IPMI and Dell ERA/O

[Jeremy Chadwick]

On Fri, Aug 29, 2008 at 02:17:17AM -0400, Zaphod Beeblebrox wrote:
> Curiously, IPMI shares the ethernet ports with the onboard ethernet
> controllers without FreeBSD's knowledge.  It does use a different MAC
> address.  It is also apparently capable of using vlans (haven't tested this
> yet).  I'm most nervous about how this might behave if the port was being
> nailed with traffic --- but I can't easily test this to my satisfaction.
> What controls the contention for the port between whatever IPMI magic is
> going on and the OS use of the port?

My general opinion is to avoid IPMI at all costs.  The concept itself
is great, and the design idea is okay, but the implementation is an
atrocity.  If you *must* use IPMI, get an implementation that uses
its own, dedicated NIC.

That said, the feature you're referring to (IPMI piggybacking on top of
an existing NIC on the mainboard) is called "ASF" from a NIC driver
perspective.  The NIC driver for the OS *must* have full awareness of
said piggybacking, and if it doesn't, a couple different things can
happen:

a) NIC simply does not work
b) NIC works, but behaves oddly -- usually this is tracked down to
   the local network seeing the MAC address continually change for
   the IP address associated with the machine
c) NIC works, but IPMI and other features do not work

There are a couple different drivers for FreeBSD which have ASF
knowledge; bge(4) does, and I believe em(4) does (I could be wrong
here).  bge(4) has a loader.conf tunable that tells the driver to
understand ASF or not.

In general, it's horrible, and I feel sorry for driver authors having to
deal with it.

-- 
| Jeremy Chadwick                                jdc at parodius.com |
| Parodius Networking                       http://www.parodius.com/ |
| UNIX Systems Administrator                  Mountain View, CA, USA |
| Making life hard for others since 1977.              PGP: 4BD6C0CB |