ring-buffer: Make benchmark handle missed events
[linux-2.6.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
32 int trace_define_field(struct ftrace_event_call *call, const char *type,
33                        const char *name, int offset, int size, int is_signed,
34                        int filter_type)
35 {
36         struct ftrace_event_field *field;
37
38         field = kzalloc(sizeof(*field), GFP_KERNEL);
39         if (!field)
40                 goto err;
41
42         field->name = kstrdup(name, GFP_KERNEL);
43         if (!field->name)
44                 goto err;
45
46         field->type = kstrdup(type, GFP_KERNEL);
47         if (!field->type)
48                 goto err;
49
50         if (filter_type == FILTER_OTHER)
51                 field->filter_type = filter_assign_type(type);
52         else
53                 field->filter_type = filter_type;
54
55         field->offset = offset;
56         field->size = size;
57         field->is_signed = is_signed;
58
59         list_add(&field->link, &call->fields);
60
61         return 0;
62
63 err:
64         if (field)
65                 kfree(field->name);
66         kfree(field);
67
68         return -ENOMEM;
69 }
70 EXPORT_SYMBOL_GPL(trace_define_field);
71
72 #define __common_field(type, item)                                      \
73         ret = trace_define_field(call, #type, "common_" #item,          \
74                                  offsetof(typeof(ent), item),           \
75                                  sizeof(ent.item),                      \
76                                  is_signed_type(type), FILTER_OTHER);   \
77         if (ret)                                                        \
78                 return ret;
79
80 static int trace_define_common_fields(struct ftrace_event_call *call)
81 {
82         int ret;
83         struct trace_entry ent;
84
85         __common_field(unsigned short, type);
86         __common_field(unsigned char, flags);
87         __common_field(unsigned char, preempt_count);
88         __common_field(int, pid);
89         __common_field(int, lock_depth);
90
91         return ret;
92 }
93
94 void trace_destroy_fields(struct ftrace_event_call *call)
95 {
96         struct ftrace_event_field *field, *next;
97
98         list_for_each_entry_safe(field, next, &call->fields, link) {
99                 list_del(&field->link);
100                 kfree(field->type);
101                 kfree(field->name);
102                 kfree(field);
103         }
104 }
105
106 int trace_event_raw_init(struct ftrace_event_call *call)
107 {
108         int id;
109
110         id = register_ftrace_event(call->event);
111         if (!id)
112                 return -ENODEV;
113         call->id = id;
114         INIT_LIST_HEAD(&call->fields);
115
116         return 0;
117 }
118 EXPORT_SYMBOL_GPL(trace_event_raw_init);
119
120 static int ftrace_event_enable_disable(struct ftrace_event_call *call,
121                                         int enable)
122 {
123         int ret = 0;
124
125         switch (enable) {
126         case 0:
127                 if (call->enabled) {
128                         call->enabled = 0;
129                         tracing_stop_cmdline_record();
130                         call->unregfunc(call);
131                 }
132                 break;
133         case 1:
134                 if (!call->enabled) {
135                         tracing_start_cmdline_record();
136                         ret = call->regfunc(call);
137                         if (ret) {
138                                 tracing_stop_cmdline_record();
139                                 pr_info("event trace: Could not enable event "
140                                         "%s\n", call->name);
141                                 break;
142                         }
143                         call->enabled = 1;
144                 }
145                 break;
146         }
147
148         return ret;
149 }
150
151 static void ftrace_clear_events(void)
152 {
153         struct ftrace_event_call *call;
154
155         mutex_lock(&event_mutex);
156         list_for_each_entry(call, &ftrace_events, list) {
157                 ftrace_event_enable_disable(call, 0);
158         }
159         mutex_unlock(&event_mutex);
160 }
161
162 /*
163  * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
164  */
165 static int __ftrace_set_clr_event(const char *match, const char *sub,
166                                   const char *event, int set)
167 {
168         struct ftrace_event_call *call;
169         int ret = -EINVAL;
170
171         mutex_lock(&event_mutex);
172         list_for_each_entry(call, &ftrace_events, list) {
173
174                 if (!call->name || !call->regfunc)
175                         continue;
176
177                 if (match &&
178                     strcmp(match, call->name) != 0 &&
179                     strcmp(match, call->system) != 0)
180                         continue;
181
182                 if (sub && strcmp(sub, call->system) != 0)
183                         continue;
184
185                 if (event && strcmp(event, call->name) != 0)
186                         continue;
187
188                 ftrace_event_enable_disable(call, set);
189
190                 ret = 0;
191         }
192         mutex_unlock(&event_mutex);
193
194         return ret;
195 }
196
197 static int ftrace_set_clr_event(char *buf, int set)
198 {
199         char *event = NULL, *sub = NULL, *match;
200
201         /*
202          * The buf format can be <subsystem>:<event-name>
203          *  *:<event-name> means any event by that name.
204          *  :<event-name> is the same.
205          *
206          *  <subsystem>:* means all events in that subsystem
207          *  <subsystem>: means the same.
208          *
209          *  <name> (no ':') means all events in a subsystem with
210          *  the name <name> or any event that matches <name>
211          */
212
213         match = strsep(&buf, ":");
214         if (buf) {
215                 sub = match;
216                 event = buf;
217                 match = NULL;
218
219                 if (!strlen(sub) || strcmp(sub, "*") == 0)
220                         sub = NULL;
221                 if (!strlen(event) || strcmp(event, "*") == 0)
222                         event = NULL;
223         }
224
225         return __ftrace_set_clr_event(match, sub, event, set);
226 }
227
228 /**
229  * trace_set_clr_event - enable or disable an event
230  * @system: system name to match (NULL for any system)
231  * @event: event name to match (NULL for all events, within system)
232  * @set: 1 to enable, 0 to disable
233  *
234  * This is a way for other parts of the kernel to enable or disable
235  * event recording.
