Re: How does BIOS "thermal throttling" interact with cpufreq? (on VIA C7)

[Wes Felter <wesley@felter.org>]

Phil Endecott wrote:
> Dear Cpufreq experts,
> 
> I have a Jetway J7F2 motherboard with a VIA C7 processor and have 
> recently got CPU frequency scaling working with the ondemand governor.  
> It all works very smoothly - thanks.
> 
> In the BIOS settup screens there are options for "thermal throttling".  
> They allow me to define a threshold voltage and a "duty cycle"; the 
> screens are rather vague and there's nothing about it in the board 
> documentation.  My guess is that when the CPU temperature reaches this 
> threshold the clock speed is reduced.  I have looked at the datasheet 
> for the chip that includes the temperature monitoring logic, and asked 
> about it on the lm-sensors (aka hwmon) mailing list, and there seems to 
> be an "over temp" output signal from this chip.  (The is also an "alarm" 
> output, with a separate threshold register, which I think goes to the 
> speaker.)
> 
> Does anyone know how this stuff works, and how it interacts with 
> cpufreq?  Could it be that the "over temperature" signal is actually 
> just an interrupt and software is supposed to respond by changing the 
> frequency?  Does cpufreq need to be aware of this throttling function 
> (i.e. having enabled cpufreq, have I unintentionally disabled this 
> thermal protection?)

If they've done it really right (like Intel TM1/TM2), then the CPU will 
handle everything transparently for you. If they've done it right, then 
overtemp will trigger a system management interrupt (SMI) and BIOS code 
will set the IA32_CLOCK_MODULATION register; because clock modulation is 
(supposed to be) orthogonal to EIST it should work fine. If they do it 
some other way then who knows.

Wes Felter - wesley@felter.org