On Sat, Mar 06, 2010 at 01:12:51AM +0100, Carl-Daniel Hailfinger wrote:
On 05.03.2010 19:26, ron minnich wrote:
What would coreboot need to do to "support" IPMI BMC?
Depends on the IPMI BMC. If you're lucky, it works out of the box, and if you're unlucky, you have to implement undocumented BIOS interfaces. The easiest solution is to buy a card and try it, and if it doesn't work, reverse engineer it or try another card.
Besides that, IPMI is a security nightmare (see the discussions on the linux-kernel mailing list about IPMI bypassing host network security).
Even worse - I have yet to encounter a reliable IPMI card. The sorts of problems I've encountered are:
* specific packets can 'kill' the IPMI controller. Once the thing is hung, the only fix is a *cold* boot of the entire machine. * I've seen machines crash, taking the IPMI controller with them. Makes the whole thing kind of pointless... * general reliability issues. IPMI controllers also seem to like to hang themselves occasionally
I really tried to make IPMI work reliably; I have an 80 machine cluster full of these things. I wasted a ton of time on them (3 different generations from 2 vendors - Tyan and Supermicro).
I think that those issues were largely caused by extremely crappy proprietary firmware. But there is a more fundamental issue; the IPMI BMC is pretty tightly connected into the mainboard, by design. That's bad - how can you guarantee that IPMI BMC will always be available, fully out of band, when it is not 100% independent of the mainboard?
In the end I gave up; I now use serial console servers (opengear is highly recommended) and switched PDUs (I've tried various brands, so far I like Raritan's Dominion series the best). That works, 100% of the time.
Thanks, Ward.