Re: [PATCH 6/6] Use one zonelist that is filtered by nodemask

[Nishanth Aravamudan <nacc@us.ibm.com>]

On 09.11.2007 [10:16:07 -0800], Nishanth Aravamudan wrote:
> On 09.11.2007 [09:26:01 -0800], Christoph Lameter wrote:
> > On Fri, 9 Nov 2007, Lee Schermerhorn wrote:
> > 
> > > > On the other hand, if we call alloc_pages() with GFP_THISNODE set, there
> > > > is no nid to base the allocation on, so we "fallback" to numa_node_id()
> > > > [ almost like the nid had been specified as -1 ].
> > > > 
> > > > So I guess this is logical -- but I wonder, do we have any callers of
> > > > alloc_pages(GFP_THISNODE) ? It seems like an odd thing to do, when
> > > > alloc_pages_node() exists?
> > > 
> > > I don't know if we have any current callers that do this, but absent any
> > > documentation specifying otherwise, Mel's implementation matches what
> > > I'd expect the behavior to be if I DID call alloc_pages with 'THISNODE.
> > > However, we could specify that THISNODE is ignored in __alloc_pages()
> > > and recommend the use of alloc_pages_node() passing numa_node_id() as
> > > the nid parameter to achieve the behavior.  This would eliminate the
> > > check for 'THISNODE in __alloc_pages().  Just mask it off before calling
> > > down to __alloc_pages_internal().
> > > 
> > > Does this make sense?
> > 
> > I like consistency. If someone absolutely wants a local page then
> > specifying GFP_THISNODE to __alloc_pages is okay. Leave as is I guess. 
> 
> Fair enough.
> 
> > What happens though if an MPOL_BIND policy is in effect? The node used
> > must then be the nearest node from the policy mask....
> 
> Indeed, this probably needs to be validated... Sigh, more interleaving
> of policies and everything else...

Hrm, more importantly, isn't this an existing issue? Maybe should be
resolved separately from the one zonelist patches.

-Nish

-- 
Nishanth Aravamudan <nacc@us.ibm.com>
IBM Linux Technology Center
-
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/