[foreign-memaccess] RFR: JDK-8252757: Add support for shared segments [v2]

[Maurizio Cimadamore]

On Fri, 4 Sep 2020 18:40:51 GMT, Paul Sandoz <psandoz at openjdk.org> wrote:

> his is about integrity, as long as the developer suitably manages access + close races it should not be, but perhaps we
> should be on the lookout for such bugs just in case?

Yes - there is a possibility (especially in interpreted execution) that we might safepoint _after_ memory access, thus
issuing an exception even though it looks the access was all right. But from the perspective of the system, this is
still a close vs. access race, the effects of which are non-deterministic - so I think that it's acceptable to fail in
that case (after all, this is work is mostly to avoid hard VM crashes - the user should still add relevant
synchronization logic to avoid races!).

>  otherwise i don't think the unconditional deoptimization will fly.

I'm told that, after deoptimization, we jump straight back into compiled code (`DeoptAction=none`).

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

PR: https://git.openjdk.java.net/panama-foreign/pull/304