Proposal for reorg of kernel directory

[Benjamin Herrenschmidt]

On Tue, 2005-06-21 at 21:25 +0200, Arnd Bergmann wrote:
> On Dinsdag 21 Juni 2005 21:02, Becky Bruce wrote:
> 
> > We've recently begun work on a port of the 64-bit kernel to a
> > Freescale part, and noticed that all the platform-specific code is
> > currently in the kernel directory.  With new 64-bit parts in the
> > works, we expect the number of supported platforms to increase
> > significantly and to include more embedded systems.
> 
> Hmm, at least I'd hope not to need a new platform type for every
> piece of hardware, so there would not be too many of these.

I want to avoid platform numbers explosion too. Paul, David and I have
been discussing that a bit already. The idea would be to make the
functions for accessing the flattened device-tree early during boot
available to platform code during ppc_md.probe(). That way, in most
cases, the need for platform number goes away. There is some work on the
interrupt management and other bits & pieces, but that's mostly only for
pSeries and pmac. The idea is to replace that number with the type of HV
interface if any, and a "major" platform number to differenciate things
like iSeries.

> I would like to see platform types like 'everything with 64 bit
> Freescale CPUs running on SLOF' and maybe another platform type
> for the same CPU with a flat device tree if that differs a lot.

I have no problem have as many xxxxx_setup.c files as there are board
around. We need to give them at least that flexibility :) Provided we
don't have to allocate paltform number codes for every of them. 

Ben.