I wrote a silly little testcase: https://gist.github.com/jcalvinowens/06530e6fc86f3b2c97a7027...

...but both affected znver5 machines I have "pass", so clearly there's more to it than AMD is telling us.

You should be able to get away with "-march=znver5 -mno-rdrnd -mno-rdseed" in CFLAGS if you care.

Oh, actually the original report was very specific: https://lore.kernel.org/all/20251018024010.4112396-1-gourry@...