>One thing that perplexed me when first porting 4.0.92g to NetBSD was
>that HAVE_TIMER_SETTIME seems to imply timeval's everywhere, whereas
>our timer_gettime() and timer_settime() use timespec's. That's why
>detection of those two is disabled in 4.0.92h and above. Now I figure,
>that to get any meaningful use out of the nanosecond stuff, you'd want
>the loop filter to use timeval's too. So my question is, would not
>changing timer_gettime() and timer_settime() to use timeval's break
>existing binaries?

The thing to do in that case is to simply version the old system calls.
Old binaries will continue to use the old system call numbers, but new
binaries will use the newer system calls.

But are we talking about the same thing?

% grep timer_gettime *
syscalls.master: 238     UNIMPL          timer_gettime

--Ken