]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - kernel/trace/trace_events.c
tracing: Fix disabling of soft disable
[mirror_ubuntu-bionic-kernel.git] / kernel / trace / trace_events.c
1 /*
2 * event tracer
3 *
4 * Copyright (C) 2008 Red Hat Inc, Steven Rostedt <srostedt@redhat.com>
5 *
6 * - Added format output of fields of the trace point.
7 * This was based off of work by Tom Zanussi <tzanussi@gmail.com>.
8 *
9 */
10
11 #include <linux/workqueue.h>
12 #include <linux/spinlock.h>
13 #include <linux/kthread.h>
14 #include <linux/debugfs.h>
15 #include <linux/uaccess.h>
16 #include <linux/module.h>
17 #include <linux/ctype.h>
18 #include <linux/slab.h>
19 #include <linux/delay.h>
20
21 #include <asm/setup.h>
22
23 #include "trace_output.h"
24
25 #undef TRACE_SYSTEM
26 #define TRACE_SYSTEM "TRACE_SYSTEM"
27
28 DEFINE_MUTEX(event_mutex);
29
30 DEFINE_MUTEX(event_storage_mutex);
31 EXPORT_SYMBOL_GPL(event_storage_mutex);
32
33 char event_storage[EVENT_STORAGE_SIZE];
34 EXPORT_SYMBOL_GPL(event_storage);
35
36 LIST_HEAD(ftrace_events);
37 static LIST_HEAD(ftrace_common_fields);
38
39 #define GFP_TRACE (GFP_KERNEL | __GFP_ZERO)
40
41 static struct kmem_cache *field_cachep;
42 static struct kmem_cache *file_cachep;
43
44 #define SYSTEM_FL_FREE_NAME (1 << 31)
45
46 static inline int system_refcount(struct event_subsystem *system)
47 {
48 return system->ref_count & ~SYSTEM_FL_FREE_NAME;
49 }
50
51 static int system_refcount_inc(struct event_subsystem *system)
52 {
53 return (system->ref_count++) & ~SYSTEM_FL_FREE_NAME;
54 }
55
56 static int system_refcount_dec(struct event_subsystem *system)
57 {
58 return (--system->ref_count) & ~SYSTEM_FL_FREE_NAME;
59 }
60
61 /* Double loops, do not use break, only goto's work */
62 #define do_for_each_event_file(tr, file) \
63 list_for_each_entry(tr, &ftrace_trace_arrays, list) { \
64 list_for_each_entry(file, &tr->events, list)
65
66 #define do_for_each_event_file_safe(tr, file) \
67 list_for_each_entry(tr, &ftrace_trace_arrays, list) { \
68 struct ftrace_event_file *___n; \
69 list_for_each_entry_safe(file, ___n, &tr->events, list)
70
71 #define while_for_each_event_file() \
72 }
73
74 static struct list_head *
75 trace_get_fields(struct ftrace_event_call *event_call)
76 {
77 if (!event_call->class->get_fields)
78 return &event_call->class->fields;
79 return event_call->class->get_fields(event_call);
80 }
81
82 static struct ftrace_event_field *
83 __find_event_field(struct list_head *head, char *name)
84 {
85 struct ftrace_event_field *field;
86
87 list_for_each_entry(field, head, link) {
88 if (!strcmp(field->name, name))
89 return field;
90 }
91
92 return NULL;
93 }
94
95 struct ftrace_event_field *
96 trace_find_event_field(struct ftrace_event_call *call, char *name)
97 {
98 struct ftrace_event_field *field;
99 struct list_head *head;
100
101 field = __find_event_field(&ftrace_common_fields, name);
102 if (field)
103 return field;
104
105 head = trace_get_fields(call);
106 return __find_event_field(head, name);
107 }
108
109 static int __trace_define_field(struct list_head *head, const char *type,
110 const char *name, int offset, int size,
111 int is_signed, int filter_type)
112 {
113 struct ftrace_event_field *field;
114
115 field = kmem_cache_alloc(field_cachep, GFP_TRACE);
116 if (!field)
117 return -ENOMEM;
118
119 field->name = name;
120 field->type = type;
121
122 if (filter_type == FILTER_OTHER)
123 field->filter_type = filter_assign_type(type);
124 else
125 field->filter_type = filter_type;
126
127 field->offset = offset;
128 field->size = size;
129 field->is_signed = is_signed;
130
131 list_add(&field->link, head);
132
133 return 0;
134 }
135
136 int trace_define_field(struct ftrace_event_call *call, const char *type,
137 const char *name, int offset, int size, int is_signed,
138 int filter_type)
139 {
140 struct list_head *head;
141
142 if (WARN_ON(!call->class))
143 return 0;
144
145 head = trace_get_fields(call);
146 return __trace_define_field(head, type, name, offset, size,
147 is_signed, filter_type);
148 }
149 EXPORT_SYMBOL_GPL(trace_define_field);
150
151 #define __common_field(type, item) \
152 ret = __trace_define_field(&ftrace_common_fields, #type, \
153 "common_" #item, \
154 offsetof(typeof(ent), item), \
155 sizeof(ent.item), \
156 is_signed_type(type), FILTER_OTHER); \
157 if (ret) \
158 return ret;
159
160 static int trace_define_common_fields(void)
161 {
162 int ret;
163 struct trace_entry ent;
164
165 __common_field(unsigned short, type);
166 __common_field(unsigned char, flags);
167 __common_field(unsigned char, preempt_count);
168 __common_field(int, pid);
169
170 return ret;
171 }
172
173 static void trace_destroy_fields(struct ftrace_event_call *call)
174 {
175 struct ftrace_event_field *field, *next;
176 struct list_head *head;
177
178 head = trace_get_fields(call);
179 list_for_each_entry_safe(field, next, head, link) {
180 list_del(&field->link);
181 kmem_cache_free(field_cachep, field);
182 }
183 }
184
185 int trace_event_raw_init(struct ftrace_event_call *call)
186 {
187 int id;
188
189 id = register_ftrace_event(&call->event);
190 if (!id)
191 return -ENODEV;
192
193 return 0;
194 }
195 EXPORT_SYMBOL_GPL(trace_event_raw_init);
196
197 int ftrace_event_reg(struct ftrace_event_call *call,
198 enum trace_reg type, void *data)
199 {
200 struct ftrace_event_file *file = data;
201
202 switch (type) {
203 case TRACE_REG_REGISTER:
204 return tracepoint_probe_register(call->name,
205 call->class->probe,
206 file);
207 case TRACE_REG_UNREGISTER:
208 tracepoint_probe_unregister(call->name,
209 call->class->probe,
210 file);
211 return 0;
212
213 #ifdef CONFIG_PERF_EVENTS
214 case TRACE_REG_PERF_REGISTER:
215 return tracepoint_probe_register(call->name,
216 call->class->perf_probe,
217 call);
218 case TRACE_REG_PERF_UNREGISTER:
219 tracepoint_probe_unregister(call->name,
220 call->class->perf_probe,
221 call);
222 return 0;
223 case TRACE_REG_PERF_OPEN:
224 case TRACE_REG_PERF_CLOSE:
225 case TRACE_REG_PERF_ADD:
226 case TRACE_REG_PERF_DEL:
227 return 0;
228 #endif
229 }
230 return 0;
231 }
232 EXPORT_SYMBOL_GPL(ftrace_event_reg);
233
234 void trace_event_enable_cmd_record(bool enable)
235 {
236 struct ftrace_event_file *file;
237 struct trace_array *tr;
238
239 mutex_lock(&event_mutex);
240 do_for_each_event_file(tr, file) {
241
242 if (!(file->flags & FTRACE_EVENT_FL_ENABLED))
243 continue;
244
245 if (enable) {
246 tracing_start_cmdline_record();
247 set_bit(FTRACE_EVENT_FL_RECORDED_CMD_BIT, &file->flags);
248 } else {
249 tracing_stop_cmdline_record();
250 clear_bit(FTRACE_EVENT_FL_RECORDED_CMD_BIT, &file->flags);
251 }
252 } while_for_each_event_file();
253 mutex_unlock(&event_mutex);
254 }
255
256 static int __ftrace_event_enable_disable(struct ftrace_event_file *file,
257 int enable, int soft_disable)
258 {
259 struct ftrace_event_call *call = file->event_call;
260 int ret = 0;
261 int disable;
262
263 switch (enable) {
264 case 0:
265 /*
266 * When soft_disable is set and enable is cleared, the sm_ref
267 * reference counter is decremented. If it reaches 0, we want
268 * to clear the SOFT_DISABLED flag but leave the event in the
269 * state that it was. That is, if the event was enabled and
270 * SOFT_DISABLED isn't set, then do nothing. But if SOFT_DISABLED
271 * is set we do not want the event to be enabled before we
272 * clear the bit.
273 *
274 * When soft_disable is not set but the SOFT_MODE flag is,
275 * we do nothing. Do not disable the tracepoint, otherwise
276 * "soft enable"s (clearing the SOFT_DISABLED bit) wont work.
277 */
278 if (soft_disable) {
279 if (atomic_dec_return(&file->sm_ref) > 0)
280 break;
281 disable = file->flags & FTRACE_EVENT_FL_SOFT_DISABLED;
282 clear_bit(FTRACE_EVENT_FL_SOFT_MODE_BIT, &file->flags);
283 } else
284 disable = !(file->flags & FTRACE_EVENT_FL_SOFT_MODE);
285
286 if (disable && (file->flags & FTRACE_EVENT_FL_ENABLED)) {
287 clear_bit(FTRACE_EVENT_FL_ENABLED_BIT, &file->flags);
288 if (file->flags & FTRACE_EVENT_FL_RECORDED_CMD) {
289 tracing_stop_cmdline_record();
290 clear_bit(FTRACE_EVENT_FL_RECORDED_CMD_BIT, &file->flags);
291 }
292 call->class->reg(call, TRACE_REG_UNREGISTER, file);
293 }
294 /* If in SOFT_MODE, just set the SOFT_DISABLE_BIT, else clear it */
295 if (file->flags & FTRACE_EVENT_FL_SOFT_MODE)
296 set_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &file->flags);
297 else
298 clear_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &file->flags);
299 break;
300 case 1:
301 /*
302 * When soft_disable is set and enable is set, we want to
303 * register the tracepoint for the event, but leave the event
304 * as is. That means, if the event was already enabled, we do
305 * nothing (but set SOFT_MODE). If the event is disabled, we
306 * set SOFT_DISABLED before enabling the event tracepoint, so
307 * it still seems to be disabled.
