[lm-sensors] sensors-detect killed my CPU

Achim Gottinger achim at ag-web.biz
Mon May 5 21:44:35 CEST 2008


Jean Delvare schrieb:
> Hi Achim,
>
> On Mon, 05 May 2008 14:15:34 +0200, achim wrote:
>   
>> He He, the BKDG is tough to read. I tried to figure out the SMbus
>> adresses.
>>
>> http://www.abload.de/image.php?img=baredit-sidebandeis.jpg
>>
>> This is the register where the adress should be in bits 4:6 whom are all
>> zero. However bit 3 is set and that should mean that the sideband
>> interface is used.
>>
>> http://www.abload.de/image.php?img=sidebandumy.jpg
>>     
>
> This isn't how I read the specification. Bit 3 set means that the SBI
> is _disabled_. But it your case, bit 3 is _not_ set, is it?
>
> I wonder where the 4 higher SMBus address bits are set.
>
>   
Yepp it's unset means SBI is enabled. Can be a bios issue that the 
correct address does not shop up here.
That board also does strange things with the cpu and nb vid. No matter 
what i set in the bios the current-p-state register allways
shows 1.3V. The sensor-chip reads the correct values.
The voltage register works to set p-states. I tweaked the p-state-0 and 
p-state-1 registers once in a way that CnQ worked with
400MHz at 0.8V in idle and 2.6GHz at 1.4V under load. With CnQ enabled 
it uses the vid's from those two registers (per cpu).

I'd love to get the readings of those SVI SMbus interfaces. That way it 
should be possible to read the current northbridge (on chip) voltage, 
whom is not accessible via the sensor-chip and i can verify that voltage 
only with an DMM at the moment.
> So there must be yet another chip which doesn't like being probed. This
> confirms my impression that the safest fix is to blacklist the
> motherboard and disable the SMBus entirely. Although I understand that
> you don't want to do this now, as you are still investigating
It's i2cdetect 0 itself. I need to reboot after i ran this to dump the 
special chips. Otherwise I only get XX's. This issue only occures with 
the phenom and not with the X2 cpu.
All chips beside those at 0x2e,0x47 and 0x6e can be fully dumped without 
causing the XX-issue.
Today I wrote a long problem report for the sapphire support and asked 
them what chip they use at 0x2e. Hope they will shed some light on this 
issue.
I'll post their response here.





More information about the lm-sensors mailing list