Re: A20 frequency: allow overclocking/overvolage or not

[Michael <macallan%netbsd.org@localhost>]

Hello,

On Thu, 15 Oct 2015 16:00:48 +0200
Manuel Bouyer <bouyer%antioche.eu.org@localhost> wrote:

> so I have code to allow changing the A20 core frequency (to e.g. lower
> the power input when idle). I used CPU freq/voltage tables from linux.
> It is possible to run the A20 at 1008Mhz (instead of 960) by increasing
> the CPU volatge to 1425mv, which is out of spec (the datasheet doens't
> mention frequency, but the max voltage for CPU core is 1400mv).
> My question is: should I commit the code with 1008Mhz, without it,
> or conditionally allow it (either with sysctl or a compile-time option) ?
> 
> The problem with allowing it unconditionally is that, even through
> the system boots at 960Mhz/1400mv, starting estd will cause the system to
> use the out of spec voltage under load.
> 
> My prefered way of handling this would be to add two sysctls:
> machdep.frequency.min and machdep.frequency.max, user-settable.
> machdep.frequency.available will then only report the entries from
> the CPU freq/voltage table that are between the bounds, and estd will only
> use thoses.

Sounds like the sane thing to do, to me at least.

That said, we should probably think about a sane way to make estd or
sysmon watch the CPU temperature and lower speed when it reaches a
given threshold. I've seen hardware that overheats and crashes even
within specs when running at full speed for prolonged periods of time.
Or as a way to silence it.
I've implemented a hack to do that on some macppc laptops which lack fan
speed control ( so it's either all or nothing. This particular laptop
also has no way to talk to the fan controller at all. ), would be nice
to generalize it.

have fun
Michael