Re: SLUB regression in current Linus

[Eric Dumazet <eric.dumazet@gmail.com>]

Le mercredi 25 mai 2011 à 09:47 -0500, Christoph Lameter a écrit :
> On Wed, 25 May 2011, James Morris wrote:
> 
> > It turned out the system was still unstable with the attached config
> > (e.g. spontaneous reboot).
> 
> Ahh. Thank you.
> 
> Here is the fix:
> 
> Subject: slub: Fix double bit unlock in debug mode
> 
> Commit 442b06bcea23a01934d3da7ec5898fa154a6cafb added a deactivate_slab()
> in the debug case in __slab_alloc(). deactivate_slab() unlocks the current
> slab used for allocation. Going to the label unlock_out: does it again.
> 
> So simply return the object. In the debug case we do not need all the other
> processing that unlock_out: does.
> 
> Signed-off-by: Christoph Lameter <cl@linux.com>
> 
> ---
>  mm/slub.c |    3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
> 
> Index: linux-2.6/mm/slub.c
> ===================================================================
> --- linux-2.6.orig/mm/slub.c	2011-05-25 09:41:27.000000000 -0500
> +++ linux-2.6/mm/slub.c	2011-05-25 09:41:39.000000000 -0500
> @@ -1884,7 +1884,8 @@ debug:
>  	deactivate_slab(s, c);
>  	c->page = NULL;

is this c->page = NULL; really necessary ?

Thanks !

>  	c->node = NUMA_NO_NODE;
> -	goto unlock_out;
> +	local_irq_restore(flags);
> +	return object;
>  }
> 
>  /*