On Tue, 18 Jan 2005 13:58:20 -0800, Greg KH <[EMAIL PROTECTED]> wrote: > On Tue, Jan 18, 2005 at 04:49:34PM -0500, Dmitry Torokhov wrote: > > On Tue, 18 Jan 2005 13:30:02 -0800, Greg KH <[EMAIL PROTECTED]> wrote: > > > On Tue, Jan 18, 2005 at 03:56:35PM +0100, Hannes Reinecke wrote: > > > > Hi all, > > > > > > > > the input subsystem is using call_usermodehelper directly, which breaks > > > > all sorts of assertions especially when using udev. > > > > And it's definitely going to fail once someone is trying to use netlink > > > > messages for hotplug event delivery. > > > > > > > > To remedy this I've implemented a new sysfs class 'input_device' which > > > > is a representation of 'struct input_dev'. So each device listed in > > > > '/proc/bus/input/devices' gets a class device associated with it. > > > > And we'll get proper hotplug events for each input_device which can be > > > > handled by udev accordingly. > > > > > > Hm, why another input class? We already have /sys/class/input, which we > > > get hotplug events for. We also have the individual input device > > > hotplug events, which is what I think we really want here, right? > > > > These are a bit different classes. One is a generic input device class > > device. Then you have several class device interfaces (evdev, > > mousedev, joydev, tsdev, keyboard) that together with generic input > > device produce concrete input devices (mouse, js, ts) that you have > > implemented with class_simple. > > Hm, but we still need to make the input_dev a "real" struct device, > right? And if you do that, then you just hooked up your hotplug event > properly, with no userspace breakage.
I wasn't planning on doing that. The real devices are serio ports, gameport ports and USB devices.They require power and resource management and so forth. input_device is just a product of binding a port to appropriate driver and seems to me like an ideal class_device candidate. Then you add couple of class interfaces and get another class_device layer as a result. > Then, if you want to still make the evdev, mousedev, and so on as > class_device interfaces, that's fine, but the main point of this patch > was to allow the call_usermodehelper call to be removed, so that the > input subsytem will work properly with the kernel event and hotplug > systems. > I was mostly talking about the need of 2 separate classes and this patch lays groundwork for it althou lifetime rules in input system need to be cleaned up before we can go all the way. -- Dmitry - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/