?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
IPVS: fix netns if reading ip_vs_* procfs entries
[linux-2.6/linux-mips.git] / kernel / trace / trace_events.c
blobe88f74fe1d4ce02fedf3e6052feaf2cbdd2017da
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 LIST_HEAD(ftrace_common_fields);
38
39 struct list_head *
40 trace_get_fields(struct ftrace_event_call *event_call)
41 {
42 if (!event_call->class->get_fields)
43 return &event_call->class->fields;
44 return event_call->class->get_fields(event_call);
45 }
46
47 static int __trace_define_field(struct list_head *head, const char *type,
48 const char *name, int offset, int size,
49 int is_signed, int filter_type)
50 {
51 struct ftrace_event_field *field;
52
53 field = kzalloc(sizeof(*field), GFP_KERNEL);
54 if (!field)
55 goto err;
56
57 field->name = kstrdup(name, GFP_KERNEL);
58 if (!field->name)
59 goto err;
60
61 field->type = kstrdup(type, GFP_KERNEL);
62 if (!field->type)
63 goto err;
64
65 if (filter_type == FILTER_OTHER)
66 field->filter_type = filter_assign_type(type);
67 else
68 field->filter_type = filter_type;
69
70 field->offset = offset;
71 field->size = size;
72 field->is_signed = is_signed;
73
74 list_add(&field->link, head);
75
76 return 0;
77
78 err:
79 if (field)
80 kfree(field->name);
81 kfree(field);
82
83 return -ENOMEM;
84 }
85
86 int trace_define_field(struct ftrace_event_call *call, const char *type,
87 const char *name, int offset, int size, int is_signed,
88 int filter_type)
89 {
90 struct list_head *head;
91
92 if (WARN_ON(!call->class))
93 return 0;
94
95 head = trace_get_fields(call);
96 return __trace_define_field(head, type, name, offset, size,
97 is_signed, filter_type);
98 }
99 EXPORT_SYMBOL_GPL(trace_define_field);
100
101 #define __common_field(type, item) \
102 ret = __trace_define_field(&ftrace_common_fields, #type, \
103 "common_" #item, \
104 offsetof(typeof(ent), item), \
105 sizeof(ent.item), \
106 is_signed_type(type), FILTER_OTHER); \
107 if (ret) \
108 return ret;
109
110 static int trace_define_common_fields(void)
111 {
112 int ret;
113 struct trace_entry ent;
114
115 __common_field(unsigned short, type);
116 __common_field(unsigned char, flags);
117 __common_field(unsigned char, preempt_count);
118 __common_field(int, pid);
119
120 return ret;
121 }
122
123 void trace_destroy_fields(struct ftrace_event_call *call)
124 {
125 struct ftrace_event_field *field, *next;
126 struct list_head *head;
127
128 head = trace_get_fields(call);
129 list_for_each_entry_safe(field, next, head, link) {
130 list_del(&field->link);
131 kfree(field->type);
132 kfree(field->name);
133 kfree(field);
134 }
135 }
136
137 int trace_event_raw_init(struct ftrace_event_call *call)
138 {
139 int id;
140
141 id = register_ftrace_event(&call->event);
142 if (!id)
143 return -ENODEV;
144
145 return 0;
146 }
147 EXPORT_SYMBOL_GPL(trace_event_raw_init);
148
149 int ftrace_event_reg(struct ftrace_event_call *call, enum trace_reg type)
150 {
151 switch (type) {
152 case TRACE_REG_REGISTER:
153 return tracepoint_probe_register(call->name,
154 call->class->probe,
155 call);
156 case TRACE_REG_UNREGISTER:
157 tracepoint_probe_unregister(call->name,
158 call->class->probe,
159 call);
160 return 0;
161
162 #ifdef CONFIG_PERF_EVENTS
163 case TRACE_REG_PERF_REGISTER:
164 return tracepoint_probe_register(call->name,
165 call->class->perf_probe,
166 call);
167 case TRACE_REG_PERF_UNREGISTER:
168 tracepoint_probe_unregister(call->name,
169 call->class->perf_probe,
170 call);
171 return 0;
172 #endif
173 }
174 return 0;
175 }
176 EXPORT_SYMBOL_GPL(ftrace_event_reg);
177
178 void trace_event_enable_cmd_record(bool enable)
179 {
180 struct ftrace_event_call *call;
181
182 mutex_lock(&event_mutex);
183 list_for_each_entry(call, &ftrace_events, list) {
184 if (!(call->flags & TRACE_EVENT_FL_ENABLED))
185 continue;
186
187 if (enable) {
188 tracing_start_cmdline_record();
189 call->flags |= TRACE_EVENT_FL_RECORDED_CMD;
190 } else {
191 tracing_stop_cmdline_record();
192 call->flags &= ~TRACE_EVENT_FL_RECORDED_CMD;
193 }
194 }
195 mutex_unlock(&event_mutex);
196 }
197
198 static int ftrace_event_enable_disable(struct ftrace_event_call *call,
199 int enable)
200 {
201 int ret = 0;
202
203 switch (enable) {
204 case 0:
205 if (call->flags & TRACE_EVENT_FL_ENABLED) {
206 call->flags &= ~TRACE_EVENT_FL_ENABLED;
207 if (call->flags & TRACE_EVENT_FL_RECORDED_CMD) {
208 tracing_stop_cmdline_record();
209 call->flags &= ~TRACE_EVENT_FL_RECORDED_CMD;
210 }
211 call->class->reg(call, TRACE_REG_UNREGISTER);
212 }
213 break;
214 case 1:
215 if (!(call->flags & TRACE_EVENT_FL_ENABLED)) {
216 if (trace_flags & TRACE_ITER_RECORD_CMD) {
217 tracing_start_cmdline_record();
218 call->flags |= TRACE_EVENT_FL_RECORDED_CMD;
219 }
220 ret = call->class->reg(call, TRACE_REG_REGISTER);
221 if (ret) {
222 tracing_stop_cmdline_record();
223 pr_info("event trace: Could not enable event "
224 "%s\n", call->name);
225 break;
226 }
227 call->flags |= TRACE_EVENT_FL_ENABLED;
228 }
229 break;
230 }
231
232 return ret;
233 }
234
235 static void ftrace_clear_events(void)
236 {
237 struct ftrace_event_call *call;
238
239 mutex_lock(&event_mutex);
240 list_for_each_entry(call, &ftrace_events, list) {
241 ftrace_event_enable_disable(call, 0);
242 }
243 mutex_unlock(&event_mutex);
244 }
245
246 /*
247 * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
248 */
249 static int __ftrace_set_clr_event(const char *match, const char *sub,
250 const char *event, int set)
251 {
252 struct ftrace_event_call *call;
253 int ret = -EINVAL;
254
255 mutex_lock(&event_mutex);
256 list_for_each_entry(call, &ftrace_events, list) {
257
258 if (!call->name || !call->class || !call->class->reg)
259 continue;
260
261 if (match &&
262 strcmp(match, call->name) != 0 &&
263 strcmp(match, call->class->system) != 0)
264 continue;
265
266 if (sub && strcmp(sub, call->class->system) != 0)
267 continue;
268
269 if (event && strcmp(event, call->name) != 0)
270 continue;
271
272 ftrace_event_enable_disable(call, set);
273
274 ret = 0;
275 }
276 mutex_unlock(&event_mutex);
277
278 return ret;
279 }
280
281 static int ftrace_set_clr_event(char *buf, int set)
282 {
283 char *event = NULL, *sub = NULL, *match;
284
285 /*
286 * The buf format can be <subsystem>:<event-name>
287 * *:<event-name> means any event by that name.
