RFR: 8306767: Concurrent repacking of extra data in MethodData is potentially unsafe [v24]

[Erik Österlund]

On Mon, 22 Jan 2024 11:14:41 GMT, Emanuel Peter <epeter at openjdk.org> wrote:

>> As explained in a [comment below](https://github.com/openjdk/jdk/pull/16840#issuecomment-1833529561), we have to ensure that reading/writing/cleaning the extra data all needs to be guarded by the `extra_data_lock`, and that no safepoint should happen while holding that lock, so that the lock is not broken.
>> 
>> I introduced `check_extra_data_locked`, where I check that we hold the lock, and if we are a java thread (only those ever safepoint), that we currently are in a `NoSafepointVerifier` scope, hence we verify that no safepoint will be taken.
>> 
>> I placed `check_extra_data_locked` in all the places where we access the extra data, and then placed locks (with implicit no-safepoint-verifiers) at the call-site of those places.
>> 
>> I also needed to change the rank of `extra_data_lock` to `nosafepoint` and set the `Mutex::_no_safepoint_check_flag` when taking the lock. Otherwise I could not take the lock from a VM thread.
>> 
>> **Testing**
>> Testing: tier1-3 and stress.
>
> Emanuel Peter has updated the pull request incrementally with one additional commit since the last revision:
> 
>   cleanup unnecessary changes

This looks very promising to me. Very nice with the asserts checking the appropriate locking is there when using the extra data section. Great job!

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

Marked as reviewed by eosterlund (Reviewer).

PR Review: https://git.openjdk.org/jdk/pull/16840#pullrequestreview-1841077500