Merge tag 'nfs-for-6.12-1' of git://git.linux-nfs.org/projects/anna/linux-nfs
[linux-2.6-block.git] / net / rfkill / core.c
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * Copyright (C) 2006 - 2007 Ivo van Doorn
4  * Copyright (C) 2007 Dmitry Torokhov
5  * Copyright 2009 Johannes Berg <johannes@sipsolutions.net>
6  */
7
8 #include <linux/kernel.h>
9 #include <linux/module.h>
10 #include <linux/init.h>
11 #include <linux/workqueue.h>
12 #include <linux/capability.h>
13 #include <linux/list.h>
14 #include <linux/mutex.h>
15 #include <linux/rfkill.h>
16 #include <linux/sched.h>
17 #include <linux/spinlock.h>
18 #include <linux/device.h>
19 #include <linux/miscdevice.h>
20 #include <linux/wait.h>
21 #include <linux/poll.h>
22 #include <linux/fs.h>
23 #include <linux/slab.h>
24
25 #include "rfkill.h"
26
27 #define POLL_INTERVAL           (5 * HZ)
28
29 #define RFKILL_BLOCK_HW         BIT(0)
30 #define RFKILL_BLOCK_SW         BIT(1)
31 #define RFKILL_BLOCK_SW_PREV    BIT(2)
32 #define RFKILL_BLOCK_ANY        (RFKILL_BLOCK_HW |\
33                                  RFKILL_BLOCK_SW |\
34                                  RFKILL_BLOCK_SW_PREV)
35 #define RFKILL_BLOCK_SW_SETCALL BIT(31)
36
37 struct rfkill {
38         spinlock_t              lock;
39
40         enum rfkill_type        type;
41
42         unsigned long           state;
43         unsigned long           hard_block_reasons;
44
45         u32                     idx;
46
47         bool                    registered;
48         bool                    persistent;
49         bool                    polling_paused;
50         bool                    suspended;
51         bool                    need_sync;
52
53         const struct rfkill_ops *ops;
54         void                    *data;
55
56 #ifdef CONFIG_RFKILL_LEDS
57         struct led_trigger      led_trigger;
58         const char              *ledtrigname;
59 #endif
60
61         struct device           dev;
62         struct list_head        node;
63
64         struct delayed_work     poll_work;
65         struct work_struct      uevent_work;
66         struct work_struct      sync_work;
67         char                    name[];
68 };
69 #define to_rfkill(d)    container_of(d, struct rfkill, dev)
70
71 struct rfkill_int_event {
72         struct list_head        list;
73         struct rfkill_event_ext ev;
74 };
75
76 struct rfkill_data {
77         struct list_head        list;
78         struct list_head        events;
79         struct mutex            mtx;
80         wait_queue_head_t       read_wait;
81         bool                    input_handler;
82         u8                      max_size;
83 };
84
85
86 MODULE_AUTHOR("Ivo van Doorn <IvDoorn@gmail.com>");
87 MODULE_AUTHOR("Johannes Berg <johannes@sipsolutions.net>");
88 MODULE_DESCRIPTION("RF switch support");
89 MODULE_LICENSE("GPL");
90
91
92 /*
93  * The locking here should be made much smarter, we currently have
94  * a bit of a stupid situation because drivers might want to register
95  * the rfkill struct under their own lock, and take this lock during
96  * rfkill method calls -- which will cause an AB-BA deadlock situation.
97  *
98  * To fix that, we need to rework this code here to be mostly lock-free
99  * and only use the mutex for list manipulations, not to protect the
100  * various other global variables. Then we can avoid holding the mutex
101  * around driver operations, and all is happy.
102  */
103 static LIST_HEAD(rfkill_list);  /* list of registered rf switches */
104 static DEFINE_MUTEX(rfkill_global_mutex);
105 static LIST_HEAD(rfkill_fds);   /* list of open fds of /dev/rfkill */
106
107 static unsigned int rfkill_default_state = 1;
108 module_param_named(default_state, rfkill_default_state, uint, 0444);
109 MODULE_PARM_DESC(default_state,
110                  "Default initial state for all radio types, 0 = radio off");
111
112 static struct {
113         bool cur, sav;
114 } rfkill_global_states[NUM_RFKILL_TYPES];
115
116 static bool rfkill_epo_lock_active;
117
118
119 #ifdef CONFIG_RFKILL_LEDS
120 static void rfkill_led_trigger_event(struct rfkill *rfkill)
121 {
122         struct led_trigger *trigger;
123
124         if (!rfkill->registered)
125                 return;
126
127         trigger = &rfkill->led_trigger;
128
129         if (rfkill->state & RFKILL_BLOCK_ANY)
130                 led_trigger_event(trigger, LED_OFF);
131         else
132                 led_trigger_event(trigger, LED_FULL);
133 }
134
135 static int rfkill_led_trigger_activate(struct led_classdev *led)
136 {
137         struct rfkill *rfkill;
138
139         rfkill = container_of(led->trigger, struct rfkill, led_trigger);
140
141         rfkill_led_trigger_event(rfkill);
142
143         return 0;
144 }
145
146 const char *rfkill_get_led_trigger_name(struct rfkill *rfkill)
147 {
148         return rfkill->led_trigger.name;
149 }
150 EXPORT_SYMBOL(rfkill_get_led_trigger_name);
151
152 void rfkill_set_led_trigger_name(struct rfkill *rfkill, const char *name)
153 {
154         BUG_ON(!rfkill);
155
156         rfkill->ledtrigname = name;
157 }
158 EXPORT_SYMBOL(rfkill_set_led_trigger_name);
159
160 static int rfkill_led_trigger_register(struct rfkill *rfkill)
161 {
162         rfkill->led_trigger.name = rfkill->ledtrigname
163                                         ? : dev_name(&rfkill->dev);
164         rfkill->led_trigger.activate = rfkill_led_trigger_activate;
165         return led_trigger_register(&rfkill->led_trigger);
166 }
167
168 static void rfkill_led_trigger_unregister(struct rfkill *rfkill)
169 {
170         led_trigger_unregister(&rfkill->led_trigger);
171 }
172
173 static struct led_trigger rfkill_any_led_trigger;
174 static struct led_trigger rfkill_none_led_trigger;
175 static struct work_struct rfkill_global_led_trigger_work;
176
177 static void rfkill_global_led_trigger_worker(struct work_struct *work)
