powernowd vs. ondemand governor

[Matt Zimmerman]

On Mon, Feb 16, 2009 at 01:20:21PM -0800, Steve Langasek wrote:
> On Mon, Feb 16, 2009 at 02:58:01PM +0000, Matt Zimmerman wrote:
> > We currently ship powernowd in the desktop seed, its main purpose being to
> > load cpufreq modules set up the kernel's CPU frequency scaling governor.  If
> > (and only if) that fails, it starts up powernowd instead.  I'm not sure on
> > which platforms that's still needed, if any.
> 
> > Is this still an appropriate default?  Aren't the necessary modules loaded
> > automatically now?  Would it be a better idea to set the default to ondemand
> > in the kernel and drop powernowd altogether?
> 
> None of the cpufreq modules (under
> /lib/modules/$kvers/kernel/arch/x86/kernel/cpu/cpufreq/) have any module
> aliases, so I don't think they get loaded except via the powernowd init
> script.

I'll give it a shot and see.

> Furthermore, my understanding is that even if udev could autoload the
> modules, there's no way to set a policy for which governor to use without
> running a script exactly like the current powernowd init script.

Yes, we might need to provide some method to override it, but
CONFIG_CPU_FREQ_DEFAULT_GOV_ONDEMAND would take care of the default.

> As for platforms where it's needed, I have one system (an older AMD64) that
> doesn't work with any of the platform cpufreq modules, so powernowd is the
> only option there.

That's useful information, thanks.

Is the fact that powernowd is now unmaintained upstream cause for further
consideration?

-- 
 - mdz