Hi all,

I have a situation where a tcp connection from a Windows to a NetBSD 1.5.1
system is being terminated at the socket level with the error "Connection
reset by peer" (on the Windows box), and at the same time on the NetBSD
system I see the kernel printout : "tlp0: transmit timeout". Does this ring
any bells with anyone?

I will give a bit more information :

1. In general the socket on the target machine closes down cleanly although
I have seen an instance where it also terminated with the "Connection reset
by peer" error. In *every* case I have seen the "tlp0: transmit timeout"
message printed out, either a few seconds before or a few seconds after the
connection was dropped on the NetBSD system

2. It generally takes some time to fail (up to a couple of hours) and the
NetBSD system is pushing out between 2 and 4 Mbits/s.

3. I am only seeing this when connecting to machines with tlp0 cards
(Davicom chipset) - I have been unable to reproduce this connecting to
NetBSD systems with Realtek 8139 cards, and the tests have been running for
days.

4. I am also *only* seeing it when I am connecting from a Windows machine
(which makes me think it could be something funny in the Windows sockets
layer). I have had stable connections running from other NetBSD systems for
days with identical code (although different compiler). I have tried
different Win98,2000 and XP with the same results.

5. I have tested with the 2 systems separated by a 100 meg switch, and with
them going through a gateway machine (down to 10 meg) with the same results.


Is there anyone out there who has seen something like this? My feeling is
that this could be a Windows funny, but the fact that Netbsd is printing
some sort of warning worries me. Is the tlp0 driver rock solid on 1.5.1?

Regards
 Stuart