Suggestion for underlying design of power settings infrastructure (was: Re: Power Management and OpenMoko - third power setting)

[Werner Almesberger]

Michael 'Mickey' Lauer wrote:
> Agreed. If we use a clever policy system, there's almost no use for manual 
> intervention for powering on and off subsystems.

By the way, the biggest problem we're currently facing from the kernel
side is that there doesn't seem to be a good way to make sure that
things get suspended/resumes in a healthy order.

The main issue is of course with the PMU, which sits on the I2C bus.
So if we shut down I2C or the PMU before everything else is gone, or
do bring them up after something else wants to access them, bad things
happen.

The "try again later" fixes Andy has made in the last few days all cure
some secondary issues resulting from this overall problem, but we
haven't found a good general solution yet. (Our current priority is to
get suspend/resume in a shape where it will not cause undue problems
during idle current measurements at the factory.)

- Werner