Re: [UPDATED v3][PATCH 4/7] regulator: framework core

[Liam Girdwood <lg@opensource.wolfsonmicro.com>]

On Fri, 2008-03-07 at 23:41 -0800, Greg KH wrote:
> On Fri, Mar 07, 2008 at 10:19:51PM +0000, Liam Girdwood wrote:
> > On Fri, 2008-03-07 at 08:10 -0800, Greg KH wrote:
> > > On Thu, Mar 06, 2008 at 06:11:54PM +0000, Liam Girdwood wrote:
> > 
> > > > +/**
> > > > + * struct regulator_cdev
> > > > + *
> > > > + * Voltage / Current regulator class device. One for each regulator.
> > > > + */
> > > > +struct regulator_cdev {
> > > > +	struct regulator_desc *desc;
> > > > +	int use_count;
> > > > +
> > > > +	struct list_head list;
> > > > +	struct list_head consumer_list;
> > > > +	struct blocking_notifier_head notifier;
> > > > +	struct mutex mutex; /* consumer lock */
> > > > +	struct module *owner;
> > > > +	struct class_device cdev;
> > > 
> > > Can you change this to use a "struct device" instead?  We are trying to
> > > get rid of class_device, and there are only 3 users of it in the kernel
> > > today (memorystick, infiniband, and scsi), and I have patches pending to
> > > fix all of these.  For 2.6.26 I would like to be rid of it finally.
> > > 
> > > If you want, I would be glad to fix this up for you, it should be a
> > > pretty simple replacement.
> > > 
> > 
> > Yes please. Much appreciated :)
> 
> Ok, it's below, on top of your patch series.  If you want me to merge it
> in, I can do that as well.  It's a bit big, as I renamed "cdev" to "dev"
> in your structure names and variable names as it isn't a class device
> anymore.

Thanks. I've now committed into Wolfson git. I'll give it a spin on real
hardware early next week and then send to Andrew.

> 
> > > > +	struct regulation_constraints *constraints;
> > > > +	struct regulator_cdev *parent;	/* for tree */
> > > 
> > > And if when you convert, you can get rid of this pointer, it would not
> > > be needed.
> > 
> > Fwiw, the regulator-parent relationship is established by the platform
> > code after the regulator driver has registered all the regulator
> > devices. I assume we can just device_move() to re-parent. 
> 
> Ick, no, just pass the parent into the regulator_register() function.
> Otherwise you will get some wierd uevents spit out (run udevmonitor to
> see what I mean.)  You should set up the parent linkage before you
> register the device with the driver core.

Sorry, I think the 'parent' naming is a little confusing. The device
parent would be set at register time and the 'supply' would be set later
during platform power config.

Liam