Re: [PATCH v2] Add suspend/resume for HPET

[David Brownell <david-b@pacbell.net>]

On Monday 02 April 2007 1:04 pm, Alan Stern wrote:
> On Mon, 2 Apr 2007, David Brownell wrote:

> > This is the kind of thing that the pm_parent relationship was (AFAICT)
> > originally supposed to handle.  Of course, it doesn't/can't, given the
> > current implementation ... that relationship is never used.
> 
> Just so.  In fact, there almost certainly are other dependencies that 
> nobody is aware of, simply because they have never had a chance to bite.

In any given system, yes there are bugs lurking.  But I was more concerned
with a provably wrong assumption made by the current framework.  Such things
cause cascading fragility.

As Thomas mentioned, HPET isn't the only place where a "linear" model fails.


> Such things can be rather difficult to pin down when they occur.  I would
> be happy enough to leave matters as they are, with a strict LIFO approach.

I wouldn't.  Much better to have a solid handle on the interdependencies
than to need to cope, long term, with a framework that doesn't allow that.

Remember also that a LIFO model assumes that there's only one sequence by
which the hardware powers up/down ... i.e. that there's no runtime PM going
on, whereby large chunks are regularly powered down/up based on usage.
Better runtime PM becomes more important as system complexity rises.

- Dave