Re: An embedded-Linux power-management framework

On Sat, 2002-08-03 at 02:28, Bishop Brock wrote:m

We are proposing a scalable solution that naturally extends the
capabilities of cpufreq. We encapsulate all of the mutually constrained
physical and abstract parameters of a power management policy into
a system-dependent object called an "operating point", and propose
standard ways to manipulate operating points and sets of operating points.
This mechanism will be integrated with the device model in such a way
that as devices come on and off line, the optimum operating point is always
available.

Each operating point being some appropriate categorised and chip
supported combination of power state ? So for something simple like a
mobile K6 we'd have something like

			Full on
			STPCLK and Powernow combinations
			hlt

We also add another dimension not present in the current solution, i.e.,
kernel control of dynamic power management at critical points in the
OS, particularly the entry and exit of the idle (halt) state, consistent

You install your own idle handler to replace the default one - there are
some examples of people doing this already. The amd76x_pm code for
example installs its own idle handler which is needed because both
processors must be idle to drop into most powersaving states.

with policies defined in advance.  As we noted in the original posting,
significant system-wide energy savings are possible by rapidly scaling
frequencies during even short idle periods, and only the OS has the
visibility required to actively manage these events.

Clearly true



-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf