tracing: Remove test of NULL define_fields callback
[linux-flexiantxendom0-3.2.10.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 static int ftrace_event_enable_disable(struct ftrace_event_call *call,
145                                         int enable)
146 {
147         int ret = 0;
148
149         switch (enable) {
150         case 0:
151                 if (call->flags & TRACE_EVENT_FL_ENABLED) {
152                         call->flags &= ~TRACE_EVENT_FL_ENABLED;
153                         tracing_stop_cmdline_record();
154                         if (call->class->reg)
155                                 call->class->reg(call, TRACE_REG_UNREGISTER);
156                         else
157                                 tracepoint_probe_unregister(call->name,
158                                                             call->class->probe,
159                                                             call);
160                 }
161                 break;
162         case 1:
163                 if (!(call->flags & TRACE_EVENT_FL_ENABLED)) {
164                         tracing_start_cmdline_record();
165                         if (call->class->reg)
166                                 ret = call->class->reg(call, TRACE_REG_REGISTER);
167                         else
168                                 ret = tracepoint_probe_register(call->name,
169                                                                 call->class->probe,
170                                                                 call);
171                         if (ret) {
172                                 tracing_stop_cmdline_record();
173                                 pr_info("event trace: Could not enable event "
174                                         "%s\n", call->name);
175                                 break;
176                         }
177                         call->flags |= TRACE_EVENT_FL_ENABLED;
178                 }
179                 break;
180         }
181
182         return ret;
183 }
184
185 static void ftrace_clear_events(void)
186 {
187         struct ftrace_event_call *call;
188
189         mutex_lock(&event_mutex);
190         list_for_each_entry(call, &ftrace_events, list) {
191                 ftrace_event_enable_disable(call, 0);
192         }
193         mutex_unlock(&event_mutex);
194 }
195
196 /*
197  * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
198  */
199 static int __ftrace_set_clr_event(const char *match, const char *sub,
200                                   const char *event, int set)
201 {
202         struct ftrace_event_call *call;
203         int ret = -EINVAL;
204
205         mutex_lock(&event_mutex);
206         list_for_each_entry(call, &ftrace_events, list) {
207
208                 if (!call->name || !call->class ||
209                     (!call->class->probe && !call->class->reg))
210                         continue;
211
212                 if (match &&
213                     strcmp(match, call->name) != 0 &&
214                     strcmp(match, call->class->system) != 0)
215                         continue;
216
217                 if (sub && strcmp(sub, call->class->system) != 0)
218                         continue;
219
220                 if (event && strcmp(event, call->name) != 0)
221                         continue;
222
223                 ftrace_event_enable_disable(call, set);
224
225                 ret = 0;
226         }
227         mutex_unlock(&event_mutex);
228
229         return ret;
230 }
231
232 static int ftrace_set_clr_event(char *buf, int set)
233 {
234         char *event = NULL, *sub = NULL, *match;
235
236         /*
237          * The buf format can be <subsystem>:<event-name>
238          *  *:<event-name> means any event by that name.
239          *  :<event-name> is the same.
240          *
241          *  <subsystem>:* means all events in that subsystem
242          *  <subsystem>: means the same.
243          *
244          *  <name> (no ':') means all events in a subsystem with
245          *  the name <name> or any event that matches <name>
246          */
247
248         match = strsep(&buf, ":");
249         if (buf) {
250                 sub = match;
251                 event = buf;
252                 match = NULL;
253
254                 if (!strlen(sub) || strcmp(sub, "*") == 0)
255                         sub = NULL;
256                 if (!strlen(event) || strcmp(event, "*") == 0)
257                         event = NULL;
258         }
259
260         return __ftrace_set_clr_event(match, sub, event, set);
261 }
262
263 /**
264  * trace_set_clr_event - enable or disable an event
265  * @system: system name to match (NULL for any system)
266  * @event: event name to match (NULL for all events, within system)
267  * @set: 1 to enable, 0 to disable
268  *
269  * This is a way for other parts of the kernel to enable or disable
270  * event recording.
271  *
272  * Returns 0 on success, -EINVAL if the parameters do not match any
273  * registered events.
274  */
275 int trace_set_clr_event(const char *system, const char *event, int set)
276 {
277         return __ftrace_set_clr_event(NULL, system, event, set);
278 }
279
280 /* 128 should be much more than enough */
281 #define EVENT_BUF_SIZE          127
282
283 static ssize_t
284 ftrace_event_write(struct file *file, const char __user *ubuf,
285                    size_t cnt, loff_t *ppos)
286 {
287         struct trace_parser parser;
288         ssize_t read, ret;
289
290         if (!cnt)
291                 return 0;
292
293         ret = tracing_update_buffers();
294         if (ret < 0)
295                 return ret;
296
297         if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
298                 return -ENOMEM;
299
300         read = trace_get_user(&parser, ubuf, cnt, ppos);
301
302         if (read >= 0 && trace_parser_loaded((&parser))) {
303                 int set = 1;
304
305                 if (*parser.buffer == '!')
306                         set = 0;
307
308                 parser.buffer[parser.idx] = 0;
309
310                 ret = ftrace_set_clr_event(parser.buffer + !set, set);
311                 if (ret)
312                         goto out_put;
313         }
314
315         ret = read;
316
317  out_put:
318         trace_parser_put(&parser);
319
320         return ret;
321 }
322
323 static void *
324 t_next(struct seq_file *m, void *v, loff_t *pos)
325 {
326         struct ftrace_event_call *call = v;
327
328         (*pos)++;
329
330         list_for_each_entry_continue(call, &ftrace_events, list) {
331                 /*
332                  * The ftrace subsystem is for showing formats only.
