?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 4.19.168
[linux/fpc-iii.git] / kernel / trace / trace_events.c
blob0fc06a7da87fb6308b96fd2281dfd228018d3e86
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * event tracer
4 *
5 * Copyright (C) 2008 Red Hat Inc, Steven Rostedt <srostedt@redhat.com>
6 *
7 * - Added format output of fields of the trace point.
8 * This was based off of work by Tom Zanussi <tzanussi@gmail.com>.
9 *
10 */
11
12 #define pr_fmt(fmt) fmt
13
14 #include <linux/workqueue.h>
15 #include <linux/spinlock.h>
16 #include <linux/kthread.h>
17 #include <linux/tracefs.h>
18 #include <linux/uaccess.h>
19 #include <linux/module.h>
20 #include <linux/ctype.h>
21 #include <linux/sort.h>
22 #include <linux/slab.h>
23 #include <linux/delay.h>
24
25 #include <trace/events/sched.h>
26
27 #include <asm/setup.h>
28
29 #include "trace_output.h"
30
31 #undef TRACE_SYSTEM
32 #define TRACE_SYSTEM "TRACE_SYSTEM"
33
34 DEFINE_MUTEX(event_mutex);
35
36 LIST_HEAD(ftrace_events);
37 static LIST_HEAD(ftrace_generic_fields);
38 static LIST_HEAD(ftrace_common_fields);
39
40 #define GFP_TRACE (GFP_KERNEL | __GFP_ZERO)
41
42 static struct kmem_cache *field_cachep;
43 static struct kmem_cache *file_cachep;
44
45 static inline int system_refcount(struct event_subsystem *system)
46 {
47 return system->ref_count;
48 }
49
50 static int system_refcount_inc(struct event_subsystem *system)
51 {
52 return system->ref_count++;
53 }
54
55 static int system_refcount_dec(struct event_subsystem *system)
56 {
57 return --system->ref_count;
58 }
59
60 /* Double loops, do not use break, only goto's work */
61 #define do_for_each_event_file(tr, file) \
62 list_for_each_entry(tr, &ftrace_trace_arrays, list) { \
63 list_for_each_entry(file, &tr->events, list)
64
65 #define do_for_each_event_file_safe(tr, file) \
66 list_for_each_entry(tr, &ftrace_trace_arrays, list) { \
67 struct trace_event_file *___n; \
68 list_for_each_entry_safe(file, ___n, &tr->events, list)
69
70 #define while_for_each_event_file() \
71 }
72
73 static struct list_head *
74 trace_get_fields(struct trace_event_call *event_call)
75 {
76 if (!event_call->class->get_fields)
77 return &event_call->class->fields;
78 return event_call->class->get_fields(event_call);
79 }
80
81 static struct ftrace_event_field *
82 __find_event_field(struct list_head *head, char *name)
83 {
84 struct ftrace_event_field *field;
85
86 list_for_each_entry(field, head, link) {
87 if (!strcmp(field->name, name))
88 return field;
89 }
90
91 return NULL;
92 }
93
94 struct ftrace_event_field *
95 trace_find_event_field(struct trace_event_call *call, char *name)
96 {
97 struct ftrace_event_field *field;
98 struct list_head *head;
99
100 head = trace_get_fields(call);
101 field = __find_event_field(head, name);
102 if (field)
103 return field;
104
105 field = __find_event_field(&ftrace_generic_fields, name);
106 if (field)
107 return field;
108
109 return __find_event_field(&ftrace_common_fields, name);
110 }
111
112 static int __trace_define_field(struct list_head *head, const char *type,
113 const char *name, int offset, int size,
114 int is_signed, int filter_type)
115 {
116 struct ftrace_event_field *field;
117
118 field = kmem_cache_alloc(field_cachep, GFP_TRACE);
119 if (!field)
120 return -ENOMEM;
121
122 field->name = name;
123 field->type = type;
124
125 if (filter_type == FILTER_OTHER)
126 field->filter_type = filter_assign_type(type);
127 else
128 field->filter_type = filter_type;
129
130 field->offset = offset;
131 field->size = size;
132 field->is_signed = is_signed;
133
134 list_add(&field->link, head);
135
136 return 0;
137 }
138
139 int trace_define_field(struct trace_event_call *call, const char *type,
140 const char *name, int offset, int size, int is_signed,
141 int filter_type)
142 {
143 struct list_head *head;
144
145 if (WARN_ON(!call->class))
146 return 0;
147
148 head = trace_get_fields(call);
149 return __trace_define_field(head, type, name, offset, size,
150 is_signed, filter_type);
151 }
152 EXPORT_SYMBOL_GPL(trace_define_field);
153
154 #define __generic_field(type, item, filter_type) \
155 ret = __trace_define_field(&ftrace_generic_fields, #type, \
156 #item, 0, 0, is_signed_type(type), \
157 filter_type); \
158 if (ret) \
159 return ret;
160
161 #define __common_field(type, item) \
162 ret = __trace_define_field(&ftrace_common_fields, #type, \
163 "common_" #item, \
164 offsetof(typeof(ent), item), \
165 sizeof(ent.item), \
166 is_signed_type(type), FILTER_OTHER); \
167 if (ret) \
168 return ret;
169
170 static int trace_define_generic_fields(void)
171 {
172 int ret;
173
174 __generic_field(int, CPU, FILTER_CPU);
175 __generic_field(int, cpu, FILTER_CPU);
176 __generic_field(char *, COMM, FILTER_COMM);
177 __generic_field(char *, comm, FILTER_COMM);
178
179 return ret;
180 }
181
182 static int trace_define_common_fields(void)
183 {
184 int ret;
185 struct trace_entry ent;
186
187 __common_field(unsigned short, type);
188 __common_field(unsigned char, flags);
189 __common_field(unsigned char, preempt_count);
190 __common_field(int, pid);
191
192 return ret;
193 }
194
195 static void trace_destroy_fields(struct trace_event_call *call)
196 {
197 struct ftrace_event_field *field, *next;
198 struct list_head *head;
199
200 head = trace_get_fields(call);
201 list_for_each_entry_safe(field, next, head, link) {
202 list_del(&field->link);
203 kmem_cache_free(field_cachep, field);
204 }
205 }
206
207 /*
208 * run-time version of trace_event_get_offsets_<call>() that returns the last
209 * accessible offset of trace fields excluding __dynamic_array bytes
210 */
211 int trace_event_get_offsets(struct trace_event_call *call)
212 {
213 struct ftrace_event_field *tail;
214 struct list_head *head;
215
216 head = trace_get_fields(call);
217 /*
218 * head->next points to the last field with the largest offset,
219 * since it was added last by trace_define_field()
220 */
221 tail = list_first_entry(head, struct ftrace_event_field, link);
222 return tail->offset + tail->size;
223 }
224
225 int trace_event_raw_init(struct trace_event_call *call)
226 {
227 int id;
228
229 id = register_trace_event(&call->event);
230 if (!id)
231 return -ENODEV;
232
233 return 0;
234 }
235 EXPORT_SYMBOL_GPL(trace_event_raw_init);
236
237 bool trace_event_ignore_this_pid(struct trace_event_file *trace_file)
238 {
239 struct trace_array *tr = trace_file->tr;
240 struct trace_array_cpu *data;
241 struct trace_pid_list *pid_list;
242
243 pid_list = rcu_dereference_raw(tr->filtered_pids);
244 if (!pid_list)
245 return false;
246
247 data = this_cpu_ptr(tr->trace_buffer.data);
248
249 return data->ignore_pid;
250 }
251 EXPORT_SYMBOL_GPL(trace_event_ignore_this_pid);
252
253 void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer,
254 struct trace_event_file *trace_file,
255 unsigned long len)
256 {
257 struct trace_event_call *event_call = trace_file->event_call;
258
259 if ((trace_file->flags & EVENT_FILE_FL_PID_FILTER) &&
260 trace_event_ignore_this_pid(trace_file))
261 return NULL;
262
263 local_save_flags(fbuffer->flags);
264 fbuffer->pc = preempt_count();
265 /*
266 * If CONFIG_PREEMPT is enabled, then the tracepoint itself disables
267 * preemption (adding one to the preempt_count). Since we are
268 * interested in the preempt_count at the time the tracepoint was
269 * hit, we need to subtract one to offset the increment.
270 */
271 if (IS_ENABLED(CONFIG_PREEMPT))
272 fbuffer->pc--;
273 fbuffer->trace_file = trace_file;
274
275 fbuffer->event =
276 trace_event_buffer_lock_reserve(&fbuffer->buffer, trace_file,
277 event_call->event.type, len,
278 fbuffer->flags, fbuffer->pc);
279 if (!fbuffer->event)
280 return NULL;
281
282 fbuffer->entry = ring_buffer_event_data(fbuffer->event);
283 return fbuffer->entry;
284 }
285 EXPORT_SYMBOL_GPL(trace_event_buffer_reserve);
286
287 int trace_event_reg(struct trace_event_call *call,
288 enum trace_reg type, void *data)
289 {
290 struct trace_event_file *file = data;
291
292 WARN_ON(!(call->flags & TRACE_EVENT_FL_TRACEPOINT));
293 switch (type) {
294 case TRACE_REG_REGISTER:
295 return tracepoint_probe_register(call->tp,
296 call->class->probe,
297 file);
298 case TRACE_REG_UNREGISTER:
299 tracepoint_probe_unregister(call->tp,
300 call->class->probe,
301 file);
302 return 0;
303
304 #ifdef CONFIG_PERF_EVENTS
305 case TRACE_REG_PERF_REGISTER:
306 return tracepoint_probe_register(call->tp,
307 call->class->perf_probe,
308 call);
309 case TRACE_REG_PERF_UNREGISTER:
310 tracepoint_probe_unregister(call->tp,
311 call->class->perf_probe,
312 call);
313 return 0;
314 case TRACE_REG_PERF_OPEN:
315 case TRACE_REG_PERF_CLOSE:
316 case TRACE_REG_PERF_ADD:
317 case TRACE_REG_PERF_DEL:
318 return 0;
319 #endif
320 }
321 return 0;
322 }
323 EXPORT_SYMBOL_GPL(trace_event_reg);
324
325 void trace_event_enable_cmd_record(bool enable)
326 {
327 struct trace_event_file *file;
328 struct trace_array *tr;
329
330 lockdep_assert_held(&event_mutex);
331
332 do_for_each_event_file(tr, file) {
333
334 if (!(file->flags & EVENT_FILE_FL_ENABLED))
335 continue;
336
337 if (enable) {
338 tracing_start_cmdline_record();
339 set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
340 } else {
341 tracing_stop_cmdline_record();
342 clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
343 }
344 } while_for_each_event_file();
345 }
346
347 void trace_event_enable_tgid_record(bool enable)
348 {
349 struct trace_event_file *file;
350 struct trace_array *tr;
351
352 lockdep_assert_held(&event_mutex);
353
354 do_for_each_event_file(tr, file) {
355 if (!(file->flags & EVENT_FILE_FL_ENABLED))
356 continue;
357
358 if (enable) {
359 tracing_start_tgid_record();
360 set_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
361 } else {
362 tracing_stop_tgid_record();
363 clear_bit(EVENT_FILE_FL_RECORDED_TGID_BIT,
364 &file->flags);
365 }
366 } while_for_each_event_file();
367 }
368
369 static int __ftrace_event_enable_disable(struct trace_event_file *file,
370 int enable, int soft_disable)
371 {
372 struct trace_event_call *call = file->event_call;
373 struct trace_array *tr = file->tr;
374 unsigned long file_flags = file->flags;
375 int ret = 0;
376 int disable;
377
378 switch (enable) {
379 case 0:
380 /*
381 * When soft_disable is set and enable is cleared, the sm_ref
382 * reference counter is decremented. If it reaches 0, we want
383 * to clear the SOFT_DISABLED flag but leave the event in the
384 * state that it was. That is, if the event was enabled and
385 * SOFT_DISABLED isn't set, then do nothing. But if SOFT_DISABLED
386 * is set we do not want the event to be enabled before we
387 * clear the bit.
388 *
389 * When soft_disable is not set but the SOFT_MODE flag is,
390 * we do nothing. Do not disable the tracepoint, otherwise
391 * "soft enable"s (clearing the SOFT_DISABLED bit) wont work.
