projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
ARC: support HIGHMEM even without PAE40
[deliverable/linux.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 #define pr_fmt(fmt) fmt
12
13 #include <linux/workqueue.h>
14 #include <linux/spinlock.h>
15 #include <linux/kthread.h>
16 #include <linux/tracefs.h>
17 #include <linux/uaccess.h>
18 #include <linux/bsearch.h>
19 #include <linux/module.h>
20 #include <linux/ctype.h>
21 #include <linux/sort.h>
22 #include <linux/slab.h>
23 #include <linux/delay.h>
24
25 #include <trace/events/sched.h>
26
27 #include <asm/setup.h>
28
29 #include "trace_output.h"
30
31 #undef TRACE_SYSTEM
32 #define TRACE_SYSTEM "TRACE_SYSTEM"
33
34 DEFINE_MUTEX(event_mutex);
35
36 LIST_HEAD(ftrace_events);
37 static LIST_HEAD(ftrace_generic_fields);
38 static LIST_HEAD(ftrace_common_fields);
39
40 #define GFP_TRACE (GFP_KERNEL | __GFP_ZERO)
41
42 static struct kmem_cache *field_cachep;
43 static struct kmem_cache *file_cachep;
44
45 static inline int system_refcount(struct event_subsystem *system)
46 {
47 return system->ref_count;
48 }
49
50 static int system_refcount_inc(struct event_subsystem *system)
51 {
52 return system->ref_count++;
53 }
54
55 static int system_refcount_dec(struct event_subsystem *system)
56 {
57 return --system->ref_count;
58 }
59
60 /* Double loops, do not use break, only goto's work */
61 #define do_for_each_event_file(tr, file) \
62 list_for_each_entry(tr, &ftrace_trace_arrays, list) { \
63 list_for_each_entry(file, &tr->events, list)
64
65 #define do_for_each_event_file_safe(tr, file) \
66 list_for_each_entry(tr, &ftrace_trace_arrays, list) { \
67 struct trace_event_file *___n; \
68 list_for_each_entry_safe(file, ___n, &tr->events, list)
69
70 #define while_for_each_event_file() \
71 }
72
73 static struct list_head *
74 trace_get_fields(struct trace_event_call *event_call)
75 {
76 if (!event_call->class->get_fields)
77 return &event_call->class->fields;
78 return event_call->class->get_fields(event_call);
79 }
80
81 static struct ftrace_event_field *
82 __find_event_field(struct list_head *head, char *name)
83 {
84 struct ftrace_event_field *field;
85
86 list_for_each_entry(field, head, link) {
87 if (!strcmp(field->name, name))
88 return field;
89 }
90
91 return NULL;
92 }
93
94 struct ftrace_event_field *
95 trace_find_event_field(struct trace_event_call *call, char *name)
96 {
97 struct ftrace_event_field *field;
98 struct list_head *head;
99
100 head = trace_get_fields(call);
101 field = __find_event_field(head, name);
102 if (field)
103 return field;
104
105 field = __find_event_field(&ftrace_generic_fields, name);
106 if (field)
107 return field;
108
109 return __find_event_field(&ftrace_common_fields, name);
110 }
111
112 static int __trace_define_field(struct list_head *head, const char *type,
113 const char *name, int offset, int size,
114 int is_signed, int filter_type)
115 {
116 struct ftrace_event_field *field;
117
118 field = kmem_cache_alloc(field_cachep, GFP_TRACE);
119 if (!field)
120 return -ENOMEM;
121
122 field->name = name;
123 field->type = type;
124
125 if (filter_type == FILTER_OTHER)
126 field->filter_type = filter_assign_type(type);
127 else
128 field->filter_type = filter_type;
129
130 field->offset = offset;
131 field->size = size;
132 field->is_signed = is_signed;
133
134 list_add(&field->link, head);
135
136 return 0;
137 }
138
139 int trace_define_field(struct trace_event_call *call, const char *type,
140 const char *name, int offset, int size, int is_signed,
141 int filter_type)
142 {
143 struct list_head *head;
144
145 if (WARN_ON(!call->class))
146 return 0;
147
148 head = trace_get_fields(call);
149 return __trace_define_field(head, type, name, offset, size,
150 is_signed, filter_type);
151 }
152 EXPORT_SYMBOL_GPL(trace_define_field);
153
154 #define __generic_field(type, item, filter_type) \
155 ret = __trace_define_field(&ftrace_generic_fields, #type, \
156 #item, 0, 0, is_signed_type(type), \
157 filter_type); \
158 if (ret) \
159 return ret;
160
161 #define __common_field(type, item) \
162 ret = __trace_define_field(&ftrace_common_fields, #type, \
163 "common_" #item, \
164 offsetof(typeof(ent), item), \
165 sizeof(ent.item), \
166 is_signed_type(type), FILTER_OTHER); \
167 if (ret) \
168 return ret;
169
170 static int trace_define_generic_fields(void)
171 {
172 int ret;
173
174 __generic_field(int, CPU, FILTER_CPU);
175 __generic_field(int, cpu, FILTER_CPU);
176 __generic_field(char *, COMM, FILTER_COMM);
177 __generic_field(char *, comm, FILTER_COMM);
178
179 return ret;
180 }
181
182 static int trace_define_common_fields(void)
183 {
184 int ret;
185 struct trace_entry ent;
186
187 __common_field(unsigned short, type);
188 __common_field(unsigned char, flags);
189 __common_field(unsigned char, preempt_count);
190 __common_field(int, pid);
191
192 return ret;
193 }
194
195 static void trace_destroy_fields(struct trace_event_call *call)
196 {
197 struct ftrace_event_field *field, *next;
198 struct list_head *head;
199
200 head = trace_get_fields(call);
201 list_for_each_entry_safe(field, next, head, link) {
202 list_del(&field->link);
203 kmem_cache_free(field_cachep, field);
204 }
205 }
206
207 int trace_event_raw_init(struct trace_event_call *call)
208 {
209 int id;
210
211 id = register_trace_event(&call->event);
212 if (!id)
213 return -ENODEV;
214
215 return 0;
216 }
217 EXPORT_SYMBOL_GPL(trace_event_raw_init);
218
219 bool trace_event_ignore_this_pid(struct trace_event_file *trace_file)
220 {
221 struct trace_array *tr = trace_file->tr;
222 struct trace_array_cpu *data;
223 struct trace_pid_list *pid_list;
224
225 pid_list = rcu_dereference_sched(tr->filtered_pids);
226 if (!pid_list)
227 return false;
228
229 data = this_cpu_ptr(tr->trace_buffer.data);
230
231 return data->ignore_pid;
232 }
233 EXPORT_SYMBOL_GPL(trace_event_ignore_this_pid);
234
235 void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer,
236 struct trace_event_file *trace_file,
237 unsigned long len)
238 {
239 struct trace_event_call *event_call = trace_file->event_call;
240
241 if ((trace_file->flags & EVENT_FILE_FL_PID_FILTER) &&
242 trace_event_ignore_this_pid(trace_file))
243 return NULL;
244
245 local_save_flags(fbuffer->flags);
246 fbuffer->pc = preempt_count();
247 fbuffer->trace_file = trace_file;
248
249 fbuffer->event =
250 trace_event_buffer_lock_reserve(&fbuffer->buffer, trace_file,
251 event_call->event.type, len,
252 fbuffer->flags, fbuffer->pc);
253 if (!fbuffer->event)
254 return NULL;
255
256 fbuffer->entry = ring_buffer_event_data(fbuffer->event);
257 return fbuffer->entry;
258 }
259 EXPORT_SYMBOL_GPL(trace_event_buffer_reserve);
260
261 static DEFINE_SPINLOCK(tracepoint_iter_lock);
262
263 static void output_printk(struct trace_event_buffer *fbuffer)
264 {
265 struct trace_event_call *event_call;
266 struct trace_event *event;
267 unsigned long flags;
268 struct trace_iterator *iter = tracepoint_print_iter;
269
270 if (!iter)
271 return;
272
273 event_call = fbuffer->trace_file->event_call;
274 if (!event_call || !event_call->event.funcs ||
275 !event_call->event.funcs->trace)
276 return;
277
278 event = &fbuffer->trace_file->event_call->event;
279
280 spin_lock_irqsave(&tracepoint_iter_lock, flags);
281 trace_seq_init(&iter->seq);
282 iter->ent = fbuffer->entry;
283 event_call->event.funcs->trace(iter, 0, event);
284 trace_seq_putc(&iter->seq, 0);
285 printk("%s", iter->seq.buffer);
286
287 spin_unlock_irqrestore(&tracepoint_iter_lock, flags);
288 }
289
290 void trace_event_buffer_commit(struct trace_event_buffer *fbuffer)
291 {
292 if (tracepoint_printk)
293 output_printk(fbuffer);
294
295 event_trigger_unlock_commit(fbuffer->trace_file, fbuffer->buffer,
296 fbuffer->event, fbuffer->entry,
297 fbuffer->flags, fbuffer->pc);
298 }
299 EXPORT_SYMBOL_GPL(trace_event_buffer_commit);
300
301 int trace_event_reg(struct trace_event_call *call,
302 enum trace_reg type, void *data)
303 {
304 struct trace_event_file *file = data;
305
306 WARN_ON(!(call->flags & TRACE_EVENT_FL_TRACEPOINT));
307 switch (type) {
308 case TRACE_REG_REGISTER:
309 return tracepoint_probe_register(call->tp,
310 call->class->probe,
311 file);
312 case TRACE_REG_UNREGISTER:
313 tracepoint_probe_unregister(call->tp,
314 call->class->probe,
315 file);
316 return 0;
317
318 #ifdef CONFIG_PERF_EVENTS
319 case TRACE_REG_PERF_REGISTER:
320 return tracepoint_probe_register(call->tp,
321 call->class->perf_probe,
322 call);
323 case TRACE_REG_PERF_UNREGISTER:
324 tracepoint_probe_unregister(call->tp,
325 call->class->perf_probe,
326 call);
327 return 0;
328 case TRACE_REG_PERF_OPEN:
329 case TRACE_REG_PERF_CLOSE:
330 case TRACE_REG_PERF_ADD:
331 case TRACE_REG_PERF_DEL:
332 return 0;
333 #endif
334 }
335 return 0;
336 }
337 EXPORT_SYMBOL_GPL(trace_event_reg);
338
339 void trace_event_enable_cmd_record(bool enable)
340 {
341 struct trace_event_file *file;
342 struct trace_array *tr;
343
344 mutex_lock(&event_mutex);
345 do_for_each_event_file(tr, file) {
346
347 if (!(file->flags & EVENT_FILE_FL_ENABLED))
348 continue;
349
350 if (enable) {
351 tracing_start_cmdline_record();
352 set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
353 } else {
354 tracing_stop_cmdline_record();
355 clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
356 }
357 } while_for_each_event_file();
358 mutex_unlock(&event_mutex);
359 }
360
361 static int __ftrace_event_enable_disable(struct trace_event_file *file,
362 int enable, int soft_disable)
363 {
364 struct trace_event_call *call = file->event_call;
365 struct trace_array *tr = file->tr;
366 int ret = 0;
367 int disable;
368
369 switch (enable) {
370 case 0:
371 /*
372 * When soft_disable is set and enable is cleared, the sm_ref
373 * reference counter is decremented. If it reaches 0, we want
374 * to clear the SOFT_DISABLED flag but leave the event in the
375 * state that it was. That is, if the event was enabled and
376 * SOFT_DISABLED isn't set, then do nothing. But if SOFT_DISABLED
377 * is set we do not want the event to be enabled before we
378 * clear the bit.
379 *
380 * When soft_disable is not set but the SOFT_MODE flag is,
381 * we do nothing. Do not disable the tracepoint, otherwise
382 * "soft enable"s (clearing the SOFT_DISABLED bit) wont work.
383 */
384 if (soft_disable) {
385 if (atomic_dec_return(&file->sm_ref) > 0)
386 break;
387 disable = file->flags & EVENT_FILE_FL_SOFT_DISABLED;
388 clear_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
389 } else
390 disable = !(file->flags & EVENT_FILE_FL_SOFT_MODE);
391
392 if (disable && (file->flags & EVENT_FILE_FL_ENABLED)) {
393 clear_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
394 if (file->flags & EVENT_FILE_FL_RECORDED_CMD) {
395 tracing_stop_cmdline_record();
396 clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
397 }
398 call->class->reg(call, TRACE_REG_UNREGISTER, file);
399 }
400 /* If in SOFT_MODE, just set the SOFT_DISABLE_BIT, else clear it */
401 if (file->flags & EVENT_FILE_FL_SOFT_MODE)
402 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
403 else
404 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
405 break;
406 case 1:
407 /*
408 * When soft_disable is set and enable is set, we want to
409 * register the tracepoint for the event, but leave the event
410 * as is. That means, if the event was already enabled, we do
411 * nothing (but set SOFT_MODE). If the event is disabled, we
412 * set SOFT_DISABLED before enabling the event tracepoint, so
413 * it still seems to be disabled.
