Re: CPUFreq ability to overclock

Re: CPUFreq ability to overclock

[Jeremy Fitzhardinge]

Eduard-Gabriel Munteanu wrote:
>  I've seen a comment about scaling voltages in cpufreq.c, but it seems 
> there is no actual support for that. 

Cpufreq operates in terms of "operating points" which are 
voltage/frequency pairs, rather than just voltages.  All the CPU drivers 
which support it have a voltage corresponding to each frequency to make 
an operating point.

    J

CPUFreq ability to overclock

[Eduard-Gabriel Munteanu]

*This message was transferred with a trial version of CommuniGate(r) Pro*
I'm thinking to implement a way to have CPUFreq's maximum FSB increased 
to allow overclocking. Having that done, it should make sense to have a 
governor (userspace maybe, with a 3rd-party tool) thermal-throttle the 
CPU. There is one slight problem: can we control Vcore from within the 
kernel, on-the-fly, without rebooting? I've seen a comment about scaling 
voltages in cpufreq.c, but it seems there is no actual support for that.

My idea is to make cpufreq_policy.max (same for cpufreq_cpuinfo's 
member) a soft limit (to prevent unwanted overclocking from usual 
governors) and have another member hold the hard limit. Chipset drivers 
should calculate the hard limit based on installed processor or 
boot-time FSB.

I'm looking forward to comments and suggestions.