333                  * They can not be enabled or disabled via the event files.
334                  */
335                 if (call->class && (call->class->probe || call->class->reg))
336                         return call;
337         }
338
339         return NULL;
340 }
341
342 static void *t_start(struct seq_file *m, loff_t *pos)
343 {
344         struct ftrace_event_call *call;
345         loff_t l;
346
347         mutex_lock(&event_mutex);
348
349         call = list_entry(&ftrace_events, struct ftrace_event_call, list);
350         for (l = 0; l <= *pos; ) {
351                 call = t_next(m, call, &l);
352                 if (!call)
353                         break;
354         }
355         return call;
356 }
357
358 static void *
359 s_next(struct seq_file *m, void *v, loff_t *pos)
360 {
361         struct ftrace_event_call *call = v;
362
363         (*pos)++;
364
365         list_for_each_entry_continue(call, &ftrace_events, list) {
366                 if (call->flags & TRACE_EVENT_FL_ENABLED)
367                         return call;
368         }
369
370         return NULL;
371 }
372
373 static void *s_start(struct seq_file *m, loff_t *pos)
374 {
375         struct ftrace_event_call *call;
376         loff_t l;
377
378         mutex_lock(&event_mutex);
379
380         call = list_entry(&ftrace_events, struct ftrace_event_call, list);
381         for (l = 0; l <= *pos; ) {
382                 call = s_next(m, call, &l);
383                 if (!call)
384                         break;
385         }
386         return call;
387 }
388
389 static int t_show(struct seq_file *m, void *v)
390 {
391         struct ftrace_event_call *call = v;
392
393         if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
394                 seq_printf(m, "%s:", call->class->system);
395         seq_printf(m, "%s\n", call->name);
396
397         return 0;
398 }
399
400 static void t_stop(struct seq_file *m, void *p)
401 {
402         mutex_unlock(&event_mutex);
403 }
404
405 static int
406 ftrace_event_seq_open(struct inode *inode, struct file *file)
407 {
408         const struct seq_operations *seq_ops;
409
410         if ((file->f_mode & FMODE_WRITE) &&
411             (file->f_flags & O_TRUNC))
412                 ftrace_clear_events();
413
414         seq_ops = inode->i_private;
415         return seq_open(file, seq_ops);
416 }
417
418 static ssize_t
419 event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
420                   loff_t *ppos)
421 {
422         struct ftrace_event_call *call = filp->private_data;
423         char *buf;
424
425         if (call->flags & TRACE_EVENT_FL_ENABLED)
426                 buf = "1\n";
427         else
428                 buf = "0\n";
429
430         return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
431 }
432
433 static ssize_t
434 event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
435                    loff_t *ppos)
436 {
437         struct ftrace_event_call *call = filp->private_data;
438         char buf[64];
439         unsigned long val;
440         int ret;
441
442         if (cnt >= sizeof(buf))
443                 return -EINVAL;
444
445         if (copy_from_user(&buf, ubuf, cnt))
446                 return -EFAULT;
447
448         buf[cnt] = 0;
449
450         ret = strict_strtoul(buf, 10, &val);
451         if (ret < 0)
452                 return ret;
453
454         ret = tracing_update_buffers();
455         if (ret < 0)
456                 return ret;
457
458         switch (val) {
459         case 0:
460         case 1:
461                 mutex_lock(&event_mutex);
462                 ret = ftrace_event_enable_disable(call, val);
463                 mutex_unlock(&event_mutex);
464                 break;
465
466         default:
467                 return -EINVAL;
468         }
469
470         *ppos += cnt;
471
472         return ret ? ret : cnt;
473 }
474
475 static ssize_t
476 system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
477                    loff_t *ppos)
478 {
479         const char set_to_char[4] = { '?', '0', '1', 'X' };
480         const char *system = filp->private_data;
481         struct ftrace_event_call *call;
482         char buf[2];
483         int set = 0;
484         int ret;
485
486         mutex_lock(&event_mutex);
487         list_for_each_entry(call, &ftrace_events, list) {
488                 if (!call->name || !call->class ||
489                     (!call->class->probe && !call->class->reg))
490                         continue;
491
492                 if (system && strcmp(call->class->system, system) != 0)
493                         continue;
494
495                 /*
496                  * We need to find out if all the events are set
497                  * or if all events or cleared, or if we have
498                  * a mixture.
499                  */
500                 set |= (1 << !!(call->flags & TRACE_EVENT_FL_ENABLED));
501
502                 /*
503                  * If we have a mixture, no need to look further.
504                  */
505                 if (set == 3)
506                         break;
507         }
508         mutex_unlock(&event_mutex);
509
510         buf[0] = set_to_char[set];
511         buf[1] = '\n';
512
513         ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
514
515         return ret;
516 }
517
518 static ssize_t
519 system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
520                     loff_t *ppos)
521 {
522         const char *system = filp->private_data;
523         unsigned long val;
524         char buf[64];
525         ssize_t ret;
526
527         if (cnt >= sizeof(buf))
528                 return -EINVAL;
529
530         if (copy_from_user(&buf, ubuf, cnt))
531                 return -EFAULT;
532
533         buf[cnt] = 0;
534
535         ret = strict_strtoul(buf, 10, &val);
536         if (ret < 0)
537                 return ret;
538
539         ret = tracing_update_buffers();
540         if (ret < 0)
541                 return ret;
542
543         if (val != 0 && val != 1)
544                 return -EINVAL;
545
546         ret = __ftrace_set_clr_event(NULL, system, NULL, val);
547         if (ret)
548                 goto out;
549
550         ret = cnt;
551
552 out:
553         *ppos += cnt;
554
555         return ret;
556 }
557
558 static void print_event_fields(struct trace_seq *s, struct list_head *head)
559 {
560         struct ftrace_event_field *field;
561
562         list_for_each_entry_reverse(field, head, link) {
563                 /*
564                  * Smartly shows the array type(except dynamic array).
565                  * Normal:
566                  *      field:TYPE VAR
567                  * If TYPE := TYPE[LEN], it is shown:
568                  *      field:TYPE VAR[LEN]
569                  */
570                 const char *array_descriptor = strchr(field->type, '[');
571
572                 if (!strncmp(field->type, "__data_loc", 10))
573                         array_descriptor = NULL;
574
575                 if (!array_descriptor) {
576                         trace_seq_printf(s, "\tfield:%s %s;\toffset:%u;"
577                                         "\tsize:%u;\tsigned:%d;\n",
578                                         field->type, field->name, field->offset,
579                                         field->size, !!field->is_signed);
580                 } else {
581                         trace_seq_printf(s, "\tfield:%.*s %s%s;\toffset:%u;"
582                                         "\tsize:%u;\tsigned:%d;\n",
583                                         (int)(array_descriptor - field->type),
584                                         field->type, field->name,
585                                         array_descriptor, field->offset,
586                                         field->size, !!field->is_signed);
587                 }
588         }
589 }
590
591 static ssize_t
592 event_format_read(struct file *filp, char __user *ubuf, size_t cnt,
593                   loff_t *ppos)
594 {
595         struct ftrace_event_call *call = filp->private_data;
596         struct list_head *head;
597         struct trace_seq *s;
598         char *buf;
599         int r;
600
601         if (*ppos)
602                 return 0;
603
604         s = kmalloc(sizeof(*s), GFP_KERNEL);
605         if (!s)
606                 return -ENOMEM;
607
608         trace_seq_init(s);
609
610         trace_seq_printf(s, "name: %s\n", call->name);
611         trace_seq_printf(s, "ID: %d\n", call->event.type);
612         trace_seq_printf(s, "format:\n");
613
614         /* print common fields */
615         print_event_fields(s, &ftrace_common_fields);
616
617         trace_seq_putc(s, '\n');
618
619         /* print event specific fields */
620         head = trace_get_fields(call);
621         print_event_fields(s, head);
622
623         r = trace_seq_printf(s, "\nprint fmt: %s\n", call->print_fmt);
624
625         if (!r) {
626                 /*
627                  * ug!  The format output is bigger than a PAGE!!
628                  */
629                 buf = "FORMAT TOO BIG\n";
630                 r = simple_read_from_buffer(ubuf, cnt, ppos,
631                                               buf, strlen(buf));
632                 goto out;
633         }
634
635         r = simple_read_from_buffer(ubuf, cnt, ppos,
636                                     s->buffer, s->len);
637  out:
638         kfree(s);
639         return r;
640 }
641
642 static ssize_t
643 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
644 {
645         struct ftrace_event_call *call = filp->private_data;
646         struct trace_seq *s;
647         int r;
648
649         if (*ppos)
650                 return 0;
651
652         s = kmalloc(sizeof(*s), GFP_KERNEL);
653         if (!s)
654                 return -ENOMEM;
655
656         trace_seq_init(s);
657         trace_seq_printf(s, "%d\n", call->event.type);
658
659         r = simple_read_from_buffer(ubuf, cnt, ppos,
660                                     s->buffer, s->len);
661         kfree(s);
662         return r;
663 }
664
665 static ssize_t
666 event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
667                   loff_t *ppos)
668 {
669         struct ftrace_event_call *call = filp->private_data;
670         struct trace_seq *s;
671         int r;
672
673         if (*ppos)
674                 return 0;
675
676         s = kmalloc(sizeof(*s), GFP_KERNEL);
677         if (!s)
678                 return -ENOMEM;
679
680         trace_seq_init(s);
681
682         print_event_filter(call, s);
683         r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
684
685         kfree(s);
686
687         return r;
688 }
689
690 static ssize_t
691 event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
692                    loff_t *ppos)
693 {
694         struct ftrace_event_call *call = filp->private_data;
695         char *buf;
696         int err;
697
698         if (cnt >= PAGE_SIZE)
699                 return -EINVAL;
700
701         buf = (char *)__get_free_page(GFP_TEMPORARY);
702         if (!buf)
703                 return -ENOMEM;
704
705         if (copy_from_user(buf, ubuf, cnt)) {
706                 free_page((unsigned long) buf);
707                 return -EFAULT;
708         }
709         buf[cnt] = '\0';
710
711         err = apply_event_filter(call, buf);
712         free_page((unsigned long) buf);
713         if (err < 0)
714                 return err;
715
716         *ppos += cnt;
717
718         return cnt;
719 }
720
721 static ssize_t
722 subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
723                       loff_t *ppos)
724 {
725         struct event_subsystem *system = filp->private_data;
726         struct trace_seq *s;
727         int r;
728
729         if (*ppos)
730                 return 0;
731
732         s = kmalloc(sizeof(*s), GFP_KERNEL);
733         if (!s)
734                 return -ENOMEM;
735
736         trace_seq_init(s);
737
738         print_subsystem_event_filter(system, s);
739         r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
740
