On Thu, Oct 3, 2024 at 6:53 PM Shuah Khan <skhan@xxxxxxxxxxxxxxxxxxx> wrote: > > The x86 test from 6.12 works just fine on 6.11. > > Yes x86 test is a good example to look at that handles 32-bit > and 640-bit issues you brought up in your email. The x86 test is only for 64-bit. It doesn't get compiled on 32-bit. That changes in the series I attached. > 1. Ideally selftests should compile on all architectures. > Exception to this are few architecture specific > features which can be selectively compiled either > with ifdef statements in the code or Makefile arch > checks. The thing is, this *is* a test with an architecture-specific feature: it's not available on all archs, especially not 32-bit ones. There's the workaround (in the attached series) that I think might help things. But for the chacha one, it just won't compile on systems that lack the implementation. > 2. Selftest from mainline should run on stable releases > handling missing features and missing config options > with skip so we don't have to deal false failures. Noted. I would like to point out, though, that this is the "opposite" stability guarantee that Linux usually offers of old binaries working on new systems. You want new binaries working on old systems. > 3. Reported results are clear to the users and testers. Patch 3/3 fixes this up. > Thank you for the patches. I will review your patches and give you > feedback. Great, okay. I think the series matches your expectations. I have additional ideas too for the chacha test, if you think it's necessary to go further, but maybe things are good enough now. Jason
