Port-m88k archive

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]

Re: NetBSD/luna88k

> Cool!

Thanks, and I'm really glad you are still on port-m88k :-)

> I always wanted to get back to the m88k projects, but sadly
> haven't been able to work on it since that time. I didn't
> realize my m88k/NetBSD website was offline. I'll have to
> check into that when I get a chance.

It would be great if we can get svn repo data on your work.

It looks there are only one major problem around syscall stub in libc
to get single user with /bin/sh, i.e. your NetBSD/m88k work seems
almost complete.

On NetBSD, src/lib/libc/sys/Makefile.inc has ${NOERR} definitions
for syscalls that never fail, and in that case conditional branches
to cerror() are omitted.

On the other hand OpenBSD doesn't have ${NOERR} handling and its
kernel syscall functions unconditionally adjust return addresses
to skip branch instruction after tb0.

NetBSD/m88k kernel uses the sililar systemcall implementation
with OpenBSD/m88k, then all syscalls from libc that never fail
(like getpid(2) and getuid(2) etc.) don't work properly.

It looks this fix solves the following problems you noted in your Trac:
* init(8) exits with code 1
 -> caused by getuid(2) failure in sbin/init/init.c
* debug serial garbage output problem with /bin/sh 
 -> maybe unexpected syscalls (write(2)?) were invoked
    after a wrong return of ${NOERR} syscalls

Another notable fix is snooping settings for m88200:
but I'm not sure this is related to your CPU error exception issue
on youe MVME187.  Maybe several DMA and cache issue remains, but
fortunately luna88k devices don't use DMA.

After one more fix around vunmapbuf(9)
now NetBSD/luna88k almost gets multi user login shell, but
it looks there is something wrong around /usr/bin/login binary.

Note Kenji Aoyama, a maintainer of OpenBSD/luna88k, reported
my NetBSD/luna88k kernel (that worked on nono) didn't boot
on his (real) LUNA-88K2 with three CPUs.

I'm not sure it's K2 specific, multi CPU specific, or differences
between emulators and real hardware.

> Would be nice if it could be brought up to date. I see there's
> this now: https://github.com/redstar/llvm-m88k

Yeah, several people also say that. Maybe we should make NetBSD/m88k
stable first as OpenBSD/m88k, then worth to try next toolchains.

Again thanks,
Izumi Tsutsui

Home | Main Index | Thread Index | Old Index