308 */
309 if (!soft_disable)
310 clear_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &file->flags);
311 else {
312 if (atomic_inc_return(&file->sm_ref) > 1)
313 break;
314 set_bit(FTRACE_EVENT_FL_SOFT_MODE_BIT, &file->flags);
315 }
316
317 if (!(file->flags & FTRACE_EVENT_FL_ENABLED)) {
318
319 /* Keep the event disabled, when going to SOFT_MODE. */
320 if (soft_disable)
321 set_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &file->flags);
322
323 if (trace_flags & TRACE_ITER_RECORD_CMD) {
324 tracing_start_cmdline_record();
325 set_bit(FTRACE_EVENT_FL_RECORDED_CMD_BIT, &file->flags);
326 }
327 ret = call->class->reg(call, TRACE_REG_REGISTER, file);
328 if (ret) {
329 tracing_stop_cmdline_record();
330 pr_info("event trace: Could not enable event "
331 "%s\n", call->name);
332 break;
333 }
334 set_bit(FTRACE_EVENT_FL_ENABLED_BIT, &file->flags);
335
336 /* WAS_ENABLED gets set but never cleared. */
337 call->flags |= TRACE_EVENT_FL_WAS_ENABLED;
338 }
339 break;
340 }
341
342 return ret;
343 }
344
345 static int ftrace_event_enable_disable(struct ftrace_event_file *file,
346 int enable)
347 {
348 return __ftrace_event_enable_disable(file, enable, 0);
349 }
350
351 static void ftrace_clear_events(struct trace_array *tr)
352 {
353 struct ftrace_event_file *file;
354
355 mutex_lock(&event_mutex);
356 list_for_each_entry(file, &tr->events, list) {
357 ftrace_event_enable_disable(file, 0);
358 }
359 mutex_unlock(&event_mutex);
360 }
361
362 static void __put_system(struct event_subsystem *system)
363 {
364 struct event_filter *filter = system->filter;
365
366 WARN_ON_ONCE(system_refcount(system) == 0);
367 if (system_refcount_dec(system))
368 return;
369
370 list_del(&system->list);
371
372 if (filter) {
373 kfree(filter->filter_string);
374 kfree(filter);
375 }
376 if (system->ref_count & SYSTEM_FL_FREE_NAME)
377 kfree(system->name);
378 kfree(system);
379 }
380
381 static void __get_system(struct event_subsystem *system)
382 {
383 WARN_ON_ONCE(system_refcount(system) == 0);
384 system_refcount_inc(system);
385 }
386
387 static void __get_system_dir(struct ftrace_subsystem_dir *dir)
388 {
389 WARN_ON_ONCE(dir->ref_count == 0);
390 dir->ref_count++;
391 __get_system(dir->subsystem);
392 }
393
394 static void __put_system_dir(struct ftrace_subsystem_dir *dir)
395 {
396 WARN_ON_ONCE(dir->ref_count == 0);
397 /* If the subsystem is about to be freed, the dir must be too */
398 WARN_ON_ONCE(system_refcount(dir->subsystem) == 1 && dir->ref_count != 1);
399
400 __put_system(dir->subsystem);
401 if (!--dir->ref_count)
402 kfree(dir);
403 }
404
405 static void put_system(struct ftrace_subsystem_dir *dir)
406 {
407 mutex_lock(&event_mutex);
408 __put_system_dir(dir);
409 mutex_unlock(&event_mutex);
410 }
411
412 /*
413 * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
414 */
415 static int __ftrace_set_clr_event(struct trace_array *tr, const char *match,
416 const char *sub, const char *event, int set)
417 {
418 struct ftrace_event_file *file;
419 struct ftrace_event_call *call;
420 int ret = -EINVAL;
421
422 mutex_lock(&event_mutex);
423 list_for_each_entry(file, &tr->events, list) {
424
425 call = file->event_call;
426
427 if (!call->name || !call->class || !call->class->reg)
428 continue;
429
430 if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
431 continue;
432
433 if (match &&
434 strcmp(match, call->name) != 0 &&
435 strcmp(match, call->class->system) != 0)
436 continue;
437
438 if (sub && strcmp(sub, call->class->system) != 0)
439 continue;
440
441 if (event && strcmp(event, call->name) != 0)
442 continue;
443
444 ftrace_event_enable_disable(file, set);
445
446 ret = 0;
447 }
448 mutex_unlock(&event_mutex);
449
450 return ret;
451 }
452
453 static int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set)
454 {
455 char *event = NULL, *sub = NULL, *match;
456
457 /*
458 * The buf format can be <subsystem>:<event-name>
459 * *:<event-name> means any event by that name.
460 * :<event-name> is the same.
461 *
462 * <subsystem>:* means all events in that subsystem
463 * <subsystem>: means the same.
464 *
465 * <name> (no ':') means all events in a subsystem with
466 * the name <name> or any event that matches <name>
467 */
468
469 match = strsep(&buf, ":");
470 if (buf) {
471 sub = match;
472 event = buf;
473 match = NULL;
474
475 if (!strlen(sub) || strcmp(sub, "*") == 0)
476 sub = NULL;
477 if (!strlen(event) || strcmp(event, "*") == 0)
478 event = NULL;
479 }
480
481 return __ftrace_set_clr_event(tr, match, sub, event, set);
482 }
483
484 /**
485 * trace_set_clr_event - enable or disable an event
486 * @system: system name to match (NULL for any system)
487 * @event: event name to match (NULL for all events, within system)
488 * @set: 1 to enable, 0 to disable
489 *
490 * This is a way for other parts of the kernel to enable or disable
491 * event recording.
492 *
493 * Returns 0 on success, -EINVAL if the parameters do not match any
494 * registered events.
495 */
496 int trace_set_clr_event(const char *system, const char *event, int set)
497 {
498 struct trace_array *tr = top_trace_array();
499
500 return __ftrace_set_clr_event(tr, NULL, system, event, set);
501 }
502 EXPORT_SYMBOL_GPL(trace_set_clr_event);
503
504 /* 128 should be much more than enough */
505 #define EVENT_BUF_SIZE 127
506
507 static ssize_t
508 ftrace_event_write(struct file *file, const char __user *ubuf,
509 size_t cnt, loff_t *ppos)
510 {
511 struct trace_parser parser;
512 struct seq_file *m = file->private_data;
513 struct trace_array *tr = m->private;
514 ssize_t read, ret;
515
516 if (!cnt)
517 return 0;
518
519 ret = tracing_update_buffers();
520 if (ret < 0)
521 return ret;
522
523 if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
524 return -ENOMEM;
525
526 read = trace_get_user(&parser, ubuf, cnt, ppos);
527
528 if (read >= 0 && trace_parser_loaded((&parser))) {
529 int set = 1;
530
531 if (*parser.buffer == '!')
532 set = 0;
533
534 parser.buffer[parser.idx] = 0;
535
536 ret = ftrace_set_clr_event(tr, parser.buffer + !set, set);
537 if (ret)
538 goto out_put;
539 }
540
541 ret = read;
542
543 out_put:
544 trace_parser_put(&parser);
545
546 return ret;
547 }
548
549 static void *
550 t_next(struct seq_file *m, void *v, loff_t *pos)
551 {
552 struct ftrace_event_file *file = v;
553 struct ftrace_event_call *call;
554 struct trace_array *tr = m->private;
555
556 (*pos)++;
557
558 list_for_each_entry_continue(file, &tr->events, list) {
559 call = file->event_call;
560 /*
561 * The ftrace subsystem is for showing formats only.
562 * They can not be enabled or disabled via the event files.
563 */
564 if (call->class && call->class->reg)
565 return file;
566 }
567
568 return NULL;
569 }
570
571 static void *t_start(struct seq_file *m, loff_t *pos)
572 {
573 struct ftrace_event_file *file;
574 struct trace_array *tr = m->private;
575 loff_t l;
576
577 mutex_lock(&event_mutex);
578
579 file = list_entry(&tr->events, struct ftrace_event_file, list);
580 for (l = 0; l <= *pos; ) {
581 file = t_next(m, file, &l);
582 if (!file)
583 break;
584 }
585 return file;
586 }
587
588 static void *
589 s_next(struct seq_file *m, void *v, loff_t *pos)
590 {
591 struct ftrace_event_file *file = v;
592 struct trace_array *tr = m->private;
593
594 (*pos)++;
595
596 list_for_each_entry_continue(file, &tr->events, list) {
597 if (file->flags & FTRACE_EVENT_FL_ENABLED)
598 return file;
599 }
600
601 return NULL;
602 }
603
604 static void *s_start(struct seq_file *m, loff_t *pos)
605 {
606 struct ftrace_event_file *file;
607 struct trace_array *tr = m->private;
608 loff_t l;
609
610 mutex_lock(&event_mutex);
611
612 file = list_entry(&tr->events, struct ftrace_event_file, list);
613 for (l = 0; l <= *pos; ) {
614 file = s_next(m, file, &l);
615 if (!file)
616 break;
617 }
618 return file;
619 }
620
621 static int t_show(struct seq_file *m, void *v)
622 {
623 struct ftrace_event_file *file = v;
624 struct ftrace_event_call *call = file->event_call;
625
626 if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
627 seq_printf(m, "%s:", call->class->system);
628 seq_printf(m, "%s\n", call->name);
629
630 return 0;
631 }
632
633 static void t_stop(struct seq_file *m, void *p)
634 {
635 mutex_unlock(&event_mutex);
636 }
637
638 static ssize_t
639 event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
640 loff_t *ppos)
641 {
642 struct ftrace_event_file *file = filp->private_data;
643 char buf[4] = "0";
644
645 if (file->flags & FTRACE_EVENT_FL_ENABLED &&
646 !(file->flags & FTRACE_EVENT_FL_SOFT_DISABLED))
647 strcpy(buf, "1");
648
649 if (file->flags & FTRACE_EVENT_FL_SOFT_DISABLED ||
650 file->flags & FTRACE_EVENT_FL_SOFT_MODE)
651 strcat(buf, "*");
652
653 strcat(buf, "\n");
654
655 return simple_read_from_buffer(ubuf, cnt, ppos, buf, strlen(buf));
656 }
657
658 static ssize_t
659 event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
660 loff_t *ppos)
661 {
662 struct ftrace_event_file *file = filp->private_data;
663 unsigned long val;
664 int ret;
665
666 if (!file)
667 return -EINVAL;
668
669 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
670 if (ret)
671 return ret;
672
673 ret = tracing_update_buffers();
674 if (ret < 0)
675 return ret;
676
677 switch (val) {
678 case 0:
679 case 1:
680 mutex_lock(&event_mutex);
681 ret = ftrace_event_enable_disable(file, val);
682 mutex_unlock(&event_mutex);
683 break;
684
685 default:
686 return -EINVAL;
687 }
688
689 *ppos += cnt;
690
691 return ret ? ret : cnt;
692 }
693
694 static ssize_t
695 system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
696 loff_t *ppos)
697 {
698 const char set_to_char[4] = { '?', '0', '1', 'X' };
699 struct ftrace_subsystem_dir *dir = filp->private_data;
700 struct event_subsystem *system = dir->subsystem;
701 struct ftrace_event_call *call;
702 struct ftrace_event_file *file;
703 struct trace_array *tr = dir->tr;
704 char buf[2];
705 int set = 0;
706 int ret;
707
708 mutex_lock(&event_mutex);
709 list_for_each_entry(file, &tr->events, list) {
710 call = file->event_call;
711 if (!call->name || !call->class || !call->class->reg)
712 continue;
713
714 if (system && strcmp(call->class->system, system->name) != 0)
715 continue;
716
717 /*
718 * We need to find out if all the events are set
719 * or if all events or cleared, or if we have
720 * a mixture.
