> And, well just a thought... TFTP booting of Linux/MIPS kernel is nice
> but what about the initialization of the bootinfo struct (and the
> soon-coming NASP) made by Milo on ARC machines (perhaps soon a
> DEC-Milo?) ? I mean, we have to compile a kernel for "this" machine
> (with proper hard-coded infos) in order to boot it by TFTP, or can you
> think of any other means ?
I have already built another means: the kernel entry point for
DECStation kernels is the dec_entry routine in the decstation.S file,
rather than the usual kernel_entry. This routine then queries the
boot prom for info about the machine, fills in the bootinfo structure,
and then moves the exception handlers down in memory prior to jumping
off to the standard kernel entry point.
In fact, I think a similar method is also used by the SPARC team...
where they query the Sun boot proms. It's only on really brain-dead
PC architecture clones that you must query the bus... :-)
The dec_entry routine is very simple at the moment, but due to DEC's
reasonably structured TurboChannel architecture, it will be possible
to discover all of the devices either on the motherboard or plugged in
to TurboChannel slots. (See also the MACH sources for clues as to how
this is done. ICK!) The bootinfo structure on the DEC will be a
little more complicated because of this...
All of the TurboChannel goo can get quite complicated, and would
probably be too large to fit adequately into milo given it's need to
be no bigger than a boot block... this way milo for the DEC (dilo? :-)
can be small and neat, and it need only be able to find the kernel
image on a DEC hard drive, and even then it can use the boot prom to
deal with the hardware.
Paul M. Antoine, Net: email@example.com
Softway Pty Ltd WWW: www.softway.com.au
PO Box 305, Strawberry Hills, NSW 2012, Australia Tel: +61 2 698 2322
Level 2, 79 Myrtle St, Chippendale, NSW 2008, Australia Fax: +61 2 699 9174
"It is the lack of acceptance of diversity which threatens to
destroy society, NOT the free expression of it." - Me.