[PATCH] Input: add flags bitfield

[PATCH] Input: add flags bitfield

[Henrique de Moraes Holschuh]

Add a flags bitfield to the input_dev structure, which can be used for
internal coordination among kernel input devices and input handlers without
the need to use ever-expanding blacklists on the input handlers.

Add initial flag bits which allows an input driver to request that joystick
emulation (joydev) or mouse emulation (mousedev) not be attached to an
input device.

This will be used by accelerometer drivers exporting a raw interface which
is not to be used as a joystick device (not to confuse this with the usual
fuzzed joystick interface these drivers export for enhanced Neverball
productivity), for example.

Signed-off-by: Henrique de Moraes Holschuh <hmh@hmh.eng.br>
Cc: Dmitry Torokhov <dtor@mail.ru>
Cc: Richard Purdie <rpurdie@rpsys.net>
---
 drivers/input/joydev.c   |    3 +++
 drivers/input/mousedev.c |    3 +++
 include/linux/input.h    |   10 ++++++++++
 3 files changed, 16 insertions(+), 0 deletions(-)

diff --git a/drivers/input/joydev.c b/drivers/input/joydev.c
index 22b2789..3837389 100644
--- a/drivers/input/joydev.c
+++ b/drivers/input/joydev.c
@@ -727,6 +727,9 @@ static int joydev_connect(struct input_handler *handler, struct input_dev *dev,
 	int i, j, t, minor;
 	int error;
 
+	if (dev->flags & EV_F_NOJOYSTICK)
+		return -ENODEV;
+
 	for (minor = 0; minor < JOYDEV_MINORS; minor++)
 		if (!joydev_table[minor])
 			break;
diff --git a/drivers/input/mousedev.c b/drivers/input/mousedev.c
index bbbe5e8..716975f 100644
--- a/drivers/input/mousedev.c
+++ b/drivers/input/mousedev.c
@@ -962,6 +962,9 @@ static int mousedev_connect(struct input_handler *handler,
 	int minor;
 	int error;
 
+	if (dev->flags & EV_F_NOMOUSE)
+		return -ENODEV;
+
 	for (minor = 0; minor < MOUSEDEV_MINORS; minor++)
 		if (!mousedev_table[minor])
 			break;
diff --git a/include/linux/input.h b/include/linux/input.h
index 1bdc39a..a5654fe 100644
--- a/include/linux/input.h
+++ b/include/linux/input.h
@@ -951,12 +951,20 @@ struct ff_effect {
 #include <linux/timer.h>
 #include <linux/mod_devicetable.h>
 
+/*
+ * input device special flags
+ */
+
+#define EV_F_NOMOUSE		0x0001	/* Do not attach mousedev */
+#define EV_F_NOJOYSTICK		0x0002	/* Do not attach joydev */
+
 /**
  * struct input_dev - represents an input device
  * @name: name of the device
  * @phys: physical path to the device in the system hierarchy
  * @uniq: unique identification code for the device (if device has it)
  * @id: id of the device (struct input_id)
+ * @flags: bitmap of special flags (EV_F_*)
  * @evbit: bitmap of types of events supported by the device (EV_KEY,
  *	EV_REL, etc.)
  * @keybit: bitmap of keys/buttons this device has
@@ -1034,6 +1042,8 @@ struct input_dev {
 	const char *uniq;
 	struct input_id id;
 
+	unsigned long flags;
+
 	unsigned long evbit[BITS_TO_LONGS(EV_CNT)];
 	unsigned long keybit[BITS_TO_LONGS(KEY_CNT)];
 	unsigned long relbit[BITS_TO_LONGS(REL_CNT)];
-- 
1.5.4.4

Re: [PATCH] Input: add flags bitfield

[Dmitry Torokhov]

Hi Henrique,

On Sun, Mar 16, 2008 at 05:14:11PM -0300, Henrique de Moraes Holschuh wrote:
> Add a flags bitfield to the input_dev structure, which can be used for
> internal coordination among kernel input devices and input handlers without
> the need to use ever-expanding blacklists on the input handlers.
> 
> Add initial flag bits which allows an input driver to request that joystick
> emulation (joydev) or mouse emulation (mousedev) not be attached to an
> input device.
> 
> This will be used by accelerometer drivers exporting a raw interface which
> is not to be used as a joystick device (not to confuse this with the usual
> fuzzed joystick interface these drivers export for enhanced Neverball
> productivity), for example.
> 

I'd rather not apply this patch because it pushes kowledge of existing
input interfaces into device drivers. What we could do instead is add
a 'type' field to the input device structure and then input interfaces
(evdev/mousedev, etc) could have an option of matching either by device
type or by device capabilities or both. Your raw devices could have type
of accelerometer and joydev would bind to devices with type "joystick"
or "unknown" + certain capabilities. Will this work?