392 */
393 if (soft_disable) {
394 if (atomic_dec_return(&file->sm_ref) > 0)
395 break;
396 disable = file->flags & EVENT_FILE_FL_SOFT_DISABLED;
397 clear_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
398 } else
399 disable = !(file->flags & EVENT_FILE_FL_SOFT_MODE);
400
401 if (disable && (file->flags & EVENT_FILE_FL_ENABLED)) {
402 clear_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
403 if (file->flags & EVENT_FILE_FL_RECORDED_CMD) {
404 tracing_stop_cmdline_record();
405 clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
406 }
407
408 if (file->flags & EVENT_FILE_FL_RECORDED_TGID) {
409 tracing_stop_tgid_record();
410 clear_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
411 }
412
413 call->class->reg(call, TRACE_REG_UNREGISTER, file);
414 }
415 /* If in SOFT_MODE, just set the SOFT_DISABLE_BIT, else clear it */
416 if (file->flags & EVENT_FILE_FL_SOFT_MODE)
417 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
418 else
419 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
420 break;
421 case 1:
422 /*
423 * When soft_disable is set and enable is set, we want to
424 * register the tracepoint for the event, but leave the event
425 * as is. That means, if the event was already enabled, we do
426 * nothing (but set SOFT_MODE). If the event is disabled, we
427 * set SOFT_DISABLED before enabling the event tracepoint, so
428 * it still seems to be disabled.
429 */
430 if (!soft_disable)
431 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
432 else {
433 if (atomic_inc_return(&file->sm_ref) > 1)
434 break;
435 set_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
436 }
437
438 if (!(file->flags & EVENT_FILE_FL_ENABLED)) {
439 bool cmd = false, tgid = false;
440
441 /* Keep the event disabled, when going to SOFT_MODE. */
442 if (soft_disable)
443 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
444
445 if (tr->trace_flags & TRACE_ITER_RECORD_CMD) {
446 cmd = true;
447 tracing_start_cmdline_record();
448 set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
449 }
450
451 if (tr->trace_flags & TRACE_ITER_RECORD_TGID) {
452 tgid = true;
453 tracing_start_tgid_record();
454 set_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
455 }
456
457 ret = call->class->reg(call, TRACE_REG_REGISTER, file);
458 if (ret) {
459 if (cmd)
460 tracing_stop_cmdline_record();
461 if (tgid)
462 tracing_stop_tgid_record();
463 pr_info("event trace: Could not enable event "
464 "%s\n", trace_event_name(call));
465 break;
466 }
467 set_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
468
469 /* WAS_ENABLED gets set but never cleared. */
470 set_bit(EVENT_FILE_FL_WAS_ENABLED_BIT, &file->flags);
471 }
472 break;
473 }
474
475 /* Enable or disable use of trace_buffered_event */
476 if ((file_flags & EVENT_FILE_FL_SOFT_DISABLED) !=
477 (file->flags & EVENT_FILE_FL_SOFT_DISABLED)) {
478 if (file->flags & EVENT_FILE_FL_SOFT_DISABLED)
479 trace_buffered_event_enable();
480 else
481 trace_buffered_event_disable();
482 }
483
484 return ret;
485 }
486
487 int trace_event_enable_disable(struct trace_event_file *file,
488 int enable, int soft_disable)
489 {
490 return __ftrace_event_enable_disable(file, enable, soft_disable);
491 }
492
493 static int ftrace_event_enable_disable(struct trace_event_file *file,
494 int enable)
495 {
496 return __ftrace_event_enable_disable(file, enable, 0);
497 }
498
499 static void ftrace_clear_events(struct trace_array *tr)
500 {
501 struct trace_event_file *file;
502
503 mutex_lock(&event_mutex);
504 list_for_each_entry(file, &tr->events, list) {
505 ftrace_event_enable_disable(file, 0);
506 }
507 mutex_unlock(&event_mutex);
508 }
509
510 static void
511 event_filter_pid_sched_process_exit(void *data, struct task_struct *task)
512 {
513 struct trace_pid_list *pid_list;
514 struct trace_array *tr = data;
515
516 pid_list = rcu_dereference_raw(tr->filtered_pids);
517 trace_filter_add_remove_task(pid_list, NULL, task);
518 }
519
520 static void
521 event_filter_pid_sched_process_fork(void *data,
522 struct task_struct *self,
523 struct task_struct *task)
524 {
525 struct trace_pid_list *pid_list;
526 struct trace_array *tr = data;
527
528 pid_list = rcu_dereference_sched(tr->filtered_pids);
529 trace_filter_add_remove_task(pid_list, self, task);
530 }
531
532 void trace_event_follow_fork(struct trace_array *tr, bool enable)
533 {
534 if (enable) {
535 register_trace_prio_sched_process_fork(event_filter_pid_sched_process_fork,
536 tr, INT_MIN);
537 register_trace_prio_sched_process_free(event_filter_pid_sched_process_exit,
538 tr, INT_MAX);
539 } else {
540 unregister_trace_sched_process_fork(event_filter_pid_sched_process_fork,
541 tr);
542 unregister_trace_sched_process_free(event_filter_pid_sched_process_exit,
543 tr);
544 }
545 }
546
547 static void
548 event_filter_pid_sched_switch_probe_pre(void *data, bool preempt,
549 struct task_struct *prev, struct task_struct *next)
550 {
551 struct trace_array *tr = data;
552 struct trace_pid_list *pid_list;
553
554 pid_list = rcu_dereference_sched(tr->filtered_pids);
555
556 this_cpu_write(tr->trace_buffer.data->ignore_pid,
557 trace_ignore_this_task(pid_list, prev) &&
558 trace_ignore_this_task(pid_list, next));
559 }
560
561 static void
562 event_filter_pid_sched_switch_probe_post(void *data, bool preempt,
563 struct task_struct *prev, struct task_struct *next)
564 {
565 struct trace_array *tr = data;
566 struct trace_pid_list *pid_list;
567
568 pid_list = rcu_dereference_sched(tr->filtered_pids);
569
570 this_cpu_write(tr->trace_buffer.data->ignore_pid,
571 trace_ignore_this_task(pid_list, next));
572 }
573
574 static void
575 event_filter_pid_sched_wakeup_probe_pre(void *data, struct task_struct *task)
576 {
577 struct trace_array *tr = data;
578 struct trace_pid_list *pid_list;
579
580 /* Nothing to do if we are already tracing */
581 if (!this_cpu_read(tr->trace_buffer.data->ignore_pid))
582 return;
583
584 pid_list = rcu_dereference_sched(tr->filtered_pids);
585
586 this_cpu_write(tr->trace_buffer.data->ignore_pid,
587 trace_ignore_this_task(pid_list, task));
588 }
589
590 static void
591 event_filter_pid_sched_wakeup_probe_post(void *data, struct task_struct *task)
592 {
593 struct trace_array *tr = data;
594 struct trace_pid_list *pid_list;
595
596 /* Nothing to do if we are not tracing */
597 if (this_cpu_read(tr->trace_buffer.data->ignore_pid))
598 return;
599
600 pid_list = rcu_dereference_sched(tr->filtered_pids);
601
602 /* Set tracing if current is enabled */
603 this_cpu_write(tr->trace_buffer.data->ignore_pid,
604 trace_ignore_this_task(pid_list, current));
605 }
606
607 static void __ftrace_clear_event_pids(struct trace_array *tr)
608 {
609 struct trace_pid_list *pid_list;
610 struct trace_event_file *file;
611 int cpu;
612
613 pid_list = rcu_dereference_protected(tr->filtered_pids,
614 lockdep_is_held(&event_mutex));
615 if (!pid_list)
616 return;
617
618 unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_pre, tr);
619 unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_post, tr);
620
621 unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre, tr);
622 unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_post, tr);
623
624 unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre, tr);
625 unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post, tr);
626
627 unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_pre, tr);
628 unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_post, tr);
629
630 list_for_each_entry(file, &tr->events, list) {
631 clear_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
632 }
633
634 for_each_possible_cpu(cpu)
635 per_cpu_ptr(tr->trace_buffer.data, cpu)->ignore_pid = false;
636
637 rcu_assign_pointer(tr->filtered_pids, NULL);
638
639 /* Wait till all users are no longer using pid filtering */
640 tracepoint_synchronize_unregister();
641
642 trace_free_pid_list(pid_list);
643 }
644
645 static void ftrace_clear_event_pids(struct trace_array *tr)
646 {
647 mutex_lock(&event_mutex);
648 __ftrace_clear_event_pids(tr);
649 mutex_unlock(&event_mutex);
650 }
651
652 static void __put_system(struct event_subsystem *system)
653 {
654 struct event_filter *filter = system->filter;
655
656 WARN_ON_ONCE(system_refcount(system) == 0);
657 if (system_refcount_dec(system))
658 return;
659
660 list_del(&system->list);
661
662 if (filter) {
663 kfree(filter->filter_string);
664 kfree(filter);
665 }
666 kfree_const(system->name);
667 kfree(system);
668 }
669
670 static void __get_system(struct event_subsystem *system)
671 {
672 WARN_ON_ONCE(system_refcount(system) == 0);
673 system_refcount_inc(system);
674 }
675
676 static void __get_system_dir(struct trace_subsystem_dir *dir)
677 {
678 WARN_ON_ONCE(dir->ref_count == 0);
679 dir->ref_count++;
680 __get_system(dir->subsystem);
681 }
682
683 static void __put_system_dir(struct trace_subsystem_dir *dir)
684 {
685 WARN_ON_ONCE(dir->ref_count == 0);
686 /* If the subsystem is about to be freed, the dir must be too */
687 WARN_ON_ONCE(system_refcount(dir->subsystem) == 1 && dir->ref_count != 1);
688
689 __put_system(dir->subsystem);
690 if (!--dir->ref_count)
691 kfree(dir);
692 }
693
694 static void put_system(struct trace_subsystem_dir *dir)
695 {
696 mutex_lock(&event_mutex);
697 __put_system_dir(dir);
698 mutex_unlock(&event_mutex);
699 }
700
701 static void remove_subsystem(struct trace_subsystem_dir *dir)
702 {
703 if (!dir)
704 return;
705
706 if (!--dir->nr_events) {
707 tracefs_remove_recursive(dir->entry);
708 list_del(&dir->list);
709 __put_system_dir(dir);
710 }
711 }
712
713 static void remove_event_file_dir(struct trace_event_file *file)
714 {
715 struct dentry *dir = file->dir;
716 struct dentry *child;
717
718 if (dir) {
719 spin_lock(&dir->d_lock); /* probably unneeded */
720 list_for_each_entry(child, &dir->d_subdirs, d_child) {
721 if (d_really_is_positive(child)) /* probably unneeded */
722 d_inode(child)->i_private = NULL;
723 }
724 spin_unlock(&dir->d_lock);
725
726 tracefs_remove_recursive(dir);
727 }
728
729 list_del(&file->list);
730 remove_subsystem(file->system);
731 free_event_filter(file->filter);
732 kmem_cache_free(file_cachep, file);
733 }
734
735 /*
736 * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
737 */
738 static int
739 __ftrace_set_clr_event_nolock(struct trace_array *tr, const char *match,
740 const char *sub, const char *event, int set)
741 {
742 struct trace_event_file *file;
743 struct trace_event_call *call;
744 const char *name;
745 int ret = -EINVAL;
746 int eret = 0;
747
748 list_for_each_entry(file, &tr->events, list) {
749
750 call = file->event_call;
751 name = trace_event_name(call);
752
753 if (!name || !call->class || !call->class->reg)
754 continue;
755
756 if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
757 continue;
758
759 if (match &&
760 strcmp(match, name) != 0 &&
761 strcmp(match, call->class->system) != 0)
762 continue;
763
764 if (sub && strcmp(sub, call->class->system) != 0)
765 continue;
766
767 if (event && strcmp(event, name) != 0)
768 continue;
769
770 ret = ftrace_event_enable_disable(file, set);
771
772 /*
773 * Save the first error and return that. Some events
774 * may still have been enabled, but let the user
775 * know that something went wrong.
776 */
777 if (ret && !eret)
778 eret = ret;
779
780 ret = eret;
781 }
782
783 return ret;
784 }
785
786 static int __ftrace_set_clr_event(struct trace_array *tr, const char *match,
787 const char *sub, const char *event, int set)
788 {
789 int ret;
790
791 mutex_lock(&event_mutex);
792 ret = __ftrace_set_clr_event_nolock(tr, match, sub, event, set);
793 mutex_unlock(&event_mutex);
794
795 return ret;
796 }
797
798 static int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set)
799 {
800 char *event = NULL, *sub = NULL, *match;
801 int ret;
802
803 if (!tr)
804 return -ENOENT;
805 /*
806 * The buf format can be <subsystem>:<event-name>
807 * *:<event-name> means any event by that name.
808 * :<event-name> is the same.
809 *
810 * <subsystem>:* means all events in that subsystem
811 * <subsystem>: means the same.
812 *
813 * <name> (no ':') means all events in a subsystem with
814 * the name <name> or any event that matches <name>
815 */
816
817 match = strsep(&buf, ":");
818 if (buf) {
819 sub = match;
820 event = buf;
821 match = NULL;
822
823 if (!strlen(sub) || strcmp(sub, "*") == 0)
824 sub = NULL;
825 if (!strlen(event) || strcmp(event, "*") == 0)
826 event = NULL;
827 }
828
829 ret = __ftrace_set_clr_event(tr, match, sub, event, set);
830
831 /* Put back the colon to allow this to be called again */
832 if (buf)
833 *(buf - 1) = ':';
834
835 return ret;
836 }
837
838 /**
839 * trace_set_clr_event - enable or disable an event
840 * @system: system name to match (NULL for any system)
841 * @event: event name to match (NULL for all events, within system)
842 * @set: 1 to enable, 0 to disable
843 *
844 * This is a way for other parts of the kernel to enable or disable
845 * event recording.
