[lm-sensors] System instability with Supermicro X8DTi-F and

[lm-sensors] System instability with Supermicro X8DTi-F and

[Erik Brakkee]

Hi,


I encountered major system instability when using the w83795 module 
together with fancontrol on a X8DTI-F motherboard running opensuse 11.3. 
The processor is an Intel Xeon L5630.
I did the experiments early this year with version 0.7 of the driver. 
Actually fancontrol really worked well, keeping the system really quiet.

However, I experienced random hangs of the system and also there were a 
lot of events in the logs such as these:

    1 | 12/16/2010 | 14:30:02 | Fan #0x13 | Lower Non-critical going low
    2 | 12/16/2010 | 14:30:02 | Fan #0x13 | Lower Critical going low
    3 | 12/16/2010 | 14:30:03 | Fan #0x13 | Lower Non-recoverable going low
    4 | 12/16/2010 | 14:30:05 | Fan #0x13 | Lower Non-recoverable going low
    5 | 12/16/2010 | 14:30:06 | Fan #0x13 | Lower Critical going low
    6 | 12/16/2010 | 14:30:06 | Fan #0x13 | Lower Non-critical going low

The theory somehow was that there was concurrent access by the embedded BMC controller and the driver.

After disabling fancontrol the system went from highly unstable to rock solid. I even tried monitoring the system at periodic intervals using the w83795 driver and that resulted in the same types of events, exactly coinciding with the periodic polling.

In other words, I am quite clear that this is related to the driver or at least an interworking between the driver and the embedded BMC.

I have now switched off fancontrol permanently but would like to eventually enable it again because the system is just much more quiet (it is a home server with low load).

Any suggestions? Have similar problems been found?

Cheers
   Erik





_______________________________________________
lm-sensors mailing list
lm-sensors@lm-sensors.org
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors

Re: [lm-sensors] System instability with Supermicro X8DTi-F and

[Philip Pokorny]

Your motherboard BIOS includes fan speed control and a BMC for sensor  
monitoring

Use ipmitool and perhaps bmcsensors module to monitor sensors.

Go into the bios and under hardware monitoring arrow down to see fan  
speeds and then select the ES fan profile (lowest) rather than Off,  
performance, balanced. That should make the fan speeds much slower and  
quieter

Phil P.

-- 
Philip Pokorny, RHCE
Chief Hardware Architect
PENGUIN COMPUTING, Inc
www.penguincomputing.com

On Apr 12, 2011, at 4:37 AM, Erik Brakkee <erik@brakkee.org> wrote:

> Hi,
>
>
> I encountered major system instability when using the w83795 module  
> together with fancontrol on a X8DTI-F motherboard running opensuse  
> 11.3. The processor is an Intel Xeon L5630.
> I did the experiments early this year with version 0.7 of the  
> driver. Actually fancontrol really worked well, keeping the system  
> really quiet.
>
> However, I experienced random hangs of the system and also there  
> were a lot of events in the logs such as these:
>
>   1 | 12/16/2010 | 14:30:02 | Fan #0x13 | Lower Non-critical going low
>   2 | 12/16/2010 | 14:30:02 | Fan #0x13 | Lower Critical going low
>   3 | 12/16/2010 | 14:30:03 | Fan #0x13 | Lower Non-recoverable  
> going low
>   4 | 12/16/2010 | 14:30:05 | Fan #0x13 | Lower Non-recoverable  
> going low
>   5 | 12/16/2010 | 14:30:06 | Fan #0x13 | Lower Critical going low
>   6 | 12/16/2010 | 14:30:06 | Fan #0x13 | Lower Non-critical going low
>
> The theory somehow was that there was concurrent access by the  
> embedded BMC controller and the driver.
>
> After disabling fancontrol the system went from highly unstable to  
> rock solid. I even tried monitoring the system at periodic intervals  
> using the w83795 driver and that resulted in the same types of  
> events, exactly coinciding with the periodic polling.
>
> In other words, I am quite clear that this is related to the driver  
> or at least an interworking between the driver and the embedded BMC.
>
> I have now switched off fancontrol permanently but would like to  
> eventually enable it again because the system is just much more  
> quiet (it is a home server with low load).
>
> Any suggestions? Have similar problems been found?
>
> Cheers
>  Erik
>
>
>
>
>
> _______________________________________________
> lm-sensors mailing list
> lm-sensors@lm-sensors.org
> http://lists.lm-sensors.org/mailman/listinfo/lm-sensors

_______________________________________________
lm-sensors mailing list
lm-sensors@lm-sensors.org
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors

Re: [lm-sensors] System instability with Supermicro X8DTi-F and

[Erik Brakkee]