721 */
722 set |= (1 << !!(file->flags & FTRACE_EVENT_FL_ENABLED));
723
724 /*
725 * If we have a mixture, no need to look further.
726 */
727 if (set == 3)
728 break;
729 }
730 mutex_unlock(&event_mutex);
731
732 buf[0] = set_to_char[set];
733 buf[1] = '\n';
734
735 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
736
737 return ret;
738 }
739
740 static ssize_t
741 system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
742 loff_t *ppos)
743 {
744 struct ftrace_subsystem_dir *dir = filp->private_data;
745 struct event_subsystem *system = dir->subsystem;
746 const char *name = NULL;
747 unsigned long val;
748 ssize_t ret;
749
750 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
751 if (ret)
752 return ret;
753
754 ret = tracing_update_buffers();
755 if (ret < 0)
756 return ret;
757
758 if (val != 0 && val != 1)
759 return -EINVAL;
760
761 /*
762 * Opening of "enable" adds a ref count to system,
763 * so the name is safe to use.
764 */
765 if (system)
766 name = system->name;
767
768 ret = __ftrace_set_clr_event(dir->tr, NULL, name, NULL, val);
769 if (ret)
770 goto out;
771
772 ret = cnt;
773
774 out:
775 *ppos += cnt;
776
777 return ret;
778 }
779
780 enum {
781 FORMAT_HEADER = 1,
782 FORMAT_FIELD_SEPERATOR = 2,
783 FORMAT_PRINTFMT = 3,
784 };
785
786 static void *f_next(struct seq_file *m, void *v, loff_t *pos)
787 {
788 struct ftrace_event_call *call = m->private;
789 struct ftrace_event_field *field;
790 struct list_head *common_head = &ftrace_common_fields;
791 struct list_head *head = trace_get_fields(call);
792
793 (*pos)++;
794
795 switch ((unsigned long)v) {
796 case FORMAT_HEADER:
797 if (unlikely(list_empty(common_head)))
798 return NULL;
799
800 field = list_entry(common_head->prev,
801 struct ftrace_event_field, link);
802 return field;
803
804 case FORMAT_FIELD_SEPERATOR:
805 if (unlikely(list_empty(head)))
806 return NULL;
807
808 field = list_entry(head->prev, struct ftrace_event_field, link);
809 return field;
810
811 case FORMAT_PRINTFMT:
812 /* all done */
813 return NULL;
814 }
815
816 field = v;
817 if (field->link.prev == common_head)
818 return (void *)FORMAT_FIELD_SEPERATOR;
819 else if (field->link.prev == head)
820 return (void *)FORMAT_PRINTFMT;
821
822 field = list_entry(field->link.prev, struct ftrace_event_field, link);
823
824 return field;
825 }
826
827 static void *f_start(struct seq_file *m, loff_t *pos)
828 {
829 loff_t l = 0;
830 void *p;
831
832 /* Start by showing the header */
833 if (!*pos)
834 return (void *)FORMAT_HEADER;
835
836 p = (void *)FORMAT_HEADER;
837 do {
838 p = f_next(m, p, &l);
839 } while (p && l < *pos);
840
841 return p;
842 }
843
844 static int f_show(struct seq_file *m, void *v)
845 {
846 struct ftrace_event_call *call = m->private;
847 struct ftrace_event_field *field;
848 const char *array_descriptor;
849
850 switch ((unsigned long)v) {
851 case FORMAT_HEADER:
852 seq_printf(m, "name: %s\n", call->name);
853 seq_printf(m, "ID: %d\n", call->event.type);
854 seq_printf(m, "format:\n");
855 return 0;
856
857 case FORMAT_FIELD_SEPERATOR:
858 seq_putc(m, '\n');
859 return 0;
860
861 case FORMAT_PRINTFMT:
862 seq_printf(m, "\nprint fmt: %s\n",
863 call->print_fmt);
864 return 0;
865 }
866
867 field = v;
868
869 /*
870 * Smartly shows the array type(except dynamic array).
871 * Normal:
872 * field:TYPE VAR
873 * If TYPE := TYPE[LEN], it is shown:
874 * field:TYPE VAR[LEN]
875 */
876 array_descriptor = strchr(field->type, '[');
877
878 if (!strncmp(field->type, "__data_loc", 10))
879 array_descriptor = NULL;
880
881 if (!array_descriptor)
882 seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
883 field->type, field->name, field->offset,
884 field->size, !!field->is_signed);
885 else
886 seq_printf(m, "\tfield:%.*s %s%s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
887 (int)(array_descriptor - field->type),
888 field->type, field->name,
889 array_descriptor, field->offset,
890 field->size, !!field->is_signed);
891
892 return 0;
893 }
894
895 static void f_stop(struct seq_file *m, void *p)
896 {
897 }
898
899 static const struct seq_operations trace_format_seq_ops = {
900 .start = f_start,
901 .next = f_next,
902 .stop = f_stop,
903 .show = f_show,
904 };
905
906 static int trace_format_open(struct inode *inode, struct file *file)
907 {
908 struct ftrace_event_call *call = inode->i_private;
909 struct seq_file *m;
910 int ret;
911
912 ret = seq_open(file, &trace_format_seq_ops);
913 if (ret < 0)
914 return ret;
915
916 m = file->private_data;
917 m->private = call;
918
919 return 0;
920 }
921
922 static ssize_t
923 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
924 {
925 struct ftrace_event_call *call = filp->private_data;
926 struct trace_seq *s;
927 int r;
928
929 if (*ppos)
930 return 0;
931
932 s = kmalloc(sizeof(*s), GFP_KERNEL);
933 if (!s)
934 return -ENOMEM;
935
936 trace_seq_init(s);
937 trace_seq_printf(s, "%d\n", call->event.type);
938
939 r = simple_read_from_buffer(ubuf, cnt, ppos,
940 s->buffer, s->len);
941 kfree(s);
942 return r;
943 }
944
945 static ssize_t
946 event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
947 loff_t *ppos)
948 {
949 struct ftrace_event_call *call = filp->private_data;
950 struct trace_seq *s;
951 int r;
952
953 if (*ppos)
954 return 0;
955
956 s = kmalloc(sizeof(*s), GFP_KERNEL);
957 if (!s)
958 return -ENOMEM;
959
960 trace_seq_init(s);
961
962 print_event_filter(call, s);
963 r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
964
965 kfree(s);
966
967 return r;
968 }
969
970 static ssize_t
971 event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
972 loff_t *ppos)
973 {
974 struct ftrace_event_call *call = filp->private_data;
975 char *buf;
976 int err;
977
978 if (cnt >= PAGE_SIZE)
979 return -EINVAL;
980
981 buf = (char *)__get_free_page(GFP_TEMPORARY);
982 if (!buf)
983 return -ENOMEM;
984
985 if (copy_from_user(buf, ubuf, cnt)) {
986 free_page((unsigned long) buf);
987 return -EFAULT;
988 }
989 buf[cnt] = '\0';
990
991 err = apply_event_filter(call, buf);
992 free_page((unsigned long) buf);
993 if (err < 0)
994 return err;
995
996 *ppos += cnt;
997
998 return cnt;
999 }
1000
1001 static LIST_HEAD(event_subsystems);
1002
1003 static int subsystem_open(struct inode *inode, struct file *filp)
1004 {
1005 struct event_subsystem *system = NULL;
1006 struct ftrace_subsystem_dir *dir = NULL; /* Initialize for gcc */
1007 struct trace_array *tr;
1008 int ret;
1009
1010 /* Make sure the system still exists */
1011 mutex_lock(&event_mutex);
1012 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
1013 list_for_each_entry(dir, &tr->systems, list) {
1014 if (dir == inode->i_private) {
1015 /* Don't open systems with no events */
1016 if (dir->nr_events) {
1017 __get_system_dir(dir);
1018 system = dir->subsystem;
1019 }
1020 goto exit_loop;
1021 }
1022 }
1023 }
1024 exit_loop:
1025 mutex_unlock(&event_mutex);
1026
1027 if (!system)
1028 return -ENODEV;
1029
1030 /* Some versions of gcc think dir can be uninitialized here */
1031 WARN_ON(!dir);
1032
1033 ret = tracing_open_generic(inode, filp);
1034 if (ret < 0)
1035 put_system(dir);
1036
1037 return ret;
1038 }
1039
1040 static int system_tr_open(struct inode *inode, struct file *filp)
1041 {
1042 struct ftrace_subsystem_dir *dir;
1043 struct trace_array *tr = inode->i_private;
1044 int ret;
1045
1046 /* Make a temporary dir that has no system but points to tr */
1047 dir = kzalloc(sizeof(*dir), GFP_KERNEL);
1048 if (!dir)
1049 return -ENOMEM;
1050
1051 dir->tr = tr;
1052
1053 ret = tracing_open_generic(inode, filp);
1054 if (ret < 0)
1055 kfree(dir);
1056
1057 filp->private_data = dir;
1058
1059 return ret;
1060 }
1061
1062 static int subsystem_release(struct inode *inode, struct file *file)
1063 {
1064 struct ftrace_subsystem_dir *dir = file->private_data;
1065
1066 /*
1067 * If dir->subsystem is NULL, then this is a temporary
1068 * descriptor that was made for a trace_array to enable
1069 * all subsystems.
