Re: Kernel 2.2.14

[Benjamin Herrenschmidt <bh40@calva.net>]

On Thu, Jan 13, 2000, Adam C Powell IV <hazelsct@mit.edu> wrote:

>A "Me too": on StarMax (pmac 4400 clone, 603e-160), 2.2.14 doesn't even start
>atyfb (onboard Mach64) or offb (unsupported Cirrus Logic MacPicasso).  Quik
>(2.0-2) segfaults on me, corrupting my boot blocks, with or without 2.2.14 in
>quik.conf, but booting vmlinux.coff on the floppy crashes too with the OF
>error
>message below.  2.2.15pre1 from floppy fails in exactly the same way.
>

This is a problem I fixed yesterday evening ;) The fix will be in my next
kernel (will be rev.12). I'll upload it tonight or tomorrow.

Basically, what happens is that the cleanup of the MMU on kernel entry I
introduced recently to be more reliable with yaboot broke the "oldworld"
OF booters. Those booters used to play with BAT mappings on the behalf of
Open Firmware, causing OF to return bogus physical address to my code.

You can manually add a temporary fix until my fix is merged:

Edit arch/ppc/kernel/prom.c, in function prom_entry(), at the end, add:

/* If we are already running at 0xc0000000, assume the bootloader has
 * setup a BAT for us, causing OF returned phys to be bogus
 */ 
if (reloc_offset() == 0)
    phys = 0;