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