> It works terrible fine. I bet that you can4t guess the value of a counter
> that runs at 1Mhz.

Sounds good when asynchronous and fast compared to the read-rate. Like in
your particular implementation.

A dangerous implementation would be to use the internal TMR of a controller
instead of an asynchronous one.  Probably you have your keyboard reader
in the TMR interrupt.  Probably you need a new "random" value when the
user starts a game - by pressing the START _key_!  THAT would be a bad
implementation of the (good) idea.

When it comes to security/cryptography, you must avoid that the user can
influence the random number algorithm.  In your example that could be
to remove the external counter and replace it by his own generated signal,
and thereby enforce the generation of weak session keys etc.