Re: [PATCH/RFC 1/4]device core changes

[Greg KH <greg@kroah.com>]

On Tue, Nov 09, 2004 at 11:35:41AM +0800, Li Shaohua wrote:
> On Tue, 2004-11-09 at 08:50, Li Shaohua wrote:
> > On Tue, 2004-11-09 at 06:58, Greg KH wrote:
> > > On Mon, Nov 08, 2004 at 12:11:11PM +0800, Li Shaohua wrote:
> > > > Hi,
> > > > This is the device core change required. Add .platform_bind method for
> > > > bus_type, so platform can do addition things when add a new device. A
> > > > case is ACPI, we want to utilize some ACPI methods for physical devices.
> > > > 1. Why doesn't use 'platform_notify'?
> > > > Current device core has a 'platform_notify' mechanism, but it's not
> > > > sufficient for this. Only sepcific bus type know how to parse dev.bus_id
> > > > and know how to encode specific device's address into ACPI _ADR syntax.
> > > 
> > > I don't see why platform_notify is not sufficient.  This is the exact
> > > reason it was added to the code.
> > As I said in the email, we need know the bus type to decode and encode
> > address. If you use platform_notify, you must do something like this:
> > switch (dev->bus)
> > {
> > case pci_bus_type:
> > bind PCI devices with ACPI devices
> > break;
> > case ide_bus_type:
> > bind IDE devices with ACPI devices
> > break;
> > ....
> > }
> > But note this method requires all bus types are build-in. If a bus type
> > is in a loadable module (such as IDE bus), the method will failed. I
> > searched current tree, only ARM implemented 'platform_notify', but ARM
> > only cares PCI bus, ACPI cares about all bus types.
> > > 
> Oops, it's my bad. we can identify the bus type from bus_type->name, but
> it looks like a little ugly. Why the bus_type hasn't a flag to identify
> which bus it is?

Because if you have a struct bus * you _have_ to know what type it is.

> Anyway, thanks Greg. I will add as you said.

Hm, hopefully Pat will chime in about what would be best for this, as he
created the platform_notify interface.

thanks,

greg k-h