288 * :<event-name> is the same.
289 *
290 * <subsystem>:* means all events in that subsystem
291 * <subsystem>: means the same.
292 *
293 * <name> (no ':') means all events in a subsystem with
294 * the name <name> or any event that matches <name>
295 */
296
297 match = strsep(&buf, ":");
298 if (buf) {
299 sub = match;
300 event = buf;
301 match = NULL;
302
303 if (!strlen(sub) || strcmp(sub, "*") == 0)
304 sub = NULL;
305 if (!strlen(event) || strcmp(event, "*") == 0)
306 event = NULL;
307 }
308
309 return __ftrace_set_clr_event(match, sub, event, set);
310 }
311
312 /**
313 * trace_set_clr_event - enable or disable an event
314 * @system: system name to match (NULL for any system)
315 * @event: event name to match (NULL for all events, within system)
316 * @set: 1 to enable, 0 to disable
317 *
318 * This is a way for other parts of the kernel to enable or disable
319 * event recording.
320 *
321 * Returns 0 on success, -EINVAL if the parameters do not match any
322 * registered events.
323 */
324 int trace_set_clr_event(const char *system, const char *event, int set)
325 {
326 return __ftrace_set_clr_event(NULL, system, event, set);
327 }
328 EXPORT_SYMBOL_GPL(trace_set_clr_event);
329
330 /* 128 should be much more than enough */
331 #define EVENT_BUF_SIZE 127
332
333 static ssize_t
334 ftrace_event_write(struct file *file, const char __user *ubuf,
335 size_t cnt, loff_t *ppos)
336 {
337 struct trace_parser parser;
338 ssize_t read, ret;
339
340 if (!cnt)
341 return 0;
342
343 ret = tracing_update_buffers();
344 if (ret < 0)
345 return ret;
346
347 if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
348 return -ENOMEM;
349
350 read = trace_get_user(&parser, ubuf, cnt, ppos);
351
352 if (read >= 0 && trace_parser_loaded((&parser))) {
353 int set = 1;
354
355 if (*parser.buffer == '!')
356 set = 0;
357
358 parser.buffer[parser.idx] = 0;
359
360 ret = ftrace_set_clr_event(parser.buffer + !set, set);
361 if (ret)
362 goto out_put;
363 }
364
365 ret = read;
366
367 out_put:
368 trace_parser_put(&parser);
369
370 return ret;
371 }
372
373 static void *
374 t_next(struct seq_file *m, void *v, loff_t *pos)
375 {
376 struct ftrace_event_call *call = v;
377
378 (*pos)++;
379
380 list_for_each_entry_continue(call, &ftrace_events, list) {
381 /*
382 * The ftrace subsystem is for showing formats only.
383 * They can not be enabled or disabled via the event files.
