Re: [PATCH 22/35] writeback: trace global dirty page states

[Hugh Dickins]

On Fri, 17 Dec 2010, Wu Fengguang wrote:
> On Mon, Dec 13, 2010 at 10:47:08PM +0800, Wu, Fengguang wrote:
> 
> > +	TP_fast_assign(
> > +		strlcpy(__entry->bdi,
> > +			dev_name(mapping->backing_dev_info->dev), 32);
> > +		__entry->ino			= mapping->host->i_ino;
> 
> I got an oops against the above line on shmem. Can be fixed by the
> below patch, but still not 100% confident..
> 
> Thanks,
> Fengguang
> ---
> Subject: writeback fix dereferencing NULL shmem mapping->host
> Date: Thu Dec 16 22:22:00 CST 2010
> 
> The oops happens when doing "cp /proc/vmstat /dev/shm". It seems to be
> triggered on accessing host->i_ino, since the offset of i_ino is exactly
> 0x50. However I'm afraid the problem is not fully understand
> 
> 1) it's not normal that tmpfs will have mapping->host == NULL
> 
> 2) I tried removing the dereference as the below diff, however it
>    didn't stop the oops. This is very weird.
> 
> TRACE_EVENT balance_dirty_state:
> 
>  	TP_fast_assign(
>  		strlcpy(__entry->bdi,
>  			dev_name(mapping->backing_dev_info->dev), 32);

I believe this line above is actually the problem: you can imagine that
tmpfs leaves backing_dev_info->dev NULL, and dev_name() appears to
access dev->init_name at 64-bit offset 0x50 down struct device.

> -		__entry->ino			= mapping->host->i_ino;
>  		__entry->nr_dirty		= nr_dirty;
>  		__entry->nr_writeback		= nr_writeback;
>  		__entry->nr_unstable		= nr_unstable;
...
> 
> CC: Hugh Dickins <hugh.dickins@xxxxxxxxxxxxx>

I prefer hughd@xxxxxxxxxx, but the tiscali address survived unexpectedly.

> Signed-off-by: Wu Fengguang <fengguang.wu@xxxxxxxxx>
> ---
>  mm/page-writeback.c |    3 +++
>  1 file changed, 3 insertions(+)
> 
> --- linux-next.orig/mm/page-writeback.c	2010-12-17 09:30:11.000000000 +0800
> +++ linux-next/mm/page-writeback.c	2010-12-17 09:31:05.000000000 +0800
> @@ -907,6 +907,9 @@ void balance_dirty_pages_ratelimited_nr(
>  {
>  	struct backing_dev_info *bdi = mapping->backing_dev_info;
>  
> +	if (!mapping_cap_writeback_dirty(mapping))
> +		return;
> +
>  	current->nr_dirtied += nr_pages_dirtied;
>  
>  	if (unlikely(!current->nr_dirtied_pause))

That would not really be the right patch to fix your oops, but it
or something like would be a very sensible patch in its own right:
looking back through old patches I never got around to sending in,
I can see I had a very similar one two years ago, to save wasting
time on dirty page accounting here when it's inappropriate.
Though mine was testing !mapping_cap_account_dirty(mapping).

Hugh
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/