Subject: Re: RFC: est.c driver synced with OpenBSD.
To: Simon Burge <simonb@NetBSD.org>
From: Juan RP <juan@xtrarom.org>
List: tech-kern
Date: 08/31/2006 08:46:19
On Tue, 29 Aug 2006 22:03:50 +1000
Simon Burge <simonb@NetBSD.org> wrote:

> and sysctl says:
> 
>   machdep.est.frequency.target = 1000
>   machdep.est.frequency.current = 1000
>   machdep.est.frequency.available = 1833 1000
> 
> It defaulted to 1830 MHz, and I've since changed it 1000 MHz.  The
> lmbench "mhz" program says the frequencies appear to be correct.
 
Good news, Simon made a patch that will create a fake table with all values
(not only highest and lowest) using an algorithm:

http://www.xtrarom.org/~juan/juan-est-plus-fake-table.diff

I tried it on my Pentium M 745 1800MHz and it did show 13 frequencies.
The known table for this CPU only has 6 freqs, so I thought the algorithm
wasn't working correctly, but I think it's working fine because openssl
speed seems to ok.

dmesg with this patch: http://www.xtrarom.org/~juan/new_est-1.dmesg

And tests made with openssl speed:

http://www.xtrarom.org/~juan/new_est_tests.txt

Could you please test this patch on NetBSD if you have a SpeedStep
capable CPU and report results of all frequencies running openssl speed
(only first three values is enough).

Thanks.