741         kfree(s);
742
743         return r;
744 }
745
746 static ssize_t
747 subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
748                        loff_t *ppos)
749 {
750         struct event_subsystem *system = filp->private_data;
751         char *buf;
752         int err;
753
754         if (cnt >= PAGE_SIZE)
755                 return -EINVAL;
756
757         buf = (char *)__get_free_page(GFP_TEMPORARY);
758         if (!buf)
759                 return -ENOMEM;
760
761         if (copy_from_user(buf, ubuf, cnt)) {
762                 free_page((unsigned long) buf);
763                 return -EFAULT;
764         }
765         buf[cnt] = '\0';
766
767         err = apply_subsystem_event_filter(system, buf);
768         free_page((unsigned long) buf);
769         if (err < 0)
770                 return err;
771
772         *ppos += cnt;
773
774         return cnt;
775 }
776
777 static ssize_t
778 show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
779 {
780         int (*func)(struct trace_seq *s) = filp->private_data;
781         struct trace_seq *s;
782         int r;
783
784         if (*ppos)
785                 return 0;
786
787         s = kmalloc(sizeof(*s), GFP_KERNEL);
788         if (!s)
789                 return -ENOMEM;
790
791         trace_seq_init(s);
792
793         func(s);
794         r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
795
796         kfree(s);
797
798         return r;
799 }
800
801 static const struct seq_operations show_event_seq_ops = {
802         .start = t_start,
803         .next = t_next,
804         .show = t_show,
805         .stop = t_stop,
806 };
807
808 static const struct seq_operations show_set_event_seq_ops = {
809         .start = s_start,
810         .next = s_next,
811         .show = t_show,
812         .stop = t_stop,
813 };
814
815 static const struct file_operations ftrace_avail_fops = {
816         .open = ftrace_event_seq_open,
817         .read = seq_read,
818         .llseek = seq_lseek,
819         .release = seq_release,
820 };
821
822 static const struct file_operations ftrace_set_event_fops = {
823         .open = ftrace_event_seq_open,
824         .read = seq_read,
825         .write = ftrace_event_write,
826         .llseek = seq_lseek,
827         .release = seq_release,
828 };
829
830 static const struct file_operations ftrace_enable_fops = {
831         .open = tracing_open_generic,
832         .read = event_enable_read,
833         .write = event_enable_write,
834 };
835
836 static const struct file_operations ftrace_event_format_fops = {
837         .open = tracing_open_generic,
838         .read = event_format_read,
839 };
840
841 static const struct file_operations ftrace_event_id_fops = {
842         .open = tracing_open_generic,
843         .read = event_id_read,
844 };
845
846 static const struct file_operations ftrace_event_filter_fops = {
847         .open = tracing_open_generic,
848         .read = event_filter_read,
849         .write = event_filter_write,
850 };
851
852 static const struct file_operations ftrace_subsystem_filter_fops = {
853         .open = tracing_open_generic,
854         .read = subsystem_filter_read,
855         .write = subsystem_filter_write,
856 };
857
858 static const struct file_operations ftrace_system_enable_fops = {
859         .open = tracing_open_generic,
860         .read = system_enable_read,
861         .write = system_enable_write,
862 };
863
864 static const struct file_operations ftrace_show_header_fops = {
865         .open = tracing_open_generic,
866         .read = show_header,
867 };
868
869 static struct dentry *event_trace_events_dir(void)
870 {
871         static struct dentry *d_tracer;
872         static struct dentry *d_events;
873
874         if (d_events)
875                 return d_events;
876
877         d_tracer = tracing_init_dentry();
878         if (!d_tracer)
879                 return NULL;
880
881         d_events = debugfs_create_dir("events", d_tracer);
882         if (!d_events)
883                 pr_warning("Could not create debugfs "
884                            "'events' directory\n");
885
886         return d_events;
887 }
888
889 static LIST_HEAD(event_subsystems);
890
891 static struct dentry *
892 event_subsystem_dir(const char *name, struct dentry *d_events)
893 {
894         struct event_subsystem *system;
895         struct dentry *entry;
896
897         /* First see if we did not already create this dir */
898         list_for_each_entry(system, &event_subsystems, list) {
899                 if (strcmp(system->name, name) == 0) {
900                         system->nr_events++;
901                         return system->entry;
902                 }
903         }
904
905         /* need to create new entry */
906         system = kmalloc(sizeof(*system), GFP_KERNEL);
907         if (!system) {
908                 pr_warning("No memory to create event subsystem %s\n",
909                            name);
910                 return d_events;
911         }
912
913         system->entry = debugfs_create_dir(name, d_events);
914         if (!system->entry) {
915                 pr_warning("Could not create event subsystem %s\n",
916                            name);
917                 kfree(system);
918                 return d_events;
919         }
920
921         system->nr_events = 1;
922         system->name = kstrdup(name, GFP_KERNEL);
923         if (!system->name) {
924                 debugfs_remove(system->entry);
925                 kfree(system);
926                 return d_events;
927         }
928
929         list_add(&system->list, &event_subsystems);
930
931         system->filter = NULL;
932
933         system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
934         if (!system->filter) {
935                 pr_warning("Could not allocate filter for subsystem "
936                            "'%s'\n", name);
937                 return system->entry;
938         }
939
940         entry = debugfs_create_file("filter", 0644, system->entry, system,
941                                     &ftrace_subsystem_filter_fops);
942         if (!entry) {
943                 kfree(system->filter);
944                 system->filter = NULL;
945                 pr_warning("Could not create debugfs "
946                            "'%s/filter' entry\n", name);
947         }
948
949         trace_create_file("enable", 0644, system->entry,
950                           (void *)system->name,
951                           &ftrace_system_enable_fops);
952
953         return system->entry;
954 }
955
956 static int
957 event_create_dir(struct ftrace_event_call *call, struct dentry *d_events,
958                  const struct file_operations *id,
959                  const struct file_operations *enable,
960                  const struct file_operations *filter,
961                  const struct file_operations *format)
962 {
963         struct list_head *head;
964         int ret;
965
966         /*
967          * If the trace point header did not define TRACE_SYSTEM
968          * then the system would be called "TRACE_SYSTEM".