414 */
415 if (!soft_disable)
416 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
417 else {
418 if (atomic_inc_return(&file->sm_ref) > 1)
419 break;
420 set_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
421 }
422
423 if (!(file->flags & EVENT_FILE_FL_ENABLED)) {
424
425 /* Keep the event disabled, when going to SOFT_MODE. */
426 if (soft_disable)
427 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
428
429 if (tr->trace_flags & TRACE_ITER_RECORD_CMD) {
430 tracing_start_cmdline_record();
431 set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
432 }
433 ret = call->class->reg(call, TRACE_REG_REGISTER, file);
434 if (ret) {
435 tracing_stop_cmdline_record();
436 pr_info("event trace: Could not enable event "
437 "%s\n", trace_event_name(call));
438 break;
439 }
440 set_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
441
442 /* WAS_ENABLED gets set but never cleared. */
443 call->flags |= TRACE_EVENT_FL_WAS_ENABLED;
444 }
445 break;
446 }
447
448 return ret;
449 }
450
451 int trace_event_enable_disable(struct trace_event_file *file,
452 int enable, int soft_disable)
453 {
454 return __ftrace_event_enable_disable(file, enable, soft_disable);
455 }
456
457 static int ftrace_event_enable_disable(struct trace_event_file *file,
458 int enable)
459 {
460 return __ftrace_event_enable_disable(file, enable, 0);
461 }
462
463 static void ftrace_clear_events(struct trace_array *tr)
464 {
465 struct trace_event_file *file;
466
467 mutex_lock(&event_mutex);
468 list_for_each_entry(file, &tr->events, list) {
469 ftrace_event_enable_disable(file, 0);
470 }
471 mutex_unlock(&event_mutex);
472 }
473
474 static int cmp_pid(const void *key, const void *elt)
475 {
476 const pid_t *search_pid = key;
477 const pid_t *pid = elt;
478
479 if (*search_pid == *pid)
480 return 0;
481 if (*search_pid < *pid)
482 return -1;
483 return 1;
484 }
485
486 static bool
487 check_ignore_pid(struct trace_pid_list *filtered_pids, struct task_struct *task)
488 {
489 pid_t search_pid;
490 pid_t *pid;
491
492 /*
493 * Return false, because if filtered_pids does not exist,
494 * all pids are good to trace.
495 */
496 if (!filtered_pids)
497 return false;
498
499 search_pid = task->pid;
500
501 pid = bsearch(&search_pid, filtered_pids->pids,
502 filtered_pids->nr_pids, sizeof(pid_t),
503 cmp_pid);
504 if (!pid)
505 return true;
506
507 return false;
508 }
509
510 static void
511 event_filter_pid_sched_switch_probe_pre(void *data, bool preempt,
512 struct task_struct *prev, struct task_struct *next)
513 {
514 struct trace_array *tr = data;
515 struct trace_pid_list *pid_list;
516
517 pid_list = rcu_dereference_sched(tr->filtered_pids);
518
519 this_cpu_write(tr->trace_buffer.data->ignore_pid,
520 check_ignore_pid(pid_list, prev) &&
521 check_ignore_pid(pid_list, next));
522 }
523
524 static void
525 event_filter_pid_sched_switch_probe_post(void *data, bool preempt,
526 struct task_struct *prev, struct task_struct *next)
527 {
528 struct trace_array *tr = data;
529 struct trace_pid_list *pid_list;
530
531 pid_list = rcu_dereference_sched(tr->filtered_pids);
532
533 this_cpu_write(tr->trace_buffer.data->ignore_pid,
534 check_ignore_pid(pid_list, next));
535 }
536
537 static void
538 event_filter_pid_sched_wakeup_probe_pre(void *data, struct task_struct *task)
539 {
540 struct trace_array *tr = data;
541 struct trace_pid_list *pid_list;
542
543 /* Nothing to do if we are already tracing */
544 if (!this_cpu_read(tr->trace_buffer.data->ignore_pid))
545 return;
546
547 pid_list = rcu_dereference_sched(tr->filtered_pids);
548
549 this_cpu_write(tr->trace_buffer.data->ignore_pid,
550 check_ignore_pid(pid_list, task));
551 }
552
553 static void
554 event_filter_pid_sched_wakeup_probe_post(void *data, struct task_struct *task)
555 {
556 struct trace_array *tr = data;
557 struct trace_pid_list *pid_list;
558
559 /* Nothing to do if we are not tracing */
560 if (this_cpu_read(tr->trace_buffer.data->ignore_pid))
561 return;
562
563 pid_list = rcu_dereference_sched(tr->filtered_pids);
564
565 /* Set tracing if current is enabled */
566 this_cpu_write(tr->trace_buffer.data->ignore_pid,
567 check_ignore_pid(pid_list, current));
568 }
569
570 static void __ftrace_clear_event_pids(struct trace_array *tr)
571 {
572 struct trace_pid_list *pid_list;
573 struct trace_event_file *file;
574 int cpu;
575
576 pid_list = rcu_dereference_protected(tr->filtered_pids,
577 lockdep_is_held(&event_mutex));
578 if (!pid_list)
579 return;
580
581 unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_pre, tr);
582 unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_post, tr);
583
584 unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre, tr);
585 unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_post, tr);
586
587 unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre, tr);
588 unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post, tr);
589
590 unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_pre, tr);
591 unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_post, tr);
592
593 list_for_each_entry(file, &tr->events, list) {
594 clear_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
595 }
596
597 for_each_possible_cpu(cpu)
598 per_cpu_ptr(tr->trace_buffer.data, cpu)->ignore_pid = false;
599
600 rcu_assign_pointer(tr->filtered_pids, NULL);
601
602 /* Wait till all users are no longer using pid filtering */
603 synchronize_sched();
604
605 free_pages((unsigned long)pid_list->pids, pid_list->order);
606 kfree(pid_list);
607 }
608
609 static void ftrace_clear_event_pids(struct trace_array *tr)
610 {
611 mutex_lock(&event_mutex);
612 __ftrace_clear_event_pids(tr);
613 mutex_unlock(&event_mutex);
614 }
615
616 static void __put_system(struct event_subsystem *system)
617 {
618 struct event_filter *filter = system->filter;
619
620 WARN_ON_ONCE(system_refcount(system) == 0);
621 if (system_refcount_dec(system))
622 return;
623
624 list_del(&system->list);
625
626 if (filter) {
627 kfree(filter->filter_string);
628 kfree(filter);
629 }
630 kfree_const(system->name);
631 kfree(system);
632 }
633
634 static void __get_system(struct event_subsystem *system)
635 {
636 WARN_ON_ONCE(system_refcount(system) == 0);
637 system_refcount_inc(system);
638 }
639
640 static void __get_system_dir(struct trace_subsystem_dir *dir)
641 {
642 WARN_ON_ONCE(dir->ref_count == 0);
643 dir->ref_count++;
644 __get_system(dir->subsystem);
645 }
646
647 static void __put_system_dir(struct trace_subsystem_dir *dir)
648 {
649 WARN_ON_ONCE(dir->ref_count == 0);
650 /* If the subsystem is about to be freed, the dir must be too */
651 WARN_ON_ONCE(system_refcount(dir->subsystem) == 1 && dir->ref_count != 1);
652
653 __put_system(dir->subsystem);
654 if (!--dir->ref_count)
655 kfree(dir);
656 }
657
658 static void put_system(struct trace_subsystem_dir *dir)
659 {
660 mutex_lock(&event_mutex);
661 __put_system_dir(dir);
662 mutex_unlock(&event_mutex);
663 }
664
665 static void remove_subsystem(struct trace_subsystem_dir *dir)
666 {
667 if (!dir)
668 return;
669
670 if (!--dir->nr_events) {
671 tracefs_remove_recursive(dir->entry);
672 list_del(&dir->list);
673 __put_system_dir(dir);
674 }
675 }
676
677 static void remove_event_file_dir(struct trace_event_file *file)
678 {
679 struct dentry *dir = file->dir;
680 struct dentry *child;
681
682 if (dir) {
683 spin_lock(&dir->d_lock); /* probably unneeded */
684 list_for_each_entry(child, &dir->d_subdirs, d_child) {
685 if (d_really_is_positive(child)) /* probably unneeded */
686 d_inode(child)->i_private = NULL;
687 }
688 spin_unlock(&dir->d_lock);
689
690 tracefs_remove_recursive(dir);
691 }
692
693 list_del(&file->list);
694 remove_subsystem(file->system);
695 free_event_filter(file->filter);
696 kmem_cache_free(file_cachep, file);
697 }
698
699 /*
700 * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
701 */
702 static int
703 __ftrace_set_clr_event_nolock(struct trace_array *tr, const char *match,
704 const char *sub, const char *event, int set)
705 {
706 struct trace_event_file *file;
707 struct trace_event_call *call;
708 const char *name;
709 int ret = -EINVAL;
710
711 list_for_each_entry(file, &tr->events, list) {
712
713 call = file->event_call;
714 name = trace_event_name(call);
715
716 if (!name || !call->class || !call->class->reg)
717 continue;
718
719 if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
720 continue;
721
722 if (match &&
723 strcmp(match, name) != 0 &&
724 strcmp(match, call->class->system) != 0)
725 continue;
726
727 if (sub && strcmp(sub, call->class->system) != 0)
728 continue;
729
730 if (event && strcmp(event, name) != 0)
731 continue;
732
733 ftrace_event_enable_disable(file, set);
734
735 ret = 0;
736 }
737
738 return ret;
739 }
740
741 static int __ftrace_set_clr_event(struct trace_array *tr, const char *match,
742 const char *sub, const char *event, int set)
743 {
744 int ret;
745
746 mutex_lock(&event_mutex);
747 ret = __ftrace_set_clr_event_nolock(tr, match, sub, event, set);
748 mutex_unlock(&event_mutex);
749
750 return ret;
751 }
752
753 static int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set)
754 {
755 char *event = NULL, *sub = NULL, *match;
756 int ret;
757
758 /*
759 * The buf format can be <subsystem>:<event-name>
760 * *:<event-name> means any event by that name.
761 * :<event-name> is the same.
762 *
763 * <subsystem>:* means all events in that subsystem
764 * <subsystem>: means the same.
765 *
766 * <name> (no ':') means all events in a subsystem with
767 * the name <name> or any event that matches <name>
768 */
769
770 match = strsep(&buf, ":");
771 if (buf) {
772 sub = match;
773 event = buf;
774 match = NULL;
775
776 if (!strlen(sub) || strcmp(sub, "*") == 0)
777 sub = NULL;
778 if (!strlen(event) || strcmp(event, "*") == 0)
779 event = NULL;
780 }
781
782 ret = __ftrace_set_clr_event(tr, match, sub, event, set);
783
784 /* Put back the colon to allow this to be called again */
785 if (buf)
786 *(buf - 1) = ':';
787
788 return ret;
789 }
790
791 /**
792 * trace_set_clr_event - enable or disable an event
793 * @system: system name to match (NULL for any system)
794 * @event: event name to match (NULL for all events, within system)
795 * @set: 1 to enable, 0 to disable
796 *
797 * This is a way for other parts of the kernel to enable or disable
798 * event recording.
799 *
800 * Returns 0 on success, -EINVAL if the parameters do not match any
801 * registered events.
802 */
803 int trace_set_clr_event(const char *system, const char *event, int set)
804 {
805 struct trace_array *tr = top_trace_array();
806
807 if (!tr)
808 return -ENODEV;
809
810 return __ftrace_set_clr_event(tr, NULL, system, event, set);
811 }
812 EXPORT_SYMBOL_GPL(trace_set_clr_event);
813
814 /* 128 should be much more than enough */
815 #define EVENT_BUF_SIZE 127
816
817 static ssize_t
818 ftrace_event_write(struct file *file, const char __user *ubuf,
819 size_t cnt, loff_t *ppos)
820 {
821 struct trace_parser parser;
822 struct seq_file *m = file->private_data;
823 struct trace_array *tr = m->private;
824 ssize_t read, ret;
825
826 if (!cnt)
827 return 0;
828
829 ret = tracing_update_buffers();
830 if (ret < 0)
831 return ret;
832
833 if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
834 return -ENOMEM;
835
836 read = trace_get_user(&parser, ubuf, cnt, ppos);
837
838 if (read >= 0 && trace_parser_loaded((&parser))) {
839 int set = 1;
840
841 if (*parser.buffer == '!')
