1===============================
2Creating an input device driver
3===============================
4
5The simplest example
6~~~~~~~~~~~~~~~~~~~~
7
8Here comes a very simple example of an input device driver. The device has
9just one button and the button is accessible at i/o port BUTTON_PORT. When
10pressed or released a BUTTON_IRQ happens. The driver could look like::
11
12    #include <linux/input.h>
13    #include <linux/module.h>
14    #include <linux/init.h>
15
16    #include <asm/irq.h>
17    #include <asm/io.h>
18
19    static struct input_dev *button_dev;
20
21    static irqreturn_t button_interrupt(int irq, void *dummy)
22    {
23	    input_report_key(button_dev, BTN_0, inb(BUTTON_PORT) & 1);
24	    input_sync(button_dev);
25	    return IRQ_HANDLED;
26    }
27
28    static int __init button_init(void)
29    {
30	    int error;
31
32	    if (request_irq(BUTTON_IRQ, button_interrupt, 0, "button", NULL)) {
33		    printk(KERN_ERR "button.c: Can't allocate irq %d\n", button_irq);
34		    return -EBUSY;
35	    }
36
37	    button_dev = input_allocate_device();
38	    if (!button_dev) {
39		    printk(KERN_ERR "button.c: Not enough memory\n");
40		    error = -ENOMEM;
41		    goto err_free_irq;
42	    }
43
44	    button_dev->evbit[0] = BIT_MASK(EV_KEY);
45	    button_dev->keybit[BIT_WORD(BTN_0)] = BIT_MASK(BTN_0);
46
47	    error = input_register_device(button_dev);
48	    if (error) {
49		    printk(KERN_ERR "button.c: Failed to register device\n");
50		    goto err_free_dev;
51	    }
52
53	    return 0;
54
55    err_free_dev:
56	    input_free_device(button_dev);
57    err_free_irq:
58	    free_irq(BUTTON_IRQ, button_interrupt);
59	    return error;
60    }
61
62    static void __exit button_exit(void)
63    {
64	    input_unregister_device(button_dev);
65	    free_irq(BUTTON_IRQ, button_interrupt);
66    }
67
68    module_init(button_init);
69    module_exit(button_exit);
70
71What the example does
72~~~~~~~~~~~~~~~~~~~~~
73
74First it has to include the <linux/input.h> file, which interfaces to the
75input subsystem. This provides all the definitions needed.
76
77In the _init function, which is called either upon module load or when
78booting the kernel, it grabs the required resources (it should also check
79for the presence of the device).
80
81Then it allocates a new input device structure with input_allocate_device()
82and sets up input bitfields. This way the device driver tells the other
83parts of the input systems what it is - what events can be generated or
84accepted by this input device. Our example device can only generate EV_KEY
85type events, and from those only BTN_0 event code. Thus we only set these
86two bits. We could have used::
87
88	set_bit(EV_KEY, button_dev.evbit);
89	set_bit(BTN_0, button_dev.keybit);
90
91as well, but with more than single bits the first approach tends to be
92shorter.
93
94Then the example driver registers the input device structure by calling::
95
96	input_register_device(&button_dev);
97
98This adds the button_dev structure to linked lists of the input driver and
99calls device handler modules _connect functions to tell them a new input
100device has appeared. input_register_device() may sleep and therefore must
101not be called from an interrupt or with a spinlock held.
102
103While in use, the only used function of the driver is::
104
105	button_interrupt()
106
107which upon every interrupt from the button checks its state and reports it
108via the::
109
110	input_report_key()
111
112call to the input system. There is no need to check whether the interrupt
113routine isn't reporting two same value events (press, press for example) to
114the input system, because the input_report_* functions check that
115themselves.
116
117Then there is the::
118
119	input_sync()
120
121call to tell those who receive the events that we've sent a complete report.
122This doesn't seem important in the one button case, but is quite important
123for example for mouse movement, where you don't want the X and Y values
124to be interpreted separately, because that'd result in a different movement.