846 *
847 * Returns 0 on success, -EINVAL if the parameters do not match any
848 * registered events.
849 */
850 int trace_set_clr_event(const char *system, const char *event, int set)
851 {
852 struct trace_array *tr = top_trace_array();
853
854 if (!tr)
855 return -ENODEV;
856
857 return __ftrace_set_clr_event(tr, NULL, system, event, set);
858 }
859 EXPORT_SYMBOL_GPL(trace_set_clr_event);
860
861 /* 128 should be much more than enough */
862 #define EVENT_BUF_SIZE 127
863
864 static ssize_t
865 ftrace_event_write(struct file *file, const char __user *ubuf,
866 size_t cnt, loff_t *ppos)
867 {
868 struct trace_parser parser;
869 struct seq_file *m = file->private_data;
870 struct trace_array *tr = m->private;
871 ssize_t read, ret;
872
873 if (!cnt)
874 return 0;
875
876 ret = tracing_update_buffers();
877 if (ret < 0)
878 return ret;
879
880 if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
881 return -ENOMEM;
882
883 read = trace_get_user(&parser, ubuf, cnt, ppos);
884
885 if (read >= 0 && trace_parser_loaded((&parser))) {
886 int set = 1;
887
888 if (*parser.buffer == '!')
889 set = 0;
890
891 ret = ftrace_set_clr_event(tr, parser.buffer + !set, set);
892 if (ret)
893 goto out_put;
894 }
895
896 ret = read;
897
898 out_put:
899 trace_parser_put(&parser);
900
901 return ret;
902 }
903
904 static void *
905 t_next(struct seq_file *m, void *v, loff_t *pos)
906 {
907 struct trace_event_file *file = v;
908 struct trace_event_call *call;
909 struct trace_array *tr = m->private;
910
911 (*pos)++;
912
913 list_for_each_entry_continue(file, &tr->events, list) {
914 call = file->event_call;
915 /*
916 * The ftrace subsystem is for showing formats only.
917 * They can not be enabled or disabled via the event files.
918 */
919 if (call->class && call->class->reg &&
920 !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
921 return file;
922 }
923
924 return NULL;
925 }
926
927 static void *t_start(struct seq_file *m, loff_t *pos)
928 {
929 struct trace_event_file *file;
930 struct trace_array *tr = m->private;
931 loff_t l;
932
933 mutex_lock(&event_mutex);
934
935 file = list_entry(&tr->events, struct trace_event_file, list);
936 for (l = 0; l <= *pos; ) {
937 file = t_next(m, file, &l);
938 if (!file)
939 break;
940 }
941 return file;
942 }
943
944 static void *
945 s_next(struct seq_file *m, void *v, loff_t *pos)
946 {
947 struct trace_event_file *file = v;
948 struct trace_array *tr = m->private;
949
950 (*pos)++;
951
952 list_for_each_entry_continue(file, &tr->events, list) {
953 if (file->flags & EVENT_FILE_FL_ENABLED)
954 return file;
955 }
956
957 return NULL;
958 }
959
960 static void *s_start(struct seq_file *m, loff_t *pos)
961 {
962 struct trace_event_file *file;
963 struct trace_array *tr = m->private;
964 loff_t l;
965
966 mutex_lock(&event_mutex);
967
968 file = list_entry(&tr->events, struct trace_event_file, list);
969 for (l = 0; l <= *pos; ) {
970 file = s_next(m, file, &l);
971 if (!file)
972 break;
973 }
974 return file;
975 }
976
977 static int t_show(struct seq_file *m, void *v)
978 {
979 struct trace_event_file *file = v;
980 struct trace_event_call *call = file->event_call;
981
982 if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
983 seq_printf(m, "%s:", call->class->system);
984 seq_printf(m, "%s\n", trace_event_name(call));
985
986 return 0;
987 }
988
989 static void t_stop(struct seq_file *m, void *p)
990 {
991 mutex_unlock(&event_mutex);
992 }
993
994 static void *
995 p_next(struct seq_file *m, void *v, loff_t *pos)
996 {
997 struct trace_array *tr = m->private;
998 struct trace_pid_list *pid_list = rcu_dereference_sched(tr->filtered_pids);
999
1000 return trace_pid_next(pid_list, v, pos);
1001 }
1002
1003 static void *p_start(struct seq_file *m, loff_t *pos)
1004 __acquires(RCU)
1005 {
1006 struct trace_pid_list *pid_list;
1007 struct trace_array *tr = m->private;
1008
1009 /*
1010 * Grab the mutex, to keep calls to p_next() having the same
1011 * tr->filtered_pids as p_start() has.
1012 * If we just passed the tr->filtered_pids around, then RCU would
1013 * have been enough, but doing that makes things more complex.
1014 */
1015 mutex_lock(&event_mutex);
1016 rcu_read_lock_sched();
1017
1018 pid_list = rcu_dereference_sched(tr->filtered_pids);
1019
1020 if (!pid_list)
1021 return NULL;
1022
1023 return trace_pid_start(pid_list, pos);
1024 }
1025
1026 static void p_stop(struct seq_file *m, void *p)
1027 __releases(RCU)
1028 {
1029 rcu_read_unlock_sched();
1030 mutex_unlock(&event_mutex);
1031 }
1032
1033 static ssize_t
1034 event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1035 loff_t *ppos)
1036 {
1037 struct trace_event_file *file;
1038 unsigned long flags;
1039 char buf[4] = "0";
1040
1041 mutex_lock(&event_mutex);
1042 file = event_file_data(filp);
1043 if (likely(file))
1044 flags = file->flags;
1045 mutex_unlock(&event_mutex);
1046
1047 if (!file)
1048 return -ENODEV;
1049
1050 if (flags & EVENT_FILE_FL_ENABLED &&
1051 !(flags & EVENT_FILE_FL_SOFT_DISABLED))
1052 strcpy(buf, "1");
1053
1054 if (flags & EVENT_FILE_FL_SOFT_DISABLED ||
1055 flags & EVENT_FILE_FL_SOFT_MODE)
1056 strcat(buf, "*");
1057
1058 strcat(buf, "\n");
1059
1060 return simple_read_from_buffer(ubuf, cnt, ppos, buf, strlen(buf));
1061 }
1062
1063 static ssize_t
1064 event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1065 loff_t *ppos)
1066 {
1067 struct trace_event_file *file;
1068 unsigned long val;
1069 int ret;
1070
1071 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1072 if (ret)
1073 return ret;
1074
1075 ret = tracing_update_buffers();
1076 if (ret < 0)
1077 return ret;
1078
1079 switch (val) {
1080 case 0:
1081 case 1:
1082 ret = -ENODEV;
1083 mutex_lock(&event_mutex);
1084 file = event_file_data(filp);
1085 if (likely(file))
1086 ret = ftrace_event_enable_disable(file, val);
1087 mutex_unlock(&event_mutex);
1088 break;
1089
1090 default:
1091 return -EINVAL;
1092 }
1093
1094 *ppos += cnt;
1095
1096 return ret ? ret : cnt;
1097 }
1098
1099 static ssize_t
1100 system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1101 loff_t *ppos)
1102 {
1103 const char set_to_char[4] = { '?', '0', '1', 'X' };
1104 struct trace_subsystem_dir *dir = filp->private_data;
1105 struct event_subsystem *system = dir->subsystem;
1106 struct trace_event_call *call;
1107 struct trace_event_file *file;
1108 struct trace_array *tr = dir->tr;
1109 char buf[2];
1110 int set = 0;
1111 int ret;
1112
1113 mutex_lock(&event_mutex);
1114 list_for_each_entry(file, &tr->events, list) {
1115 call = file->event_call;
1116 if (!trace_event_name(call) || !call->class || !call->class->reg)
1117 continue;
1118
1119 if (system && strcmp(call->class->system, system->name) != 0)
1120 continue;
1121
1122 /*
1123 * We need to find out if all the events are set
1124 * or if all events or cleared, or if we have
1125 * a mixture.
1126 */
1127 set |= (1 << !!(file->flags & EVENT_FILE_FL_ENABLED));
1128
1129 /*
1130 * If we have a mixture, no need to look further.
1131 */
1132 if (set == 3)
1133 break;
1134 }
1135 mutex_unlock(&event_mutex);
1136
1137 buf[0] = set_to_char[set];
1138 buf[1] = '\n';
1139
1140 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
1141
1142 return ret;
1143 }
1144
1145 static ssize_t
1146 system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1147 loff_t *ppos)
1148 {
1149 struct trace_subsystem_dir *dir = filp->private_data;
1150 struct event_subsystem *system = dir->subsystem;
1151 const char *name = NULL;
1152 unsigned long val;
1153 ssize_t ret;
1154
1155 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1156 if (ret)
1157 return ret;
1158
1159 ret = tracing_update_buffers();
1160 if (ret < 0)
1161 return ret;
1162
1163 if (val != 0 && val != 1)
1164 return -EINVAL;
1165
1166 /*
1167 * Opening of "enable" adds a ref count to system,
1168 * so the name is safe to use.
1169 */
1170 if (system)
1171 name = system->name;
1172
1173 ret = __ftrace_set_clr_event(dir->tr, NULL, name, NULL, val);
1174 if (ret)
1175 goto out;
1176
1177 ret = cnt;
1178
1179 out:
1180 *ppos += cnt;
1181
1182 return ret;
1183 }
1184
1185 enum {
1186 FORMAT_HEADER = 1,
1187 FORMAT_FIELD_SEPERATOR = 2,
1188 FORMAT_PRINTFMT = 3,
1189 };
1190
1191 static void *f_next(struct seq_file *m, void *v, loff_t *pos)
1192 {
1193 struct trace_event_call *call = event_file_data(m->private);
1194 struct list_head *common_head = &ftrace_common_fields;
1195 struct list_head *head = trace_get_fields(call);
1196 struct list_head *node = v;
1197
1198 (*pos)++;
1199
1200 switch ((unsigned long)v) {
1201 case FORMAT_HEADER:
1202 node = common_head;
1203 break;
1204
1205 case FORMAT_FIELD_SEPERATOR:
1206 node = head;
1207 break;
1208
1209 case FORMAT_PRINTFMT:
1210 /* all done */
1211 return NULL;
1212 }
1213
1214 node = node->prev;
1215 if (node == common_head)
1216 return (void *)FORMAT_FIELD_SEPERATOR;
1217 else if (node == head)
1218 return (void *)FORMAT_PRINTFMT;
1219 else
1220 return node;
1221 }
1222
1223 static int f_show(struct seq_file *m, void *v)
1224 {
1225 struct trace_event_call *call = event_file_data(m->private);
1226 struct ftrace_event_field *field;
1227 const char *array_descriptor;
1228
1229 switch ((unsigned long)v) {
1230 case FORMAT_HEADER:
1231 seq_printf(m, "name: %s\n", trace_event_name(call));
1232 seq_printf(m, "ID: %d\n", call->event.type);
1233 seq_puts(m, "format:\n");
1234 return 0;
1235
1236 case FORMAT_FIELD_SEPERATOR:
1237 seq_putc(m, '\n');
1238 return 0;
1239
1240 case FORMAT_PRINTFMT:
1241 seq_printf(m, "\nprint fmt: %s\n",
1242 call->print_fmt);
1243 return 0;
1244 }
1245
1246 field = list_entry(v, struct ftrace_event_field, link);
1247 /*
1248 * Smartly shows the array type(except dynamic array).
