Power parents

Power parents

[Alan Stern]

Ben:

Pavel Machek suggested you were a good person to ask this question.

I see that the power tree agrees pretty much with the device tree, but
there is the possibility of having a different parent pointer.  However
the device_pm_set_parent() routine isn't called anywhere in the kernel.  
Does that mean it can be eliminated, making the two trees identical?

Alan Stern

Re: Power parents

[Benjamin Herrenschmidt]

On Fri, 2004-10-08 at 04:46, Alan Stern wrote:
> Ben:
> 
> Pavel Machek suggested you were a good person to ask this question.
> 
> I see that the power tree agrees pretty much with the device tree, but
> there is the possibility of having a different parent pointer.  However
> the device_pm_set_parent() routine isn't called anywhere in the kernel.  
> Does that mean it can be eliminated, making the two trees identical?

Currently the trees are identical yes. I may still want to "insert"
special nodes in the Power tree though...

Ben.

Re: Power parents

[Alan Stern]

On Fri, 8 Oct 2004, Benjamin Herrenschmidt wrote:

> > I see that the power tree agrees pretty much with the device tree, but
> > there is the possibility of having a different parent pointer.  However
> > the device_pm_set_parent() routine isn't called anywhere in the kernel.  
> > Does that mean it can be eliminated, making the two trees identical?
> 
> Currently the trees are identical yes. I may still want to "insert"
> special nodes in the Power tree though...

My reason for asking is because having special parent pointers like that,
without corresponding children pointers, makes it impossible to guarantee
that parents aren't suspended until after their children when doing a
selective suspend.

What sort of special nodes are you thinking of inserting, and how would 
they affect selective suspend?

Alan Stern

Re: Power parents

[Benjamin Herrenschmidt]

On Sat, 2004-10-09 at 00:07, Alan Stern wrote:

> My reason for asking is because having special parent pointers like that,
> without corresponding children pointers, makes it impossible to guarantee
> that parents aren't suspended until after their children when doing a
> selective suspend.

Oh, I also remember at one point toying with moving some devices to
different parents based on weird mobo design, but I dropped the idea.

> What sort of special nodes are you thinking of inserting, and how would 
> they affect selective suspend?

Oh, I'm thinking about inserting a "platform" entry between some USB
controllers and the PCI parent. On PowerMac, some of these USB controllers
are power managed via special stuffs in Apple ASICs that aren't directly
related to the PCI bus they sit on. Right now, I have hacks in the USB
HCD drivers themselves for that, but it would be cleaner just to "insert"
an entry in the power tree to do the job.

Ben.