AW: Re: __setup_cpu_be problem

[Benjamin Herrenschmidt]

On Mon, 2006-02-13 at 23:17 +0100, Arnd Bergmann wrote:
> On Monday 13 February 2006 19:19, Hartmut Penner wrote:
> >       the initial value set by FW of the HID6 is 0x00010034_00000000.
> 
> Ok, good: Both large page sizes are set to 16M, neither is 1M or 64k.

That should be changed. One should be set to 64K and the other to 16M.
At this point, it's not yet clear how the kernel will make use of 64K
pages, it requires a feature bit that is never set (indicating that
cache inhibited L pages are supported). It will be provided by
ibm,pa-feature property in the long run but last I looked, it wasn't yet
implemented by any firmware.

> That means that we can just rip out the HID6 setup from the kernel
> without losing the ability for 16M pages. Geoff, please submit a
> patch to replace __setup_cpu_be with __setup_cpu_power4 if that
> solves your problem.
>
> The new 64k page support has never worked so far on Cell because
> of missing spufs code for this, so we don't get a regression either
> way. We still need the firmware changes (HID6 setup and the device
> tree properties) in order to support 64k pages, but we don't need
> to worry about breaking stuff in the process.

Ben.