In article <49e4cb7f1ddave_daniels@argonet.co.uk> you write:
>In article <Pine.NEB.4.21.0007261156420.292-100000@odysseus.mono.org>,
>   David Brownlee <abs@netbsd.org> wrote:
>> Swithing to a more standard bootloader not only gives something
>> 	that can be more easily built and debugged from within NetBSD,
>> 	but also gives the option to boot from other filesystems and
>> 	network devices.
>
>So you would take the standard bootloader, modify it to run
>under RISC OS using RISC OS SWIs to update the page tables and
>so forth?

That's what I'd like to see, yes (though I'm not offering to do it for
arm32).  For arm26, it's not actually necessary to munge the page-tables,
since the kernel runs out of direct-mapped physical memory, which is always
there in SVC mode.

> Do not forget that even when control has transferred
>to the NetBSD kernel, it is still running under RISC OS until
>it takes control of the machine.

What do you mean by "running under RISC OS"?  My current loader kills
interrupts before launching the kernel, and the kernel doesn't fault until
well after it's replaced zero page, so no RISC OS code runs after the first
instruction of the kernel has run.  Is the arm32 kernel different in this
regard?

-- 
Ben Harris