Date:	Wed, 15 Aug 2007 16:48:13 -0400
From:	Lee Schermerhorn <Lee.Schermerhorn@...com>
To:	Christoph Lameter <clameter@....com>
Cc:	"Serge E. Hallyn" <serge@...lyn.com>,
	Dhaval Giani <dhaval@...ux.vnet.ibm.com>, bob.picco@...com,
	nacc@...ibm.com, kamezawa.hiroyu@...fujitsu.com, mel@...net.ie,
	akpm@...ux-foundation.org, Balbir Singh <balbir@...ibm.com>,
	Srivatsa Vaddagiri <vatsa@...ux.vnet.ibm.com>,
	lkml <linux-kernel@...r.kernel.org>,
	ckrm-tech <ckrm-tech@...ts.sourceforge.net>
Subject: Re: Regression in 2.6.23-rc2-mm2, mounting cpusets causes a hang

On Wed, 2007-08-15 at 13:36 -0700, Christoph Lameter wrote:
> On Wed, 15 Aug 2007, Lee Schermerhorn wrote:
> 
> > > So its always true for node 0. The "bit" is set.
> > 
> > The issue is with the N_*_MEMORY masks.  They don't get initialized
> > properly because node_set_state() is a no-op if !NUMA.  So, where we
> > look for intersections with or where we AND with the N_*_MEMORY masks we
> > get the empty set.
> 
> That is intentional. Memory is always present if you are on !NUMA. You can 
> simply use a default nodemask where only node 0 is set. That is what the 
> fallback provides. Maybe it does not provide the right thing for cpusets?
> 
> > > We are trying to get cpusets to work with !NUMA?
> > 
> > Well, yes.  In Serge's case, he's trying to use cpusets with !NUMA.
> > He'll have to comment on the reasons for that.  Looking at all of the
> > #ifdefs and init/Kconfig, CPUSET does not depend on NUMA--only SMP and
> > CONTAINERS [altho' methinks CPUSET should select CONTAINERS rather than
> > depend on it...].  So, you can use cpusets to partition of cpus in
> > non-NUMA configs.
> 
> Looks like we need to fix cpuset nodemasks for the !NUMA case then?
> It cannot expect to find valid nodemasks if !NUMA.

Well, OK.  But Paul really hates #ifdefs in kernel/cpusets.c.  He's
asked me to remove them before, so I avoided them here.  Cpusets really
should use only nodes with memory--i.e., the N_HIGH_MEMORY state.  

> 
> > In the more general case, tho', I'm looking at all uses of the
> > node_online_map and for_each_online_node, for instances where they
> > should be replaced with one of the *_MEMORY masks.  IMO, generic code
> > that is compiled independent of any CONFIG option, like NUMA,  should
> > just work, independent of the config.  Currently, as Serge has shown,
> 
> AFAIK this works except for cpusets.

So far.  I'm replacing other usage of node_online_map with the
N_HIGH_MEMORY mask, as we discussed.  I should have that patch ready to
post tomorrow.

> 
> > this is not the case.  So, I think we should fix the *_MEMORY maps to be
> > correctly populated in both the NUMA and !NUMA cases.  A couple of
> > options:
> 
> There is no point in having a variable if you know the results because of 
> !NUMA. That is the way nodemask.h has always operated.

But, the mask--the N_HIGH_MEMORY array element, that is--is there for
both NUMA and !NUMA [== N_NORMAL_MEMORY for !CONFIG_HIGHMEM].  We just
don't initialize it for the !NUMA case, currently.  

> 
> > Thoughts?
> 
> Lets get either rid of the definitions for the nodemasks in the !NUMA 
> case or fix their contents to have the right constant value expected in 
> cpusets.

That's what the patch I posted today [option 2] does--statically
initializes the N_*_MEMORY and N_CPU masks to indicate that node 0
exists.  Serge and Dhaval have tested it on their platform and it solves
the cpuset mount problem.

Lee

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

Powered by blists - more mailing lists