freebsd stable - Dec 2005 - Lockup with powerd on amd64/tyan k8sd pro

Hi,

I've just purchased a new dual-Opteron 2MHz Tyan K8SD Pro machine, with 
one of the SiS SATA raid controllers (which isn't supported, but I'm 
using gmirror to great success, though that was configured after the 
problem below).

I've installed 6-stable as of about two weeks ago. I tried out using 
powerd in 'debug' mode (adaptive by default I guess), more for my own 
interest's sake, and it locked up the machine solid. Prior to the 
lock-up the messages reported is was dropping through the various MHz 
steps, and then the lock-up when it reached what I'm guessing was the 
lowest level. I was working on the first vty and the lock-up meant I 
couldn't switch consoles, or reboot with ctrl-alt-delete etc, but I 
didn't see any panic messages. Being a kernel miser I've stripped out 
debugging options and so on, but just raising it in case its of 
interest. I'll upgrade to the latest stable and try it out again now, 
but also willing to try other suggestions if its not a known issue.

Aside from that, this is also a bit of a recommendation for the 
configuration I've got, it seems extremely slick. I've compiled the JDK 
1.5 port and its the fastest one I've tested under FreeBSD, albeit thats 
the hardware and CPU's that will be making most of the difference versus 
my other lowly machines - I was just very happy to see it working on 
amd64 (JBoss under load with Postgres moving the data is the next test).

Cheers,
Joe

On Thu, Dec 01, 2005 at 07:07:35PM +1100, Joe Shevland
wrote:
> I've installed 6-stable as of about two weeks ago. I tried out using 
> powerd in 'debug' mode (adaptive by default I guess), more for my
own
> interest's sake, and it locked up the machine solid. Prior to the 
> lock-up the messages reported is was dropping through the various MHz 
> steps, and then the lock-up when it reached what I'm guessing was the 
> lowest level.
Could you post a full sysctl dev.cpu and the same with sysctl
dev.powernow?
Is the lowest frequency the only one which cause that look-up?
If you change manually the frequency via a sysctl

dev.cpu.0.freq=<something other than the lowest one>

will tell us if that's the case.

I think anyway that I found what is wrong.  I think the 2 processors are
doing a transition at the very same time, and it's possible that will
confuse the northbridge.

I don't have time to write a fix right now.  I will post a patch to be
tested latter.

Thanks,

-- 
Bruno Ducrot

--  Which is worse:  ignorance or apathy?
--  Don't know.  Don't care.