Boost C++ Libraries: Ticket #3526: Data races in boost::thread detected by drd/helgrind

attachment set

Thu, 15 Oct 2009 11:06:07 GMT

attachment → main.cpp

My test case to run with valgrind

Steven Watanabe — Mon, 19 Oct 2009 19:03:34 GMT

Can you post the output?

anonymous — Mon, 19 Oct 2009 20:17:51 GMT

Yes, of course. Here is the output for current boost svn version (now in Arch Linux): {BUILD}

make all
Building file: ../src/test.cpp
Invoking: GCC C++ Compiler
g++ -O0 -g3 -Wall -c -fmessage-length=0 -MMD -MP -MF"src/test.d" -MT"src/test.d" -o"src/test.o" "../src/test.cpp"
Finished building: ../src/test.cpp
Building target: test
Invoking: GCC C++ Linker
g++  -o"test"  ./src/test.o   -lboost_thread
Finished building target: test

{RUN}

[mist@mist-pc test]$ valgrind --tool=drd ./Debug/test
==13143== drd, a thread error detector
==13143== Copyright (C) 2006-2009, and GNU GPL'd, by Bart Van Assche.
==13143== Using Valgrind-3.5.0 and LibVEX; rerun with -h for copyright info
==13143== Command: ./Debug/test
==13143==
==13143== Thread 3:
==13143== Conflicting load by thread 3 at 0x05043df0 size 8
==13143==    at 0x4E3A684: ??? (in /usr/local/lib/libboost_thread.so.1.41.0)
==13143==    by 0x4E3AA88: boost::detail::set_current_thread_data(boost::detail::thread_data_base*) (in /usr/local/lib/libboost_thread.so.1.41.0)
==13143==    by 0x4E3B084: thread_proxy (in /usr/local/lib/libboost_thread.so.1.41.0)
==13143==    by 0x4C2CC70: vgDrd_thread_wrapper (in /usr/lib/valgrind/vgpreload_drd-amd64-linux.so)
==13143==    by 0x5D4B579: start_thread (in /lib/libpthread-2.10.1.so)
==13143==    by 0x58B514C: clone (in /lib/libc-2.10.1.so)
==13143== Allocation context: BSS section of /usr/local/lib/libboost_thread.so.1.41.0
==13143== Other segment start (thread 2)
==13143==    at 0x4C25533: pthread_mutex_lock (in /usr/lib/valgrind/vgpreload_drd-amd64-linux.so)
==13143==    by 0x4E3A73D: ??? (in /usr/local/lib/libboost_thread.so.1.41.0)
==13143==    by 0x4E3AA88: boost::detail::set_current_thread_data(boost::detail::thread_data_base*) (in /usr/local/lib/libboost_thread.so.1.41.0)
==13143==    by 0x4E3B084: thread_proxy (in /usr/local/lib/libboost_thread.so.1.41.0)
==13143==    by 0x4C2CC70: vgDrd_thread_wrapper (in /usr/lib/valgrind/vgpreload_drd-amd64-linux.so)
==13143==    by 0x5D4B579: start_thread (in /lib/libpthread-2.10.1.so)
==13143==    by 0x58B514C: clone (in /lib/libc-2.10.1.so)
==13143== Other segment end (thread 2)
==13143==    at 0x4C2615F: pthread_mutex_unlock (in /usr/lib/valgrind/vgpreload_drd-amd64-linux.so)
==13143==    by 0x4E3AA88: boost::detail::set_current_thread_data(boost::detail::thread_data_base*) (in /usr/local/lib/libboost_thread.so.1.41.0)
==13143==    by 0x4E3B084: thread_proxy (in /usr/local/lib/libboost_thread.so.1.41.0)
==13143==    by 0x4C2CC70: vgDrd_thread_wrapper (in /usr/lib/valgrind/vgpreload_drd-amd64-linux.so)
==13143==    by 0x5D4B579: start_thread (in /lib/libpthread-2.10.1.so)
==13143==    by 0x58B514C: clone (in /lib/libc-2.10.1.so)
==13143== Other segment start (thread 2)
==13143==    at 0x4C25533: pthread_mutex_lock (in /usr/lib/valgrind/vgpreload_drd-amd64-linux.so)
==13143==    by 0x4E3A6AB: ??? (in /usr/local/lib/libboost_thread.so.1.41.0)
==13143==    by 0x4E3AA88: boost::detail::set_current_thread_data(boost::detail::thread_data_base*) (in /usr/local/lib/libboost_thread.so.1.41.0)
==13143==    by 0x4E3B084: thread_proxy (in /usr/local/lib/libboost_thread.so.1.41.0)
==13143==    by 0x4C2CC70: vgDrd_thread_wrapper (in /usr/lib/valgrind/vgpreload_drd-amd64-linux.so)
==13143==    by 0x5D4B579: start_thread (in /lib/libpthread-2.10.1.so)
==13143==    by 0x58B514C: clone (in /lib/libc-2.10.1.so)
==13143== Other segment end (thread 2)
==13143==    at 0x4C2615F: pthread_mutex_unlock (in /usr/lib/valgrind/vgpreload_drd-amd64-linux.so)
==13143==    by 0x4E3A722: ??? (in /usr/local/lib/libboost_thread.so.1.41.0)
==13143==    by 0x4E3AA88: boost::detail::set_current_thread_data(boost::detail::thread_data_base*) (in /usr/local/lib/libboost_thread.so.1.41.0)
==13143==    by 0x4E3B084: thread_proxy (in /usr/local/lib/libboost_thread.so.1.41.0)
==13143==    by 0x4C2CC70: vgDrd_thread_wrapper (in /usr/lib/valgrind/vgpreload_drd-amd64-linux.so)
==13143==    by 0x5D4B579: start_thread (in /lib/libpthread-2.10.1.so)
==13143==    by 0x58B514C: clone (in /lib/libc-2.10.1.so)
==13143==
==13143==
==13143== For counts of detected and suppressed errors, rerun with: -v
==13143== ERROR SUMMARY: 1 errors from 1 contexts (suppressed: 34 from 29)