178 {
179         enum led_brightness brightness = LED_OFF;
180         struct rfkill *rfkill;
181
182         mutex_lock(&rfkill_global_mutex);
183         list_for_each_entry(rfkill, &rfkill_list, node) {
184                 if (!(rfkill->state & RFKILL_BLOCK_ANY)) {
185                         brightness = LED_FULL;
186                         break;
187                 }
188         }
189         mutex_unlock(&rfkill_global_mutex);
190
191         led_trigger_event(&rfkill_any_led_trigger, brightness);
192         led_trigger_event(&rfkill_none_led_trigger,
193                           brightness == LED_OFF ? LED_FULL : LED_OFF);
194 }
195
196 static void rfkill_global_led_trigger_event(void)
197 {
198         schedule_work(&rfkill_global_led_trigger_work);
199 }
200
201 static int rfkill_global_led_trigger_register(void)
202 {
203         int ret;
204
205         INIT_WORK(&rfkill_global_led_trigger_work,
206                         rfkill_global_led_trigger_worker);
207
208         rfkill_any_led_trigger.name = "rfkill-any";
209         ret = led_trigger_register(&rfkill_any_led_trigger);
210         if (ret)
211                 return ret;
212
213         rfkill_none_led_trigger.name = "rfkill-none";
214         ret = led_trigger_register(&rfkill_none_led_trigger);
215         if (ret)
216                 led_trigger_unregister(&rfkill_any_led_trigger);
217         else
218                 /* Delay activation until all global triggers are registered */
219                 rfkill_global_led_trigger_event();
220
221         return ret;
222 }
223
224 static void rfkill_global_led_trigger_unregister(void)
225 {
226         led_trigger_unregister(&rfkill_none_led_trigger);
227         led_trigger_unregister(&rfkill_any_led_trigger);
228         cancel_work_sync(&rfkill_global_led_trigger_work);
229 }
230 #else
231 static void rfkill_led_trigger_event(struct rfkill *rfkill)
232 {
233 }
234
235 static inline int rfkill_led_trigger_register(struct rfkill *rfkill)
236 {
237         return 0;
238 }
239
240 static inline void rfkill_led_trigger_unregister(struct rfkill *rfkill)
241 {
242 }
243
244 static void rfkill_global_led_trigger_event(void)
245 {
246 }
247
248 static int rfkill_global_led_trigger_register(void)
249 {
250         return 0;
251 }
252
253 static void rfkill_global_led_trigger_unregister(void)
254 {
255 }
256 #endif /* CONFIG_RFKILL_LEDS */
257
258 static void rfkill_fill_event(struct rfkill_event_ext *ev,
259                               struct rfkill *rfkill,
260                               enum rfkill_operation op)
261 {
262         unsigned long flags;
263
264         ev->idx = rfkill->idx;
265         ev->type = rfkill->type;
266         ev->op = op;
267
268         spin_lock_irqsave(&rfkill->lock, flags);
269         ev->hard = !!(rfkill->state & RFKILL_BLOCK_HW);
270         ev->soft = !!(rfkill->state & (RFKILL_BLOCK_SW |
271                                         RFKILL_BLOCK_SW_PREV));
272         ev->hard_block_reasons = rfkill->hard_block_reasons;
273         spin_unlock_irqrestore(&rfkill->lock, flags);
274 }
275
276 static void rfkill_send_events(struct rfkill *rfkill, enum rfkill_operation op)
277 {
278         struct rfkill_data *data;
279         struct rfkill_int_event *ev;
280
281         list_for_each_entry(data, &rfkill_fds, list) {
282                 ev = kzalloc(sizeof(*ev), GFP_KERNEL);
283                 if (!ev)
284                         continue;
285                 rfkill_fill_event(&ev->ev, rfkill, op);
286                 mutex_lock(&data->mtx);
287                 list_add_tail(&ev->list, &data->events);
288                 mutex_unlock(&data->mtx);
289                 wake_up_interruptible(&data->read_wait);
290         }
291 }
292
293 static void rfkill_event(struct rfkill *rfkill)
294 {
295         if (!rfkill->registered)
296                 return;
297
298         kobject_uevent(&rfkill->dev.kobj, KOBJ_CHANGE);
299
300         /* also send event to /dev/rfkill */
301         rfkill_send_events(rfkill, RFKILL_OP_CHANGE);
302 }
303
304 /**
305  * rfkill_set_block - wrapper for set_block method
306  *
307  * @rfkill: the rfkill struct to use
308  * @blocked: the new software state
309  *
310  * Calls the set_block method (when applicable) and handles notifications
311  * etc. as well.
312  */
313 static void rfkill_set_block(struct rfkill *rfkill, bool blocked)
314 {
315         unsigned long flags;
316         bool prev, curr;
317         int err;
318
319         if (unlikely(rfkill->dev.power.power_state.event & PM_EVENT_SLEEP))
320                 return;
321
322         /*
323          * Some platforms (...!) generate input events which affect the
324          * _hard_ kill state -- whenever something tries to change the
325          * current software state query the hardware state too.
326          */
327         if (rfkill->ops->query)
328                 rfkill->ops->query(rfkill, rfkill->data);
329
330         spin_lock_irqsave(&rfkill->lock, flags);
331         prev = rfkill->state & RFKILL_BLOCK_SW;
332
333         if (prev)
334                 rfkill->state |= RFKILL_BLOCK_SW_PREV;
335         else
336                 rfkill->state &= ~RFKILL_BLOCK_SW_PREV;
337
338         if (blocked)
339                 rfkill->state |= RFKILL_BLOCK_SW;
340         else
341                 rfkill->state &= ~RFKILL_BLOCK_SW;
342
343         rfkill->state |= RFKILL_BLOCK_SW_SETCALL;
344         spin_unlock_irqrestore(&rfkill->lock, flags);
345
346         err = rfkill->ops->set_block(rfkill->data, blocked);
347
348         spin_lock_irqsave(&rfkill->lock, flags);
349         if (err) {
350                 /*
351                  * Failed -- reset status to _PREV, which may be different
352                  * from what we have set _PREV to earlier in this function
353                  * if rfkill_set_sw_state was invoked.
