Merge branch 'x86-setup-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git...
[pandora-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 LIST_HEAD(ftrace_events);
31 LIST_HEAD(ftrace_common_fields);
32
33 struct list_head *
34 trace_get_fields(struct ftrace_event_call *event_call)
35 {
36         if (!event_call->class->get_fields)
37                 return &event_call->class->fields;
38         return event_call->class->get_fields(event_call);
39 }
40
41 static int __trace_define_field(struct list_head *head, const char *type,
42                                 const char *name, int offset, int size,
43                                 int is_signed, int filter_type)
44 {
45         struct ftrace_event_field *field;
46
47         field = kzalloc(sizeof(*field), GFP_KERNEL);
48         if (!field)
49                 goto err;
50
51         field->name = kstrdup(name, GFP_KERNEL);
52         if (!field->name)
53                 goto err;
54
55         field->type = kstrdup(type, GFP_KERNEL);
56         if (!field->type)
57                 goto err;
58
59         if (filter_type == FILTER_OTHER)
60                 field->filter_type = filter_assign_type(type);
61         else
62                 field->filter_type = filter_type;
63
64         field->offset = offset;
65         field->size = size;
66         field->is_signed = is_signed;
67
68         list_add(&field->link, head);
69
70         return 0;
71
72 err:
73         if (field)
74                 kfree(field->name);
75         kfree(field);
76
77         return -ENOMEM;
78 }
79
80 int trace_define_field(struct ftrace_event_call *call, const char *type,
81                        const char *name, int offset, int size, int is_signed,
82                        int filter_type)
83 {
84         struct list_head *head;
85
86         if (WARN_ON(!call->class))
87                 return 0;
88
89         head = trace_get_fields(call);
90         return __trace_define_field(head, type, name, offset, size,
91                                     is_signed, filter_type);
92 }
93 EXPORT_SYMBOL_GPL(trace_define_field);
94
95 #define __common_field(type, item)                                      \
96         ret = __trace_define_field(&ftrace_common_fields, #type,        \
97                                    "common_" #item,                     \
98                                    offsetof(typeof(ent), item),         \
99                                    sizeof(ent.item),                    \
100                                    is_signed_type(type), FILTER_OTHER); \
101         if (ret)                                                        \
102                 return ret;
103
104 static int trace_define_common_fields(void)
105 {
106         int ret;
107         struct trace_entry ent;
108
109         __common_field(unsigned short, type);
110         __common_field(unsigned char, flags);
111         __common_field(unsigned char, preempt_count);
112         __common_field(int, pid);
113         __common_field(int, lock_depth);
114
115         return ret;
116 }
117
118 void trace_destroy_fields(struct ftrace_event_call *call)
119 {
120         struct ftrace_event_field *field, *next;
121         struct list_head *head;
122
123         head = trace_get_fields(call);
124         list_for_each_entry_safe(field, next, head, link) {
125                 list_del(&field->link);
126                 kfree(field->type);
127                 kfree(field->name);
128                 kfree(field);
129         }
130 }
131
132 int trace_event_raw_init(struct ftrace_event_call *call)
133 {
134         int id;
135
136         id = register_ftrace_event(&call->event);
137         if (!id)
138                 return -ENODEV;
139
140         return 0;
141 }
142 EXPORT_SYMBOL_GPL(trace_event_raw_init);
143
144 int ftrace_event_reg(struct ftrace_event_call *call, enum trace_reg type)
145 {
146         switch (type) {
147         case TRACE_REG_REGISTER:
148                 return tracepoint_probe_register(call->name,
149                                                  call->class->probe,
150                                                  call);
151         case TRACE_REG_UNREGISTER:
152                 tracepoint_probe_unregister(call->name,
153                                             call->class->probe,
154                                             call);
155                 return 0;
156
157 #ifdef CONFIG_PERF_EVENTS
158         case TRACE_REG_PERF_REGISTER:
159                 return tracepoint_probe_register(call->name,
160                                                  call->class->perf_probe,
161                                                  call);
162         case TRACE_REG_PERF_UNREGISTER:
163                 tracepoint_probe_unregister(call->name,
164                                             call->class->perf_probe,
165                                             call);
166                 return 0;
167 #endif
168         }
169         return 0;
170 }
171 EXPORT_SYMBOL_GPL(ftrace_event_reg);
172
173 void trace_event_enable_cmd_record(bool enable)
174 {
175         struct ftrace_event_call *call;
176
177         mutex_lock(&event_mutex);
178         list_for_each_entry(call, &ftrace_events, list) {
179                 if (!(call->flags & TRACE_EVENT_FL_ENABLED))
180                         continue;
181
182                 if (enable) {
183                         tracing_start_cmdline_record();
184                         call->flags |= TRACE_EVENT_FL_RECORDED_CMD;
185                 } else {
186                         tracing_stop_cmdline_record();
187                         call->flags &= ~TRACE_EVENT_FL_RECORDED_CMD;
188                 }
189         }
190         mutex_unlock(&event_mutex);
191 }
192
193 static int ftrace_event_enable_disable(struct ftrace_event_call *call,
194                                         int enable)
195 {
196         int ret = 0;
197
198         switch (enable) {
199         case 0:
200                 if (call->flags & TRACE_EVENT_FL_ENABLED) {