969          */
970         if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
971                 d_events = event_subsystem_dir(call->class->system, d_events);
972
973         call->dir = debugfs_create_dir(call->name, d_events);
974         if (!call->dir) {
975                 pr_warning("Could not create debugfs "
976                            "'%s' directory\n", call->name);
977                 return -1;
978         }
979
980         if (call->class->probe || call->class->reg)
981                 trace_create_file("enable", 0644, call->dir, call,
982                                   enable);
983
984 #ifdef CONFIG_PERF_EVENTS
985         if (call->event.type && (call->class->perf_probe || call->class->reg))
986                 trace_create_file("id", 0444, call->dir, call,
987                                   id);
988 #endif
989
990         /*
991          * Other events may have the same class. Only update
992          * the fields if they are not already defined.
993          */
994         head = trace_get_fields(call);
995         if (list_empty(head)) {
996                 ret = call->class->define_fields(call);
997                 if (ret < 0) {
998                         pr_warning("Could not initialize trace point"
999                                    " events/%s\n", call->name);
1000                         return ret;
1001                 }
1002         }
1003         trace_create_file("filter", 0644, call->dir, call,
1004                           filter);
1005
1006         trace_create_file("format", 0444, call->dir, call,
1007                           format);
1008
1009         return 0;
1010 }
1011
1012 static int __trace_add_event_call(struct ftrace_event_call *call)
1013 {
1014         struct dentry *d_events;
1015         int ret;
1016
1017         if (!call->name)
1018                 return -EINVAL;
1019
1020         if (call->class->raw_init) {
1021                 ret = call->class->raw_init(call);
1022                 if (ret < 0) {
1023                         if (ret != -ENOSYS)
1024                                 pr_warning("Could not initialize trace "
1025                                 "events/%s\n", call->name);
1026                         return ret;
1027                 }
1028         }
1029
1030         d_events = event_trace_events_dir();
1031         if (!d_events)
1032                 return -ENOENT;
1033
1034         ret = event_create_dir(call, d_events, &ftrace_event_id_fops,
1035                                 &ftrace_enable_fops, &ftrace_event_filter_fops,
1036                                 &ftrace_event_format_fops);
1037         if (!ret)
1038                 list_add(&call->list, &ftrace_events);
1039
1040         return ret;
1041 }
1042
1043 /* Add an additional event_call dynamically */
1044 int trace_add_event_call(struct ftrace_event_call *call)
1045 {
1046         int ret;
1047         mutex_lock(&event_mutex);
1048         ret = __trace_add_event_call(call);
1049         mutex_unlock(&event_mutex);
1050         return ret;
1051 }
1052
1053 static void remove_subsystem_dir(const char *name)
1054 {
1055         struct event_subsystem *system;
1056
1057         if (strcmp(name, TRACE_SYSTEM) == 0)
1058                 return;
1059
1060         list_for_each_entry(system, &event_subsystems, list) {
1061                 if (strcmp(system->name, name) == 0) {
1062                         if (!--system->nr_events) {
1063                                 struct event_filter *filter = system->filter;
1064
1065                                 debugfs_remove_recursive(system->entry);
1066                                 list_del(&system->list);
1067                                 if (filter) {
1068                                         kfree(filter->filter_string);
1069                                         kfree(filter);
1070                                 }
1071                                 kfree(system->name);
1072                                 kfree(system);
1073                         }
1074                         break;
1075                 }
1076         }
1077 }
1078
1079 /*
1080  * Must be called under locking both of event_mutex and trace_event_mutex.
1081  */
1082 static void __trace_remove_event_call(struct ftrace_event_call *call)
1083 {
1084         ftrace_event_enable_disable(call, 0);
1085         if (call->event.funcs)
1086                 __unregister_ftrace_event(&call->event);
1087         debugfs_remove_recursive(call->dir);
1088         list_del(&call->list);
1089         trace_destroy_fields(call);
1090         destroy_preds(call);
1091         remove_subsystem_dir(call->class->system);
1092 }
1093
1094 /* Remove an event_call */
1095 void trace_remove_event_call(struct ftrace_event_call *call)
1096 {
1097         mutex_lock(&event_mutex);
1098         down_write(&trace_event_mutex);
1099         __trace_remove_event_call(call);
1100         up_write(&trace_event_mutex);
1101         mutex_unlock(&event_mutex);
1102 }
1103
1104 #define for_each_event(event, start, end)                       \
1105         for (event = start;                                     \
1106              (unsigned long)event < (unsigned long)end;         \
1107              event++)
1108
1109 #ifdef CONFIG_MODULES
1110
1111 static LIST_HEAD(ftrace_module_file_list);
1112
1113 /*
1114  * Modules must own their file_operations to keep up with
1115  * reference counting.