384 */
385 if (call->class && call->class->reg)
386 return call;
387 }
388
389 return NULL;
390 }
391
392 static void *t_start(struct seq_file *m, loff_t *pos)
393 {
394 struct ftrace_event_call *call;
395 loff_t l;
396
397 mutex_lock(&event_mutex);
398
399 call = list_entry(&ftrace_events, struct ftrace_event_call, list);
400 for (l = 0; l <= *pos; ) {
401 call = t_next(m, call, &l);
402 if (!call)
403 break;
404 }
405 return call;
406 }
407
408 static void *
409 s_next(struct seq_file *m, void *v, loff_t *pos)
410 {
411 struct ftrace_event_call *call = v;
412
413 (*pos)++;
414
415 list_for_each_entry_continue(call, &ftrace_events, list) {
416 if (call->flags & TRACE_EVENT_FL_ENABLED)
417 return call;
418 }
419
420 return NULL;
421 }
422
423 static void *s_start(struct seq_file *m, loff_t *pos)
424 {
425 struct ftrace_event_call *call;
426 loff_t l;
427
428 mutex_lock(&event_mutex);
429
430 call = list_entry(&ftrace_events, struct ftrace_event_call, list);
431 for (l = 0; l <= *pos; ) {
432 call = s_next(m, call, &l);
433 if (!call)
434 break;
435 }
436 return call;
437 }
438
439 static int t_show(struct seq_file *m, void *v)
440 {
441 struct ftrace_event_call *call = v;
442
443 if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
444 seq_printf(m, "%s:", call->class->system);
445 seq_printf(m, "%s\n", call->name);
446
447 return 0;
448 }
449
450 static void t_stop(struct seq_file *m, void *p)
451 {
452 mutex_unlock(&event_mutex);
453 }
454
455 static int
456 ftrace_event_seq_open(struct inode *inode, struct file *file)
457 {
458 const struct seq_operations *seq_ops;
459
460 if ((file->f_mode & FMODE_WRITE) &&
461 (file->f_flags & O_TRUNC))
462 ftrace_clear_events();
463
464 seq_ops = inode->i_private;
465 return seq_open(file, seq_ops);
466 }
467
468 static ssize_t
469 event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
470 loff_t *ppos)
471 {
472 struct ftrace_event_call *call = filp->private_data;
473 char *buf;
474
475 if (call->flags & TRACE_EVENT_FL_ENABLED)
476 buf = "1\n";
477 else
478 buf = "0\n";
479
480 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
481 }
482
483 static ssize_t
484 event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
485 loff_t *ppos)
486 {
487 struct ftrace_event_call *call = filp->private_data;
488 char buf[64];
489 unsigned long val;
490 int ret;
491
492 if (cnt >= sizeof(buf))
493 return -EINVAL;
494
495 if (copy_from_user(&buf, ubuf, cnt))
496 return -EFAULT;
497
498 buf[cnt] = 0;
499
500 ret = strict_strtoul(buf, 10, &val);
501 if (ret < 0)
502 return ret;
503
504 ret = tracing_update_buffers();
505 if (ret < 0)
506 return ret;
507
508 switch (val) {
509 case 0:
510 case 1:
511 mutex_lock(&event_mutex);
512 ret = ftrace_event_enable_disable(call, val);
513 mutex_unlock(&event_mutex);
514 break;
515
516 default:
517 return -EINVAL;
518 }
519
520 *ppos += cnt;
521
522 return ret ? ret : cnt;
523 }
524
525 static ssize_t
526 system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
527 loff_t *ppos)
528 {
529 const char set_to_char[4] = { '?', '0', '1', 'X' };
530 const char *system = filp->private_data;
531 struct ftrace_event_call *call;
532 char buf[2];
533 int set = 0;
534 int ret;
535
536 mutex_lock(&event_mutex);
537 list_for_each_entry(call, &ftrace_events, list) {
538 if (!call->name || !call->class || !call->class->reg)
539 continue;
540
541 if (system && strcmp(call->class->system, system) != 0)
542 continue;
543
544 /*
545 * We need to find out if all the events are set
546 * or if all events or cleared, or if we have
547 * a mixture.
548 */
549 set |= (1 << !!(call->flags & TRACE_EVENT_FL_ENABLED));
550
551 /*
552 * If we have a mixture, no need to look further.
553 */
554 if (set == 3)
555 break;
556 }
557 mutex_unlock(&event_mutex);
558
559 buf[0] = set_to_char[set];
560 buf[1] = '\n';
561
562 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
563
564 return ret;
565 }
566
567 static ssize_t
568 system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
569 loff_t *ppos)
570 {
571 const char *system = filp->private_data;
572 unsigned long val;
573 char buf[64];
574 ssize_t ret;
575
576 if (cnt >= sizeof(buf))
577 return -EINVAL;
578
579 if (copy_from_user(&buf, ubuf, cnt))
580 return -EFAULT;
581
582 buf[cnt] = 0;
583
584 ret = strict_strtoul(buf, 10, &val);
585 if (ret < 0)
586 return ret;
587
588 ret = tracing_update_buffers();
589 if (ret < 0)
590 return ret;
591
592 if (val != 0 && val != 1)
593 return -EINVAL;
594
595 ret = __ftrace_set_clr_event(NULL, system, NULL, val);
596 if (ret)
597 goto out;
598
599 ret = cnt;
600
601 out:
602 *ppos += cnt;
603
604 return ret;
605 }
606
607 enum {
608 FORMAT_HEADER = 1,
609 FORMAT_FIELD_SEPERATOR = 2,
610 FORMAT_PRINTFMT = 3,
611 };
612
613 static void *f_next(struct seq_file *m, void *v, loff_t *pos)
614 {
615 struct ftrace_event_call *call = m->private;
616 struct ftrace_event_field *field;
617 struct list_head *common_head = &ftrace_common_fields;
618 struct list_head *head = trace_get_fields(call);
619
620 (*pos)++;
621
622 switch ((unsigned long)v) {
623 case FORMAT_HEADER:
624 if (unlikely(list_empty(common_head)))
625 return NULL;
626
627 field = list_entry(common_head->prev,
628 struct ftrace_event_field, link);
629 return field;
630
631 case FORMAT_FIELD_SEPERATOR:
632 if (unlikely(list_empty(head)))
633 return NULL;
634
635 field = list_entry(head->prev, struct ftrace_event_field, link);
636 return field;
637
638 case FORMAT_PRINTFMT:
639 /* all done */
640 return NULL;
641 }
642
643 field = v;
644 if (field->link.prev == common_head)
645 return (void *)FORMAT_FIELD_SEPERATOR;
646 else if (field->link.prev == head)
647 return (void *)FORMAT_PRINTFMT;
648
649 field = list_entry(field->link.prev, struct ftrace_event_field, link);
650
651 return field;
652 }
653
654 static void *f_start(struct seq_file *m, loff_t *pos)
655 {
656 loff_t l = 0;
657 void *p;
658
659 /* Start by showing the header */
660 if (!*pos)
661 return (void *)FORMAT_HEADER;
662
663 p = (void *)FORMAT_HEADER;
664 do {
665 p = f_next(m, p, &l);
666 } while (p && l < *pos);
667
668 return p;
669 }
670
671 static int f_show(struct seq_file *m, void *v)
672 {
673 struct ftrace_event_call *call = m->private;
674 struct ftrace_event_field *field;
675 const char *array_descriptor;
676
677 switch ((unsigned long)v) {
678 case FORMAT_HEADER:
679 seq_printf(m, "name: %s\n", call->name);
680 seq_printf(m, "ID: %d\n", call->event.type);
681 seq_printf(m, "format:\n");
682 return 0;
683
684 case FORMAT_FIELD_SEPERATOR:
685 seq_putc(m, '\n');
686 return 0;
687
688 case FORMAT_PRINTFMT:
689 seq_printf(m, "\nprint fmt: %s\n",
690 call->print_fmt);
691 return 0;
692 }
693
694 field = v;
695
696 /*
697 * Smartly shows the array type(except dynamic array).
698 * Normal:
699 * field:TYPE VAR
700 * If TYPE := TYPE[LEN], it is shown:
701 * field:TYPE VAR[LEN]
702 */
703 array_descriptor = strchr(field->type, '[');
704
705 if (!strncmp(field->type, "__data_loc", 10))
706 array_descriptor = NULL;
707
708 if (!array_descriptor)
709 seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
710 field->type, field->name, field->offset,
711 field->size, !!field->is_signed);
712 else
713 seq_printf(m, "\tfield:%.*s %s%s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
714 (int)(array_descriptor - field->type),
715 field->type, field->name,
716 array_descriptor, field->offset,
717 field->size, !!field->is_signed);
718
719 return 0;
720 }
721
722 static void f_stop(struct seq_file *m, void *p)
723 {
724 }
725
726 static const struct seq_operations trace_format_seq_ops = {
727 .start = f_start,
728 .next = f_next,
729 .stop = f_stop,
730 .show = f_show,
731 };
732
733 static int trace_format_open(struct inode *inode, struct file *file)
734 {
735 struct ftrace_event_call *call = inode->i_private;
736 struct seq_file *m;
737 int ret;
738
739 ret = seq_open(file, &trace_format_seq_ops);
740 if (ret < 0)
741 return ret;
742
743 m = file->private_data;
744 m->private = call;
745
746 return 0;
747 }
748
749 static ssize_t
750 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
751 {
752 struct ftrace_event_call *call = filp->private_data;
753 struct trace_seq *s;
754 int r;
755
756 if (*ppos)
757 return 0;
758
759 s = kmalloc(sizeof(*s), GFP_KERNEL);
760 if (!s)
761 return -ENOMEM;
762
763 trace_seq_init(s);
764 trace_seq_printf(s, "%d\n", call->event.type);
765
766 r = simple_read_from_buffer(ubuf, cnt, ppos,
767 s->buffer, s->len);
768 kfree(s);
769 return r;
770 }
771
772 static ssize_t
773 event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
774 loff_t *ppos)
775 {
776 struct ftrace_event_call *call = filp->private_data;
777 struct trace_seq *s;
778 int r;
779
780 if (*ppos)
781 return 0;
782
783 s = kmalloc(sizeof(*s), GFP_KERNEL);
784 if (!s)
785 return -ENOMEM;
786
787 trace_seq_init(s);
788
789 print_event_filter(call, s);
790 r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
791
792 kfree(s);
793
794 return r;
795 }
796
797 static ssize_t
798 event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
799 loff_t *ppos)
800 {
801 struct ftrace_event_call *call = filp->private_data;
802 char *buf;
803 int err;
804
805 if (cnt >= PAGE_SIZE)
806 return -EINVAL;
807
808 buf = (char *)__get_free_page(GFP_TEMPORARY);
809 if (!buf)
810 return -ENOMEM;
811
812 if (copy_from_user(buf, ubuf, cnt)) {
813 free_page((unsigned long) buf);
814 return -EFAULT;
815 }
816 buf[cnt] = '\0';
817
818 err = apply_event_filter(call, buf);
819 free_page((unsigned long) buf);
820 if (err < 0)
821 return err;
822
823 *ppos += cnt;
824
825 return cnt;
826 }
827
828 static ssize_t
829 subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
830 loff_t *ppos)
831 {
832 struct event_subsystem *system = filp->private_data;
833 struct trace_seq *s;
834 int r;
835
836 if (*ppos)
837 return 0;
838
839 s = kmalloc(sizeof(*s), GFP_KERNEL);
840 if (!s)
841 return -ENOMEM;
842
843 trace_seq_init(s);
844
845 print_subsystem_event_filter(system, s);
846 r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
847
848 kfree(s);
849
850 return r;
851 }
852
853 static ssize_t
854 subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
855 loff_t *ppos)
856 {
857 struct event_subsystem *system = filp->private_data;
858 char *buf;
859 int err;
860
861 if (cnt >= PAGE_SIZE)
862 return -EINVAL;
863
864 buf = (char *)__get_free_page(GFP_TEMPORARY);
865 if (!buf)
866 return -ENOMEM;
867
868 if (copy_from_user(buf, ubuf, cnt)) {
869 free_page((unsigned long) buf);
870 return -EFAULT;
871 }
872 buf[cnt] = '\0';
873
874 err = apply_subsystem_event_filter(system, buf);
875 free_page((unsigned long) buf);
876 if (err < 0)
877 return err;
878
879 *ppos += cnt;
880
881 return cnt;
882 }
883
884 static ssize_t
885 show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
886 {
887 int (*func)(struct trace_seq *s) = filp->private_data;
888 struct trace_seq *s;
889 int r;
890
891 if (*ppos)
892 return 0;
893
894 s = kmalloc(sizeof(*s), GFP_KERNEL);
895 if (!s)
896 return -ENOMEM;
897
898 trace_seq_init(s);
899
900 func(s);
901 r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
902
903 kfree(s);
904
905 return r;
906 }
907
908 static const struct seq_operations show_event_seq_ops = {
909 .start = t_start,
910 .next = t_next,
911 .show = t_show,
912 .stop = t_stop,
913 };
914
915 static const struct seq_operations show_set_event_seq_ops = {
916 .start = s_start,
917 .next = s_next,
918 .show = t_show,
919 .stop = t_stop,
920 };
921
922 static const struct file_operations ftrace_avail_fops = {
923 .open = ftrace_event_seq_open,
924 .read = seq_read,
925 .llseek = seq_lseek,
926 .release = seq_release,
927 };
928
929 static const struct file_operations ftrace_set_event_fops = {
