Crazy clock

Atte Peltomaki koston at iki.fi
Tue Dec 14 23:13:51 PST 2004


> :> :The problem is that the BIOS might mess with the timer directly like my
> :> :notebook does. We can't detect changes from System Management Mode.
> :> 
> :>     Ugh.  You're right.  SMM bypasses everything.  Somebody needs to shoot
> :>     Intel.
> :
> :Apart from the shooting bit, is there something I could do to debug the
> :situation? 
> :
> :-Atte
> 
>     No.  I think if that user-timer-1 option works you should just stick with
>     it.
> 
>     Obviously I should add support for other timebases, despite the fact that
>     most of them are broken.  I think FreeBSD did get that one right though
>     their support infrastructure is ridiculously complex for the purpose.
>     e.g. the ACPI timer (which is an unlatched ripple counter on many chipsets
>     and thus cannot simply be read once to get a stable value), and the LAPIC
>     timer.

After some time of using the box, it seems that the user-timer-1 kernel
option doesn't completely fix the problem. Clock drifting is
significantly reduced, now the clock wonders perhaps an hour per day.
Used to go about twice the speed. 


-Atte





More information about the Users mailing list