-- 
Dmitry

Re: [PATCH] Input: add flags bitfield

[Henrique de Moraes Holschuh]

On Fri, 21 Mar 2008 16:07:01 -0400, "Dmitry Torokhov"
<dmitry.torokhov@gmail.com> said:
> On Sun, Mar 16, 2008 at 05:14:11PM -0300, Henrique de Moraes Holschuh wrote:
> > Add a flags bitfield to the input_dev structure, which can be used for
> > internal coordination among kernel input devices and input handlers without
> > the need to use ever-expanding blacklists on the input handlers.
> > 
> > Add initial flag bits which allows an input driver to request that joystick
> > emulation (joydev) or mouse emulation (mousedev) not be attached to an
> > input device.
> > 
> > This will be used by accelerometer drivers exporting a raw interface which
> > is not to be used as a joystick device (not to confuse this with the usual
> > fuzzed joystick interface these drivers export for enhanced Neverball
> > productivity), for example.
> 
> I'd rather not apply this patch because it pushes kowledge of existing
> input interfaces into device drivers. What we could do instead is add
> a 'type' field to the input device structure and then input interfaces
> (evdev/mousedev, etc) could have an option of matching either by device
> type or by device capabilities or both. Your raw devices could have type
> of accelerometer and joydev would bind to devices with type "joystick"
> or "unknown" + certain capabilities. Will this work?

It would solve my problem, yes.

But I'd prefer if joydev and mousedev did not bind to
unknown+capabilities,
just in case.  Looks like bad form to me, and might bite us back later
on.
We can properly fix all drivers in-tree to have suitable types for
joydev
and/or mousedev binds, rfkill binds, and so on after all.

What did I miss?  Are there drivers for some sort of generic device that
would be unable to set their type bitfields at runtime?

BTW: as far as I can see, type should be a bitfield or a list, to allow
a
device to have more than one type.  Is that fine with you?

-- 
  "One disk to rule them all, One disk to find them. One disk to bring
  them all and in the darkness grind them. In the Land of Redmond
  where the shadows lie." -- The Silicon Valley Tarot
  Henrique Holschuh

Re: [PATCH] Input: add flags bitfield

[Henrique de Moraes Holschuh]

On Fri, 21 Mar 2008, Henrique de Moraes Holschuh wrote:
> On Fri, 21 Mar 2008 16:07:01 -0400, "Dmitry Torokhov"
> <dmitry.torokhov@gmail.com> said:
> > On Sun, Mar 16, 2008 at 05:14:11PM -0300, Henrique de Moraes Holschuh wrote:
> > > Add a flags bitfield to the input_dev structure, which can be used for
> > > internal coordination among kernel input devices and input handlers without
> > > the need to use ever-expanding blacklists on the input handlers.
> > > 
> > > Add initial flag bits which allows an input driver to request that joystick
> > > emulation (joydev) or mouse emulation (mousedev) not be attached to an
> > > input device.
> > > 
> > > This will be used by accelerometer drivers exporting a raw interface which
> > > is not to be used as a joystick device (not to confuse this with the usual
> > > fuzzed joystick interface these drivers export for enhanced Neverball
> > > productivity), for example.
> > 
> > I'd rather not apply this patch because it pushes kowledge of existing
> > input interfaces into device drivers. What we could do instead is add
> > a 'type' field to the input device structure and then input interfaces
> > (evdev/mousedev, etc) could have an option of matching either by device
> > type or by device capabilities or both. Your raw devices could have type
> > of accelerometer and joydev would bind to devices with type "joystick"
> > or "unknown" + certain capabilities. Will this work?
> 
> It would solve my problem, yes.
> 
> But I'd prefer if joydev and mousedev did not bind to
> unknown+capabilities, just in case.  Looks like bad form to me, and
> might bite us back later on.  We can properly fix all drivers in-tree
> to have suitable types for joydev and/or mousedev binds, rfkill binds,
> and so on after all.

Drat, removing unknown+capabilities means I'd have to hunt down every
frigging input device in the tree to annotate its type... otherwise, it
would cause regressions re. the handlers.  It is a work that needs to be
done anyway, only adding type metadata to new devices and leaving the
rest to report "unknown" is just icky.

Come to think of it, what about uinput?  It would need to be able to set
the device type as well, otherwise mousedev and joydev, for example,
would not attach to uinput-created input devices.

