Who should I believe?<p>a) The people on the LKML, with experience in both the kernel and userland side of the problem, trying to figure out the best way to manage userland expectations of RNG quality, in a context where entropy gathering is pretty much a glorified hack, of diminishing effectiveness<p>b) The security people on Twitter posting funny memes and simply stating that things should be "secure by default"<p>It would be trivial for the kernel developers to return a hash of whatever entropy they gathered so far, without ever blocking, and call that the "default secure mode", and implement a toy RNG for the non-secure, non-default mode, and call it a day. But they're trying to solve a more difficult and more important problem.