Throw if JDK-8292158 may cause AES-CTR encryption corruption

[schlosna]

Before this PR

JDK bug https://bugs.openjdk.org/browse/JDK-8292158 could cause corruption of AES-CTR streams when running on CPUs with AVX-512 vectorized AES support (e.g. Intel Ice Lake, as used in AWS 6th generation EC2 instances).

This issue should be resolved in OpenJDK 11.0.18, 15.0.10, 17.0.6, 19.0.2, and 20.0.0 when released.

One can temporarily workaround this issue be adding one of the following sets of JVM arguments:

• -XX:+UnlockDiagnosticVMOptions -XX:-UseAESCTRIntrinsics to disable only AES-CTR intrinsics
• -XX:+UnlockDiagnosticVMOptions -XX:-UseAESIntrinsics to disable all AES intrinsics
• -XX:-UseAES to disable all AES native instructions
• -XX:UseAVX=2 to disable all AVX-512 instructions

See palantir/sls-packaging#1411

After this PR

==COMMIT_MSG==
To avoid potential stream corruption, throw if JDK-8292158 may cause AES-CTR encryption corruption.

Determine if JVM is impacted by JDK-8292158 which can corrupt AES-CTR encryption streams. This bug impacts JDKs up to 11.0.18, 15.0.10, 17.0.6, 19.0.2 and when running on CPUs with AVX-512 vectorized AES support.

See https://bugs.openjdk.org/browse/JDK-8292158 introduced by https://bugs.openjdk.org/browse/JDK-8233741
==COMMIT_MSG==

Possible downsides?

This will fail fast and trigger runtime failures when running on un-mitigated JVMs rather than possible data corruption.

[changelog-app]

Generate changelog in changelog/@unreleased

Type

• Feature
• Improvement
• Fix
• Break
• Deprecation
• Manual task
• Migration

Description

To avoid potential stream corruption, throw if JDK-8292158 may cause AES-CTR encryption corruption.

Determine if JVM is impacted by JDK-8292158 which can corrupt AES-CTR encryption streams. This bug impacts JDKs up to 11.0.18, 15.0.10, 17.0.6, 19.0.2 and when running on CPUs with AVX-512 vectorized AES support.

See https://bugs.openjdk.org/browse/JDK-8292158 introduced by https://bugs.openjdk.org/browse/JDK-8233741

Check the box to generate changelog(s)

• Generate changelog entry

[carterkozak]

Could we run the failing test case instead of attempting to determine whether or not the jvm+hardware combination is impacted?

[carterkozak]

CI tests will be difficult/impossible, but a manual verification should be sufficient.

[schlosna]

Manually tested this on Intel Ice Lake and it identifies and catches the corruption