Problem with DCE and std::thread (probably a DCE bug?)

[Andreas Pantelopoulos]

Hello to all,

I posted yesterday a problem I had with porting a software switch executable in DCE, that uses std::thread. In this post I provide more details and a way to re-produce the problem easily. I managed to reproduce the problem with a very simple c++ program that uses two threads. The program exits with a SIGSEGV. Attached you will find the code for this simple program. I also attach the output of the gdb backtrace when the error occurs. I also modified the task-manager so as to print whenever it switches between threads. The cpp program is really simple: The main boots up a thread and this threads execute in a while loop, in which it sleeps for 2 seconds in each loop. The main thread also executes the same loop.

I attach the gdb result of "backtrace" command after the error occurs. What really bugs me is that the error is received in different points in each run. I believe this is a bug in DCE, or I understand and do something really really wrong, or both. 

My dce version is the 1.9 and the ns-3 version is the 3.26.

I compile the program with " g++ -std=c++11 -fPIC simple.cpp -pie -rdynamic -lpthread -o simple " and then I copy the executable in the /usr/local/bin (or wherever the DCE_PATH or DCE_ROOT points to). In order to execute it quickly you can modify the "myscripts/sleep" example to point to the "simple" binary. Also, comment the Simulator::stop() line in the script.

Any help with this issue would be greatly appreciated.

Thanks,

Andreas Pantelopoulos

[Andreas Pantelopoulos]

Continuing on the following post, I notice that if I comment any of the two std::this_thread::sleep_for() or both of them, the program executes just fine. Looking at the source code of sleep_for (https://gcc.gnu.org/onlinedocs/gcc-4.6.2/libstdc++/api/a01083_source.html) I observe that it calls nanosleep() which is captured by dce and wrapped around dce_nanosleep. Maybe there is a race condition in dce_nanosleep ?

Thanks,

Andreas

[Andreas Pantelopoulos]

A little update for anyone that cares to help. 

So, std::thread indeed builds on good-old pthreads in my linux box, and all the calls from pthreads should be intercepted by the dce runtime system. However, by using manual breakpoints in gdb, I determined that when the call to pthread_create happens through std::thread, the DCE does not intercept it in order to create the thread in it's own thread management system. However, the rest system calls are, and thus a nasty race condition occurs. 

If anyone is willing to help solving that, preferably someone who has developed for DCE in the past, that would be awesome!

Andreas

[Andreas Pantelopoulos]

In case anyone is wondering or experiences same issues with std::thread and ns3, the problem is solved by compiling and linking against libc++ from llvm. 

Also, compile against clang++.

Thanks for the help,

Andreas

[Tommaso Pecorella]

Great to know Andreas.

I'm sorry if you had to figure out by yourself, but I can't build DCE on my computer, so I'm kinda useless on DCE-related issues.

T.