Re: 3c905c arp/configuration problems
On Mon, Jan 14, 2002 at 10:54:38PM -0500, Angus D Madden wrote:
>
> I have a small (2 machine) home network. The gateway is
> running 2.4.16 with Debian Woody. Both of the NICs on the gateway
> are 3c905C's and are up and running with no problems.
>
> The client box also has a 3c905C. It was initially running WindowsME
> with dhcp used to configure the network interface. Everything was
> working fine, network wise.
>
> Repeated problems with the Widnows and an eventual disk failure have
> prompted me to install Debian on the client box.
>
> I could not get the network interface to function after I
> installed Debian (from potato rev 2 install disks 2.2.17 kernel).
> The 3x59x module inserts cleanly, I can bring the interface up, but I
> can't get anywhere on the network.
>
> I know that the cards autonegotiate cleanly because the right lights
> come up on the switch (100 Mbps, Full Duplex).
Does the connect light come up on the NIC in the client? The NIC in the gateway?
> When I ping from the client box to the gateway, I can see the traffic on
> the switch, and I can see the echo request reach the gateway. The
> gateway then tries to send an echo-reply. First, it tries to determine
> the ethernet address of the client box - unsuccessfully. Using tcpdump
> I can see that the arp request get sent by the gateway but not received
> by the client.
Thus traffic send from gateway to client doesnot get through.
> Likewise, when arp requests are sent from the client to the gateway, I
> can see them on the gateway. The gateway sends the is-at
> reply, but this is never seen by the client.
Same. Traffic send from gateway to client doesnot get through.
> After the pinging process, the arp caches on both machines show
> (incomplete) for a HWaddress. I have tried to manually set the
> HWaddress for both machines, but that has been unsuccessful (it gets
> erased whenever one of the cards talks to the network).
Because the who-has, is-at handshake fails.
Most probably I would pin this down to a broken cable line between the gateway and the client somewhere. Either a broken twisted pair in one of your cat-5 cables, a broken pin in one of your NICS or a dry solder joint creating a broken connection in a NIC. I've had this *exact* connundrum before and it turned out to be hardware (in my case it was the port I was using on the switch which was dead, and another port worked fine). Basically all the traffic is one way. The client can talk to the gateway but the gateway cant talk to the client.
It maybe the vortex, because they've had lots of problems with HP changing the spec's on them and breaking the OSS drivers. Theres a page by the guy that wrote the original driver explaining the hassles.
>
> I have booted the client with a Widnows rescue disk in order to use the
> 3com dos utilities and everything looked fine. I can follow with the
> vortex debug output and vortex-diag output if necessary.
>
> Here is a sample from the vortex debug in syslog (abbreviated):
>
> eth0: Trying to send a packet, Tx index 3
> eth0: interrupt, status e201, latency 1 ticks
> eth0: In interrupt loop, status e201
> eth0: exiting interrupt, status e201
>
> ifconfig on the client machine shows plenty of Tx packets but zero Rx
> packets.
See. Broken wire!
> I think I can rule out hardware problems because everything worked when
> the client was windows.
Are you sure something hasn't changed. No-ones tripped over the cat5 or stepped on the switch or anything?
> Can anyone give me a pointer on resolving this? I've been googling and
> rtfming all day and I can't get my head around it.
First time I had this prob it took me weeks. Then when I changed the switch port and it all worked I spent a week banging my head against my desk :)
Kind Regards
Crispin Wellington
Reply to: