RFR: 8319896: Remove monitor deflation from final audit [v2]

[Daniel D. Daugherty]

On Thu, 16 Nov 2023 14:47:46 GMT, Stefan Karlsson <stefank at openjdk.org> wrote:

>> In [JDK-8318757](https://bugs.openjdk.org/browse/JDK-8318757) we can see that running a monitor deflation pass during a safepointed operation can interleave with the async monitor deflation, which causes various problems. The fix for that is to stop deflating monitors in the thread dump operation and instead only collect relevant monitors.
>> 
>> There is yet another place where we call monitor deflation from outside of the monitor deflation thread. That place is the "final audit" part, which walks over monitors and performs verification and logging just before we exit the JVM. Before the walk over the list of monitors we perform a monitor deflation pass to prune the system from "uninteresting" monitors.
>> 
>> I propose that we remove the monitor deflation from the final audit, and that we instead only visit "interesting" monitors (those that have an owner or "is busy"). After this change it's only the monitor deflation thread that performs monitor deflation. It is unclear to me if the final audit can actually interleave with the async monitor deflation, but this removal makes it easier to reason around monitor deflation since it is only one thread that is performing it.
>> 
>> This PR is dependent on #16519. I intend to run extended testing on both of these PRs over the weekend.
>
> Stefan Karlsson has updated the pull request with a new target base due to a merge or a rebase. The pull request now contains 18 commits:
> 
>  - Merge remote-tracking branch 'upstream/master' into 8319896_remove_deflation_from_final_audit
>  - 8319896: Remove monitor deflation from final audit
>  - Remove the limit for deflation requests
>  - Remove reinitialization in test
>  - Update comments
>  - Tweak the flag comment a bit
>  - Add AsyncMonitorDeflationForThreadDumpLimit flag
>  - Typos
>  - Remove comment in do_monitors
>  - Make monitors array public static
>  - ... and 8 more: https://git.openjdk.org/jdk/compare/9e7a3ae2...a3df9b5d

I do like the idea of simplifying life by having only the MonitorDeflation thread be
responsible for deflating ObjectMonitors. That does simplify life and makes it much
easier to reason about monitor deflation. When I added the deflation to the
VM_ThreadDump VM operation and to the final thread auditing I was in a mode
where I was trying very hard to clean up all the garbage. While this was a useful
exercise when we were developing async monitor deflation because we were able
to find "lost" ObjectMonitors, it is no longer necessary to be so crazy about it.
The async monitor deflation code has matured for a very long time...

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

PR Comment: https://git.openjdk.org/jdk/pull/16605#issuecomment-1814880350