[PATCH] discontig.c

[PATCH] discontig.c

[Robert Picco]

Hi David and Andrew:

I found a problem with HP NUMA configuration in 2.6.7-rc3-mm1.  The 
node_online_map is incorrect
because of actions taken by reassign_cpu_only_nodes in 
arch/ia64/mm/discontig.c.  The mm1 patch
which changes the mempolicy for the init_task uses this invalid 
node_online_map and eventually uses a
non-existent node's zonelist when calling __alloc_pages. 

The patch below takes care of this issue.  It puts the non-existent 
nodes offline.

thanks,

Bob

diff -ruN -X /home/picco/losl/dontdiff linux-2.6.7-rc3-m1.orig/arch/ia64/mm/discontig.c linux-2.6.7-rc3-m1.hpet/arch/ia64/mm/discontig.c
--- linux-2.6.7-rc3-m1.orig/arch/ia64/mm/discontig.c	2004-06-07 15:14:58.000000000 -0400
+++ linux-2.6.7-rc3-m1.hpet/arch/ia64/mm/discontig.c	2004-06-09 18:44:02.000000000 -0400
@@ -154,6 +154,9 @@
 
 	memcpy(numa_slit, numa_slit_fix, sizeof (numa_slit));
 
+	for (i = nnode; i < numnodes; i++)
+		node_set_offline(i);
+
 	numnodes = nnode;
 
 	return;

Re: [PATCH] discontig.c

[David Mosberger]

>>>>> On Thu, 10 Jun 2004 11:23:28 -0400, Robert Picco <Robert.Picco@hp.com> said:

  Rob> Hi David and Andrew: I found a problem with HP NUMA
  Rob> configuration in 2.6.7-rc3-mm1.  The node_online_map is
  Rob> incorrect because of actions taken by reassign_cpu_only_nodes
  Rob> in arch/ia64/mm/discontig.c.  The mm1 patch which changes the
  Rob> mempolicy for the init_task uses this invalid node_online_map
  Rob> and eventually uses a non-existent node's zonelist when calling
  Rob> __alloc_pages.

The patch looks fine to me.  If there are no objections from others, I
shall be happy to apply it in my tree, if that's what you want me to
do.

	--david

Re: [PATCH] discontig.c

[Jesse Barnes]

On Thursday, June 10, 2004 11:10 am, David Mosberger wrote:
> >>>>> On Thu, 10 Jun 2004 11:23:28 -0400, Robert Picco
> >>>>> <Robert.Picco@hp.com> said:
>
>   Rob> Hi David and Andrew: I found a problem with HP NUMA
>   Rob> configuration in 2.6.7-rc3-mm1.  The node_online_map is
>   Rob> incorrect because of actions taken by reassign_cpu_only_nodes
>   Rob> in arch/ia64/mm/discontig.c.  The mm1 patch which changes the
>   Rob> mempolicy for the init_task uses this invalid node_online_map
>   Rob> and eventually uses a non-existent node's zonelist when calling
>   Rob> __alloc_pages.
>
> The patch looks fine to me.  If there are no objections from others, I
> shall be happy to apply it in my tree, if that's what you want me to
> do.

The node_online bitmap is already initialized to 0, isn't it?  Oh no, I see.  
arch_numa_init sets all the available nodes online, but then 
reassign_cpu_nodes truncates that number.  The patch looks ok to me.

Jesse