In performance testing the ServiceEntryPoint benchmark, which basically just calls the ping command, we see that a good chunk of time is spent in getLockerInfo. We spend 3% of the total time in the benchmark inside this function, mostly downstream of CurOp::_sumBlockedTimeTotal(), but in other areas as well.

See if there are any potential ways to cache info here, or make this algorithmically tighter (do we need to emplace items into a vector then std::sort them?)

Below screenshot is from running hotspot on a locally-derived perf.data. To do this yourself, you must X-forward your Linux workstation.