125
126dev->open() and dev->close()
127~~~~~~~~~~~~~~~~~~~~~~~~~~~~
128
129In case the driver has to repeatedly poll the device, because it doesn't
130have an interrupt coming from it and the polling is too expensive to be done
131all the time, or if the device uses a valuable resource (e.g. interrupt), it
132can use the open and close callback to know when it can stop polling or
133release the interrupt and when it must resume polling or grab the interrupt
134again. To do that, we would add this to our example driver::
135
136    static int button_open(struct input_dev *dev)
137    {
138	    if (request_irq(BUTTON_IRQ, button_interrupt, 0, "button", NULL)) {
139		    printk(KERN_ERR "button.c: Can't allocate irq %d\n", button_irq);
140		    return -EBUSY;
141	    }
142
143	    return 0;
144    }
145
146    static void button_close(struct input_dev *dev)
147    {
148	    free_irq(IRQ_AMIGA_VERTB, button_interrupt);
149    }
150
151    static int __init button_init(void)
152    {
153	    ...
154	    button_dev->open = button_open;
155	    button_dev->close = button_close;
156	    ...
157    }
158
159Note that input core keeps track of number of users for the device and
160makes sure that dev->open() is called only when the first user connects
161to the device and that dev->close() is called when the very last user
162disconnects. Calls to both callbacks are serialized.
163
164The open() callback should return a 0 in case of success or any non-zero value
165in case of failure. The close() callback (which is void) must always succeed.
166
167Inhibiting input devices
168~~~~~~~~~~~~~~~~~~~~~~~~
169
170Inhibiting a device means ignoring input events from it. As such it is about
171maintaining relationships with input handlers - either already existing
172relationships, or relationships to be established while the device is in
173inhibited state.
174
175If a device is inhibited, no input handler will receive events from it.
176
177The fact that nobody wants events from the device is exploited further, by
178calling device's close() (if there are users) and open() (if there are users) on
179inhibit and uninhibit operations, respectively. Indeed, the meaning of close()
180is to stop providing events to the input core and that of open() is to start
181providing events to the input core.
182
183Calling the device's close() method on inhibit (if there are users) allows the
184driver to save power. Either by directly powering down the device or by
185releasing the runtime-PM reference it got in open() when the driver is using
186runtime-PM.
187
188Inhibiting and uninhibiting are orthogonal to opening and closing the device by
189input handlers. Userspace might want to inhibit a device in anticipation before
190any handler is positively matched against it.
191
192Inhibiting and uninhibiting are orthogonal to device's being a wakeup source,
193too. Being a wakeup source plays a role when the system is sleeping, not when
194the system is operating.  How drivers should program their interaction between
195inhibiting, sleeping and being a wakeup source is driver-specific.
196
197Taking the analogy with the network devices - bringing a network interface down
198doesn't mean that it should be impossible be wake the system up on LAN through
199this interface. So, there may be input drivers which should be considered wakeup
200sources even when inhibited. Actually, in many I2C input devices their interrupt
201is declared a wakeup interrupt and its handling happens in driver's core, which
202is not aware of input-specific inhibit (nor should it be).  Composite devices
203containing several interfaces can be inhibited on a per-interface basis and e.g.
204inhibiting one interface shouldn't affect the device's capability of being a
205wakeup source.
206
207If a device is to be considered a wakeup source while inhibited, special care
208must be taken when programming its suspend(), as it might need to call device's
209open(). Depending on what close() means for the device in question, not
210opening() it before going to sleep might make it impossible to provide any
211wakeup events. The device is going to sleep anyway.
212
213Basic event types
214~~~~~~~~~~~~~~~~~
215
216The most simple event type is EV_KEY, which is used for keys and buttons.
217It's reported to the input system via::
218
219	input_report_key(struct input_dev *dev, int code, int value)
220
221See uapi/linux/input-event-codes.h for the allowable values of code (from 0 to
222KEY_MAX). Value is interpreted as a truth value, i.e. any non-zero value means
223key pressed, zero value means key released. The input code generates events only
224in case the value is different from before.
225
226In addition to EV_KEY, there are two more basic event types: EV_REL and
227EV_ABS. They are used for relative and absolute values supplied by the
228device. A relative value may be for example a mouse movement in the X axis.
229The mouse reports it as a relative difference from the last position,
230because it doesn't have any absolute coordinate system to work in. Absolute
231events are namely for joysticks and digitizers - devices that do work in an
232absolute coordinate systems.