236  *
237  * Returns 0 on success, -EINVAL if the parameters do not match any
238  * registered events.
239  */
240 int trace_set_clr_event(const char *system, const char *event, int set)
241 {
242         return __ftrace_set_clr_event(NULL, system, event, set);
243 }
244
245 /* 128 should be much more than enough */
246 #define EVENT_BUF_SIZE          127
247
248 static ssize_t
249 ftrace_event_write(struct file *file, const char __user *ubuf,
250                    size_t cnt, loff_t *ppos)
251 {
252         struct trace_parser parser;
253         ssize_t read, ret;
254
255         if (!cnt)
256                 return 0;
257
258         ret = tracing_update_buffers();
259         if (ret < 0)
260                 return ret;
261
262         if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
263                 return -ENOMEM;
264
265         read = trace_get_user(&parser, ubuf, cnt, ppos);
266
267         if (read >= 0 && trace_parser_loaded((&parser))) {
268                 int set = 1;
269
270                 if (*parser.buffer == '!')
271                         set = 0;
272
273                 parser.buffer[parser.idx] = 0;
274
275                 ret = ftrace_set_clr_event(parser.buffer + !set, set);
276                 if (ret)
277                         goto out_put;
278         }
279
280         ret = read;
281
282  out_put:
283         trace_parser_put(&parser);
284
285         return ret;
286 }
287
288 static void *
289 t_next(struct seq_file *m, void *v, loff_t *pos)
290 {
291         struct ftrace_event_call *call = v;
292
293         (*pos)++;
294
295         list_for_each_entry_continue(call, &ftrace_events, list) {
296                 /*
297                  * The ftrace subsystem is for showing formats only.
298                  * They can not be enabled or disabled via the event files.
299                  */
300                 if (call->regfunc)
301                         return call;
302         }
303
304         return NULL;
305 }
306
307 static void *t_start(struct seq_file *m, loff_t *pos)
308 {
309         struct ftrace_event_call *call;
310         loff_t l;
311
312         mutex_lock(&event_mutex);
313
314         call = list_entry(&ftrace_events, struct ftrace_event_call, list);
315         for (l = 0; l <= *pos; ) {
316                 call = t_next(m, call, &l);
317                 if (!call)
318                         break;
319         }
320         return call;
321 }
322
323 static void *
324 s_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                 if (call->enabled)
332                         return call;
333         }
334
335         return NULL;
336 }
337
338 static void *s_start(struct seq_file *m, loff_t *pos)
339 {
340         struct ftrace_event_call *call;
341         loff_t l;
342
343         mutex_lock(&event_mutex);
344
345         call = list_entry(&ftrace_events, struct ftrace_event_call, list);
346         for (l = 0; l <= *pos; ) {
347                 call = s_next(m, call, &l);
348                 if (!call)
349                         break;
350         }
351         return call;
352 }
353
354 static int t_show(struct seq_file *m, void *v)
355 {
356         struct ftrace_event_call *call = v;
357
358         if (strcmp(call->system, TRACE_SYSTEM) != 0)
359                 seq_printf(m, "%s:", call->system);
360         seq_printf(m, "%s\n", call->name);
361
362         return 0;
363 }
364
365 static void t_stop(struct seq_file *m, void *p)
366 {
367         mutex_unlock(&event_mutex);
368 }
369
370 static int
371 ftrace_event_seq_open(struct inode *inode, struct file *file)
372 {
373         const struct seq_operations *seq_ops;
374
375         if ((file->f_mode & FMODE_WRITE) &&
376             (file->f_flags & O_TRUNC))
377                 ftrace_clear_events();
378
379         seq_ops = inode->i_private;
380         return seq_open(file, seq_ops);
381 }
382
383 static ssize_t
384 event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
385                   loff_t *ppos)
386 {
387         struct ftrace_event_call *call = filp->private_data;
388         char *buf;
389
390         if (call->enabled)
391                 buf = "1\n";
392         else
393                 buf = "0\n";
394
395         return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
396 }
397
398 static ssize_t
399 event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
400                    loff_t *ppos)
401 {
402         struct ftrace_event_call *call = filp->private_data;
403         char buf[64];
404         unsigned long val;
405         int ret;
406
407         if (cnt >= sizeof(buf))
408                 return -EINVAL;
409
410         if (copy_from_user(&buf, ubuf, cnt))
411                 return -EFAULT;
412
413         buf[cnt] = 0;
414
415         ret = strict_strtoul(buf, 10, &val);
416         if (ret < 0)
417                 return ret;
418
419         ret = tracing_update_buffers();
420         if (ret < 0)
421                 return ret;
422
423         switch (val) {
424         case 0:
425         case 1:
426                 mutex_lock(&event_mutex);
427                 ret = ftrace_event_enable_disable(call, val);
428                 mutex_unlock(&event_mutex);
429                 break;
430
431         default:
432                 return -EINVAL;
433         }
434
435         *ppos += cnt;
436
437         return ret ? ret : cnt;
438 }
439
440 static ssize_t
441 system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
442                    loff_t *ppos)
443 {
444         const char set_to_char[4] = { '?', '0', '1', 'X' };
445         const char *system = filp->private_data;
446         struct ftrace_event_call *call;
447         char buf[2];
448         int set = 0;
449         int ret;
450
451         mutex_lock(&event_mutex);
452         list_for_each_entry(call, &ftrace_events, list) {
453                 if (!call->name || !call->regfunc)
454                         continue;
455
456                 if (system && strcmp(call->system, system) != 0)
457                         continue;
458
459                 /*
460                  * We need to find out if all the events are set
461                  * or if all events or cleared, or if we have
462                  * a mixture.
