Trying to freeze and failing ...

[ron minnich]

Can anyone who has built a k7sem recently verify that the 'FIXED MTRR set 
to UC' code is not fouling up the K7sem?

I really do want to freeze the 1.0 tree but we're not converging. Or we 
almost are save for "the last few bugs". I just got some K7SEM's but won't 
have time to look at this for a bit -- if anyone can test this for me with 
the latest CVS that would be helpful.

Andrew Ip, you are probably the best person to test this, can you take a 
look?

ron

[Andrew Ip]

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.

--- freebios/src/cpu/k7/cpufixup.c.bak      2003-04-09 08:01:08.000000000 +0800
+++ freebios/src/cpu/k7/cpufixup.c  2003-06-24 12:41:43.000000000 +0800
@@ -51,7 +51,7 @@
        rdmsr(SYSCFG_MSR, lo, hi);
        printk_spew("SYSCFG was 0x%x:0x%x\n", hi, lo);
        lo |= SYSCFG_MSR_MtrrVarDramEn;
-       lo |= SYSCFG_MSR_SysEccEn;
+//     lo |= SYSCFG_MSR_SysEccEn;
        wrmsr(SYSCFG_MSR, lo, hi);
        rdmsr(SYSCFG_MSR, lo, hi);
        printk_spew("SYSCFG IS NOW 0x%x:0x%x\n", hi, lo);

-Andrew

[ron minnich]

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.

ron

ron minnich  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

[Andrew Ip]

> 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.

I think we can do the following,
if ecc option is on, then 
check for ecc availibility
if ecc is availible, then enable ecc; otherwise, skip ecc.

-Andrew

Andrew Ip  writes:
> > 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.
> I think we can do the following,
> if ecc option is on, then 
> check for ecc availibility
> if ecc is availible, then enable ecc; otherwise, skip ecc.

Sounds reasonable if we have chipsets that do not support ECC.
The conditional logic should be a function defined in a header so
it does not clutter up the code.

Eric

[ron minnich]

On 24 Jun 2003, Eric W. Biederman wrote:
> ron minnich  writes:
> > 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.

obviously, but I would like to get to the point that people can at least
boot from linuxbios instead of watching the explosions. I don't have the
time to do any more than this right now.  Maybe later.

If anybody wants to look at this it's fine by me.

ron

[ron minnich]

On Tue, 24 Jun 2003, Andrew Ip wrote:
> I think we can do the following,
> if ecc option is on, then 
> check for ecc availibility
> if ecc is availible, then enable ecc; otherwise, skip ecc.

can you write the code real quick and run it by me for a sanity check?

ron