Re: Cpufreq for opteron

[Andi Kleen]

> --- Begin Message ---
> Dave Jones <davej@xxxxxxxxxx> writes:
>
> > This is likely to be lots of 'fun'. The multiple stage state machine
> > that the opteron powernow uses could be preempted at any stage.
> > Might not be that big a deal for UP (except for any timing specific
> > routines, that need explicit disable/enable around them). But for SMP,
> > where you could wind up on a different CPU when you return to kernel
> > space, 'bad shit' will happen. Good luck!
>
> You just have to set the scheduler affinity mask of the current process
> to a single CPU while executing this - or run it in a workqueue
> which is already bound.
>
> But it doesn't matter right now because the driver doesn't support SMP.
>
> >  > +static int onbattery = 1;	/* Set if running on battery, reset otherwise. */
> >  > +			   /* Of no relevance unless batterypstates <     */
> >  > +			   /* numpstates, as defined in the PSB/PST.      */
> > 
> > Where is this set ? My guess is you're going to need ACPI hooks
> > to do this, in which case it shouldn't be static.
>
> See the comment. For the current code/machines it does not make any difference.
>
> It may be for future chips, but that will need an updated driver anyways.
>
> >  > +	/* WARNING - the cpufreq calls end up doing nothing in a SMP kernel.       */
> >  > +	/* This code will not work too well in such a kernel. This module protects */
> >  > +	/* itself from being compiled ifdef CONFIG_SMP.                            */
> > 
> > Again, why ? Have you actually tried this ?
> > If you have any ideas whats wrong here, we'd like to get this fixed up.
>
> SMP cpufreq will need a lot more work.
>
> Also current Opterons don't support advanced power saving in SMP systems
> (neither does Intel btw)
>
> -Andi
>
> --- End Message ---