842 set = 0;
843
844 parser.buffer[parser.idx] = 0;
845
846 ret = ftrace_set_clr_event(tr, parser.buffer + !set, set);
847 if (ret)
848 goto out_put;
849 }
850
851 ret = read;
852
853 out_put:
854 trace_parser_put(&parser);
855
856 return ret;
857 }
858
859 static void *
860 t_next(struct seq_file *m, void *v, loff_t *pos)
861 {
862 struct trace_event_file *file = v;
863 struct trace_event_call *call;
864 struct trace_array *tr = m->private;
865
866 (*pos)++;
867
868 list_for_each_entry_continue(file, &tr->events, list) {
869 call = file->event_call;
870 /*
871 * The ftrace subsystem is for showing formats only.
872 * They can not be enabled or disabled via the event files.
873 */
874 if (call->class && call->class->reg &&
875 !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
876 return file;
877 }
878
879 return NULL;
880 }
881
882 static void *t_start(struct seq_file *m, loff_t *pos)
883 {
884 struct trace_event_file *file;
885 struct trace_array *tr = m->private;
886 loff_t l;
887
888 mutex_lock(&event_mutex);
889
890 file = list_entry(&tr->events, struct trace_event_file, list);
891 for (l = 0; l <= *pos; ) {
892 file = t_next(m, file, &l);
893 if (!file)
894 break;
895 }
896 return file;
897 }
898
899 static void *
900 s_next(struct seq_file *m, void *v, loff_t *pos)
901 {
902 struct trace_event_file *file = v;
903 struct trace_array *tr = m->private;
904
905 (*pos)++;
906
907 list_for_each_entry_continue(file, &tr->events, list) {
908 if (file->flags & EVENT_FILE_FL_ENABLED)
909 return file;
910 }
911
912 return NULL;
913 }
914
915 static void *s_start(struct seq_file *m, loff_t *pos)
916 {
917 struct trace_event_file *file;
918 struct trace_array *tr = m->private;
919 loff_t l;
920
921 mutex_lock(&event_mutex);
922
923 file = list_entry(&tr->events, struct trace_event_file, list);
924 for (l = 0; l <= *pos; ) {
925 file = s_next(m, file, &l);
926 if (!file)
927 break;
928 }
929 return file;
930 }
931
932 static int t_show(struct seq_file *m, void *v)
933 {
934 struct trace_event_file *file = v;
935 struct trace_event_call *call = file->event_call;
936
937 if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
938 seq_printf(m, "%s:", call->class->system);
939 seq_printf(m, "%s\n", trace_event_name(call));
940
941 return 0;
942 }
943
944 static void t_stop(struct seq_file *m, void *p)
945 {
946 mutex_unlock(&event_mutex);
947 }
948
949 static void *p_start(struct seq_file *m, loff_t *pos)
950 __acquires(RCU)
951 {
952 struct trace_pid_list *pid_list;
953 struct trace_array *tr = m->private;
954
955 /*
956 * Grab the mutex, to keep calls to p_next() having the same
957 * tr->filtered_pids as p_start() has.
958 * If we just passed the tr->filtered_pids around, then RCU would
959 * have been enough, but doing that makes things more complex.
960 */
961 mutex_lock(&event_mutex);
962 rcu_read_lock_sched();
963
964 pid_list = rcu_dereference_sched(tr->filtered_pids);
965
966 if (!pid_list || *pos >= pid_list->nr_pids)
967 return NULL;
968
969 return (void *)&pid_list->pids[*pos];
970 }
971
972 static void p_stop(struct seq_file *m, void *p)
973 __releases(RCU)
974 {
975 rcu_read_unlock_sched();
976 mutex_unlock(&event_mutex);
977 }
978
979 static void *
980 p_next(struct seq_file *m, void *v, loff_t *pos)
981 {
982 struct trace_array *tr = m->private;
983 struct trace_pid_list *pid_list = rcu_dereference_sched(tr->filtered_pids);
984
985 (*pos)++;
986
987 if (*pos >= pid_list->nr_pids)
988 return NULL;
989
990 return (void *)&pid_list->pids[*pos];
991 }
992
993 static int p_show(struct seq_file *m, void *v)
994 {
995 pid_t *pid = v;
996
997 seq_printf(m, "%d\n", *pid);
998 return 0;
999 }
1000
1001 static ssize_t
1002 event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1003 loff_t *ppos)
1004 {
1005 struct trace_event_file *file;
1006 unsigned long flags;
1007 char buf[4] = "0";
1008
1009 mutex_lock(&event_mutex);
1010 file = event_file_data(filp);
1011 if (likely(file))
1012 flags = file->flags;
1013 mutex_unlock(&event_mutex);
1014
1015 if (!file)
1016 return -ENODEV;
1017
1018 if (flags & EVENT_FILE_FL_ENABLED &&
1019 !(flags & EVENT_FILE_FL_SOFT_DISABLED))
1020 strcpy(buf, "1");
1021
1022 if (flags & EVENT_FILE_FL_SOFT_DISABLED ||
1023 flags & EVENT_FILE_FL_SOFT_MODE)
1024 strcat(buf, "*");
1025
1026 strcat(buf, "\n");
1027
1028 return simple_read_from_buffer(ubuf, cnt, ppos, buf, strlen(buf));
1029 }
1030
1031 static ssize_t
1032 event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1033 loff_t *ppos)
1034 {
1035 struct trace_event_file *file;
1036 unsigned long val;
1037 int ret;
1038
1039 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1040 if (ret)
1041 return ret;
1042
1043 ret = tracing_update_buffers();
1044 if (ret < 0)
1045 return ret;
1046
1047 switch (val) {
1048 case 0:
1049 case 1:
1050 ret = -ENODEV;
1051 mutex_lock(&event_mutex);
1052 file = event_file_data(filp);
1053 if (likely(file))
1054 ret = ftrace_event_enable_disable(file, val);
1055 mutex_unlock(&event_mutex);
1056 break;
1057
1058 default:
1059 return -EINVAL;
1060 }
1061
1062 *ppos += cnt;
1063
1064 return ret ? ret : cnt;
1065 }
1066
1067 static ssize_t
1068 system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1069 loff_t *ppos)
1070 {
1071 const char set_to_char[4] = { '?', '0', '1', 'X' };
1072 struct trace_subsystem_dir *dir = filp->private_data;
1073 struct event_subsystem *system = dir->subsystem;
1074 struct trace_event_call *call;
1075 struct trace_event_file *file;
1076 struct trace_array *tr = dir->tr;
1077 char buf[2];
1078 int set = 0;
1079 int ret;
1080
1081 mutex_lock(&event_mutex);
1082 list_for_each_entry(file, &tr->events, list) {
1083 call = file->event_call;
1084 if (!trace_event_name(call) || !call->class || !call->class->reg)
1085 continue;
1086
1087 if (system && strcmp(call->class->system, system->name) != 0)
1088 continue;
1089
1090 /*
1091 * We need to find out if all the events are set
1092 * or if all events or cleared, or if we have
1093 * a mixture.
1094 */
1095 set |= (1 << !!(file->flags & EVENT_FILE_FL_ENABLED));
1096
1097 /*
1098 * If we have a mixture, no need to look further.
1099 */
1100 if (set == 3)
1101 break;
1102 }
1103 mutex_unlock(&event_mutex);
1104
1105 buf[0] = set_to_char[set];
1106 buf[1] = '\n';
1107
1108 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
1109
1110 return ret;
1111 }
1112
1113 static ssize_t
1114 system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1115 loff_t *ppos)
1116 {
1117 struct trace_subsystem_dir *dir = filp->private_data;
1118 struct event_subsystem *system = dir->subsystem;
1119 const char *name = NULL;
1120 unsigned long val;
1121 ssize_t ret;
1122
1123 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1124 if (ret)
1125 return ret;
1126
1127 ret = tracing_update_buffers();
1128 if (ret < 0)
1129 return ret;
1130
1131 if (val != 0 && val != 1)
1132 return -EINVAL;
1133
1134 /*
1135 * Opening of "enable" adds a ref count to system,
1136 * so the name is safe to use.
1137 */
1138 if (system)
1139 name = system->name;
1140
1141 ret = __ftrace_set_clr_event(dir->tr, NULL, name, NULL, val);
1142 if (ret)
1143 goto out;
1144
1145 ret = cnt;
1146
1147 out:
1148 *ppos += cnt;
1149
1150 return ret;
1151 }
1152
1153 enum {
1154 FORMAT_HEADER = 1,
1155 FORMAT_FIELD_SEPERATOR = 2,
1156 FORMAT_PRINTFMT = 3,
1157 };
1158
1159 static void *f_next(struct seq_file *m, void *v, loff_t *pos)
1160 {
1161 struct trace_event_call *call = event_file_data(m->private);
1162 struct list_head *common_head = &ftrace_common_fields;
1163 struct list_head *head = trace_get_fields(call);
1164 struct list_head *node = v;
1165
1166 (*pos)++;
1167
1168 switch ((unsigned long)v) {
1169 case FORMAT_HEADER:
1170 node = common_head;
1171 break;
1172
1173 case FORMAT_FIELD_SEPERATOR:
1174 node = head;
1175 break;
1176
1177 case FORMAT_PRINTFMT:
1178 /* all done */
1179 return NULL;
1180 }
1181
1182 node = node->prev;
1183 if (node == common_head)
1184 return (void *)FORMAT_FIELD_SEPERATOR;
1185 else if (node == head)
1186 return (void *)FORMAT_PRINTFMT;
1187 else
1188 return node;
1189 }
1190
1191 static int f_show(struct seq_file *m, void *v)
1192 {
1193 struct trace_event_call *call = event_file_data(m->private);
1194 struct ftrace_event_field *field;
1195 const char *array_descriptor;
1196
1197 switch ((unsigned long)v) {
1198 case FORMAT_HEADER:
1199 seq_printf(m, "name: %s\n", trace_event_name(call));
1200 seq_printf(m, "ID: %d\n", call->event.type);
1201 seq_puts(m, "format:\n");
1202 return 0;
1203
1204 case FORMAT_FIELD_SEPERATOR:
1205 seq_putc(m, '\n');
1206 return 0;
1207
1208 case FORMAT_PRINTFMT:
1209 seq_printf(m, "\nprint fmt: %s\n",
1210 call->print_fmt);
1211 return 0;
1212 }
1213
1214 field = list_entry(v, struct ftrace_event_field, link);
1215 /*
1216 * Smartly shows the array type(except dynamic array).
