RFR: 8247972: incorrect implementation of JVM TI GetObjectMonitorUsage [v3]

[David Holmes]

On Wed, 7 Feb 2024 07:02:11 GMT, Serguei Spitsyn <sspitsyn at openjdk.org> wrote:

>> The implementation of the JVM TI `GetObjectMonitorUsage` does not match the spec.
>> The function returns the following structure:
>> 
>> 
>> typedef struct {
>>     jthread owner;
>>     jint entry_count;
>>     jint waiter_count;
>>     jthread* waiters;
>>     jint notify_waiter_count;
>>     jthread* notify_waiters;
>> } jvmtiMonitorUsage;
>> 
>> 
>> The following four fields are defined this way:
>> 
>> waiter_count  [jint] The number of threads waiting to own this monitor
>> waiters  [jthread*] The waiter_count waiting threads
>> notify_waiter_count  [jint]  The number of threads waiting to be notified by this monitor
>> notify_waiters  [jthread*] The notify_waiter_count threads waiting to be notified
>> 
>> The `waiters` has to include all threads waiting to enter the monitor or to re-enter it in `Object.wait()`.
>> The implementation also includes the threads waiting to be notified in `Object.wait()` which is wrong.
>> The `notify_waiters` has to include all threads waiting to be notified in `Object.wait()`.
>> The implementation also includes the threads waiting to re-enter the monitor in `Object.wait()` which is wrong.
>> This update makes it right.
>> 
>> The implementation of the JDWP command `ObjectReference.MonitorInfo (5)` is based on the JVM TI `GetObjectMonitorInfo()`. This update has a tweak to keep the existing behavior of this command.
>> 
>> The follwoing JVMTI vmTestbase tests are fixed to adopt to the `GetObjectMonitorUsage()` correct behavior:
>> 
>>   jvmti/GetObjectMonitorUsage/objmonusage001
>>   jvmti/GetObjectMonitorUsage/objmonusage003
>> 
>> 
>> The following JVMTI JCK tests have to be fixed to adopt to correct behavior:
>> 
>> vm/jvmti/GetObjectMonitorUsage/gomu001/gomu00101/gomu00101.html
>> vm/jvmti/GetObjectMonitorUsage/gomu001/gomu00101/gomu00101a.html
>> vm/jvmti/GetObjectMonitorUsage/gomu001/gomu00102/gomu00102.html
>> vm/jvmti/GetObjectMonitorUsage/gomu001/gomu00102/gomu00102a.html
>> 
>> 
>> 
>> A JCK bug will be filed and the tests have to be added into the JCK problem list located in the closed repository.
>> 
>> Also, please see and review the related CSR:
>>  [8324677](https://bugs.openjdk.org/browse/JDK-8324677): incorrect implementation of JVM TI GetObjectMonitorUsage
>> 
>> The Release-Note is:
>> [8325314](https://bugs.openjdk.org/browse/JDK-8325314): Release Note: incorrect implementation of JVM TI GetObjectMonitorUsage
>>  
>> Testing:
>>  - tested with mach5 tiers 1-6
>
> Serguei Spitsyn has updated the pull request with a new target base due to a merge or a rebase. The incremental webrev excludes the unrelated changes brought in by the merge/rebase. The pull request contains three additional commits since the last revision:
> 
>  - Merge
>  - review: thread in notify waiter list can't be BLOCKED
>  - 8324677: Specification clarification needed for JVM TI GetObjectMonitorUsage

Based on Dan's analysis I would have to go back and redo the complete analysis for myself. :( I think the only way to make sense of this is to actually set up scenarios where we have different threads contending on entry, different threads waiting and different threads re-entering after being notified, and see what values actually get returned in each case.

I think the `pending_current_monitor` issue may be a distinct/different issue. I can easily imagine that this was overlooked when we introduced the "wait morphing" rather than have the notified/timed-out/interrupted waiters actually place themselves on the entry queue.

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

PR Comment: https://git.openjdk.org/jdk/pull/17680#issuecomment-1933469232