RFR: 8210832: Remove sneaky locking in class Monitor

[Thomas Schatzl]

Hi,

On Tue, 2019-01-29 at 19:53 +1000, David Holmes wrote:
> Hi Per,
> 
> If I may jump in on one thing you suggest ... destructors. Do we
> ever actually destroy Mutex or Monitor instances? There are inherent

It would be really useful to keep being able to do this: I have been
on-and-off working on creating remembered sets for regions on demand
(where currently for each region's remembered set there is a Mutex
associated to it, allocated at startup at the moment), and free them
when no longer necessary.

Doing so improves memory footprint to avoid having a big "empty"
remembered set around that's actually never used for many regions,
particularly at startup. More flexibility with assigning remembered
sets also allows other changes that improve overall gc performance
quite a bit.

> races that can make it very dangerous to try and actually delete the
> low-level PlatformMonitor and destroy the pthread_mutex or
> pthread_cond, and even release the memory used. The related

Thanks,
  Thomas