1217 * Normal:
1218 * field:TYPE VAR
1219 * If TYPE := TYPE[LEN], it is shown:
1220 * field:TYPE VAR[LEN]
1221 */
1222 array_descriptor = strchr(field->type, '[');
1223
1224 if (!strncmp(field->type, "__data_loc", 10))
1225 array_descriptor = NULL;
1226
1227 if (!array_descriptor)
1228 seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1229 field->type, field->name, field->offset,
1230 field->size, !!field->is_signed);
1231 else
1232 seq_printf(m, "\tfield:%.*s %s%s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1233 (int)(array_descriptor - field->type),
1234 field->type, field->name,
1235 array_descriptor, field->offset,
1236 field->size, !!field->is_signed);
1237
1238 return 0;
1239 }
1240
1241 static void *f_start(struct seq_file *m, loff_t *pos)
1242 {
1243 void *p = (void *)FORMAT_HEADER;
1244 loff_t l = 0;
1245
1246 /* ->stop() is called even if ->start() fails */
1247 mutex_lock(&event_mutex);
1248 if (!event_file_data(m->private))
1249 return ERR_PTR(-ENODEV);
1250
1251 while (l < *pos && p)
1252 p = f_next(m, p, &l);
1253
1254 return p;
1255 }
1256
1257 static void f_stop(struct seq_file *m, void *p)
1258 {
1259 mutex_unlock(&event_mutex);
1260 }
1261
1262 static const struct seq_operations trace_format_seq_ops = {
1263 .start = f_start,
1264 .next = f_next,
1265 .stop = f_stop,
1266 .show = f_show,
1267 };
1268
1269 static int trace_format_open(struct inode *inode, struct file *file)
1270 {
1271 struct seq_file *m;
1272 int ret;
1273
1274 ret = seq_open(file, &trace_format_seq_ops);
1275 if (ret < 0)
1276 return ret;
1277
1278 m = file->private_data;
1279 m->private = file;
1280
1281 return 0;
1282 }
1283
1284 static ssize_t
1285 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1286 {
1287 int id = (long)event_file_data(filp);
1288 char buf[32];
1289 int len;
1290
1291 if (*ppos)
1292 return 0;
1293
1294 if (unlikely(!id))
1295 return -ENODEV;
1296
1297 len = sprintf(buf, "%d\n", id);
1298
1299 return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
1300 }
1301
1302 static ssize_t
1303 event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1304 loff_t *ppos)
1305 {
1306 struct trace_event_file *file;
1307 struct trace_seq *s;
1308 int r = -ENODEV;
1309
1310 if (*ppos)
1311 return 0;
1312
1313 s = kmalloc(sizeof(*s), GFP_KERNEL);
1314
1315 if (!s)
1316 return -ENOMEM;
1317
1318 trace_seq_init(s);
1319
1320 mutex_lock(&event_mutex);
1321 file = event_file_data(filp);
1322 if (file)
1323 print_event_filter(file, s);
1324 mutex_unlock(&event_mutex);
1325
1326 if (file)
1327 r = simple_read_from_buffer(ubuf, cnt, ppos,
1328 s->buffer, trace_seq_used(s));
1329
1330 kfree(s);
1331
1332 return r;
1333 }
1334
1335 static ssize_t
1336 event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1337 loff_t *ppos)
1338 {
1339 struct trace_event_file *file;
1340 char *buf;
1341 int err = -ENODEV;
1342
1343 if (cnt >= PAGE_SIZE)
1344 return -EINVAL;
1345
1346 buf = memdup_user_nul(ubuf, cnt);
1347 if (IS_ERR(buf))
1348 return PTR_ERR(buf);
1349
1350 mutex_lock(&event_mutex);
1351 file = event_file_data(filp);
1352 if (file)
1353 err = apply_event_filter(file, buf);
1354 mutex_unlock(&event_mutex);
1355
1356 kfree(buf);
1357 if (err < 0)
1358 return err;
1359
1360 *ppos += cnt;
1361
1362 return cnt;
1363 }
1364
1365 static LIST_HEAD(event_subsystems);
1366
1367 static int subsystem_open(struct inode *inode, struct file *filp)
1368 {
1369 struct event_subsystem *system = NULL;
1370 struct trace_subsystem_dir *dir = NULL; /* Initialize for gcc */
1371 struct trace_array *tr;
1372 int ret;
1373
1374 if (tracing_is_disabled())
1375 return -ENODEV;
1376
1377 /* Make sure the system still exists */
1378 mutex_lock(&trace_types_lock);
1379 mutex_lock(&event_mutex);
1380 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
1381 list_for_each_entry(dir, &tr->systems, list) {
1382 if (dir == inode->i_private) {
1383 /* Don't open systems with no events */
1384 if (dir->nr_events) {
1385 __get_system_dir(dir);
1386 system = dir->subsystem;
1387 }
1388 goto exit_loop;
1389 }
1390 }
1391 }
1392 exit_loop:
1393 mutex_unlock(&event_mutex);
1394 mutex_unlock(&trace_types_lock);
1395
1396 if (!system)
1397 return -ENODEV;
1398
1399 /* Some versions of gcc think dir can be uninitialized here */
1400 WARN_ON(!dir);
1401
1402 /* Still need to increment the ref count of the system */
1403 if (trace_array_get(tr) < 0) {
1404 put_system(dir);
1405 return -ENODEV;
1406 }
1407
1408 ret = tracing_open_generic(inode, filp);
1409 if (ret < 0) {
1410 trace_array_put(tr);
1411 put_system(dir);
1412 }
1413
1414 return ret;
1415 }
1416
1417 static int system_tr_open(struct inode *inode, struct file *filp)
1418 {
1419 struct trace_subsystem_dir *dir;
1420 struct trace_array *tr = inode->i_private;
1421 int ret;
1422
1423 if (tracing_is_disabled())
1424 return -ENODEV;
1425
1426 if (trace_array_get(tr) < 0)
1427 return -ENODEV;
1428
1429 /* Make a temporary dir that has no system but points to tr */
1430 dir = kzalloc(sizeof(*dir), GFP_KERNEL);
1431 if (!dir) {
1432 trace_array_put(tr);
1433 return -ENOMEM;
1434 }
1435
1436 dir->tr = tr;
1437
1438 ret = tracing_open_generic(inode, filp);
1439 if (ret < 0) {
1440 trace_array_put(tr);
1441 kfree(dir);
1442 return ret;
1443 }
1444
1445 filp->private_data = dir;
1446
1447 return 0;
1448 }
1449
1450 static int subsystem_release(struct inode *inode, struct file *file)
1451 {
1452 struct trace_subsystem_dir *dir = file->private_data;
1453
1454 trace_array_put(dir->tr);
1455
1456 /*
1457 * If dir->subsystem is NULL, then this is a temporary
1458 * descriptor that was made for a trace_array to enable
1459 * all subsystems.
1460 */
1461 if (dir->subsystem)
1462 put_system(dir);
1463 else
1464 kfree(dir);
1465
1466 return 0;
1467 }
1468
1469 static ssize_t
1470 subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1471 loff_t *ppos)
1472 {
1473 struct trace_subsystem_dir *dir = filp->private_data;
1474 struct event_subsystem *system = dir->subsystem;
1475 struct trace_seq *s;
1476 int r;
1477
1478 if (*ppos)
1479 return 0;
1480
1481 s = kmalloc(sizeof(*s), GFP_KERNEL);
1482 if (!s)
1483 return -ENOMEM;
1484
1485 trace_seq_init(s);
1486
1487 print_subsystem_event_filter(system, s);
1488 r = simple_read_from_buffer(ubuf, cnt, ppos,
1489 s->buffer, trace_seq_used(s));
1490
1491 kfree(s);
1492
1493 return r;
1494 }
1495
1496 static ssize_t
1497 subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1498 loff_t *ppos)
1499 {
1500 struct trace_subsystem_dir *dir = filp->private_data;
1501 char *buf;
1502 int err;
1503
1504 if (cnt >= PAGE_SIZE)
1505 return -EINVAL;
1506
1507 buf = memdup_user_nul(ubuf, cnt);
1508 if (IS_ERR(buf))
1509 return PTR_ERR(buf);
1510
1511 err = apply_subsystem_event_filter(dir, buf);
1512 kfree(buf);
1513 if (err < 0)
1514 return err;
1515
1516 *ppos += cnt;
1517
1518 return cnt;
1519 }
1520
1521 static ssize_t
1522 show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1523 {
1524 int (*func)(struct trace_seq *s) = filp->private_data;
1525 struct trace_seq *s;
1526 int r;
1527
1528 if (*ppos)
1529 return 0;
1530
1531 s = kmalloc(sizeof(*s), GFP_KERNEL);
1532 if (!s)
1533 return -ENOMEM;
1534
1535 trace_seq_init(s);
1536
1537 func(s);
1538 r = simple_read_from_buffer(ubuf, cnt, ppos,
1539 s->buffer, trace_seq_used(s));
1540
1541 kfree(s);
1542
1543 return r;
1544 }
1545
1546 static int max_pids(struct trace_pid_list *pid_list)
1547 {
1548 return (PAGE_SIZE << pid_list->order) / sizeof(pid_t);
1549 }
1550
1551 static void ignore_task_cpu(void *data)
1552 {
1553 struct trace_array *tr = data;
1554 struct trace_pid_list *pid_list;
1555
1556 /*
1557 * This function is called by on_each_cpu() while the
1558 * event_mutex is held.
1559 */
1560 pid_list = rcu_dereference_protected(tr->filtered_pids,
1561 mutex_is_locked(&event_mutex));
1562
1563 this_cpu_write(tr->trace_buffer.data->ignore_pid,
1564 check_ignore_pid(pid_list, current));
1565 }
1566
1567 static ssize_t
1568 ftrace_event_pid_write(struct file *filp, const char __user *ubuf,
1569 size_t cnt, loff_t *ppos)
1570 {
1571 struct seq_file *m = filp->private_data;
1572 struct trace_array *tr = m->private;
1573 struct trace_pid_list *filtered_pids = NULL;
1574 struct trace_pid_list *pid_list = NULL;
1575 struct trace_event_file *file;
1576 struct trace_parser parser;
1577 unsigned long val;
1578 loff_t this_pos;
1579 ssize_t read = 0;
1580 ssize_t ret = 0;
1581 pid_t pid;
1582 int i;
1583
1584 if (!cnt)
1585 return 0;
1586
1587 ret = tracing_update_buffers();
1588 if (ret < 0)
1589 return ret;
1590
1591 if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
1592 return -ENOMEM;
1593
1594 mutex_lock(&event_mutex);
1595 /*
1596 * Load as many pids into the array before doing a
1597 * swap from the tr->filtered_pids to the new list.
1598 */
1599 while (cnt > 0) {
1600
1601 this_pos = 0;
1602
1603 ret = trace_get_user(&parser, ubuf, cnt, &this_pos);
1604 if (ret < 0 || !trace_parser_loaded(&parser))
1605 break;
1606
1607 read += ret;
1608 ubuf += ret;
1609 cnt -= ret;
1610
1611 parser.buffer[parser.idx] = 0;
1612
1613 ret = -EINVAL;
1614 if (kstrtoul(parser.buffer, 0, &val))
1615 break;
1616 if (val > INT_MAX)
1617 break;
1618
1619 pid = (pid_t)val;
1620
1621 ret = -ENOMEM;
1622 if (!pid_list) {
1623 pid_list = kmalloc(sizeof(*pid_list), GFP_KERNEL);
1624 if (!pid_list)
1625 break;
1626
1627 filtered_pids = rcu_dereference_protected(tr->filtered_pids,
1628 lockdep_is_held(&event_mutex));
1629 if (filtered_pids)
1630 pid_list->order = filtered_pids->order;
1631 else
1632 pid_list->order = 0;
1633
1634 pid_list->pids = (void *)__get_free_pages(GFP_KERNEL,
1635 pid_list->order);
1636 if (!pid_list->pids)
1637 break;
1638
1639 if (filtered_pids) {
1640 pid_list->nr_pids = filtered_pids->nr_pids;
1641 memcpy(pid_list->pids, filtered_pids->pids,
1642 pid_list->nr_pids * sizeof(pid_t));
1643 } else
1644 pid_list->nr_pids = 0;
1645 }
1646
1647 if (pid_list->nr_pids >= max_pids(pid_list)) {
1648 pid_t *pid_page;
1649
1650 pid_page = (void *)__get_free_pages(GFP_KERNEL,
1651 pid_list->order + 1);
1652 if (!pid_page)
1653 break;
1654 memcpy(pid_page, pid_list->pids,
1655 pid_list->nr_pids * sizeof(pid_t));
1656 free_pages((unsigned long)pid_list->pids, pid_list->order);
1657
1658 pid_list->order++;
1659 pid_list->pids = pid_page;
1660 }
1661
1662 pid_list->pids[pid_list->nr_pids++] = pid;
1663 trace_parser_clear(&parser);
1664 ret = 0;
1665 }
1666 trace_parser_put(&parser);
1667
1668 if (ret < 0) {
1669 if (pid_list)
1670 free_pages((unsigned long)pid_list->pids, pid_list->order);
1671 kfree(pid_list);
1672 mutex_unlock(&event_mutex);
1673 return ret;
1674 }
1675
1676 if (!pid_list) {
1677 mutex_unlock(&event_mutex);
1678 return ret;
1679 }
1680
1681 sort(pid_list->pids, pid_list->nr_pids, sizeof(pid_t), cmp_pid, NULL);
1682
1683 /* Remove duplicates */
1684 for (i = 1; i < pid_list->nr_pids; i++) {
1685 int start = i;
1686
1687 while (i < pid_list->nr_pids &&
1688 pid_list->pids[i - 1] == pid_list->pids[i])
1689 i++;
1690
1691 if (start != i) {
1692 if (i < pid_list->nr_pids) {
1693 memmove(&pid_list->pids[start], &pid_list->pids[i],
1694 (pid_list->nr_pids - i) * sizeof(pid_t));
1695 pid_list->nr_pids -= i - start;
1696 i = start;
1697 } else
1698 pid_list->nr_pids = start;
1699 }
1700 }
1701
1702 rcu_assign_pointer(tr->filtered_pids, pid_list);
1703
1704 list_for_each_entry(file, &tr->events, list) {
1705 set_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
1706 }
1707
1708 if (filtered_pids) {
1709 synchronize_sched();
1710
1711 free_pages((unsigned long)filtered_pids->pids, filtered_pids->order);
1712 kfree(filtered_pids);
1713 } else {
1714 /*
1715 * Register a probe that is called before all other probes
1716 * to set ignore_pid if next or prev do not match.
1717 * Register a probe this is called after all other probes
1718 * to only keep ignore_pid set if next pid matches.
1719 */
1720 register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_pre,
1721 tr, INT_MAX);
1722 register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_post,
1723 tr, 0);
1724
1725 register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre,
1726 tr, INT_MAX);
1727 register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_post,
1728 tr, 0);
1729
1730 register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre,
1731 tr, INT_MAX);
1732 register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post,
1733 tr, 0);
1734
1735 register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_pre,
1736 tr, INT_MAX);
1737 register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_post,
1738 tr, 0);
1739 }
1740
1741 /*
1742 * Ignoring of pids is done at task switch. But we have to
1743 * check for those tasks that are currently running.
1744 * Always do this in case a pid was appended or removed.
