Mill Computing, Inc. Forums The Mill Architecture Security Reply To: Security

joseph.h.garvin
Participant
Post count: 22

I really like the Mill’s approach to stack safety and that in particular it prevents Return Oriented Programming.

Has making random number generation support built in been considered? Besides the stack, the other major bane of embedded security is random number generation. Bad random numbers weaken crypto, and there have been a ton of vulnerabilities relating to routers picking their random seed at boot up time before sufficient entropy has been collected by the OS, leading to predictable random numbers and allowing entry to attackers. Intel has RDRAND but that doesn’t help most of the embedded world. Obviously there’s no reason in principle why a hardware entropy source couldn’t be integrated into the Mill, but it would be great for it to be part of the minimum configuration (Tin) so that its hard to screw up and help cement a reputation for the Mill as being more secure than other CPUs. This could also be a great opportunity to take advantage of multiple outputs on the belt and/or the Mill vector types, lots of simulations will use random matrices/vectors. If you wanted to be really fancy you could support different distributions (e.g. Zipfian vs. Gaussian).