Subject: Re: bootxx proto don't know ELF...
To: Allen Wittenauer <email@example.com>
From: Mark Abene <firstname.lastname@example.org>
Date: 03/15/1999 07:36:02
On Sat, Mar 13, 1999 at 08:52:48AM -0800, Allen Wittenauer wrote:
> pk is still working on that part of the build process yet. When the ELF
> snapshot came out, Paul mentioned that you still couldn't use it to build a
> kernel yet (with the reason why being left as an exercise for the reader, as
> Casper Dik likes to say)...and now you found the reason why. =) [Other than
> that, the ELF system appears to be working perfectly. Granted, my ELF machine
> sans kernel/booter is _completely_ ELF w/out any a.out emulation running.]
> To build a kernel, you'll still need an a.out machine. I haven't tried setting
> the OBJECT_FMT flags, though. So that might work.
> For your Voyager, you probably better stick with a.out on it and wait for the
> bootxx issues to be dealt with.
Got it. Actually, what I ended up doing was a fresh install of 19990228.ELF,
and then untarring the base.tgz and comp.tgz from 19990103 into /emul/aout.
Then sticking "/emul/aout/usr/bin" at the front of my PATH was sufficient
to build the NetBSD-current kernel using the a.out toolchain in /emul/aout.
Oh, I also had to .include the /sys/compat/aout "files" file in sparc/conf's
"files" file, and add an underscore in front of the ENTRY-point symbol name
in the am7930 and fd related assembly files. Not to mention a "COMPAT_AOUT"
in my conf file.
This was sufficient to allow me to build a working -current kernel in a.out
format which boots fine, and give me the opportunity to build the -current
userland with 19990228.ELF's ELF toolchain.
While everything is now peachy-keen, my current problem with -current is that
the machine panics on boot when pcmcia0 attaches to nell0, with a "data fault"
exception. Not a very descriptive panic. If I disable nell in my conf file,
all is well. Unfortunately, at least in 19990103, nell autoconfig'ed fine
and would detect card insertion/removal without problem, even if the debug
output seemed a little confused about the contents of registers.
Now that there are some actual drivers for pcmcia cards, I was sad to see
that the stp4020 driver now seems "broke" in -current.