RFR: 8369190: JavaFrameAnchor on AArch64 appears to be missing barriers [v2]

[Justin King]

On Mon, 6 Oct 2025 15:38:34 GMT, Justin King <jcking at openjdk.org> wrote:

>> src/hotspot/cpu/aarch64/javaFrameAnchor_aarch64.hpp line 43:
>> 
>>> 41:   void clear(void) {
>>> 42:     // Must clear sp first and place a store-store barrier (dmb ISHST) immediately after,
>>> 43:     // to ensure ACGT does not observe a corrupted frame.
>> 
>> What is ACGT that you want to slow down AArch64 HotSpot for it?
>
> `AsyncGetCallTrace`, meant `AGCT`.

Thinking about this, do we "actually" need store barriers at all, so long as the stores are in the correct program order and we put compiler reordering barriers? I assume when the signal is delivered to the thread, all writes are flushed before the signal handler is invoked such that it looks like program order was observed? AGCT is only ever used from the current thread.

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

PR Review Comment: https://git.openjdk.org/jdk/pull/27645#discussion_r2407232782