tech-kern archive

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]

Re: Power Management Framework architectural design

Joerg Sonnenberger wrote:

> On Thu, Apr 10, 2008 at 10:02:49AM +0000, Juha Keski-Saari wrote:
> > Wouldn't this remove the particular problem in the case of OMAP2 and
> > enable platform dependent logic, unless there is a platform with no
> > clock control and driven by power domain states... but even then
> > this idea could be functional by motivating the power domain
> > transitions through its member states.
> Let me repeat the question. The problem is that on most architectures
> we have very limited information about power domains, clock domains
> are even less well supported. For PCI centric systems you can
> normally only approximate this by considering each (bus,device) pair
> a single clock domain and (bus,device,function) a power domain. This
> doesn't always apply, when you add SuperIO chipsets, but it is a good
> first case approximation. ACPI exports the power domain data in some
> cases, but it is far from being complete. It also provides limited
> mechanisms to power down a power domain. For PCI again bus specific
> options are often possible. So the fine grained control in the
> embedded space is very specific and therefore I am not sure how to
> properly abstract it so that both cases can be fulfilled.
> Joerg

Thank you for the clarification, with that information I can refine my
suggestion. Since the power and clock domains concept is not a global
one, probably we cannot impose change upon the PMF on its account. How
do you feel about optional agents for power and clock domains, so each
environment can choose whether they want to include the relevant
functionality, if it is valid in the context of that environment. Then
by that configuration the device instance would be extended with the
interface to the agents that are configured in. On a platform that does
not want to know about power and clock domains, the drivers can be
oblivious and PMF unchanged; On one that does want to use these power
saving options, additional functionality is available. The role of
power domain as well as clock domain management could remain as
mentioned before in the platforms that support their functionality.
These agents could be loosely coupled, with just an interface to the
drivers. One risk I can see is division of information without a
central component, but I'm not sure if it is critical here.

So in essence I agree, the PMF should not be extended to cater for
these additional features, and would like to hear if you concur with
the proposed way the extension would be done outside PMF.

Juha K-S

Home | Main Index | Thread Index | Old Index