1070 */
1071 if (dir->subsystem)
1072 put_system(dir);
1073 else
1074 kfree(dir);
1075
1076 return 0;
1077 }
1078
1079 static ssize_t
1080 subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1081 loff_t *ppos)
1082 {
1083 struct ftrace_subsystem_dir *dir = filp->private_data;
1084 struct event_subsystem *system = dir->subsystem;
1085 struct trace_seq *s;
1086 int r;
1087
1088 if (*ppos)
1089 return 0;
1090
1091 s = kmalloc(sizeof(*s), GFP_KERNEL);
1092 if (!s)
1093 return -ENOMEM;
1094
1095 trace_seq_init(s);
1096
1097 print_subsystem_event_filter(system, s);
1098 r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
1099
1100 kfree(s);
1101
1102 return r;
1103 }
1104
1105 static ssize_t
1106 subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1107 loff_t *ppos)
1108 {
1109 struct ftrace_subsystem_dir *dir = filp->private_data;
1110 char *buf;
1111 int err;
1112
1113 if (cnt >= PAGE_SIZE)
1114 return -EINVAL;
1115
1116 buf = (char *)__get_free_page(GFP_TEMPORARY);
1117 if (!buf)
1118 return -ENOMEM;
1119
1120 if (copy_from_user(buf, ubuf, cnt)) {
1121 free_page((unsigned long) buf);
1122 return -EFAULT;
1123 }
1124 buf[cnt] = '\0';
1125
1126 err = apply_subsystem_event_filter(dir, buf);
1127 free_page((unsigned long) buf);
1128 if (err < 0)
1129 return err;
1130
1131 *ppos += cnt;
1132
1133 return cnt;
1134 }
1135
1136 static ssize_t
1137 show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1138 {
1139 int (*func)(struct trace_seq *s) = filp->private_data;
1140 struct trace_seq *s;
1141 int r;
1142
1143 if (*ppos)
1144 return 0;
1145
1146 s = kmalloc(sizeof(*s), GFP_KERNEL);
1147 if (!s)
1148 return -ENOMEM;
1149
1150 trace_seq_init(s);
1151
1152 func(s);
1153 r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
1154
1155 kfree(s);
1156
1157 return r;
1158 }
1159
1160 static int ftrace_event_avail_open(struct inode *inode, struct file *file);
1161 static int ftrace_event_set_open(struct inode *inode, struct file *file);
1162
1163 static const struct seq_operations show_event_seq_ops = {
1164 .start = t_start,
1165 .next = t_next,
1166 .show = t_show,
1167 .stop = t_stop,
1168 };
1169
1170 static const struct seq_operations show_set_event_seq_ops = {
1171 .start = s_start,
1172 .next = s_next,
1173 .show = t_show,
1174 .stop = t_stop,
1175 };
1176
1177 static const struct file_operations ftrace_avail_fops = {
1178 .open = ftrace_event_avail_open,
1179 .read = seq_read,
1180 .llseek = seq_lseek,
1181 .release = seq_release,
1182 };
1183
1184 static const struct file_operations ftrace_set_event_fops = {
1185 .open = ftrace_event_set_open,
1186 .read = seq_read,
1187 .write = ftrace_event_write,
1188 .llseek = seq_lseek,
1189 .release = seq_release,
1190 };
1191
1192 static const struct file_operations ftrace_enable_fops = {
1193 .open = tracing_open_generic,
1194 .read = event_enable_read,
1195 .write = event_enable_write,
1196 .llseek = default_llseek,
1197 };
1198
1199 static const struct file_operations ftrace_event_format_fops = {
1200 .open = trace_format_open,
1201 .read = seq_read,
1202 .llseek = seq_lseek,
1203 .release = seq_release,
1204 };
1205
1206 static const struct file_operations ftrace_event_id_fops = {
1207 .open = tracing_open_generic,
1208 .read = event_id_read,
1209 .llseek = default_llseek,
1210 };
1211
1212 static const struct file_operations ftrace_event_filter_fops = {
1213 .open = tracing_open_generic,
1214 .read = event_filter_read,
1215 .write = event_filter_write,
1216 .llseek = default_llseek,
1217 };
1218
1219 static const struct file_operations ftrace_subsystem_filter_fops = {
1220 .open = subsystem_open,
1221 .read = subsystem_filter_read,
1222 .write = subsystem_filter_write,
1223 .llseek = default_llseek,
1224 .release = subsystem_release,
1225 };
1226
1227 static const struct file_operations ftrace_system_enable_fops = {
1228 .open = subsystem_open,
1229 .read = system_enable_read,
1230 .write = system_enable_write,
1231 .llseek = default_llseek,
1232 .release = subsystem_release,
1233 };
1234
1235 static const struct file_operations ftrace_tr_enable_fops = {
1236 .open = system_tr_open,
1237 .read = system_enable_read,
1238 .write = system_enable_write,
1239 .llseek = default_llseek,
1240 .release = subsystem_release,
1241 };
1242
1243 static const struct file_operations ftrace_show_header_fops = {
1244 .open = tracing_open_generic,
1245 .read = show_header,
1246 .llseek = default_llseek,
1247 };
1248
1249 static int
1250 ftrace_event_open(struct inode *inode, struct file *file,
1251 const struct seq_operations *seq_ops)
1252 {
1253 struct seq_file *m;
1254 int ret;
1255
1256 ret = seq_open(file, seq_ops);
1257 if (ret < 0)
1258 return ret;
1259 m = file->private_data;
1260 /* copy tr over to seq ops */
1261 m->private = inode->i_private;
1262
1263 return ret;
1264 }
1265
1266 static int
1267 ftrace_event_avail_open(struct inode *inode, struct file *file)
1268 {
1269 const struct seq_operations *seq_ops = &show_event_seq_ops;
1270
1271 return ftrace_event_open(inode, file, seq_ops);
1272 }
1273
1274 static int
1275 ftrace_event_set_open(struct inode *inode, struct file *file)
1276 {
1277 const struct seq_operations *seq_ops = &show_set_event_seq_ops;
1278 struct trace_array *tr = inode->i_private;
1279
1280 if ((file->f_mode & FMODE_WRITE) &&
1281 (file->f_flags & O_TRUNC))
1282 ftrace_clear_events(tr);
1283
1284 return ftrace_event_open(inode, file, seq_ops);
1285 }
1286
1287 static struct event_subsystem *
1288 create_new_subsystem(const char *name)
1289 {
1290 struct event_subsystem *system;
1291
1292 /* need to create new entry */
1293 system = kmalloc(sizeof(*system), GFP_KERNEL);
1294 if (!system)
1295 return NULL;
1296
1297 system->ref_count = 1;
1298
1299 /* Only allocate if dynamic (kprobes and modules) */
1300 if (!core_kernel_data((unsigned long)name)) {
1301 system->ref_count |= SYSTEM_FL_FREE_NAME;
1302 system->name = kstrdup(name, GFP_KERNEL);
1303 if (!system->name)
1304 goto out_free;
1305 } else
1306 system->name = name;
1307
1308 system->filter = NULL;
1309
1310 system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
1311 if (!system->filter)
1312 goto out_free;
1313
1314 list_add(&system->list, &event_subsystems);
1315
1316 return system;
1317
1318 out_free:
1319 if (system->ref_count & SYSTEM_FL_FREE_NAME)
1320 kfree(system->name);
1321 kfree(system);
1322 return NULL;
1323 }
1324
1325 static struct dentry *
1326 event_subsystem_dir(struct trace_array *tr, const char *name,
1327 struct ftrace_event_file *file, struct dentry *parent)
1328 {
1329 struct ftrace_subsystem_dir *dir;
1330 struct event_subsystem *system;
1331 struct dentry *entry;
1332
1333 /* First see if we did not already create this dir */
1334 list_for_each_entry(dir, &tr->systems, list) {
1335 system = dir->subsystem;
1336 if (strcmp(system->name, name) == 0) {
1337 dir->nr_events++;
1338 file->system = dir;
1339 return dir->entry;
1340 }
1341 }
1342
1343 /* Now see if the system itself exists. */
1344 list_for_each_entry(system, &event_subsystems, list) {
1345 if (strcmp(system->name, name) == 0)
1346 break;
1347 }
1348 /* Reset system variable when not found */
1349 if (&system->list == &event_subsystems)
1350 system = NULL;
1351
1352 dir = kmalloc(sizeof(*dir), GFP_KERNEL);
1353 if (!dir)
1354 goto out_fail;
1355
1356 if (!system) {
1357 system = create_new_subsystem(name);
1358 if (!system)
1359 goto out_free;
1360 } else
1361 __get_system(system);
1362
1363 dir->entry = debugfs_create_dir(name, parent);
1364 if (!dir->entry) {
1365 pr_warning("Failed to create system directory %s\n", name);
1366 __put_system(system);
1367 goto out_free;
1368 }
1369
1370 dir->tr = tr;
1371 dir->ref_count = 1;
1372 dir->nr_events = 1;
1373 dir->subsystem = system;
1374 file->system = dir;
1375
1376 entry = debugfs_create_file("filter", 0644, dir->entry, dir,
1377 &ftrace_subsystem_filter_fops);
1378 if (!entry) {
1379 kfree(system->filter);
1380 system->filter = NULL;
1381 pr_warning("Could not create debugfs '%s/filter' entry\n", name);
1382 }
1383
1384 trace_create_file("enable", 0644, dir->entry, dir,
1385 &ftrace_system_enable_fops);
1386
1387 list_add(&dir->list, &tr->systems);
1388
1389 return dir->entry;
1390
1391 out_free:
1392 kfree(dir);
1393 out_fail:
1394 /* Only print this message if failed on memory allocation */
1395 if (!dir || !system)
1396 pr_warning("No memory to create event subsystem %s\n",
1397 name);
1398 return NULL;
1399 }
1400
1401 static int
1402 event_create_dir(struct dentry *parent,
1403 struct ftrace_event_file *file,
1404 const struct file_operations *id,
1405 const struct file_operations *enable,
1406 const struct file_operations *filter,
1407 const struct file_operations *format)
1408 {
1409 struct ftrace_event_call *call = file->event_call;
1410 struct trace_array *tr = file->tr;
1411 struct list_head *head;
1412 struct dentry *d_events;
1413 int ret;
1414
1415 /*
1416 * If the trace point header did not define TRACE_SYSTEM
1417 * then the system would be called "TRACE_SYSTEM".