201                         call->flags &= ~TRACE_EVENT_FL_ENABLED;
202                         if (call->flags & TRACE_EVENT_FL_RECORDED_CMD) {
203                                 tracing_stop_cmdline_record();
204                                 call->flags &= ~TRACE_EVENT_FL_RECORDED_CMD;
205                         }
206                         call->class->reg(call, TRACE_REG_UNREGISTER);
207                 }
208                 break;
209         case 1:
210                 if (!(call->flags & TRACE_EVENT_FL_ENABLED)) {
211                         if (trace_flags & TRACE_ITER_RECORD_CMD) {
212                                 tracing_start_cmdline_record();
213                                 call->flags |= TRACE_EVENT_FL_RECORDED_CMD;
214                         }
215                         ret = call->class->reg(call, TRACE_REG_REGISTER);
216                         if (ret) {
217                                 tracing_stop_cmdline_record();
218                                 pr_info("event trace: Could not enable event "
219                                         "%s\n", call->name);
220                                 break;
221                         }
222                         call->flags |= TRACE_EVENT_FL_ENABLED;
223                 }
224                 break;
225         }
226
227         return ret;
228 }
229
230 static void ftrace_clear_events(void)
231 {
232         struct ftrace_event_call *call;
233
234         mutex_lock(&event_mutex);
235         list_for_each_entry(call, &ftrace_events, list) {
236                 ftrace_event_enable_disable(call, 0);
237         }
238         mutex_unlock(&event_mutex);
239 }
240
241 /*
242  * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
243  */
244 static int __ftrace_set_clr_event(const char *match, const char *sub,
245                                   const char *event, int set)
246 {
247         struct ftrace_event_call *call;
248         int ret = -EINVAL;
249
250         mutex_lock(&event_mutex);
251         list_for_each_entry(call, &ftrace_events, list) {
252
253                 if (!call->name || !call->class || !call->class->reg)
254                         continue;
255
256                 if (match &&
257                     strcmp(match, call->name) != 0 &&
258                     strcmp(match, call->class->system) != 0)
259                         continue;
260
261                 if (sub && strcmp(sub, call->class->system) != 0)
262                         continue;
263
264                 if (event && strcmp(event, call->name) != 0)
265                         continue;
266
267                 ftrace_event_enable_disable(call, set);
268
269                 ret = 0;
270         }
271         mutex_unlock(&event_mutex);
272
273         return ret;
274 }
275
276 static int ftrace_set_clr_event(char *buf, int set)
277 {
278         char *event = NULL, *sub = NULL, *match;
279
280         /*
281          * The buf format can be <subsystem>:<event-name>
282          *  *:<event-name> means any event by that name.
283          *  :<event-name> is the same.
284          *
285          *  <subsystem>:* means all events in that subsystem
286          *  <subsystem>: means the same.
287          *
288          *  <name> (no ':') means all events in a subsystem with
289          *  the name <name> or any event that matches <name>
290          */
291
292         match = strsep(&buf, ":");
293         if (buf) {
294                 sub = match;
295                 event = buf;
296                 match = NULL;
297
298                 if (!strlen(sub) || strcmp(sub, "*") == 0)
299                         sub = NULL;
300                 if (!strlen(event) || strcmp(event, "*") == 0)
301                         event = NULL;
302         }
303
304         return __ftrace_set_clr_event(match, sub, event, set);
305 }
306
307 /**
308  * trace_set_clr_event - enable or disable an event
309  * @system: system name to match (NULL for any system)
310  * @event: event name to match (NULL for all events, within system)
311  * @set: 1 to enable, 0 to disable
312  *
313  * This is a way for other parts of the kernel to enable or disable
314  * event recording.
315  *
316  * Returns 0 on success, -EINVAL if the parameters do not match any
317  * registered events.
318  */
319 int trace_set_clr_event(const char *system, const char *event, int set)
320 {
321         return __ftrace_set_clr_event(NULL, system, event, set);
322 }
323
324 /* 128 should be much more than enough */
325 #define EVENT_BUF_SIZE          127
326
327 static ssize_t
328 ftrace_event_write(struct file *file, const char __user *ubuf,
329                    size_t cnt, loff_t *ppos)
330 {
331         struct trace_parser parser;
332         ssize_t read, ret;
333
334         if (!cnt)
335                 return 0;
336
337         ret = tracing_update_buffers();
338         if (ret < 0)
339                 return ret;
340
341         if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
342                 return -ENOMEM;
343
344         read = trace_get_user(&parser, ubuf, cnt, ppos);
345
346         if (read >= 0 && trace_parser_loaded((&parser))) {
347                 int set = 1;
348
349                 if (*parser.buffer == '!')
350                         set = 0;
351
352                 parser.buffer[parser.idx] = 0;
353
354                 ret = ftrace_set_clr_event(parser.buffer + !set, set);
355                 if (ret)
356                         goto out_put;
357         }
358
359         ret = read;
360
361  out_put:
362         trace_parser_put(&parser);
363
364         return ret;
365 }
366
367 static void *
368 t_next(struct seq_file *m, void *v, loff_t *pos)
369 {
370         struct ftrace_event_call *call = v;
371
372         (*pos)++;
373
374         list_for_each_entry_continue(call, &ftrace_events, list) {
375                 /*
376                  * The ftrace subsystem is for showing formats only.
377                  * They can not be enabled or disabled via the event files.
378                  */
379                 if (call->class && call->class->reg)
380                         return call;
381         }
382
383         return NULL;
384 }
385
386 static void *t_start(struct seq_file *m, loff_t *pos)
387 {
388         struct ftrace_event_call *call;
389         loff_t l;
390
391         mutex_lock(&event_mutex);
392
393         call = list_entry(&ftrace_events, struct ftrace_event_call, list);
394         for (l = 0; l <= *pos; ) {
395                 call = t_next(m, call, &l);
396                 if (!call)
397                         break;
398         }
399         return call;
400 }
401
402 static void *
403 s_next(struct seq_file *m, void *v, loff_t *pos)
404 {
405         struct ftrace_event_call *call = v;
406
407         (*pos)++;
408
409         list_for_each_entry_continue(call, &ftrace_events, list) {
410                 if (call->flags & TRACE_EVENT_FL_ENABLED)
411                         return call;
412         }
413
414         return NULL;
415 }
416
417 static void *s_start(struct seq_file *m, loff_t *pos)
418 {
419         struct ftrace_event_call *call;
420         loff_t l;
421
422         mutex_lock(&event_mutex);
423
424         call = list_entry(&ftrace_events, struct ftrace_event_call, list);
425         for (l = 0; l <= *pos; ) {
426                 call = s_next(m, call, &l);
427                 if (!call)
428                         break;
429         }
430         return call;
431 }
432
433 static int t_show(struct seq_file *m, void *v)
434 {
435         struct ftrace_event_call *call = v;
436
437         if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
438                 seq_printf(m, "%s:", call->class->system);
439         seq_printf(m, "%s\n", call->name);
440
441         return 0;
442 }
443
444 static void t_stop(struct seq_file *m, void *p)
445 {
446         mutex_unlock(&event_mutex);
447 }
448
449 static int
450 ftrace_event_seq_open(struct inode *inode, struct file *file)
451 {
452         const struct seq_operations *seq_ops;
453
454         if ((file->f_mode & FMODE_WRITE) &&
455             (file->f_flags & O_TRUNC))
456                 ftrace_clear_events();
457
458         seq_ops = inode->i_private;
459         return seq_open(file, seq_ops);
460 }
461
462 static ssize_t
463 event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
464                   loff_t *ppos)
465 {
466         struct ftrace_event_call *call = filp->private_data;
467         char *buf;
468
469         if (call->flags & TRACE_EVENT_FL_ENABLED)
470                 buf = "1\n";
471         else
472                 buf = "0\n";
473
474         return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
475 }
476
477 static ssize_t
478 event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
479                    loff_t *ppos)
480 {
481         struct ftrace_event_call *call = filp->private_data;
482         char buf[64];
483         unsigned long val;
484         int ret;
485
486         if (cnt >= sizeof(buf))
487                 return -EINVAL;
488
489         if (copy_from_user(&buf, ubuf, cnt))
490                 return -EFAULT;
491
492         buf[cnt] = 0;
493
494         ret = strict_strtoul(buf, 10, &val);
495         if (ret < 0)
496                 return ret;
497
498         ret = tracing_update_buffers();
499         if (ret < 0)
500                 return ret;
501
502         switch (val) {
503         case 0:
504         case 1:
505                 mutex_lock(&event_mutex);
506                 ret = ftrace_event_enable_disable(call, val);
507                 mutex_unlock(&event_mutex);
508                 break;
509
510         default:
511                 return -EINVAL;
512         }
513
514         *ppos += cnt;
515
516         return ret ? ret : cnt;
517 }
518
519 static ssize_t
520 system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
521                    loff_t *ppos)
522 {
523         const char set_to_char[4] = { '?', '0', '1', 'X' };
524         const char *system = filp->private_data;
525         struct ftrace_event_call *call;
526         char buf[2];
527         int set = 0;
528         int ret;
529
530         mutex_lock(&event_mutex);
531         list_for_each_entry(call, &ftrace_events, list) {
532                 if (!call->name || !call->class || !call->class->reg)
533                         continue;
534
535                 if (system && strcmp(call->class->system, system) != 0)
536                         continue;
537
538                 /*
539                  * We need to find out if all the events are set
540                  * or if all events or cleared, or if we have
541                  * a mixture.
542                  */
543                 set |= (1 << !!(call->flags & TRACE_EVENT_FL_ENABLED));
544
545                 /*
546                  * If we have a mixture, no need to look further.
547                  */
548                 if (set == 3)
549                         break;
550         }
551         mutex_unlock(&event_mutex);
552
553         buf[0] = set_to_char[set];
554         buf[1] = '\n';
555
556         ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
557
558         return ret;
559 }
560
561 static ssize_t
562 system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
563                     loff_t *ppos)
564 {
565         const char *system = filp->private_data;
566         unsigned long val;
567         char buf[64];
568         ssize_t ret;
569
570         if (cnt >= sizeof(buf))
571                 return -EINVAL;
572
573         if (copy_from_user(&buf, ubuf, cnt))
574                 return -EFAULT;
575
576         buf[cnt] = 0;
577
578         ret = strict_strtoul(buf, 10, &val);
579         if (ret < 0)
580                 return ret;
581
582         ret = tracing_update_buffers();
583         if (ret < 0)
584                 return ret;
585
586         if (val != 0 && val != 1)
587                 return -EINVAL;
588
589         ret = __ftrace_set_clr_event(NULL, system, NULL, val);
590         if (ret)
591                 goto out;
592
593         ret = cnt;
594
595 out:
596         *ppos += cnt;
597
598         return ret;
599 }
600
601 static void print_event_fields(struct trace_seq *s, struct list_head *head)
602 {
603         struct ftrace_event_field *field;
604
605         list_for_each_entry_reverse(field, head, link) {
606                 /*
607                  * Smartly shows the array type(except dynamic array).
608                  * Normal:
609                  *      field:TYPE VAR
610                  * If TYPE := TYPE[LEN], it is shown:
611                  *      field:TYPE VAR[LEN]
612                  */
613                 const char *array_descriptor = strchr(field->type, '[');
614
615                 if (!strncmp(field->type, "__data_loc", 10))
616                         array_descriptor = NULL;
617
618                 if (!array_descriptor) {
619                         trace_seq_printf(s, "\tfield:%s %s;\toffset:%u;"
620                                         "\tsize:%u;\tsigned:%d;\n",
621                                         field->type, field->name, field->offset,
622                                         field->size, !!field->is_signed);
623                 } else {
624                         trace_seq_printf(s, "\tfield:%.*s %s%s;\toffset:%u;"
625                                         "\tsize:%u;\tsigned:%d;\n",
626                                         (int)(array_descriptor - field->type),
627                                         field->type, field->name,
628                                         array_descriptor, field->offset,
629                                         field->size, !!field->is_signed);
630                 }
631         }
632 }
633
634 static ssize_t
635 event_format_read(struct file *filp, char __user *ubuf, size_t cnt,
636                   loff_t *ppos)
637 {
638         struct ftrace_event_call *call = filp->private_data;
639         struct list_head *head;
640         struct trace_seq *s;
641         char *buf;
642         int r;
643
644         if (*ppos)
645                 return 0;
646
647         s = kmalloc(sizeof(*s), GFP_KERNEL);
648         if (!s)
649                 return -ENOMEM;
650
651         trace_seq_init(s);
652
653         trace_seq_printf(s, "name: %s\n", call->name);
654         trace_seq_printf(s, "ID: %d\n", call->event.type);
655         trace_seq_printf(s, "format:\n");
656
657         /* print common fields */
658         print_event_fields(s, &ftrace_common_fields);
659
660         trace_seq_putc(s, '\n');
661
662         /* print event specific fields */
663         head = trace_get_fields(call);
664         print_event_fields(s, head);
665
666         r = trace_seq_printf(s, "\nprint fmt: %s\n", call->print_fmt);
667
668         if (!r) {
669                 /*
670                  * ug!  The format output is bigger than a PAGE!!