Mon, 19 Oct 2009 21:04:15 GMT

(it was output for program attached in ticket)

Steven Watanabe — Tue, 20 Oct 2009 02:02:11 GMT

Okay. The problem is [source:trunk/boost/thread/pthread/once.hpp@45865#L49 once.hpp line 49] and [source:trunk/boost/thread/pthread/once.hpp@45865#L72 once.hpp line 72]. For this to work, we need atomic read and write. I'm not sure whether this is a real problem or not. At least it should be explicit.

Tue, 20 Oct 2009 11:23:24 GMT

Thanks! Mhm, if it will be stated as a not-real-problem, may be it is possible to contact with valgrind developers to give ability to switch off such message? It makes rather difficult to debug real thread usage errors.

version, severity changed; cc set

bvanassche@… — Sat, 05 Jun 2010 10:40:33 GMT

cc bvanassche@… added
version Boost 1.40.0 → Boost Development Trunk
severity Regression → Problem

This is a real bug. With the current implementation it is possible that call_once() invokes its second argument (Function f) twice. flag.epoch should be tested again after detail::once_epoch_mutex has been locked, just like as when applying the double-checked optimization to the singleton pattern.

bvanassche@… — Sat, 05 Jun 2010 11:14:28 GMT

Replying to bvanassche@…:

This is a real bug. With the current implementation it is possible that call_once() invokes its second argument (Function f) twice. flag.epoch should be tested again after detail::once_epoch_mutex has been locked, just like as when applying the double-checked optimization to the singleton pattern.

Please ignore the above comment -- I overlooked the pthread_mutex_scoped_lock statement near the beginning of call_once() which guarantees thread-safe updating of flag.epoch and also that f() is invoked by only one thread.

status, severity changed; resolution set

bvanassche@… — Sat, 05 Jun 2010 11:38:31 GMT

status new → closed
resolution → invalid
severity Problem → Regression

The reason DRD printed the above report is as follows:

boost::detail::get_current_thread_data() and boost::detail::set_current_thread_data() both invoke boost::call_once(current_thread_tls_init_flag,create_current_thread_tls_key) and next use current_thread_tls_key.
The first call of boost::call_once() invokes create_current_thread_tls_key() and hence sets current_thread_tls_key. This last function is invoked without holding the mutex detail::once_epoch_mutex.
The second and subsequent invocations of boost::call_once() neither lock detail::once_epoch_mutex nor trigger a call to create_current_thread_tls_key().

Or: although every read of current_thread_tls_key happens after its initialization, this ordering is invisible to data race detection tools. That's why DRD complains about it. A suppression pattern for this access pattern will be added in Valgrind 3.6.0.