1249 * Normal:
1250 * field:TYPE VAR
1251 * If TYPE := TYPE[LEN], it is shown:
1252 * field:TYPE VAR[LEN]
1253 */
1254 array_descriptor = strchr(field->type, '[');
1255
1256 if (!strncmp(field->type, "__data_loc", 10))
1257 array_descriptor = NULL;
1258
1259 if (!array_descriptor)
1260 seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1261 field->type, field->name, field->offset,
1262 field->size, !!field->is_signed);
1263 else
1264 seq_printf(m, "\tfield:%.*s %s%s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1265 (int)(array_descriptor - field->type),
1266 field->type, field->name,
1267 array_descriptor, field->offset,
1268 field->size, !!field->is_signed);
1269
1270 return 0;
1271 }
1272
1273 static void *f_start(struct seq_file *m, loff_t *pos)
1274 {
1275 void *p = (void *)FORMAT_HEADER;
1276 loff_t l = 0;
1277
1278 /* ->stop() is called even if ->start() fails */
1279 mutex_lock(&event_mutex);
1280 if (!event_file_data(m->private))
1281 return ERR_PTR(-ENODEV);
1282
1283 while (l < *pos && p)
1284 p = f_next(m, p, &l);
1285
1286 return p;
1287 }
1288
1289 static void f_stop(struct seq_file *m, void *p)
1290 {
1291 mutex_unlock(&event_mutex);
1292 }
1293
1294 static const struct seq_operations trace_format_seq_ops = {
1295 .start = f_start,
1296 .next = f_next,
1297 .stop = f_stop,
1298 .show = f_show,
1299 };
1300
1301 static int trace_format_open(struct inode *inode, struct file *file)
1302 {
1303 struct seq_file *m;
1304 int ret;
1305
1306 ret = seq_open(file, &trace_format_seq_ops);
1307 if (ret < 0)
1308 return ret;
1309
1310 m = file->private_data;
1311 m->private = file;
1312
1313 return 0;
1314 }
1315
1316 static ssize_t
1317 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1318 {
1319 int id = (long)event_file_data(filp);
1320 char buf[32];
1321 int len;
1322
1323 if (unlikely(!id))
1324 return -ENODEV;
1325
1326 len = sprintf(buf, "%d\n", id);
1327
1328 return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
1329 }
1330
1331 static ssize_t
1332 event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1333 loff_t *ppos)
1334 {
1335 struct trace_event_file *file;
1336 struct trace_seq *s;
1337 int r = -ENODEV;
1338
1339 if (*ppos)
1340 return 0;
1341
1342 s = kmalloc(sizeof(*s), GFP_KERNEL);
1343
1344 if (!s)
1345 return -ENOMEM;
1346
1347 trace_seq_init(s);
1348
1349 mutex_lock(&event_mutex);
1350 file = event_file_data(filp);
1351 if (file)
1352 print_event_filter(file, s);
1353 mutex_unlock(&event_mutex);
1354
1355 if (file)
1356 r = simple_read_from_buffer(ubuf, cnt, ppos,
1357 s->buffer, trace_seq_used(s));
1358
1359 kfree(s);
1360
1361 return r;
1362 }
1363
1364 static ssize_t
1365 event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1366 loff_t *ppos)
1367 {
1368 struct trace_event_file *file;
1369 char *buf;
1370 int err = -ENODEV;
1371
1372 if (cnt >= PAGE_SIZE)
1373 return -EINVAL;
1374
1375 buf = memdup_user_nul(ubuf, cnt);
1376 if (IS_ERR(buf))
1377 return PTR_ERR(buf);
1378
1379 mutex_lock(&event_mutex);
1380 file = event_file_data(filp);
1381 if (file)
1382 err = apply_event_filter(file, buf);
1383 mutex_unlock(&event_mutex);
1384
1385 kfree(buf);
1386 if (err < 0)
1387 return err;
1388
1389 *ppos += cnt;
1390
1391 return cnt;
1392 }
1393
1394 static LIST_HEAD(event_subsystems);
1395
1396 static int subsystem_open(struct inode *inode, struct file *filp)
1397 {
1398 struct event_subsystem *system = NULL;
1399 struct trace_subsystem_dir *dir = NULL; /* Initialize for gcc */
1400 struct trace_array *tr;
1401 int ret;
1402
1403 if (tracing_is_disabled())
1404 return -ENODEV;
1405
1406 /* Make sure the system still exists */
1407 mutex_lock(&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
1446 static int system_tr_open(struct inode *inode, struct file *filp)
1447 {
1448 struct trace_subsystem_dir *dir;
1449 struct trace_array *tr = inode->i_private;
1450 int ret;
1451
1452 if (tracing_is_disabled())
1453 return -ENODEV;
1454
1455 if (trace_array_get(tr) < 0)
1456 return -ENODEV;
1457
1458 /* Make a temporary dir that has no system but points to tr */
1459 dir = kzalloc(sizeof(*dir), GFP_KERNEL);
1460 if (!dir) {
1461 trace_array_put(tr);
1462 return -ENOMEM;
1463 }
1464
1465 dir->tr = tr;
1466
1467 ret = tracing_open_generic(inode, filp);
1468 if (ret < 0) {
1469 trace_array_put(tr);
1470 kfree(dir);
1471 return ret;
1472 }
1473
1474 filp->private_data = dir;
1475
1476 return 0;
1477 }
1478
1479 static int subsystem_release(struct inode *inode, struct file *file)
1480 {
1481 struct trace_subsystem_dir *dir = file->private_data;
1482
1483 trace_array_put(dir->tr);
1484
1485 /*
1486 * If dir->subsystem is NULL, then this is a temporary
1487 * descriptor that was made for a trace_array to enable
1488 * all subsystems.
1489 */
1490 if (dir->subsystem)
1491 put_system(dir);
1492 else
1493 kfree(dir);
1494
1495 return 0;
1496 }
1497
1498 static ssize_t
1499 subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1500 loff_t *ppos)
1501 {
1502 struct trace_subsystem_dir *dir = filp->private_data;
1503 struct event_subsystem *system = dir->subsystem;
1504 struct trace_seq *s;
1505 int r;
1506
1507 if (*ppos)
1508 return 0;
1509
1510 s = kmalloc(sizeof(*s), GFP_KERNEL);
1511 if (!s)
1512 return -ENOMEM;
1513
1514 trace_seq_init(s);
1515
1516 print_subsystem_event_filter(system, s);
1517 r = simple_read_from_buffer(ubuf, cnt, ppos,
1518 s->buffer, trace_seq_used(s));
1519
1520 kfree(s);
1521
1522 return r;
1523 }
1524
1525 static ssize_t
1526 subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1527 loff_t *ppos)
1528 {
1529 struct trace_subsystem_dir *dir = filp->private_data;
1530 char *buf;
1531 int err;
1532
1533 if (cnt >= PAGE_SIZE)
1534 return -EINVAL;
1535
1536 buf = memdup_user_nul(ubuf, cnt);
1537 if (IS_ERR(buf))
1538 return PTR_ERR(buf);
1539
1540 err = apply_subsystem_event_filter(dir, buf);
1541 kfree(buf);
1542 if (err < 0)
1543 return err;
1544
1545 *ppos += cnt;
1546
1547 return cnt;
1548 }
1549
1550 static ssize_t
1551 show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1552 {
1553 int (*func)(struct trace_seq *s) = filp->private_data;
1554 struct trace_seq *s;
1555 int r;
1556
1557 if (*ppos)
1558 return 0;
1559
1560 s = kmalloc(sizeof(*s), GFP_KERNEL);
1561 if (!s)
1562 return -ENOMEM;
1563
1564 trace_seq_init(s);
1565
1566 func(s);
1567 r = simple_read_from_buffer(ubuf, cnt, ppos,
1568 s->buffer, trace_seq_used(s));
1569
1570 kfree(s);
1571
1572 return r;
1573 }
1574
1575 static void ignore_task_cpu(void *data)
1576 {
1577 struct trace_array *tr = data;
1578 struct trace_pid_list *pid_list;
1579
1580 /*
1581 * This function is called by on_each_cpu() while the
1582 * event_mutex is held.
1583 */
1584 pid_list = rcu_dereference_protected(tr->filtered_pids,
1585 mutex_is_locked(&event_mutex));
1586
1587 this_cpu_write(tr->trace_buffer.data->ignore_pid,
1588 trace_ignore_this_task(pid_list, current));
1589 }
1590
1591 static ssize_t
1592 ftrace_event_pid_write(struct file *filp, const char __user *ubuf,
1593 size_t cnt, loff_t *ppos)
1594 {
1595 struct seq_file *m = filp->private_data;
1596 struct trace_array *tr = m->private;
1597 struct trace_pid_list *filtered_pids = NULL;
1598 struct trace_pid_list *pid_list;
1599 struct trace_event_file *file;
1600 ssize_t ret;
1601
1602 if (!cnt)
1603 return 0;
1604
1605 ret = tracing_update_buffers();
1606 if (ret < 0)
1607 return ret;
1608
1609 mutex_lock(&event_mutex);
1610
1611 filtered_pids = rcu_dereference_protected(tr->filtered_pids,
1612 lockdep_is_held(&event_mutex));
1613
1614 ret = trace_pid_write(filtered_pids, &pid_list, ubuf, cnt);
1615 if (ret < 0)
1616 goto out;
1617
1618 rcu_assign_pointer(tr->filtered_pids, pid_list);
1619
1620 list_for_each_entry(file, &tr->events, list) {
1621 set_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
1622 }
1623
1624 if (filtered_pids) {
1625 tracepoint_synchronize_unregister();
1626 trace_free_pid_list(filtered_pids);
1627 } else if (pid_list) {
1628 /*
1629 * Register a probe that is called before all other probes
1630 * to set ignore_pid if next or prev do not match.
1631 * Register a probe this is called after all other probes
1632 * to only keep ignore_pid set if next pid matches.
1633 */
1634 register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_pre,
1635 tr, INT_MAX);
1636 register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_post,
1637 tr, 0);
1638
1639 register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre,
1640 tr, INT_MAX);
1641 register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_post,
1642 tr, 0);
1643
1644 register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre,
1645 tr, INT_MAX);
1646 register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post,
1647 tr, 0);
1648
1649 register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_pre,
1650 tr, INT_MAX);
1651 register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_post,
1652 tr, 0);
1653 }
1654
1655 /*
1656 * Ignoring of pids is done at task switch. But we have to
1657 * check for those tasks that are currently running.
1658 * Always do this in case a pid was appended or removed.