1418 */
1419 if (strcmp(call->class->system, TRACE_SYSTEM) != 0) {
1420 d_events = event_subsystem_dir(tr, call->class->system, file, parent);
1421 if (!d_events)
1422 return -ENOMEM;
1423 } else
1424 d_events = parent;
1425
1426 file->dir = debugfs_create_dir(call->name, d_events);
1427 if (!file->dir) {
1428 pr_warning("Could not create debugfs '%s' directory\n",
1429 call->name);
1430 return -1;
1431 }
1432
1433 if (call->class->reg && !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
1434 trace_create_file("enable", 0644, file->dir, file,
1435 enable);
1436
1437 #ifdef CONFIG_PERF_EVENTS
1438 if (call->event.type && call->class->reg)
1439 trace_create_file("id", 0444, file->dir, call,
1440 id);
1441 #endif
1442
1443 /*
1444 * Other events may have the same class. Only update
1445 * the fields if they are not already defined.
1446 */
1447 head = trace_get_fields(call);
1448 if (list_empty(head)) {
1449 ret = call->class->define_fields(call);
1450 if (ret < 0) {
1451 pr_warning("Could not initialize trace point"
1452 " events/%s\n", call->name);
1453 return -1;
1454 }
1455 }
1456 trace_create_file("filter", 0644, file->dir, call,
1457 filter);
1458
1459 trace_create_file("format", 0444, file->dir, call,
1460 format);
1461
1462 return 0;
1463 }
1464
1465 static void remove_subsystem(struct ftrace_subsystem_dir *dir)
1466 {
1467 if (!dir)
1468 return;
1469
1470 if (!--dir->nr_events) {
1471 debugfs_remove_recursive(dir->entry);
1472 list_del(&dir->list);
1473 __put_system_dir(dir);
1474 }
1475 }
1476
1477 static void remove_event_from_tracers(struct ftrace_event_call *call)
1478 {
1479 struct ftrace_event_file *file;
1480 struct trace_array *tr;
1481
1482 do_for_each_event_file_safe(tr, file) {
1483
1484 if (file->event_call != call)
1485 continue;
1486
1487 list_del(&file->list);
1488 debugfs_remove_recursive(file->dir);
1489 remove_subsystem(file->system);
1490 kmem_cache_free(file_cachep, file);
1491
1492 /*
1493 * The do_for_each_event_file_safe() is
1494 * a double loop. After finding the call for this
1495 * trace_array, we use break to jump to the next
1496 * trace_array.
1497 */
1498 break;
1499 } while_for_each_event_file();
1500 }
1501
1502 static void event_remove(struct ftrace_event_call *call)
1503 {
1504 struct trace_array *tr;
1505 struct ftrace_event_file *file;
1506
1507 do_for_each_event_file(tr, file) {
1508 if (file->event_call != call)
1509 continue;
1510 ftrace_event_enable_disable(file, 0);
1511 /*
1512 * The do_for_each_event_file() is
1513 * a double loop. After finding the call for this
1514 * trace_array, we use break to jump to the next
1515 * trace_array.
1516 */
1517 break;
1518 } while_for_each_event_file();
1519
1520 if (call->event.funcs)
1521 __unregister_ftrace_event(&call->event);
1522 remove_event_from_tracers(call);
1523 list_del(&call->list);
1524 }
1525
1526 static int event_init(struct ftrace_event_call *call)
1527 {
1528 int ret = 0;
1529
1530 if (WARN_ON(!call->name))
1531 return -EINVAL;
1532
1533 if (call->class->raw_init) {
1534 ret = call->class->raw_init(call);
1535 if (ret < 0 && ret != -ENOSYS)
1536 pr_warn("Could not initialize trace events/%s\n",
1537 call->name);
1538 }
1539
1540 return ret;
1541 }
1542
1543 static int
1544 __register_event(struct ftrace_event_call *call, struct module *mod)
1545 {
1546 int ret;
1547
1548 ret = event_init(call);
1549 if (ret < 0)
1550 return ret;
1551
1552 list_add(&call->list, &ftrace_events);
1553 call->mod = mod;
1554
1555 return 0;
1556 }
1557
1558 static struct ftrace_event_file *
1559 trace_create_new_event(struct ftrace_event_call *call,
1560 struct trace_array *tr)
1561 {
1562 struct ftrace_event_file *file;
1563
1564 file = kmem_cache_alloc(file_cachep, GFP_TRACE);
1565 if (!file)
1566 return NULL;
1567
1568 file->event_call = call;
1569 file->tr = tr;
1570 atomic_set(&file->sm_ref, 0);
1571 list_add(&file->list, &tr->events);
1572
1573 return file;
1574 }
1575
1576 /* Add an event to a trace directory */
1577 static int
1578 __trace_add_new_event(struct ftrace_event_call *call,
1579 struct trace_array *tr,
1580 const struct file_operations *id,
1581 const struct file_operations *enable,
1582 const struct file_operations *filter,
1583 const struct file_operations *format)
1584 {
1585 struct ftrace_event_file *file;
1586
1587 file = trace_create_new_event(call, tr);
1588 if (!file)
1589 return -ENOMEM;
1590
1591 return event_create_dir(tr->event_dir, file, id, enable, filter, format);
1592 }
1593
1594 /*
1595 * Just create a decriptor for early init. A descriptor is required
1596 * for enabling events at boot. We want to enable events before
1597 * the filesystem is initialized.
1598 */
1599 static __init int
1600 __trace_early_add_new_event(struct ftrace_event_call *call,
1601 struct trace_array *tr)
1602 {
1603 struct ftrace_event_file *file;
1604
1605 file = trace_create_new_event(call, tr);
1606 if (!file)
1607 return -ENOMEM;
1608
1609 return 0;
1610 }
1611
1612 struct ftrace_module_file_ops;
1613 static void __add_event_to_tracers(struct ftrace_event_call *call,
1614 struct ftrace_module_file_ops *file_ops);
1615
1616 /* Add an additional event_call dynamically */
1617 int trace_add_event_call(struct ftrace_event_call *call)
1618 {
1619 int ret;
1620 mutex_lock(&event_mutex);
1621
1622 ret = __register_event(call, NULL);
1623 if (ret >= 0)
1624 __add_event_to_tracers(call, NULL);
1625
1626 mutex_unlock(&event_mutex);
1627 return ret;
1628 }
1629
1630 /*
1631 * Must be called under locking both of event_mutex and trace_event_sem.
1632 */
1633 static void __trace_remove_event_call(struct ftrace_event_call *call)
1634 {
1635 event_remove(call);
1636 trace_destroy_fields(call);
1637 destroy_preds(call);
1638 }
1639
1640 /* Remove an event_call */
1641 void trace_remove_event_call(struct ftrace_event_call *call)
1642 {
1643 mutex_lock(&event_mutex);
1644 down_write(&trace_event_sem);
1645 __trace_remove_event_call(call);
1646 up_write(&trace_event_sem);
1647 mutex_unlock(&event_mutex);
1648 }
1649
1650 #define for_each_event(event, start, end) \
1651 for (event = start; \
1652 (unsigned long)event < (unsigned long)end; \
1653 event++)
1654
1655 #ifdef CONFIG_MODULES
1656
1657 static LIST_HEAD(ftrace_module_file_list);
1658
1659 /*
1660 * Modules must own their file_operations to keep up with
1661 * reference counting.
1662 */
1663 struct ftrace_module_file_ops {
1664 struct list_head list;
1665 struct module *mod;
1666 struct file_operations id;
1667 struct file_operations enable;
1668 struct file_operations format;
1669 struct file_operations filter;
1670 };
1671
1672 static struct ftrace_module_file_ops *
1673 find_ftrace_file_ops(struct ftrace_module_file_ops *file_ops, struct module *mod)
1674 {
1675 /*
1676 * As event_calls are added in groups by module,
1677 * when we find one file_ops, we don't need to search for
1678 * each call in that module, as the rest should be the
1679 * same. Only search for a new one if the last one did
1680 * not match.
1681 */
1682 if (file_ops && mod == file_ops->mod)
1683 return file_ops;
1684
1685 list_for_each_entry(file_ops, &ftrace_module_file_list, list) {
1686 if (file_ops->mod == mod)
1687 return file_ops;
1688 }
1689 return NULL;
1690 }
1691
1692 static struct ftrace_module_file_ops *
1693 trace_create_file_ops(struct module *mod)
1694 {
1695 struct ftrace_module_file_ops *file_ops;
1696
1697 /*
1698 * This is a bit of a PITA. To allow for correct reference
1699 * counting, modules must "own" their file_operations.
1700 * To do this, we allocate the file operations that will be
1701 * used in the event directory.