1745 */
1746 on_each_cpu(ignore_task_cpu, tr, 1);
1747
1748 mutex_unlock(&event_mutex);
1749
1750 ret = read;
1751 *ppos += read;
1752
1753 return ret;
1754 }
1755
1756 static int ftrace_event_avail_open(struct inode *inode, struct file *file);
1757 static int ftrace_event_set_open(struct inode *inode, struct file *file);
1758 static int ftrace_event_set_pid_open(struct inode *inode, struct file *file);
1759 static int ftrace_event_release(struct inode *inode, struct file *file);
1760
1761 static const struct seq_operations show_event_seq_ops = {
1762 .start = t_start,
1763 .next = t_next,
1764 .show = t_show,
1765 .stop = t_stop,
1766 };
1767
1768 static const struct seq_operations show_set_event_seq_ops = {
1769 .start = s_start,
1770 .next = s_next,
1771 .show = t_show,
1772 .stop = t_stop,
1773 };
1774
1775 static const struct seq_operations show_set_pid_seq_ops = {
1776 .start = p_start,
1777 .next = p_next,
1778 .show = p_show,
1779 .stop = p_stop,
1780 };
1781
1782 static const struct file_operations ftrace_avail_fops = {
1783 .open = ftrace_event_avail_open,
1784 .read = seq_read,
1785 .llseek = seq_lseek,
1786 .release = seq_release,
1787 };
1788
1789 static const struct file_operations ftrace_set_event_fops = {
1790 .open = ftrace_event_set_open,
1791 .read = seq_read,
1792 .write = ftrace_event_write,
1793 .llseek = seq_lseek,
1794 .release = ftrace_event_release,
1795 };
1796
1797 static const struct file_operations ftrace_set_event_pid_fops = {
1798 .open = ftrace_event_set_pid_open,
1799 .read = seq_read,
1800 .write = ftrace_event_pid_write,
1801 .llseek = seq_lseek,
1802 .release = ftrace_event_release,
1803 };
1804
1805 static const struct file_operations ftrace_enable_fops = {
1806 .open = tracing_open_generic,
1807 .read = event_enable_read,
1808 .write = event_enable_write,
1809 .llseek = default_llseek,
1810 };
1811
1812 static const struct file_operations ftrace_event_format_fops = {
1813 .open = trace_format_open,
1814 .read = seq_read,
1815 .llseek = seq_lseek,
1816 .release = seq_release,
1817 };
1818
1819 static const struct file_operations ftrace_event_id_fops = {
1820 .read = event_id_read,
1821 .llseek = default_llseek,
1822 };
1823
1824 static const struct file_operations ftrace_event_filter_fops = {
1825 .open = tracing_open_generic,
1826 .read = event_filter_read,
1827 .write = event_filter_write,
1828 .llseek = default_llseek,
1829 };
1830
1831 static const struct file_operations ftrace_subsystem_filter_fops = {
1832 .open = subsystem_open,
1833 .read = subsystem_filter_read,
1834 .write = subsystem_filter_write,
1835 .llseek = default_llseek,
1836 .release = subsystem_release,
1837 };
1838
1839 static const struct file_operations ftrace_system_enable_fops = {
1840 .open = subsystem_open,
1841 .read = system_enable_read,
1842 .write = system_enable_write,
1843 .llseek = default_llseek,
1844 .release = subsystem_release,
1845 };
1846
1847 static const struct file_operations ftrace_tr_enable_fops = {
1848 .open = system_tr_open,
1849 .read = system_enable_read,
1850 .write = system_enable_write,
1851 .llseek = default_llseek,
1852 .release = subsystem_release,
1853 };
1854
1855 static const struct file_operations ftrace_show_header_fops = {
1856 .open = tracing_open_generic,
1857 .read = show_header,
1858 .llseek = default_llseek,
1859 };
1860
1861 static int
1862 ftrace_event_open(struct inode *inode, struct file *file,
1863 const struct seq_operations *seq_ops)
1864 {
1865 struct seq_file *m;
1866 int ret;
1867
1868 ret = seq_open(file, seq_ops);
1869 if (ret < 0)
1870 return ret;
1871 m = file->private_data;
1872 /* copy tr over to seq ops */
1873 m->private = inode->i_private;
1874
1875 return ret;
1876 }
1877
1878 static int ftrace_event_release(struct inode *inode, struct file *file)
1879 {
1880 struct trace_array *tr = inode->i_private;
1881
1882 trace_array_put(tr);
1883
1884 return seq_release(inode, file);
1885 }
1886
1887 static int
1888 ftrace_event_avail_open(struct inode *inode, struct file *file)
1889 {
1890 const struct seq_operations *seq_ops = &show_event_seq_ops;
1891
1892 return ftrace_event_open(inode, file, seq_ops);
1893 }
1894
1895 static int
1896 ftrace_event_set_open(struct inode *inode, struct file *file)
1897 {
1898 const struct seq_operations *seq_ops = &show_set_event_seq_ops;
1899 struct trace_array *tr = inode->i_private;
1900 int ret;
1901
1902 if (trace_array_get(tr) < 0)
1903 return -ENODEV;
1904
1905 if ((file->f_mode & FMODE_WRITE) &&
1906 (file->f_flags & O_TRUNC))
1907 ftrace_clear_events(tr);
1908
1909 ret = ftrace_event_open(inode, file, seq_ops);
1910 if (ret < 0)
1911 trace_array_put(tr);
1912 return ret;
1913 }
1914
1915 static int
1916 ftrace_event_set_pid_open(struct inode *inode, struct file *file)
1917 {
1918 const struct seq_operations *seq_ops = &show_set_pid_seq_ops;
1919 struct trace_array *tr = inode->i_private;
1920 int ret;
1921
1922 if (trace_array_get(tr) < 0)
1923 return -ENODEV;
1924
1925 if ((file->f_mode & FMODE_WRITE) &&
1926 (file->f_flags & O_TRUNC))
1927 ftrace_clear_event_pids(tr);
1928
1929 ret = ftrace_event_open(inode, file, seq_ops);
1930 if (ret < 0)
1931 trace_array_put(tr);
1932 return ret;
1933 }
1934
1935 static struct event_subsystem *
1936 create_new_subsystem(const char *name)
1937 {
1938 struct event_subsystem *system;
1939
1940 /* need to create new entry */
1941 system = kmalloc(sizeof(*system), GFP_KERNEL);
1942 if (!system)
1943 return NULL;
1944
1945 system->ref_count = 1;
1946
1947 /* Only allocate if dynamic (kprobes and modules) */
1948 system->name = kstrdup_const(name, GFP_KERNEL);
1949 if (!system->name)
1950 goto out_free;
1951
1952 system->filter = NULL;
1953
1954 system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
1955 if (!system->filter)
1956 goto out_free;
1957
1958 list_add(&system->list, &event_subsystems);
1959
1960 return system;
1961
1962 out_free:
1963 kfree_const(system->name);
1964 kfree(system);
1965 return NULL;
1966 }
1967
1968 static struct dentry *
1969 event_subsystem_dir(struct trace_array *tr, const char *name,
1970 struct trace_event_file *file, struct dentry *parent)
1971 {
1972 struct trace_subsystem_dir *dir;
1973 struct event_subsystem *system;
1974 struct dentry *entry;
1975
1976 /* First see if we did not already create this dir */
1977 list_for_each_entry(dir, &tr->systems, list) {
1978 system = dir->subsystem;
1979 if (strcmp(system->name, name) == 0) {
1980 dir->nr_events++;
1981 file->system = dir;
1982 return dir->entry;
1983 }
1984 }
1985
1986 /* Now see if the system itself exists. */
1987 list_for_each_entry(system, &event_subsystems, list) {
1988 if (strcmp(system->name, name) == 0)
1989 break;
1990 }
1991 /* Reset system variable when not found */
1992 if (&system->list == &event_subsystems)
1993 system = NULL;
1994
1995 dir = kmalloc(sizeof(*dir), GFP_KERNEL);
1996 if (!dir)
1997 goto out_fail;
1998
1999 if (!system) {
2000 system = create_new_subsystem(name);
2001 if (!system)
2002 goto out_free;
2003 } else
2004 __get_system(system);
2005
2006 dir->entry = tracefs_create_dir(name, parent);
2007 if (!dir->entry) {
2008 pr_warn("Failed to create system directory %s\n", name);
2009 __put_system(system);
2010 goto out_free;
2011 }
2012
2013 dir->tr = tr;
2014 dir->ref_count = 1;
2015 dir->nr_events = 1;
2016 dir->subsystem = system;
2017 file->system = dir;
2018
2019 entry = tracefs_create_file("filter", 0644, dir->entry, dir,
2020 &ftrace_subsystem_filter_fops);
2021 if (!entry) {
2022 kfree(system->filter);
2023 system->filter = NULL;
2024 pr_warn("Could not create tracefs '%s/filter' entry\n", name);
2025 }
2026
2027 trace_create_file("enable", 0644, dir->entry, dir,
2028 &ftrace_system_enable_fops);
2029
2030 list_add(&dir->list, &tr->systems);
2031
2032 return dir->entry;
2033
2034 out_free:
2035 kfree(dir);
2036 out_fail:
2037 /* Only print this message if failed on memory allocation */
2038 if (!dir || !system)
2039 pr_warn("No memory to create event subsystem %s\n", name);
2040 return NULL;
2041 }
2042
2043 static int
2044 event_create_dir(struct dentry *parent, struct trace_event_file *file)
2045 {
2046 struct trace_event_call *call = file->event_call;
2047 struct trace_array *tr = file->tr;
2048 struct list_head *head;
2049 struct dentry *d_events;
2050 const char *name;
2051 int ret;
2052
2053 /*
2054 * If the trace point header did not define TRACE_SYSTEM
2055 * then the system would be called "TRACE_SYSTEM".
2056 */
2057 if (strcmp(call->class->system, TRACE_SYSTEM) != 0) {
2058 d_events = event_subsystem_dir(tr, call->class->system, file, parent);
2059 if (!d_events)
2060 return -ENOMEM;
2061 } else
2062 d_events = parent;
2063
2064 name = trace_event_name(call);
2065 file->dir = tracefs_create_dir(name, d_events);
2066 if (!file->dir) {
2067 pr_warn("Could not create tracefs '%s' directory\n", name);
2068 return -1;
2069 }
2070
2071 if (call->class->reg && !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
2072 trace_create_file("enable", 0644, file->dir, file,
2073 &ftrace_enable_fops);
2074
2075 #ifdef CONFIG_PERF_EVENTS
2076 if (call->event.type && call->class->reg)
2077 trace_create_file("id", 0444, file->dir,
2078 (void *)(long)call->event.type,
2079 &ftrace_event_id_fops);
2080 #endif
2081
2082 /*
2083 * Other events may have the same class. Only update
2084 * the fields if they are not already defined.
2085 */
2086 head = trace_get_fields(call);
2087 if (list_empty(head)) {
2088 ret = call->class->define_fields(call);
2089 if (ret < 0) {
2090 pr_warn("Could not initialize trace point events/%s\n",
2091 name);
2092 return -1;
2093 }
2094 }
2095 trace_create_file("filter", 0644, file->dir, file,
2096 &ftrace_event_filter_fops);
2097
2098 trace_create_file("trigger", 0644, file->dir, file,
2099 &event_trigger_fops);
2100
2101 trace_create_file("format", 0444, file->dir, call,
2102 &ftrace_event_format_fops);
2103
2104 return 0;
2105 }
2106
2107 static void remove_event_from_tracers(struct trace_event_call *call)
2108 {
2109 struct trace_event_file *file;
2110 struct trace_array *tr;
2111
2112 do_for_each_event_file_safe(tr, file) {
2113 if (file->event_call != call)
2114 continue;
2115
2116 remove_event_file_dir(file);
2117 /*
2118 * The do_for_each_event_file_safe() is
2119 * a double loop. After finding the call for this
2120 * trace_array, we use break to jump to the next
2121 * trace_array.
2122 */
2123 break;
2124 } while_for_each_event_file();
2125 }
2126
2127 static void event_remove(struct trace_event_call *call)
2128 {
2129 struct trace_array *tr;
2130 struct trace_event_file *file;
2131
2132 do_for_each_event_file(tr, file) {
2133 if (file->event_call != call)
2134 continue;
2135 ftrace_event_enable_disable(file, 0);
2136 /*
2137 * The do_for_each_event_file() is
2138 * a double loop. After finding the call for this
2139 * trace_array, we use break to jump to the next
2140 * trace_array.
