[crac] RFR: 8355974: [CRaC] Move CPUFeatures verification to the parent process of JVM [v11]

[Jan Kratochvil]

On Tue, 13 May 2025 18:29:16 GMT, Timofei Pushkin <tpushkin at openjdk.org> wrote:

>> I realized the above won't work with the current restore code: if we won't record the features because the user asked us to ignore them but the platform and engine supports them then the restore code will error because the features are not present.
>> 
>> But without that the UX is not great and I don't really understand how `CPUFeatures=ignore` is supposed to work (it is not restore-settable but it is not used on checkpoint).
>> 
>> So maybe it's better to change the restore code to warn instead of erroring when the features are not present (and the warning should ideally be silenceable by `CPUFeatures=ignore` or `IgnoreCPUFeatures`).
>
> This can be postponed until the `CPUFeatures=ignore`/`IgnoreCPUFeatures` cleanup.

I have applied the patch as it makes sense. You are right it cannot be currently used as it depends on `-XX:CPUFeatures=ignore` but that is not restore-settable (only `-XX:+IgnoreCPUFeatures` is but that is a different flag). That should be fixed by ZULU-75248.

-------------

PR Review Comment: https://git.openjdk.org/crac/pull/227#discussion_r2092969611