Thu, 19 Jan 2012 08:23:02 GMT

I still get the data race reports with valgrind-3.6.1. Did the suppression pattern not get into valgrind 3.6? Or is there another problem?

bvanassche@… — Thu, 19 Jan 2012 12:55:01 GMT

An appropriate suppression pattern was added in glibc-2.X-drd.supp via SVN revisions 8848, 10684 and 11152. As far as I can see these changes are present in Valgrind 3.6.0. Note: these changes only affect DRD but not Helgrind.

Thu, 19 Jan 2012 13:14:15 GMT

I get the following output with valgrind 3.6.1 and drd:

[froth@megaera ~]$ valgrind --tool=drd ./a.out
==9839== drd, a thread error detector
==9839== Copyright (C) 2006-2010, and GNU GPL'd, by Bart Van Assche.
==9839== Using Valgrind-3.6.1 and LibVEX; rerun with -h for copyright info
==9839== Command: ./a.out
==9839==
==9839== Thread 3:
==9839== Conflicting load by thread 3 at 0x040431ec size 4
==9839==    at 0x4035873: ??? (in /usr/lib/libboost_thread-mt.so.1.47.0)
==9839==    by 0x42F17CD2: start_thread (in /lib/libpthread-2.14.90.so)
==9839==    by 0x42E5583D: clone (in /lib/libc-2.14.90.so)
==9839== Allocation context: BSS section of /usr/lib/libboost_thread-mt.so.1.47.0
==9839== Other segment start (thread 2)
==9839==    at 0x400AC73: pthread_mutex_lock (drd_pthread_intercepts.c:587)
==9839==    by 0x42E63C03: pthread_mutex_lock (in /lib/libc-2.14.90.so)
==9839==    by 0x403595B: ??? (in /usr/lib/libboost_thread-mt.so.1.47.0)
==9839==    by 0x4035B84: boost::detail::set_current_thread_data(boost::detail::thread_data_base*) (in /usr/lib/libboost_thread-mt.so.1.47.0)
==9839==    by 0x42F17CD2: start_thread (in /lib/libpthread-2.14.90.so)
==9839==    by 0x42E5583D: clone (in /lib/libc-2.14.90.so)
==9839== Other segment end (thread 2)
==9839==    at 0x400B508: pthread_mutex_unlock (drd_pthread_intercepts.c:640)
==9839==    by 0x42E63C43: pthread_mutex_unlock (in /lib/libc-2.14.90.so)
==9839==    by 0x40358CD: ??? (in /usr/lib/libboost_thread-mt.so.1.47.0)
==9839==    by 0x4035B84: boost::detail::set_current_thread_data(boost::detail::thread_data_base*) (in /usr/lib/libboost_thread-mt.so.1.47.0)
==9839==    by 0x42F17CD2: start_thread (in /lib/libpthread-2.14.90.so)
==9839==    by 0x42E5583D: clone (in /lib/libc-2.14.90.so)
==9839== Other segment start (thread 2)
==9839==    at 0x400AC73: pthread_mutex_lock (drd_pthread_intercepts.c:587)
==9839==    by 0x42E63C03: pthread_mutex_lock (in /lib/libc-2.14.90.so)
==9839==    by 0x403589A: ??? (in /usr/lib/libboost_thread-mt.so.1.47.0)
==9839==    by 0x4035B84: boost::detail::set_current_thread_data(boost::detail::thread_data_base*) (in /usr/lib/libboost_thread-mt.so.1.47.0)
==9839==    by 0x42F17CD2: start_thread (in /lib/libpthread-2.14.90.so)
==9839==    by 0x42E5583D: clone (in /lib/libc-2.14.90.so)
==9839== Other segment end (thread 2)
==9839==    at 0x400B508: pthread_mutex_unlock (drd_pthread_intercepts.c:640)
==9839==    by 0x42E63C43: pthread_mutex_unlock (in /lib/libc-2.14.90.so)
==9839==    by 0x403593B: ??? (in /usr/lib/libboost_thread-mt.so.1.47.0)
==9839==    by 0x4035B84: boost::detail::set_current_thread_data(boost::detail::thread_data_base*) (in /usr/lib/libboost_thread-mt.so.1.47.0)
==9839==    by 0x42F17CD2: start_thread (in /lib/libpthread-2.14.90.so)
==9839==    by 0x42E5583D: clone (in /lib/libc-2.14.90.so)
==9839==
==9839== Conflicting load by thread 3 at 0x040431e8 size 4
==9839==    at 0x4035879: ??? (in /usr/lib/libboost_thread-mt.so.1.47.0)
==9839==    by 0x42F17CD2: start_thread (in /lib/libpthread-2.14.90.so)
==9839==    by 0x42E5583D: clone (in /lib/libc-2.14.90.so)
==9839== Allocation context: BSS section of /usr/lib/libboost_thread-mt.so.1.47.0
==9839== Other segment start (thread 2)
==9839==    at 0x400AC73: pthread_mutex_lock (drd_pthread_intercepts.c:587)
==9839==    by 0x42E63C03: pthread_mutex_lock (in /lib/libc-2.14.90.so)
==9839==    by 0x403595B: ??? (in /usr/lib/libboost_thread-mt.so.1.47.0)
==9839==    by 0x4035B84: boost::detail::set_current_thread_data(boost::detail::thread_data_base*) (in /usr/lib/libboost_thread-mt.so.1.47.0)
==9839==    by 0x42F17CD2: start_thread (in /lib/libpthread-2.14.90.so)
==9839==    by 0x42E5583D: clone (in /lib/libc-2.14.90.so)
==9839== Other segment end (thread 2)
==9839==    at 0x400B508: pthread_mutex_unlock (drd_pthread_intercepts.c:640)
==9839==    by 0x42E63C43: pthread_mutex_unlock (in /lib/libc-2.14.90.so)
==9839==    by 0x40358CD: ??? (in /usr/lib/libboost_thread-mt.so.1.47.0)
==9839==    by 0x4035B84: boost::detail::set_current_thread_data(boost::detail::thread_data_base*) (in /usr/lib/libboost_thread-mt.so.1.47.0)
==9839==    by 0x42F17CD2: start_thread (in /lib/libpthread-2.14.90.so)
==9839==    by 0x42E5583D: clone (in /lib/libc-2.14.90.so)
==9839== Other segment start (thread 2)
==9839==    at 0x400AC73: pthread_mutex_lock (drd_pthread_intercepts.c:587)
==9839==    by 0x42E63C03: pthread_mutex_lock (in /lib/libc-2.14.90.so)
==9839==    by 0x403589A: ??? (in /usr/lib/libboost_thread-mt.so.1.47.0)
==9839==    by 0x4035B84: boost::detail::set_current_thread_data(boost::detail::thread_data_base*) (in /usr/lib/libboost_thread-mt.so.1.47.0)
==9839==    by 0x42F17CD2: start_thread (in /lib/libpthread-2.14.90.so)
==9839==    by 0x42E5583D: clone (in /lib/libc-2.14.90.so)
==9839== Other segment end (thread 2)
==9839==    at 0x400B508: pthread_mutex_unlock (drd_pthread_intercepts.c:640)
==9839==    by 0x42E63C43: pthread_mutex_unlock (in /lib/libc-2.14.90.so)
==9839==    by 0x403593B: ??? (in /usr/lib/libboost_thread-mt.so.1.47.0)
==9839==    by 0x4035B84: boost::detail::set_current_thread_data(boost::detail::thread_data_base*) (in /usr/lib/libboost_thread-mt.so.1.47.0)
==9839==    by 0x42F17CD2: start_thread (in /lib/libpthread-2.14.90.so)
==9839==    by 0x42E5583D: clone (in /lib/libc-2.14.90.so)
==9839==
==9839== Thread 1:
==9839== Conflicting load by thread 1 at 0x040431ec size 4
==9839==    at 0x4035873: ??? (in /usr/lib/libboost_thread-mt.so.1.47.0)
==9839==    by 0x42D7C6B2: (below main) (in /lib/libc-2.14.90.so)
==9839== Allocation context: BSS section of /usr/lib/libboost_thread-mt.so.1.47.0
==9839== Other segment start (thread 2)
==9839==    at 0x400B508: pthread_mutex_unlock (drd_pthread_intercepts.c:640)
==9839==    by 0x42E63C43: pthread_mutex_unlock (in /lib/libc-2.14.90.so)
==9839==    by 0x403593B: ??? (in /usr/lib/libboost_thread-mt.so.1.47.0)
==9839==    by 0x4035B84: boost::detail::set_current_thread_data(boost::detail::thread_data_base*) (in /usr/lib/libboost_thread-mt.so.1.47.0)
==9839==    by 0x42F17CD2: start_thread (in /lib/libpthread-2.14.90.so)
==9839==    by 0x42E5583D: clone (in /lib/libc-2.14.90.so)
