At 9:35 AM +0200 4/23/05, R=E9mi Zara wrote:

>So I would also say corrupted kernel.

Not necessarily.  There are several places in the kernel that contain
Altivec code by default, but have tests to see if the CPU can use Altivec
before it is executed.  This is why it is important to locate the code that
is supposed to be executing in the kernel object file and determine if it
does match.

>If the bootloader jumps into the kernel with the compressed one, I
>suppose it should also jump into it with the uncompressed one.

If you're loading via enet, there is a 4M limit for file transfers.

>With an uncompressed kernel, the boot hangs before the copyright
>notice. In the install doc, it is sais this may be due to a "real-base"
>problem.
>My settings are in line with the doc. Why would a bad real-base provoke
>a hang ?

Because the loading of the kernel into memory will overwrite Open Firmware.
Real-base is the location of Open Firmware.  On an OF 1.0.5 system, if
real-base is set to -1, Open Firmware will load at 0x400000.

>I see in your code that you use "ofw_dbg". Do you know if something
>equivalent is available for NetBSD ?

It is basically a front end for writing to the OF output-device.  I had to
add it to figure out how far the kernel was getting.  You'll want to review
the code in powerpc/powerpc/openfirm.c, in particular OF_write.  You can
find the ofw_dbg methods I wrote at

http://dialectronics.com/OldWorldMacs/code/machdep.c

and the printf wrapper at

http://dialectronics.com/OldWorldMacs/code/ofw_printsub.c

>Is there a quick way to drop back into openfirmware, by provoking an
>exception for instance, so that I can see how far it goes into the
>kernel ?

OF_enter() will drop into OF and "go" may allow you to resume.

tim