[vortex-bug] 3c556 TX bug ???

[Felix Lindner]

Andrew Morton wrote:
> 
> Felix Lindner wrote:
> > 3c59x: Donald Becker and others. www.scyld.com/network/vortex.html
> > 02:06.0: 3Com PCI 3c556 Laptop Tornado at 0xe800. Vers LK1.1.16
> >
> > When doing a netcat to a echo-like server such as this:
> > nc -n -v 1.2.3.4 7 </dev/urandom
> > and having a tcpdump running beside that, the communication would stop
> > after aprx 30 seconds. Then, I could see traffic on the network using
> > the tcpdump but my NIC did not send anything anymore. I assume the TX
> > queue raised an error and awaits a reset from the driver it never gets.
> >
> > Any ideas on what this is? Could that be related to the driver messages
> > I see once in a while in syslog:
> >  eth0: Transmit error, Tx status register 90.
> 
> Transmit underrun.  The NIC wasn't able to read
> data from main memory across PCI fast enough.
> 
> That's unusual.
> 
> The driver does issue a Tx Reset in response to this.
> 
> I'm not sure what could be causing this.

The later (Tx status register 90), I see pretty often with this system,
but I never experienced the issue with no sending before. 
Uh, and here comes the OS info I was not including (sorry for that):
Linux 2.4.13.
When I see the Transmit underruns in syslog(1), the interface will work
as it did before. But with the explained netcat issue, I had to
'ifconfig eth0 down' and 'up' it to make it work again. 

Thanks for the response,
Felix