930 .open = ftrace_event_seq_open,
931 .read = seq_read,
932 .write = ftrace_event_write,
933 .llseek = seq_lseek,
934 .release = seq_release,
935 };
936
937 static const struct file_operations ftrace_enable_fops = {
938 .open = tracing_open_generic,
939 .read = event_enable_read,
940 .write = event_enable_write,
941 .llseek = default_llseek,
942 };
943
944 static const struct file_operations ftrace_event_format_fops = {
945 .open = trace_format_open,
946 .read = seq_read,
947 .llseek = seq_lseek,
948 .release = seq_release,
949 };
950
951 static const struct file_operations ftrace_event_id_fops = {
952 .open = tracing_open_generic,
953 .read = event_id_read,
954 .llseek = default_llseek,
955 };
956
957 static const struct file_operations ftrace_event_filter_fops = {
958 .open = tracing_open_generic,
959 .read = event_filter_read,
960 .write = event_filter_write,
961 .llseek = default_llseek,
962 };
963
964 static const struct file_operations ftrace_subsystem_filter_fops = {
965 .open = tracing_open_generic,
966 .read = subsystem_filter_read,
967 .write = subsystem_filter_write,
968 .llseek = default_llseek,
969 };
970
971 static const struct file_operations ftrace_system_enable_fops = {
972 .open = tracing_open_generic,
973 .read = system_enable_read,
974 .write = system_enable_write,
975 .llseek = default_llseek,
976 };
977
978 static const struct file_operations ftrace_show_header_fops = {
979 .open = tracing_open_generic,
980 .read = show_header,
981 .llseek = default_llseek,
982 };
983
984 static struct dentry *event_trace_events_dir(void)
985 {
986 static struct dentry *d_tracer;
987 static struct dentry *d_events;
988
989 if (d_events)
990 return d_events;
991
992 d_tracer = tracing_init_dentry();
993 if (!d_tracer)
994 return NULL;
995
996 d_events = debugfs_create_dir("events", d_tracer);
997 if (!d_events)
998 pr_warning("Could not create debugfs "
999 "'events' directory\n");
1000
1001 return d_events;
1002 }
1003
1004 static LIST_HEAD(event_subsystems);
1005
1006 static struct dentry *
1007 event_subsystem_dir(const char *name, struct dentry *d_events)
1008 {
1009 struct event_subsystem *system;
1010 struct dentry *entry;
1011
1012 /* First see if we did not already create this dir */
1013 list_for_each_entry(system, &event_subsystems, list) {
1014 if (strcmp(system->name, name) == 0) {
1015 system->nr_events++;
1016 return system->entry;
1017 }
1018 }
1019
1020 /* need to create new entry */
1021 system = kmalloc(sizeof(*system), GFP_KERNEL);
1022 if (!system) {
1023 pr_warning("No memory to create event subsystem %s\n",
1024 name);
1025 return d_events;
1026 }
1027
1028 system->entry = debugfs_create_dir(name, d_events);
1029 if (!system->entry) {
1030 pr_warning("Could not create event subsystem %s\n",
1031 name);
1032 kfree(system);
1033 return d_events;
1034 }
1035
1036 system->nr_events = 1;
1037 system->name = kstrdup(name, GFP_KERNEL);
1038 if (!system->name) {
1039 debugfs_remove(system->entry);
1040 kfree(system);
1041 return d_events;
1042 }
1043
1044 list_add(&system->list, &event_subsystems);
1045
1046 system->filter = NULL;
1047
1048 system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
1049 if (!system->filter) {
1050 pr_warning("Could not allocate filter for subsystem "
1051 "'%s'\n", name);
1052 return system->entry;
1053 }
1054
1055 entry = debugfs_create_file("filter", 0644, system->entry, system,
1056 &ftrace_subsystem_filter_fops);
1057 if (!entry) {
1058 kfree(system->filter);
1059 system->filter = NULL;
1060 pr_warning("Could not create debugfs "
1061 "'%s/filter' entry\n", name);
1062 }
1063
1064 trace_create_file("enable", 0644, system->entry,
1065 (void *)system->name,
1066 &ftrace_system_enable_fops);
1067
1068 return system->entry;
1069 }
1070
1071 static int
1072 event_create_dir(struct ftrace_event_call *call, struct dentry *d_events,
1073 const struct file_operations *id,
1074 const struct file_operations *enable,
1075 const struct file_operations *filter,
1076 const struct file_operations *format)
1077 {
1078 struct list_head *head;
1079 int ret;
1080
1081 /*
1082 * If the trace point header did not define TRACE_SYSTEM
1083 * then the system would be called "TRACE_SYSTEM".
1084 */
1085 if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
1086 d_events = event_subsystem_dir(call->class->system, d_events);
1087
1088 call->dir = debugfs_create_dir(call->name, d_events);
1089 if (!call->dir) {
1090 pr_warning("Could not create debugfs "
1091 "'%s' directory\n", call->name);
1092 return -1;
1093 }
1094
1095 if (call->class->reg)
1096 trace_create_file("enable", 0644, call->dir, call,
1097 enable);
1098
1099 #ifdef CONFIG_PERF_EVENTS
1100 if (call->event.type && call->class->reg)
1101 trace_create_file("id", 0444, call->dir, call,
1102 id);
1103 #endif
1104
1105 /*
1106 * Other events may have the same class. Only update
1107 * the fields if they are not already defined.
1108 */
1109 head = trace_get_fields(call);
1110 if (list_empty(head)) {
1111 ret = call->class->define_fields(call);
1112 if (ret < 0) {
1113 pr_warning("Could not initialize trace point"
1114 " events/%s\n", call->name);
1115 return ret;
1116 }
1117 }
1118 trace_create_file("filter", 0644, call->dir, call,
1119 filter);
1120
1121 trace_create_file("format", 0444, call->dir, call,
1122 format);
1123
1124 return 0;
1125 }
1126
1127 static int
1128 __trace_add_event_call(struct ftrace_event_call *call, struct module *mod,
1129 const struct file_operations *id,
1130 const struct file_operations *enable,
1131 const struct file_operations *filter,
1132 const struct file_operations *format)
1133 {
1134 struct dentry *d_events;
1135 int ret;
1136
1137 /* The linker may leave blanks */
1138 if (!call->name)
1139 return -EINVAL;
1140
1141 if (call->class->raw_init) {
1142 ret = call->class->raw_init(call);
1143 if (ret < 0) {
1144 if (ret != -ENOSYS)
1145 pr_warning("Could not initialize trace events/%s\n",
1146 call->name);
1147 return ret;
1148 }
1149 }
1150
1151 d_events = event_trace_events_dir();
1152 if (!d_events)
1153 return -ENOENT;
1154
1155 ret = event_create_dir(call, d_events, id, enable, filter, format);
1156 if (!ret)
1157 list_add(&call->list, &ftrace_events);
1158 call->mod = mod;
1159
1160 return ret;
1161 }
1162
1163 /* Add an additional event_call dynamically */
1164 int trace_add_event_call(struct ftrace_event_call *call)
1165 {
1166 int ret;
1167 mutex_lock(&event_mutex);
1168 ret = __trace_add_event_call(call, NULL, &ftrace_event_id_fops,
1169 &ftrace_enable_fops,
1170 &ftrace_event_filter_fops,
1171 &ftrace_event_format_fops);
1172 mutex_unlock(&event_mutex);
1173 return ret;
1174 }
1175
1176 static void remove_subsystem_dir(const char *name)
1177 {
1178 struct event_subsystem *system;
1179
1180 if (strcmp(name, TRACE_SYSTEM) == 0)
1181 return;
1182
1183 list_for_each_entry(system, &event_subsystems, list) {
1184 if (strcmp(system->name, name) == 0) {
1185 if (!--system->nr_events) {
1186 struct event_filter *filter = system->filter;
1187
1188 debugfs_remove_recursive(system->entry);
1189 list_del(&system->list);
1190 if (filter) {
1191 kfree(filter->filter_string);
1192 kfree(filter);
1193 }
1194 kfree(system->name);
1195 kfree(system);
1196 }
1197 break;
1198 }
1199 }
1200 }
1201
1202 /*
1203 * Must be called under locking both of event_mutex and trace_event_mutex.
1204 */
1205 static void __trace_remove_event_call(struct ftrace_event_call *call)
1206 {
1207 ftrace_event_enable_disable(call, 0);
1208 if (call->event.funcs)
1209 __unregister_ftrace_event(&call->event);
1210 debugfs_remove_recursive(call->dir);
1211 list_del(&call->list);
1212 trace_destroy_fields(call);
1213 destroy_preds(call);
1214 remove_subsystem_dir(call->class->system);
1215 }
1216
1217 /* Remove an event_call */
1218 void trace_remove_event_call(struct ftrace_event_call *call)
1219 {
1220 mutex_lock(&event_mutex);
1221 down_write(&trace_event_mutex);
1222 __trace_remove_event_call(call);
1223 up_write(&trace_event_mutex);
1224 mutex_unlock(&event_mutex);
1225 }
1226
1227 #define for_each_event(event, start, end) \
1228 for (event = start; \
1229 (unsigned long)event < (unsigned long)end; \
1230 event++)
1231
1232 #ifdef CONFIG_MODULES
1233
1234 static LIST_HEAD(ftrace_module_file_list);
1235
1236 /*
1237 * Modules must own their file_operations to keep up with
1238 * reference counting.
1239 */
1240 struct ftrace_module_file_ops {
1241 struct list_head list;
1242 struct module *mod;
1243 struct file_operations id;
1244 struct file_operations enable;
1245 struct file_operations format;
1246 struct file_operations filter;
1247 };
1248
1249 static struct ftrace_module_file_ops *
1250 trace_create_file_ops(struct module *mod)
1251 {
1252 struct ftrace_module_file_ops *file_ops;
1253
1254 /*
1255 * This is a bit of a PITA. To allow for correct reference
1256 * counting, modules must "own" their file_operations.
1257 * To do this, we allocate the file operations that will be
1258 * used in the event directory.
1259 */
1260
1261 file_ops = kmalloc(sizeof(*file_ops), GFP_KERNEL);
1262 if (!file_ops)
1263 return NULL;
1264
1265 file_ops->mod = mod;
1266
1267 file_ops->id = ftrace_event_id_fops;
1268 file_ops->id.owner = mod;
1269
1270 file_ops->enable = ftrace_enable_fops;
1271 file_ops->enable.owner = mod;
1272
1273 file_ops->filter = ftrace_event_filter_fops;
1274 file_ops->filter.owner = mod;
1275
1276 file_ops->format = ftrace_event_format_fops;
1277 file_ops->format.owner = mod;
1278
1279 list_add(&file_ops->list, &ftrace_module_file_list);
1280
1281 return file_ops;
1282 }
1283
1284 static void trace_module_add_events(struct module *mod)
1285 {
1286 struct ftrace_module_file_ops *file_ops = NULL;
1287 struct ftrace_event_call **call, **start, **end;
1288
1289 start = mod->trace_events;
1290 end = mod->trace_events + mod->num_trace_events;
1291
1292 if (start == end)
1293 return;
1294
1295 file_ops = trace_create_file_ops(mod);
1296 if (!file_ops)
1297 return;
1298
1299 for_each_event(call, start, end) {
1300 __trace_add_event_call(*call, mod,
1301 &file_ops->id, &file_ops->enable,
1302 &file_ops->filter, &file_ops->format);
1303 }
1304 }
1305
1306 static void trace_module_remove_events(struct module *mod)
1307 {
1308 struct ftrace_module_file_ops *file_ops;
1309 struct ftrace_event_call *call, *p;
1310 bool found = false;
1311
1312 down_write(&trace_event_mutex);
1313 list_for_each_entry_safe(call, p, &ftrace_events, list) {
1314 if (call->mod == mod) {
1315 found = true;
1316 __trace_remove_event_call(call);
1317 }
1318 }
1319
1320 /* Now free the file_operations */
1321 list_for_each_entry(file_ops, &ftrace_module_file_list, list) {
1322 if (file_ops->mod == mod)
1323 break;
1324 }
1325 if (&file_ops->list != &ftrace_module_file_list) {
1326 list_del(&file_ops->list);
1327 kfree(file_ops);
1328 }
1329
1330 /*
1331 * It is safest to reset the ring buffer if the module being unloaded
1332 * registered any events.
