How to benchmark a C program

Question

I wrote server and client programs in C and ran them on my Beaglebone platform with different Linux versions (I compiled it). However, the performance of the old Linux version(2.6) is better than the new Linux version(4.9).

So I need to do some profiling to find the time cost of functions, etc.

//server snippet
while(1){
    if(read(new_socket, buf, sizeof(int)) <= 0)
        break;

    // timecost start
    if(!init){
        gettimeofday(&begin, NULL);
        gettimeofday(&start, NULL);
        init = 1;
    }

    c++;
    if (c / 1000000 != print){
        gettimeofday(&end, NULL);
        printf("%d (%d) tot: %ld, last: %ld\n", print, count,
            get_usec(begin, end), get_usec(start, end));
        print = c / 1000000;
        gettimeofday(&start, NULL);
    }
    // timecost end

    count++;
    send(new_socket, buf, sizeof(int), 0);
}

//client snippet
while (1)
{
    send(client_fd, buf, sizeof(int), 0);
    if (read(client_fd, buf, sizeof(int)) <= 0)
        break;

    // timecost start
    if(!init){
        gettimeofday(&begin, NULL);
        gettimeofday(&start, NULL);
        init = 1;
    }

    c++;
    if (c / 1000000 != print){
        gettimeofday(&end, NULL);
        printf("%d (%d) tot: %ld, last: %ld\n", print, count,
                get_usec(begin, end), get_usec(start, end));
        print = c / 1000000;
        gettimeofday(&start, NULL);
    }
    // timecost end

    count++;
}

the result I got on old/new linux kernels,

2.x kernel:
0 (1999999) tot: 4609998, last: 4609984
1 (3999999) tot: 9150170, last: 4540120

4.x kernel:
0 (1999999) tot: 9451743, last: 9451729
1 (3999999) tot: 18892372, last: 9440623

Are there any tools that can be used to debug/profile to find the system call usage/ kernel time cost?

I did a perf record/report on the server process, every function costs more CPU on old kernel.

old:
    39.54%   server  [kernel.kallsyms]  [k] finish_task_switch
     5.02%   server  [kernel.kallsyms]  [k] _raw_spin_unlock_irqrestore
     2.59%   server  [kernel.kallsyms]  [k] nf_hook_slow
new: 
    24.74%  server   [kernel.kallsyms]  [k] finish_task_switch
     2.67%  server   [kernel.kallsyms]  [k] nf_hook_slow