GDB is a debugger used to investigate software problem. GDB is an acronym saying GDB: The GNU Project Debugger
With a debugger, it is possible to quickly find the cause of a problem in a piece of software. Often it is used to resolve segmentation faults. If you desire to resolve a problem relating to memory (for example, a memory leak), we recommend using Valgrind.
Finding a bug with the debugger
In this section, the following program is used:
This program generates a segmentation fault when it is ran.
[name@server $] g++ -g program.cpp -o program [name@server $] ./program Segmentation fault (core dumped) [name@server $]
We may then run the program inside the debugger. Note that we compiled using the option -g to include debugging symbols within the binary and allow the debugger to provide more information on the bug. We run the program inside the debugger using
[name@server $] gdb ./program (gdb) run Starting program: /home/seb/program ./program Program received signal SIGSEGV, Segmentation fault. 0x0000000000400c17 in main (argc=2, argv=0x7fffffffda88) at program.cpp:15 15 cout << numbers << endl; Missing separate debuginfos, use: debuginfo-install glibc-2.16-31.fc18.x86_64 libgcc-4.7.2-8.fc18.x86_64 libstdc++-4.7.2-8.fc18.x86_64 (gdb) bt #0 0x0000000000400c17 in main (argc=2, argv=0x7fffffffda88) at program.cpp:15
So, the above error is caused by line 15. The code tries to use index 1000000, but the array only contains 1000 elements.
Finding the cause of a segmentation fault using a core file
In this example, we use the same program as in the previous section. We however do so without using the debugger directly. This is useful for a bug that happens a long time after the program has started.
To find the cause for this error, a core file must be generated. To do this, you must activate the creation of such files.
[name@server $] ulimit -c unlimited
Executing the same program again, a core file is written.
[name@server $] ./program Segmentation fault (core dumped) [name@server $] file core.18158 core.18158: ELF 64-bit LSB core file x86-64, version 1 (SYSV), SVR4-style, from './program' [name@server $]
Using the program binary executable and the core file, it is possible to trace its execution up to the error.
[name@server $] gdb -q ./program Reading symbols from /home/seb/program...done. (gdb) core-file core.18246 [New LWP 18246] Core was generated by `./program'. Program terminated with signal 11, Segmentation fault. #0 0x0000000000400c17 in main (argc=1, argv=0x7fff2315c848) at program.cpp:15 15 cout << numbers << endl; Missing separate debuginfos, use: debuginfo-install glibc-2.16-31.fc18.x86_64 libgcc-4.7.2-8.fc18.x86_64 libstdc++-4.7.2-8.fc18.x86_64 (gdb) bt #0 0x0000000000400c17 in main (argc=1, argv=0x7fff2315c848) at program.cpp:15
We here get the same result as if we had run it inside the debugger.
Attaching the debugger to a running process
It is possible to debug a process that is already running, for example a job running on one of the compute nodes. To do so, we first need the process ID.
[name@server $] ps aux | grep firefox | grep -v grep seb 12691 6.4 7.5 1539672 282656 ? Sl 08:53 6:48 /usr/lib64/firefox/firefox http://www.google.ca/
After that, it is possible to attach the debugger directly.
[name@server $] gdb attach 12691
After having done this, a lot of information is displayed.
There are many commands available within GDB. One of the most useful is backtrace, or bt. This commands shows the current call stack.
(gdb) bt #0 0x00000033646e99ad in poll () from /lib64/libc.so.6 #1 0x0000003db86849f3 in PollWrapper(_GPollFD*, unsigned int, int) () from /usr/lib64/firefox/xulrunner/libxul.so #2 0x0000003366e47d24 in g_main_context_iterate.isra.24 () from /lib64/libglib-2.0.so.0 #3 0x0000003366e47e44 in g_main_context_iteration () from /lib64/libglib-2.0.so.0 #4 0x0000003db86849a2 in nsAppShell::ProcessNextNativeEvent(bool) () from /usr/lib64/firefox/xulrunner/libxul.so #5 0x0000003db869a7d1 in nsBaseAppShell::DoProcessNextNativeEvent(bool, unsigned int) () from /usr/lib64/firefox/xulrunner/libxul.so #6 0x0000003db869a8ea in nsBaseAppShell::OnProcessNextEvent(nsIThreadInternal*, bool, unsigned int) () from /usr/lib64/firefox/xulrunner/libxul.so #7 0x0000003db89810c2 in nsThread::ProcessNextEvent(bool, bool*) () from /usr/lib64/firefox/xulrunner/libxul.so #8 0x0000003db89563eb in NS_ProcessNextEvent(nsIThread*, bool) () from /usr/lib64/firefox/xulrunner/libxul.so #9 0x0000003db873056f in mozilla::ipc::MessagePump::Run(base::MessagePump::Delegate*) () from /usr/lib64/firefox/xulrunner/libxul.so #10 0x0000003db89a4ab7 in MessageLoop::Run() () from /usr/lib64/firefox/xulrunner/libxul.so #11 0x0000003db869a1b3 in nsBaseAppShell::Run() () from /usr/lib64/firefox/xulrunner/libxul.so #12 0x0000003db857d92d in nsAppStartup::Run() () from /usr/lib64/firefox/xulrunner/libxul.so #13 0x0000003db7d18f4a in XREMain::XRE_mainRun() () from /usr/lib64/firefox/xulrunner/libxul.so #14 0x0000003db7d1b007 in XREMain::XRE_main(int, char**, nsXREAppData const*) () from /usr/lib64/firefox/xulrunner/libxul.so #15 0x0000003db7d1b259 in XRE_main () from /usr/lib64/firefox/xulrunner/libxul.so #16 0x0000000000402c23 in do_main(int, char**, nsIFile*) () #17 0x0000000000402403 in main () (gdb) quit A debugging session is active. Inferior 1 [process 12691] will be detached. Quit anyway? (y or n) y Detaching from program: /usr/lib64/firefox/firefox, process 12691
In the previous sections, we used the run and backtrace commands. Many more commands are available to debug in an interactive way, by stopping the program. For example, you can set breakpoints on functions or lines of code, or whenever a given variable is modified. When execution is interrupted, you can analyse the state of the program by printing the value of variables. The following table contains a list of the main commands.
|where / backtrace||bt||-||displays the backtrace|
|break||b||src.c:line_number or function||sets a break point at the given line of code or function|
|watch||-||variable name||interrupts the program when a variable is modified|
|continue||c||-||resume the program|
|step||s||-||execute the next operation|
|p||variable name||displays the content of a variable|
|list||l||src.c:number||displays the given line of code|
Displaying STL structures
By default, GDB does not display C++ STL structures very well. Many solutions are given here. The simplest solution is probably this one, which is to copy this file in your home folder, with the name ~/.gdbinit.