1702 */
1703
1704 file_ops = kmalloc(sizeof(*file_ops), GFP_KERNEL);
1705 if (!file_ops)
1706 return NULL;
1707
1708 file_ops->mod = mod;
1709
1710 file_ops->id = ftrace_event_id_fops;
1711 file_ops->id.owner = mod;
1712
1713 file_ops->enable = ftrace_enable_fops;
1714 file_ops->enable.owner = mod;
1715
1716 file_ops->filter = ftrace_event_filter_fops;
1717 file_ops->filter.owner = mod;
1718
1719 file_ops->format = ftrace_event_format_fops;
1720 file_ops->format.owner = mod;
1721
1722 list_add(&file_ops->list, &ftrace_module_file_list);
1723
1724 return file_ops;
1725 }
1726
1727 static void trace_module_add_events(struct module *mod)
1728 {
1729 struct ftrace_module_file_ops *file_ops = NULL;
1730 struct ftrace_event_call **call, **start, **end;
1731
1732 start = mod->trace_events;
1733 end = mod->trace_events + mod->num_trace_events;
1734
1735 if (start == end)
1736 return;
1737
1738 file_ops = trace_create_file_ops(mod);
1739 if (!file_ops)
1740 return;
1741
1742 for_each_event(call, start, end) {
1743 __register_event(*call, mod);
1744 __add_event_to_tracers(*call, file_ops);
1745 }
1746 }
1747
1748 static void trace_module_remove_events(struct module *mod)
1749 {
1750 struct ftrace_module_file_ops *file_ops;
1751 struct ftrace_event_call *call, *p;
1752 bool clear_trace = false;
1753
1754 down_write(&trace_event_sem);
1755 list_for_each_entry_safe(call, p, &ftrace_events, list) {
1756 if (call->mod == mod) {
1757 if (call->flags & TRACE_EVENT_FL_WAS_ENABLED)
1758 clear_trace = true;
1759 __trace_remove_event_call(call);
1760 }
1761 }
1762
1763 /* Now free the file_operations */
1764 list_for_each_entry(file_ops, &ftrace_module_file_list, list) {
1765 if (file_ops->mod == mod)
1766 break;
1767 }
1768 if (&file_ops->list != &ftrace_module_file_list) {
1769 list_del(&file_ops->list);
1770 kfree(file_ops);
1771 }
1772 up_write(&trace_event_sem);
1773
1774 /*
1775 * It is safest to reset the ring buffer if the module being unloaded
1776 * registered any events that were used. The only worry is if
1777 * a new module gets loaded, and takes on the same id as the events
1778 * of this module. When printing out the buffer, traced events left
1779 * over from this module may be passed to the new module events and
1780 * unexpected results may occur.
1781 */
1782 if (clear_trace)
1783 tracing_reset_all_online_cpus();
1784 }
1785
1786 static int trace_module_notify(struct notifier_block *self,
1787 unsigned long val, void *data)
1788 {
1789 struct module *mod = data;
1790
1791 mutex_lock(&event_mutex);
1792 switch (val) {
1793 case MODULE_STATE_COMING:
1794 trace_module_add_events(mod);
1795 break;
1796 case MODULE_STATE_GOING:
1797 trace_module_remove_events(mod);
1798 break;
1799 }
1800 mutex_unlock(&event_mutex);
1801
1802 return 0;
1803 }
1804
1805 static int
1806 __trace_add_new_mod_event(struct ftrace_event_call *call,
1807 struct trace_array *tr,
1808 struct ftrace_module_file_ops *file_ops)
1809 {
1810 return __trace_add_new_event(call, tr,
1811 &file_ops->id, &file_ops->enable,
1812 &file_ops->filter, &file_ops->format);
1813 }
1814
1815 #else
1816 static inline struct ftrace_module_file_ops *
1817 find_ftrace_file_ops(struct ftrace_module_file_ops *file_ops, struct module *mod)
1818 {
1819 return NULL;
1820 }
1821 static inline int trace_module_notify(struct notifier_block *self,
1822 unsigned long val, void *data)
1823 {
1824 return 0;
1825 }
1826 static inline int
1827 __trace_add_new_mod_event(struct ftrace_event_call *call,
1828 struct trace_array *tr,
1829 struct ftrace_module_file_ops *file_ops)
1830 {
1831 return -ENODEV;
1832 }
1833 #endif /* CONFIG_MODULES */
1834
1835 /* Create a new event directory structure for a trace directory. */
1836 static void
1837 __trace_add_event_dirs(struct trace_array *tr)
1838 {
1839 struct ftrace_module_file_ops *file_ops = NULL;
1840 struct ftrace_event_call *call;
1841 int ret;
1842
1843 list_for_each_entry(call, &ftrace_events, list) {
1844 if (call->mod) {
1845 /*
1846 * Directories for events by modules need to
1847 * keep module ref counts when opened (as we don't
1848 * want the module to disappear when reading one
1849 * of these files). The file_ops keep account of
1850 * the module ref count.
1851 */
1852 file_ops = find_ftrace_file_ops(file_ops, call->mod);
1853 if (!file_ops)
1854 continue; /* Warn? */
1855 ret = __trace_add_new_mod_event(call, tr, file_ops);
1856 if (ret < 0)
1857 pr_warning("Could not create directory for event %s\n",
1858 call->name);
1859 continue;
1860 }
1861 ret = __trace_add_new_event(call, tr,
1862 &ftrace_event_id_fops,
1863 &ftrace_enable_fops,
1864 &ftrace_event_filter_fops,
1865 &ftrace_event_format_fops);
1866 if (ret < 0)
1867 pr_warning("Could not create directory for event %s\n",
1868 call->name);
1869 }
1870 }
1871
1872 #ifdef CONFIG_DYNAMIC_FTRACE
1873
1874 /* Avoid typos */
1875 #define ENABLE_EVENT_STR "enable_event"
1876 #define DISABLE_EVENT_STR "disable_event"
1877
1878 struct event_probe_data {
1879 struct ftrace_event_file *file;
1880 unsigned long count;
1881 int ref;
1882 bool enable;
1883 };
1884
1885 static struct ftrace_event_file *
1886 find_event_file(struct trace_array *tr, const char *system, const char *event)
1887 {
1888 struct ftrace_event_file *file;
1889 struct ftrace_event_call *call;
1890
1891 list_for_each_entry(file, &tr->events, list) {
1892
1893 call = file->event_call;
1894
1895 if (!call->name || !call->class || !call->class->reg)
1896 continue;
1897
1898 if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
1899 continue;
1900
1901 if (strcmp(event, call->name) == 0 &&
1902 strcmp(system, call->class->system) == 0)
1903 return file;
1904 }
1905 return NULL;
1906 }
1907
1908 static void
1909 event_enable_probe(unsigned long ip, unsigned long parent_ip, void **_data)
1910 {
1911 struct event_probe_data **pdata = (struct event_probe_data **)_data;
1912 struct event_probe_data *data = *pdata;
1913
1914 if (!data)
1915 return;
1916
1917 if (data->enable)
1918 clear_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &data->file->flags);
1919 else
1920 set_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &data->file->flags);
1921 }
1922
1923 static void
1924 event_enable_count_probe(unsigned long ip, unsigned long parent_ip, void **_data)
1925 {
1926 struct event_probe_data **pdata = (struct event_probe_data **)_data;
1927 struct event_probe_data *data = *pdata;
1928
1929 if (!data)
1930 return;
1931
1932 if (!data->count)
1933 return;
1934
1935 /* Skip if the event is in a state we want to switch to */
1936 if (data->enable == !(data->file->flags & FTRACE_EVENT_FL_SOFT_DISABLED))
1937 return;
1938
1939 if (data->count != -1)
1940 (data->count)--;
1941
1942 event_enable_probe(ip, parent_ip, _data);
1943 }
1944
1945 static int
1946 event_enable_print(struct seq_file *m, unsigned long ip,
1947 struct ftrace_probe_ops *ops, void *_data)
1948 {
1949 struct event_probe_data *data = _data;
1950
1951 seq_printf(m, "%ps:", (void *)ip);
1952
1953 seq_printf(m, "%s:%s:%s",
1954 data->enable ? ENABLE_EVENT_STR : DISABLE_EVENT_STR,
1955 data->file->event_call->class->system,
1956 data->file->event_call->name);
1957
1958 if (data->count == -1)
1959 seq_printf(m, ":unlimited\n");
1960 else
1961 seq_printf(m, ":count=%ld\n", data->count);
1962
1963 return 0;
1964 }
1965
1966 static int
1967 event_enable_init(struct ftrace_probe_ops *ops, unsigned long ip,
1968 void **_data)
1969 {
1970 struct event_probe_data **pdata = (struct event_probe_data **)_data;
1971 struct event_probe_data *data = *pdata;
1972
1973 data->ref++;
1974 return 0;
1975 }
1976
1977 static void
1978 event_enable_free(struct ftrace_probe_ops *ops, unsigned long ip,
1979 void **_data)
1980 {
1981 struct event_probe_data **pdata = (struct event_probe_data **)_data;
1982 struct event_probe_data *data = *pdata;
1983
1984 if (WARN_ON_ONCE(data->ref <= 0))
1985 return;
1986
1987 data->ref--;
1988 if (!data->ref) {
1989 /* Remove the SOFT_MODE flag */
1990 __ftrace_event_enable_disable(data->file, 0, 1);
1991 module_put(data->file->event_call->mod);
1992 kfree(data);
1993 }
1994 *pdata = NULL;
1995 }
1996
1997 static struct ftrace_probe_ops event_enable_probe_ops = {
1998 .func = event_enable_probe,
1999 .print = event_enable_print,
2000 .init = event_enable_init,
2001 .free = event_enable_free,
2002 };
2003
2004 static struct ftrace_probe_ops event_enable_count_probe_ops = {
2005 .func = event_enable_count_probe,
2006 .print = event_enable_print,
2007 .init = event_enable_init,
2008 .free = event_enable_free,
2009 };
2010
2011 static struct ftrace_probe_ops event_disable_probe_ops = {
2012 .func = event_enable_probe,
2013 .print = event_enable_print,
2014 .init = event_enable_init,
2015 .free = event_enable_free,
2016 };
2017
2018 static struct ftrace_probe_ops event_disable_count_probe_ops = {
2019 .func = event_enable_count_probe,
2020 .print = event_enable_print,
2021 .init = event_enable_init,
2022 .free = event_enable_free,
2023 };
2024
2025 static int
2026 event_enable_func(struct ftrace_hash *hash,
2027 char *glob, char *cmd, char *param, int enabled)
2028 {
2029 struct trace_array *tr = top_trace_array();
2030 struct ftrace_event_file *file;
2031 struct ftrace_probe_ops *ops;
2032 struct event_probe_data *data;
2033 const char *system;
2034 const char *event;
2035 char *number;
2036 bool enable;
2037 int ret;
2038
2039 /* hash funcs only work with set_ftrace_filter */
2040 if (!enabled || !param)
2041 return -EINVAL;
2042
2043 system = strsep(&param, ":");
2044 if (!param)
2045 return -EINVAL;
2046
2047 event = strsep(&param, ":");
2048
2049 mutex_lock(&event_mutex);
2050
2051 ret = -EINVAL;
2052 file = find_event_file(tr, system, event);
2053 if (!file)
2054 goto out;
2055
2056 enable = strcmp(cmd, ENABLE_EVENT_STR) == 0;
2057
2058 if (enable)
2059 ops = param ? &event_enable_count_probe_ops : &event_enable_probe_ops;
2060 else
2061 ops = param ? &event_disable_count_probe_ops : &event_disable_probe_ops;
2062
2063 if (glob[0] == '!') {
2064 unregister_ftrace_function_probe_func(glob+1, ops);
2065 ret = 0;
2066 goto out;
2067 }
2068
2069 ret = -ENOMEM;
2070 data = kzalloc(sizeof(*data), GFP_KERNEL);
2071 if (!data)
2072 goto out;
2073
2074 data->enable = enable;
2075 data->count = -1;
2076 data->file = file;
2077
2078 if (!param)
2079 goto out_reg;
2080
2081 number = strsep(&param, ":");
2082
2083 ret = -EINVAL;
2084 if (!strlen(number))
2085 goto out_free;
2086
2087 /*
2088 * We use the callback data field (which is a pointer)
2089 * as our counter.