1659 */
1660 on_each_cpu(ignore_task_cpu, tr, 1);
1661
1662 out:
1663 mutex_unlock(&event_mutex);
1664
1665 if (ret > 0)
1666 *ppos += ret;
1667
1668 return ret;
1669 }
1670
1671 static int ftrace_event_avail_open(struct inode *inode, struct file *file);
1672 static int ftrace_event_set_open(struct inode *inode, struct file *file);
1673 static int ftrace_event_set_pid_open(struct inode *inode, struct file *file);
1674 static int ftrace_event_release(struct inode *inode, struct file *file);
1675
1676 static const struct seq_operations show_event_seq_ops = {
1677 .start = t_start,
1678 .next = t_next,
1679 .show = t_show,
1680 .stop = t_stop,
1681 };
1682
1683 static const struct seq_operations show_set_event_seq_ops = {
1684 .start = s_start,
1685 .next = s_next,
1686 .show = t_show,
1687 .stop = t_stop,
1688 };
1689
1690 static const struct seq_operations show_set_pid_seq_ops = {
1691 .start = p_start,
1692 .next = p_next,
1693 .show = trace_pid_show,
1694 .stop = p_stop,
1695 };
1696
1697 static const struct file_operations ftrace_avail_fops = {
1698 .open = ftrace_event_avail_open,
1699 .read = seq_read,
1700 .llseek = seq_lseek,
1701 .release = seq_release,
1702 };
1703
1704 static const struct file_operations ftrace_set_event_fops = {
1705 .open = ftrace_event_set_open,
1706 .read = seq_read,
1707 .write = ftrace_event_write,
1708 .llseek = seq_lseek,
1709 .release = ftrace_event_release,
1710 };
1711
1712 static const struct file_operations ftrace_set_event_pid_fops = {
1713 .open = ftrace_event_set_pid_open,
1714 .read = seq_read,
1715 .write = ftrace_event_pid_write,
1716 .llseek = seq_lseek,
1717 .release = ftrace_event_release,
1718 };
1719
1720 static const struct file_operations ftrace_enable_fops = {
1721 .open = tracing_open_generic,
1722 .read = event_enable_read,
1723 .write = event_enable_write,
1724 .llseek = default_llseek,
1725 };
1726
1727 static const struct file_operations ftrace_event_format_fops = {
1728 .open = trace_format_open,
1729 .read = seq_read,
1730 .llseek = seq_lseek,
1731 .release = seq_release,
1732 };
1733
1734 static const struct file_operations ftrace_event_id_fops = {
1735 .read = event_id_read,
1736 .llseek = default_llseek,
1737 };
1738
1739 static const struct file_operations ftrace_event_filter_fops = {
1740 .open = tracing_open_generic,
1741 .read = event_filter_read,
1742 .write = event_filter_write,
1743 .llseek = default_llseek,
1744 };
1745
1746 static const struct file_operations ftrace_subsystem_filter_fops = {
1747 .open = subsystem_open,
1748 .read = subsystem_filter_read,
1749 .write = subsystem_filter_write,
1750 .llseek = default_llseek,
1751 .release = subsystem_release,
1752 };
1753
1754 static const struct file_operations ftrace_system_enable_fops = {
1755 .open = subsystem_open,
1756 .read = system_enable_read,
1757 .write = system_enable_write,
1758 .llseek = default_llseek,
1759 .release = subsystem_release,
1760 };
1761
1762 static const struct file_operations ftrace_tr_enable_fops = {
1763 .open = system_tr_open,
1764 .read = system_enable_read,
1765 .write = system_enable_write,
1766 .llseek = default_llseek,
1767 .release = subsystem_release,
1768 };
1769
1770 static const struct file_operations ftrace_show_header_fops = {
1771 .open = tracing_open_generic,
1772 .read = show_header,
1773 .llseek = default_llseek,
1774 };
1775
1776 static int
1777 ftrace_event_open(struct inode *inode, struct file *file,
1778 const struct seq_operations *seq_ops)
1779 {
1780 struct seq_file *m;
1781 int ret;
1782
1783 ret = seq_open(file, seq_ops);
1784 if (ret < 0)
1785 return ret;
1786 m = file->private_data;
1787 /* copy tr over to seq ops */
1788 m->private = inode->i_private;
1789
1790 return ret;
1791 }
1792
1793 static int ftrace_event_release(struct inode *inode, struct file *file)
1794 {
1795 struct trace_array *tr = inode->i_private;
1796
1797 trace_array_put(tr);
1798
1799 return seq_release(inode, file);
1800 }
1801
1802 static int
1803 ftrace_event_avail_open(struct inode *inode, struct file *file)
1804 {
1805 const struct seq_operations *seq_ops = &show_event_seq_ops;
1806
1807 return ftrace_event_open(inode, file, seq_ops);
1808 }
1809
1810 static int
1811 ftrace_event_set_open(struct inode *inode, struct file *file)
1812 {
1813 const struct seq_operations *seq_ops = &show_set_event_seq_ops;
1814 struct trace_array *tr = inode->i_private;
1815 int ret;
1816
1817 if (trace_array_get(tr) < 0)
1818 return -ENODEV;
1819
1820 if ((file->f_mode & FMODE_WRITE) &&
1821 (file->f_flags & O_TRUNC))
1822 ftrace_clear_events(tr);
1823
1824 ret = ftrace_event_open(inode, file, seq_ops);
1825 if (ret < 0)
1826 trace_array_put(tr);
1827 return ret;
1828 }
1829
1830 static int
1831 ftrace_event_set_pid_open(struct inode *inode, struct file *file)
1832 {
1833 const struct seq_operations *seq_ops = &show_set_pid_seq_ops;
1834 struct trace_array *tr = inode->i_private;
1835 int ret;
1836
1837 if (trace_array_get(tr) < 0)
1838 return -ENODEV;
1839
1840 if ((file->f_mode & FMODE_WRITE) &&
1841 (file->f_flags & O_TRUNC))
1842 ftrace_clear_event_pids(tr);
1843
1844 ret = ftrace_event_open(inode, file, seq_ops);
1845 if (ret < 0)
1846 trace_array_put(tr);
1847 return ret;
1848 }
1849
1850 static struct event_subsystem *
1851 create_new_subsystem(const char *name)
1852 {
1853 struct event_subsystem *system;
1854
1855 /* need to create new entry */
1856 system = kmalloc(sizeof(*system), GFP_KERNEL);
1857 if (!system)
1858 return NULL;
1859
1860 system->ref_count = 1;
1861
1862 /* Only allocate if dynamic (kprobes and modules) */
1863 system->name = kstrdup_const(name, GFP_KERNEL);
1864 if (!system->name)
1865 goto out_free;
1866
1867 system->filter = NULL;
1868
1869 system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
1870 if (!system->filter)
1871 goto out_free;
1872
1873 list_add(&system->list, &event_subsystems);
1874
1875 return system;
1876
1877 out_free:
1878 kfree_const(system->name);
1879 kfree(system);
1880 return NULL;
1881 }
1882
1883 static struct dentry *
1884 event_subsystem_dir(struct trace_array *tr, const char *name,
1885 struct trace_event_file *file, struct dentry *parent)
1886 {
1887 struct trace_subsystem_dir *dir;
1888 struct event_subsystem *system;
1889 struct dentry *entry;
1890
1891 /* First see if we did not already create this dir */
1892 list_for_each_entry(dir, &tr->systems, list) {
1893 system = dir->subsystem;
1894 if (strcmp(system->name, name) == 0) {
1895 dir->nr_events++;
1896 file->system = dir;
1897 return dir->entry;
1898 }
1899 }
1900
1901 /* Now see if the system itself exists. */
1902 list_for_each_entry(system, &event_subsystems, list) {
1903 if (strcmp(system->name, name) == 0)
1904 break;
1905 }
1906 /* Reset system variable when not found */
1907 if (&system->list == &event_subsystems)
1908 system = NULL;
1909
1910 dir = kmalloc(sizeof(*dir), GFP_KERNEL);
1911 if (!dir)
1912 goto out_fail;
1913
1914 if (!system) {
1915 system = create_new_subsystem(name);
1916 if (!system)
1917 goto out_free;
1918 } else
1919 __get_system(system);
1920
1921 dir->entry = tracefs_create_dir(name, parent);
1922 if (!dir->entry) {
1923 pr_warn("Failed to create system directory %s\n", name);
1924 __put_system(system);
1925 goto out_free;
1926 }
1927
1928 dir->tr = tr;
1929 dir->ref_count = 1;
1930 dir->nr_events = 1;
1931 dir->subsystem = system;
1932 file->system = dir;
1933
1934 entry = tracefs_create_file("filter", 0644, dir->entry, dir,
1935 &ftrace_subsystem_filter_fops);
1936 if (!entry) {
1937 kfree(system->filter);
1938 system->filter = NULL;
1939 pr_warn("Could not create tracefs '%s/filter' entry\n", name);
1940 }
1941
1942 trace_create_file("enable", 0644, dir->entry, dir,
1943 &ftrace_system_enable_fops);
1944
1945 list_add(&dir->list, &tr->systems);
1946
1947 return dir->entry;
1948
1949 out_free:
1950 kfree(dir);
1951 out_fail:
1952 /* Only print this message if failed on memory allocation */
1953 if (!dir || !system)
1954 pr_warn("No memory to create event subsystem %s\n", name);
1955 return NULL;
1956 }
1957
1958 static int
1959 event_create_dir(struct dentry *parent, struct trace_event_file *file)
1960 {
1961 struct trace_event_call *call = file->event_call;
1962 struct trace_array *tr = file->tr;
1963 struct list_head *head;
1964 struct dentry *d_events;
1965 const char *name;
1966 int ret;
1967
1968 /*
1969 * If the trace point header did not define TRACE_SYSTEM
1970 * then the system would be called "TRACE_SYSTEM".
1971 */
1972 if (strcmp(call->class->system, TRACE_SYSTEM) != 0) {
1973 d_events = event_subsystem_dir(tr, call->class->system, file, parent);
1974 if (!d_events)
1975 return -ENOMEM;
1976 } else
1977 d_events = parent;
1978
1979 name = trace_event_name(call);
1980 file->dir = tracefs_create_dir(name, d_events);
1981 if (!file->dir) {
1982 pr_warn("Could not create tracefs '%s' directory\n", name);
1983 return -1;
1984 }
1985
1986 if (call->class->reg && !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
1987 trace_create_file("enable", 0644, file->dir, file,
1988 &ftrace_enable_fops);
1989
1990 #ifdef CONFIG_PERF_EVENTS
1991 if (call->event.type && call->class->reg)
1992 trace_create_file("id", 0444, file->dir,
1993 (void *)(long)call->event.type,
1994 &ftrace_event_id_fops);
1995 #endif
1996
1997 /*
1998 * Other events may have the same class. Only update
1999 * the fields if they are not already defined.
2000 */
2001 head = trace_get_fields(call);
2002 if (list_empty(head)) {
2003 ret = call->class->define_fields(call);
2004 if (ret < 0) {
2005 pr_warn("Could not initialize trace point events/%s\n",
2006 name);
2007 return -1;
2008 }
2009 }
2010
2011 /*
2012 * Only event directories that can be enabled should have
2013 * triggers or filters.
2014 */
2015 if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)) {
2016 trace_create_file("filter", 0644, file->dir, file,
2017 &ftrace_event_filter_fops);
2018
2019 trace_create_file("trigger", 0644, file->dir, file,
2020 &event_trigger_fops);
2021 }
2022
2023 #ifdef CONFIG_HIST_TRIGGERS
2024 trace_create_file("hist", 0444, file->dir, file,
2025 &event_hist_fops);
2026 #endif
2027 trace_create_file("format", 0444, file->dir, call,
2028 &ftrace_event_format_fops);
2029
2030 return 0;
2031 }
2032
2033 static void remove_event_from_tracers(struct trace_event_call *call)
2034 {
2035 struct trace_event_file *file;
2036 struct trace_array *tr;
2037
2038 do_for_each_event_file_safe(tr, file) {
2039 if (file->event_call != call)
2040 continue;
2041
2042 remove_event_file_dir(file);
2043 /*
2044 * The do_for_each_event_file_safe() is
2045 * a double loop. After finding the call for this
2046 * trace_array, we use break to jump to the next
2047 * trace_array.
2048 */
2049 break;
2050 } while_for_each_event_file();
2051 }
2052
2053 static void event_remove(struct trace_event_call *call)
2054 {
2055 struct trace_array *tr;
2056 struct trace_event_file *file;
2057
2058 do_for_each_event_file(tr, file) {
2059 if (file->event_call != call)
2060 continue;
2061
2062 if (file->flags & EVENT_FILE_FL_WAS_ENABLED)
2063 tr->clear_trace = true;
2064
2065 ftrace_event_enable_disable(file, 0);
2066 /*
2067 * The do_for_each_event_file() is
2068 * a double loop. After finding the call for this
2069 * trace_array, we use break to jump to the next
2070 * trace_array.
2071 */
2072 break;
2073 } while_for_each_event_file();
2074
2075 if (call->event.funcs)
2076 __unregister_trace_event(&call->event);
2077 remove_event_from_tracers(call);
2078 list_del(&call->list);
2079 }
2080
2081 static int event_init(struct trace_event_call *call)
2082 {
2083 int ret = 0;
2084 const char *name;
2085
2086 name = trace_event_name(call);
2087 if (WARN_ON(!name))
2088 return -EINVAL;
2089
2090 if (call->class->raw_init) {
2091 ret = call->class->raw_init(call);
2092 if (ret < 0 && ret != -ENOSYS)
2093 pr_warn("Could not initialize trace events/%s\n", name);
2094 }
2095
2096 return ret;
2097 }
2098
2099 static int
2100 __register_event(struct trace_event_call *call, struct module *mod)
2101 {
2102 int ret;
2103
2104 ret = event_init(call);
2105 if (ret < 0)
2106 return ret;
2107
2108 list_add(&call->list, &ftrace_events);
2109 call->mod = mod;
2110
2111 return 0;
2112 }
2113
2114 static char *eval_replace(char *ptr, struct trace_eval_map *map, int len)
2115 {
2116 int rlen;
2117 int elen;
2118
2119 /* Find the length of the eval value as a string */
2120 elen = snprintf(ptr, 0, "%ld", map->eval_value);
2121 /* Make sure there's enough room to replace the string with the value */
2122 if (len < elen)
2123 return NULL;
2124
2125 snprintf(ptr, elen + 1, "%ld", map->eval_value);
2126
2127 /* Get the rest of the string of ptr */
2128 rlen = strlen(ptr + len);
2129 memmove(ptr + elen, ptr + len, rlen);
2130 /* Make sure we end the new string */
2131 ptr[elen + rlen] = 0;
2132
2133 return ptr + elen;
2134 }
2135
2136 static void update_event_printk(struct trace_event_call *call,
2137 struct trace_eval_map *map)
2138 {
2139 char *ptr;
2140 int quote = 0;
2141 int len = strlen(map->eval_string);
2142
2143 for (ptr = call->print_fmt; *ptr; ptr++) {
2144 if (*ptr == '\\') {
2145 ptr++;
2146 /* paranoid */
2147 if (!*ptr)
2148 break;
2149 continue;
2150 }
2151 if (*ptr == '"') {
2152 quote ^= 1;
2153 continue;
2154 }
2155 if (quote)
2156 continue;
2157 if (isdigit(*ptr)) {
2158 /* skip numbers */
2159 do {
2160 ptr++;
2161 /* Check for alpha chars like ULL */
2162 } while (isalnum(*ptr));
2163 if (!*ptr)
2164 break;
2165 /*
2166 * A number must have some kind of delimiter after
2167 * it, and we can ignore that too.
2168 */
2169 continue;
2170 }
2171 if (isalpha(*ptr) || *ptr == '_') {
2172 if (strncmp(map->eval_string, ptr, len) == 0 &&
2173 !isalnum(ptr[len]) && ptr[len] != '_') {
2174 ptr = eval_replace(ptr, map, len);
2175 /* enum/sizeof string smaller than value */
2176 if (WARN_ON_ONCE(!ptr))
2177 return;
2178 /*
2179 * No need to decrement here, as eval_replace()
2180 * returns the pointer to the character passed
2181 * the eval, and two evals can not be placed
2182 * back to back without something in between.
2183 * We can skip that something in between.
2184 */
2185 continue;
2186 }
2187 skip_more:
2188 do {
2189 ptr++;
2190 } while (isalnum(*ptr) || *ptr == '_');
2191 if (!*ptr)
2192 break;
2193 /*
2194 * If what comes after this variable is a '.' or
2195 * '->' then we can continue to ignore that string.