463                  */
464                 set |= (1 << !!call->enabled);
465
466                 /*
467                  * If we have a mixture, no need to look further.
468                  */
469                 if (set == 3)
470                         break;
471         }
472         mutex_unlock(&event_mutex);
473
474         buf[0] = set_to_char[set];
475         buf[1] = '\n';
476
477         ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
478
479         return ret;
480 }
481
482 static ssize_t
483 system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
484                     loff_t *ppos)
485 {
486         const char *system = filp->private_data;
487         unsigned long val;
488         char buf[64];
489         ssize_t ret;
490
491         if (cnt >= sizeof(buf))
492                 return -EINVAL;
493
494         if (copy_from_user(&buf, ubuf, cnt))
495                 return -EFAULT;
496
497         buf[cnt] = 0;
498
499         ret = strict_strtoul(buf, 10, &val);
500         if (ret < 0)
501                 return ret;
502
503         ret = tracing_update_buffers();
504         if (ret < 0)
505                 return ret;
506
507         if (val != 0 && val != 1)
508                 return -EINVAL;
509
510         ret = __ftrace_set_clr_event(NULL, system, NULL, val);
511         if (ret)
512                 goto out;
513
514         ret = cnt;
515
516 out:
517         *ppos += cnt;
518
519         return ret;
520 }
521
522 static ssize_t
523 event_format_read(struct file *filp, char __user *ubuf, size_t cnt,
524                   loff_t *ppos)
525 {
526         struct ftrace_event_call *call = filp->private_data;
527         struct ftrace_event_field *field;
528         struct trace_seq *s;
529         int common_field_count = 5;
530         char *buf;
531         int r = 0;
532
533         if (*ppos)
534                 return 0;
535
536         s = kmalloc(sizeof(*s), GFP_KERNEL);
537         if (!s)
538                 return -ENOMEM;
539
540         trace_seq_init(s);
541
542         trace_seq_printf(s, "name: %s\n", call->name);
543         trace_seq_printf(s, "ID: %d\n", call->id);
544         trace_seq_printf(s, "format:\n");
545
546         list_for_each_entry_reverse(field, &call->fields, link) {
547                 /*
548                  * Smartly shows the array type(except dynamic array).
549                  * Normal:
550                  *      field:TYPE VAR
551                  * If TYPE := TYPE[LEN], it is shown:
552                  *      field:TYPE VAR[LEN]
553                  */
554                 const char *array_descriptor = strchr(field->type, '[');
555
556                 if (!strncmp(field->type, "__data_loc", 10))
557                         array_descriptor = NULL;
558
559                 if (!array_descriptor) {
560                         r = trace_seq_printf(s, "\tfield:%s %s;\toffset:%u;"
561                                         "\tsize:%u;\tsigned:%d;\n",
562                                         field->type, field->name, field->offset,
563                                         field->size, !!field->is_signed);
564                 } else {
565                         r = trace_seq_printf(s, "\tfield:%.*s %s%s;\toffset:%u;"
566                                         "\tsize:%u;\tsigned:%d;\n",
567                                         (int)(array_descriptor - field->type),
568                                         field->type, field->name,
569                                         array_descriptor, field->offset,
570                                         field->size, !!field->is_signed);
571                 }
572
573                 if (--common_field_count == 0)
574                         r = trace_seq_printf(s, "\n");
575
576                 if (!r)
577                         break;
578         }
579
580         if (r)
581                 r = trace_seq_printf(s, "\nprint fmt: %s\n",
582                                 call->print_fmt);
583
584         if (!r) {
585                 /*
586                  * ug!  The format output is bigger than a PAGE!!