2090 */
2091 ret = kstrtoul(number, 0, &data->count);
2092 if (ret)
2093 goto out_free;
2094
2095 out_reg:
2096 /* Don't let event modules unload while probe registered */
2097 ret = try_module_get(file->event_call->mod);
2098 if (!ret) {
2099 ret = -EBUSY;
2100 goto out_free;
2101 }
2102
2103 ret = __ftrace_event_enable_disable(file, 1, 1);
2104 if (ret < 0)
2105 goto out_put;
2106 ret = register_ftrace_function_probe(glob, ops, data);
2107 /*
2108 * The above returns on success the # of functions enabled,
2109 * but if it didn't find any functions it returns zero.
2110 * Consider no functions a failure too.
2111 */
2112 if (!ret) {
2113 ret = -ENOENT;
2114 goto out_disable;
2115 } else if (ret < 0)
2116 goto out_disable;
2117 /* Just return zero, not the number of enabled functions */
2118 ret = 0;
2119 out:
2120 mutex_unlock(&event_mutex);
2121 return ret;
2122
2123 out_disable:
2124 __ftrace_event_enable_disable(file, 0, 1);
2125 out_put:
2126 module_put(file->event_call->mod);
2127 out_free:
2128 kfree(data);
2129 goto out;
2130 }
2131
2132 static struct ftrace_func_command event_enable_cmd = {
2133 .name = ENABLE_EVENT_STR,
2134 .func = event_enable_func,
2135 };
2136
2137 static struct ftrace_func_command event_disable_cmd = {
2138 .name = DISABLE_EVENT_STR,
2139 .func = event_enable_func,
2140 };
2141
2142 static __init int register_event_cmds(void)
2143 {
2144 int ret;
2145
2146 ret = register_ftrace_command(&event_enable_cmd);
2147 if (WARN_ON(ret < 0))
2148 return ret;
2149 ret = register_ftrace_command(&event_disable_cmd);
2150 if (WARN_ON(ret < 0))
2151 unregister_ftrace_command(&event_enable_cmd);
2152 return ret;
2153 }
2154 #else
2155 static inline int register_event_cmds(void) { return 0; }
2156 #endif /* CONFIG_DYNAMIC_FTRACE */
2157
2158 /*
2159 * The top level array has already had its ftrace_event_file
2160 * descriptors created in order to allow for early events to
2161 * be recorded. This function is called after the debugfs has been
2162 * initialized, and we now have to create the files associated
2163 * to the events.
2164 */
2165 static __init void
2166 __trace_early_add_event_dirs(struct trace_array *tr)
2167 {
2168 struct ftrace_event_file *file;
2169 int ret;
2170
2171
2172 list_for_each_entry(file, &tr->events, list) {
2173 ret = event_create_dir(tr->event_dir, file,
2174 &ftrace_event_id_fops,
2175 &ftrace_enable_fops,
2176 &ftrace_event_filter_fops,
2177 &ftrace_event_format_fops);
2178 if (ret < 0)
2179 pr_warning("Could not create directory for event %s\n",
2180 file->event_call->name);
2181 }
2182 }
2183
2184 /*
2185 * For early boot up, the top trace array requires to have
2186 * a list of events that can be enabled. This must be done before
2187 * the filesystem is set up in order to allow events to be traced
2188 * early.
2189 */
2190 static __init void
2191 __trace_early_add_events(struct trace_array *tr)
2192 {
2193 struct ftrace_event_call *call;
2194 int ret;
2195
2196 list_for_each_entry(call, &ftrace_events, list) {
2197 /* Early boot up should not have any modules loaded */
2198 if (WARN_ON_ONCE(call->mod))
2199 continue;
2200
2201 ret = __trace_early_add_new_event(call, tr);
2202 if (ret < 0)
2203 pr_warning("Could not create early event %s\n",
2204 call->name);
2205 }
2206 }
2207
2208 /* Remove the event directory structure for a trace directory. */
2209 static void
2210 __trace_remove_event_dirs(struct trace_array *tr)
2211 {
2212 struct ftrace_event_file *file, *next;
2213
2214 list_for_each_entry_safe(file, next, &tr->events, list) {
2215 list_del(&file->list);
2216 debugfs_remove_recursive(file->dir);
2217 remove_subsystem(file->system);
2218 kmem_cache_free(file_cachep, file);
2219 }
2220 }
2221
2222 static void
2223 __add_event_to_tracers(struct ftrace_event_call *call,
2224 struct ftrace_module_file_ops *file_ops)
2225 {
2226 struct trace_array *tr;
2227
2228 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
2229 if (file_ops)
2230 __trace_add_new_mod_event(call, tr, file_ops);
2231 else
2232 __trace_add_new_event(call, tr,
2233 &ftrace_event_id_fops,
2234 &ftrace_enable_fops,
2235 &ftrace_event_filter_fops,
2236 &ftrace_event_format_fops);
2237 }
2238 }
2239
2240 static struct notifier_block trace_module_nb = {
2241 .notifier_call = trace_module_notify,
2242 .priority = 0,
2243 };
2244
2245 extern struct ftrace_event_call *__start_ftrace_events[];
2246 extern struct ftrace_event_call *__stop_ftrace_events[];
2247
2248 static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
2249
2250 static __init int setup_trace_event(char *str)
2251 {
2252 strlcpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
2253 ring_buffer_expanded = true;
2254 tracing_selftest_disabled = true;
2255
2256 return 1;
2257 }
2258 __setup("trace_event=", setup_trace_event);
2259
2260 /* Expects to have event_mutex held when called */
2261 static int
2262 create_event_toplevel_files(struct dentry *parent, struct trace_array *tr)
2263 {
2264 struct dentry *d_events;
2265 struct dentry *entry;
2266
2267 entry = debugfs_create_file("set_event", 0644, parent,
2268 tr, &ftrace_set_event_fops);
2269 if (!entry) {
2270 pr_warning("Could not create debugfs 'set_event' entry\n");
2271 return -ENOMEM;
2272 }
2273
2274 d_events = debugfs_create_dir("events", parent);
2275 if (!d_events) {
2276 pr_warning("Could not create debugfs 'events' directory\n");
2277 return -ENOMEM;
2278 }
2279
2280 /* ring buffer internal formats */
2281 trace_create_file("header_page", 0444, d_events,
2282 ring_buffer_print_page_header,
2283 &ftrace_show_header_fops);
2284
2285 trace_create_file("header_event", 0444, d_events,
2286 ring_buffer_print_entry_header,
2287 &ftrace_show_header_fops);
2288
2289 trace_create_file("enable", 0644, d_events,
2290 tr, &ftrace_tr_enable_fops);
2291
2292 tr->event_dir = d_events;
2293
2294 return 0;
2295 }
2296
2297 /**
2298 * event_trace_add_tracer - add a instance of a trace_array to events
2299 * @parent: The parent dentry to place the files/directories for events in
2300 * @tr: The trace array associated with these events
2301 *
2302 * When a new instance is created, it needs to set up its events
2303 * directory, as well as other files associated with events. It also
2304 * creates the event hierachry in the @parent/events directory.
2305 *
2306 * Returns 0 on success.
2307 */
2308 int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr)
2309 {
2310 int ret;
2311
2312 mutex_lock(&event_mutex);
2313
2314 ret = create_event_toplevel_files(parent, tr);
2315 if (ret)
2316 goto out_unlock;
2317
2318 down_write(&trace_event_sem);
2319 __trace_add_event_dirs(tr);
2320 up_write(&trace_event_sem);
2321
2322 out_unlock:
2323 mutex_unlock(&event_mutex);
2324
2325 return ret;
2326 }
2327
2328 /*
2329 * The top trace array already had its file descriptors created.
2330 * Now the files themselves need to be created.
2331 */
2332 static __init int
2333 early_event_add_tracer(struct dentry *parent, struct trace_array *tr)
2334 {
2335 int ret;
2336
2337 mutex_lock(&event_mutex);
2338
2339 ret = create_event_toplevel_files(parent, tr);
2340 if (ret)
2341 goto out_unlock;
2342
2343 down_write(&trace_event_sem);
2344 __trace_early_add_event_dirs(tr);
2345 up_write(&trace_event_sem);
2346
2347 out_unlock:
2348 mutex_unlock(&event_mutex);
2349
2350 return ret;
2351 }
2352
2353 int event_trace_del_tracer(struct trace_array *tr)
2354 {
2355 /* Disable any running events */
2356 __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
2357
2358 mutex_lock(&event_mutex);
2359
2360 down_write(&trace_event_sem);
2361 __trace_remove_event_dirs(tr);
2362 debugfs_remove_recursive(tr->event_dir);
2363 up_write(&trace_event_sem);
2364
2365 tr->event_dir = NULL;
2366
2367 mutex_unlock(&event_mutex);
2368
2369 return 0;
2370 }
2371
2372 static __init int event_trace_memsetup(void)
2373 {
2374 field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC);
2375 file_cachep = KMEM_CACHE(ftrace_event_file, SLAB_PANIC);
2376 return 0;
2377 }
2378
2379 static __init int event_trace_enable(void)
2380 {
2381 struct trace_array *tr = top_trace_array();
2382 struct ftrace_event_call **iter, *call;
2383 char *buf = bootup_event_buf;
2384 char *token;
2385 int ret;
2386
2387 for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
2388
2389 call = *iter;
2390 ret = event_init(call);
2391 if (!ret)
2392 list_add(&call->list, &ftrace_events);
2393 }
2394
2395 /*
2396 * We need the top trace array to have a working set of trace
2397 * points at early init, before the debug files and directories
2398 * are created. Create the file entries now, and attach them
2399 * to the actual file dentries later.
2400 */
2401 __trace_early_add_events(tr);
2402
2403 while (true) {
2404 token = strsep(&buf, ",");
2405
2406 if (!token)
2407 break;
2408 if (!*token)
2409 continue;
2410
2411 ret = ftrace_set_clr_event(tr, token, 1);
2412 if (ret)
2413 pr_warn("Failed to enable trace event: %s\n", token);
2414 }
2415
2416 trace_printk_start_comm();
2417
2418 register_event_cmds();
2419
2420 return 0;
2421 }
2422
2423 static __init int event_trace_init(void)
2424 {
2425 struct trace_array *tr;
2426 struct dentry *d_tracer;
2427 struct dentry *entry;
2428 int ret;
2429
2430 tr = top_trace_array();
2431
2432 d_tracer = tracing_init_dentry();
2433 if (!d_tracer)
2434 return 0;
2435
2436 entry = debugfs_create_file("available_events", 0444, d_tracer,
2437 tr, &ftrace_avail_fops);
2438 if (!entry)
2439 pr_warning("Could not create debugfs "
2440 "'available_events' entry\n");
2441
2442 if (trace_define_common_fields())
2443 pr_warning("tracing: Failed to allocate common fields");
2444
2445 ret = early_event_add_tracer(d_tracer, tr);
2446 if (ret)
2447 return ret;
2448
2449 ret = register_module_notifier(&trace_module_nb);
2450 if (ret)
2451 pr_warning("Failed to register trace events module notifier\n");
2452
2453 return 0;
2454 }
2455 early_initcall(event_trace_memsetup);
2456 core_initcall(event_trace_enable);
2457 fs_initcall(event_trace_init);
2458
2459 #ifdef CONFIG_FTRACE_STARTUP_TEST
2460
2461 static DEFINE_SPINLOCK(test_spinlock);
2462 static DEFINE_SPINLOCK(test_spinlock_irq);
2463 static DEFINE_MUTEX(test_mutex);
2464
2465 static __init void test_work(struct work_struct *dummy)
2466 {
2467 spin_lock(&test_spinlock);
2468 spin_lock_irq(&test_spinlock_irq);
2469 udelay(1);
2470 spin_unlock_irq(&test_spinlock_irq);
2471 spin_unlock(&test_spinlock);
2472
2473 mutex_lock(&test_mutex);
2474 msleep(1);
2475 mutex_unlock(&test_mutex);
2476 }
2477
2478 static __init int event_test_thread(void *unused)
2479 {
2480 void *test_malloc;
2481
2482 test_malloc = kmalloc(1234, GFP_KERNEL);
2483 if (!test_malloc)
2484 pr_info("failed to kmalloc\n");
2485
2486 schedule_on_each_cpu(test_work);
2487
2488 kfree(test_malloc);
2489
2490 set_current_state(TASK_INTERRUPTIBLE);
2491 while (!kthread_should_stop())
2492 schedule();
2493
2494 return 0;
2495 }
2496
2497 /*
2498 * Do various things that may trigger events.
2499 */
2500 static __init void event_test_stuff(void)
2501 {
2502 struct task_struct *test_thread;
2503
2504 test_thread = kthread_run(event_test_thread, NULL, "test-events");
2505 msleep(1);
2506 kthread_stop(test_thread);
2507 }
2508
2509 /*
2510 * For every trace event defined, we will test each trace point separately,
2511 * and then by groups, and finally all trace points.
2512 */
2513 static __init void event_trace_self_tests(void)
2514 {
2515 struct ftrace_subsystem_dir *dir;
2516 struct ftrace_event_file *file;
2517 struct ftrace_event_call *call;
2518 struct event_subsystem *system;
2519 struct trace_array *tr;
2520 int ret;
2521
2522 tr = top_trace_array();
2523
2524 pr_info("Running tests on trace events:\n");
2525
2526 list_for_each_entry(file, &tr->events, list) {
2527
2528 call = file->event_call;
2529
2530 /* Only test those that have a probe */
2531 if (!call->class || !call->class->probe)
2532 continue;
2533
2534 /*
2535 * Testing syscall events here is pretty useless, but
2536 * we still do it if configured. But this is time consuming.
2537 * What we really need is a user thread to perform the
2538 * syscalls as we test.
2539 */
2540 #ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
2541 if (call->class->system &&
2542 strcmp(call->class->system, "syscalls") == 0)
2543 continue;
2544 #endif
2545
2546 pr_info("Testing event %s: ", call->name);
2547
2548 /*
2549 * If an event is already enabled, someone is using
2550 * it and the self test should not be on.
2551 */
2552 if (file->flags & FTRACE_EVENT_FL_ENABLED) {
2553 pr_warning("Enabled event during self test!\n");
2554 WARN_ON_ONCE(1);
2555 continue;
2556 }
2557
2558 ftrace_event_enable_disable(file, 1);
2559 event_test_stuff();
2560 ftrace_event_enable_disable(file, 0);
2561
2562 pr_cont("OK\n");
2563 }
2564
2565 /* Now test at the sub system level */
2566
2567 pr_info("Running tests on trace event systems:\n");
2568
2569 list_for_each_entry(dir, &tr->systems, list) {
2570
2571 system = dir->subsystem;
2572
2573 /* the ftrace system is special, skip it */
2574 if (strcmp(system->name, "ftrace") == 0)
2575 continue;
2576
2577 pr_info("Testing event system %s: ", system->name);
2578
2579 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1);
2580 if (WARN_ON_ONCE(ret)) {
2581 pr_warning("error enabling system %s\n",
2582 system->name);
2583 continue;
2584 }
2585
2586 event_test_stuff();
2587
2588 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0);
2589 if (WARN_ON_ONCE(ret)) {
2590 pr_warning("error disabling system %s\n",
2591 system->name);
2592 continue;
2593 }
2594
2595 pr_cont("OK\n");
2596 }
2597
2598 /* Test with all events enabled */
2599
2600 pr_info("Running tests on all trace events:\n");
2601 pr_info("Testing all events: ");
2602
2603 ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1);
2604 if (WARN_ON_ONCE(ret)) {
2605 pr_warning("error enabling all events\n");
2606 return;
2607 }
2608
2609 event_test_stuff();
2610
2611 /* reset sysname */
2612 ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
2613 if (WARN_ON_ONCE(ret)) {
2614 pr_warning("error disabling all events\n");
2615 return;
2616 }
2617
2618 pr_cont("OK\n");
2619 }
2620
2621 #ifdef CONFIG_FUNCTION_TRACER
2622
2623 static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
2624
2625 static void
2626 function_test_events_call(unsigned long ip, unsigned long parent_ip,
2627 struct ftrace_ops *op, struct pt_regs *pt_regs)
2628 {
2629 struct ring_buffer_event *event;
2630 struct ring_buffer *buffer;
2631 struct ftrace_entry *entry;
2632 unsigned long flags;
2633 long disabled;
2634 int cpu;
2635 int pc;
2636
2637 pc = preempt_count();
2638 preempt_disable_notrace();
2639 cpu = raw_smp_processor_id();
2640 disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
2641
2642 if (disabled != 1)
2643 goto out;
2644
2645 local_save_flags(flags);
2646
2647 event = trace_current_buffer_lock_reserve(&buffer,
2648 TRACE_FN, sizeof(*entry),
2649 flags, pc);
2650 if (!event)
2651 goto out;
2652 entry = ring_buffer_event_data(event);
2653 entry->ip = ip;
2654 entry->parent_ip = parent_ip;
2655
2656 trace_buffer_unlock_commit(buffer, event, flags, pc);
2657
2658 out:
2659 atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
2660 preempt_enable_notrace();
2661 }
2662
2663 static struct ftrace_ops trace_ops __initdata =
2664 {
2665 .func = function_test_events_call,
2666 .flags = FTRACE_OPS_FL_RECURSION_SAFE,
2667 };
2668
2669 static __init void event_trace_self_test_with_function(void)
2670 {
2671 int ret;
2672 ret = register_ftrace_function(&trace_ops);
2673 if (WARN_ON(ret < 0)) {
2674 pr_info("Failed to enable function tracer for event tests\n");
2675 return;
2676 }
2677 pr_info("Running tests again, along with the function tracer\n");
2678 event_trace_self_tests();
2679 unregister_ftrace_function(&trace_ops);
2680 }
2681 #else
2682 static __init void event_trace_self_test_with_function(void)
2683 {
2684 }
2685 #endif
2686
2687 static __init int event_trace_self_tests_init(void)
2688 {
2689 if (!tracing_selftest_disabled) {
2690 event_trace_self_tests();
2691 event_trace_self_test_with_function();
2692 }
2693
2694 return 0;
2695 }
2696
2697 late_initcall(event_trace_self_tests_init);
2698
2699 #endif