Re: m68k linuxthreads broken?
> Some experiments with modified recompiles and strace show that fork()
> puts these processes in a broken state where lots of C library calls,
> including syslog(), make them dump core. It' really fork(). If I use a
> flag like --debug so they don't have to detach from the console they don't
> fork and they work.
[...]
> Now: is this problem known? Could somebody try to reproduce it? I want to
> check whether it's just my obscure box before I file a bug report. Results
> with non-Mac m68ks would be interesting as well since the Mac kernels are
> somewhat special and it could, in theory, be a kernel problem.
Two remarks: It is entirely possible that linuxthreads is broken on m68k.
I doubt fork() in general is broken since that's used quite often in other
code and would have been noticed before. Meaning the plain libc fork(), I
have no idea what linuxthreads does different with forks.
Second, it would help a lot if you could write a small demo program that
shows how a simple fork() and some action on behalf of the forked process
makes it dump core. Testing this on a range of machines with different
processor, speed, kernel and libc versions should help to narrow down the
problem.
I doubt there's something special to the Mac kernel causing this. I can
imagine something particular with 040 processors and shared libraries,
rather.
Michael
Reply to: