Cryptographic random number generators seeded by physical entropy sources are employed in many embedded security systems, including self-encrypting disk drives, being manufactured by the millions every year. Random numbers are used for generating encryption keys and for facilitating secure communication, and they are also provided to users for their applications. We discuss common randomness requirements, techniques for estimating the entropy of physical sources, investigate specific nonrandom physical properties, estimate the autocorrelation, then mix reduce the data until all common randomness tests pass. This method is applied to a randomness source in disk drives: the always changing coefficients of an adaptive filter for the read channel equalization. These coefficients, affected by many kinds of physical noise, are used in the reseeding process of a cryptographic pseudorandom number generator in a family of self encrypting disk drives currently in the market.