Re: [Libunwind-devel] FreeBSD port

[Konstantin Belousov]

On Sun, Apr 11, 2010 at 12:36:11PM -0700, Arun Sharma wrote:
> On Sun, Apr 11, 2010 at 12:16 PM, Konstantin Belousov
> <address@hidden>wrote:
> 
> > Are the results of running make check for Linux x86 and x86_64 available
> > somewhere ? Are all tests passed ? I am asking because I am investigating
> > the bug where dwarf unwinder on x86 computes %ebp location for the
> > main() procedure bogusly, pointing to it %esi, that is wrong and causes
> > unw_step() to fail (on x86). I assume that this is a bug of my port,
> > but still worth checking.
> >
> 
> I see 4 failures on x86_64 and 6 on x86. Most of it due to unimplemented
> functionality.
> 
> One possible reason for the failures may be due to the fact that dwarf and
> libunwind use different register numbering. The map is given by:
> 
> x86/Gglobal.c:dwarf_to_unw_regnum_map

Upon further investigation, it looks like a toolchain issue.
I used a custom-built gcc 4.4.3/binutils 2.20.1 for the development.
The 'make check' gives me 9 failures both on x86 and x86_64.

After switching to the system compiler, 4.2.1 + patches, and system
binutils, 2.15 + patches, I am down to 5 failures on x86 and 6 on x86_64.
In particular, the issue with unwinding the main() frame for Gtest-bt
disappeared.

pgpr2Ir7c8WwT.pgp
Description: PGP signature