2141 */
2142 break;
2143 } while_for_each_event_file();
2144
2145 if (call->event.funcs)
2146 __unregister_trace_event(&call->event);
2147 remove_event_from_tracers(call);
2148 list_del(&call->list);
2149 }
2150
2151 static int event_init(struct trace_event_call *call)
2152 {
2153 int ret = 0;
2154 const char *name;
2155
2156 name = trace_event_name(call);
2157 if (WARN_ON(!name))
2158 return -EINVAL;
2159
2160 if (call->class->raw_init) {
2161 ret = call->class->raw_init(call);
2162 if (ret < 0 && ret != -ENOSYS)
2163 pr_warn("Could not initialize trace events/%s\n", name);
2164 }
2165
2166 return ret;
2167 }
2168
2169 static int
2170 __register_event(struct trace_event_call *call, struct module *mod)
2171 {
2172 int ret;
2173
2174 ret = event_init(call);
2175 if (ret < 0)
2176 return ret;
2177
2178 list_add(&call->list, &ftrace_events);
2179 call->mod = mod;
2180
2181 return 0;
2182 }
2183
2184 static char *enum_replace(char *ptr, struct trace_enum_map *map, int len)
2185 {
2186 int rlen;
2187 int elen;
2188
2189 /* Find the length of the enum value as a string */
2190 elen = snprintf(ptr, 0, "%ld", map->enum_value);
2191 /* Make sure there's enough room to replace the string with the value */
2192 if (len < elen)
2193 return NULL;
2194
2195 snprintf(ptr, elen + 1, "%ld", map->enum_value);
2196
2197 /* Get the rest of the string of ptr */
2198 rlen = strlen(ptr + len);
2199 memmove(ptr + elen, ptr + len, rlen);
2200 /* Make sure we end the new string */
2201 ptr[elen + rlen] = 0;
2202
2203 return ptr + elen;
2204 }
2205
2206 static void update_event_printk(struct trace_event_call *call,
2207 struct trace_enum_map *map)
2208 {
2209 char *ptr;
2210 int quote = 0;
2211 int len = strlen(map->enum_string);
2212
2213 for (ptr = call->print_fmt; *ptr; ptr++) {
2214 if (*ptr == '\\') {
2215 ptr++;
2216 /* paranoid */
2217 if (!*ptr)
2218 break;
2219 continue;
2220 }
2221 if (*ptr == '"') {
2222 quote ^= 1;
2223 continue;
2224 }
2225 if (quote)
2226 continue;
2227 if (isdigit(*ptr)) {
2228 /* skip numbers */
2229 do {
2230 ptr++;
2231 /* Check for alpha chars like ULL */
2232 } while (isalnum(*ptr));
2233 if (!*ptr)
2234 break;
2235 /*
2236 * A number must have some kind of delimiter after
2237 * it, and we can ignore that too.
2238 */
2239 continue;
2240 }
2241 if (isalpha(*ptr) || *ptr == '_') {
2242 if (strncmp(map->enum_string, ptr, len) == 0 &&
2243 !isalnum(ptr[len]) && ptr[len] != '_') {
2244 ptr = enum_replace(ptr, map, len);
2245 /* Hmm, enum string smaller than value */
2246 if (WARN_ON_ONCE(!ptr))
2247 return;
2248 /*
2249 * No need to decrement here, as enum_replace()
2250 * returns the pointer to the character passed
2251 * the enum, and two enums can not be placed
2252 * back to back without something in between.
2253 * We can skip that something in between.
2254 */
2255 continue;
2256 }
2257 skip_more:
2258 do {
2259 ptr++;
2260 } while (isalnum(*ptr) || *ptr == '_');
2261 if (!*ptr)
2262 break;
2263 /*
2264 * If what comes after this variable is a '.' or
2265 * '->' then we can continue to ignore that string.
2266 */
2267 if (*ptr == '.' || (ptr[0] == '-' && ptr[1] == '>')) {
2268 ptr += *ptr == '.' ? 1 : 2;
2269 if (!*ptr)
2270 break;
2271 goto skip_more;
2272 }
2273 /*
2274 * Once again, we can skip the delimiter that came
2275 * after the string.
2276 */
2277 continue;
2278 }
2279 }
2280 }
2281
2282 void trace_event_enum_update(struct trace_enum_map **map, int len)
2283 {
2284 struct trace_event_call *call, *p;
2285 const char *last_system = NULL;
2286 int last_i;
2287 int i;
2288
2289 down_write(&trace_event_sem);
2290 list_for_each_entry_safe(call, p, &ftrace_events, list) {
2291 /* events are usually grouped together with systems */
2292 if (!last_system || call->class->system != last_system) {
2293 last_i = 0;
2294 last_system = call->class->system;
2295 }
2296
2297 for (i = last_i; i < len; i++) {
2298 if (call->class->system == map[i]->system) {
2299 /* Save the first system if need be */
2300 if (!last_i)
2301 last_i = i;
2302 update_event_printk(call, map[i]);
2303 }
2304 }
2305 }
2306 up_write(&trace_event_sem);
2307 }
2308
2309 static struct trace_event_file *
2310 trace_create_new_event(struct trace_event_call *call,
2311 struct trace_array *tr)
2312 {
2313 struct trace_event_file *file;
2314
2315 file = kmem_cache_alloc(file_cachep, GFP_TRACE);
2316 if (!file)
2317 return NULL;
2318
2319 file->event_call = call;
2320 file->tr = tr;
2321 atomic_set(&file->sm_ref, 0);
2322 atomic_set(&file->tm_ref, 0);
2323 INIT_LIST_HEAD(&file->triggers);
2324 list_add(&file->list, &tr->events);
2325
2326 return file;
2327 }
2328
2329 /* Add an event to a trace directory */
2330 static int
2331 __trace_add_new_event(struct trace_event_call *call, struct trace_array *tr)
2332 {
2333 struct trace_event_file *file;
2334
2335 file = trace_create_new_event(call, tr);
2336 if (!file)
2337 return -ENOMEM;
2338
2339 return event_create_dir(tr->event_dir, file);
2340 }
2341
2342 /*
2343 * Just create a decriptor for early init. A descriptor is required
2344 * for enabling events at boot. We want to enable events before
2345 * the filesystem is initialized.
2346 */
2347 static __init int
2348 __trace_early_add_new_event(struct trace_event_call *call,
2349 struct trace_array *tr)
2350 {
2351 struct trace_event_file *file;
2352
2353 file = trace_create_new_event(call, tr);
2354 if (!file)
2355 return -ENOMEM;
2356
2357 return 0;
2358 }
2359
2360 struct ftrace_module_file_ops;
2361 static void __add_event_to_tracers(struct trace_event_call *call);
2362
2363 /* Add an additional event_call dynamically */
2364 int trace_add_event_call(struct trace_event_call *call)
2365 {
2366 int ret;
2367 mutex_lock(&trace_types_lock);
2368 mutex_lock(&event_mutex);
2369
2370 ret = __register_event(call, NULL);
2371 if (ret >= 0)
2372 __add_event_to_tracers(call);
2373
2374 mutex_unlock(&event_mutex);
2375 mutex_unlock(&trace_types_lock);
2376 return ret;
2377 }
2378
2379 /*
2380 * Must be called under locking of trace_types_lock, event_mutex and
2381 * trace_event_sem.
2382 */
2383 static void __trace_remove_event_call(struct trace_event_call *call)
2384 {
2385 event_remove(call);
2386 trace_destroy_fields(call);
2387 free_event_filter(call->filter);
2388 call->filter = NULL;
2389 }
2390
2391 static int probe_remove_event_call(struct trace_event_call *call)
2392 {
2393 struct trace_array *tr;
2394 struct trace_event_file *file;
2395
2396 #ifdef CONFIG_PERF_EVENTS
2397 if (call->perf_refcount)
2398 return -EBUSY;
2399 #endif
2400 do_for_each_event_file(tr, file) {
2401 if (file->event_call != call)
2402 continue;
2403 /*
2404 * We can't rely on ftrace_event_enable_disable(enable => 0)
2405 * we are going to do, EVENT_FILE_FL_SOFT_MODE can suppress
2406 * TRACE_REG_UNREGISTER.
2407 */
2408 if (file->flags & EVENT_FILE_FL_ENABLED)
2409 return -EBUSY;
2410 /*
2411 * The do_for_each_event_file_safe() is
2412 * a double loop. After finding the call for this
2413 * trace_array, we use break to jump to the next
2414 * trace_array.
2415 */
2416 break;
2417 } while_for_each_event_file();
2418
2419 __trace_remove_event_call(call);
2420
2421 return 0;
2422 }
2423
2424 /* Remove an event_call */
2425 int trace_remove_event_call(struct trace_event_call *call)
2426 {
2427 int ret;
2428
2429 mutex_lock(&trace_types_lock);
2430 mutex_lock(&event_mutex);
2431 down_write(&trace_event_sem);
2432 ret = probe_remove_event_call(call);
2433 up_write(&trace_event_sem);
2434 mutex_unlock(&event_mutex);
2435 mutex_unlock(&trace_types_lock);
2436
2437 return ret;
2438 }
2439
2440 #define for_each_event(event, start, end) \
2441 for (event = start; \
2442 (unsigned long)event < (unsigned long)end; \
2443 event++)
2444
2445 #ifdef CONFIG_MODULES
2446
2447 static void trace_module_add_events(struct module *mod)
2448 {
2449 struct trace_event_call **call, **start, **end;
2450
2451 if (!mod->num_trace_events)
2452 return;
2453
2454 /* Don't add infrastructure for mods without tracepoints */
2455 if (trace_module_has_bad_taint(mod)) {
2456 pr_err("%s: module has bad taint, not creating trace events\n",
2457 mod->name);
2458 return;
2459 }
2460
2461 start = mod->trace_events;
2462 end = mod->trace_events + mod->num_trace_events;
2463
2464 for_each_event(call, start, end) {
2465 __register_event(*call, mod);
2466 __add_event_to_tracers(*call);
2467 }
2468 }
2469
2470 static void trace_module_remove_events(struct module *mod)
2471 {
2472 struct trace_event_call *call, *p;
2473 bool clear_trace = false;
2474
2475 down_write(&trace_event_sem);
2476 list_for_each_entry_safe(call, p, &ftrace_events, list) {
2477 if (call->mod == mod) {
2478 if (call->flags & TRACE_EVENT_FL_WAS_ENABLED)
2479 clear_trace = true;
2480 __trace_remove_event_call(call);
2481 }
2482 }
2483 up_write(&trace_event_sem);
2484
2485 /*
2486 * It is safest to reset the ring buffer if the module being unloaded
2487 * registered any events that were used. The only worry is if
2488 * a new module gets loaded, and takes on the same id as the events
2489 * of this module. When printing out the buffer, traced events left
2490 * over from this module may be passed to the new module events and
2491 * unexpected results may occur.