671                  */
672                 buf = "FORMAT TOO BIG\n";
673                 r = simple_read_from_buffer(ubuf, cnt, ppos,
674                                               buf, strlen(buf));
675                 goto out;
676         }
677
678         r = simple_read_from_buffer(ubuf, cnt, ppos,
679                                     s->buffer, s->len);
680  out:
681         kfree(s);
682         return r;
683 }
684
685 static ssize_t
686 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
687 {
688         struct ftrace_event_call *call = filp->private_data;
689         struct trace_seq *s;
690         int r;
691
692         if (*ppos)
693                 return 0;
694
695         s = kmalloc(sizeof(*s), GFP_KERNEL);
696         if (!s)
697                 return -ENOMEM;
698
699         trace_seq_init(s);
700         trace_seq_printf(s, "%d\n", call->event.type);
701
702         r = simple_read_from_buffer(ubuf, cnt, ppos,
703                                     s->buffer, s->len);
704         kfree(s);
705         return r;
706 }
707
708 static ssize_t
709 event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
710                   loff_t *ppos)
711 {
712         struct ftrace_event_call *call = filp->private_data;
713         struct trace_seq *s;
714         int r;
715
716         if (*ppos)
717                 return 0;
718
719         s = kmalloc(sizeof(*s), GFP_KERNEL);
720         if (!s)
721                 return -ENOMEM;
722
723         trace_seq_init(s);
724
725         print_event_filter(call, s);
726         r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
727
728         kfree(s);
729
730         return r;
731 }
732
733 static ssize_t
734 event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
735                    loff_t *ppos)
736 {
737         struct ftrace_event_call *call = filp->private_data;
738         char *buf;
739         int err;
740
741         if (cnt >= PAGE_SIZE)
742                 return -EINVAL;
743
744         buf = (char *)__get_free_page(GFP_TEMPORARY);
745         if (!buf)
746                 return -ENOMEM;
747
748         if (copy_from_user(buf, ubuf, cnt)) {
749                 free_page((unsigned long) buf);
750                 return -EFAULT;
751         }
752         buf[cnt] = '\0';
753
754         err = apply_event_filter(call, buf);
755         free_page((unsigned long) buf);
756         if (err < 0)
757                 return err;
758
759         *ppos += cnt;
760
761         return cnt;
762 }
763
764 static ssize_t
765 subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
766                       loff_t *ppos)
767 {
768         struct event_subsystem *system = filp->private_data;
769         struct trace_seq *s;
770         int r;
771
772         if (*ppos)
773                 return 0;
774
775         s = kmalloc(sizeof(*s), GFP_KERNEL);
776         if (!s)
777                 return -ENOMEM;
778
779         trace_seq_init(s);
780
781         print_subsystem_event_filter(system, s);
782         r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
783
784         kfree(s);
785
786         return r;
787 }
788
789 static ssize_t
790 subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
791                        loff_t *ppos)
792 {
793         struct event_subsystem *system = filp->private_data;
794         char *buf;
795         int err;
796
797         if (cnt >= PAGE_SIZE)
798                 return -EINVAL;
799
800         buf = (char *)__get_free_page(GFP_TEMPORARY);
801         if (!buf)
802                 return -ENOMEM;
803
804         if (copy_from_user(buf, ubuf, cnt)) {
805                 free_page((unsigned long) buf);
806                 return -EFAULT;
807         }
808         buf[cnt] = '\0';
809
810         err = apply_subsystem_event_filter(system, buf);
811         free_page((unsigned long) buf);
812         if (err < 0)
813                 return err;
814
815         *ppos += cnt;
816
817         return cnt;
818 }
819
820 static ssize_t
821 show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
822 {
823         int (*func)(struct trace_seq *s) = filp->private_data;
824         struct trace_seq *s;
825         int r;
826
827         if (*ppos)
828                 return 0;
829
830         s = kmalloc(sizeof(*s), GFP_KERNEL);
831         if (!s)
832                 return -ENOMEM;
833
834         trace_seq_init(s);
835
836         func(s);
837         r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
838
839         kfree(s);
840
841         return r;
842 }
843
844 static const struct seq_operations show_event_seq_ops = {
845         .start = t_start,
846         .next = t_next,
847         .show = t_show,
848         .stop = t_stop,
849 };
850
851 static const struct seq_operations show_set_event_seq_ops = {
852         .start = s_start,
853         .next = s_next,
854         .show = t_show,
855         .stop = t_stop,
856 };
857
858 static const struct file_operations ftrace_avail_fops = {
859         .open = ftrace_event_seq_open,
860         .read = seq_read,
861         .llseek = seq_lseek,
862         .release = seq_release,
863 };
864
865 static const struct file_operations ftrace_set_event_fops = {
866         .open = ftrace_event_seq_open,
867         .read = seq_read,
868         .write = ftrace_event_write,
869         .llseek = seq_lseek,
870         .release = seq_release,
871 };
872
873 static const struct file_operations ftrace_enable_fops = {
874         .open = tracing_open_generic,
875         .read = event_enable_read,
876         .write = event_enable_write,
877 };
878
