On Fri, 27 Jan 2023 18:14:56 +0000 (UTC) Thorsten Glaser <t...@mirbsd.de> wrote:
found 925358 1:7.2+dfsg-1+b2
thanks
Ben Hutchings dixit:
>and it certainly has been buggy on some architectures in the past. It
>seems to be solid now on real hardware.
It probably is. All tests now pass on ARAnyM, even using the mksh
binary built under qemu-user-static (and failing tests there), so
it’s an emulation issue in qemu.
The easiest way to reproduce this locally is (in a sid chroot if
needed, so explicit invocation of qemu, not using binfmt-misc, as
that may use the host’s):
$ wget
https://deb.debian.org/debian-ports/pool-m68k/main/m/mksh/mksh_59c-21_m68k.deb
$ sha256sum mksh_59c-21_m68k.deb
62df852fc4163b8fbda39bd6e06d146f1da18883789077a68ffa6d4b9562651a
mksh_59c-21_m68k.deb
$ paxtar xaf mksh_59c-21_m68k.deb
$ paxtar xaf data.tar.xz
$ chmod +x usr/lib/klibc/bin/mksh # buildd chmod -x’s it if tests fail
$ /usr/bin/qemu-m68k-static usr/lib/klibc/bin/mksh -c 'echo hi'
hi
$ /usr/bin/qemu-m68k-static usr/lib/klibc/bin/mksh -c 'echo hi; /bin/echo hx;
echo hy'
hi
hx
_
(it just sits there, ignores ^C, uses 100% CPU and needs kill -9)
I was able to re-verify this with the packaged version. In the meantime,
the precisely same binary passes every single test I can throw at it
when running under aranym 1.1.0-2 (kernel 5.16.11-1 but I’ve no doubt
upgrading that won’t change a thing).
I tried this with current qemu 8.0 and 8.1-rc4. I can reproduce it with
mksh attached to Message#47, it hangs in both cases. However, the above
with mksh_59c-29_m68k.deb (or s390x.deb) works just fine with both qemu
versions, and works with qemu 7.2 too. I tried mksh_59c-21_m68k.deb as
well (with the same sha256sum), and it works fine here, no hangs. On
bookworm system but with different qemu-user-static packages.
Hmm..
/mjt