1116  */
1117 struct ftrace_module_file_ops {
1118         struct list_head                list;
1119         struct module                   *mod;
1120         struct file_operations          id;
1121         struct file_operations          enable;
1122         struct file_operations          format;
1123         struct file_operations          filter;
1124 };
1125
1126 static struct ftrace_module_file_ops *
1127 trace_create_file_ops(struct module *mod)
1128 {
1129         struct ftrace_module_file_ops *file_ops;
1130
1131         /*
1132          * This is a bit of a PITA. To allow for correct reference
1133          * counting, modules must "own" their file_operations.
1134          * To do this, we allocate the file operations that will be
1135          * used in the event directory.
1136          */
1137
1138         file_ops = kmalloc(sizeof(*file_ops), GFP_KERNEL);
1139         if (!file_ops)
1140                 return NULL;
1141
1142         file_ops->mod = mod;
1143
1144         file_ops->id = ftrace_event_id_fops;
1145         file_ops->id.owner = mod;
1146
1147         file_ops->enable = ftrace_enable_fops;
1148         file_ops->enable.owner = mod;
1149
1150         file_ops->filter = ftrace_event_filter_fops;
1151         file_ops->filter.owner = mod;
1152
1153         file_ops->format = ftrace_event_format_fops;
1154         file_ops->format.owner = mod;
1155
1156         list_add(&file_ops->list, &ftrace_module_file_list);
1157
1158         return file_ops;
1159 }
1160
1161 static void trace_module_add_events(struct module *mod)
1162 {
1163         struct ftrace_module_file_ops *file_ops = NULL;
1164         struct ftrace_event_call *call, *start, *end;
1165         struct dentry *d_events;
1166         int ret;
1167
1168         start = mod->trace_events;
1169         end = mod->trace_events + mod->num_trace_events;
1170
1171         if (start == end)
1172                 return;
1173
1174         d_events = event_trace_events_dir();
1175         if (!d_events)
1176                 return;
1177
1178         for_each_event(call, start, end) {
1179                 /* The linker may leave blanks */
1180                 if (!call->name)
1181                         continue;
1182                 if (call->class->raw_init) {
1183                         ret = call->class->raw_init(call);
1184                         if (ret < 0) {
1185                                 if (ret != -ENOSYS)
1186                                         pr_warning("Could not initialize trace "
1187                                         "point events/%s\n", call->name);
1188                                 continue;
1189                         }
1190                 }
1191                 /*
1192                  * This module has events, create file ops for this module
1193                  * if not already done.
1194                  */
1195                 if (!file_ops) {
1196                         file_ops = trace_create_file_ops(mod);
1197                         if (!file_ops)
1198                                 return;
1199                 }
1200                 call->mod = mod;
1201                 ret = event_create_dir(call, d_events,
1202                                        &file_ops->id, &file_ops->enable,
1203                                        &file_ops->filter, &file_ops->format);
1204                 if (!ret)
1205                         list_add(&call->list, &ftrace_events);
1206         }
1207 }
1208
1209 static void trace_module_remove_events(struct module *mod)
1210 {
1211         struct ftrace_module_file_ops *file_ops;
1212         struct ftrace_event_call *call, *p;
1213         bool found = false;
1214
1215         down_write(&trace_event_mutex);
1216         list_for_each_entry_safe(call, p, &ftrace_events, list) {
1217                 if (call->mod == mod) {
1218                         found = true;
1219                         __trace_remove_event_call(call);
1220                 }
1221         }
1222
1223         /* Now free the file_operations */
1224         list_for_each_entry(file_ops, &ftrace_module_file_list, list) {
1225                 if (file_ops->mod == mod)
1226                         break;
1227         }
1228         if (&file_ops->list != &ftrace_module_file_list) {
1229                 list_del(&file_ops->list);
1230                 kfree(file_ops);
1231         }
1232
1233         /*
1234          * It is safest to reset the ring buffer if the module being unloaded
1235          * registered any events.
