[gpm]Bug with IMPS/2 w/ Microsoft Intellimouse 1.3A on Linux

[Pozsar Balazs]

> > I had the same problem, and I figured out two things:
> > - Sometimes the mouse is confused and imps2 init will fail if I do not
> > send a reset to it.
> > - moving the mouse before, or during initialization will make it fail.
>
> This is a problem I think, for all mouse (non trivial) initialization:
> What happens is that when you move the mouse it sends motiondata, as
> datastreaming is on per default. When you afterwards try to communicate
> (init) with the mouse and read the responds you read motiondata instead of
> the expected answar and bail out.
>
> So any mouseinitialization where a respond is expected/needed should first
> turn off datastreaming and flush the fd. Then begin to communicating with
> the mouse. When ended turn on the datastreaming again.

I all figured out this, so that's exactly how the patch works :)

> > The attached patch solved all my problems, so can try it, but this is not
> > official :)
>
> The reason for succes with the patch, is that by waiting for mouse_id_code
> after reset (00) you effectually flush the fd, and then starts
> initialization and read the responds correct. The patch have the faults
> that

> A) datamotion packets can legally take the value of 0.

I was thinking about this too. I do not know why, but the patch worked in
every time: I couldn't make the init with moving the mouse around or
pressing the buttons or scrolling the wheel (but I do tried.).

I thought that we should wait for an 0xAA first, then for a 0x00, and that
should be enough, but I didn't have time for testing this.

> B) moving the
> mouse while initializing can cause the same problem. But it will solve the
> problem in most cases.

Well, as I said above I tested above I tested it on three mice, and I
couldn't make the init fail. Can you? :)

ps: I am about to send a more complete patch, with a seperate reset_ps2
function, and calling that from each i_*ps2, this is not the final version
yet that should be applied. More testing is needed of course.

-- 
pozsy