gettid() system call volume with kernel mode.

I've been doing a bit of profiling of our applications lately using trace-compass and the corresponding `SCHED_INSTRUMENTATION_xx` profiling options.

During periods of heavy thread synchronization, I'm seeing ~1000 calls to `gettid()` over a 10 second period. I understand that knowing the thread ID of the calling task is required mutexs, but I'm looking at possible solutions to save or cache this value, as, if I understand correctly, the TID doesn't change once the process is created.

I have a proposed solution in the linked PR, which used TLS to store the value, when available. However, this approach seems a little bit off, and possible could be done in a better way.