Re: Bug#67099: segfault on m68k
On Jul 14, Bjoern Brill wrote:
> I suspect the following reason: on this m68k box, most programs linked
> against libpthread.so segfault without visible reason, usually a short
> time after fork()ing (for example, nscd silently dies because of this, as
> does ypbind). Now /usr/bin/python IS linked against libpthread and we have
> definitely fork()ed since we get a SIGCHLD.
Hmm, perhaps Python on m68k shouldn't be linked against pthread, at
least on potato. I guess the longterm solution is for someone to look
at what's wrong with (I assume) the handcoded assembly in
libpthread. But there isn't that much there...
OTOH it could be nastiness with cas (compare and set) and its buddies
on certain systems, in which case it's a hardware limitation. Maybe
disabling compare_and_set() in libpthread might help, though I dunno.
In any case, I know it works on 2 68060 Amiga systems (kullervo and
CTS's); maybe we can narrow it down to a particular 0x0 and/or
hardware platform (Mac, Amiga, Atari) and go from there.