👋 Welcome back zgu! A progress list of the required criteria for merging this PR into master will be added to the body of your pull request. There are additional pull request commands available for use with this pull request.

@zhengyu123 The following label will be automatically applied to this pull request:

• hotspot-runtime

When this pull request is ready to be reviewed, an "RFR" email will be sent to the corresponding mailing list. If you would like to change these labels, use the /label pull request command.

Webrevs

• 02: Full - Incremental (6b430244)
• 01: Full - Incremental (d3cb57b6)
• 00: Full (d23018fe)

Hi Zhengyu,

I do not understand this change. The "counter" is what makes the lock acts as a lock. Locks generally require acquire/release semantics wrt. to the locking and unlocking around a critical section. So how can you switch to use mo_relaxed?

David

To clarify a writeLock needs acquire on lock and release on unlock; the readLock only needs acquire on lock.

I do not understand this change. The "counter" is what makes the lock acts as a lock. Locks generally require acquire/release semantics wrt. to the locking and unlocking around a critical section. So how can you switch to use mo_relaxed?

+1. We cannot do this relaxation here, unless we also add explicit fences. I think the best we can do here is to "relax" mo_conservative (double-sided barrier) to mo_seqcst (which still have the memory ordering semantics we want for locks).

Hi Zhengyu,

I do not understand this change. The "counter" is what makes the lock acts as a lock. Locks generally require acquire/release semantics wrt. to the locking and unlocking around a critical section. So how can you switch to use mo_relaxed?

David

Right, I am embarrassed. This is a lock, does have memory semantics.

Thanks,

-Zhengyu

I'm trying to reason about this but am struggling to understand how this "lock" is actually used. It isn't really a lock at all but a kind of access-latch: anyone can access things in normal operation but when we want to shutdown then we basically put up a closed sign, wait for all current users to leave and then do a reset. There are no critical sections as such and no related memory operations that this "lock" is guarding. So maybe it doesn't need any memory ordering after all. ??

It has been a long time.

I reread the code. Yes, it is a kind of count-down-latch, which allows read-only walk to complete before list can be destroyed.

Now, I suspect that AccessLock is not sufficient, cause it does not guard against insertion.

I would like to withdraw this PR for now.