This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: libjava testsuite hanging on OpenBSD
- From: Loren James Rittle <rittle at latour dot rsch dot comm dot mot dot com>
- To: espie at nerim dot net
- Cc: gcc at gcc dot gnu dot org
- Date: Thu, 30 May 2002 13:34:50 -0500 (CDT)
- Subject: Re: libjava testsuite hanging on OpenBSD
- References: <200205300045.g4U0jbOV079105@latour.rsch.comm.mot.com> <200205300046.g4U0kdli079116@latour.rsch.comm.mot.com> <20020530112635.A11096@tetto.liafa.jussieu.fr>
- Reply-to: rittle at labs dot mot dot com
In article <20020530112635.A11096@tetto.liafa.jussieu.fr>,
Marc Espie<espie@nerim.net> writes:
> Hum, and gctest works when linked statically...
I'm glad you tested this as well since this likely means you were
dealt a fairly easy problem to fix (since you are using ELF). ;-)
In conjunction with looking over the boehm-gc OpenBSD configuration,
this probably means that the GC is not being informed of the complete
map of shared library segments.
I suggest that you need to start by (a) setting up to use the same
code path in boehm-gc/dyn_load.c as used by FREEBSD/ELF; (b) add
DYNAMIC_LOADING to the OpenBSD configuration(s) in
boehm-gc/include/private/gcconfig.h.
I would also suggest a full audit of OPENBSD guarded stuff in
boehm-gc/include/private/gcconfig.h against FREEBSD/NETBSD guarded
stuff. I will note a special issue that hit FreeBSD (first observed
on alpha but also rarely seen for i386 due to GNU ld linker script and
the FreeBSD ELF loader): There may be an unmapped memory hole between
etext and edata. The GC doesn't like such holes unless it is informed
about them.
Regards,
Loren