Re: Segfault when watchdog object is deleted (was: Re: [FFADO-devel] Segfault on jackd/ffado shutdown?)

Hi

Thanks for your thoughts.

I've changed the thread subject to more accurately describe what's going on
since the segfault is now known to be unrelated to the streaming system.  It
occurs even when the watchdog object is used without any other ffado
subsystems.

On Tue, Jul 09, 2013 at 03:57:55PM +0200, Adrian Knoth wrote:
> On 07/09/2013 03:26 PM, Jonathan Woithe wrote:
> 
> > It would appear that there's something about the way Posix threads are
> > used by FFADO which is causing trouble either due to recent gcc (eg:
> > 4.7.1) or libc (eg: 2.15).
> 
> Debian is at gcc-4.8.1 and libc 2.17, and we haven't received such bug
> reports, neither in the past nor recently.

Great - that's good to know.  Thanks.

> Which more or less eliminates toolchain issues. Or we patch/configure
> our toolchain in a certain way that prevents the bug from showing up.

Agreed.

> I vaguely remember a discussion on jack-devel and Paul saying "You must
> never ever rely on pthread cancellation, it's broken." Maybe you're just
> seeing this.

Perhaps.  I also recall this discussion, but I don't think the reasons
behind the statement were ever described in detail (at least not at that
time).  I might search the list jack archives and see what turns up.

> Or maybe your toolchain got some weird patching.

The toolchain in Slackware 14.0 is very close to being unpatched.  For gcc
there are two patches: gcc-4.7.1.adalib.pic.diff and
gcc-no_fixincludes.diff, neither of which affect C/C++ code generation as
far as I can tell.  Libc has a number of patches, most of which are clearly
not relevant here.  I might go through these and see if anything crops up.

For reference, I didn't see this segfault under Slackware 13.37 (gcc 4.5.2,
glibc 2.13).

> Last not least: Check the svn for the commit that fixed a similar issue
> we had some months ago. Patch was by David Henningsson from Canonical, I
> guess it was svn r1999.

The symptoms this fixed was quite different and related to the way iterators
were used in an invalid way to delete lists of objects associated with the
streaming system.  The watchdog object is deleted with a singular "delete"
operator.  However, I will take a closer look at the higher level calls
which lead to the segfault just in case this is another obscure
iterator-related problem.

Regards
  jonathan

1373413843.30084_0.ltw:2, <20130709232739.GA21826 at marvin dot atrad dot com dot au>