1333 */
1334 if (found)
1335 tracing_reset_current_online_cpus();
1336 up_write(&trace_event_mutex);
1337 }
1338
1339 static int trace_module_notify(struct notifier_block *self,
1340 unsigned long val, void *data)
1341 {
1342 struct module *mod = data;
1343
1344 mutex_lock(&event_mutex);
1345 switch (val) {
1346 case MODULE_STATE_COMING:
1347 trace_module_add_events(mod);
1348 break;
1349 case MODULE_STATE_GOING:
1350 trace_module_remove_events(mod);
1351 break;
1352 }
1353 mutex_unlock(&event_mutex);
1354
1355 return 0;
1356 }
1357 #else
1358 static int trace_module_notify(struct notifier_block *self,
1359 unsigned long val, void *data)
1360 {
1361 return 0;
1362 }
1363 #endif /* CONFIG_MODULES */
1364
1365 static struct notifier_block trace_module_nb = {
1366 .notifier_call = trace_module_notify,
1367 .priority = 0,
1368 };
1369
1370 extern struct ftrace_event_call *__start_ftrace_events[];
1371 extern struct ftrace_event_call *__stop_ftrace_events[];
1372
1373 static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
1374
1375 static __init int setup_trace_event(char *str)
1376 {
1377 strlcpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
1378 ring_buffer_expanded = 1;
1379 tracing_selftest_disabled = 1;
1380
1381 return 1;
1382 }
1383 __setup("trace_event=", setup_trace_event);
1384
1385 static __init int event_trace_init(void)
1386 {
1387 struct ftrace_event_call **call;
1388 struct dentry *d_tracer;
1389 struct dentry *entry;
1390 struct dentry *d_events;
1391 int ret;
1392 char *buf = bootup_event_buf;
1393 char *token;
1394
1395 d_tracer = tracing_init_dentry();
1396 if (!d_tracer)
1397 return 0;
1398
1399 entry = debugfs_create_file("available_events", 0444, d_tracer,
1400 (void *)&show_event_seq_ops,
1401 &ftrace_avail_fops);
1402 if (!entry)
1403 pr_warning("Could not create debugfs "
1404 "'available_events' entry\n");
1405
1406 entry = debugfs_create_file("set_event", 0644, d_tracer,
1407 (void *)&show_set_event_seq_ops,
1408 &ftrace_set_event_fops);
1409 if (!entry)
1410 pr_warning("Could not create debugfs "
1411 "'set_event' entry\n");
1412
1413 d_events = event_trace_events_dir();
1414 if (!d_events)
1415 return 0;
1416
1417 /* ring buffer internal formats */
1418 trace_create_file("header_page", 0444, d_events,
1419 ring_buffer_print_page_header,
1420 &ftrace_show_header_fops);
1421
1422 trace_create_file("header_event", 0444, d_events,
1423 ring_buffer_print_entry_header,
1424 &ftrace_show_header_fops);
1425
1426 trace_create_file("enable", 0644, d_events,
1427 NULL, &ftrace_system_enable_fops);
1428
1429 if (trace_define_common_fields())
1430 pr_warning("tracing: Failed to allocate common fields");
1431
1432 for_each_event(call, __start_ftrace_events, __stop_ftrace_events) {
1433 __trace_add_event_call(*call, NULL, &ftrace_event_id_fops,
1434 &ftrace_enable_fops,
1435 &ftrace_event_filter_fops,
1436 &ftrace_event_format_fops);
1437 }
1438
1439 while (true) {
1440 token = strsep(&buf, ",");
1441
1442 if (!token)
1443 break;
1444 if (!*token)
1445 continue;
1446
1447 ret = ftrace_set_clr_event(token, 1);
1448 if (ret)
1449 pr_warning("Failed to enable trace event: %s\n", token);
1450 }
1451
1452 ret = register_module_notifier(&trace_module_nb);
1453 if (ret)
1454 pr_warning("Failed to register trace events module notifier\n");
1455
1456 return 0;
1457 }
1458 fs_initcall(event_trace_init);
1459
1460 #ifdef CONFIG_FTRACE_STARTUP_TEST
1461
1462 static DEFINE_SPINLOCK(test_spinlock);
1463 static DEFINE_SPINLOCK(test_spinlock_irq);
1464 static DEFINE_MUTEX(test_mutex);
1465
1466 static __init void test_work(struct work_struct *dummy)
1467 {
1468 spin_lock(&test_spinlock);
1469 spin_lock_irq(&test_spinlock_irq);
1470 udelay(1);
1471 spin_unlock_irq(&test_spinlock_irq);
1472 spin_unlock(&test_spinlock);
1473
1474 mutex_lock(&test_mutex);
1475 msleep(1);
1476 mutex_unlock(&test_mutex);
1477 }
1478
1479 static __init int event_test_thread(void *unused)
1480 {
1481 void *test_malloc;
1482
1483 test_malloc = kmalloc(1234, GFP_KERNEL);
1484 if (!test_malloc)
1485 pr_info("failed to kmalloc\n");
1486
1487 schedule_on_each_cpu(test_work);
1488
1489 kfree(test_malloc);
1490
1491 set_current_state(TASK_INTERRUPTIBLE);
1492 while (!kthread_should_stop())
1493 schedule();
1494
1495 return 0;
1496 }
1497
1498 /*
1499 * Do various things that may trigger events.
