RFR: 8263377: Store method handle linkers in the 'non-nmethods' heap

[Jorn Vernee]

On Fri, 3 Jun 2022 17:51:24 GMT, Jorn Vernee <jvernee at openjdk.org> wrote:

>> 8263377: Store method handle linkers in the 'non-nmethods' heap
>
> src/hotspot/share/oops/method.cpp line 1016:
> 
>> 1014:   CodeBlob* nm = code();       // Put it into local variable to guard against concurrent updates
>> 1015:   if (nm != NULL) {
>> 1016:     nm->as_compiled_method()->make_not_entrant();
> 
> Looks like this code is guarding against a situation where 2 threads race to generate a native wrapper.
> 
> I think there is a potential race here with the new method handle linker blobs if 2 threads race to generate a linker, and one ends up overwriting the other as well.
> 
> I think this cast would fail, but we would also have a memory leak if we only made CompiledMethods non-entrant.
> 
> i.e. it seems there can be cases where 2 threads race to generate a MH linker, and both try to install it. We should make sure that there are no leaks in that case.
> 
> I think it might be enough to see here if `nm` is not null, and if it's a CompiledMethod, make it not entrant, and if it's an MH linker, free it directly (can delegate to `RuntimeBlob::free`). Though, I'm not sure if it's okay here to grab the code cache lock for that. Maybe there's a way to mark it as unused, and have it be cleaned up later.

Ah, actually, this code is just trying to prevent the existing native wrapper from being used in the future. Since it embeds the target native function as a constant (https://github.com/openjdk/jdk/blob/master/src/hotspot/cpu/x86/sharedRuntime_x86_64.cpp#L1927)

Forget what I said :) (*sigh* Friday evenings)

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

PR: https://git.openjdk.java.net/jdk/pull/8760