On Thu, 25 Jun 2026 11:34:28 GMT, Johan Sjölen <[email protected]> wrote:
>> src/java.base/share/classes/java/util/concurrent/LinkedTransferQueue.java
>> line 392:
>>
>>> 390:
>>> 391: final Thread getAcqWaiter() {
>>> 392: return (Thread)WAITER.getAcquire(this);
>>
>> @DougLea Are you OK with this change, or should we make the access even
>> stronger?
>
> This isn't sufficient, it needs a Java volatile load as it needs to be
> ordered with a different variable. This just happens to work with our current
> choice of loadAcquire instruction on ARM64 (as @merykitty has pointed out).
Maybe a full fence is the cleaner expression here? It directly says that
everything before the fence, including the successful exchange, stays before
everything after it, including ordinary loads.
@DougLea is the expert, though :P
-------------
PR Review Comment: https://git.openjdk.org/jdk/pull/31465#discussion_r3474253040