587                  */
588                 buf = "FORMAT TOO BIG\n";
589                 r = simple_read_from_buffer(ubuf, cnt, ppos,
590                                               buf, strlen(buf));
591                 goto out;
592         }
593
594         r = simple_read_from_buffer(ubuf, cnt, ppos,
595                                     s->buffer, s->len);
596  out:
597         kfree(s);
598         return r;
599 }
600
601 static ssize_t
602 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
603 {
604         struct ftrace_event_call *call = filp->private_data;
605         struct trace_seq *s;
606         int r;
607
608         if (*ppos)
609                 return 0;
610
611         s = kmalloc(sizeof(*s), GFP_KERNEL);
612         if (!s)
613                 return -ENOMEM;
614
615         trace_seq_init(s);
616         trace_seq_printf(s, "%d\n", call->id);
617
618         r = simple_read_from_buffer(ubuf, cnt, ppos,
619                                     s->buffer, s->len);
620         kfree(s);
621         return r;
622 }
623
624 static ssize_t
625 event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
626                   loff_t *ppos)
627 {
628         struct ftrace_event_call *call = filp->private_data;
629         struct trace_seq *s;
630         int r;
631
632         if (*ppos)
633                 return 0;
634
635         s = kmalloc(sizeof(*s), GFP_KERNEL);
636         if (!s)
637                 return -ENOMEM;
638
639         trace_seq_init(s);
640
641         print_event_filter(call, s);
642         r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
643
644         kfree(s);
645
646         return r;
647 }
648
649 static ssize_t
650 event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
651                    loff_t *ppos)
652 {
653         struct ftrace_event_call *call = filp->private_data;
654         char *buf;
655         int err;
656
657         if (cnt >= PAGE_SIZE)
658                 return -EINVAL;
659
660         buf = (char *)__get_free_page(GFP_TEMPORARY);
661         if (!buf)
662                 return -ENOMEM;
663
664         if (copy_from_user(buf, ubuf, cnt)) {
665                 free_page((unsigned long) buf);
666                 return -EFAULT;
667         }
668         buf[cnt] = '\0';
669
670         err = apply_event_filter(call, buf);
671         free_page((unsigned long) buf);
672         if (err < 0)
673                 return err;
674
675         *ppos += cnt;
676
677         return cnt;
678 }
679
680 static ssize_t
681 subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
682                       loff_t *ppos)
683 {
684         struct event_subsystem *system = filp->private_data;
685         struct trace_seq *s;
686         int r;
687
688         if (*ppos)
689                 return 0;
690
691         s = kmalloc(sizeof(*s), GFP_KERNEL);
692         if (!s)
693                 return -ENOMEM;
694
695         trace_seq_init(s);
696
697         print_subsystem_event_filter(system, s);
698         r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
699
700         kfree(s);
701
702         return r;
703 }
704
705 static ssize_t
706 subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
707                        loff_t *ppos)
708 {
709         struct event_subsystem *system = filp->private_data;
710         char *buf;
711         int err;
712
713         if (cnt >= PAGE_SIZE)
714                 return -EINVAL;
715
716         buf = (char *)__get_free_page(GFP_TEMPORARY);
717         if (!buf)
718                 return -ENOMEM;
719
720         if (copy_from_user(buf, ubuf, cnt)) {
721                 free_page((unsigned long) buf);
722                 return -EFAULT;
723         }
724         buf[cnt] = '\0';
725
726         err = apply_subsystem_event_filter(system, buf);
727         free_page((unsigned long) buf);
728         if (err < 0)
729                 return err;
730
731         *ppos += cnt;
732
733         return cnt;
734 }
735
736 static ssize_t
737 show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
738 {
739         int (*func)(struct trace_seq *s) = filp->private_data;
740         struct trace_seq *s;
741         int r;
742
743         if (*ppos)
744                 return 0;
745
746         s = kmalloc(sizeof(*s), GFP_KERNEL);
747         if (!s)
748                 return -ENOMEM;
749
750         trace_seq_init(s);
751
752         func(s);
753         r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
754
755         kfree(s);
756
757         return r;
758 }
759
760 static const struct seq_operations show_event_seq_ops = {
761         .start = t_start,
762         .next = t_next,
763         .show = t_show,
764         .stop = t_stop,
765 };
766
767 static const struct seq_operations show_set_event_seq_ops = {
768         .start = s_start,
769         .next = s_next,
770         .show = t_show,
771         .stop = t_stop,
772 };
773
774 static const struct file_operations ftrace_avail_fops = {
775         .open = ftrace_event_seq_open,
776         .read = seq_read,
777         .llseek = seq_lseek,
778         .release = seq_release,
779 };
780
781 static const struct file_operations ftrace_set_event_fops = {
782         .open = ftrace_event_seq_open,
783         .read = seq_read,
784         .write = ftrace_event_write,
785         .llseek = seq_lseek,
786         .release = seq_release,
787 };
788
789 static const struct file_operations ftrace_enable_fops = {
790         .open = tracing_open_generic,
791         .read = event_enable_read,
792         .write = event_enable_write,
793 };
794
795 static const struct file_operations ftrace_event_format_fops = {
796         .open = tracing_open_generic,
797         .read = event_format_read,
798 };
799
800 static const struct file_operations ftrace_event_id_fops = {
801         .open = tracing_open_generic,
802         .read = event_id_read,
803 };
804
805 static const struct file_operations ftrace_event_filter_fops = {
806         .open = tracing_open_generic,
807         .read = event_filter_read,
808         .write = event_filter_write,
809 };
810
811 static const struct file_operations ftrace_subsystem_filter_fops = {
812         .open = tracing_open_generic,
813         .read = subsystem_filter_read,
814         .write = subsystem_filter_write,
815 };
816
817 static const struct file_operations ftrace_system_enable_fops = {
818         .open = tracing_open_generic,
819         .read = system_enable_read,
820         .write = system_enable_write,
821 };
822
823 static const struct file_operations ftrace_show_header_fops = {
824         .open = tracing_open_generic,
825         .read = show_header,
826 };
827
828 static struct dentry *event_trace_events_dir(void)
829 {
830         static struct dentry *d_tracer;
831         static struct dentry *d_events;
832
833         if (d_events)
834                 return d_events;
835
836         d_tracer = tracing_init_dentry();
837         if (!d_tracer)
838                 return NULL;
839
840         d_events = debugfs_create_dir("events", d_tracer);
