Not sure I understand the /dev/urandom concern in that comment. If this doesn't need to be cryptographically good and you don't want logspam every time a new number is needed why not seed this once with /dev/urandom. Surely a single log message is better than trying to reinvent the wheel, especially since this isn't the first time I've seen this particular code linked on HN due to the assumptions not holding true.

Dude is stubborn in the face of reality. It's not as if there hadn't been RDRAND hardware bugs before.