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

[Greg KH <greg@kroah.com>]

On Thu, Nov 11, 2004 at 03:03:33PM +0800, Li Shaohua wrote:
> On Wed, 2004-11-10 at 12:28, Russell King wrote:
> > On Wed, Nov 10, 2004 at 09:45:37AM +0800, Li Shaohua wrote:
> > > On Wed, 2004-11-10 at 09:24, Greg KH wrote:
> > > > Maybe your other patches weren't so bad...  If we implement them, can we
> > > > drop the platform notify stuff?
> > > Currently only ARM use 'platform_notify', and we can easily convert it
> > > to use per-bus 'platform_bind'. One concern of per-bus 'platform_bind'
> > > is we will have many '#ifdef ..' if many platforms implement their
> > > per-bus 'platform_bind'.
> > 
> > Except none of the merged ARM platforms use platform_notify, and I haven't
> > seen any suggestion in the ARM world of why it would be needed.
> Ok, let me summarize it. we now have two options:
> 1. using 'platform_notify'
> platform_notify only has one parameter 'struct device', we must know the
> exact bus type of a device. We can identify the bus type from its name
> (such as 'pci', 'ide'), but it's quite some ugly. Or we can add a 'type'
> flag in the 'struct bus_type' to indicate the exact bus type which Greg
> doesn't like it. One shortcoming is the method hasn't good flexibility,
> we must add a new type whenever a new bus type is added.
> 2. using per-bus type 'platform_bind'
> Every bus type defines a 'platform_bind', so we know the exact bus type
> naturally in platform_bind. The method can't handle special devices,
> such as PCI root bridge, which hasn't a bus type, so no 'platform_bind'
> is invoked for them. we must use some tricky methods to work around.
> Another concern is the chaos if many platforms define 'platform_bind'
> for a bus type, which isn't a big problem currently.
> Greg, it seems you tend to option 2, isn't it?

I don't tend toward option 2, I just don't see much of any workable
option right now :(

thanks,

greg k-h