841         if (!d_events)
842                 pr_warning("Could not create debugfs "
843                            "'events' directory\n");
844
845         return d_events;
846 }
847
848 static LIST_HEAD(event_subsystems);
849
850 static struct dentry *
851 event_subsystem_dir(const char *name, struct dentry *d_events)
852 {
853         struct event_subsystem *system;
854         struct dentry *entry;
855
856         /* First see if we did not already create this dir */
857         list_for_each_entry(system, &event_subsystems, list) {
858                 if (strcmp(system->name, name) == 0) {
859                         system->nr_events++;
860                         return system->entry;
861                 }
862         }
863
864         /* need to create new entry */
865         system = kmalloc(sizeof(*system), GFP_KERNEL);
866         if (!system) {
867                 pr_warning("No memory to create event subsystem %s\n",
868                            name);
869                 return d_events;
870         }
871
872         system->entry = debugfs_create_dir(name, d_events);
873         if (!system->entry) {
874                 pr_warning("Could not create event subsystem %s\n",
875                            name);
876                 kfree(system);
877                 return d_events;
878         }
879
880         system->nr_events = 1;
881         system->name = kstrdup(name, GFP_KERNEL);
882         if (!system->name) {
883                 debugfs_remove(system->entry);
884                 kfree(system);
885                 return d_events;
886         }
887
888         list_add(&system->list, &event_subsystems);
889
890         system->filter = NULL;
891
892         system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
893         if (!system->filter) {
894                 pr_warning("Could not allocate filter for subsystem "
895                            "'%s'\n", name);
896                 return system->entry;
897         }
898
899         entry = debugfs_create_file("filter", 0644, system->entry, system,
900                                     &ftrace_subsystem_filter_fops);
901         if (!entry) {
902                 kfree(system->filter);
903                 system->filter = NULL;
904                 pr_warning("Could not create debugfs "
905                            "'%s/filter' entry\n", name);
906         }
907
908         trace_create_file("enable", 0644, system->entry,
909                           (void *)system->name,
910                           &ftrace_system_enable_fops);
911
912         return system->entry;
913 }
914
915 static int
916 event_create_dir(struct ftrace_event_call *call, struct dentry *d_events,
917                  const struct file_operations *id,
918                  const struct file_operations *enable,
919                  const struct file_operations *filter,
920                  const struct file_operations *format)
921 {
922         int ret;
923
924         /*
925          * If the trace point header did not define TRACE_SYSTEM
926          * then the system would be called "TRACE_SYSTEM".
927          */
928         if (strcmp(call->system, TRACE_SYSTEM) != 0)
929                 d_events = event_subsystem_dir(call->system, d_events);
930
931         call->dir = debugfs_create_dir(call->name, d_events);
932         if (!call->dir) {
933                 pr_warning("Could not create debugfs "
934                            "'%s' directory\n", call->name);
935                 return -1;
936         }
937
938         if (call->regfunc)
939                 trace_create_file("enable", 0644, call->dir, call,
940                                   enable);
941
942         if (call->id && call->perf_event_enable)
943                 trace_create_file("id", 0444, call->dir, call,
944                                   id);
945
946         if (call->define_fields) {
947                 ret = trace_define_common_fields(call);
948                 if (!ret)
949                         ret = call->define_fields(call);
950                 if (ret < 0) {
951                         pr_warning("Could not initialize trace point"
952                                    " events/%s\n", call->name);
953                         return ret;
954                 }
955                 trace_create_file("filter", 0644, call->dir, call,
956                                   filter);
957         }
958
959         trace_create_file("format", 0444, call->dir, call,
960                           format);
961
962         return 0;
963 }
964
965 static int __trace_add_event_call(struct ftrace_event_call *call)
966 {
967         struct dentry *d_events;
968         int ret;
969
970         if (!call->name)
971                 return -EINVAL;
972
973         if (call->raw_init) {
974                 ret = call->raw_init(call);
975                 if (ret < 0) {
976                         if (ret != -ENOSYS)
977                                 pr_warning("Could not initialize trace "
978                                 "events/%s\n", call->name);
979                         return ret;
980                 }
981         }
982
983         d_events = event_trace_events_dir();
984         if (!d_events)
985                 return -ENOENT;
986
987         ret = event_create_dir(call, d_events, &ftrace_event_id_fops,
988                                 &ftrace_enable_fops, &ftrace_event_filter_fops,
989                                 &ftrace_event_format_fops);
990         if (!ret)
991                 list_add(&call->list, &ftrace_events);
992
993         return ret;
994 }
995
996 /* Add an additional event_call dynamically */
997 int trace_add_event_call(struct ftrace_event_call *call)
998 {
999         int ret;
1000         mutex_lock(&event_mutex);
1001         ret = __trace_add_event_call(call);
1002         mutex_unlock(&event_mutex);
1003         return ret;
1004 }
1005
1006 static void remove_subsystem_dir(const char *name)
1007 {
1008         struct event_subsystem *system;
1009
1010         if (strcmp(name, TRACE_SYSTEM) == 0)
1011                 return;
1012
1013         list_for_each_entry(system, &event_subsystems, list) {
1014                 if (strcmp(system->name, name) == 0) {
1015                         if (!--system->nr_events) {
1016                                 struct event_filter *filter = system->filter;
1017
1018                                 debugfs_remove_recursive(system->entry);
1019                                 list_del(&system->list);
1020                                 if (filter) {
1021                                         kfree(filter->filter_string);
1022                                         kfree(filter);
1023                                 }
1024                                 kfree(system->name);
1025                                 kfree(system);
1026                         }
1027                         break;
1028                 }
1029         }
1030 }
1031
1032 /*
1033  * Must be called under locking both of event_mutex and trace_event_mutex.
