RFR: 8346123: [REDO] NMT should not use ThreadCritical [v15]

[Robert Toyonaga]

On Mon, 13 Jan 2025 07:40:10 GMT, Thomas Stuefe <stuefe at openjdk.org> wrote:

>> Robert Toyonaga has updated the pull request incrementally with one additional commit since the last revision:
>> 
>>   small fix to is_single_threaded and whitespace
>
> Do we still think using hotspot Mutexes is a good goal? As opposed to Roberts's original attempt of using a non-asserting platform mutex? What do people think? I myself am not so sure anymore.
> 
> We get deadlock prevention but also get brittle at initialization.

@tstuefe 
> Do we still think using hotspot Mutexes is a good goal? As opposed to Roberts's original attempt of using a non-asserting platform mutex? What do people think? I myself am not so sure anymore.
We get deadlock prevention but also get brittle at initialization.

Being brittle at initialization is unfortunate, and there is also the point @kimbarrett mentions about this [only working with VM threads](https://github.com/openjdk/jdk/pull/22745#discussion_r1894657610). I am not exactly sure what the best choice is here, but I do think deadlock detection is very nice since the NMT locking is quite broad in scope. I believe, in most cases, we lock around both the NMT accounting and the actual memory operation.

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

PR Comment: https://git.openjdk.org/jdk/pull/22745#issuecomment-2587616074