> Hmmm... yes I do. There's really no way to avoid it at the moment, and
> maybe not at all. When we have hd/fd working and finish writing boot
> loader, then the kernel can be in ELF and just the boot loader will need
> to be in COFF. However, anyone who wants to be able to remote-boot a
> DECstation is always going to need a COFF kernel image capability, unless
> we re-write the boot prom!
Is there a reason why you don't use a multistage bootstrap loader that
allows you to use a small COFF bootloader which then loads the real
kernel image in whatever format you wish to use?
> > I'll probably switch to a ELF -> COFF converter for making the Milo
> > bootfile, so for me COFF in the binutils might become entrily super-
> > fluous. If you however need it - it's just a fingersnap to add it
> > back in.
Ok. During the weekend I've found some bugs in the a.out support of
binutils 2.7. I just had enough luck so far, that's why I didn't notice
this problem before. Well, now I finally give a.out what it deserves -
a coffein. I've written a converter that converts a ELF kernel image
into a bootable a.out image for Milo. This works reliable and requires
no changes to Milo because the format of the ARC kernel images stays
unchanged. A more sophisticated format including compression is on the
to-do list, also.
Crossformat linking of ELF binaries to a.out or COFF should work also in
theory for this task but this feature is still that buggy in the binutils
("born to debugged") that I didn't even attempt to debug it. In short
this all means about binutils: For best results use
- binutils 2.6 for a.out
- binutils 2.7 for ELF
(The others probably should know this, too, so I cc this to the list.)
> Thanks for your continuing good work on Linux/MIPS!