Re: [Valgrind-users] Strange errors in backtrace

[Josef W. <Jos...@gm...>]

On Thursday 24 July 2003 07:05, Denis Perchine wrote:
> Hello,
>
> > > I have strange errors in backtrace family functions.
> > > The trace attached is with backtrace.c and backtracesyms.c from
> > > glibc-2.3.2 compiled with debug and directly linked to see line number.
> > >
> > > Interesting thing is that I can not reproduce this on the test case. It
> > > happens only in big program. Any ideas why this could happend?
> >
> > I don't know about this implementation of backtrace(), but such
> > functions often rummage around in places a portable correct program
> > never would.  That doesn't mean there's necessarily a problem in this
> > code, but it's worth inspecting (and adding Valgrind macros to clarify
> > anything correct-but-odd to V itself).
> >
> > I'm guessing here, but it may only show up on large code because of the
> > depth of its stack use.
>
> It looks like this. Here it is debug output. It walks the stack until it is
> in [esp;__libc_stack_end] boundaries. And here is the strange part.
> 1. Addresses in backtrace different by one with valgrind trace.
> 2. valgrind starts to issue errors just after backtrace walked over
> the stack shown by valgrind. Might be this could be fixed by fixing a value
> of __libc_stack_end by valgrind?

Hi,

I don't know about the rest, but a stack-trace going deeper than the stack 
shown by valgrind simply means memory access to addresses V can't check 
because it memory was written to before V switched to its simulated CPU.

Josef