VGA MMIO: does it really not need volatile writes?

[meithecatte]

Currently, the code used to access VGA does plain memory pokes:

    for (i, &byte) in HELLO.iter().enumerate() {
        unsafe {
            *vga_buffer.offset(i as isize * 2) = byte;
            *vga_buffer.offset(i as isize * 2 + 1) = 0xb;
        }
    }

This surprised me — I expected to see something like ptr::write_volatile here. Are you sure this code isn't working by accident?

I'm not too familiar with the exact semantics, but my guess would be that without volatile, these memory accesses are not considered side-effects, and therefore the compiler could, for example:

• prove that our code won't read them (because of an upcoming infinite loop, for example), and decide remove them
• move the write around some code that might not terminate, for example:

  print("doing the thing... ");
  some_mutex.lock().push(42);
  print("it worked!");

  could get optimised to

  some_mutex.lock().push(42);
  print("doing the thing... it worked!");

  This would suck big time if you're trying to track down a deadlock.

Is this something you've given some thought, or merely an oversight?

[xuanplus]

You will find it in next post: VGA Text Mode