Hi Eric, On Tue, Jul 19, 2022 at 10:51 PM Eric W. Biederman <ebiederm@xxxxxxxxxxxx> wrote: > Acked-by: "Eric W. Biederman" <ebiederm@xxxxxxxxxxxx> Thank you. > The whole design seems very strange to me. I would think instead of > having a current hardware random number generator the kernel would > pull from every hardware random generator available. Further that > we can get a userspace read all of the way into driver code for > a hardware random generator seems weird. I would think in > practice we would want all of this filtered through /dev/random, > /dev/urandom, and the get_entropy syscall. Yes indeed. In general, the hwrng interface is kind of badly designed and a bit of a nuisance. I've spent the last few months reworking random.c and that's finally nearing okay enough shape. Possibly after I'll turn my attention to a real overhaul of hwrng too (assuming Herbert gives me lattitude to do that, I guess; I don't maintain that code). The main goal anyhow ought be to just non-invasively shephard bits into random.c, with additional frills being merely additional. Jason
