> The get-versions are cheaper, but they only provide a cached timestamp > from basically the last hardclock() event. So they are not really > suitable for > high resolution time stamping - but probably ok for file system time > stamps and probably device drivers.
for fwiw, i think we may want to have an option to use either the real or the get version here. the default probably should be the real one, but some use cases probably can cope with less precision (and thus, less overhead.) this is almost certainly the largest chunk of time taken for any KERNHIST entry addition (there are a couple of slow parts - an atomic_cas and the microtime().) this is a future enhancement not directly related to what Paul is doing here.. .mrg.