Michael van Elst wrote:
> nbsd@hck.sk (Juraj Hercek) writes:
>
>   
>> There was some interest in having undervolting feature in current 
>> kernel. I did a patch some time ago which was merely a rewrite of est 
>> (http://mail-index.netbsd.org/tech-kern/2006/06/10/0006.html). It wasn't 
>> a way to go because of many changes needed to adjust it when something 
>> changed in original est. New patch is completely different in a such way 
>> that it tries to make *minimal* changes to the original est.
>>     
>
> I also have added undervolting support recently. The code is much
> simpler (and doesn't need unconst :-)) but probably a bit more
> cumbersome to use.
>
> In particular, I only have a single integer knob that sets the
> undervolting for the currently active frequency. Unlike yours
> the knob specifies the voltage drop and not the absolute value.
> I.e. in sysctl.conf I use (Pentium M 780):
>
> machdep.est.frequency.target=2267
> machdep.est.undervolt=192
> machdep.est.frequency.target=1867
> machdep.est.undervolt=192
> machdep.est.frequency.target=1600
> machdep.est.undervolt=224
> machdep.est.frequency.target=1333
> machdep.est.undervolt=224
> machdep.est.frequency.target=1067
> machdep.est.undervolt=256
> machdep.est.frequency.target=800
> machdep.est.undervolt=272
>
>   
Heh, nice Idea, if I would have known someone else is working on 
undervolt patch I probably wouldn't create the new one. On the other 
hand, I've "just" updated my old patch from previous year (taking most 
of the code from it).
> I found that the mprime program is a good indicator for stability. It
> is not in pkgsrc but you can find it here: http://www.mersenne.org/
>
>   
Yes, I've heard about it. However, I wasn't afraid of stability of given 
voltages, just the proper functioning of the patch (using properly 
undervolted voltages for all frequencies) - I've taken the voltage ramp 
from linux where I use undervolting patch for about 2 years (on the same 
notebook).

It would be interesting if undervolting would work work properly with 
NetBSD's XEN - I've adjusted XEN for swithing frequencies/voltages using 
hypervisor call some time ago, but had some troubles with time "running 
too fast" so I gave up. Is anyone working on this?

-- Juraj