At 9:17 PM +0200 4/23/05, R=E9mi Zara wrote:

>0 > %SRR0 10 - dis
>0034C950: 93C10048
>0034C954: 93E1D94C
>0034C958: 3C608000
>0034C95C: 7C3F0B78
>0034C960: 7CBE2B08
>0034C964: 90010008
>0034C968: 3BF6D700
>0034C96C: 3CA0F008
>0034C970: 38001FFC
>0034C974: 9001080C
>0034C978: 93010030
>
>and the assembly should be (this is in initppc):
>
>   34c950:	93 c1 00 48 	stw     r30,72(r1)
>   34c954:	93 e1 00 4c 	stw     r31,76(r1)
>   34c958:	3c 60 80 00 	lis     r3,-32768
>   34c95c:	7c 3f 0b 78 	mr      r31,r1
>   34c960:	7c be 2b 78 	mr      r30,r5
>   34c964:	90 01 00 08 	stw     r0,8(r1)
>   34c968:	3b 60 00 00 	li      r27,0
>   34c96c:	3c a0 f0 00 	lis     r5,-4096
>   34c970:	38 00 1f fc 	li      r0,8188
>   34c974:	90 01 00 0c 	stw     r0,12(r1)
>   34c978:	93 01 00 30 	stw     r24,48(r1)
>

Just a byte here and there.  A d94c instead of 004c in the second
instruction.  080c instead of 000c.  The 3BF6D700 is way wrong.  Just to
cross-check - can you find the disassembly results within the kernel but in
a different address location?

I haven't tried my bootloader modifications on a 601, so I can't predict
the results, but I use a slightly different approach than the stock NetBSD
bootloader.  Instead of reading from the boot file incrementally as libsa
decompresses the kernel, I load the entire file into memory and then fake a
file system to decompress it.  I also use tftp for both the bootloader and
kernel, instead of NFS.  If you're building MD images, this can help fast
testing of changes.  There isn't a true ethernet driver in the NetBSD
bootloader, so the network driver uses OF routines to do an NFS client.
This could introduce errors.

If you don't mind, try

http://www.dialectronics.com/bootloader/boot25k.xcf

You'll need to move the kernel image to the same directory as the
bootloader within your tftpd setup.  I can't guarantee any different
behavior, and it might be a waste of your time, but it could help narrow
down the problem.

thanks,
tim