3c59x: recurring transmit errors

[Ben Liblit]

I am running Linux 2.0.35 with 3c59x driver version 0.99E on a 100
megabit network.

    3c59x.c:v0.99E 5/12/98 Donald Becker http://cesdis.gsfc.nasa.gov/linux/drivers/vortex.html
    eth0: 3Com 3c595 Vortex 100baseTX at 0xff80, 00:a0:24:c5:ef:f0, IRQ 7
      Internal config register is 101001b, transceivers 0xe10a.
      64K word-wide RAM 3:1 Rx:Tx split, autoselect/10baseT interface.
    eth0: Overriding PCI latency timer (CFLT) setting of 64, new value is 248.
    eth0: Initial media type 100baseTX.
    eth0: vortex_open() InternalConfig 0141001b.
    eth0: vortex_open() irq 7 media status 8802.
    eth0: Media selection timer tick happened, 100baseTX.
    eth0: Media 100baseTX has link beat, 8882.
    eth0: Media selection timer finished, 100baseTX.

Several times a day, I seem to lose all network connectivity through
this interface.  This is an NFS-intensive environment, so NFS timeouts
are generally the first symptom.  The driver reports transmit errors
with one of three different status registers:

    eth0: Transmit error, Tx status register 90.
    eth0: Transmit error, Tx status register c0.
    eth0: Transmit error, Tx status register 90.
    eth0: Transmit error, Tx status register 90.
    NFS server asterope.CS.Berkeley.EDU not responding, still trying.
    NFS server asterope.CS.Berkeley.EDU not responding, still trying.
    eth0: Transmit error, Tx status register d0.
    NFS server asterope.CS.Berkeley.EDU OK.
    NFS server asterope.CS.Berkeley.EDU OK.

"ifconfig" records the following tally of problems:

    TX packets:734220 errors:0 dropped:0 overruns:8 carrier:8 coll:872

The network generally returns after about one to five minutes.  During
that span, the CPU is pegged at 100% by a pair of "nfsiod" processes.
A "traceroute" reports "host unreachable" for any address I might care
to try.

During these network outages, other machines on the same subnet
continue to operate normally.  Also, I have never experienced similar
outages when running NT on the same machine.  So at this point I am
inclined to suspect the Linux 3c59x driver.

Any ideas?