Hi,

I am running windows XP, virtualised using KVM (with VT-d) and I am trying to upgrade the firmware on the radios we make. This uses a serial port and I can do a lot of interaction with the radio, like reading its configuration etc, but when I try to download a new firmware to it, it fails. After a lot of research into exactly what's happening we have discovered that it occurs when an ascii NULL is being sent. It comes out the other side of KVM as either 0xFF or 0x80 (depending on how you measure it). 0xFF came from a wire sniffer and 0x80 came from directing output to a file, but curiously, in the file, I sent 10 NULLs and only every second character came out as 0x80. It isn't the serial driver in the kernel as I can download firmware using wine (I just can't do anything else, like read the radio's config). Does anyone have any ideas on what might be happening?

Thanks,
Michael

In 'normal data' there would be frequent transitions from 1 to 0 and back. Normally, the hardware would be able to 're-sync' on those if there was a tendency to drift. I wonder whether, with 10 bytes of null (or even 2 bytes of null) the hardware (in the KVM) is drifting off spec and causing the problems.

Have you tried without the KVM? DOes the problem persist?

Just a thought.

Perhaps my original post was misleading. I meant the Kernel Virtual Machine not a KVM switch. I don't think it's a timing issue because, before the NULL occurs in the file, something like 400 characters have already been sent successfully. It only doesn't handle NULL chars. I don't know the rs232 spec in the slightest, but maybe it's KVM's implementation that allows it to drift off-spec thus fooling the devices. I will ask around. In the meantime, if anyone has any ideas on how I could fix KVM, that would be great. I can program, I'm just not good with hardware issues Thanks,

Michael