1034  */
1035 static void __trace_remove_event_call(struct ftrace_event_call *call)
1036 {
1037         ftrace_event_enable_disable(call, 0);
1038         if (call->event)
1039                 __unregister_ftrace_event(call->event);
1040         debugfs_remove_recursive(call->dir);
1041         list_del(&call->list);
1042         trace_destroy_fields(call);
1043         destroy_preds(call);
1044         remove_subsystem_dir(call->system);
1045 }
1046
1047 /* Remove an event_call */
1048 void trace_remove_event_call(struct ftrace_event_call *call)
1049 {
1050         mutex_lock(&event_mutex);
1051         down_write(&trace_event_mutex);
1052         __trace_remove_event_call(call);
1053         up_write(&trace_event_mutex);
1054         mutex_unlock(&event_mutex);
1055 }
1056
1057 #define for_each_event(event, start, end)                       \
1058         for (event = start;                                     \
1059              (unsigned long)event < (unsigned long)end;         \
1060              event++)
1061
1062 #ifdef CONFIG_MODULES
1063
1064 static LIST_HEAD(ftrace_module_file_list);
1065
1066 /*
1067  * Modules must own their file_operations to keep up with
1068  * reference counting.
1069  */
1070 struct ftrace_module_file_ops {
1071         struct list_head                list;
1072         struct module                   *mod;
1073         struct file_operations          id;
1074         struct file_operations          enable;
1075         struct file_operations          format;
1076         struct file_operations          filter;
1077 };
1078
1079 static struct ftrace_module_file_ops *
1080 trace_create_file_ops(struct module *mod)
1081 {
1082         struct ftrace_module_file_ops *file_ops;
1083
1084         /*
1085          * This is a bit of a PITA. To allow for correct reference
1086          * counting, modules must "own" their file_operations.
1087          * To do this, we allocate the file operations that will be
1088          * used in the event directory.
1089          */
1090
1091         file_ops = kmalloc(sizeof(*file_ops), GFP_KERNEL);
1092         if (!file_ops)
1093                 return NULL;
1094
1095         file_ops->mod = mod;
1096
1097         file_ops->id = ftrace_event_id_fops;
1098         file_ops->id.owner = mod;
1099
1100         file_ops->enable = ftrace_enable_fops;
1101         file_ops->enable.owner = mod;
1102
1103         file_ops->filter = ftrace_event_filter_fops;
1104         file_ops->filter.owner = mod;
1105
1106         file_ops->format = ftrace_event_format_fops;
1107         file_ops->format.owner = mod;
1108
1109         list_add(&file_ops->list, &ftrace_module_file_list);
1110
1111         return file_ops;
1112 }
1113
1114 static void trace_module_add_events(struct module *mod)
1115 {
1116         struct ftrace_module_file_ops *file_ops = NULL;
1117         struct ftrace_event_call *call, *start, *end;
1118         struct dentry *d_events;
1119         int ret;
1120
1121         start = mod->trace_events;
1122         end = mod->trace_events + mod->num_trace_events;
1123
1124         if (start == end)
1125                 return;
1126
1127         d_events = event_trace_events_dir();
1128         if (!d_events)
1129                 return;
1130
1131         for_each_event(call, start, end) {
1132                 /* The linker may leave blanks */
1133                 if (!call->name)
1134                         continue;
1135                 if (call->raw_init) {
1136                         ret = call->raw_init(call);
1137                         if (ret < 0) {
1138                                 if (ret != -ENOSYS)
1139                                         pr_warning("Could not initialize trace "
1140                                         "point events/%s\n", call->name);
1141                                 continue;
1142                         }
1143                 }
1144                 /*
1145                  * This module has events, create file ops for this module
1146                  * if not already done.
1147                  */
1148                 if (!file_ops) {
1149                         file_ops = trace_create_file_ops(mod);
1150                         if (!file_ops)
1151                                 return;
1152                 }
1153                 call->mod = mod;
1154                 ret = event_create_dir(call, d_events,
1155                                        &file_ops->id, &file_ops->enable,
1156                                        &file_ops->filter, &file_ops->format);
1157                 if (!ret)
1158                         list_add(&call->list, &ftrace_events);
1159         }
1160 }
1161
1162 static void trace_module_remove_events(struct module *mod)
1163 {
1164         struct ftrace_module_file_ops *file_ops;
1165         struct ftrace_event_call *call, *p;
1166         bool found = false;
1167
1168         down_write(&trace_event_mutex);
1169         list_for_each_entry_safe(call, p, &ftrace_events, list) {
1170                 if (call->mod == mod) {
1171                         found = true;
1172                         __trace_remove_event_call(call);
1173                 }
1174         }
1175
1176         /* Now free the file_operations */
1177         list_for_each_entry(file_ops, &ftrace_module_file_list, list) {
1178                 if (file_ops->mod == mod)
1179                         break;
1180         }
1181         if (&file_ops->list != &ftrace_module_file_list) {
1182                 list_del(&file_ops->list);
1183                 kfree(file_ops);
1184         }
1185
1186         /*
1187          * It is safest to reset the ring buffer if the module being unloaded
1188          * registered any events.