==9839== Other segment end (thread 2)
==9839==    at 0x400B508: pthread_mutex_unlock (drd_pthread_intercepts.c:640)
==9839==    by 0x42E63C43: pthread_mutex_unlock (in /lib/libc-2.14.90.so)
==9839==    by 0x40358CD: ??? (in /usr/lib/libboost_thread-mt.so.1.47.0)
==9839==    by 0x4035B84: boost::detail::set_current_thread_data(boost::detail::thread_data_base*) (in /usr/lib/libboost_thread-mt.so.1.47.0)
==9839==    by 0x42F17CD2: start_thread (in /lib/libpthread-2.14.90.so)
==9839==    by 0x42E5583D: clone (in /lib/libc-2.14.90.so)
==9839== Other segment start (thread 2)
==9839==    at 0x42E55828: clone (in /lib/libc-2.14.90.so)
==9839== Other segment end (thread 2)
==9839==    at 0x400B508: pthread_mutex_unlock (drd_pthread_intercepts.c:640)
==9839==    by 0x42E63C43: pthread_mutex_unlock (in /lib/libc-2.14.90.so)
==9839==    by 0x403593B: ??? (in /usr/lib/libboost_thread-mt.so.1.47.0)
==9839==    by 0x4035B84: boost::detail::set_current_thread_data(boost::detail::thread_data_base*) (in /usr/lib/libboost_thread-mt.so.1.47.0)
==9839==    by 0x42F17CD2: start_thread (in /lib/libpthread-2.14.90.so)
==9839==    by 0x42E5583D: clone (in /lib/libc-2.14.90.so)
==9839==
==9839== Conflicting load by thread 1 at 0x040431e8 size 4
==9839==    at 0x4035879: ??? (in /usr/lib/libboost_thread-mt.so.1.47.0)
==9839==    by 0x42D7C6B2: (below main) (in /lib/libc-2.14.90.so)
==9839== Allocation context: BSS section of /usr/lib/libboost_thread-mt.so.1.47.0
==9839== Other segment start (thread 2)
==9839==    at 0x400B508: pthread_mutex_unlock (drd_pthread_intercepts.c:640)
==9839==    by 0x42E63C43: pthread_mutex_unlock (in /lib/libc-2.14.90.so)
==9839==    by 0x403593B: ??? (in /usr/lib/libboost_thread-mt.so.1.47.0)
==9839==    by 0x4035B84: boost::detail::set_current_thread_data(boost::detail::thread_data_base*) (in /usr/lib/libboost_thread-mt.so.1.47.0)
==9839==    by 0x42F17CD2: start_thread (in /lib/libpthread-2.14.90.so)
==9839==    by 0x42E5583D: clone (in /lib/libc-2.14.90.so)
==9839== Other segment end (thread 2)
==9839==    at 0x400B508: pthread_mutex_unlock (drd_pthread_intercepts.c:640)
==9839==    by 0x42E63C43: pthread_mutex_unlock (in /lib/libc-2.14.90.so)
==9839==    by 0x40358CD: ??? (in /usr/lib/libboost_thread-mt.so.1.47.0)
==9839==    by 0x4035B84: boost::detail::set_current_thread_data(boost::detail::thread_data_base*) (in /usr/lib/libboost_thread-mt.so.1.47.0)
==9839==    by 0x42F17CD2: start_thread (in /lib/libpthread-2.14.90.so)
==9839==    by 0x42E5583D: clone (in /lib/libc-2.14.90.so)
==9839== Other segment start (thread 2)
==9839==    at 0x42E55828: clone (in /lib/libc-2.14.90.so)
==9839== Other segment end (thread 2)
==9839==    at 0x400B508: pthread_mutex_unlock (drd_pthread_intercepts.c:640)
==9839==    by 0x42E63C43: pthread_mutex_unlock (in /lib/libc-2.14.90.so)
==9839==    by 0x403593B: ??? (in /usr/lib/libboost_thread-mt.so.1.47.0)
==9839==    by 0x4035B84: boost::detail::set_current_thread_data(boost::detail::thread_data_base*) (in /usr/lib/libboost_thread-mt.so.1.47.0)
==9839==    by 0x42F17CD2: start_thread (in /lib/libpthread-2.14.90.so)
==9839==    by 0x42E5583D: clone (in /lib/libc-2.14.90.so)
==9839==
==9839==
==9839== For counts of detected and suppressed errors, rerun with: -v
==9839== ERROR SUMMARY: 4 errors from 4 contexts (suppressed: 42 from 22)