RFR: 8331539: [REDO] NMT: add/make a mandatory MEMFLAGS argument to family of os::reserve/commit/uncommit memory API [v2]

[Afshin Zafari]

On Fri, 24 May 2024 10:32:17 GMT, Stefan Karlsson <stefank at openjdk.org> wrote:

>> I have already tried to move `ThreadCritical` into the `MemTracker` (in another PR), but it failed. AFAIR, the unmapping/releasing the memory should be in critical section too. The current implementation follows this order: 1) create critical section 2) unmap/release 3) if successful, call MemTracker. The step 2) should be in critical section.
>
> Hmm. os::release_memory also calls `record_virtual_memory_release`, and then this code calls it again with a second ThreadCritical, but then it is called again with `extra_memory`. I still find this addition of `extra_memory` highly dubious.

Some facts: 
- `MemTracker::record_virtual_memory_release()` has no `ThreadCritical` internally and therefore should be called inside a critical section.
- When `os::release_memory()` returns, the `ThreadCritical` that is created there is destroyed and a new one should be created again here.
- Releasing a sub-region that flagged for CDS and is contained in a larger CDS region is ignored at `MemTracker::record_virtual_memory_release()`. It is a valid case due to the way that CDS reserves and/or releases regions.
- This exceptional case is notified to `MemTracker` by passing `true` as `extra_memory`.
- Inside `MemTracker`, the `extra_memory == true` is used in the places where the exceptional case should/would be addressed.

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

PR Review Comment: https://git.openjdk.org/jdk/pull/19343#discussion_r1613352945