[3.3.0] Sporadic dilithium_kat_4x4_AES_Deterministic failure about botan

[thesamesam]

Huh, I hit this one before in #3623 (comment).

[randombit]

This is a surprising test to sporadically fail since it's specifically testing that the (completely deterministic) Dilithium key generation process behaves as expected. If it failed, I would expect it to be due to a miscompilation and always fail.

Just did ./botan-test dilithium_kat_4x4_AES_Deterministic --test-runs=1000 and all passed. I'm going to try 10,000 runs now. If I can't repro with that it's probably either an interleaving problem running tests in multiple threads, or a hardware fluke on your machine, and in either case would need further digging.

[reneme]

Last time @thesamesam saw such sporadic test failures, it had to do with tests racing each other on CPUID and (independently) a racy registration of ECC curves.

Given that the failing test deals with AES: Could this again be a CPUID race?

Edit: Perhaps #3860?

[reneme]

@thesamesam Could you try whether this patch helps?