RFR: 8320304: Refactor and simplify monitor deflation functions

[Stefan Karlsson]

On Fri, 17 Nov 2023 16:31:08 GMT, Patricio Chilano Mateo <pchilanomate at openjdk.org> wrote:

>> The recent rewrites to remove safepointed monitor deflation allows us to simplify the monitor deflation code a bit. The code is now guaranteed to be called from the MonitorDeflationThread, which is a JavaThread, and never from the VMThread.
>> 
>> I would like to propose two changes to the code:
>> 1) Remove all code that supports monitor deflation from other threads than the MonitorDeflationThread
>> 2) Extract out the logging code so that it is easier to see the structure of the monitor deflation code
>> 
>> I have run minimal manual of this patch. I'd like to get some feedback on the proposal before I proceed to run more in-depth testing.
>
> src/hotspot/share/runtime/synchronizer.cpp line 1731:
> 
>> 1729:     ThreadBlockInVM tbivm(current);
>> 1730: 
>> 1731:     log.after_blocked();
> 
> This sequence is pre-existent but the check for safepoint/handshakes happens in ~ThreadBlockInVM so there is no point in logging before and after ThreadBlockInVM since we are not going to block. Also with this sequence the timer will still run if we do block in ~ThreadBlockInVM. So I would remove them and move log.end() at the end of this block before the ~ThreadBlockInVM runs since we are done by then. If we want to log when we the deflater actually returns then we can add a log there too.

Good point. This code is going to be rewritten for https://bugs.openjdk.org/browse/JDK-8319137 and https://bugs.openjdk.org/browse/JDK-8319048. The comment here https://github.com/openjdk/jdk/pull/16412#issuecomment-1818358260 is a suggestion for how to solve both issues, and I think it should include the fixes to the logging.

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

PR Review Comment: https://git.openjdk.org/jdk/pull/16706#discussion_r1398877806