On Wed, 2009-11-11 at 01:59 -0800, Dmitry Torokhov wrote: > On Wed, Nov 11, 2009 at 10:52:53AM +0200, Artem Bityutskiy wrote: > > Sorry Dmitry, I'm new to the input subsystem, may be this is why I still > > do not see any problem with introducing a nice and useful ioctl? I do not insist on ioctl, if you can show a nice way to use sysfs. So far you did suggest this way, and I also do not see any nice way. Really, we are happy to use sysfs, just show us how. > There are several reasons: > > - ioctls are not useful from scripts Not valid point. There are tons of ioctls all over the place. And no one prevents us from creating a userspace program which the scripts can use. And your sysfs proposal is unusable for both the scripts and the C programs. It is not any better. > - 32/64 bit comapt issues Invalid point. Real compat issues exist only for ioctls which pass pointers. We are not going to do this at all. > - currently there islimited number of ioctls in input core, they are all > concentrated in joydev and evdev, almost all of them work on all > devices, verifying locking is easy OK, but this does not look like a blocker for introducing one more ioctl. > > On Wed, 2009-11-11 at 00:19 -0800, Dmitry Torokhov wrote: > > > > > I understand the appeal of working with the keycodes but what you are > > > > > asking is not simply ignoring certain keypresses (that is easy, just > > > > > ignore corresponding events in userspace), you want to prevent system > > > > > from waking up. In other words you want to "control PM layer through > > > > > input layer" and I believe this is wrong. > > > > > > > > Err, this is really about disabling keys. > > > > > > If you want to stop delivery of certain keycodes to userspace then OK, > > > an option to subscrbe to certain events only instead of getting all > > > events from the device. But again, it will only work for that particular > > > user only, not everyone. > > > > Yes, this will work for some particular input devices only, those which > > support the keys masking. What is wrong with this? We have the '-ENOTTY' > > error code, which for ioctl's means "ioctl is not supported". > > > > But no one prevents us from making this work for _every_ input device by > > implementing a default ioctl handler in evdev.c, if needed. This handler > > can mask certain keys in SW. > > > > IOW: > > 1. if my HW supports HW key masking, the evdev ioctl calls my > > handler, which is implemented in my driver. > > 2. if my HW does not support this, the evdev ioctl masks the keys in > > SW. > > > > We are interested in 1. And if later someone needs 2., he/she can very > > easily implement that. But for now we can just return -ENOTTY (or > > -EINVAL, which seems to be used in the input subsystem). > > And #2 is what I think input layer should provide. Something that > works for all devices and is isolated (one client does not affect > others). We can do this, if you believe it normal if we implement code we do not use and thus, we cannot really test. I do not think it is normal, but if you insist, we can try doing. > > > > If we think about this in an > > > > abstract way, you have an input device with many keys. And the device's > > > > HW allows you to disable any of it's keys. When you disable the key, the > > > > input device ignores it and does not generate any interrupt for it. > > > > > > And the reason you want to stop interrupts? > > > > The reason is that I want simply mute certain keys. This is the only > > reason. > > > > Now the question is _how_ I want to mute the key. My HW supports muting, > > so I want to mute it in HW level, just because it is much better and > > more efficient. Just because in this case I will not have any interrupt > > and my device will wake up less. > > And the reason you care about number of wake ups is not saving power by > any chance? Yes it is. But there could be many other reasons as well. E.g., we may want to achieve smaller latency by lessening amount of interrupts. Really, you can look at this as we simply want to disable specific keys. For exampe, for power savings. Or for smaller latency because of less interrupts. Whatever. -- Best Regards, Artem Bityutskiy (Артём Битюцкий) -- To unsubscribe from this list: send the line "unsubscribe linux-input" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
