Subject: Re: [boost] Hangs with Boost 1.56 on clang armhf
From: Niall Douglas (s_sourceforge_at_[hidden])
Date: 2014-08-02 22:13:13
On 30 Jul 2014 at 16:47, Niall Douglas wrote:
> I'm seeing consistent hangs of AFIO on Boost 1.56 on Ubuntu 14.04 LTS
> for clang armhf. GCC armhf works fine, so this is a clang armhf
> related issue.
> [snip]
> Same problem - tests hang. I tried running a test manually and I get
> this:
>
> terminate called after throwing an instance of
> 'boost::thread_interrupted'
> terminate called recursively
> terminate called recursively
> terminate called recursively
> terminate called recursively
> terminate called recursively
> terminate called recursively
> terminate called recursively
> terminate called recursively
> terminate called recursively
> terminate called recursively
> terminate called recursively
> terminate called recursively
> terminate called recursively
> terminate called recursively
> terminate called recursively
> terminate called recursively
> terminate called recursively
> terminate called recursively
> terminate called recursively
> terminate called recursively
>
> Remember, this is identical code to any other Linux target, plus this
> works a-ok for GCC armhf.
>
> A quick backtrace with gdb reveals this:
>
> terminate called after throwing an instance of
> 'boost::thread_interrupted'
> [Thread 0xb5b76450 (LWP 12610) exited]
>
> Program received signal SIGABRT, Aborted.
> [Switching to Thread 0xb6376450 (LWP 12609)]
> __libc_do_syscall () at
> ../ports/sysdeps/unix/sysv/linux/arm/libc-do-syscall.S:44
> 44 ../ports/sysdeps/unix/sysv/linux/arm/libc-do-syscall.S: No
> such file or directory.
> (gdb) bt
> #0 __libc_do_syscall () at
> ../ports/sysdeps/unix/sysv/linux/arm/libc-do-syscall.S:44
> #1 0xb639f0fe in __GI_raise (sig=sig_at_entry=6) at
> ../nptl/sysdeps/unix/sysv/linux/raise.c:56
> #2 0xb63a1956 in __GI_abort () at abort.c:89
> #3 0xb65377a8 in __gnu_cxx::__verbose_terminate_handler() ()
> from /usr/lib/arm-linux-gnueabihf/libstdc++.so.6
> #4 0xb65361c8 in ?? () from
> /usr/lib/arm-linux-gnueabihf/libstdc++.so.6
> Backtrace stopped: previous frame identical to this frame (corrupt
> stack?)
>
>
> What looks to be happening here is that thread cancellation, which
> works by having an exception thrown, appears to not be caught by the
> thread being cancelled for some odd reason.
>
> I don't have time to investigate this until maybe Saturday. I guess
> we'll just have to ship 1.56 and look into this for 1.57.
Good news on this, though it took a day of head scratching to figure
it out.
It turns out that ARM clang 3.3 and earlier implements C++ exceptions
using the inefficient SJLJ exception ABI - you know, the one that
calls setjmp all the time for every place an unwind might happen.
Anyway, they finally got round to implementing the ARM EHABI
exception ABI which is zero runtime cost and they went ahead and
turned it on by default in 3.4, or at least it is being turned on by
default in the Debian/Ubuntu binaries as well as the LLVM binaries.
Unfortunately, it was very broken indeed. It produces ARM binaries
which simply cannot catch non-trivial C++ exceptions, though
otherwise work fine. This caused Boost.Thread when interrupting a
thread wait to enter an infinite loop at described above, indeed if
you EVER catch a type with RTTI it infinite loops.
Fortunately, Chromium realised this shortly after the 3.4 release,
and they've been hard at work making a EHABI implementation which
actually works for 3.5. I just finished compiling 3.5 from trunk and
I can confirm that all AFIO unit test pass swimmingly with it for
armhf.
You can read more about the 3.5 ARM exception handling improvements
at http://llvm.org/docs/ReleaseNotes.html#changes-to-the-arm-backend.
I'll go notify Debian and Ubuntu upstream that clang 3.4 on ARM is
useless for C++. Something in the Boost 1.56 release notes might be
helpful too, basically don't use ARM EHABI before clang 3.5.
Niall
-- ned Productions Limited Consulting http://www.nedproductions.biz/ http://ie.linkedin.com/in/nialldouglas/