RFR: 8284734: Undiscovered references if using ReferentBasedDiscovery

[Kim Barrett]

On Mon, 18 Apr 2022 09:02:07 GMT, Kim Barrett <kbarrett at openjdk.org> wrote:

>> Small change of fixing a crash if using `ReferentBasedDiscovery`.
>> 
>> Test: hotspot_gc
>
> src/hotspot/share/gc/shared/referenceProcessor.cpp line 1008:
> 
>> 1006:     //  2. CM restarts after mark-stack overflow.
>> 1007:     assert(UseG1GC, "inv");
>> 1008:     return true;
> 
> This whole block that is checking discovered is, for non-G1 is really just an assert that discovered is nullptr.  It would be better to entirely avoid the fetch of discovered here for other collectors in non-debug builds.  Maybe this block ought to instead be something like
> 
>   if (UseG1GC) {
>     // G1 might process (and so attempt to discover) a reference multiple
>     // times, due to CM restart after mark-stack overflow.
>     const oop discovered = ...;
>     if (discovered != nullptr) {
>       assert(oopDesc::is_oop(discovered), ...);
>       return true;
>     }
>   } else {
>     assert(java_lang_ref_Reference::discovered(obj) == nullptr, ...);
>   }

Can the UseG1GC test instead be discovery_is_concurrent()?  The old comments talk about concurrent discovery.  But can a G1 young-only collection during a concurrent cycle attempt to discover a reference that is already in the CM reference processor?  I'm not sure, and the discovery policy might be relevant to the answer.

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

PR: https://git.openjdk.java.net/jdk/pull/8196