Re: [Linux-ia64] [patch] 2.4 timer_interrupt/gettimeoffset machvec

[David Mosberger <davidm@napali.hpl.hp.com>]

>>>>> On Tue, 8 Apr 2003 14:49:25 -0700, Jesse Barnes <jbarnes@sgi.com> said:

  >> The motivation for changing timer_interrupt() seems suspect to me.
  >> The timer interupt is generated PER CPU, so there is no drift issue at
  >> all (in fact, we intentionally _skew_ things at boot time to reduce
  >> the likelihood of getting all timer interrupts at the same time).

  Jesse> The problem is that we want high resolution gettimeofday, which means
  Jesse> that we need to know how long its been since wall time was updated
  Jesse> when gettimeoffset is called.  We do that right now by snapshoting the
  Jesse> RTC in timer_interrupt...

Clearly you need to establish the relationship between the external
clock and time-of-day somewhere, but I'm not so sure this should be
done in the arch-specific timer interrupt handler.  I suspect you
really want to do it where the time-of-day gets updated.  Also, I
think this should be treated much more as a driver issue rather than a
platform-issue (suppose someone plugged in an adapter card providing a
low-latency, atomic accurracy & high precision lock, you'd presumably
want to be able to use that card in favor of whatever other hardware
might be there.

	--david