354                  */
355                 if (rfkill->state & RFKILL_BLOCK_SW_PREV)
356                         rfkill->state |= RFKILL_BLOCK_SW;
357                 else
358                         rfkill->state &= ~RFKILL_BLOCK_SW;
359         }
360         rfkill->state &= ~RFKILL_BLOCK_SW_SETCALL;
361         rfkill->state &= ~RFKILL_BLOCK_SW_PREV;
362         curr = rfkill->state & RFKILL_BLOCK_SW;
363         spin_unlock_irqrestore(&rfkill->lock, flags);
364
365         rfkill_led_trigger_event(rfkill);
366         rfkill_global_led_trigger_event();
367
368         if (prev != curr)
369                 rfkill_event(rfkill);
370 }
371
372 static void rfkill_sync(struct rfkill *rfkill)
373 {
374         lockdep_assert_held(&rfkill_global_mutex);
375
376         if (!rfkill->need_sync)
377                 return;
378
379         rfkill_set_block(rfkill, rfkill_global_states[rfkill->type].cur);
380         rfkill->need_sync = false;
381 }
382
383 static void rfkill_update_global_state(enum rfkill_type type, bool blocked)
384 {
385         int i;
386
387         if (type != RFKILL_TYPE_ALL) {
388                 rfkill_global_states[type].cur = blocked;
389                 return;
390         }
391
392         for (i = 0; i < NUM_RFKILL_TYPES; i++)
393                 rfkill_global_states[i].cur = blocked;
394 }
395
396 #ifdef CONFIG_RFKILL_INPUT
397 static atomic_t rfkill_input_disabled = ATOMIC_INIT(0);
398
399 /**
400  * __rfkill_switch_all - Toggle state of all switches of given type
401  * @type: type of interfaces to be affected
402  * @blocked: the new state
403  *
404  * This function sets the state of all switches of given type,
405  * unless a specific switch is suspended.
406  *
407  * Caller must have acquired rfkill_global_mutex.
408  */
409 static void __rfkill_switch_all(const enum rfkill_type type, bool blocked)
410 {
411         struct rfkill *rfkill;
412
413         rfkill_update_global_state(type, blocked);
414         list_for_each_entry(rfkill, &rfkill_list, node) {
415                 if (rfkill->type != type && type != RFKILL_TYPE_ALL)
416                         continue;
417
418                 rfkill_set_block(rfkill, blocked);
419         }
420 }
421
422 /**
423  * rfkill_switch_all - Toggle state of all switches of given type
424  * @type: type of interfaces to be affected
425  * @blocked: the new state
426  *
427  * Acquires rfkill_global_mutex and calls __rfkill_switch_all(@type, @state).
428  * Please refer to __rfkill_switch_all() for details.
429  *
430  * Does nothing if the EPO lock is active.
431  */
432 void rfkill_switch_all(enum rfkill_type type, bool blocked)
433 {
434         if (atomic_read(&rfkill_input_disabled))
435                 return;
436
437         mutex_lock(&rfkill_global_mutex);
438
439         if (!rfkill_epo_lock_active)
440                 __rfkill_switch_all(type, blocked);
441
442         mutex_unlock(&rfkill_global_mutex);
443 }
444
445 /**
446  * rfkill_epo - emergency power off all transmitters
447  *
448  * This kicks all non-suspended rfkill devices to RFKILL_STATE_SOFT_BLOCKED,
449  * ignoring everything in its path but rfkill_global_mutex and rfkill->mutex.
450  *
451  * The global state before the EPO is saved and can be restored later
452  * using rfkill_restore_states().
453  */
454 void rfkill_epo(void)
455 {
456         struct rfkill *rfkill;
457         int i;
458
459         if (atomic_read(&rfkill_input_disabled))
460                 return;
461
462         mutex_lock(&rfkill_global_mutex);
463
464         rfkill_epo_lock_active = true;
465         list_for_each_entry(rfkill, &rfkill_list, node)
466                 rfkill_set_block(rfkill, true);
467
468         for (i = 0; i < NUM_RFKILL_TYPES; i++) {
469                 rfkill_global_states[i].sav = rfkill_global_states[i].cur;
470                 rfkill_global_states[i].cur = true;
471         }
472
473         mutex_unlock(&rfkill_global_mutex);
474 }
475
476 /**
477  * rfkill_restore_states - restore global states
478  *
479  * Restore (and sync switches to) the global state from the
480  * states in rfkill_default_states.  This can undo the effects of
481  * a call to rfkill_epo().
482  */
483 void rfkill_restore_states(void)
484 {
485         int i;
486
487         if (atomic_read(&rfkill_input_disabled))
488                 return;
489
490         mutex_lock(&rfkill_global_mutex);
491
492         rfkill_epo_lock_active = false;
493         for (i = 0; i < NUM_RFKILL_TYPES; i++)
494                 __rfkill_switch_all(i, rfkill_global_states[i].sav);
495         mutex_unlock(&rfkill_global_mutex);
496 }
497
498 /**
499  * rfkill_remove_epo_lock - unlock state changes
500  *
501  * Used by rfkill-input manually unlock state changes, when
502  * the EPO switch is deactivated.
503  */
504 void rfkill_remove_epo_lock(void)
505 {
506         if (atomic_read(&rfkill_input_disabled))
507                 return;
508
509         mutex_lock(&rfkill_global_mutex);
510         rfkill_epo_lock_active = false;
511         mutex_unlock(&rfkill_global_mutex);
512 }
513
514 /**
515  * rfkill_is_epo_lock_active - returns true EPO is active
516  *
517  * Returns 0 (false) if there is NOT an active EPO condition,
518  * and 1 (true) if there is an active EPO condition, which
519  * locks all radios in one of the BLOCKED states.
520  *
521  * Can be called in atomic context.
522  */
523 bool rfkill_is_epo_lock_active(void)
524 {
525         return rfkill_epo_lock_active;
526 }
527
528 /**
529  * rfkill_get_global_sw_state - returns global state for a type
530  * @type: the type to get the global state of
531  *
532  * Returns the current global state for a given wireless
533  * device type.
