Cost vs benefit?

I'd have thought that it would be better, for tracepoint functions, to require that they don't clobber registers, even if that isn't the usual convention on some particular arch? If the tracepoints are almost always not active, it doesn't make much sense to have the inactive tracepoint implemented as (save registers, nop, restore registers). Well, I guess in the inactive case you could replace the save/restore with nop's as well, but that still makes a much longer sequence than necessary.