On Fri, 17 May 2024 22:31:32 GMT, Leonid Mesnik <lmes...@openjdk.org> wrote:
>> The JvmtiTrace::safe_get_thread_name sometimes crashes when called while >> current thread is in native thread state. >> >> It happens when thread_name is set for tracing from jvmti functions. >> See: >> https://github.com/openjdk/jdk/blob/master/src/hotspot/share/prims/jvmtiEnter.xsl#L649 >> >> >> The setup is called and the thread name is used in tracing before the thread >> transition. There is no good location where this method could be called from >> vm thread_state only. Some functions like raw monitor enter/exit never >> transition in vm state. So sometimes it is needed to call this function from >> native thread state. >> >> The change should affect JVMTI trace mode only (-XX:TraceJVMTI). >> >> Verified by running jvmti/jdi/jdb tests with tracing enabled. > > Leonid Mesnik has updated the pull request incrementally with one additional > commit since the last revision: > > wrong thread state Okay now I get it. Even once the function is made truly safe, we are always calling it from an unsafe state and so will get the default `Thread::name` response. So now, after any transition to the VM the name is read again to get a good value. This seems a good enhancement though I have to wonder if the apparent changing of the thread name in the tracing might cause problems. The tracing really needs to include a unique thread identifier. Thanks src/hotspot/share/prims/jvmtiEventController.cpp line 961: > 959: JvmtiEventControllerPrivate::change_field_watch(jvmtiEvent event_type, > bool added) { > 960: int *count_addr; > 961: Nit: this file doesn't need to be touched. ------------- Marked as reviewed by dholmes (Reviewer). PR Review: https://git.openjdk.org/jdk/pull/19275#pullrequestreview-2065287663 PR Review Comment: https://git.openjdk.org/jdk/pull/19275#discussion_r1606209679