1236          */
1237         if (found)
1238                 tracing_reset_current_online_cpus();
1239         up_write(&trace_event_mutex);
1240 }
1241
1242 static int trace_module_notify(struct notifier_block *self,
1243                                unsigned long val, void *data)
1244 {
1245         struct module *mod = data;
1246
1247         mutex_lock(&event_mutex);
1248         switch (val) {
1249         case MODULE_STATE_COMING:
1250                 trace_module_add_events(mod);
1251                 break;
1252         case MODULE_STATE_GOING:
1253                 trace_module_remove_events(mod);
1254                 break;
1255         }
1256         mutex_unlock(&event_mutex);
1257
1258         return 0;
1259 }
1260 #else
1261 static int trace_module_notify(struct notifier_block *self,
1262                                unsigned long val, void *data)
1263 {
1264         return 0;
1265 }
1266 #endif /* CONFIG_MODULES */
1267
1268 static struct notifier_block trace_module_nb = {
1269         .notifier_call = trace_module_notify,
1270         .priority = 0,
1271 };
1272
1273 extern struct ftrace_event_call __start_ftrace_events[];
1274 extern struct ftrace_event_call __stop_ftrace_events[];
1275
1276 static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
1277
1278 static __init int setup_trace_event(char *str)
1279 {
1280         strlcpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
1281         ring_buffer_expanded = 1;
1282         tracing_selftest_disabled = 1;
1283
1284         return 1;
1285 }
1286 __setup("trace_event=", setup_trace_event);
1287
1288 static __init int event_trace_init(void)
1289 {
1290         struct ftrace_event_call *call;
1291         struct dentry *d_tracer;
1292         struct dentry *entry;
1293         struct dentry *d_events;
1294         int ret;
1295         char *buf = bootup_event_buf;
1296         char *token;
1297
1298         d_tracer = tracing_init_dentry();
1299         if (!d_tracer)
1300                 return 0;
1301
1302         entry = debugfs_create_file("available_events", 0444, d_tracer,
1303                                     (void *)&show_event_seq_ops,
1304                                     &ftrace_avail_fops);
1305         if (!entry)
1306                 pr_warning("Could not create debugfs "
1307                            "'available_events' entry\n");
1308
1309         entry = debugfs_create_file("set_event", 0644, d_tracer,
1310                                     (void *)&show_set_event_seq_ops,
1311                                     &ftrace_set_event_fops);
1312         if (!entry)
1313                 pr_warning("Could not create debugfs "
1314                            "'set_event' entry\n");
1315
1316         d_events = event_trace_events_dir();
1317         if (!d_events)
1318                 return 0;
1319
1320         /* ring buffer internal formats */
1321         trace_create_file("header_page", 0444, d_events,
1322                           ring_buffer_print_page_header,
1323                           &ftrace_show_header_fops);
1324
1325         trace_create_file("header_event", 0444, d_events,
1326                           ring_buffer_print_entry_header,
1327                           &ftrace_show_header_fops);
1328
1329         trace_create_file("enable", 0644, d_events,
1330                           NULL, &ftrace_system_enable_fops);
1331
1332         if (trace_define_common_fields())
1333                 pr_warning("tracing: Failed to allocate common fields");
1334
1335         for_each_event(call, __start_ftrace_events, __stop_ftrace_events) {
1336                 /* The linker may leave blanks */
1337                 if (!call->name)
1338                         continue;
1339                 if (call->class->raw_init) {
1340                         ret = call->class->raw_init(call);
1341                         if (ret < 0) {
1342                                 if (ret != -ENOSYS)
1343                                         pr_warning("Could not initialize trace "
1344                                         "point events/%s\n", call->name);
1345                                 continue;
1346                         }
1347                 }
1348                 ret = event_create_dir(call, d_events, &ftrace_event_id_fops,
1349                                        &ftrace_enable_fops,
1350                                        &ftrace_event_filter_fops,
1351                                        &ftrace_event_format_fops);
1352                 if (!ret)
1353                         list_add(&call->list, &ftrace_events);
1354         }
1355
1356         while (true) {
1357                 token = strsep(&buf, ",");
1358
1359                 if (!token)
1360                         break;
1361                 if (!*token)
1362                         continue;
1363
1364                 ret = ftrace_set_clr_event(token, 1);
1365                 if (ret)
1366                         pr_warning("Failed to enable trace event: %s\n", token);
1367         }
1368
1369         ret = register_module_notifier(&trace_module_nb);
1370         if (ret)
1371                 pr_warning("Failed to register trace events module notifier\n");
1372
1373         return 0;
1374 }
1375 fs_initcall(event_trace_init);
1376
1377 #ifdef CONFIG_FTRACE_STARTUP_TEST
1378
1379 static DEFINE_SPINLOCK(test_spinlock);
1380 static DEFINE_SPINLOCK(test_spinlock_irq);
1381 static DEFINE_MUTEX(test_mutex);
1382
1383 static __init void test_work(struct work_struct *dummy)
1384 {
1385         spin_lock(&test_spinlock);
1386         spin_lock_irq(&test_spinlock_irq);
1387         udelay(1);
1388         spin_unlock_irq(&test_spinlock_irq);
1389         spin_unlock(&test_spinlock);
1390
1391         mutex_lock(&test_mutex);
1392         msleep(1);
1393         mutex_unlock(&test_mutex);
1394 }
1395
1396 static __init int event_test_thread(void *unused)
1397 {
1398         void *test_malloc;
1399
1400         test_malloc = kmalloc(1234, GFP_KERNEL);
1401         if (!test_malloc)
1402                 pr_info("failed to kmalloc\n");
1403
1404         schedule_on_each_cpu(test_work);
1405
1406         kfree(test_malloc);
1407
1408         set_current_state(TASK_INTERRUPTIBLE);
1409         while (!kthread_should_stop())
1410                 schedule();
1411
1412         return 0;
1413 }
1414
1415 /*
1416  * Do various things that may trigger events.
1417  */
1418 static __init void event_test_stuff(void)
1419 {
1420         struct task_struct *test_thread;
1421
1422         test_thread = kthread_run(event_test_thread, NULL, "test-events");
1423         msleep(1);
1424         kthread_stop(test_thread);
1425 }
1426
1427 /*
1428  * For every trace event defined, we will test each trace point separately,
1429  * and then by groups, and finally all trace points.
