Re: [RFC/Patch](memory hotplug) fix null pointer access of kmem_cache_node after memory hotplug

[Christoph Lameter <clameter@sgi.com>]

On Wed, 19 Sep 2007, Yasunori Goto wrote:

> build_zonelist() is called very early stage of bootstrap, But it is
> called final stage of hot-add.
> When build_zonelist() is called at hot-add, all kernel module can
> use new memory of the node. So, I'm afraid like following worst case.
> 
>    build_zonelist()              
>         :                     new_nodes_page = new_slab();
>         :                         :
>         :                         :
>         :                     discard_slab(new_nodes_page)
>         :                         (access kmem_cache_node)
>         :
>    kmem_cache_node setting,

So we cannot do this without holding off other kernel accesses since it is 
not serialized like bootstrap. Sigh.
 
> > > I think this "delay creation" fix is better way than it.
> > 
> > Looks like this is a way to on demand node structure creation?
> 
> Yes.

Could be useful in general if you can make that work reliably. We can just 
start out with a single per node structure for the boot node and then add 
others on demand?

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>