Re: [PATCH mmotm] memcg: fix mem_cgroup_update_mapped_file_stat oops

[KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>]

On Wed, 29 Apr 2009 22:13:33 +0100 (BST)
Hugh Dickins <hugh@veritas.com> wrote:

> CONFIG_SPARSEMEM=y CONFIG_CGROUP_MEM_RES_CTLR=y cgroup_disable=memory
> bootup is oopsing in mem_cgroup_update_mapped_file_stat().  !SPARSEMEM
> is fine because its lookup_page_cgroup() contains an explicit check for
> NULL node_page_cgroup, but the SPARSEMEM version was missing a check for
> NULL section->page_cgroup.
> 
Ouch, it's curious this bug alive now.. thank you.

Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>

I think this patch itself is sane but.. Balbir, could you see "caller" ?
It seems strange.

> Signed-off-by: Hugh Dickins <hugh@veritas.com>
> ---
> Should go in as a fix to
> memcg-add-file-based-rss-accounting.patch
> but it's curious that's the first thing to suffer from this divergence.
> 
> Perhaps this is the wrong fix, and there should be an explicit
> mem_cgroup_disable() check somewhere else; but it would then seem
> dangerous that SPARSEMEM and !SPARSEMEM diverge in this way,
> and there are lots of lookup_page_cgroup NULL tests around.
> 
>  mm/page_cgroup.c |    2 ++
>  1 file changed, 2 insertions(+)
> 
> --- 2.6.30-rc3-mm1/mm/page_cgroup.c	2009-04-29 21:01:06.000000000 +0100
> +++ mmotm/mm/page_cgroup.c	2009-04-29 21:12:04.000000000 +0100
> @@ -99,6 +99,8 @@ struct page_cgroup *lookup_page_cgroup(s
>  	unsigned long pfn = page_to_pfn(page);
>  	struct mem_section *section = __pfn_to_section(pfn);
>  
> +	if (!section->page_cgroup)
> +		return NULL;
>  	return section->page_cgroup + pfn;
>  }
>  
>