1189          */
1190         if (found)
1191                 tracing_reset_current_online_cpus();
1192         up_write(&trace_event_mutex);
1193 }
1194
1195 static int trace_module_notify(struct notifier_block *self,
1196                                unsigned long val, void *data)
1197 {
1198         struct module *mod = data;
1199
1200         mutex_lock(&event_mutex);
1201         switch (val) {
1202         case MODULE_STATE_COMING:
1203                 trace_module_add_events(mod);
1204                 break;
1205         case MODULE_STATE_GOING:
1206                 trace_module_remove_events(mod);
1207                 break;
1208         }
1209         mutex_unlock(&event_mutex);
1210
1211         return 0;
1212 }
1213 #else
1214 static int trace_module_notify(struct notifier_block *self,
1215                                unsigned long val, void *data)
1216 {
1217         return 0;
1218 }
1219 #endif /* CONFIG_MODULES */
1220
1221 static struct notifier_block trace_module_nb = {
1222         .notifier_call = trace_module_notify,
1223         .priority = 0,
1224 };
1225
1226 extern struct ftrace_event_call __start_ftrace_events[];
1227 extern struct ftrace_event_call __stop_ftrace_events[];
1228
1229 static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
1230
1231 static __init int setup_trace_event(char *str)
1232 {
1233         strlcpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
1234         ring_buffer_expanded = 1;
1235         tracing_selftest_disabled = 1;
1236
1237         return 1;
1238 }
1239 __setup("trace_event=", setup_trace_event);
1240
1241 static __init int event_trace_init(void)
1242 {
1243         struct ftrace_event_call *call;
1244         struct dentry *d_tracer;
1245         struct dentry *entry;
1246         struct dentry *d_events;
1247         int ret;
1248         char *buf = bootup_event_buf;
1249         char *token;
1250
1251         d_tracer = tracing_init_dentry();
1252         if (!d_tracer)
1253                 return 0;
1254
1255         entry = debugfs_create_file("available_events", 0444, d_tracer,
1256                                     (void *)&show_event_seq_ops,
1257                                     &ftrace_avail_fops);
1258         if (!entry)
1259                 pr_warning("Could not create debugfs "
1260                            "'available_events' entry\n");
1261
1262         entry = debugfs_create_file("set_event", 0644, d_tracer,
1263                                     (void *)&show_set_event_seq_ops,
1264                                     &ftrace_set_event_fops);
1265         if (!entry)
1266                 pr_warning("Could not create debugfs "
1267                            "'set_event' entry\n");
1268
1269         d_events = event_trace_events_dir();
1270         if (!d_events)
1271                 return 0;
1272
1273         /* ring buffer internal formats */
1274         trace_create_file("header_page", 0444, d_events,
1275                           ring_buffer_print_page_header,
1276                           &ftrace_show_header_fops);
1277
1278         trace_create_file("header_event", 0444, d_events,
1279                           ring_buffer_print_entry_header,
1280                           &ftrace_show_header_fops);
1281
1282         trace_create_file("enable", 0644, d_events,
1283                           NULL, &ftrace_system_enable_fops);
1284
1285         for_each_event(call, __start_ftrace_events, __stop_ftrace_events) {
1286                 /* The linker may leave blanks */
1287                 if (!call->name)
1288                         continue;
1289                 if (call->raw_init) {
1290                         ret = call->raw_init(call);
1291                         if (ret < 0) {
1292                                 if (ret != -ENOSYS)
1293                                         pr_warning("Could not initialize trace "
1294                                         "point events/%s\n", call->name);
1295                                 continue;
1296                         }
1297                 }
1298                 ret = event_create_dir(call, d_events, &ftrace_event_id_fops,
1299                                        &ftrace_enable_fops,
1300                                        &ftrace_event_filter_fops,
1301                                        &ftrace_event_format_fops);
1302                 if (!ret)
1303                         list_add(&call->list, &ftrace_events);
1304         }
1305
1306         while (true) {
1307                 token = strsep(&buf, ",");
1308
1309                 if (!token)
1310                         break;
1311                 if (!*token)
1312                         continue;
1313
1314                 ret = ftrace_set_clr_event(token, 1);
1315                 if (ret)
1316                         pr_warning("Failed to enable trace event: %s\n", token);
1317         }
1318
1319         ret = register_module_notifier(&trace_module_nb);
1320         if (ret)
1321                 pr_warning("Failed to register trace events module notifier\n");
1322
1323         return 0;
1324 }
1325 fs_initcall(event_trace_init);
1326
1327 #ifdef CONFIG_FTRACE_STARTUP_TEST
1328
1329 static DEFINE_SPINLOCK(test_spinlock);
1330 static DEFINE_SPINLOCK(test_spinlock_irq);
1331 static DEFINE_MUTEX(test_mutex);
1332
1333 static __init void test_work(struct work_struct *dummy)
1334 {
1335         spin_lock(&test_spinlock);
1336         spin_lock_irq(&test_spinlock_irq);
1337         udelay(1);
1338         spin_unlock_irq(&test_spinlock_irq);
1339         spin_unlock(&test_spinlock);
1340
1341         mutex_lock(&test_mutex);
1342         msleep(1);
1343         mutex_unlock(&test_mutex);
1344 }
1345
1346 static __init int event_test_thread(void *unused)
1347 {
1348         void *test_malloc;
1349
1350         test_malloc = kmalloc(1234, GFP_KERNEL);
1351         if (!test_malloc)
1352                 pr_info("failed to kmalloc\n");
1353
1354         schedule_on_each_cpu(test_work);
1355
1356         kfree(test_malloc);
1357
1358         set_current_state(TASK_INTERRUPTIBLE);
1359         while (!kthread_should_stop())
1360                 schedule();
1361
1362         return 0;
1363 }
1364
1365 /*
1366  * Do various things that may trigger events.
