Which motherboards work well with em(4)?

[Chris Howells]

On Tue, February 21, 2006 1:49 am, Jack Vogel wrote:

>> Lol. I did wonder if the Intel NICs might be programmed to do this in
>> protest at being in a board with a Via chipset and AMD CPU ;)
>
> Nah, not at all. There is a certain OEM that has a large number of
> AMD based servers and they buy Intel LOMs to put on their system
> boards (they are named after that big yellow ball in the sky :).

Hah.

> hmm, well these arent anything bleeding edge, so isnt any hardware
> issues that occur to me

OK. I have reported this problem in the past and a few people like Gleb
Smirnoff and Christian Peron have helped in diagnosing and providing
patches. The problem always seems to be fixed then comes back some
unspecified time later. According to an email from Christian earlier today
I am the only person still reporting the problem :(

Since they are different ethernet controllers I am wondering if it could
be the motherboard failing to deal with the interrupts, or maybe even the
ethernet switch. Polling appears not to help though. The switch is a cheap
SMC EZ 5 port, I've been comtemplating getting a cheap 5 port 3Com switch
from eBay to test.

> Yes, although after I sent this email i noticed a couple reports of panics
> that made me a bit leary.

I'm installing -current on a spare disk at the moment, so hopefully I'll
have a go with it shortly.

> It sounds like you have transmission hangs, about how frequently is
> this happening, and is there any other info about the situation that
> you can characterize? Like is the NIC running at gig speed, or is it
> lower, full duplex, etc etc...

It's really hard to explain because it's a very intermittent. But both
NICs are connected to the same 5 port switch SMC EZ GigE switch. They are
both in auto negotiate mode (I tried forcing full duplex but it didn't
make any difference).

My network layout is something like this:

  Windows XP PC, nVidia GigE
         |
    SMC EZ GigE Switch
         |
   SMC EZ GigE Switch  -------`
  |           |                |
6.0-REL     6.1-pre           Netgear 100MBit switch
machine     machine


The cards just stop passing any data, seemingly at random. 'ifconfig em0
down; ifconfig em0 up' fixes it, as does, IIRC, unplugging the ethernet
cable from the switch and plugging it in again. The cable is cat5e, quite
short (less than 1m), and I've tried various other cables without success.

Plugging the machine into the 100Mbit switch instead seems to prevent the
problem from occurring, so it seems to only be higher GigE speeds which
cause it.

For instance, a couple of days ago I was restoring some data from a tape
accross the network, probably maxed out at 3Meg. After a few minutes the
receiving card just wedged. Not particularly intensive.

On the other hand this afternoon I was testing and successfully copied 8GB
or so of data from the XP machine to the machine running 6.1-pre. So I
started transferring the same lot of data to another folder on the samba
share and very soon after it wedged.

Basically completely unpredictable. :(

Thanks.

-- 
Cheers, Chris Howells -- howells at kde.org
Web: http://www.kde.org.uk, PGP ID: 0x33795A2C
KDE/Qt Developer: http://www.kde.org