534  */
535 bool rfkill_get_global_sw_state(const enum rfkill_type type)
536 {
537         return rfkill_global_states[type].cur;
538 }
539 #endif
540
541 bool rfkill_set_hw_state_reason(struct rfkill *rfkill,
542                                 bool blocked,
543                                 enum rfkill_hard_block_reasons reason)
544 {
545         unsigned long flags;
546         bool ret, prev;
547
548         BUG_ON(!rfkill);
549
550         spin_lock_irqsave(&rfkill->lock, flags);
551         prev = !!(rfkill->hard_block_reasons & reason);
552         if (blocked) {
553                 rfkill->state |= RFKILL_BLOCK_HW;
554                 rfkill->hard_block_reasons |= reason;
555         } else {
556                 rfkill->hard_block_reasons &= ~reason;
557                 if (!rfkill->hard_block_reasons)
558                         rfkill->state &= ~RFKILL_BLOCK_HW;
559         }
560         ret = !!(rfkill->state & RFKILL_BLOCK_ANY);
561         spin_unlock_irqrestore(&rfkill->lock, flags);
562
563         rfkill_led_trigger_event(rfkill);
564         rfkill_global_led_trigger_event();
565
566         if (rfkill->registered && prev != blocked)
567                 schedule_work(&rfkill->uevent_work);
568
569         return ret;
570 }
571 EXPORT_SYMBOL(rfkill_set_hw_state_reason);
572
573 static void __rfkill_set_sw_state(struct rfkill *rfkill, bool blocked)
574 {
575         u32 bit = RFKILL_BLOCK_SW;
576
577         /* if in a ops->set_block right now, use other bit */
578         if (rfkill->state & RFKILL_BLOCK_SW_SETCALL)
579                 bit = RFKILL_BLOCK_SW_PREV;
580
581         if (blocked)
582                 rfkill->state |= bit;
583         else
584                 rfkill->state &= ~bit;
585 }
586
587 bool rfkill_set_sw_state(struct rfkill *rfkill, bool blocked)
588 {
589         unsigned long flags;
590         bool prev, hwblock;
591
592         BUG_ON(!rfkill);
593
594         spin_lock_irqsave(&rfkill->lock, flags);
595         prev = !!(rfkill->state & RFKILL_BLOCK_SW);
596         __rfkill_set_sw_state(rfkill, blocked);
597         hwblock = !!(rfkill->state & RFKILL_BLOCK_HW);
598         blocked = blocked || hwblock;
599         spin_unlock_irqrestore(&rfkill->lock, flags);
600
601         if (!rfkill->registered)
602                 return blocked;
603
604         if (prev != blocked && !hwblock)
605                 schedule_work(&rfkill->uevent_work);
606
607         rfkill_led_trigger_event(rfkill);
608         rfkill_global_led_trigger_event();
609
610         return blocked;
611 }
612 EXPORT_SYMBOL(rfkill_set_sw_state);
613
614 void rfkill_init_sw_state(struct rfkill *rfkill, bool blocked)
615 {
616         unsigned long flags;
617
618         BUG_ON(!rfkill);
619         BUG_ON(rfkill->registered);
620
621         spin_lock_irqsave(&rfkill->lock, flags);
622         __rfkill_set_sw_state(rfkill, blocked);
623         rfkill->persistent = true;
624         spin_unlock_irqrestore(&rfkill->lock, flags);
625 }
626 EXPORT_SYMBOL(rfkill_init_sw_state);
627
628 void rfkill_set_states(struct rfkill *rfkill, bool sw, bool hw)
629 {
630         unsigned long flags;
631         bool swprev, hwprev;
632
633         BUG_ON(!rfkill);
634
635         spin_lock_irqsave(&rfkill->lock, flags);
636
637         /*
638          * No need to care about prev/setblock ... this is for uevent only
639          * and that will get triggered by rfkill_set_block anyway.
640          */
641         swprev = !!(rfkill->state & RFKILL_BLOCK_SW);
642         hwprev = !!(rfkill->state & RFKILL_BLOCK_HW);
643         __rfkill_set_sw_state(rfkill, sw);
644         if (hw)
645                 rfkill->state |= RFKILL_BLOCK_HW;
646         else
647                 rfkill->state &= ~RFKILL_BLOCK_HW;
648
649         spin_unlock_irqrestore(&rfkill->lock, flags);
650
651         if (!rfkill->registered) {
652                 rfkill->persistent = true;
653         } else {
654                 if (swprev != sw || hwprev != hw)
655                         schedule_work(&rfkill->uevent_work);
656
657                 rfkill_led_trigger_event(rfkill);
658                 rfkill_global_led_trigger_event();
659         }
660 }
661 EXPORT_SYMBOL(rfkill_set_states);
662
663 static const char * const rfkill_types[] = {
664         NULL, /* RFKILL_TYPE_ALL */
665         "wlan",
666         "bluetooth",
667         "ultrawideband",
668         "wimax",
669         "wwan",
670         "gps",
671         "fm",
672         "nfc",
673 };
674
675 enum rfkill_type rfkill_find_type(const char *name)
676 {
677         int i;
678
679         BUILD_BUG_ON(ARRAY_SIZE(rfkill_types) != NUM_RFKILL_TYPES);
680
681         if (!name)
682                 return RFKILL_TYPE_ALL;
683
684         for (i = 1; i < NUM_RFKILL_TYPES; i++)
685                 if (!strcmp(name, rfkill_types[i]))
686                         return i;
687         return RFKILL_TYPE_ALL;
688 }
689 EXPORT_SYMBOL(rfkill_find_type);
690
691 static ssize_t name_show(struct device *dev, struct device_attribute *attr,
692                          char *buf)
693 {
694         struct rfkill *rfkill = to_rfkill(dev);
695
696         return sysfs_emit(buf, "%s\n", rfkill->name);
697 }
698 static DEVICE_ATTR_RO(name);
699
700 static ssize_t type_show(struct device *dev, struct device_attribute *attr,
701                          char *buf)
702 {
703         struct rfkill *rfkill = to_rfkill(dev);
704
705         return sysfs_emit(buf, "%s\n", rfkill_types[rfkill->type]);
706 }
707 static DEVICE_ATTR_RO(type);
708
709 static ssize_t index_show(struct device *dev, struct device_attribute *attr,
710                           char *buf)
711 {
712         struct rfkill *rfkill = to_rfkill(dev);
713
714         return sysfs_emit(buf, "%d\n", rfkill->idx);
715 }
716 static DEVICE_ATTR_RO(index);
717
718 static ssize_t persistent_show(struct device *dev,
719                                struct device_attribute *attr, char *buf)
720 {
721         struct rfkill *rfkill = to_rfkill(dev);
722
723         return sysfs_emit(buf, "%d\n", rfkill->persistent);
