Re: [Libunwind-devel] Testing time

[Lassi Tuura]

Hey,

> I have some reports of crashes on RHEL6 + GCC 4.6 + gold linker. LTO may be 
> involved, I am not sure.

So far this looks like an issue in using gcc/gold/lto. Some more info at:

 http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48384
 http://sourceware.org/bugzilla/show_bug.cgi?id=12629

Unfortunately the real details aren't on those reports, but the problem is that 
unwind info FDE address ranges are inside .eh_frame, not in .text. Otherwise it 
looks to correctly describe functions, so it seems somehow unwind info is 
generated with a wrong base address. We don't yet know what causes this, but it 
doesn't look like libunwind issue. For all we know we could have even built the 
tool chain incorrectly.

Regards,
Lassi