233
234Having the device report EV_REL buttons is as simple as with EV_KEY; simply
235set the corresponding bits and call the::
236
237	input_report_rel(struct input_dev *dev, int code, int value)
238
239function. Events are generated only for non-zero values.
240
241However EV_ABS requires a little special care. Before calling
242input_register_device, you have to fill additional fields in the input_dev
243struct for each absolute axis your device has. If our button device had also
244the ABS_X axis::
245
246	button_dev.absmin[ABS_X] = 0;
247	button_dev.absmax[ABS_X] = 255;
248	button_dev.absfuzz[ABS_X] = 4;
249	button_dev.absflat[ABS_X] = 8;
250
251Or, you can just say::
252
253	input_set_abs_params(button_dev, ABS_X, 0, 255, 4, 8);
254
255This setting would be appropriate for a joystick X axis, with the minimum of
2560, maximum of 255 (which the joystick *must* be able to reach, no problem if
257it sometimes reports more, but it must be able to always reach the min and
258max values), with noise in the data up to +- 4, and with a center flat
259position of size 8.
260
261If you don't need absfuzz and absflat, you can set them to zero, which mean
262that the thing is precise and always returns to exactly the center position
263(if it has any).
264
265BITS_TO_LONGS(), BIT_WORD(), BIT_MASK()
266~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
267
268These three macros from bitops.h help some bitfield computations::
269
270	BITS_TO_LONGS(x) - returns the length of a bitfield array in longs for
271			   x bits
272	BIT_WORD(x)	 - returns the index in the array in longs for bit x
273	BIT_MASK(x)	 - returns the index in a long for bit x
274
275The id* and name fields
276~~~~~~~~~~~~~~~~~~~~~~~
277
278The dev->name should be set before registering the input device by the input
279device driver. It's a string like 'Generic button device' containing a
280user friendly name of the device.
281
282The id* fields contain the bus ID (PCI, USB, ...), vendor ID and device ID
283of the device. The bus IDs are defined in input.h. The vendor and device IDs
284are defined in pci_ids.h, usb_ids.h and similar include files. These fields
285should be set by the input device driver before registering it.
286
287The idtype field can be used for specific information for the input device
288driver.
289
290The id and name fields can be passed to userland via the evdev interface.
291
292The keycode, keycodemax, keycodesize fields
293~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
294
295These three fields should be used by input devices that have dense keymaps.
296The keycode is an array used to map from scancodes to input system keycodes.
297The keycode max should contain the size of the array and keycodesize the
298size of each entry in it (in bytes).
299
300Userspace can query and alter current scancode to keycode mappings using
301EVIOCGKEYCODE and EVIOCSKEYCODE ioctls on corresponding evdev interface.
302When a device has all 3 aforementioned fields filled in, the driver may
303rely on kernel's default implementation of setting and querying keycode
304mappings.
305
306dev->getkeycode() and dev->setkeycode()
307~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
308
309getkeycode() and setkeycode() callbacks allow drivers to override default
310keycode/keycodesize/keycodemax mapping mechanism provided by input core
311and implement sparse keycode maps.
312
313Key autorepeat
314~~~~~~~~~~~~~~
315
316... is simple. It is handled by the input.c module. Hardware autorepeat is
317not used, because it's not present in many devices and even where it is
318present, it is broken sometimes (at keyboards: Toshiba notebooks). To enable
319autorepeat for your device, just set EV_REP in dev->evbit. All will be
320handled by the input system.
321
322Other event types, handling output events
323~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
324
325The other event types up to now are:
326
327- EV_LED - used for the keyboard LEDs.
328- EV_SND - used for keyboard beeps.
329
330They are very similar to for example key events, but they go in the other
331direction - from the system to the input device driver. If your input device
332driver can handle these events, it has to set the respective bits in evbit,
333*and* also the callback routine::
334
335    button_dev->event = button_event;
336
337    int button_event(struct input_dev *dev, unsigned int type,
338		     unsigned int code, int value)
339    {
340	    if (type == EV_SND && code == SND_BELL) {
341		    outb(value, BUTTON_BELL);
342		    return 0;
343	    }
344	    return -1;
345    }
346
347This callback routine can be called from an interrupt or a BH (although that
348isn't a rule), and thus must not sleep, and must not take too long to finish.
349