724 }
725 static DEVICE_ATTR_RO(persistent);
726
727 static ssize_t hard_show(struct device *dev, struct device_attribute *attr,
728                          char *buf)
729 {
730         struct rfkill *rfkill = to_rfkill(dev);
731
732         return sysfs_emit(buf, "%d\n", (rfkill->state & RFKILL_BLOCK_HW) ? 1 : 0);
733 }
734 static DEVICE_ATTR_RO(hard);
735
736 static ssize_t soft_show(struct device *dev, struct device_attribute *attr,
737                          char *buf)
738 {
739         struct rfkill *rfkill = to_rfkill(dev);
740
741         mutex_lock(&rfkill_global_mutex);
742         rfkill_sync(rfkill);
743         mutex_unlock(&rfkill_global_mutex);
744
745         return sysfs_emit(buf, "%d\n", (rfkill->state & RFKILL_BLOCK_SW) ? 1 : 0);
746 }
747
748 static ssize_t soft_store(struct device *dev, struct device_attribute *attr,
749                           const char *buf, size_t count)
750 {
751         struct rfkill *rfkill = to_rfkill(dev);
752         unsigned long state;
753         int err;
754
755         if (!capable(CAP_NET_ADMIN))
756                 return -EPERM;
757
758         err = kstrtoul(buf, 0, &state);
759         if (err)
760                 return err;
761
762         if (state > 1 )
763                 return -EINVAL;
764
765         mutex_lock(&rfkill_global_mutex);
766         rfkill_sync(rfkill);
767         rfkill_set_block(rfkill, state);
768         mutex_unlock(&rfkill_global_mutex);
769
770         return count;
771 }
772 static DEVICE_ATTR_RW(soft);
773
774 static ssize_t hard_block_reasons_show(struct device *dev,
775                                        struct device_attribute *attr,
776                                        char *buf)
777 {
778         struct rfkill *rfkill = to_rfkill(dev);
779
780         return sysfs_emit(buf, "0x%lx\n", rfkill->hard_block_reasons);
781 }
782 static DEVICE_ATTR_RO(hard_block_reasons);
783
784 static u8 user_state_from_blocked(unsigned long state)
785 {
786         if (state & RFKILL_BLOCK_HW)
787                 return RFKILL_USER_STATE_HARD_BLOCKED;
788         if (state & RFKILL_BLOCK_SW)
789                 return RFKILL_USER_STATE_SOFT_BLOCKED;
790
791         return RFKILL_USER_STATE_UNBLOCKED;
792 }
793
794 static ssize_t state_show(struct device *dev, struct device_attribute *attr,
795                           char *buf)
796 {
797         struct rfkill *rfkill = to_rfkill(dev);
798
799         mutex_lock(&rfkill_global_mutex);
800         rfkill_sync(rfkill);
801         mutex_unlock(&rfkill_global_mutex);
802
803         return sysfs_emit(buf, "%d\n", user_state_from_blocked(rfkill->state));
804 }
805
806 static ssize_t state_store(struct device *dev, struct device_attribute *attr,
807                            const char *buf, size_t count)
808 {
809         struct rfkill *rfkill = to_rfkill(dev);
810         unsigned long state;
811         int err;
812
813         if (!capable(CAP_NET_ADMIN))
814                 return -EPERM;
815
816         err = kstrtoul(buf, 0, &state);
817         if (err)
818                 return err;
819
820         if (state != RFKILL_USER_STATE_SOFT_BLOCKED &&
821             state != RFKILL_USER_STATE_UNBLOCKED)
822                 return -EINVAL;
823
824         mutex_lock(&rfkill_global_mutex);
825         rfkill_sync(rfkill);
826         rfkill_set_block(rfkill, state == RFKILL_USER_STATE_SOFT_BLOCKED);
827         mutex_unlock(&rfkill_global_mutex);
828
829         return count;
830 }
831 static DEVICE_ATTR_RW(state);
832
833 static struct attribute *rfkill_dev_attrs[] = {
834         &dev_attr_name.attr,
835         &dev_attr_type.attr,
836         &dev_attr_index.attr,
837         &dev_attr_persistent.attr,
838         &dev_attr_state.attr,
839         &dev_attr_soft.attr,
840         &dev_attr_hard.attr,
841         &dev_attr_hard_block_reasons.attr,
842         NULL,
843 };
844 ATTRIBUTE_GROUPS(rfkill_dev);
845
846 static void rfkill_release(struct device *dev)
847 {
848         struct rfkill *rfkill = to_rfkill(dev);
849
850         kfree(rfkill);
851 }
852
853 static int rfkill_dev_uevent(const struct device *dev, struct kobj_uevent_env *env)
854 {
855         struct rfkill *rfkill = to_rfkill(dev);
856         unsigned long flags;
857         unsigned long reasons;
858         u32 state;
859         int error;
860
861         error = add_uevent_var(env, "RFKILL_NAME=%s", rfkill->name);
862         if (error)
863                 return error;
864         error = add_uevent_var(env, "RFKILL_TYPE=%s",
865                                rfkill_types[rfkill->type]);
866         if (error)
867                 return error;
868         spin_lock_irqsave(&rfkill->lock, flags);
869         state = rfkill->state;
870         reasons = rfkill->hard_block_reasons;
871         spin_unlock_irqrestore(&rfkill->lock, flags);
872         error = add_uevent_var(env, "RFKILL_STATE=%d",
873                                user_state_from_blocked(state));
874         if (error)
875                 return error;
876         return add_uevent_var(env, "RFKILL_HW_BLOCK_REASON=0x%lx", reasons);
877 }
878
879 void rfkill_pause_polling(struct rfkill *rfkill)
880 {
881         BUG_ON(!rfkill);
882
883         if (!rfkill->ops->poll)
884                 return;
885
886         rfkill->polling_paused = true;
887         cancel_delayed_work_sync(&rfkill->poll_work);
888 }
889 EXPORT_SYMBOL(rfkill_pause_polling);
890
891 void rfkill_resume_polling(struct rfkill *rfkill)
892 {
893         BUG_ON(!rfkill);
894
895         if (!rfkill->ops->poll)
896                 return;
897
898         rfkill->polling_paused = false;
899
900         if (rfkill->suspended)
901                 return;
902
903         queue_delayed_work(system_power_efficient_wq,
904                            &rfkill->poll_work, 0);
905 }
906 EXPORT_SYMBOL(rfkill_resume_polling);
907
908 #ifdef CONFIG_PM_SLEEP
909 static int rfkill_suspend(struct device *dev)
910 {
911         struct rfkill *rfkill = to_rfkill(dev);
912
913         rfkill->suspended = true;
914         cancel_delayed_work_sync(&rfkill->poll_work);
915
916         return 0;
917 }
918
919 static int rfkill_resume(struct device *dev)
920 {
921         struct rfkill *rfkill = to_rfkill(dev);
922         bool cur;
923
924         rfkill->suspended = false;
925
926         if (!rfkill->registered)
927                 return 0;
928
929         if (!rfkill->persistent) {
930                 cur = !!(rfkill->state & RFKILL_BLOCK_SW);
931                 rfkill_set_block(rfkill, cur);
932         }
933
934         if (rfkill->ops->poll && !rfkill->polling_paused)
935                 queue_delayed_work(system_power_efficient_wq,
936                                    &rfkill->poll_work, 0);
937
938         return 0;
939 }
940
941 static SIMPLE_DEV_PM_OPS(rfkill_pm_ops, rfkill_suspend, rfkill_resume);
942 #define RFKILL_PM_OPS (&rfkill_pm_ops)
943 #else
944 #define RFKILL_PM_OPS NULL
945 #endif
946
947 static struct class rfkill_class = {
948         .name           = "rfkill",
949         .dev_release    = rfkill_release,
950         .dev_groups     = rfkill_dev_groups,
951         .dev_uevent     = rfkill_dev_uevent,
952         .pm             = RFKILL_PM_OPS,
953 };
954
955 bool rfkill_blocked(struct rfkill *rfkill)
956 {
957         unsigned long flags;
958         u32 state;
959
960         spin_lock_irqsave(&rfkill->lock, flags);
961         state = rfkill->state;
962         spin_unlock_irqrestore(&rfkill->lock, flags);
963
964         return !!(state & RFKILL_BLOCK_ANY);
965 }
966 EXPORT_SYMBOL(rfkill_blocked);
967
968 bool rfkill_soft_blocked(struct rfkill *rfkill)
969 {
970         unsigned long flags;
971         u32 state;
972
973         spin_lock_irqsave(&rfkill->lock, flags);
974         state = rfkill->state;
975         spin_unlock_irqrestore(&rfkill->lock, flags);
976
977         return !!(state & RFKILL_BLOCK_SW);
978 }
979 EXPORT_SYMBOL(rfkill_soft_blocked);
980
981 struct rfkill * __must_check rfkill_alloc(const char *name,
982                                           struct device *parent,
983                                           const enum rfkill_type type,
984                                           const struct rfkill_ops *ops,
985                                           void *ops_data)
986 {
987         struct rfkill *rfkill;
988         struct device *dev;
989
990         if (WARN_ON(!ops))
991                 return NULL;
992
993         if (WARN_ON(!ops->set_block))
994                 return NULL;
995
996         if (WARN_ON(!name))
997                 return NULL;
998
999         if (WARN_ON(type == RFKILL_TYPE_ALL || type >= NUM_RFKILL_TYPES))
1000                 return NULL;
1001
1002         rfkill = kzalloc(sizeof(*rfkill) + strlen(name) + 1, GFP_KERNEL);
1003         if (!rfkill)
1004                 return NULL;
1005
1006         spin_lock_init(&rfkill->lock);
1007         INIT_LIST_HEAD(&rfkill->node);
1008         rfkill->type = type;
1009         strcpy(rfkill->name, name);
1010         rfkill->ops = ops;
1011         rfkill->data = ops_data;
1012
1013         dev = &rfkill->dev;
1014         dev->class = &rfkill_class;
1015         dev->parent = parent;
1016         device_initialize(dev);
1017
1018         return rfkill;
1019 }
1020 EXPORT_SYMBOL(rfkill_alloc);
1021
1022 static void rfkill_poll(struct work_struct *work)
1023 {
1024         struct rfkill *rfkill;
1025
1026         rfkill = container_of(work, struct rfkill, poll_work.work);
1027
1028         /*
1029          * Poll hardware state -- driver will use one of the
1030          * rfkill_set{,_hw,_sw}_state functions and use its
1031          * return value to update the current status.
1032          */
1033         rfkill->ops->poll(rfkill, rfkill->data);
1034
1035         queue_delayed_work(system_power_efficient_wq,
1036                 &rfkill->poll_work,
1037                 round_jiffies_relative(POLL_INTERVAL));
1038 }
1039
1040 static void rfkill_uevent_work(struct work_struct *work)
1041 {
1042         struct rfkill *rfkill;
1043
1044         rfkill = container_of(work, struct rfkill, uevent_work);
1045
1046         mutex_lock(&rfkill_global_mutex);
1047         rfkill_event(rfkill);
1048         mutex_unlock(&rfkill_global_mutex);
1049 }
1050
1051 static void rfkill_sync_work(struct work_struct *work)
1052 {
1053         struct rfkill *rfkill = container_of(work, struct rfkill, sync_work);
1054
1055         mutex_lock(&rfkill_global_mutex);
1056         rfkill_sync(rfkill);
1057         mutex_unlock(&rfkill_global_mutex);
1058 }
1059
1060 int __must_check rfkill_register(struct rfkill *rfkill)
1061 {
1062         static unsigned long rfkill_no;
1063         struct device *dev;
1064         int error;
1065
1066         if (!rfkill)
1067                 return -EINVAL;
1068
1069         dev = &rfkill->dev;
1070
1071         mutex_lock(&rfkill_global_mutex);
1072
1073         if (rfkill->registered) {
1074                 error = -EALREADY;
1075                 goto unlock;
1076         }
1077
1078         rfkill->idx = rfkill_no;
1079         dev_set_name(dev, "rfkill%lu", rfkill_no);
1080         rfkill_no++;
1081
1082         list_add_tail(&rfkill->node, &rfkill_list);
1083
1084         error = device_add(dev);
1085         if (error)
1086                 goto remove;
1087
1088         error = rfkill_led_trigger_register(rfkill);
1089         if (error)
1090                 goto devdel;
1091
1092         rfkill->registered = true;
1093
1094         INIT_DELAYED_WORK(&rfkill->poll_work, rfkill_poll);
1095         INIT_WORK(&rfkill->uevent_work, rfkill_uevent_work);
1096         INIT_WORK(&rfkill->sync_work, rfkill_sync_work);
1097
1098         if (rfkill->ops->poll)
1099                 queue_delayed_work(system_power_efficient_wq,
1100                         &rfkill->poll_work,
1101                         round_jiffies_relative(POLL_INTERVAL));
1102
1103         if (!rfkill->persistent || rfkill_epo_lock_active) {
1104                 rfkill->need_sync = true;
1105                 schedule_work(&rfkill->sync_work);
1106         } else {
