On 2/5/22, Dominik Brodowski <linux@xxxxxxxxxxxxxxxxxxxx> wrote: > Why are we only using RDRAND here, and not RDSEED? Simply because that's what was already used here; I didn't revisit the old decision. It seems like any changes there should be made in a separate patch with its own justification. If you think there's good rationale, free to send that. Part of why these changes are so gradual is because much of random.c isn't my code originally. Were it mine, I'd presumably know all my various rationales and be able to rapidly think within them and reevaluate. But because that's not the case, I find that I'm spending a lot of time trying to reconstruct the original rationales of its authors. IOW, rather than saying, "I don't get this, must be bad," I'm trying to do a little bit of archeology to at least make sure I know what I'm disagreeing with, if I even disagree at all. That's time consuming in part, but also is part of doing things evolutionarily. With regards to RDRAND vs RDSEED, just off the top of my head -- I'm writing this email on my phone -- I think extract_entropy/extract_buf used to be used as part of /dev/random's blocking stream, which ostensibly could mean more frequent calls, once every 10 bytes IIRC. Nowadays it's only called once every 5 minutes (per numa node), so maybe RDSEED could make sense? Or maybe there are other reasons to unearth, or none at all. We'll have to look and see. Jason
