Markus, wait....

there is another possibility.

I already had some e-mail exchange and asynchronuous remote debug
sessions with somebody who had problems with a new kernel (-1.2) and
the Fusion40.

It looks that a (yet unidentified) bug in sys/arch/amiga/amiga_init.c
prevents booting directly via FASTMEM if certain bits of the physical
address of the fastmem are != 0. 

I don't remember the exact bits, but the Fusion Forty has the nice
err... feature to put its CPU FASTROM equivalent MMU tables and ROM
copy at the beginning of the FASTMEM, and make it look to the OS like
the fastmem start ad some odd address (offset a bit over 512 kBytes,
instead of directly at 0x11000000, where the physical memory is
addressed).

Workaround, as I recall, is to either:

a) boot via chipmem, if possible (also serves as a quick check if this
is really the problem), or

b) manipulate your memlist to start at 0x11000000 (forgot what the
tool was that Andreas used), or

c) don't let the Fusion 40 tools integrate the local memory; create a
memory list entry by hand for the memory on the F40 board, and directly call 
loadbsd.

I'm afraid we'll have to add yet another kludge list entry into
loadbsd.

Regards,
	Ignatios Souvatzis