io_block_mapping in PPC64?

[Benjamin Herrenschmidt]

On Mon, 2005-05-16 at 10:53 +0300, Tzachi Perelstein wrote:

> [Tzachi]: I do not use device-tree. Our Linux is loaded by U-Boot which
> first time supports the IBM-970FX and first time executed in 64-bit
> mode. There is no real need for such complex structures when loaded by
> U-Boot. If you want ppc64 architecture to be spread to embedded systems
> too you should avoid such restrictions. 

No, this will not happen. Look at ppc32. Every single board vendor ended
up hacking it's own board_info structure, which quickly became a total
maintainance nightmare.

It is very easy and not complex at all to provide at least a simple
device-tree to the kernel. What we might do to "help" here may be to
provide some code for things like uboot to generate it based on a simple
ASCII definition or something like that. But there is simply no way I
will let arbitrary structures be invented for every new board out there
and be passed around, turning the ppc64 kernel into the same kind of
unmanageable mess that ppc32 is nowadays.

There are also open implementations of Open Firmware floating around
(but I agree a solution that can be fitted in uboot would be useful).

I'm pretty confident that Paul Mackerras (ppc64 architecture maintainer)
agrees with me here.

Ben.