RFR: 8355769: Optimize nmethod dependency recording [v2]

[Aleksey Shipilev]

On Tue, 29 Apr 2025 09:36:41 GMT, Andrew Dinn <adinn at openjdk.org> wrote:

> I checked the two callers of this method and noticed that `MethodHandles::add_dependent_nmethod` includes `assert_locked_or_safepoint(CodeCache_lock)` while `InstanceKlass::add_dependent_nmethod` has no assert. First, should both not include an assert? Second, why allow for being at a safepoint in the case of a MethodHandle rather than just assert the lock is held?

Yeah, locks are too weak in callers. We can make them stronger, as in new commit.

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

PR Comment: https://git.openjdk.org/jdk/pull/24933#issuecomment-2838689318