monitoring hardware temperatures

[Andriy Gapon]

on 06/12/2010 09:30 Mikhail T. said the following:
> Hello!
> 
> I have a server (Dell Poweredge 2900), that's loaded with sensors.
> 
> While it was in Windows-mode, a utility was able to tell me not only the
> temperature of each CPU-core, but also that of every DIMM!.. One of them was
> running far hotter than others, and I'd like to continue keeping an eye on it
> now that the box run FreeBSD.
> 
> In FreeBSD there is coretemp(4), which is nice, but nothing else... There is no
> hw.acpi.thermal hierarchy either on this box... Yet, the box has 6 fans, two
> power-supplies, plus DIMMs -- all of them with sensors, that I can't read...
> 
> It seems, in 2007, there was an attempt to introduce OpenBSD's sensor-framework:
> 
>    http://kerneltrap.org/OpenBSD/BSDCan_2008_Hardware_Sensors_Framework
> 
> but it was backed-out after being declared a "pile of crap" and "festering
> junkpile" by our most mirthful contributor:
> 
>   
> http://docs.freebsd.org/cgi/getmsg.cgi?fetch=193129+0+archive/2007/cvs-all/20071021.cvs-all
> 
> 
> "until a proper architectural solution has been found". Has that happened in the
> three years, that passed since that lovely discussion? Or are we still waiting
> for someone to design and implement it not merely "adequately", but "perfectly"?

My opinion: if that effort was not called "sensors framework", but some name
less ambitious like "basic support for a few sensors on some desktop systems",
then it would get far less attention and criticism :-)
But I don't want to dig into this matter any deeper :-)

> If the three other BSD-cousins have had this for a while (NetBSD -- for 10
> years, apparently), continuing to insist on some future perfection seems wrong
> -- we should have this "adequate but imperfect" method if only for cross-BSD
> compatibility.
> 
> Is there, perhaps, a set of patches still secretly maintained by some die-hard?

Not being a die-hard I do keep, not so secretly, that code in my tree.

> I'd love to try it here, and will be very thankful, if it gives me the
> monitoring, that I can not obtain otherwise... Thanks! Yours,

Well, that code has support only for a few types of hardware monitoring chips
(Super I/Os with hardware monitoring function).
So, it greatly depends on exact kind of hardware and sensors that you have.
First thing you should do to is to discover what kind of hardware is used for
monitoring in your server.
In your case that data might be provided via IPMI.

Especially I am not sure about monitoring DIMM temperature - greatly depends on
the way that it is actually done.  Perhaps it's reported via SMBus by the DIMMs
themselves, not sure...

-- 
Andriy Gapon