more NetGear mising ARPs

[Neale Banks]

With 2.0.36 and tulip.c:v0.89H on a not-so-slow machine (349.80 bogomips)
I appear to be bumping into this problem :-(

Originally this machine was happily working, plugged in to a NetGear
10/100 "hub" (IIRC, a DS516) on a busyish network.  The machine was moved
to a 10Mb hub on a _quiet_ subnet and now has missing-arp problems - worse
with the Cisco 25xx which is the gateway on this subnet (the other linux
box on the subnet is OK if you first arp from that, with the Cisco it's
much less inclined to talk (even with attempting to talk from the Cisco)).

>From dmesg:

tulip.c:v0.89H 5/23/98 becker@cesdis.gsfc.nasa.gov
eth0: Lite-On 82c168 PNIC at 0xe800, 00 a0 cc 40 0a a7, IRQ 10.
eth0:  MII transceiver found at MDIO address 1, config 1000 status 782d.
eth0:  Advertising 01e1 on PHY 1, previously advertising 01e1.
eth0: Changing PNIC configuration to half-duplex, CSR6 816e0000.

>From /proc/pci:

PCI devices found:
  Bus  0, device  11, function  0:
    Ethernet controller: Lite-on LNE100TX (rev 33).
      Medium devsel.  Fast back-to-back capable.  IRQ 10.  Master Capable.  Latency=64.  
      I/O at 0xe800.
      Non-prefetchable 32 bit memory at 0xe9000000.

Turning debug up to three didn't produce anything obviously "interesting"
(but the first time I booted with debug=3 the arp to the Cisco worked,
subsequent tests suggest this may have been a fluke).

>From ifconfig:

eth0      Link encap:Ethernet  HWaddr 00:A0:CC:40:0A:A7  
          inet addr:a.b.c.20  Bcast:a.b.c.23  Mask:255.255.255.248
          UP BROADCAST RUNNING MULTICAST  MTU:1500  Metric:1
          RX packets:266 errors:0 dropped:0 overruns:0 frame:0
          TX packets:247 errors:0 dropped:0 overruns:0 carrier:0
          Collisions:0 
          Interrupt:10 Base address:0xe800

On occasions where the system is failing arp for all its neighbours, Tx
packets increments but Rx packets does _not_ and the arp entries are
marked as incomplete.

Anyone have any idea what might be happening here, or how to work around
this?

I'll try to load and compile v0.91 of tulip.c tomorrow to see how that
performs.

Thanks,
Neale.