gpm/serial bug?

Tyson D Sawyer (tyson@rwii.com)
Sat, 16 Nov 1996 00:40:35 -0500 (EST)

I just spent too many hours figuring out that gpm was running and
shouldn't have been. I had a serial device attached to /dev/cua0
on a headless, sort of embedded, Linux PC. A short time after
starting the program that controls the device on the serial port
the the kernel would start thinking that bogus data was being typed
on the console. I was getting illegal character in name type messages
from getty. If I plugged in a monitor I could see the bogus data.
If I plugged in a keyboard I could inject correct key strokes in
amongst the bogus ones. If I shut down the controlling program and
the device then the problems stopped. If I let it keep running the
system would slowely die until it would do little more than respond
to pings on the network.

I experienced the problem with 2.0.14, 2.0.22 and 2.0.25.

Questions:

1) Does gpm do funny things that would make it suspect or does this
"have to" be a kernel problem?

2) Has anyone else seen anything like this before? I am accustomed
to having one program steal serial data from an other or
other messed up behavior like that, but I have neve before
seen messed up behavior from the kernel in a situation like
this before.

Getting rid of gpm and resetting the system cleared the problem. I suspect
something like gpm behaving badly when recieving data it doesn't understand.

Thanks,
Ty