Re: BUG: Bad rss-counter state mm:ffff8800c5a96000 idx:3 val:3894

[Jeremiah Mahler <jmmahler@gmail.com>]

Andrew, Michal,

On Tue, Dec 29, 2015 at 10:30:37AM -0800, Andrew Morton wrote:
> On Tue, 29 Dec 2015 17:23:47 +0100 Michal Hocko <mhocko@kernel.org> wrote:
> 
> > [CCing Andrew]
> > 
> > On Thu 24-12-15 09:12:53, Jeremiah Mahler wrote:
> > > all,
> > > 
> > > I have started seeing a "Bad rss-counter" message in the logs with
> > > the latest linux-next 20151222+.
> > > 
> > >   [  458.282192] BUG: Bad rss-counter state mm:ffff8800c5a96000 idx:3 val:3894
> > 
> > This is MM_SHMEMPAGES so an "unamapped" shmem memory. One possible
> > reason might be an unitialized zap_details used from unmap_mapping_range
> > during truncate introduced by "mm, oom: introduce oom reaper" from the
> > mmotm tree. There is a fix for this which is still pending
> > http://lkml.kernel.org/r/1450487091-7822-1-git-send-email-sasha.levin%40oracle.com
> 
> That won't be getting fixed until linux-next returns from holidays.
> 
> This:
> 
> --- a/mm/memory.c~mm-oom-introduce-oom-reaper-fix-5-fix
> +++ a/mm/memory.c
> @@ -2415,7 +2415,7 @@ static inline void unmap_mapping_range_t
>  void unmap_mapping_range(struct address_space *mapping,
>  		loff_t const holebegin, loff_t const holelen, int even_cows)
>  {
> -	struct zap_details details;
> +	struct zap_details details = { };
>  	pgoff_t hba = holebegin >> PAGE_SHIFT;
>  	pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
>  
> _
> 

I tested both the patch that Michal mentioned as well as the change that
Andrew provided.  The both appear to fix the problem on my machine.

Thanks for the help :-)

-- 
- Jeremiah Mahler