Subject: Re: poll(2)
To: Chris G Demetriou <>
From: Jonathan Stone <jonathan@DSG.Stanford.EDU>
List: tech-kern
Date: 12/09/1996 17:53:54
[alpha references snipped; the 2^10-Hz alpha clock-tick "hz" was why
 I referred to millisecond accuracy]

Quoting cgd:

>Granted, that doesn't change what you're saying re: scheduling, but
>when you add in the fact that the Alpha uses a 1024Hz real time clock
>interrupt rate, which could be easily increased, well, you're still in
>microseconds for scheduling, but are getting closer.  It's just a
>matter of time.  8-)

Uh, ``of course.''. That's why i also said:

>Clearly we should add a nano-second resolution poll interface.
>Once we do so, regardless of the actualy in-kernel resolution,
>poll(2) and upoll(3) become poll(3) and upoll(3).

(the "y" is a typo.)

In case it's not clear, the reason for adding nanosecond resolution is
that NetBSD has acquired POSIX .12(?)-compatible syscalls with nanosecond
resolution, and if we're going to change poll(2) to higher resolution,
I'm saying move it to nanoseconds _now_, for consistency with other
nanosecond-granularity syscalls, even if the internal implementation
doesn't offer better than millisecond (or 100s of microsecond)
resolution for some time to come.