On Tue, 25 Jul 2023 21:48:24 GMT, Paul Hohensee <p...@openjdk.org> wrote:
> MonitoringSupport_lock is initialized only when UseG1GC is true, but > [JDK-8304074](https://bugs.openjdk.org/browse/JDK-8304074) uses it to > implement getTotalThreadAllocatedBytes, which is available for all garbage > collectors. While the current code sets UseG1GC regardless of which collector > is specified, see FLAG_SET_ERGO_IF_DEFAULT(UseG1GC, true) in gcConfig.cpp, if > G1 isn't included in the Hotspot build or Hotspot is not running on a server > class machine (unlikely these days), the lock will not be initialized. The > lock's initialization should be unconditional. > > I updated ThreadAllocatedMemory.java to run the test using both G1 and Serial > collectors. So... this does not visibly break right now, because the `MutexLocker` in `jmm_GetTotalThreadAllocatedMemory` silently accepts `nullptr` as the lock? I think we should `assert(MonitoringSupport_lock != nullptr)` there. This should also turn the new `UseSerialGC` test config into proper regression test: the test should fail on that assert without this patch. ------------- PR Comment: https://git.openjdk.org/jdk/pull/15028#issuecomment-1651533398
