Ronald G Minnich wrote:
>
> This causes PIII/800 systems to hang. I'm beginning to think that for
> these more tricky optimizations I'd like to do it when Linux hits runlevel
> 3. This will ensure that we can always get to runlevel 1, so that if there
> is a problem we can recover in linux.
>
> More detail:
>
> For every chipset, I would like the chipset to have 'guaranteed' timing.
> This timing should stay that way until Linux runlevel 1. Then, as part of
> the transition to Runlevel 3, a script or a smart program could set bits
> that improve performance.
>
> There are a lot of advantages to this idea.
>
> 1) Even if you get the performance bits wrong, you can always get the
> machine up to Linux. It's much easier to debug this way.
> 2) You can test new enhancements, and not worry that the machine will
> simply go comatose. We've seen this with regular BIOS, where the only
> option once you set something wrong is to open the case and clear CMOS.
> Imagine doing this to 1024 machines.
>
> Comments?
>
I agreed that the CPU Pipeline changes is problematic and not well tested.
We were comparing the performance of mpeg2dec in LiViD project with Award
BIOS. The CPU pipeline without the CPU pipeline, the performance of LinuxBIOS
is about 30% less than Award BIOS.
Ollie