1500 */
1501 static __init void event_test_stuff(void)
1502 {
1503 struct task_struct *test_thread;
1504
1505 test_thread = kthread_run(event_test_thread, NULL, "test-events");
1506 msleep(1);
1507 kthread_stop(test_thread);
1508 }
1509
1510 /*
1511 * For every trace event defined, we will test each trace point separately,
1512 * and then by groups, and finally all trace points.
1513 */
1514 static __init void event_trace_self_tests(void)
1515 {
1516 struct ftrace_event_call *call;
1517 struct event_subsystem *system;
1518 int ret;
1519
1520 pr_info("Running tests on trace events:\n");
1521
1522 list_for_each_entry(call, &ftrace_events, list) {
1523
1524 /* Only test those that have a probe */
1525 if (!call->class || !call->class->probe)
1526 continue;
1527
1528 /*
1529 * Testing syscall events here is pretty useless, but
1530 * we still do it if configured. But this is time consuming.
1531 * What we really need is a user thread to perform the
1532 * syscalls as we test.
1533 */
1534 #ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
1535 if (call->class->system &&
1536 strcmp(call->class->system, "syscalls") == 0)
1537 continue;
1538 #endif
1539
1540 pr_info("Testing event %s: ", call->name);
1541
1542 /*
1543 * If an event is already enabled, someone is using
1544 * it and the self test should not be on.
1545 */
1546 if (call->flags & TRACE_EVENT_FL_ENABLED) {
1547 pr_warning("Enabled event during self test!\n");
1548 WARN_ON_ONCE(1);
1549 continue;
1550 }
1551
1552 ftrace_event_enable_disable(call, 1);
1553 event_test_stuff();
1554 ftrace_event_enable_disable(call, 0);
1555
1556 pr_cont("OK\n");
1557 }
1558
1559 /* Now test at the sub system level */
1560
1561 pr_info("Running tests on trace event systems:\n");
1562
1563 list_for_each_entry(system, &event_subsystems, list) {
1564
1565 /* the ftrace system is special, skip it */
1566 if (strcmp(system->name, "ftrace") == 0)
1567 continue;
1568
1569 pr_info("Testing event system %s: ", system->name);
1570
1571 ret = __ftrace_set_clr_event(NULL, system->name, NULL, 1);
1572 if (WARN_ON_ONCE(ret)) {
1573 pr_warning("error enabling system %s\n",
1574 system->name);
1575 continue;
1576 }
1577
1578 event_test_stuff();
1579
1580 ret = __ftrace_set_clr_event(NULL, system->name, NULL, 0);
1581 if (WARN_ON_ONCE(ret))
1582 pr_warning("error disabling system %s\n",
1583 system->name);
1584
1585 pr_cont("OK\n");
1586 }
1587
1588 /* Test with all events enabled */
1589
1590 pr_info("Running tests on all trace events:\n");
1591 pr_info("Testing all events: ");
1592
1593 ret = __ftrace_set_clr_event(NULL, NULL, NULL, 1);
1594 if (WARN_ON_ONCE(ret)) {
1595 pr_warning("error enabling all events\n");
1596 return;
1597 }
1598
1599 event_test_stuff();
1600
1601 /* reset sysname */
1602 ret = __ftrace_set_clr_event(NULL, NULL, NULL, 0);
1603 if (WARN_ON_ONCE(ret)) {
1604 pr_warning("error disabling all events\n");
1605 return;
1606 }
1607
1608 pr_cont("OK\n");
1609 }
1610
1611 #ifdef CONFIG_FUNCTION_TRACER
1612
1613 static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
1614
1615 static void
1616 function_test_events_call(unsigned long ip, unsigned long parent_ip)
1617 {
1618 struct ring_buffer_event *event;
1619 struct ring_buffer *buffer;
1620 struct ftrace_entry *entry;
1621 unsigned long flags;
1622 long disabled;
1623 int cpu;
1624 int pc;
1625
1626 pc = preempt_count();
1627 preempt_disable_notrace();
1628 cpu = raw_smp_processor_id();
1629 disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
1630
1631 if (disabled != 1)
1632 goto out;
1633
1634 local_save_flags(flags);
1635
1636 event = trace_current_buffer_lock_reserve(&buffer,
1637 TRACE_FN, sizeof(*entry),
1638 flags, pc);
1639 if (!event)
1640 goto out;
1641 entry = ring_buffer_event_data(event);
1642 entry->ip = ip;
1643 entry->parent_ip = parent_ip;
1644
1645 trace_nowake_buffer_unlock_commit(buffer, event, flags, pc);
1646
1647 out:
1648 atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
1649 preempt_enable_notrace();
1650 }
1651
1652 static struct ftrace_ops trace_ops __initdata =
1653 {
1654 .func = function_test_events_call,
1655 };
1656
1657 static __init void event_trace_self_test_with_function(void)
1658 {
1659 register_ftrace_function(&trace_ops);
1660 pr_info("Running tests again, along with the function tracer\n");
1661 event_trace_self_tests();
1662 unregister_ftrace_function(&trace_ops);
1663 }
1664 #else
1665 static __init void event_trace_self_test_with_function(void)
1666 {
1667 }
1668 #endif
1669
1670 static __init int event_trace_self_tests_init(void)
1671 {
1672 if (!tracing_selftest_disabled) {
1673 event_trace_self_tests();
1674 event_trace_self_test_with_function();
1675 }
1676
1677 return 0;
1678 }
1679
1680 late_initcall(event_trace_self_tests_init);
1681
1682 #endif
Mirror of the linux-mips.org kernel tree