879 static const struct file_operations ftrace_event_format_fops = {
880         .open = tracing_open_generic,
881         .read = event_format_read,
882 };
883
884 static const struct file_operations ftrace_event_id_fops = {
885         .open = tracing_open_generic,
886         .read = event_id_read,
887 };
888
889 static const struct file_operations ftrace_event_filter_fops = {
890         .open = tracing_open_generic,
891         .read = event_filter_read,
892         .write = event_filter_write,
893 };
894
895 static const struct file_operations ftrace_subsystem_filter_fops = {
896         .open = tracing_open_generic,
897         .read = subsystem_filter_read,
898         .write = subsystem_filter_write,
899 };
900
901 static const struct file_operations ftrace_system_enable_fops = {
902         .open = tracing_open_generic,
903         .read = system_enable_read,
904         .write = system_enable_write,
905 };
906
907 static const struct file_operations ftrace_show_header_fops = {
908         .open = tracing_open_generic,
909         .read = show_header,
910 };
911
912 static struct dentry *event_trace_events_dir(void)
913 {
914         static struct dentry *d_tracer;
915         static struct dentry *d_events;
916
917         if (d_events)
918                 return d_events;
919
920         d_tracer = tracing_init_dentry();
921         if (!d_tracer)
922                 return NULL;
923
924         d_events = debugfs_create_dir("events", d_tracer);
925         if (!d_events)
926                 pr_warning("Could not create debugfs "
927                            "'events' directory\n");
928
929         return d_events;
930 }
931
932 static LIST_HEAD(event_subsystems);
933
934 static struct dentry *
935 event_subsystem_dir(const char *name, struct dentry *d_events)
936 {
937         struct event_subsystem *system;
938         struct dentry *entry;
939
940         /* First see if we did not already create this dir */
941         list_for_each_entry(system, &event_subsystems, list) {
942                 if (strcmp(system->name, name) == 0) {
943                         system->nr_events++;
944                         return system->entry;
945                 }
946         }
947
948         /* need to create new entry */
949         system = kmalloc(sizeof(*system), GFP_KERNEL);
950         if (!system) {
951                 pr_warning("No memory to create event subsystem %s\n",
952                            name);
953                 return d_events;
954         }
955
956         system->entry = debugfs_create_dir(name, d_events);
957         if (!system->entry) {
958                 pr_warning("Could not create event subsystem %s\n",
959                            name);
960                 kfree(system);
961                 return d_events;
962         }
963
964         system->nr_events = 1;
965         system->name = kstrdup(name, GFP_KERNEL);
966         if (!system->name) {
967                 debugfs_remove(system->entry);
968                 kfree(system);
969                 return d_events;
970         }
971
972         list_add(&system->list, &event_subsystems);
973
974         system->filter = NULL;
975
976         system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
977         if (!system->filter) {
978                 pr_warning("Could not allocate filter for subsystem "
979                            "'%s'\n", name);
980                 return system->entry;
981         }
982
983         entry = debugfs_create_file("filter", 0644, system->entry, system,
984                                     &ftrace_subsystem_filter_fops);
985         if (!entry) {
986                 kfree(system->filter);
987                 system->filter = NULL;
988                 pr_warning("Could not create debugfs "
989                            "'%s/filter' entry\n", name);
990         }
991
992         trace_create_file("enable", 0644, system->entry,
993                           (void *)system->name,
994                           &ftrace_system_enable_fops);
995
996         return system->entry;
997 }
998
999 static int
1000 event_create_dir(struct ftrace_event_call *call, struct dentry *d_events,
1001                  const struct file_operations *id,
1002                  const struct file_operations *enable,
1003                  const struct file_operations *filter,
1004                  const struct file_operations *format)
1005 {
1006         struct list_head *head;
1007         int ret;
1008
1009         /*
1010          * If the trace point header did not define TRACE_SYSTEM
1011          * then the system would be called "TRACE_SYSTEM".
1012          */
1013         if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
1014                 d_events = event_subsystem_dir(call->class->system, d_events);
1015
1016         call->dir = debugfs_create_dir(call->name, d_events);
1017         if (!call->dir) {
1018                 pr_warning("Could not create debugfs "
1019                            "'%s' directory\n", call->name);
1020                 return -1;
1021         }
1022
1023         if (call->class->reg)
1024                 trace_create_file("enable", 0644, call->dir, call,
1025                                   enable);
1026
1027 #ifdef CONFIG_PERF_EVENTS
1028         if (call->event.type && call->class->reg)
1029                 trace_create_file("id", 0444, call->dir, call,
1030                                   id);
1031 #endif
1032
1033         /*
1034          * Other events may have the same class. Only update
1035          * the fields if they are not already defined.