1107 #ifdef CONFIG_RFKILL_INPUT
1108                 bool soft_blocked = !!(rfkill->state & RFKILL_BLOCK_SW);
1109
1110                 if (!atomic_read(&rfkill_input_disabled))
1111                         __rfkill_switch_all(rfkill->type, soft_blocked);
1112 #endif
1113         }
1114
1115         rfkill_global_led_trigger_event();
1116         rfkill_send_events(rfkill, RFKILL_OP_ADD);
1117
1118         mutex_unlock(&rfkill_global_mutex);
1119         return 0;
1120
1121  devdel:
1122         device_del(&rfkill->dev);
1123  remove:
1124         list_del_init(&rfkill->node);
1125  unlock:
1126         mutex_unlock(&rfkill_global_mutex);
1127         return error;
1128 }
1129 EXPORT_SYMBOL(rfkill_register);
1130
1131 void rfkill_unregister(struct rfkill *rfkill)
1132 {
1133         BUG_ON(!rfkill);
1134
1135         if (rfkill->ops->poll)
1136                 cancel_delayed_work_sync(&rfkill->poll_work);
1137
1138         cancel_work_sync(&rfkill->uevent_work);
1139         cancel_work_sync(&rfkill->sync_work);
1140
1141         rfkill->registered = false;
1142
1143         device_del(&rfkill->dev);
1144
1145         mutex_lock(&rfkill_global_mutex);
1146         rfkill_send_events(rfkill, RFKILL_OP_DEL);
1147         list_del_init(&rfkill->node);
1148         rfkill_global_led_trigger_event();
1149         mutex_unlock(&rfkill_global_mutex);
1150
1151         rfkill_led_trigger_unregister(rfkill);
1152 }
1153 EXPORT_SYMBOL(rfkill_unregister);
1154
1155 void rfkill_destroy(struct rfkill *rfkill)
1156 {
1157         if (rfkill)
1158                 put_device(&rfkill->dev);
1159 }
1160 EXPORT_SYMBOL(rfkill_destroy);
1161
1162 static int rfkill_fop_open(struct inode *inode, struct file *file)
1163 {
1164         struct rfkill_data *data;
1165         struct rfkill *rfkill;
1166         struct rfkill_int_event *ev, *tmp;
1167
1168         data = kzalloc(sizeof(*data), GFP_KERNEL);
1169         if (!data)
1170                 return -ENOMEM;
1171
1172         data->max_size = RFKILL_EVENT_SIZE_V1;
1173
1174         INIT_LIST_HEAD(&data->events);
1175         mutex_init(&data->mtx);
1176         init_waitqueue_head(&data->read_wait);
1177
1178         mutex_lock(&rfkill_global_mutex);
1179         /*
1180          * start getting events from elsewhere but hold mtx to get
1181          * startup events added first
1182          */
1183
1184         list_for_each_entry(rfkill, &rfkill_list, node) {
1185                 ev = kzalloc(sizeof(*ev), GFP_KERNEL);
1186                 if (!ev)
1187                         goto free;
1188                 rfkill_sync(rfkill);
1189                 rfkill_fill_event(&ev->ev, rfkill, RFKILL_OP_ADD);
1190                 mutex_lock(&data->mtx);
1191                 list_add_tail(&ev->list, &data->events);
1192                 mutex_unlock(&data->mtx);
1193         }
1194         list_add(&data->list, &rfkill_fds);
1195         mutex_unlock(&rfkill_global_mutex);
1196
1197         file->private_data = data;
1198
1199         return stream_open(inode, file);
1200
1201  free:
1202         mutex_unlock(&rfkill_global_mutex);
1203         mutex_destroy(&data->mtx);
1204         list_for_each_entry_safe(ev, tmp, &data->events, list)
1205                 kfree(ev);
1206         kfree(data);
1207         return -ENOMEM;
1208 }
1209
1210 static __poll_t rfkill_fop_poll(struct file *file, poll_table *wait)
1211 {
1212         struct rfkill_data *data = file->private_data;
1213         __poll_t res = EPOLLOUT | EPOLLWRNORM;
1214
1215         poll_wait(file, &data->read_wait, wait);
1216
1217         mutex_lock(&data->mtx);
1218         if (!list_empty(&data->events))
1219                 res = EPOLLIN | EPOLLRDNORM;
1220         mutex_unlock(&data->mtx);
1221
1222         return res;
1223 }
1224
1225 static ssize_t rfkill_fop_read(struct file *file, char __user *buf,
1226                                size_t count, loff_t *pos)
1227 {
1228         struct rfkill_data *data = file->private_data;
1229         struct rfkill_int_event *ev;
1230         unsigned long sz;
1231         int ret;
1232
1233         mutex_lock(&data->mtx);
1234
1235         while (list_empty(&data->events)) {
1236                 if (file->f_flags & O_NONBLOCK) {
1237                         ret = -EAGAIN;
1238                         goto out;
1239                 }
1240                 mutex_unlock(&data->mtx);
1241                 /* since we re-check and it just compares pointers,
1242                  * using !list_empty() without locking isn't a problem
1243                  */
1244                 ret = wait_event_interruptible(data->read_wait,
1245                                                !list_empty(&data->events));
1246                 mutex_lock(&data->mtx);
1247
1248                 if (ret)
1249                         goto out;
1250         }
1251
1252         ev = list_first_entry(&data->events, struct rfkill_int_event,
1253                                 list);
1254
1255         sz = min_t(unsigned long, sizeof(ev->ev), count);
1256         sz = min_t(unsigned long, sz, data->max_size);
1257         ret = sz;
1258         if (copy_to_user(buf, &ev->ev, sz))
1259                 ret = -EFAULT;
1260
1261         list_del(&ev->list);
1262         kfree(ev);
1263  out:
1264         mutex_unlock(&data->mtx);
1265         return ret;
1266 }
1267
1268 static ssize_t rfkill_fop_write(struct file *file, const char __user *buf,
1269                                 size_t count, loff_t *pos)
1270 {
1271         struct rfkill_data *data = file->private_data;
1272         struct rfkill *rfkill;
1273         struct rfkill_event_ext ev;
1274         int ret;
1275
1276         /* we don't need the 'hard' variable but accept it */
1277         if (count < RFKILL_EVENT_SIZE_V1 - 1)
1278                 return -EINVAL;
1279
1280         /*
1281          * Copy as much data as we can accept into our 'ev' buffer,
1282          * but tell userspace how much we've copied so it can determine
1283          * our API version even in a write() call, if it cares.
