Re: sched_clock() uses are broken

[Mike Galbraith <efault@gmx.de>]

On Sun, 2006-05-07 at 17:53 +0100, Russell King wrote:
> On Sun, May 07, 2006 at 11:30:15PM +1000, Nick Piggin wrote:
> > Russell King wrote:
> > >On Sun, May 07, 2006 at 11:00:29PM +1000, Nick Piggin wrote:
> > >
> > >>Nick Piggin wrote:
> > >>
> > >>
> > >>>I stand by my first reply to your comment WRT the API.
> > >>
> > >>Actually, on rereading, it seems like I was a bit confused about
> > >>your proposal. I don't think you specified anyway the units
> > >>returned by your new sched_clock(). So it is identical to my
> > >>"corrected" interface :\
> > >
> > >
> > >Okay, so that presumably means we have to either stick with what we
> > >currently have, or go the whole hog and re-implement the sched_clock()
> > >support?
> > >
> > >IOW, my patch on 2nd May isn't of any use as it currently stands?
> > 
> > IMO it would probably be best to try to re implement it in one go.
> > It shouldn't have spread too far out of kernel/sched.c, and the arch
> > code should mostly be implementable in terms of their sched_clock().
> > Mundane but not difficult.
> 
> Having looked at this several times over the last couple of days, I've
> come to the conclusion that I'm not the right person to fix this problem.
> I've tried several methods of converting the code, but every time I
> remain unconvinced that the changes are provably correct as far as not
> missing something, so I end up throwing the changes away and starting
> again.
> 
> Yes, I admit defeat.

Ah, you feel my pain.  I look forward to the continuous nanosecond clock
that the time guys are currently talking about, and will hopefully _not_
decide is not needed.

	-Mike