Philip Pokorny wrote:
> Your motherboard BIOS includes fan speed control and a BMC for sensor 
> monitoring
>
> Use ipmitool and perhaps bmcsensors module to monitor sensors.
>
> Go into the bios and under hardware monitoring arrow down to see fan 
> speeds and then select the ES fan profile (lowest) rather than Off, 
> performance, balanced. That should make the fan speeds much slower and 
> quieter
These are the steps that I already took.  But I would like to use 
fancontrol because then I can make it really quiet most of the time, yet 
keep the server relatively cool when under high load. Are you saying 
that it is not possible to use fancontrol in combination with my 
motherboard?
>
> Phil P.
>
Nonsense and other useful things: http://brakkee.org
MountainHoppers: http://mountainhoppers.nl
Track Detective: http://trackdetective.com
Twitter: http://twitter.com/ErikBrakkee




_______________________________________________
lm-sensors mailing list
lm-sensors@lm-sensors.org
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors

Re: [lm-sensors] System instability with Supermicro X8DTi-F and

[Philip Pokorny]

Your own experience says that fancontrol, lmsensors and a BMC do not  
have a way to coordinate actions which leads to errors and faults.

And as Jean has said, the Winbond chip has a paged register set that  
requires coordination to access.  The BMC code has now way to  
coordinate access with the OS.

Jean mentions thermal trip points. I believe that the winbond chip  
includes it's own autonomous fan speed control seperate from the BMC  
monitoring. What if you just reprogram the fan control curves in the  
Winbond chip once at boot if you don't like the ES BIOS fan control  
curve?

Otherwise, no I don't think what you want is possible.

How hot were the CPU under full load with the ES fan curve that you  
felt they needed to be colder?

Phil P.

-- 
Philip Pokorny, RHCE
Chief Hardware Architect
PENGUIN COMPUTING, Inc
www.penguincomputing.com

On Apr 12, 2011, at 8:33 AM, Erik Brakkee <erik@brakkee.org> wrote:

> Philip Pokorny wrote:
>> Your motherboard BIOS includes fan speed control and a BMC for  
>> sensor monitoring
>>
>> Use ipmitool and perhaps bmcsensors module to monitor sensors.
>>
>> Go into the bios and under hardware monitoring arrow down to see  
>> fan speeds and then select the ES fan profile (lowest) rather than  
>> Off, performance, balanced. That should make the fan speeds much  
>> slower and quieter
> These are the steps that I already took.  But I would like to use  
> fancontrol because then I can make it really quiet most of the time,  
> yet keep the server relatively cool when under high load. Are you  
> saying that it is not possible to use fancontrol in combination with  
> my motherboard?
>>
>> Phil P.
>>
> Nonsense and other useful things: http://brakkee.org
> MountainHoppers: http://mountainhoppers.nl
> Track Detective: http://trackdetective.com
> Twitter: http://twitter.com/ErikBrakkee
>
>
>

_______________________________________________
lm-sensors mailing list
lm-sensors@lm-sensors.org
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors

Re: [lm-sensors] System instability with Supermicro X8DTi-F and

[Erik Brakkee]

Philip Pokorny wrote:
> Your own experience says that fancontrol, lmsensors and a BMC do not 
> have a way to coordinate actions which leads to errors and faults.
>
> And as Jean has said, the Winbond chip has a paged register set that 
> requires coordination to access.  The BMC code has now way to 
> coordinate access with the OS.
>
> Jean mentions thermal trip points. I believe that the winbond chip 
> includes it's own autonomous fan speed control seperate from the BMC 
> monitoring. What if you just reprogram the fan control curves in the 
> Winbond chip once at boot if you don't like the ES BIOS fan control 
> curve?
I am probably not going to do that because it sounds risky.
>
> Otherwise, no I don't think what you want is possible.
So that's clear now. The solution I have now is also acceptable. (Also, 
I did not receive Jean's mail).
>
> How hot were the CPU under full load with the ES fan curve that you 
> felt they needed to be colder?
Under full CPU load for a longer time, it can get up to 85-90 degrees 
Celsius when I use the lowest setting in the motherboard. With maximum 
cooling it gets to 72 degrees in such a case. When I use fancontrol I 
can make it such that it is around 59 degrees while idle and at most 75 
degrees when fully loaded. A temperature of 59 degrees is the 
temperature just after it starts up so the minimum temperature. Right  
now I am just using a setting which is one higher than the lowest in the 
motherboard and that is acceptable, but it will be a bit louder under 
idle conditions. Under full load it will get approximately 80 degrees.
>
>
> Phil P.
>


-- 

Nonsense and other useful things: http://brakkee.org
MountainHoppers: http://mountainhoppers.nl
Track Detective: http://trackdetective.com
Twitter: http://twitter.com/ErikBrakkee




_______________________________________________
lm-sensors mailing list
lm-sensors@lm-sensors.org
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors