On Wed, 3 Oct 2007, Yasunori Goto wrote:

quoted text
> > 
> > That would work. But it would be better to shrink the cache first. The 
> > first 2 slabs on a node may be empty and the shrinking will remove those. 
> > If you do not shrink then the code may falsely assume that there are 
> > objects on the node.
> 
> I'm sorry, but I don't think I understand what you mean... :-(
> Could you explain more? 
> 
> Which slabs should be shrinked? kmem_cache_node and kmem_cache_cpu?

The slab for which you are trying to set the kmem_cache_node pointer to 
NULL needs to be shrunk.
 
quoted text
> I think kmem_cache_cpu should be disabled by cpu hotplug,
> not memory/node hotplug. Basically, cpu should be offlined before
> memory offline on the node.

Hmmm.. Ok for cpu hotplug you could simply disregard the per cpu 
structure if the per cpu slab was flushed first.

However, the per node structure may hold slabs with no objects even after 
all objects were removed on a node. These need to be flushed by calling
kmem_cache_shrink() on the slab cache.

On the other hand: If you can guarantee that they will not be used and 
that no objects are in them and that you can recover the pages used in 
different ways then zapping the per node pointer like that is okay.

-
unsubscribe notice
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/