global tm_fork_start_par[128], tm_fork_start[128], tm_fork_end[128], 
       tm_exec_start[128], tm_exec_end[128], p_argstr[128];
global fork[128], postfork[128], exec[128], proc[128];

probe syscall.fork {
    tm_fork_start_par[tid()] = local_clock_us();   
}
probe syscall.fork.return {
    if($return > 1) {
        tm_fork_start[$return] = tm_fork_start_par[tid()];
        delete tm_fork_start_par[tid()];
    }
}
probe kprocess.start {
    tm_fork_end[pid()] = local_clock_us();
}
probe kprocess.exec {
    p_argstr[pid()] = argstr;
    tm_exec_start[pid()] = local_clock_us();
}
probe kprocess.exec_complete {
    tm_exec_end[pid()] = local_clock_us();
}
probe kprocess.exit {
    argstr = p_argstr[pid()];

    fork[execname(), argstr] <<< tm_fork_end[pid()] - tm_fork_start[pid()];
    postfork[execname(), argstr] <<< tm_exec_start[pid()] - tm_fork_end[pid()];
    exec[execname(), argstr] <<< tm_exec_end[pid()] - tm_exec_start[pid()];
    proc[execname(), argstr] <<< local_clock_us() - tm_exec_end[pid()];

    delete tm_fork_start[pid()];    delete tm_fork_end[pid()];
    delete tm_exec_start[pid()];    delete tm_exec_end[pid()];
    delete p_argstr[pid()];
}

probe timer.s(1) {
    printf("%48s %8s %8s %8s %8s\n",
            "COMMAND", "FORK", "POSTFORK", "EXEC", "PROC");
    foreach([execname, args] in proc) {
        printf("%10s %36s  %6dus %6dus %6dus %6dus\n", execname, args, 
                @avg(fork[execname, args]), 
                @avg(postfork[execname, args]), 
                @avg(exec[execname, args]), 
                @avg(proc[execname, args]));
    }
    delete fork; delete postfork; delete exec; delete proc;
}