Re: [PATCH] Apply memory policies to top two highest zones when highest zone is ZONE_MOVABLE

[Mel Gorman]

On (07/08/07 11:14), Andrew Morton didst pronounce:
> On Tue, 7 Aug 2007 17:55:47 +0100 mel@xxxxxxxxx (Mel Gorman) wrote:
> 
> > On (06/08/07 22:12), Andrew Morton didst pronounce:
> > > On Mon, 6 Aug 2007 22:55:41 +0100 mel@xxxxxxxxx (Mel Gorman) wrote:
> > > 
> > > > On (06/08/07 22:31), Andi Kleen didst pronounce:
> > > > > 
> > > > > > If correct, I would suggest merging the horrible hack for .23 then taking
> > > > > > it out when we merge "grouping pages by mobility".  But what if we don't do
> > > > > > that merge?
> > > > > 
> > > > > Or disable ZONE_MOVABLE until it is usable?
> > > > 
> > > > It's usable now. The issue with policies only occurs if the user specifies
> > > > kernelcore= or movablecore= on the command-line. Your language suggests
> > > > that you believe policies are not applied when ZONE_MOVABLE is configured
> > > > at build-time.
> > > 
> > > So..  the problem which we're fixing here is only present when someone
> > > use kernelcore=.  This is in fact an argument for _not_ merging the
> > > horrible-hack.
> > > 
> > 
> > It's even more constrained than that. It only applies to the MPOL_BIND
> > policy when kernelcore= is specified. The other policies work the same
> > as they ever did.
> 
> so.. should we forget about merging the horrible-hack?

Despite a fairly specific case, I'd still like to get the problem fixed
for 2.6.23. I've posted up an alternative fix under the subject "Use one
zonelist per node instead of multiple zonelists v2". It's a more invasive
fix although arguably it's better overall than the hack because it's not
dealing with a specific special case and has a sensible path forward that
makes policies a saner ultimately. However, unlike the hack it affects all
callers of the page allocator so lets see what the reaction from reviewers
is before forgetting about the hack altogether.

-- 
Mel Gorman
Part-time Phd Student                          Linux Technology Center
University of Limerick                         IBM Dublin Software Lab
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/