RFR: 8335091: NMT: VMATree reserve_mapping and commit_mapping APIs need MEMFLAGS while un/-committing API has no MEMFLAGS arg [v8]

[Afshin Zafari]

On Wed, 25 Sep 2024 06:53:48 GMT, Johan Sjölen <jsjolen at openjdk.org> wrote:

>> src/hotspot/share/nmt/vmatree.hpp line 184:
>> 
>>> 182:   SummaryDiff commit_mapping(position from, position sz, const RegionData& metadata, bool copy_flag = false) {
>>> 183:     return register_mapping(from, from + sz, StateType::Committed, metadata, copy_flag);
>>> 184:   }
>> 
>> Why do we need this API? It's only used in the test so far, so do we really need it?
>> 
>> I am not too crazy about `register_mapping` basically being a wrapper around `register_mapping(.., true)`
>> 
>> What if someone calls `commit_mapping(.., true)` ? Doesn't that turn into `register_mapping(..., StateType::Committed, metadata, true)` which is the same as `uncommit_mapping()` ? It doesn't make sense?
>
> I think it makes sense to have these short definitions that just expands to the right incantation to `register_mapping`, otherwise it would just have to happen at the place where the caller uses `VMATree`. We will need all of the APIs very soon, with the integration of Afshin's port of VirtualMemoryTracker to VMATree.
> 
>>What if someone calls commit_mapping(.., true) ? Doesn't that turn into register_mapping(..., StateType::Committed, metadata, true) which is the same as uncommit_mapping() ? It doesn't make sense?
> 
> That's true, it shoudn't be `StateType::Committed` in `uncommit_mapping`, it should be reserved!!

`StateType::Reserved` is used instead.

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

PR Review Comment: https://git.openjdk.org/jdk/pull/20330#discussion_r1775135408