Weird network freezes

[Ian Clark <mrrooster%gmail.com@localhost>]

Hi All,

For a while now I've been having some weird network issues that I'm
having real trouble tracking down.

It usually occurs when copying files from a windows maching to/from a
samba share on NetBSD, although I don't think it's related to samba, I
think that's just a good program for showing the issue.

As an example, I'll open a rar file from a samba share on my server,
and extract the contents to another share in a different folder on the
server. (so both reading and writing to the machine.)

The machine is an AMD box (Phenom, but not the buggy one), with an AMD
780G motherboard.

It has 2 intel gigabit NICs, using the wm driver, one connected to my
internal network one connected to the DSL router.

The network is gigabit, and as I was having this issue before I
recently moved house (hence new cables) I don't think it's a cable
issue.

My switch is a Netgear 24port gig switch, which replaced the much
cheaper generic brand 8 port I had before. (But didn't fix anything.)

I used to get the same issue with a realtek card, which I replaced for
the intel ones as the consensus seemed to be they were about the best
to use with NetBSD.

I can't see anything obvious in the samba logs, I can't see anything
obvious in the syslog, I can't see anything obvious in the dmesg
output. I can't see anything in the windows event logs.

The only indication I can see that anything has gone wrong is sysstat
shows an increase in input errors, and windows fails with a 'network
name not found' (although it immediatly reconnects.)

Wed Aug  3 11:35:09 BST 2011
Name  Mtu   Network       Address              Ipkts Ierrs    Opkts Oerrs Colls
wm0   1500  <Link>        00:1b:21:55:99:14 678509067   239 1075919645
    0     0
wm0   1500  192.168.0/24  ohmyno.co.uk      678509067   239 1075919645
    0     0
wm0   1500  fe80::/64     fe80::21b:21ff:fe 678509067   239 1075919645
    0     0

The network config is:

wm0: flags=8b43<UP,BROADCAST,RUNNING,PROMISC,ALLMULTI,SIMPLEX,MULTICAST>
mtu 1500
        
capabilities=7ff80<TSO4,IP4CSUM_Rx,IP4CSUM_Tx,TCP4CSUM_Rx,TCP4CSUM_Tx,UDP4CSUM_Rx,UDP4CSUM_Tx,TCP6CSUM_Rx,TCP6CSUM_Tx,UDP6CSUM_Rx,UDP6CSUM_Tx,TSO6>
        enabled=0
        address: 00:1b:21:55:99:14
        media: Ethernet autoselect (1000baseT
full-duplex,flowcontrol,rxpause,txpause)
        status: active
        inet 192.168.0.254 netmask 0xffffff00 broadcast 192.168.0.255
        inet6 fe80::21b:21ff:fe55:9914%wm0 prefixlen 64 scopeid 0x1


The machine is using pf, acting as a firewall/nat box, the machine has
a wireless card bridged with the intel NIC (but I had this problem
long before this was setup.) I've disabled all the hardware
checksumming on the intel card too (because of an issue with the
bridge, not related to this.)

I also have a torrent client running fairly constantly, which seems to
exaserbate the problem somewhat.

I'm running a 5.99.51 kernel over a 5 userland. (This problem occured
with a stock 5 kernel and uerland, built from the netbsd5 branch.)

As I think I've just about replaced everything I can and tried all the
obvious stuff does anyone have any suggestions? Is it time to start
fiddling with PCI settings in the bios for example? (The sporadic
nature of the issue and lack of obvious issues in log files makes me
think it's something more low level.)

Any help much appreciated.

Cheers,

Ian