1284          */
1285         count = min(count, sizeof(ev));
1286         count = min_t(size_t, count, data->max_size);
1287         if (copy_from_user(&ev, buf, count))
1288                 return -EFAULT;
1289
1290         if (ev.type >= NUM_RFKILL_TYPES)
1291                 return -EINVAL;
1292
1293         mutex_lock(&rfkill_global_mutex);
1294
1295         switch (ev.op) {
1296         case RFKILL_OP_CHANGE_ALL:
1297                 rfkill_update_global_state(ev.type, ev.soft);
1298                 list_for_each_entry(rfkill, &rfkill_list, node)
1299                         if (rfkill->type == ev.type ||
1300                             ev.type == RFKILL_TYPE_ALL)
1301                                 rfkill_set_block(rfkill, ev.soft);
1302                 ret = 0;
1303                 break;
1304         case RFKILL_OP_CHANGE:
1305                 list_for_each_entry(rfkill, &rfkill_list, node)
1306                         if (rfkill->idx == ev.idx &&
1307                             (rfkill->type == ev.type ||
1308                              ev.type == RFKILL_TYPE_ALL))
1309                                 rfkill_set_block(rfkill, ev.soft);
1310                 ret = 0;
1311                 break;
1312         default:
1313                 ret = -EINVAL;
1314                 break;
1315         }
1316
1317         mutex_unlock(&rfkill_global_mutex);
1318
1319         return ret ?: count;
1320 }
1321
1322 static int rfkill_fop_release(struct inode *inode, struct file *file)
1323 {
1324         struct rfkill_data *data = file->private_data;
1325         struct rfkill_int_event *ev, *tmp;
1326
1327         mutex_lock(&rfkill_global_mutex);
1328         list_del(&data->list);
1329         mutex_unlock(&rfkill_global_mutex);
1330
1331         mutex_destroy(&data->mtx);
1332         list_for_each_entry_safe(ev, tmp, &data->events, list)
1333                 kfree(ev);
1334
1335 #ifdef CONFIG_RFKILL_INPUT
1336         if (data->input_handler)
1337                 if (atomic_dec_return(&rfkill_input_disabled) == 0)
1338                         printk(KERN_DEBUG "rfkill: input handler enabled\n");
1339 #endif
1340
1341         kfree(data);
1342
1343         return 0;
1344 }
1345
1346 static long rfkill_fop_ioctl(struct file *file, unsigned int cmd,
1347                              unsigned long arg)
1348 {
1349         struct rfkill_data *data = file->private_data;
1350         int ret = -ENOTTY;
1351         u32 size;
1352
1353         if (_IOC_TYPE(cmd) != RFKILL_IOC_MAGIC)
1354                 return -ENOTTY;
1355
1356         mutex_lock(&data->mtx);
1357         switch (_IOC_NR(cmd)) {
1358 #ifdef CONFIG_RFKILL_INPUT
1359         case RFKILL_IOC_NOINPUT:
1360                 if (!data->input_handler) {
1361                         if (atomic_inc_return(&rfkill_input_disabled) == 1)
1362                                 printk(KERN_DEBUG "rfkill: input handler disabled\n");
1363                         data->input_handler = true;
1364                 }
1365                 ret = 0;
1366                 break;
1367 #endif
1368         case RFKILL_IOC_MAX_SIZE:
1369                 if (get_user(size, (__u32 __user *)arg)) {
1370                         ret = -EFAULT;
1371                         break;
1372                 }
1373                 if (size < RFKILL_EVENT_SIZE_V1 || size > U8_MAX) {
1374                         ret = -EINVAL;
1375                         break;
1376                 }
1377                 data->max_size = size;
1378                 ret = 0;
1379                 break;
1380         default:
1381                 break;
1382         }
1383         mutex_unlock(&data->mtx);
1384
1385         return ret;
1386 }
1387
1388 static const struct file_operations rfkill_fops = {
1389         .owner          = THIS_MODULE,
1390         .open           = rfkill_fop_open,
1391         .read           = rfkill_fop_read,
1392         .write          = rfkill_fop_write,
1393         .poll           = rfkill_fop_poll,
1394         .release        = rfkill_fop_release,
1395         .unlocked_ioctl = rfkill_fop_ioctl,
1396         .compat_ioctl   = compat_ptr_ioctl,
1397         .llseek         = no_llseek,
1398 };
1399
1400 #define RFKILL_NAME "rfkill"
1401
1402 static struct miscdevice rfkill_miscdev = {
1403         .fops   = &rfkill_fops,
1404         .name   = RFKILL_NAME,
1405         .minor  = RFKILL_MINOR,
1406 };
1407
1408 static int __init rfkill_init(void)
1409 {
1410         int error;
1411
1412         rfkill_update_global_state(RFKILL_TYPE_ALL, !rfkill_default_state);
1413
1414         error = class_register(&rfkill_class);
1415         if (error)
1416                 goto error_class;
1417
1418         error = misc_register(&rfkill_miscdev);
1419         if (error)
1420                 goto error_misc;
1421
1422         error = rfkill_global_led_trigger_register();
1423         if (error)
1424                 goto error_led_trigger;
1425
1426 #ifdef CONFIG_RFKILL_INPUT
1427         error = rfkill_handler_init();
1428         if (error)
1429                 goto error_input;
1430 #endif
1431
1432         return 0;
1433
1434 #ifdef CONFIG_RFKILL_INPUT
1435 error_input:
1436         rfkill_global_led_trigger_unregister();
1437 #endif
1438 error_led_trigger:
1439         misc_deregister(&rfkill_miscdev);
1440 error_misc:
1441         class_unregister(&rfkill_class);
1442 error_class:
1443         return error;
1444 }
1445 subsys_initcall(rfkill_init);
1446
1447 static void __exit rfkill_exit(void)
1448 {
1449 #ifdef CONFIG_RFKILL_INPUT
1450         rfkill_handler_exit();
1451 #endif
1452         rfkill_global_led_trigger_unregister();
1453         misc_deregister(&rfkill_miscdev);
1454         class_unregister(&rfkill_class);
1455 }
1456 module_exit(rfkill_exit);
1457
1458 MODULE_ALIAS_MISCDEV(RFKILL_MINOR);
1459 MODULE_ALIAS("devname:" RFKILL_NAME);