[Libunwind-devel] Re: Backtrace from signal handler on arm from threads

[Henrik Grindal Bakken]

Lassi Tuura <address@hidden> writes:

> Hi,
>
>> Hmm.  Looking at it, it might seem like libpthread is lacking some
>> symbols, creating problems.
>
> Symbols don't affect unwinding, as far as I know.

Yeah, that was another thing I was wondering about...  Can I in some
way use strip or other tools to remove debug symbols while keeping
unwind info alive?  It currently does not work at all if I don't have
my oversized libc and libpthread with full debug info in there.

>> One thing, though: For release software, we'd prefer to ship
>> without debugging symbols.  I can accept that that means we have
>> worse backtraces, but it's not particularly handy that the
>> unwinding segfaults.  Is this hard to avoid?
>
> See my other reply, you are dropping into frame-chain (= fallback)
> walk code. It can be exceedingly hard to get that to work
> reliably. On x86_64 we added this to be very careful when in that
> piece of code:
>
>    /* We could get here because of missing/bad unwind information.
>       Validate all addresses before dereferencing. */
>    c->validate = 1;
>
> ARM code doesn't do that, though I don't know if ARM has validating
> memory access calls in the first place. You might want to compare
> the code a bit.

I don't know either.  Will have to research some.  I'll have a look.

-- 
Henrik Grindal Bakken <address@hidden>
PGP ID: 8D436E52
Fingerprint: 131D 9590 F0CF 47EF 7963  02AF 9236 D25A 8D43 6E52