On Fri, Jan 06, 2012 at 09:14:34PM +0100, Henrik Rydberg wrote:

On Fri, Jan 06, 2012 at 12:02:40PM -0800, Dmitry Torokhov wrote:

quoted

On Fri, Jan 06, 2012 at 08:48:15PM +0100, Henrik Rydberg wrote:

quoted

quoted

quoted

Ok, maybe not to so easy after all, which probably answers my own
question. Looks like a EVIOCGMTSLOT, taking both slot and event code
as argument, would be the cleaner route to take. Another ioctl, how do we
feel about that?

What's the problem with userspace locking?

The idea was to get by without it.

Regarding ioctls, it does not seem realizable due to ioctl number
exhaustion.

I am prettu sure we cabn spare 1 ioctl number. We just need to pass slot
number not as part of ioctl number but in data instead. Like
EVIOCGKEYCODE works.

Right, thanks. Perhaps we could even pass it as result data -
returning all mt data in one go? With the purpose being to capture the
full MT state, it ought to be both simpler and faster.

Yes, if caller passes buffer size and buffer address we can return all
data in one go. We just need to make _really sure_ we are using 32/64 bit
safe interface.

-- 
Dmitry