Re: Segmentation fault in __uClibc_main on m68k

[Rich Felker <dalias@libc.org>]

On Sat, May 30, 2015 at 05:12:13AM -0500, Rob Landley wrote:
> On Sat, May 30, 2015 at 3:06 AM, Waldemar Brodkorb <wbx@openadk.org> wrote:
> > Hi,
> > Rob Landley wrote,
> >
> >> Of course somebody did a uclibc-ng fork (bought the domain name and
> >> everything), but I talked to him and his reason for doing it is there
> >> are some obscure targets even glibc doesn't support, and I expect that
> >> as musl grows support for those targets his reasons for doing it will
> >> gradually fade away. *shrug* We'll see.
> >
> > We'll see. In the FPGA world there are four main architectures,
> > NIOS2, ARC, Microblaze and Xtensa. Nios2 and Microblaze are
> > supported by GNU libc. ARC and Xtensa are only supported by
> > uClibc/uClibc-ng. There are nice developers and company's behind.
> >
> > For all the no-MMU architectures or systems there is only one option
> > at the moment. uClibc/uClibc-ng!
> 
> Actually I spent this morning walking Rich through setup of his new
> sh2 nommu board this morning.

Is there an easy way to get started testing nommu Microblaze Linux in
qemu or another emulator? Since musl already has a microblaze target,
it should be easy to support.

Based on my introduction to this stuff so far, the big thing I think
needs work in the nommu scene is getting rid of hackish toolchain
cruft. Flat binary format is a hideous hack and the exact same thing
(binaries built using the standard psABI calling/linking conventions,
but without shareable text) can be achieved with plain (non-FDPIC)
ELF, but the kernel refuses to enable binfmt_elf for nommu, despite
the fact that binfmt_elf_fdpic exists and can do the exact work
binfmt_elf should be doing on nommu when it detects that the ELF
headers require a fixed displacement between LOAD segments.

When we get musl working on nommu, I don't want to dredge back up all
the horrible hacks that were done historically in uclinux. Plain ELF
binaries for the same ISA should just work, albeit with fork() failing
with ENOSYS and some combinations of mmap flags failing, etc., and
with no sharing of text. Then we can gradually add FDPIC on an
as-needed basis. I'm working out the proof-of-concept for this
approach on sh2 right now.

Rich