Re: [Valgrind-users] Assertion 'DRD_(g_threadinfo)[tid].pt_threadid != INVALID_POSIX_THREADID' failed.

[Bart V. A. <bva...@ac...>]

On 02/21/15 01:08, Joe VanAndel wrote:
> I am using valgrind 3.10.1 (I get the same results building valgrind
> from SVN).  
> 
> My program uses boost 1_57_0, and Wt 3-3-4-rc1.  I am compiling with gcc
> 4.8.2 on CentOS 7.0.1406 
> 
> When I run my program under valgrind —tool=drd, 
> 
> valgrind aborts with the attached message.
> I have several questions:
> 
> 1) If this is a real error in my code, Wt, or Boost, how would I find
> and fix the error?
> 
> 2) Does anyone routinely use the boost DRD tool with Wt and Boost, or
> with Boost?
> 
> 
> drd: drd_thread.c:630 (vgDrd_thread_entering_pthread_create): Assertion
> 'DRD_(g_threadinfo)[tid].pt_threadid != INVALID_POSIX_THREADID' failed.
> 
> host stacktrace:
> ==21016==    at 0x38046C68: show_sched_status_wrk (m_libcassert.c:319)
> ==21016==    by 0x38046D84: report_and_quit (m_libcassert.c:390)
> ==21016==    by 0x38046F11: vgPlain_assert_fail (m_libcassert.c:456)
> ==21016==    by 0x3803639B: vgDrd_thread_entering_pthread_create
> (drd_thread.c:630)
> ==21016==    by 0x3802AC6B: handle_client_request (drd_clientreq.c:295)
> ==21016==    by 0x3805E1E0: wrap_tool_handle_client_request
> (m_tooliface.c:280)
> ==21016==    by 0x38097C9F: do_client_request (scheduler.c:2074)
> ==21016==    by 0x38097C9F: vgPlain_scheduler (scheduler.c:1407)
> ==21016==    by 0x380A687C: thread_wrapper (syswrap-linux.c:103)
> ==21016==    by 0x380A687C: run_a_thread_NORETURN (syswrap-linux.c:156)
> 
> sched status:
>   running_tid=1
> 
> Thread 1: status = VgTs_Runnable
> ==21016==    at 0x4C2F5B9: vgDrd_entering_pthread_create
> (drd_pthread_intercepts.c:331)
> ==21016==    by 0x4C2F5B9: pthread_create_intercept
> (drd_pthread_intercepts.c:484)
> ==21016==    by 0x4C2F5B9: pthread_create@* (drd_pthread_intercepts.c:501)
> ==21016==    by 0x5C14C38: boost::thread::start_thread_noexcept() (in
> /usr/local/boost_1_57_0/lib/libboost_thread-mt.so.1.57.0)
> ==21016==    by 0x6A6975: boost::thread::start_thread() (thread.hpp:179)

This might be a bug in the Valgrind core. When a thread is created the
Valgrind core invokes drd_pre_thread_create(). That function calls
DRD_(thread_pre_create)() which sets up the valgrind-to-DRD thread ID
translation. The assertion failure above means that such a translation
had not been set up before the code in the newly created thread started
to run.

If you can provide a small test program with which I can reproduce the
above behavior I will have a look into this. The boost regression test
that is present in the Valgrind source tree seems to work fine on CentOS
7.0.1046:

$ ./vg-in-place --tool=drd drd/tests/boost_thread
==24662== drd, a thread error detector
==24662== Copyright (C) 2006-2013, and GNU GPL'd, by Bart Van Assche.
==24662== Using Valgrind-3.11.0.SVN and LibVEX; rerun with -h for
copyright info
==24662== Command: drd/tests/boost_thread
==24662==
Thread 1.
Thread 2.
Finished.
==24662==
==24662== For counts of detected and suppressed errors, rerun with: -v
==24662== ERROR SUMMARY: 0 errors from 0 contexts (suppressed: 33 from 33)

Bart.