1036          */
1037         head = trace_get_fields(call);
1038         if (list_empty(head)) {
1039                 ret = call->class->define_fields(call);
1040                 if (ret < 0) {
1041                         pr_warning("Could not initialize trace point"
1042                                    " events/%s\n", call->name);
1043                         return ret;
1044                 }
1045         }
1046         trace_create_file("filter", 0644, call->dir, call,
1047                           filter);
1048
1049         trace_create_file("format", 0444, call->dir, call,
1050                           format);
1051
1052         return 0;
1053 }
1054
1055 static int
1056 __trace_add_event_call(struct ftrace_event_call *call, struct module *mod,
1057                        const struct file_operations *id,
1058                        const struct file_operations *enable,
1059                        const struct file_operations *filter,
1060                        const struct file_operations *format)
1061 {
1062         struct dentry *d_events;
1063         int ret;
1064
1065         /* The linker may leave blanks */
1066         if (!call->name)
1067                 return -EINVAL;
1068
1069         if (call->class->raw_init) {
1070                 ret = call->class->raw_init(call);
1071                 if (ret < 0) {
1072                         if (ret != -ENOSYS)
1073                                 pr_warning("Could not initialize trace events/%s\n",
1074                                            call->name);
1075                         return ret;
1076                 }
1077         }
1078
1079         d_events = event_trace_events_dir();
1080         if (!d_events)
1081                 return -ENOENT;
1082
1083         ret = event_create_dir(call, d_events, id, enable, filter, format);
1084         if (!ret)
1085                 list_add(&call->list, &ftrace_events);
1086         call->mod = mod;
1087
1088         return ret;
1089 }
1090
1091 /* Add an additional event_call dynamically */
1092 int trace_add_event_call(struct ftrace_event_call *call)
1093 {
1094         int ret;
1095         mutex_lock(&event_mutex);
1096         ret = __trace_add_event_call(call, NULL, &ftrace_event_id_fops,
1097                                      &ftrace_enable_fops,
1098                                      &ftrace_event_filter_fops,
1099                                      &ftrace_event_format_fops);
1100         mutex_unlock(&event_mutex);
1101         return ret;
1102 }
1103
1104 static void remove_subsystem_dir(const char *name)
1105 {
1106         struct event_subsystem *system;
1107
1108         if (strcmp(name, TRACE_SYSTEM) == 0)
1109                 return;
1110
1111         list_for_each_entry(system, &event_subsystems, list) {
1112                 if (strcmp(system->name, name) == 0) {
1113                         if (!--system->nr_events) {
1114                                 struct event_filter *filter = system->filter;
1115
1116                                 debugfs_remove_recursive(system->entry);
1117                                 list_del(&system->list);
1118                                 if (filter) {
1119                                         kfree(filter->filter_string);
1120                                         kfree(filter);
1121                                 }
1122                                 kfree(system->name);
1123                                 kfree(system);
1124                         }
1125                         break;
1126                 }
1127         }
1128 }
1129
1130 /*
1131  * Must be called under locking both of event_mutex and trace_event_mutex.
1132  */
1133 static void __trace_remove_event_call(struct ftrace_event_call *call)
1134 {
1135         ftrace_event_enable_disable(call, 0);
1136         if (call->event.funcs)
1137                 __unregister_ftrace_event(&call->event);
1138         debugfs_remove_recursive(call->dir);
1139         list_del(&call->list);
1140         trace_destroy_fields(call);
1141         destroy_preds(call);
1142         remove_subsystem_dir(call->class->system);
1143 }
1144
1145 /* Remove an event_call */
1146 void trace_remove_event_call(struct ftrace_event_call *call)
1147 {
1148         mutex_lock(&event_mutex);
1149         down_write(&trace_event_mutex);
1150         __trace_remove_event_call(call);
1151         up_write(&trace_event_mutex);
1152         mutex_unlock(&event_mutex);
1153 }
1154
1155 #define for_each_event(event, start, end)                       \
1156         for (event = start;                                     \
1157              (unsigned long)event < (unsigned long)end;         \
1158              event++)
1159
1160 #ifdef CONFIG_MODULES
1161
1162 static LIST_HEAD(ftrace_module_file_list);
1163
1164 /*
1165  * Modules must own their file_operations to keep up with
1166  * reference counting.
1167  */
1168 struct ftrace_module_file_ops {
1169         struct list_head                list;
1170         struct module                   *mod;
1171         struct file_operations          id;
1172         struct file_operations          enable;
1173         struct file_operations          format;
1174         struct file_operations          filter;
1175 };
1176
1177 static struct ftrace_module_file_ops *
1178 trace_create_file_ops(struct module *mod)
1179 {
1180         struct ftrace_module_file_ops *file_ops;
1181
1182         /*
1183          * This is a bit of a PITA. To allow for correct reference
1184          * counting, modules must "own" their file_operations.
1185          * To do this, we allocate the file operations that will be
1186          * used in the event directory.
1187          */
1188
1189         file_ops = kmalloc(sizeof(*file_ops), GFP_KERNEL);
1190         if (!file_ops)
1191                 return NULL;
1192
1193         file_ops->mod = mod;
1194
1195         file_ops->id = ftrace_event_id_fops;
1196         file_ops->id.owner = mod;
1197
1198         file_ops->enable = ftrace_enable_fops;
1199         file_ops->enable.owner = mod;
1200
1201         file_ops->filter = ftrace_event_filter_fops;
1202         file_ops->filter.owner = mod;
1203
1204         file_ops->format = ftrace_event_format_fops;
1205         file_ops->format.owner = mod;
1206
1207         list_add(&file_ops->list, &ftrace_module_file_list);
1208
1209         return file_ops;
1210 }
1211
1212 static void trace_module_add_events(struct module *mod)
1213 {
1214         struct ftrace_module_file_ops *file_ops = NULL;
1215         struct ftrace_event_call *call, *start, *end;
1216
1217         start = mod->trace_events;
1218         end = mod->trace_events + mod->num_trace_events;
1219
1220         if (start == end)
1221                 return;
1222
1223         file_ops = trace_create_file_ops(mod);
1224         if (!file_ops)
1225                 return;
1226
1227         for_each_event(call, start, end) {
1228                 __trace_add_event_call(call, mod,
1229                                        &file_ops->id, &file_ops->enable,
1230                                        &file_ops->filter, &file_ops->format);
1231         }
1232 }
1233
1234 static void trace_module_remove_events(struct module *mod)
1235 {
1236         struct ftrace_module_file_ops *file_ops;
1237         struct ftrace_event_call *call, *p;
1238         bool found = false;
1239
1240         down_write(&trace_event_mutex);
1241         list_for_each_entry_safe(call, p, &ftrace_events, list) {
1242                 if (call->mod == mod) {
1243                         found = true;
1244                         __trace_remove_event_call(call);
1245                 }
1246         }
1247
1248         /* Now free the file_operations */
1249         list_for_each_entry(file_ops, &ftrace_module_file_list, list) {
1250                 if (file_ops->mod == mod)
1251                         break;
1252         }
1253         if (&file_ops->list != &ftrace_module_file_list) {
1254                 list_del(&file_ops->list);
1255                 kfree(file_ops);
1256         }
1257
1258         /*
1259          * It is safest to reset the ring buffer if the module being unloaded
1260          * registered any events.
