Re: WhitePapers

[Ignatios Souvatzis <is%jocelyn.rhein.de@localhost>]

On Fri, Jan 29, 1999 at 02:46:20AM -0500, Christos Zoulas wrote:
> On Jan 28,  9:18pm, is%jocelyn.rhein.de@localhost (Ignatios Souvatzis) wrote:
> -- Subject: WhitePapers
> 
> | Here are the whitepapers...
> | 
> | if you want to comment, I'd prefer you to reply to port-amigappc (which is 
> NOT
> | public currently, but I invite you).
> 
> I don't know how the BAT works, but here are some comments on relocation:
> 
> - The kernel can be linked for the virtual address it will be finally using,
>   and loaded at some other physical address. Some careful hand crafted 
> assembly
>   in locore.s can setup the mmu properly. This avoids the need for relocation,
>   but it is not trivial to write the hand crafted assembly.

No real option for ppc, unless I want to rewrite the pmap from scratch. 
Exception handlers are initially executed with mmu off, so some data references
have to be to physical addresses anyway; and worse: as Wolfgang Solfrank told
me, some of the routines are called both from mmu off as from mmu on parts;
so the kernel has to be physically mapped.

> - Write all the code pic? Is that even feasible?

no. Not the data references. Not if you want to use the stock compiler.

> - Use a separate magic program as you mention. Seems to me that it will
>   be difficult to debug...

I did the same for the Amiga bootblock, parsing the a.out relocation table
myself. This time, we could use libbfd or objdump output. Actually, that was
most my question was about... what is easier?

> - have an ld.so in the kernel. But then ld.so will need to relocate itself, so
>   you end up in the second case.

eek. Nonono. Pleasenot.

Thanks anyway.
        Ignatios