RFR: 8363620: AArch64: reimplement emit_static_call_stub() [v6]

[Dean Long]

On Sat, 13 Dec 2025 09:40:21 GMT, Andrew Haley <aph at openjdk.org> wrote:

> > The race because of old redefined methods might be more likely.
> 
> Mm, yes. Do you know of a good stress test for that?

8225681 mentions vmTestbase/nsk/jvmti/RedefineClasses/StressRedefine, so I would start there.  However, I suspect that the race is no longer possible, because redefining a class should require a safepoint, and after 8322630 we no longer allow a safepoint between resolving the callee and updating the call site.  But maybe there is still a race before the NoSafepointVerifier in resolve_helper if two threads are trying to resolve the same call site and they allow a safepoint during find_callee_info.

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

PR Comment: https://git.openjdk.org/jdk/pull/26638#issuecomment-3657951520