2196 */
2197 if (*ptr == '.' || (ptr[0] == '-' && ptr[1] == '>')) {
2198 ptr += *ptr == '.' ? 1 : 2;
2199 if (!*ptr)
2200 break;
2201 goto skip_more;
2202 }
2203 /*
2204 * Once again, we can skip the delimiter that came
2205 * after the string.
2206 */
2207 continue;
2208 }
2209 }
2210 }
2211
2212 void trace_event_eval_update(struct trace_eval_map **map, int len)
2213 {
2214 struct trace_event_call *call, *p;
2215 const char *last_system = NULL;
2216 bool first = false;
2217 int last_i;
2218 int i;
2219
2220 down_write(&trace_event_sem);
2221 list_for_each_entry_safe(call, p, &ftrace_events, list) {
2222 /* events are usually grouped together with systems */
2223 if (!last_system || call->class->system != last_system) {
2224 first = true;
2225 last_i = 0;
2226 last_system = call->class->system;
2227 }
2228
2229 /*
2230 * Since calls are grouped by systems, the likelyhood that the
2231 * next call in the iteration belongs to the same system as the
2232 * previous call is high. As an optimization, we skip seaching
2233 * for a map[] that matches the call's system if the last call
2234 * was from the same system. That's what last_i is for. If the
2235 * call has the same system as the previous call, then last_i
2236 * will be the index of the first map[] that has a matching
2237 * system.
2238 */
2239 for (i = last_i; i < len; i++) {
2240 if (call->class->system == map[i]->system) {
2241 /* Save the first system if need be */
2242 if (first) {
2243 last_i = i;
2244 first = false;
2245 }
2246 update_event_printk(call, map[i]);
2247 }
2248 }
2249 }
2250 up_write(&trace_event_sem);
2251 }
2252
2253 static struct trace_event_file *
2254 trace_create_new_event(struct trace_event_call *call,
2255 struct trace_array *tr)
2256 {
2257 struct trace_event_file *file;
2258
2259 file = kmem_cache_alloc(file_cachep, GFP_TRACE);
2260 if (!file)
2261 return NULL;
2262
2263 file->event_call = call;
2264 file->tr = tr;
2265 atomic_set(&file->sm_ref, 0);
2266 atomic_set(&file->tm_ref, 0);
2267 INIT_LIST_HEAD(&file->triggers);
2268 list_add(&file->list, &tr->events);
2269
2270 return file;
2271 }
2272
2273 /* Add an event to a trace directory */
2274 static int
2275 __trace_add_new_event(struct trace_event_call *call, struct trace_array *tr)
2276 {
2277 struct trace_event_file *file;
2278
2279 file = trace_create_new_event(call, tr);
2280 if (!file)
2281 return -ENOMEM;
2282
2283 return event_create_dir(tr->event_dir, file);
2284 }
2285
2286 /*
2287 * Just create a decriptor for early init. A descriptor is required
2288 * for enabling events at boot. We want to enable events before
2289 * the filesystem is initialized.
2290 */
2291 static __init int
2292 __trace_early_add_new_event(struct trace_event_call *call,
2293 struct trace_array *tr)
2294 {
2295 struct trace_event_file *file;
2296
2297 file = trace_create_new_event(call, tr);
2298 if (!file)
2299 return -ENOMEM;
2300
2301 return 0;
2302 }
2303
2304 struct ftrace_module_file_ops;
2305 static void __add_event_to_tracers(struct trace_event_call *call);
2306
2307 int trace_add_event_call_nolock(struct trace_event_call *call)
2308 {
2309 int ret;
2310 lockdep_assert_held(&event_mutex);
2311
2312 mutex_lock(&trace_types_lock);
2313
2314 ret = __register_event(call, NULL);
2315 if (ret >= 0)
2316 __add_event_to_tracers(call);
2317
2318 mutex_unlock(&trace_types_lock);
2319 return ret;
2320 }
2321
2322 /* Add an additional event_call dynamically */
2323 int trace_add_event_call(struct trace_event_call *call)
2324 {
2325 int ret;
2326
2327 mutex_lock(&event_mutex);
2328 ret = trace_add_event_call_nolock(call);
2329 mutex_unlock(&event_mutex);
2330 return ret;
2331 }
2332
2333 /*
2334 * Must be called under locking of trace_types_lock, event_mutex and
2335 * trace_event_sem.
2336 */
2337 static void __trace_remove_event_call(struct trace_event_call *call)
2338 {
2339 event_remove(call);
2340 trace_destroy_fields(call);
2341 free_event_filter(call->filter);
2342 call->filter = NULL;
2343 }
2344
2345 static int probe_remove_event_call(struct trace_event_call *call)
2346 {
2347 struct trace_array *tr;
2348 struct trace_event_file *file;
2349
2350 #ifdef CONFIG_PERF_EVENTS
2351 if (call->perf_refcount)
2352 return -EBUSY;
2353 #endif
2354 do_for_each_event_file(tr, file) {
2355 if (file->event_call != call)
2356 continue;
2357 /*
2358 * We can't rely on ftrace_event_enable_disable(enable => 0)
2359 * we are going to do, EVENT_FILE_FL_SOFT_MODE can suppress
2360 * TRACE_REG_UNREGISTER.
2361 */
2362 if (file->flags & EVENT_FILE_FL_ENABLED)
2363 return -EBUSY;
2364 /*
2365 * The do_for_each_event_file_safe() is
2366 * a double loop. After finding the call for this
2367 * trace_array, we use break to jump to the next
2368 * trace_array.
2369 */
2370 break;
2371 } while_for_each_event_file();
2372
2373 __trace_remove_event_call(call);
2374
2375 return 0;
2376 }
2377
2378 /* no event_mutex version */
2379 int trace_remove_event_call_nolock(struct trace_event_call *call)
2380 {
2381 int ret;
2382
2383 lockdep_assert_held(&event_mutex);
2384
2385 mutex_lock(&trace_types_lock);
2386 down_write(&trace_event_sem);
2387 ret = probe_remove_event_call(call);
2388 up_write(&trace_event_sem);
2389 mutex_unlock(&trace_types_lock);
2390
2391 return ret;
2392 }
2393
2394 /* Remove an event_call */
2395 int trace_remove_event_call(struct trace_event_call *call)
2396 {
2397 int ret;
2398
2399 mutex_lock(&event_mutex);
2400 ret = trace_remove_event_call_nolock(call);
2401 mutex_unlock(&event_mutex);
2402
2403 return ret;
2404 }
2405
2406 #define for_each_event(event, start, end) \
2407 for (event = start; \
2408 (unsigned long)event < (unsigned long)end; \
2409 event++)
2410
2411 #ifdef CONFIG_MODULES
2412
2413 static void trace_module_add_events(struct module *mod)
2414 {
2415 struct trace_event_call **call, **start, **end;
2416
2417 if (!mod->num_trace_events)
2418 return;
2419
2420 /* Don't add infrastructure for mods without tracepoints */
2421 if (trace_module_has_bad_taint(mod)) {
2422 pr_err("%s: module has bad taint, not creating trace events\n",
2423 mod->name);
2424 return;
2425 }
2426
2427 start = mod->trace_events;
2428 end = mod->trace_events + mod->num_trace_events;
2429
2430 for_each_event(call, start, end) {
2431 __register_event(*call, mod);
2432 __add_event_to_tracers(*call);
2433 }
2434 }
2435
2436 static void trace_module_remove_events(struct module *mod)
2437 {
2438 struct trace_event_call *call, *p;
2439
2440 down_write(&trace_event_sem);
2441 list_for_each_entry_safe(call, p, &ftrace_events, list) {
2442 if (call->mod == mod)
2443 __trace_remove_event_call(call);
2444 }
2445 up_write(&trace_event_sem);
2446
2447 /*
2448 * It is safest to reset the ring buffer if the module being unloaded
2449 * registered any events that were used. The only worry is if
2450 * a new module gets loaded, and takes on the same id as the events
2451 * of this module. When printing out the buffer, traced events left
2452 * over from this module may be passed to the new module events and
2453 * unexpected results may occur.
2454 */
2455 tracing_reset_all_online_cpus();
2456 }
2457
2458 static int trace_module_notify(struct notifier_block *self,
2459 unsigned long val, void *data)
2460 {
2461 struct module *mod = data;
2462
2463 mutex_lock(&event_mutex);
2464 mutex_lock(&trace_types_lock);
2465 switch (val) {
2466 case MODULE_STATE_COMING:
2467 trace_module_add_events(mod);
2468 break;
2469 case MODULE_STATE_GOING:
2470 trace_module_remove_events(mod);
2471 break;
2472 }
2473 mutex_unlock(&trace_types_lock);
2474 mutex_unlock(&event_mutex);
2475
2476 return 0;
2477 }
2478
2479 static struct notifier_block trace_module_nb = {
2480 .notifier_call = trace_module_notify,
2481 .priority = 1, /* higher than trace.c module notify */
2482 };
2483 #endif /* CONFIG_MODULES */
2484
2485 /* Create a new event directory structure for a trace directory. */
2486 static void
2487 __trace_add_event_dirs(struct trace_array *tr)
2488 {
2489 struct trace_event_call *call;
2490 int ret;
2491
2492 list_for_each_entry(call, &ftrace_events, list) {
2493 ret = __trace_add_new_event(call, tr);
2494 if (ret < 0)
2495 pr_warn("Could not create directory for event %s\n",
2496 trace_event_name(call));
2497 }
2498 }
2499
2500 /* Returns any file that matches the system and event */
2501 struct trace_event_file *
2502 __find_event_file(struct trace_array *tr, const char *system, const char *event)
2503 {
2504 struct trace_event_file *file;
2505 struct trace_event_call *call;
2506 const char *name;
2507
2508 list_for_each_entry(file, &tr->events, list) {
2509
2510 call = file->event_call;
2511 name = trace_event_name(call);
2512
2513 if (!name || !call->class)
2514 continue;
2515
2516 if (strcmp(event, name) == 0 &&
2517 strcmp(system, call->class->system) == 0)
2518 return file;
2519 }
2520 return NULL;
2521 }
2522
2523 /* Returns valid trace event files that match system and event */
2524 struct trace_event_file *
2525 find_event_file(struct trace_array *tr, const char *system, const char *event)
2526 {
2527 struct trace_event_file *file;
2528
2529 file = __find_event_file(tr, system, event);
2530 if (!file || !file->event_call->class->reg ||
2531 file->event_call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
2532 return NULL;
2533
2534 return file;
2535 }
2536
2537 #ifdef CONFIG_DYNAMIC_FTRACE
2538
2539 /* Avoid typos */
2540 #define ENABLE_EVENT_STR "enable_event"
2541 #define DISABLE_EVENT_STR "disable_event"
2542
2543 struct event_probe_data {
2544 struct trace_event_file *file;
2545 unsigned long count;
2546 int ref;
2547 bool enable;
2548 };
2549
2550 static void update_event_probe(struct event_probe_data *data)
2551 {
2552 if (data->enable)
2553 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
2554 else
2555 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
2556 }
2557
2558 static void
2559 event_enable_probe(unsigned long ip, unsigned long parent_ip,
2560 struct trace_array *tr, struct ftrace_probe_ops *ops,
2561 void *data)
2562 {
2563 struct ftrace_func_mapper *mapper = data;
2564 struct event_probe_data *edata;
2565 void **pdata;
2566
2567 pdata = ftrace_func_mapper_find_ip(mapper, ip);
2568 if (!pdata || !*pdata)
2569 return;
2570
2571 edata = *pdata;
2572 update_event_probe(edata);
2573 }
2574
2575 static void
2576 event_enable_count_probe(unsigned long ip, unsigned long parent_ip,
2577 struct trace_array *tr, struct ftrace_probe_ops *ops,
2578 void *data)
2579 {
2580 struct ftrace_func_mapper *mapper = data;
2581 struct event_probe_data *edata;
2582 void **pdata;
2583
2584 pdata = ftrace_func_mapper_find_ip(mapper, ip);
2585 if (!pdata || !*pdata)
2586 return;
2587
2588 edata = *pdata;
2589
2590 if (!edata->count)
2591 return;
2592
2593 /* Skip if the event is in a state we want to switch to */
2594 if (edata->enable == !(edata->file->flags & EVENT_FILE_FL_SOFT_DISABLED))
2595 return;
2596
2597 if (edata->count != -1)
2598 (edata->count)--;
2599
2600 update_event_probe(edata);
2601 }
2602
2603 static int
2604 event_enable_print(struct seq_file *m, unsigned long ip,
2605 struct ftrace_probe_ops *ops, void *data)
2606 {
2607 struct ftrace_func_mapper *mapper = data;
2608 struct event_probe_data *edata;
2609 void **pdata;