-- 
  "One disk to rule them all, One disk to find them. One disk to bring
  them all and in the darkness grind them. In the Land of Redmond
  where the shadows lie." -- The Silicon Valley Tarot
  Henrique Holschuh

Re: [PATCH] Input: add flags bitfield

[Dmitry Torokhov]

On Saturday 22 March 2008, Henrique de Moraes Holschuh wrote:
> On Fri, 21 Mar 2008, Henrique de Moraes Holschuh wrote:
> > On Fri, 21 Mar 2008 16:07:01 -0400, "Dmitry Torokhov"
> > <dmitry.torokhov@gmail.com> said:
> > > On Sun, Mar 16, 2008 at 05:14:11PM -0300, Henrique de Moraes Holschuh wrote:
> > > > Add a flags bitfield to the input_dev structure, which can be used for
> > > > internal coordination among kernel input devices and input handlers without
> > > > the need to use ever-expanding blacklists on the input handlers.
> > > > 
> > > > Add initial flag bits which allows an input driver to request that joystick
> > > > emulation (joydev) or mouse emulation (mousedev) not be attached to an
> > > > input device.
> > > > 
> > > > This will be used by accelerometer drivers exporting a raw interface which
> > > > is not to be used as a joystick device (not to confuse this with the usual
> > > > fuzzed joystick interface these drivers export for enhanced Neverball
> > > > productivity), for example.
> > > 
> > > I'd rather not apply this patch because it pushes kowledge of existing
> > > input interfaces into device drivers. What we could do instead is add
> > > a 'type' field to the input device structure and then input interfaces
> > > (evdev/mousedev, etc) could have an option of matching either by device
> > > type or by device capabilities or both. Your raw devices could have type
> > > of accelerometer and joydev would bind to devices with type "joystick"
> > > or "unknown" + certain capabilities. Will this work?
> > 
> > It would solve my problem, yes.
> > 
> > But I'd prefer if joydev and mousedev did not bind to
> > unknown+capabilities, just in case.  Looks like bad form to me, and
> > might bite us back later on.  We can properly fix all drivers in-tree
> > to have suitable types for joydev and/or mousedev binds, rfkill binds,
> > and so on after all.

Ohne word - HID.

> 
> Drat, removing unknown+capabilities means I'd have to hunt down every
> frigging input device in the tree to annotate its type... otherwise, it
> would cause regressions re. the handlers.  It is a work that needs to be
> done anyway, only adding type metadata to new devices and leaving the
> rest to report "unknown" is just icky.
> 
> Come to think of it, what about uinput?  It would need to be able to set
> the device type as well, otherwise mousedev and joydev, for example,
> would not attach to uinput-created input devices.
> 

That's why I thinkg unmarked should really be default and only few selected
devices should set their type.

-- 
Dmitry

Re: [PATCH] Input: add flags bitfield

[Henrique de Moraes Holschuh]

On Sun, 23 Mar 2008, Dmitry Torokhov wrote:
> On Saturday 22 March 2008, Henrique de Moraes Holschuh wrote:
> > On Fri, 21 Mar 2008, Henrique de Moraes Holschuh wrote:
> > > But I'd prefer if joydev and mousedev did not bind to
> > > unknown+capabilities, just in case.  Looks like bad form to me, and
> > > might bite us back later on.  We can properly fix all drivers in-tree
> > > to have suitable types for joydev and/or mousedev binds, rfkill binds,
> > > and so on after all.
> 
> Ohne word - HID.

Bleh.  In that case, it will be too ugly to have it called a type, most
of the input devices will have a type of "generic", which is at the very
least damn ugly.

> That's why I thinkg unmarked should really be default and only few selected
> devices should set their type.

I do think unmarked should be the default, and if it is so hard to have
the emulation (and other handlers) ignore unmarked, we can leave that
active by default as well.  No problems here.

What I don't like is to call it a "input device type", and have bits on
it for joystick and mouse, which almost every mouse and joystick will
NOT set.

At that point, the difference between the bitfield with a handler
whitelist (with blacklist all others implied, "Cisco ACL-style"), and a
type bitfield is just in the name.  The difference from what I already
sent you would be the name, and inverted logic (I sent you a blacklist
patch, not a whitelist patch).

IMO, it doesn't make sense to leave people wondering why their HID
devices (and most other input devices) have a type of "generic", when
they indeed are mouses, joysticks, or whatever.

Another approach would be to call the flags bitfield "emulation type",
set it to "any" as default (all bits zeroed), add one bit for every
emulation-style handler we add (e.g. mousedev and joydev, but not
rfkill), and never bind a handler to a device which has that bitfield
nonzero, unless it has the emulation bit for that handler set.

Do you *really* want a device type bitfield?

-- 
  "One disk to rule them all, One disk to find them. One disk to bring
  them all and in the darkness grind them. In the Land of Redmond
  where the shadows lie." -- The Silicon Valley Tarot
  Henrique Holschuh