1261          */
1262         if (found)
1263                 tracing_reset_current_online_cpus();
1264         up_write(&trace_event_mutex);
1265 }
1266
1267 static int trace_module_notify(struct notifier_block *self,
1268                                unsigned long val, void *data)
1269 {
1270         struct module *mod = data;
1271
1272         mutex_lock(&event_mutex);
1273         switch (val) {
1274         case MODULE_STATE_COMING:
1275                 trace_module_add_events(mod);
1276                 break;
1277         case MODULE_STATE_GOING:
1278                 trace_module_remove_events(mod);
1279                 break;
1280         }
1281         mutex_unlock(&event_mutex);
1282
1283         return 0;
1284 }
1285 #else
1286 static int trace_module_notify(struct notifier_block *self,
1287                                unsigned long val, void *data)
1288 {
1289         return 0;
1290 }
1291 #endif /* CONFIG_MODULES */
1292
1293 static struct notifier_block trace_module_nb = {
1294         .notifier_call = trace_module_notify,
1295         .priority = 0,
1296 };
1297
1298 extern struct ftrace_event_call __start_ftrace_events[];
1299 extern struct ftrace_event_call __stop_ftrace_events[];
1300
1301 static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
1302
1303 static __init int setup_trace_event(char *str)
1304 {
1305         strlcpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
1306         ring_buffer_expanded = 1;
1307         tracing_selftest_disabled = 1;
1308
1309         return 1;
1310 }
1311 __setup("trace_event=", setup_trace_event);
1312
1313 static __init int event_trace_init(void)
1314 {
1315         struct ftrace_event_call *call;
1316         struct dentry *d_tracer;
1317         struct dentry *entry;
1318         struct dentry *d_events;
1319         int ret;
1320         char *buf = bootup_event_buf;
1321         char *token;
1322
1323         d_tracer = tracing_init_dentry();
1324         if (!d_tracer)
1325                 return 0;
1326
1327         entry = debugfs_create_file("available_events", 0444, d_tracer,
1328                                     (void *)&show_event_seq_ops,
1329                                     &ftrace_avail_fops);
1330         if (!entry)
1331                 pr_warning("Could not create debugfs "
1332                            "'available_events' entry\n");
1333
1334         entry = debugfs_create_file("set_event", 0644, d_tracer,
1335                                     (void *)&show_set_event_seq_ops,
1336                                     &ftrace_set_event_fops);
1337         if (!entry)
1338                 pr_warning("Could not create debugfs "
1339                            "'set_event' entry\n");
1340
1341         d_events = event_trace_events_dir();
1342         if (!d_events)
1343                 return 0;
1344
1345         /* ring buffer internal formats */
1346         trace_create_file("header_page", 0444, d_events,
1347                           ring_buffer_print_page_header,
1348                           &ftrace_show_header_fops);
1349
1350         trace_create_file("header_event", 0444, d_events,
1351                           ring_buffer_print_entry_header,
1352                           &ftrace_show_header_fops);
1353
1354         trace_create_file("enable", 0644, d_events,
1355                           NULL, &ftrace_system_enable_fops);
1356
1357         if (trace_define_common_fields())
1358                 pr_warning("tracing: Failed to allocate common fields");
1359
1360         for_each_event(call, __start_ftrace_events, __stop_ftrace_events) {
1361                 __trace_add_event_call(call, NULL, &ftrace_event_id_fops,
1362                                        &ftrace_enable_fops,
1363                                        &ftrace_event_filter_fops,
1364                                        &ftrace_event_format_fops);
1365         }
1366
1367         while (true) {
1368                 token = strsep(&buf, ",");
1369
1370                 if (!token)
1371                         break;
1372                 if (!*token)
1373                         continue;
1374
1375                 ret = ftrace_set_clr_event(token, 1);
1376                 if (ret)
1377                         pr_warning("Failed to enable trace event: %s\n", token);
1378         }
1379
1380         ret = register_module_notifier(&trace_module_nb);
1381         if (ret)
1382                 pr_warning("Failed to register trace events module notifier\n");
1383
1384         return 0;
1385 }
1386 fs_initcall(event_trace_init);
1387
1388 #ifdef CONFIG_FTRACE_STARTUP_TEST
1389
1390 static DEFINE_SPINLOCK(test_spinlock);
1391 static DEFINE_SPINLOCK(test_spinlock_irq);
1392 static DEFINE_MUTEX(test_mutex);
1393
1394 static __init void test_work(struct work_struct *dummy)
1395 {
1396         spin_lock(&test_spinlock);
1397         spin_lock_irq(&test_spinlock_irq);
1398         udelay(1);
1399         spin_unlock_irq(&test_spinlock_irq);
1400         spin_unlock(&test_spinlock);
1401
1402         mutex_lock(&test_mutex);
1403         msleep(1);
1404         mutex_unlock(&test_mutex);
1405 }
1406
1407 static __init int event_test_thread(void *unused)
1408 {
1409         void *test_malloc;
1410
1411         test_malloc = kmalloc(1234, GFP_KERNEL);
1412         if (!test_malloc)
1413                 pr_info("failed to kmalloc\n");
1414
1415         schedule_on_each_cpu(test_work);
1416
1417         kfree(test_malloc);
1418
1419         set_current_state(TASK_INTERRUPTIBLE);
1420         while (!kthread_should_stop())
1421                 schedule();
1422
1423         return 0;
1424 }
1425
1426 /*
1427  * Do various things that may trigger events.