2492 */
2493 if (clear_trace)
2494 tracing_reset_all_online_cpus();
2495 }
2496
2497 static int trace_module_notify(struct notifier_block *self,
2498 unsigned long val, void *data)
2499 {
2500 struct module *mod = data;
2501
2502 mutex_lock(&trace_types_lock);
2503 mutex_lock(&event_mutex);
2504 switch (val) {
2505 case MODULE_STATE_COMING:
2506 trace_module_add_events(mod);
2507 break;
2508 case MODULE_STATE_GOING:
2509 trace_module_remove_events(mod);
2510 break;
2511 }
2512 mutex_unlock(&event_mutex);
2513 mutex_unlock(&trace_types_lock);
2514
2515 return 0;
2516 }
2517
2518 static struct notifier_block trace_module_nb = {
2519 .notifier_call = trace_module_notify,
2520 .priority = 1, /* higher than trace.c module notify */
2521 };
2522 #endif /* CONFIG_MODULES */
2523
2524 /* Create a new event directory structure for a trace directory. */
2525 static void
2526 __trace_add_event_dirs(struct trace_array *tr)
2527 {
2528 struct trace_event_call *call;
2529 int ret;
2530
2531 list_for_each_entry(call, &ftrace_events, list) {
2532 ret = __trace_add_new_event(call, tr);
2533 if (ret < 0)
2534 pr_warn("Could not create directory for event %s\n",
2535 trace_event_name(call));
2536 }
2537 }
2538
2539 struct trace_event_file *
2540 find_event_file(struct trace_array *tr, const char *system, const char *event)
2541 {
2542 struct trace_event_file *file;
2543 struct trace_event_call *call;
2544 const char *name;
2545
2546 list_for_each_entry(file, &tr->events, list) {
2547
2548 call = file->event_call;
2549 name = trace_event_name(call);
2550
2551 if (!name || !call->class || !call->class->reg)
2552 continue;
2553
2554 if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
2555 continue;
2556
2557 if (strcmp(event, name) == 0 &&
2558 strcmp(system, call->class->system) == 0)
2559 return file;
2560 }
2561 return NULL;
2562 }
2563
2564 #ifdef CONFIG_DYNAMIC_FTRACE
2565
2566 /* Avoid typos */
2567 #define ENABLE_EVENT_STR "enable_event"
2568 #define DISABLE_EVENT_STR "disable_event"
2569
2570 struct event_probe_data {
2571 struct trace_event_file *file;
2572 unsigned long count;
2573 int ref;
2574 bool enable;
2575 };
2576
2577 static void
2578 event_enable_probe(unsigned long ip, unsigned long parent_ip, void **_data)
2579 {
2580 struct event_probe_data **pdata = (struct event_probe_data **)_data;
2581 struct event_probe_data *data = *pdata;
2582
2583 if (!data)
2584 return;
2585
2586 if (data->enable)
2587 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
2588 else
2589 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
2590 }
2591
2592 static void
2593 event_enable_count_probe(unsigned long ip, unsigned long parent_ip, void **_data)
2594 {
2595 struct event_probe_data **pdata = (struct event_probe_data **)_data;
2596 struct event_probe_data *data = *pdata;
2597
2598 if (!data)
2599 return;
2600
2601 if (!data->count)
2602 return;
2603
2604 /* Skip if the event is in a state we want to switch to */
2605 if (data->enable == !(data->file->flags & EVENT_FILE_FL_SOFT_DISABLED))
2606 return;
2607
2608 if (data->count != -1)
2609 (data->count)--;
2610
2611 event_enable_probe(ip, parent_ip, _data);
2612 }
2613
2614 static int
2615 event_enable_print(struct seq_file *m, unsigned long ip,
2616 struct ftrace_probe_ops *ops, void *_data)
2617 {
2618 struct event_probe_data *data = _data;
2619
2620 seq_printf(m, "%ps:", (void *)ip);
2621
2622 seq_printf(m, "%s:%s:%s",
2623 data->enable ? ENABLE_EVENT_STR : DISABLE_EVENT_STR,
2624 data->file->event_call->class->system,
2625 trace_event_name(data->file->event_call));
2626
2627 if (data->count == -1)
2628 seq_puts(m, ":unlimited\n");
2629 else
2630 seq_printf(m, ":count=%ld\n", data->count);
2631
2632 return 0;
2633 }
2634
2635 static int
2636 event_enable_init(struct ftrace_probe_ops *ops, unsigned long ip,
2637 void **_data)
2638 {
2639 struct event_probe_data **pdata = (struct event_probe_data **)_data;
2640 struct event_probe_data *data = *pdata;
2641
2642 data->ref++;
2643 return 0;
2644 }
2645
2646 static void
2647 event_enable_free(struct ftrace_probe_ops *ops, unsigned long ip,
2648 void **_data)
2649 {
2650 struct event_probe_data **pdata = (struct event_probe_data **)_data;
2651 struct event_probe_data *data = *pdata;
2652
2653 if (WARN_ON_ONCE(data->ref <= 0))
2654 return;
2655
2656 data->ref--;
2657 if (!data->ref) {
2658 /* Remove the SOFT_MODE flag */
2659 __ftrace_event_enable_disable(data->file, 0, 1);
2660 module_put(data->file->event_call->mod);
2661 kfree(data);
2662 }
2663 *pdata = NULL;
2664 }
2665
2666 static struct ftrace_probe_ops event_enable_probe_ops = {
2667 .func = event_enable_probe,
2668 .print = event_enable_print,
2669 .init = event_enable_init,
2670 .free = event_enable_free,
2671 };
2672
2673 static struct ftrace_probe_ops event_enable_count_probe_ops = {
2674 .func = event_enable_count_probe,
2675 .print = event_enable_print,
2676 .init = event_enable_init,
2677 .free = event_enable_free,
2678 };
2679
2680 static struct ftrace_probe_ops event_disable_probe_ops = {
2681 .func = event_enable_probe,
2682 .print = event_enable_print,
2683 .init = event_enable_init,
2684 .free = event_enable_free,
2685 };
2686
2687 static struct ftrace_probe_ops event_disable_count_probe_ops = {
2688 .func = event_enable_count_probe,
2689 .print = event_enable_print,
2690 .init = event_enable_init,
2691 .free = event_enable_free,
2692 };
2693
2694 static int
2695 event_enable_func(struct ftrace_hash *hash,
2696 char *glob, char *cmd, char *param, int enabled)
2697 {
2698 struct trace_array *tr = top_trace_array();
2699 struct trace_event_file *file;
2700 struct ftrace_probe_ops *ops;
2701 struct event_probe_data *data;
2702 const char *system;
2703 const char *event;
2704 char *number;
2705 bool enable;
2706 int ret;
2707
2708 if (!tr)
2709 return -ENODEV;
2710
2711 /* hash funcs only work with set_ftrace_filter */
2712 if (!enabled || !param)
2713 return -EINVAL;
2714
2715 system = strsep(&param, ":");
2716 if (!param)
2717 return -EINVAL;
2718
2719 event = strsep(&param, ":");
2720
2721 mutex_lock(&event_mutex);
2722
2723 ret = -EINVAL;
2724 file = find_event_file(tr, system, event);
2725 if (!file)
2726 goto out;
2727
2728 enable = strcmp(cmd, ENABLE_EVENT_STR) == 0;
2729
2730 if (enable)
2731 ops = param ? &event_enable_count_probe_ops : &event_enable_probe_ops;
2732 else
2733 ops = param ? &event_disable_count_probe_ops : &event_disable_probe_ops;
2734
2735 if (glob[0] == '!') {
2736 unregister_ftrace_function_probe_func(glob+1, ops);
2737 ret = 0;
2738 goto out;
2739 }
2740
2741 ret = -ENOMEM;
2742 data = kzalloc(sizeof(*data), GFP_KERNEL);
2743 if (!data)
2744 goto out;
2745
2746 data->enable = enable;
2747 data->count = -1;
2748 data->file = file;
2749
2750 if (!param)
2751 goto out_reg;
2752
2753 number = strsep(&param, ":");
2754
2755 ret = -EINVAL;
2756 if (!strlen(number))
2757 goto out_free;
2758
2759 /*
2760 * We use the callback data field (which is a pointer)
2761 * as our counter.
2762 */
2763 ret = kstrtoul(number, 0, &data->count);
2764 if (ret)
2765 goto out_free;
2766
2767 out_reg:
2768 /* Don't let event modules unload while probe registered */
2769 ret = try_module_get(file->event_call->mod);
2770 if (!ret) {
2771 ret = -EBUSY;
2772 goto out_free;
2773 }
2774
2775 ret = __ftrace_event_enable_disable(file, 1, 1);
2776 if (ret < 0)
2777 goto out_put;
2778 ret = register_ftrace_function_probe(glob, ops, data);
2779 /*
2780 * The above returns on success the # of functions enabled,
2781 * but if it didn't find any functions it returns zero.
2782 * Consider no functions a failure too.
2783 */
2784 if (!ret) {
2785 ret = -ENOENT;
2786 goto out_disable;
2787 } else if (ret < 0)
2788 goto out_disable;
2789 /* Just return zero, not the number of enabled functions */
2790 ret = 0;
2791 out:
2792 mutex_unlock(&event_mutex);
2793 return ret;
2794
2795 out_disable:
2796 __ftrace_event_enable_disable(file, 0, 1);
2797 out_put:
2798 module_put(file->event_call->mod);
2799 out_free:
2800 kfree(data);
2801 goto out;
2802 }
2803
2804 static struct ftrace_func_command event_enable_cmd = {
2805 .name = ENABLE_EVENT_STR,
2806 .func = event_enable_func,
2807 };
2808
2809 static struct ftrace_func_command event_disable_cmd = {
2810 .name = DISABLE_EVENT_STR,
2811 .func = event_enable_func,
2812 };
2813
2814 static __init int register_event_cmds(void)
2815 {
2816 int ret;
2817
2818 ret = register_ftrace_command(&event_enable_cmd);
2819 if (WARN_ON(ret < 0))
2820 return ret;
2821 ret = register_ftrace_command(&event_disable_cmd);
2822 if (WARN_ON(ret < 0))
2823 unregister_ftrace_command(&event_enable_cmd);
2824 return ret;
2825 }
2826 #else
2827 static inline int register_event_cmds(void) { return 0; }
2828 #endif /* CONFIG_DYNAMIC_FTRACE */
2829
2830 /*
2831 * The top level array has already had its trace_event_file
2832 * descriptors created in order to allow for early events to
2833 * be recorded. This function is called after the tracefs has been
2834 * initialized, and we now have to create the files associated
2835 * to the events.
2836 */
2837 static __init void
2838 __trace_early_add_event_dirs(struct trace_array *tr)
2839 {
2840 struct trace_event_file *file;
2841 int ret;
2842
2843
2844 list_for_each_entry(file, &tr->events, list) {
2845 ret = event_create_dir(tr->event_dir, file);
2846 if (ret < 0)
2847 pr_warn("Could not create directory for event %s\n",
2848 trace_event_name(file->event_call));
2849 }
2850 }
2851
2852 /*
2853 * For early boot up, the top trace array requires to have
2854 * a list of events that can be enabled. This must be done before
2855 * the filesystem is set up in order to allow events to be traced
2856 * early.
2857 */
2858 static __init void
2859 __trace_early_add_events(struct trace_array *tr)
2860 {
2861 struct trace_event_call *call;
2862 int ret;
2863
2864 list_for_each_entry(call, &ftrace_events, list) {
2865 /* Early boot up should not have any modules loaded */
2866 if (WARN_ON_ONCE(call->mod))
2867 continue;
2868
2869 ret = __trace_early_add_new_event(call, tr);
2870 if (ret < 0)
2871 pr_warn("Could not create early event %s\n",
2872 trace_event_name(call));
2873 }
2874 }
2875
2876 /* Remove the event directory structure for a trace directory. */
2877 static void
2878 __trace_remove_event_dirs(struct trace_array *tr)
2879 {
2880 struct trace_event_file *file, *next;
2881
2882 list_for_each_entry_safe(file, next, &tr->events, list)
2883 remove_event_file_dir(file);
2884 }
2885
2886 static void __add_event_to_tracers(struct trace_event_call *call)
2887 {
2888 struct trace_array *tr;
2889
2890 list_for_each_entry(tr, &ftrace_trace_arrays, list)
2891 __trace_add_new_event(call, tr);
2892 }
2893
2894 extern struct trace_event_call *__start_ftrace_events[];
2895 extern struct trace_event_call *__stop_ftrace_events[];
2896
2897 static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
2898
2899 static __init int setup_trace_event(char *str)
2900 {
2901 strlcpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
2902 ring_buffer_expanded = true;
2903 tracing_selftest_disabled = true;
2904
2905 return 1;
2906 }
2907 __setup("trace_event=", setup_trace_event);
2908
2909 /* Expects to have event_mutex held when called */
2910 static int
2911 create_event_toplevel_files(struct dentry *parent, struct trace_array *tr)
2912 {
2913 struct dentry *d_events;
2914 struct dentry *entry;
2915
2916 entry = tracefs_create_file("set_event", 0644, parent,
2917 tr, &ftrace_set_event_fops);
2918 if (!entry) {
2919 pr_warn("Could not create tracefs 'set_event' entry\n");
2920 return -ENOMEM;
2921 }
2922
2923 d_events = tracefs_create_dir("events", parent);
2924 if (!d_events) {
2925 pr_warn("Could not create tracefs 'events' directory\n");
2926 return -ENOMEM;
2927 }
2928
2929 entry = tracefs_create_file("set_event_pid", 0644, parent,
2930 tr, &ftrace_set_event_pid_fops);
2931
2932 /* ring buffer internal formats */
2933 trace_create_file("header_page", 0444, d_events,
2934 ring_buffer_print_page_header,
2935 &ftrace_show_header_fops);
2936
2937 trace_create_file("header_event", 0444, d_events,
2938 ring_buffer_print_entry_header,
2939 &ftrace_show_header_fops);
2940
2941 trace_create_file("enable", 0644, d_events,
2942 tr, &ftrace_tr_enable_fops);
2943
2944 tr->event_dir = d_events;
2945
2946 return 0;
2947 }
2948
2949 /**
2950 * event_trace_add_tracer - add a instance of a trace_array to events
2951 * @parent: The parent dentry to place the files/directories for events in
2952 * @tr: The trace array associated with these events
2953 *
2954 * When a new instance is created, it needs to set up its events
2955 * directory, as well as other files associated with events. It also
2956 * creates the event hierachry in the @parent/events directory.
2957 *
2958 * Returns 0 on success.
2959 */
2960 int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr)
2961 {
2962 int ret;
2963
2964 mutex_lock(&event_mutex);
2965
2966 ret = create_event_toplevel_files(parent, tr);
2967 if (ret)
2968 goto out_unlock;
2969
2970 down_write(&trace_event_sem);
2971 __trace_add_event_dirs(tr);
2972 up_write(&trace_event_sem);
2973
2974 out_unlock:
2975 mutex_unlock(&event_mutex);
2976
2977 return ret;
2978 }
2979
2980 /*
2981 * The top trace array already had its file descriptors created.
2982 * Now the files themselves need to be created.
2983 */
2984 static __init int
2985 early_event_add_tracer(struct dentry *parent, struct trace_array *tr)
2986 {
2987 int ret;
2988
2989 mutex_lock(&event_mutex);
2990
2991 ret = create_event_toplevel_files(parent, tr);
2992 if (ret)
2993 goto out_unlock;
2994
2995 down_write(&trace_event_sem);
2996 __trace_early_add_event_dirs(tr);
2997 up_write(&trace_event_sem);
2998
2999 out_unlock:
3000 mutex_unlock(&event_mutex);
3001
3002 return ret;
3003 }
3004
3005 int event_trace_del_tracer(struct trace_array *tr)
3006 {
3007 mutex_lock(&event_mutex);
3008
3009 /* Disable any event triggers and associated soft-disabled events */
3010 clear_event_triggers(tr);
3011
3012 /* Clear the pid list */
3013 __ftrace_clear_event_pids(tr);
3014
3015 /* Disable any running events */
3016 __ftrace_set_clr_event_nolock(tr, NULL, NULL, NULL, 0);
3017
3018 /* Access to events are within rcu_read_lock_sched() */
3019 synchronize_sched();
3020
3021 down_write(&trace_event_sem);
3022 __trace_remove_event_dirs(tr);
3023 tracefs_remove_recursive(tr->event_dir);
3024 up_write(&trace_event_sem);
3025
3026 tr->event_dir = NULL;
3027
3028 mutex_unlock(&event_mutex);
3029
3030 return 0;
3031 }
3032
3033 static __init int event_trace_memsetup(void)
3034 {
3035 field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC);
3036 file_cachep = KMEM_CACHE(trace_event_file, SLAB_PANIC);
3037 return 0;
3038 }
3039
3040 static __init void
3041 early_enable_events(struct trace_array *tr, bool disable_first)
3042 {
3043 char *buf = bootup_event_buf;
3044 char *token;
3045 int ret;
3046
3047 while (true) {
3048 token = strsep(&buf, ",");
3049
3050 if (!token)
3051 break;
3052
3053 if (*token) {
3054 /* Restarting syscalls requires that we stop them first */
3055 if (disable_first)
3056 ftrace_set_clr_event(tr, token, 0);
3057
3058 ret = ftrace_set_clr_event(tr, token, 1);
3059 if (ret)
3060 pr_warn("Failed to enable trace event: %s\n", token);
3061 }
3062
3063 /* Put back the comma to allow this to be called again */
3064 if (buf)
3065 *(buf - 1) = ',';
3066 }
3067 }
3068
3069 static __init int event_trace_enable(void)
3070 {
3071 struct trace_array *tr = top_trace_array();
3072 struct trace_event_call **iter, *call;
3073 int ret;
3074
3075 if (!tr)
3076 return -ENODEV;
3077
3078 for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
3079
3080 call = *iter;
3081 ret = event_init(call);
3082 if (!ret)
3083 list_add(&call->list, &ftrace_events);
3084 }
3085
3086 /*
3087 * We need the top trace array to have a working set of trace
3088 * points at early init, before the debug files and directories
3089 * are created. Create the file entries now, and attach them
3090 * to the actual file dentries later.
3091 */
3092 __trace_early_add_events(tr);
3093
3094 early_enable_events(tr, false);
3095
3096 trace_printk_start_comm();
3097
3098 register_event_cmds();
3099
3100 register_trigger_cmds();
3101
3102 return 0;
3103 }
3104
3105 /*
3106 * event_trace_enable() is called from trace_event_init() first to
3107 * initialize events and perhaps start any events that are on the
3108 * command line. Unfortunately, there are some events that will not
3109 * start this early, like the system call tracepoints that need
3110 * to set the TIF_SYSCALL_TRACEPOINT flag of pid 1. But event_trace_enable()
3111 * is called before pid 1 starts, and this flag is never set, making
3112 * the syscall tracepoint never get reached, but the event is enabled
3113 * regardless (and not doing anything).
3114 */
3115 static __init int event_trace_enable_again(void)
3116 {
3117 struct trace_array *tr;
3118
3119 tr = top_trace_array();
3120 if (!tr)
3121 return -ENODEV;
3122
3123 early_enable_events(tr, true);
3124
3125 return 0;
3126 }
3127
3128 early_initcall(event_trace_enable_again);
3129
3130 static __init int event_trace_init(void)
3131 {
3132 struct trace_array *tr;
3133 struct dentry *d_tracer;
3134 struct dentry *entry;
3135 int ret;
3136
3137 tr = top_trace_array();
3138 if (!tr)
3139 return -ENODEV;
3140
3141 d_tracer = tracing_init_dentry();
3142 if (IS_ERR(d_tracer))
3143 return 0;
3144
3145 entry = tracefs_create_file("available_events", 0444, d_tracer,
3146 tr, &ftrace_avail_fops);
3147 if (!entry)
3148 pr_warn("Could not create tracefs 'available_events' entry\n");
3149
3150 if (trace_define_generic_fields())
3151 pr_warn("tracing: Failed to allocated generic fields");
3152
3153 if (trace_define_common_fields())
3154 pr_warn("tracing: Failed to allocate common fields");
3155
3156 ret = early_event_add_tracer(d_tracer, tr);
3157 if (ret)
3158 return ret;
3159
3160 #ifdef CONFIG_MODULES
3161 ret = register_module_notifier(&trace_module_nb);
3162 if (ret)
3163 pr_warn("Failed to register trace events module notifier\n");
3164 #endif
3165 return 0;
3166 }
3167
3168 void __init trace_event_init(void)
3169 {
3170 event_trace_memsetup();
3171 init_ftrace_syscalls();
3172 event_trace_enable();
3173 }
3174
3175 fs_initcall(event_trace_init);
3176
3177 #ifdef CONFIG_FTRACE_STARTUP_TEST
3178
3179 static DEFINE_SPINLOCK(test_spinlock);
3180 static DEFINE_SPINLOCK(test_spinlock_irq);
3181 static DEFINE_MUTEX(test_mutex);
3182
3183 static __init void test_work(struct work_struct *dummy)
3184 {
3185 spin_lock(&test_spinlock);
3186 spin_lock_irq(&test_spinlock_irq);
3187 udelay(1);
3188 spin_unlock_irq(&test_spinlock_irq);
3189 spin_unlock(&test_spinlock);
3190
3191 mutex_lock(&test_mutex);
3192 msleep(1);
3193 mutex_unlock(&test_mutex);
3194 }
3195
3196 static __init int event_test_thread(void *unused)
3197 {
3198 void *test_malloc;
3199
3200 test_malloc = kmalloc(1234, GFP_KERNEL);
3201 if (!test_malloc)
3202 pr_info("failed to kmalloc\n");
3203
3204 schedule_on_each_cpu(test_work);
3205
3206 kfree(test_malloc);
3207
3208 set_current_state(TASK_INTERRUPTIBLE);
3209 while (!kthread_should_stop()) {
3210 schedule();
3211 set_current_state(TASK_INTERRUPTIBLE);
3212 }
3213 __set_current_state(TASK_RUNNING);
3214
3215 return 0;
3216 }
3217
3218 /*
3219 * Do various things that may trigger events.
3220 */
3221 static __init void event_test_stuff(void)
3222 {
3223 struct task_struct *test_thread;
3224
3225 test_thread = kthread_run(event_test_thread, NULL, "test-events");
3226 msleep(1);
3227 kthread_stop(test_thread);
3228 }
3229
3230 /*
3231 * For every trace event defined, we will test each trace point separately,
3232 * and then by groups, and finally all trace points.
3233 */
3234 static __init void event_trace_self_tests(void)
3235 {
3236 struct trace_subsystem_dir *dir;
3237 struct trace_event_file *file;
3238 struct trace_event_call *call;
3239 struct event_subsystem *system;
3240 struct trace_array *tr;
3241 int ret;
3242
3243 tr = top_trace_array();
3244 if (!tr)
3245 return;
3246
3247 pr_info("Running tests on trace events:\n");
3248
3249 list_for_each_entry(file, &tr->events, list) {
3250
3251 call = file->event_call;
3252
3253 /* Only test those that have a probe */
3254 if (!call->class || !call->class->probe)
3255 continue;
3256
3257 /*
3258 * Testing syscall events here is pretty useless, but
3259 * we still do it if configured. But this is time consuming.
3260 * What we really need is a user thread to perform the
3261 * syscalls as we test.
3262 */
3263 #ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
3264 if (call->class->system &&
3265 strcmp(call->class->system, "syscalls") == 0)
3266 continue;
3267 #endif
3268
3269 pr_info("Testing event %s: ", trace_event_name(call));
3270
3271 /*
3272 * If an event is already enabled, someone is using
3273 * it and the self test should not be on.
3274 */
3275 if (file->flags & EVENT_FILE_FL_ENABLED) {
3276 pr_warn("Enabled event during self test!\n");
3277 WARN_ON_ONCE(1);
3278 continue;
3279 }
3280
3281 ftrace_event_enable_disable(file, 1);
3282 event_test_stuff();
3283 ftrace_event_enable_disable(file, 0);
3284
3285 pr_cont("OK\n");
3286 }
3287
3288 /* Now test at the sub system level */
3289
3290 pr_info("Running tests on trace event systems:\n");
3291
3292 list_for_each_entry(dir, &tr->systems, list) {
3293
3294 system = dir->subsystem;
3295
3296 /* the ftrace system is special, skip it */
3297 if (strcmp(system->name, "ftrace") == 0)
3298 continue;
3299
3300 pr_info("Testing event system %s: ", system->name);
3301
3302 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1);
3303 if (WARN_ON_ONCE(ret)) {
3304 pr_warn("error enabling system %s\n",
3305 system->name);
3306 continue;
3307 }
3308
3309 event_test_stuff();
3310
3311 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0);
3312 if (WARN_ON_ONCE(ret)) {
3313 pr_warn("error disabling system %s\n",
3314 system->name);
3315 continue;
3316 }
3317
3318 pr_cont("OK\n");
3319 }
3320
3321 /* Test with all events enabled */
3322
3323 pr_info("Running tests on all trace events:\n");
3324 pr_info("Testing all events: ");
3325
3326 ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1);
3327 if (WARN_ON_ONCE(ret)) {
3328 pr_warn("error enabling all events\n");
3329 return;
3330 }
3331
3332 event_test_stuff();
3333
3334 /* reset sysname */
3335 ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
3336 if (WARN_ON_ONCE(ret)) {
3337 pr_warn("error disabling all events\n");
3338 return;
3339 }
3340
3341 pr_cont("OK\n");
3342 }
3343
3344 #ifdef CONFIG_FUNCTION_TRACER
3345
3346 static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
3347
3348 static struct trace_array *event_tr;
3349
3350 static void __init
3351 function_test_events_call(unsigned long ip, unsigned long parent_ip,
3352 struct ftrace_ops *op, struct pt_regs *pt_regs)
3353 {
3354 struct ring_buffer_event *event;
3355 struct ring_buffer *buffer;
3356 struct ftrace_entry *entry;
3357 unsigned long flags;
3358 long disabled;
3359 int cpu;
3360 int pc;
3361
3362 pc = preempt_count();
3363 preempt_disable_notrace();
3364 cpu = raw_smp_processor_id();
3365 disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
3366
3367 if (disabled != 1)
3368 goto out;
3369
3370 local_save_flags(flags);
3371
3372 event = trace_current_buffer_lock_reserve(&buffer,
3373 TRACE_FN, sizeof(*entry),
3374 flags, pc);
3375 if (!event)
3376 goto out;
3377 entry = ring_buffer_event_data(event);
3378 entry->ip = ip;
3379 entry->parent_ip = parent_ip;
3380
3381 trace_buffer_unlock_commit(event_tr, buffer, event, flags, pc);
3382
3383 out:
3384 atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
3385 preempt_enable_notrace();
3386 }
3387
3388 static struct ftrace_ops trace_ops __initdata =
3389 {
3390 .func = function_test_events_call,
3391 .flags = FTRACE_OPS_FL_RECURSION_SAFE,
3392 };
3393
3394 static __init void event_trace_self_test_with_function(void)
3395 {
3396 int ret;
3397 event_tr = top_trace_array();
3398 if (WARN_ON(!event_tr))
3399 return;
3400 ret = register_ftrace_function(&trace_ops);
3401 if (WARN_ON(ret < 0)) {
3402 pr_info("Failed to enable function tracer for event tests\n");
3403 return;
3404 }
3405 pr_info("Running tests again, along with the function tracer\n");
3406 event_trace_self_tests();
3407 unregister_ftrace_function(&trace_ops);
3408 }
3409 #else
3410 static __init void event_trace_self_test_with_function(void)
3411 {
3412 }
3413 #endif
3414
3415 static __init int event_trace_self_tests_init(void)
3416 {
3417 if (!tracing_selftest_disabled) {
3418 event_trace_self_tests();
3419 event_trace_self_test_with_function();
3420 }
3421
3422 return 0;
3423 }
3424
3425 late_initcall(event_trace_self_tests_init);
3426
3427 #endif
Linux kernel - Deliverables
This page took 0.101477 seconds and 5 git commands to generate.