This is cool. I've had my own hacky version of this for a while, and it's been really useful. Thanks for submitting!
Just minor notes below.

src/sim/backtrace_glibc.cc (Diff revision 1)

Can we lift SAFE_MSG into backtrace.hh maybe, so there is only one copy of it?

Show all issues

1. Andreas Sandberg Nov. 16, 2015, 10:06 a.m. (Nov. 16, 2015, 10:06 a.m.)

   How about putting it in atomicio.hh instead and make the FD a parameter? E.g.:
   #define STATIC_MSG(fd, msg) ...
   #define STATIC_ERR(msg) STATIC_MSG(STDERRR_FILENO, msg)
   

2. Joel Hestness Nov. 16, 2015, 11:32 a.m. (Nov. 16, 2015, 11:32 a.m.)

   Yes! That's even better.

src/sim/init_signals.cc (Diff revision 1)

Should this return an exit code consistent with the signal that was raised (e.g. SIGSEGV = 139)? Someone may have scripts that handle simulator error codes, and hard coding here might make them inconsistent. Or maybe it's better that this is unique, since it looks like a strange place to exit?

Show all issues

1. Andreas Sandberg Nov. 16, 2015, 10:06 a.m. (Nov. 16, 2015, 10:06 a.m.)

   I don't know what the best solution is here. This should never ever happen under normal circumstances. IIRC, the C library does something along these lines in abort():
   reset_signal_handler();
   pthread_kill(pthread_self(), SIGABRT);
   asm("hlt");
   _exit(127);
   asm("1: hlt; jmp 1b");
   
   Normally, raising a fatal signal after it has been reset to the default handler should result in immediate termination, so the code after pthread_kill shouldn't normally be executed.
   I don't think anyone will ever look at the exit value to determine the signal. We just need to tell run scripts / the user that things went horribly wrong at this point.

2. Joel Hestness Nov. 16, 2015, 11:32 a.m. (Nov. 16, 2015, 11:32 a.m.)

   I do print exit codes in bash wrapper scripts, since they tend to be pretty informative (e.g. sometimes inside a wrapper script, a segfault will not print anything to terminal or output, and the simulation just exits. In bash, running 'echo $?' immediately after the simulator command will print 128 + 11 = 139 to indicate indicate SIGSEGV. Bash exit codes are here: http://tldp.org/LDP/abs/html/exitcodes.html ).
   That said, it's probably fine to leave it hardcoded as-is, since this is pretty low probability.

Ship It!