Re: [PULL 5/5] linux-user: Fix qemu-arm to run static armhf binaries

[Michael Tokarev]

22.07.2023 00:37, Helge Deller wrote:
..
> > So, this is kinda amusing.
> > This broke arm64, ppc64el and s390x:
> >
> > arm64$ ./qemu-aarch64 /bin/sh -c '/bin/ls -dCFl *[t]* >/dev/null'
> > qemu: uncaught target signal 11 (Segmentation fault) - core dumped
> > Segmentation fault

Note: I run it on native arm64, so it is arm64 on arm64, -
this is a quick test I had from the debian qemu autopkgtest (which
run /bin/ls under qemu and natively and compares the result).  I
haven't tried to reproduce it locally on amd64 host, - I did it on
a debian arm64 porterbox (which I was logged on anyway to debug a
different issue on armel, with qemu git tree already cloned).

> Argh, that's really unfortunate.
> I just tested myself.
> Running static busybox binary did work for me:
> # ./qemu-aarch64 busybox

It didn't trigger with ls, but it did when I run something from
emulated /bin/sh.

This whole email was more like a heads-up/warning, to collect more
details later, - and maybe someone knows what the problem is already
if it is obvious.

..
> Maybe someone else can try? I leave it up to Peter if he wants to revert
> that patch right now, or if it can wait a few days until I'm back?

For the time being, how about a quick-n-hacky band-aid, to include
this fixup only where the original prob actually triggered in the
first place?

Like, if the target is armel?  Something like

#if defined(TARGET_ARM) && !defined(TARGET_AARCH64)

or what's the right preprocessor condition is?

Thanks,

/mjt