RFR: 8346123: [REDO] NMT should not use ThreadCritical [v6]

[Robert Toyonaga]

On Mon, 23 Dec 2024 07:58:32 GMT, Thomas Stuefe <stuefe at openjdk.org> wrote:

>> I don't think this should ever get called during bootstrapping because thread stacks are only accounted lazily when a snapshot is created. I don't think an NMT snapshot would ever get created during bootstrapping, but just in case, I've added a check for `MemTracker::is_bootstrapping_done()`.
>
>> I don't think this should ever get called during bootstrapping because thread stacks are only accounted lazily when a snapshot is created. I don't think an NMT snapshot would ever get created during bootstrapping, but just in case, I've added a check for `MemTracker::is_bootstrapping_done()`.
> 
> 
> It could be called as part of the NMT report error reporting during early JVM initialization => between NMT bootstrap and Mutex system startup. That report is only written though if NMT is enabled, and that requires NMT to have been bootstrapped. Is NMT bootstrap after Mutex system bootstrap? Then we are fine here.
> 
> Man, I start warming up to your original idea of just using a plain OS-side mutex for NMT lockes, @roberttoyonaga . One that always works. Problem is missing deadlock detection.

`MemTracker::initialize();` is called before Mutexes are ready. So it seems like that means this `assert_lock_strong` can be called before Mutexes can be used. So we do need that check for `MemTracker::is_bootstrapping_done()`.

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

PR Review Comment: https://git.openjdk.org/jdk/pull/22745#discussion_r1895808205