Trying to freeze and failing ...

[Eric W. Biederman]

ron minnich <rminnich at lanl.gov> writes:

> On Tue, 24 Jun 2003, Andrew Ip wrote:
> 
> > Hi,
> > 
> > > Can anyone who has built a k7sem recently verify that the 'FIXED MTRR set 
> > > to UC' code is not fouling up the K7sem? 
> > The problem is confirmed, and it is related to enabling ecc in msr.
> > If you comment the code, you should be fine.  I guess we need better
> > mechanism to enable or disable in config file.
> 
> Thanks andrew. Anybody know what about the K7 is making this explode? 
> 
> I am going to put an #if around this tomorrow, I hope someone will test. 

Doing this in the config file is inappropriate.  We need a test to see
if ECC is enabled elsewhere on the motherboard.  

A config option may make sense if ECC on/off is a global property of
the motherboard.  Otherwise is should be a callback function that checks
to see if ecc is enabled.

My apologies for not coding that way initially.

As far as what makes this explode putting in non-ECC memory and then
testing the ECC bits should cause a machine check.  Since we don't
trap machine check it would look like an unexpected machine death.

For machines with ECC support we want the explosion if something goes
badly wrong.  So we do no compute with bad data.

Eric