On Mon, 24 Sep 2007 12:42:15 +0200 Mike Galbraith <efault@gmx.de> wrote:

quoted text
> On Mon, 2007-09-24 at 12:24 +0200, Peter Zijlstra wrote:
> 
> > how about something like:
> > 
> >  s64 delta = (s64)(vruntime - min_vruntime);
> >  if (delta > 0)
> >   min_vruntime += delta;
> > 
> > That would rid us of most of the funny conditionals there.
> 
> That still left me with negative min_vruntimes.  The pinned hogs didn't
> lock my box up, but I quickly got the below, so hastily killed it.
> 
> se.wait_max              :             7.846949
> se.wait_max              :           301.951601
> se.wait_max              :             7.071359
>

Odd, the idea (which I think is clear) is that min_vruntime can wrap
around the u64 spectrum. And by using min_vruntime as offset to base
the key around, we get a signed but limited range key-space. (because
we update min_vruntime to be the leftmost task (in a monotonic fashion))

So I'm having trouble with these patches, that is, both your wrap
around condition of:

  if (likely(new_rq->cfs.min_vruntime))

as well as the last patchlet:

  if (((s64)vruntime > (s64)min_vruntime) ||

in that neither of these changes make sense in what its trying to do.

Its perfectly valid for min_vruntime to exist in 1ULL << 63.

-
unsubscribe notice
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/