2610
2611 pdata = ftrace_func_mapper_find_ip(mapper, ip);
2612
2613 if (WARN_ON_ONCE(!pdata || !*pdata))
2614 return 0;
2615
2616 edata = *pdata;
2617
2618 seq_printf(m, "%ps:", (void *)ip);
2619
2620 seq_printf(m, "%s:%s:%s",
2621 edata->enable ? ENABLE_EVENT_STR : DISABLE_EVENT_STR,
2622 edata->file->event_call->class->system,
2623 trace_event_name(edata->file->event_call));
2624
2625 if (edata->count == -1)
2626 seq_puts(m, ":unlimited\n");
2627 else
2628 seq_printf(m, ":count=%ld\n", edata->count);
2629
2630 return 0;
2631 }
2632
2633 static int
2634 event_enable_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
2635 unsigned long ip, void *init_data, void **data)
2636 {
2637 struct ftrace_func_mapper *mapper = *data;
2638 struct event_probe_data *edata = init_data;
2639 int ret;
2640
2641 if (!mapper) {
2642 mapper = allocate_ftrace_func_mapper();
2643 if (!mapper)
2644 return -ENODEV;
2645 *data = mapper;
2646 }
2647
2648 ret = ftrace_func_mapper_add_ip(mapper, ip, edata);
2649 if (ret < 0)
2650 return ret;
2651
2652 edata->ref++;
2653
2654 return 0;
2655 }
2656
2657 static int free_probe_data(void *data)
2658 {
2659 struct event_probe_data *edata = data;
2660
2661 edata->ref--;
2662 if (!edata->ref) {
2663 /* Remove the SOFT_MODE flag */
2664 __ftrace_event_enable_disable(edata->file, 0, 1);
2665 module_put(edata->file->event_call->mod);
2666 kfree(edata);
2667 }
2668 return 0;
2669 }
2670
2671 static void
2672 event_enable_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
2673 unsigned long ip, void *data)
2674 {
2675 struct ftrace_func_mapper *mapper = data;
2676 struct event_probe_data *edata;
2677
2678 if (!ip) {
2679 if (!mapper)
2680 return;
2681 free_ftrace_func_mapper(mapper, free_probe_data);
2682 return;
2683 }
2684
2685 edata = ftrace_func_mapper_remove_ip(mapper, ip);
2686
2687 if (WARN_ON_ONCE(!edata))
2688 return;
2689
2690 if (WARN_ON_ONCE(edata->ref <= 0))
2691 return;
2692
2693 free_probe_data(edata);
2694 }
2695
2696 static struct ftrace_probe_ops event_enable_probe_ops = {
2697 .func = event_enable_probe,
2698 .print = event_enable_print,
2699 .init = event_enable_init,
2700 .free = event_enable_free,
2701 };
2702
2703 static struct ftrace_probe_ops event_enable_count_probe_ops = {
2704 .func = event_enable_count_probe,
2705 .print = event_enable_print,
2706 .init = event_enable_init,
2707 .free = event_enable_free,
2708 };
2709
2710 static struct ftrace_probe_ops event_disable_probe_ops = {
2711 .func = event_enable_probe,
2712 .print = event_enable_print,
2713 .init = event_enable_init,
2714 .free = event_enable_free,
2715 };
2716
2717 static struct ftrace_probe_ops event_disable_count_probe_ops = {
2718 .func = event_enable_count_probe,
2719 .print = event_enable_print,
2720 .init = event_enable_init,
2721 .free = event_enable_free,
2722 };
2723
2724 static int
2725 event_enable_func(struct trace_array *tr, struct ftrace_hash *hash,
2726 char *glob, char *cmd, char *param, int enabled)
2727 {
2728 struct trace_event_file *file;
2729 struct ftrace_probe_ops *ops;
2730 struct event_probe_data *data;
2731 const char *system;
2732 const char *event;
2733 char *number;
2734 bool enable;
2735 int ret;
2736
2737 if (!tr)
2738 return -ENODEV;
2739
2740 /* hash funcs only work with set_ftrace_filter */
2741 if (!enabled || !param)
2742 return -EINVAL;
2743
2744 system = strsep(&param, ":");
2745 if (!param)
2746 return -EINVAL;
2747
2748 event = strsep(&param, ":");
2749
2750 mutex_lock(&event_mutex);
2751
2752 ret = -EINVAL;
2753 file = find_event_file(tr, system, event);
2754 if (!file)
2755 goto out;
2756
2757 enable = strcmp(cmd, ENABLE_EVENT_STR) == 0;
2758
2759 if (enable)
2760 ops = param ? &event_enable_count_probe_ops : &event_enable_probe_ops;
2761 else
2762 ops = param ? &event_disable_count_probe_ops : &event_disable_probe_ops;
2763
2764 if (glob[0] == '!') {
2765 ret = unregister_ftrace_function_probe_func(glob+1, tr, ops);
2766 goto out;
2767 }
2768
2769 ret = -ENOMEM;
2770
2771 data = kzalloc(sizeof(*data), GFP_KERNEL);
2772 if (!data)
2773 goto out;
2774
2775 data->enable = enable;
2776 data->count = -1;
2777 data->file = file;
2778
2779 if (!param)
2780 goto out_reg;
2781
2782 number = strsep(&param, ":");
2783
2784 ret = -EINVAL;
2785 if (!strlen(number))
2786 goto out_free;
2787
2788 /*
2789 * We use the callback data field (which is a pointer)
2790 * as our counter.
2791 */
2792 ret = kstrtoul(number, 0, &data->count);
2793 if (ret)
2794 goto out_free;
2795
2796 out_reg:
2797 /* Don't let event modules unload while probe registered */
2798 ret = try_module_get(file->event_call->mod);
2799 if (!ret) {
2800 ret = -EBUSY;
2801 goto out_free;
2802 }
2803
2804 ret = __ftrace_event_enable_disable(file, 1, 1);
2805 if (ret < 0)
2806 goto out_put;
2807
2808 ret = register_ftrace_function_probe(glob, tr, ops, data);
2809 /*
2810 * The above returns on success the # of functions enabled,
2811 * but if it didn't find any functions it returns zero.
2812 * Consider no functions a failure too.
2813 */
2814 if (!ret) {
2815 ret = -ENOENT;
2816 goto out_disable;
2817 } else if (ret < 0)
2818 goto out_disable;
2819 /* Just return zero, not the number of enabled functions */
2820 ret = 0;
2821 out:
2822 mutex_unlock(&event_mutex);
2823 return ret;
2824
2825 out_disable:
2826 __ftrace_event_enable_disable(file, 0, 1);
2827 out_put:
2828 module_put(file->event_call->mod);
2829 out_free:
2830 kfree(data);
2831 goto out;
2832 }
2833
2834 static struct ftrace_func_command event_enable_cmd = {
2835 .name = ENABLE_EVENT_STR,
2836 .func = event_enable_func,
2837 };
2838
2839 static struct ftrace_func_command event_disable_cmd = {
2840 .name = DISABLE_EVENT_STR,
2841 .func = event_enable_func,
2842 };
2843
2844 static __init int register_event_cmds(void)
2845 {
2846 int ret;
2847
2848 ret = register_ftrace_command(&event_enable_cmd);
2849 if (WARN_ON(ret < 0))
2850 return ret;
2851 ret = register_ftrace_command(&event_disable_cmd);
2852 if (WARN_ON(ret < 0))
2853 unregister_ftrace_command(&event_enable_cmd);
2854 return ret;
2855 }
2856 #else
2857 static inline int register_event_cmds(void) { return 0; }
2858 #endif /* CONFIG_DYNAMIC_FTRACE */
2859
2860 /*
2861 * The top level array has already had its trace_event_file
2862 * descriptors created in order to allow for early events to
2863 * be recorded. This function is called after the tracefs has been
2864 * initialized, and we now have to create the files associated
2865 * to the events.
2866 */
2867 static __init void
2868 __trace_early_add_event_dirs(struct trace_array *tr)
2869 {
2870 struct trace_event_file *file;
2871 int ret;
2872
2873
2874 list_for_each_entry(file, &tr->events, list) {
2875 ret = event_create_dir(tr->event_dir, file);
2876 if (ret < 0)
2877 pr_warn("Could not create directory for event %s\n",
2878 trace_event_name(file->event_call));
2879 }
2880 }
2881
2882 /*
2883 * For early boot up, the top trace array requires to have
2884 * a list of events that can be enabled. This must be done before
2885 * the filesystem is set up in order to allow events to be traced
2886 * early.
2887 */
2888 static __init void
2889 __trace_early_add_events(struct trace_array *tr)
2890 {
2891 struct trace_event_call *call;
2892 int ret;
2893
2894 list_for_each_entry(call, &ftrace_events, list) {
2895 /* Early boot up should not have any modules loaded */
2896 if (WARN_ON_ONCE(call->mod))
2897 continue;
2898
2899 ret = __trace_early_add_new_event(call, tr);
2900 if (ret < 0)
2901 pr_warn("Could not create early event %s\n",
2902 trace_event_name(call));
2903 }
2904 }
2905
2906 /* Remove the event directory structure for a trace directory. */
2907 static void
2908 __trace_remove_event_dirs(struct trace_array *tr)
2909 {
2910 struct trace_event_file *file, *next;
2911
2912 list_for_each_entry_safe(file, next, &tr->events, list)
2913 remove_event_file_dir(file);
2914 }
2915
2916 static void __add_event_to_tracers(struct trace_event_call *call)
2917 {
2918 struct trace_array *tr;
2919
2920 list_for_each_entry(tr, &ftrace_trace_arrays, list)
2921 __trace_add_new_event(call, tr);
2922 }
2923
2924 extern struct trace_event_call *__start_ftrace_events[];
2925 extern struct trace_event_call *__stop_ftrace_events[];
2926
2927 static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
2928
2929 static __init int setup_trace_event(char *str)
2930 {
2931 strlcpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
2932 ring_buffer_expanded = true;
2933 tracing_selftest_disabled = true;
2934
2935 return 1;
2936 }
2937 __setup("trace_event=", setup_trace_event);
2938
2939 /* Expects to have event_mutex held when called */
2940 static int
2941 create_event_toplevel_files(struct dentry *parent, struct trace_array *tr)
2942 {
2943 struct dentry *d_events;
2944 struct dentry *entry;
2945
2946 entry = tracefs_create_file("set_event", 0644, parent,
2947 tr, &ftrace_set_event_fops);
2948 if (!entry) {
2949 pr_warn("Could not create tracefs 'set_event' entry\n");
2950 return -ENOMEM;
2951 }
2952
2953 d_events = tracefs_create_dir("events", parent);
2954 if (!d_events) {
2955 pr_warn("Could not create tracefs 'events' directory\n");
2956 return -ENOMEM;
2957 }
2958
2959 entry = trace_create_file("enable", 0644, d_events,
2960 tr, &ftrace_tr_enable_fops);
2961 if (!entry) {
2962 pr_warn("Could not create tracefs 'enable' entry\n");
2963 return -ENOMEM;
2964 }
2965
2966 /* There are not as crucial, just warn if they are not created */
2967
2968 entry = tracefs_create_file("set_event_pid", 0644, parent,
2969 tr, &ftrace_set_event_pid_fops);
2970 if (!entry)
2971 pr_warn("Could not create tracefs 'set_event_pid' entry\n");
2972
2973 /* ring buffer internal formats */
2974 entry = trace_create_file("header_page", 0444, d_events,
2975 ring_buffer_print_page_header,
2976 &ftrace_show_header_fops);
2977 if (!entry)
2978 pr_warn("Could not create tracefs 'header_page' entry\n");
2979
2980 entry = trace_create_file("header_event", 0444, d_events,
2981 ring_buffer_print_entry_header,
2982 &ftrace_show_header_fops);
2983 if (!entry)
2984 pr_warn("Could not create tracefs 'header_event' entry\n");
2985
2986 tr->event_dir = d_events;
2987
2988 return 0;
2989 }
2990
2991 /**
2992 * event_trace_add_tracer - add a instance of a trace_array to events
2993 * @parent: The parent dentry to place the files/directories for events in
2994 * @tr: The trace array associated with these events
2995 *
2996 * When a new instance is created, it needs to set up its events
2997 * directory, as well as other files associated with events. It also
2998 * creates the event hierachry in the @parent/events directory.
2999 *
3000 * Returns 0 on success.
3001 *
3002 * Must be called with event_mutex held.
3003 */
3004 int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr)
3005 {
3006 int ret;
3007
3008 lockdep_assert_held(&event_mutex);
3009
3010 ret = create_event_toplevel_files(parent, tr);
3011 if (ret)
3012 goto out;
3013
3014 down_write(&trace_event_sem);
3015 __trace_add_event_dirs(tr);
3016 up_write(&trace_event_sem);
3017
3018 out:
3019 return ret;
3020 }
3021
3022 /*
3023 * The top trace array already had its file descriptors created.
3024 * Now the files themselves need to be created.
