Adjust BSD os::thread_cpu_time() implementation:

[bsdkurt]

• For BSD's that have getrusage(RUSAGE_THREAD), prefer its use over pthread_getcpuclockid(3) since it can support both user and system cpu time.
• On OpenBSD, stop using pthread_getcpuclockid(3) and provide new implementation that supports both user and system cpu time and doesn't crash the JVM when called on a thread that has exited.
• Remove unnecessary checks for Bsd::_getcpuclockid != NULL in various methods since the underlying implementation checks for it as well as os::is_thread_cpu_time_supported().

Notes: On OpenBSD:

gmake run-test-only TEST="runtime/jni/terminatedThread/TestTerminatedThread.java"

Crashed the jvm by calling pthread_getcpuclockid on a terminated thread. It appears there isn't a safe way to check if a thread is terminated in the JVM and the pthread_getcpuclockid implementation doesn't work when the caller wants just user time anyway so I coded up a alternate solution.

Making getrusage(RUSAGE_THREAD) preferred for *BSD for the the current thread wasn't strictly necessary but I think makes sense if FreeBSD and NetBSD provide fully functional implementations similar to what I have done.

[bsdkurt]

@battleblow I decided to not make the change to RUSAGE_THREAD and leave it the way it was. NetBSD doesn't have RUSAGE_THREAD, so if you convert the FreeBSD implementation over to KERN_PROC_PID w/threads as well it can be removed then.

Here are the relevant tests for testing:

gmake run-test-only TEST="hotspot/jtreg/runtime/jni/terminatedThread/TestTerminatedThread.java jdk/jfr/event/runtime/TestThreadCpuTimeEvent.java"

okay to merge the current version?

[battleblow]

@battleblow I decided to not make the change to RUSAGE_THREAD and leave it the way it was. NetBSD doesn't have RUSAGE_THREAD, so if you convert the FreeBSD implementation over to KERN_PROC_PID w/threads as well it can be removed then.

Here are the relevant tests for testing:

gmake run-test-only TEST="hotspot/jtreg/runtime/jni/terminatedThread/TestTerminatedThread.java jdk/jfr/event/runtime/TestThreadCpuTimeEvent.java"

okay to merge the current version?

Yes, please go ahead and merge. I think it is reasonable to leave it as is for now. Ideally we'd be able to instrument this and see how often each version is called, but I can't think of an easy way to do it.