1428  */
1429 static __init void event_test_stuff(void)
1430 {
1431         struct task_struct *test_thread;
1432
1433         test_thread = kthread_run(event_test_thread, NULL, "test-events");
1434         msleep(1);
1435         kthread_stop(test_thread);
1436 }
1437
1438 /*
1439  * For every trace event defined, we will test each trace point separately,
1440  * and then by groups, and finally all trace points.
1441  */
1442 static __init void event_trace_self_tests(void)
1443 {
1444         struct ftrace_event_call *call;
1445         struct event_subsystem *system;
1446         int ret;
1447
1448         pr_info("Running tests on trace events:\n");
1449
1450         list_for_each_entry(call, &ftrace_events, list) {
1451
1452                 /* Only test those that have a probe */
1453                 if (!call->class || !call->class->probe)
1454                         continue;
1455
1456 /*
1457  * Testing syscall events here is pretty useless, but
1458  * we still do it if configured. But this is time consuming.
1459  * What we really need is a user thread to perform the
1460  * syscalls as we test.
1461  */
1462 #ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
1463                 if (call->class->system &&
1464                     strcmp(call->class->system, "syscalls") == 0)
1465                         continue;
1466 #endif
1467
1468                 pr_info("Testing event %s: ", call->name);
1469
1470                 /*
1471                  * If an event is already enabled, someone is using
1472                  * it and the self test should not be on.
1473                  */
1474                 if (call->flags & TRACE_EVENT_FL_ENABLED) {
1475                         pr_warning("Enabled event during self test!\n");
1476                         WARN_ON_ONCE(1);
1477                         continue;
1478                 }
1479
1480                 ftrace_event_enable_disable(call, 1);
1481                 event_test_stuff();
1482                 ftrace_event_enable_disable(call, 0);
1483
1484                 pr_cont("OK\n");
1485         }
1486
1487         /* Now test at the sub system level */
1488
1489         pr_info("Running tests on trace event systems:\n");
1490
1491         list_for_each_entry(system, &event_subsystems, list) {
1492
1493                 /* the ftrace system is special, skip it */
1494                 if (strcmp(system->name, "ftrace") == 0)
1495                         continue;
1496
1497                 pr_info("Testing event system %s: ", system->name);
1498
1499                 ret = __ftrace_set_clr_event(NULL, system->name, NULL, 1);
1500                 if (WARN_ON_ONCE(ret)) {
1501                         pr_warning("error enabling system %s\n",
1502                                    system->name);
1503                         continue;
1504                 }
1505
1506                 event_test_stuff();
1507
1508                 ret = __ftrace_set_clr_event(NULL, system->name, NULL, 0);
1509                 if (WARN_ON_ONCE(ret))
1510                         pr_warning("error disabling system %s\n",
1511                                    system->name);
1512
1513                 pr_cont("OK\n");
1514         }
1515
1516         /* Test with all events enabled */
1517
1518         pr_info("Running tests on all trace events:\n");
1519         pr_info("Testing all events: ");
1520
1521         ret = __ftrace_set_clr_event(NULL, NULL, NULL, 1);
1522         if (WARN_ON_ONCE(ret)) {
1523                 pr_warning("error enabling all events\n");
1524                 return;
1525         }
1526
1527         event_test_stuff();
1528
1529         /* reset sysname */
1530         ret = __ftrace_set_clr_event(NULL, NULL, NULL, 0);
1531         if (WARN_ON_ONCE(ret)) {
1532                 pr_warning("error disabling all events\n");
1533                 return;
1534         }
1535
1536         pr_cont("OK\n");
1537 }
1538
1539 #ifdef CONFIG_FUNCTION_TRACER
1540
1541 static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
1542
1543 static void
1544 function_test_events_call(unsigned long ip, unsigned long parent_ip)
1545 {
1546         struct ring_buffer_event *event;
1547         struct ring_buffer *buffer;
1548         struct ftrace_entry *entry;
1549         unsigned long flags;
1550         long disabled;
1551         int cpu;
1552         int pc;
1553
1554         pc = preempt_count();
1555         preempt_disable_notrace();
1556         cpu = raw_smp_processor_id();
1557         disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
1558
1559         if (disabled != 1)
1560                 goto out;
1561
1562         local_save_flags(flags);
1563
1564         event = trace_current_buffer_lock_reserve(&buffer,
1565                                                   TRACE_FN, sizeof(*entry),
1566                                                   flags, pc);
1567         if (!event)
1568                 goto out;
1569         entry   = ring_buffer_event_data(event);
1570         entry->ip                       = ip;
1571         entry->parent_ip                = parent_ip;
1572
1573         trace_nowake_buffer_unlock_commit(buffer, event, flags, pc);
1574
1575  out:
1576         atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
1577         preempt_enable_notrace();
1578 }
1579
1580 static struct ftrace_ops trace_ops __initdata  =
1581 {
1582         .func = function_test_events_call,
1583 };
1584
1585 static __init void event_trace_self_test_with_function(void)
1586 {
1587         register_ftrace_function(&trace_ops);
1588         pr_info("Running tests again, along with the function tracer\n");
1589         event_trace_self_tests();
1590         unregister_ftrace_function(&trace_ops);
1591 }
1592 #else
1593 static __init void event_trace_self_test_with_function(void)
1594 {
1595 }
1596 #endif
1597
1598 static __init int event_trace_self_tests_init(void)
1599 {
1600         if (!tracing_selftest_disabled) {
1601                 event_trace_self_tests();
1602                 event_trace_self_test_with_function();
1603         }
1604
1605         return 0;
1606 }
1607
1608 late_initcall(event_trace_self_tests_init);
1609
1610 #endif