3025 */
3026 static __init int
3027 early_event_add_tracer(struct dentry *parent, struct trace_array *tr)
3028 {
3029 int ret;
3030
3031 mutex_lock(&event_mutex);
3032
3033 ret = create_event_toplevel_files(parent, tr);
3034 if (ret)
3035 goto out_unlock;
3036
3037 down_write(&trace_event_sem);
3038 __trace_early_add_event_dirs(tr);
3039 up_write(&trace_event_sem);
3040
3041 out_unlock:
3042 mutex_unlock(&event_mutex);
3043
3044 return ret;
3045 }
3046
3047 /* Must be called with event_mutex held */
3048 int event_trace_del_tracer(struct trace_array *tr)
3049 {
3050 lockdep_assert_held(&event_mutex);
3051
3052 /* Disable any event triggers and associated soft-disabled events */
3053 clear_event_triggers(tr);
3054
3055 /* Clear the pid list */
3056 __ftrace_clear_event_pids(tr);
3057
3058 /* Disable any running events */
3059 __ftrace_set_clr_event_nolock(tr, NULL, NULL, NULL, 0);
3060
3061 /* Make sure no more events are being executed */
3062 tracepoint_synchronize_unregister();
3063
3064 down_write(&trace_event_sem);
3065 __trace_remove_event_dirs(tr);
3066 tracefs_remove_recursive(tr->event_dir);
3067 up_write(&trace_event_sem);
3068
3069 tr->event_dir = NULL;
3070
3071 return 0;
3072 }
3073
3074 static __init int event_trace_memsetup(void)
3075 {
3076 field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC);
3077 file_cachep = KMEM_CACHE(trace_event_file, SLAB_PANIC);
3078 return 0;
3079 }
3080
3081 static __init void
3082 early_enable_events(struct trace_array *tr, bool disable_first)
3083 {
3084 char *buf = bootup_event_buf;
3085 char *token;
3086 int ret;
3087
3088 while (true) {
3089 token = strsep(&buf, ",");
3090
3091 if (!token)
3092 break;
3093
3094 if (*token) {
3095 /* Restarting syscalls requires that we stop them first */
3096 if (disable_first)
3097 ftrace_set_clr_event(tr, token, 0);
3098
3099 ret = ftrace_set_clr_event(tr, token, 1);
3100 if (ret)
3101 pr_warn("Failed to enable trace event: %s\n", token);
3102 }
3103
3104 /* Put back the comma to allow this to be called again */
3105 if (buf)
3106 *(buf - 1) = ',';
3107 }
3108 }
3109
3110 static __init int event_trace_enable(void)
3111 {
3112 struct trace_array *tr = top_trace_array();
3113 struct trace_event_call **iter, *call;
3114 int ret;
3115
3116 if (!tr)
3117 return -ENODEV;
3118
3119 for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
3120
3121 call = *iter;
3122 ret = event_init(call);
3123 if (!ret)
3124 list_add(&call->list, &ftrace_events);
3125 }
3126
3127 /*
3128 * We need the top trace array to have a working set of trace
3129 * points at early init, before the debug files and directories
3130 * are created. Create the file entries now, and attach them
3131 * to the actual file dentries later.
3132 */
3133 __trace_early_add_events(tr);
3134
3135 early_enable_events(tr, false);
3136
3137 trace_printk_start_comm();
3138
3139 register_event_cmds();
3140
3141 register_trigger_cmds();
3142
3143 return 0;
3144 }
3145
3146 /*
3147 * event_trace_enable() is called from trace_event_init() first to
3148 * initialize events and perhaps start any events that are on the
3149 * command line. Unfortunately, there are some events that will not
3150 * start this early, like the system call tracepoints that need
3151 * to set the TIF_SYSCALL_TRACEPOINT flag of pid 1. But event_trace_enable()
3152 * is called before pid 1 starts, and this flag is never set, making
3153 * the syscall tracepoint never get reached, but the event is enabled
3154 * regardless (and not doing anything).
3155 */
3156 static __init int event_trace_enable_again(void)
3157 {
3158 struct trace_array *tr;
3159
3160 tr = top_trace_array();
3161 if (!tr)
3162 return -ENODEV;
3163
3164 early_enable_events(tr, true);
3165
3166 return 0;
3167 }
3168
3169 early_initcall(event_trace_enable_again);
3170
3171 __init int event_trace_init(void)
3172 {
3173 struct trace_array *tr;
3174 struct dentry *d_tracer;
3175 struct dentry *entry;
3176 int ret;
3177
3178 tr = top_trace_array();
3179 if (!tr)
3180 return -ENODEV;
3181
3182 d_tracer = tracing_init_dentry();
3183 if (IS_ERR(d_tracer))
3184 return 0;
3185
3186 entry = tracefs_create_file("available_events", 0444, d_tracer,
3187 tr, &ftrace_avail_fops);
3188 if (!entry)
3189 pr_warn("Could not create tracefs 'available_events' entry\n");
3190
3191 if (trace_define_generic_fields())
3192 pr_warn("tracing: Failed to allocated generic fields");
3193
3194 if (trace_define_common_fields())
3195 pr_warn("tracing: Failed to allocate common fields");
3196
3197 ret = early_event_add_tracer(d_tracer, tr);
3198 if (ret)
3199 return ret;
3200
3201 #ifdef CONFIG_MODULES
3202 ret = register_module_notifier(&trace_module_nb);
3203 if (ret)
3204 pr_warn("Failed to register trace events module notifier\n");
3205 #endif
3206 return 0;
3207 }
3208
3209 void __init trace_event_init(void)
3210 {
3211 event_trace_memsetup();
3212 init_ftrace_syscalls();
3213 event_trace_enable();
3214 }
3215
3216 #ifdef CONFIG_FTRACE_STARTUP_TEST
3217
3218 static DEFINE_SPINLOCK(test_spinlock);
3219 static DEFINE_SPINLOCK(test_spinlock_irq);
3220 static DEFINE_MUTEX(test_mutex);
3221
3222 static __init void test_work(struct work_struct *dummy)
3223 {
3224 spin_lock(&test_spinlock);
3225 spin_lock_irq(&test_spinlock_irq);
3226 udelay(1);
3227 spin_unlock_irq(&test_spinlock_irq);
3228 spin_unlock(&test_spinlock);
3229
3230 mutex_lock(&test_mutex);
3231 msleep(1);
3232 mutex_unlock(&test_mutex);
3233 }
3234
3235 static __init int event_test_thread(void *unused)
3236 {
3237 void *test_malloc;
3238
3239 test_malloc = kmalloc(1234, GFP_KERNEL);
3240 if (!test_malloc)
3241 pr_info("failed to kmalloc\n");
3242
3243 schedule_on_each_cpu(test_work);
3244
3245 kfree(test_malloc);
3246
3247 set_current_state(TASK_INTERRUPTIBLE);
3248 while (!kthread_should_stop()) {
3249 schedule();
3250 set_current_state(TASK_INTERRUPTIBLE);
3251 }
3252 __set_current_state(TASK_RUNNING);
3253
3254 return 0;
3255 }
3256
3257 /*
3258 * Do various things that may trigger events.
3259 */
3260 static __init void event_test_stuff(void)
3261 {
3262 struct task_struct *test_thread;
3263
3264 test_thread = kthread_run(event_test_thread, NULL, "test-events");
3265 msleep(1);
3266 kthread_stop(test_thread);
3267 }
3268
3269 /*
3270 * For every trace event defined, we will test each trace point separately,
3271 * and then by groups, and finally all trace points.
3272 */
3273 static __init void event_trace_self_tests(void)
3274 {
3275 struct trace_subsystem_dir *dir;
3276 struct trace_event_file *file;
3277 struct trace_event_call *call;
3278 struct event_subsystem *system;
3279 struct trace_array *tr;
3280 int ret;
3281
3282 tr = top_trace_array();
3283 if (!tr)
3284 return;
3285
3286 pr_info("Running tests on trace events:\n");
3287
3288 list_for_each_entry(file, &tr->events, list) {
3289
3290 call = file->event_call;
3291
3292 /* Only test those that have a probe */
3293 if (!call->class || !call->class->probe)
3294 continue;
3295
3296 /*
3297 * Testing syscall events here is pretty useless, but
3298 * we still do it if configured. But this is time consuming.
3299 * What we really need is a user thread to perform the
3300 * syscalls as we test.
3301 */
3302 #ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
3303 if (call->class->system &&
3304 strcmp(call->class->system, "syscalls") == 0)
3305 continue;
3306 #endif
3307
3308 pr_info("Testing event %s: ", trace_event_name(call));
3309
3310 /*
3311 * If an event is already enabled, someone is using
3312 * it and the self test should not be on.
3313 */
3314 if (file->flags & EVENT_FILE_FL_ENABLED) {
3315 pr_warn("Enabled event during self test!\n");
3316 WARN_ON_ONCE(1);
3317 continue;
3318 }
3319
3320 ftrace_event_enable_disable(file, 1);
3321 event_test_stuff();
3322 ftrace_event_enable_disable(file, 0);
3323
3324 pr_cont("OK\n");
3325 }
3326
3327 /* Now test at the sub system level */
3328
3329 pr_info("Running tests on trace event systems:\n");
3330
3331 list_for_each_entry(dir, &tr->systems, list) {
3332
3333 system = dir->subsystem;
3334
3335 /* the ftrace system is special, skip it */
3336 if (strcmp(system->name, "ftrace") == 0)
3337 continue;
3338
3339 pr_info("Testing event system %s: ", system->name);
3340
3341 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1);
3342 if (WARN_ON_ONCE(ret)) {
3343 pr_warn("error enabling system %s\n",
3344 system->name);
3345 continue;
3346 }
3347
3348 event_test_stuff();
3349
3350 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0);
3351 if (WARN_ON_ONCE(ret)) {
3352 pr_warn("error disabling system %s\n",
3353 system->name);
3354 continue;
3355 }
3356
3357 pr_cont("OK\n");
3358 }
3359
3360 /* Test with all events enabled */
3361
3362 pr_info("Running tests on all trace events:\n");
3363 pr_info("Testing all events: ");
3364
3365 ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1);
3366 if (WARN_ON_ONCE(ret)) {
3367 pr_warn("error enabling all events\n");
3368 return;
3369 }
3370
3371 event_test_stuff();
3372
3373 /* reset sysname */
3374 ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
3375 if (WARN_ON_ONCE(ret)) {
3376 pr_warn("error disabling all events\n");
3377 return;
3378 }
3379
3380 pr_cont("OK\n");
3381 }
3382
3383 #ifdef CONFIG_FUNCTION_TRACER
3384
3385 static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
3386
3387 static struct trace_event_file event_trace_file __initdata;
3388
3389 static void __init
3390 function_test_events_call(unsigned long ip, unsigned long parent_ip,
3391 struct ftrace_ops *op, struct pt_regs *pt_regs)
3392 {
3393 struct ring_buffer_event *event;
3394 struct ring_buffer *buffer;
3395 struct ftrace_entry *entry;
3396 unsigned long flags;
3397 long disabled;
3398 int cpu;
3399 int pc;
3400
3401 pc = preempt_count();
3402 preempt_disable_notrace();
3403 cpu = raw_smp_processor_id();
3404 disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
3405
3406 if (disabled != 1)
3407 goto out;
3408
3409 local_save_flags(flags);
3410
3411 event = trace_event_buffer_lock_reserve(&buffer, &event_trace_file,
3412 TRACE_FN, sizeof(*entry),
3413 flags, pc);
3414 if (!event)
3415 goto out;
3416 entry = ring_buffer_event_data(event);
3417 entry->ip = ip;
3418 entry->parent_ip = parent_ip;
3419
3420 event_trigger_unlock_commit(&event_trace_file, buffer, event,
3421 entry, flags, pc);
3422 out:
3423 atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
3424 preempt_enable_notrace();
3425 }
3426
3427 static struct ftrace_ops trace_ops __initdata =
3428 {
3429 .func = function_test_events_call,
3430 .flags = FTRACE_OPS_FL_RECURSION_SAFE,
3431 };
3432
3433 static __init void event_trace_self_test_with_function(void)
3434 {
3435 int ret;
3436
3437 event_trace_file.tr = top_trace_array();
3438 if (WARN_ON(!event_trace_file.tr))
3439 return;
3440
3441 ret = register_ftrace_function(&trace_ops);
3442 if (WARN_ON(ret < 0)) {
3443 pr_info("Failed to enable function tracer for event tests\n");
3444 return;
3445 }
3446 pr_info("Running tests again, along with the function tracer\n");
3447 event_trace_self_tests();
3448 unregister_ftrace_function(&trace_ops);
3449 }
3450 #else
3451 static __init void event_trace_self_test_with_function(void)
3452 {
3453 }
3454 #endif
3455
3456 static __init int event_trace_self_tests_init(void)
3457 {
3458 if (!tracing_selftest_disabled) {
3459 event_trace_self_tests();
3460 event_trace_self_test_with_function();
3461 }
3462
3463 return 0;
3464 }
3465
3466 late_initcall(event_trace_self_tests_init);
3467
3468 #endif
Linux with added FPC-III support