 struct ring_buffer_event *event;
3279 struct ring_buffer *buffer;
3280 struct ftrace_entry *entry;
3281 unsigned long flags;
3282 long disabled;
3283 int cpu;
3284 int pc;
3285
3286 pc = preempt_count();
3287 preempt_disable_notrace();
3288 cpu = raw_smp_processor_id();
3289 disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
3290
3291 if (disabled != 1)
3292 goto out;
3293
3294 local_save_flags(flags);
3295
3296 event = trace_event_buffer_lock_reserve(&buffer, &event_trace_file,
3297 TRACE_FN, sizeof(*entry),
3298 flags, pc);
3299 if (!event)
3300 goto out;
3301 entry = ring_buffer_event_data(event);
3302 entry->ip = ip;
3303 entry->parent_ip = parent_ip;
3304
3305 event_trigger_unlock_commit(&event_trace_file, buffer, event,
3306 entry, flags, pc);
3307 out:
3308 atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
3309 preempt_enable_notrace();
3310 }
3311
3312 static struct ftrace_ops trace_ops __initdata =
3313 {
3314 .func = function_test_events_call,
3315 .flags = FTRACE_OPS_FL_RECURSION_SAFE,
3316 };
3317
3318 static __init void event_trace_self_test_with_function(void)
3319 {
3320 int ret;
3321
3322 event_trace_file.tr = top_trace_array();
3323 if (WARN_ON(!event_trace_file.tr))
3324 return;
3325
3326 ret = register_ftrace_function(&trace_ops);
3327 if (WARN_ON(ret < 0)) {
3328 pr_info("Failed to enable function tracer for event tests\n");
3329 return;
3330 }
3331 pr_info("Running tests again, along with the function tracer\n");
3332 event_trace_self_tests();
3333 unregister_ftrace_function(&trace_ops);
3334 }
3335 #else
3336 static __init void event_trace_self_test_with_function(void)
3337 {
3338 }
3339 #endif
3340
3341 static __init int event_trace_self_tests_init(void)
3342 {
3343 if (!tracing_selftest_disabled) {
3344 event_trace_self_tests();
3345 event_trace_self_test_with_function();
3346 }
3347
3348 return 0;
3349 }
3350
3351 late_initcall(event_trace_self_tests_init);
3352
3353 #endif
Linux kernel - Deliverables
This page took 0.100013 seconds and 5 git commands to generate.