1367  */
1368 static __init void event_test_stuff(void)
1369 {
1370         struct task_struct *test_thread;
1371
1372         test_thread = kthread_run(event_test_thread, NULL, "test-events");
1373         msleep(1);
1374         kthread_stop(test_thread);
1375 }
1376
1377 /*
1378  * For every trace event defined, we will test each trace point separately,
1379  * and then by groups, and finally all trace points.
1380  */
1381 static __init void event_trace_self_tests(void)
1382 {
1383         struct ftrace_event_call *call;
1384         struct event_subsystem *system;
1385         int ret;
1386
1387         pr_info("Running tests on trace events:\n");
1388
1389         list_for_each_entry(call, &ftrace_events, list) {
1390
1391                 /* Only test those that have a regfunc */
1392                 if (!call->regfunc)
1393                         continue;
1394
1395 /*
1396  * Testing syscall events here is pretty useless, but
1397  * we still do it if configured. But this is time consuming.
1398  * What we really need is a user thread to perform the
1399  * syscalls as we test.
1400  */
1401 #ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
1402                 if (call->system &&
1403                     strcmp(call->system, "syscalls") == 0)
1404                         continue;
1405 #endif
1406
1407                 pr_info("Testing event %s: ", call->name);
1408
1409                 /*
1410                  * If an event is already enabled, someone is using
1411                  * it and the self test should not be on.
1412                  */
1413                 if (call->enabled) {
1414                         pr_warning("Enabled event during self test!\n");
1415                         WARN_ON_ONCE(1);
1416                         continue;
1417                 }
1418
1419                 ftrace_event_enable_disable(call, 1);
1420                 event_test_stuff();
1421                 ftrace_event_enable_disable(call, 0);
1422
1423                 pr_cont("OK\n");
1424         }
1425
1426         /* Now test at the sub system level */
1427
1428         pr_info("Running tests on trace event systems:\n");
1429
1430         list_for_each_entry(system, &event_subsystems, list) {
1431
1432                 /* the ftrace system is special, skip it */
1433                 if (strcmp(system->name, "ftrace") == 0)
1434                         continue;
1435
1436                 pr_info("Testing event system %s: ", system->name);
1437
1438                 ret = __ftrace_set_clr_event(NULL, system->name, NULL, 1);
1439                 if (WARN_ON_ONCE(ret)) {
1440                         pr_warning("error enabling system %s\n",
1441                                    system->name);
1442                         continue;
1443                 }
1444
1445                 event_test_stuff();
1446
1447                 ret = __ftrace_set_clr_event(NULL, system->name, NULL, 0);
1448                 if (WARN_ON_ONCE(ret))
1449                         pr_warning("error disabling system %s\n",
1450                                    system->name);
1451
1452                 pr_cont("OK\n");
1453         }
1454
1455         /* Test with all events enabled */
1456
1457         pr_info("Running tests on all trace events:\n");
1458         pr_info("Testing all events: ");
1459
1460         ret = __ftrace_set_clr_event(NULL, NULL, NULL, 1);
1461         if (WARN_ON_ONCE(ret)) {
1462                 pr_warning("error enabling all events\n");
1463                 return;
1464         }
1465
1466         event_test_stuff();
1467
1468         /* reset sysname */
1469         ret = __ftrace_set_clr_event(NULL, NULL, NULL, 0);
1470         if (WARN_ON_ONCE(ret)) {
1471                 pr_warning("error disabling all events\n");
1472                 return;
1473         }
1474
1475         pr_cont("OK\n");
1476 }
1477
1478 #ifdef CONFIG_FUNCTION_TRACER
1479
1480 static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
1481
1482 static void
1483 function_test_events_call(unsigned long ip, unsigned long parent_ip)
1484 {
1485         struct ring_buffer_event *event;
1486         struct ring_buffer *buffer;
1487         struct ftrace_entry *entry;
1488         unsigned long flags;
1489         long disabled;
1490         int resched;
1491         int cpu;
1492         int pc;
1493
1494         pc = preempt_count();
1495         resched = ftrace_preempt_disable();
1496         cpu = raw_smp_processor_id();
1497         disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
1498
1499         if (disabled != 1)
1500                 goto out;
1501
1502         local_save_flags(flags);
1503
1504         event = trace_current_buffer_lock_reserve(&buffer,
1505                                                   TRACE_FN, sizeof(*entry),
1506                                                   flags, pc);
1507         if (!event)
1508                 goto out;
1509         entry   = ring_buffer_event_data(event);
1510         entry->ip                       = ip;
1511         entry->parent_ip                = parent_ip;
1512
1513         trace_nowake_buffer_unlock_commit(buffer, event, flags, pc);
1514
1515  out:
1516         atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
1517         ftrace_preempt_enable(resched);
1518 }
1519
1520 static struct ftrace_ops trace_ops __initdata  =
1521 {
1522         .func = function_test_events_call,
1523 };
1524
1525 static __init void event_trace_self_test_with_function(void)
1526 {
1527         register_ftrace_function(&trace_ops);
1528         pr_info("Running tests again, along with the function tracer\n");
1529         event_trace_self_tests();
1530         unregister_ftrace_function(&trace_ops);
1531 }
1532 #else
1533 static __init void event_trace_self_test_with_function(void)
1534 {
1535 }
1536 #endif
1537
1538 static __init int event_trace_self_tests_init(void)
1539 {
1540         if (!tracing_selftest_disabled) {
1541                 event_trace_self_tests();
1542                 event_trace_self_test_with_function();
1543         }
1544
1545         return 0;
1546 }
1547
1548 late_initcall(event_trace_self_tests_init);
1549
1550 #endif