1430  */
1431 static __init void event_trace_self_tests(void)
1432 {
1433         struct ftrace_event_call *call;
1434         struct event_subsystem *system;
1435         int ret;
1436
1437         pr_info("Running tests on trace events:\n");
1438
1439         list_for_each_entry(call, &ftrace_events, list) {
1440
1441                 /* Only test those that have a probe */
1442                 if (!call->class || !call->class->probe)
1443                         continue;
1444
1445 /*
1446  * Testing syscall events here is pretty useless, but
1447  * we still do it if configured. But this is time consuming.
1448  * What we really need is a user thread to perform the
1449  * syscalls as we test.
1450  */
1451 #ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
1452                 if (call->class->system &&
1453                     strcmp(call->class->system, "syscalls") == 0)
1454                         continue;
1455 #endif
1456
1457                 pr_info("Testing event %s: ", call->name);
1458
1459                 /*
1460                  * If an event is already enabled, someone is using
1461                  * it and the self test should not be on.
1462                  */
1463                 if (call->flags & TRACE_EVENT_FL_ENABLED) {
1464                         pr_warning("Enabled event during self test!\n");
1465                         WARN_ON_ONCE(1);
1466                         continue;
1467                 }
1468
1469                 ftrace_event_enable_disable(call, 1);
1470                 event_test_stuff();
1471                 ftrace_event_enable_disable(call, 0);
1472
1473                 pr_cont("OK\n");
1474         }
1475
1476         /* Now test at the sub system level */
1477
1478         pr_info("Running tests on trace event systems:\n");
1479
1480         list_for_each_entry(system, &event_subsystems, list) {
1481
1482                 /* the ftrace system is special, skip it */
1483                 if (strcmp(system->name, "ftrace") == 0)
1484                         continue;
1485
1486                 pr_info("Testing event system %s: ", system->name);
1487
1488                 ret = __ftrace_set_clr_event(NULL, system->name, NULL, 1);
1489                 if (WARN_ON_ONCE(ret)) {
1490                         pr_warning("error enabling system %s\n",
1491                                    system->name);
1492                         continue;
1493                 }
1494
1495                 event_test_stuff();
1496
1497                 ret = __ftrace_set_clr_event(NULL, system->name, NULL, 0);
1498                 if (WARN_ON_ONCE(ret))
1499                         pr_warning("error disabling system %s\n",
1500                                    system->name);
1501
1502                 pr_cont("OK\n");
1503         }
1504
1505         /* Test with all events enabled */
1506
1507         pr_info("Running tests on all trace events:\n");
1508         pr_info("Testing all events: ");
1509
1510         ret = __ftrace_set_clr_event(NULL, NULL, NULL, 1);
1511         if (WARN_ON_ONCE(ret)) {
1512                 pr_warning("error enabling all events\n");
1513                 return;
1514         }
1515
1516         event_test_stuff();
1517
1518         /* reset sysname */
1519         ret = __ftrace_set_clr_event(NULL, NULL, NULL, 0);
1520         if (WARN_ON_ONCE(ret)) {
1521                 pr_warning("error disabling all events\n");
1522                 return;
1523         }
1524
1525         pr_cont("OK\n");
1526 }
1527
1528 #ifdef CONFIG_FUNCTION_TRACER
1529
1530 static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
1531
1532 static void
1533 function_test_events_call(unsigned long ip, unsigned long parent_ip)
1534 {
1535         struct ring_buffer_event *event;
1536         struct ring_buffer *buffer;
1537         struct ftrace_entry *entry;
1538         unsigned long flags;
1539         long disabled;
1540         int cpu;
1541         int pc;
1542
1543         pc = preempt_count();
1544         preempt_disable_notrace();
1545         cpu = raw_smp_processor_id();
1546         disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
1547
1548         if (disabled != 1)
1549                 goto out;
1550
1551         local_save_flags(flags);
1552
1553         event = trace_current_buffer_lock_reserve(&buffer,
1554                                                   TRACE_FN, sizeof(*entry),
1555                                                   flags, pc);
1556         if (!event)
1557                 goto out;
1558         entry   = ring_buffer_event_data(event);
1559         entry->ip                       = ip;
1560         entry->parent_ip                = parent_ip;
1561
1562         trace_nowake_buffer_unlock_commit(buffer, event, flags, pc);
1563
1564  out:
1565         atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
1566         preempt_enable_notrace();
1567 }
1568
1569 static struct ftrace_ops trace_ops __initdata  =
1570 {
1571         .func = function_test_events_call,
1572 };
1573
1574 static __init void event_trace_self_test_with_function(void)
1575 {
1576         register_ftrace_function(&trace_ops);
1577         pr_info("Running tests again, along with the function tracer\n");
1578         event_trace_self_tests();
1579         unregister_ftrace_function(&trace_ops);
1580 }
1581 #else
1582 static __init void event_trace_self_test_with_function(void)
1583 {
1584 }
1585 #endif
1586
1587 static __init int event_trace_self_tests_init(void)
1588 {
1589         if (!tracing_selftest_disabled) {
1590                 event_trace_self_tests();
1591                 event_trace_self_test_with_function();
1592         }
1593
1594         return 0;
1595 }
1596
1597 late_initcall(event_trace_self_tests_init);
1598
1599 #endif