Re: 2.6.23-rc6-mm1: BUG kmalloc-16: Object padding overwritten (sysfs?)

[Alexey Dobriyan <adobriyan@sw.ru>]

On Thu, Sep 20, 2007 at 10:36:13AM -0700, Christoph Lameter wrote:
> On Thu, 20 Sep 2007, Alexey Dobriyan wrote:
> > The winner is slub-avoid-touching-page-struct-when-freeing-to-per-cpu-slab.patch
> > Blind bisecting pointed to it and reverting the patch from full -mm makes
> > the problem go away
> 
> Hmmm.. This means likely that the c->node is used somewhere for 
> indexing.... Ahhh... If we count objects for sysfs output then c->node may 
> be used to index into the statistics array. The offset from the poison 
> also makes sense now since we increment values there.
> 
> Does this patch fix the issue?

Yes, it does.

> SLUB: Fix slab object counting.
> 
> We can only use the node value of the per cpu structure for counting if it 
> is positive. A negative value indicates that the slab is not valid.
> 
> Signed-off-by: Christoph Lameter <clameter@sgi.com>
> 
> ---
>  mm/slub.c |    8 ++++++--
>  1 file changed, 6 insertions(+), 2 deletions(-)
> 
> Index: linux-2.6.23-rc6-mm1/mm/slub.c
> ===================================================================
> --- linux-2.6.23-rc6-mm1.orig/mm/slub.c	2007-09-20 10:31:04.000000000 -0700
> +++ linux-2.6.23-rc6-mm1/mm/slub.c	2007-09-20 10:32:19.000000000 -0700
> @@ -3412,12 +3412,16 @@ static unsigned long slab_objects(struct
>  
>  	for_each_possible_cpu(cpu) {
>  		struct page *page;
> +		int node;
>  		struct kmem_cache_cpu *c = get_cpu_slab(s, cpu);
>  
>  		if (!c)
>  			continue;
>  
>  		page = c->page;
> +		node = c->node;
> +		if (node < 0)
> +			continue;
>  		if (page) {
>  			if (flags & SO_CPU) {
>  				int x = 0;
> @@ -3427,9 +3431,9 @@ static unsigned long slab_objects(struct
>  				else
>  					x = 1;
>  				total += x;
> -				nodes[c->node] += x;
> +				nodes[node] += x;
>  			}
> -			per_cpu[c->node]++;
> +			per_cpu[node]++;
>  		}
>  	}
>