Re: rhashtable: Use __vmalloc with GFP_ATOMIC for table allocation

[Herbert Xu]

On Fri, Dec 04, 2015 at 04:53:34PM -0500, David Miller wrote:
> From: Herbert Xu <herbert@xxxxxxxxxxxxxxxxxxx>
> Date: Fri, 4 Dec 2015 22:39:56 +0800
> 
> > When an rhashtable user pounds rhashtable hard with back-to-back
> > insertions we may end up growing the table in GFP_ATOMIC context.
> > Unfortunately when the table reaches a certain size this often
> > fails because we don't have enough physically contiguous pages
> > to hold the new table.
> > 
> > Eric Dumazet suggested (and in fact wrote this patch) using
> > __vmalloc instead which can be used in GFP_ATOMIC context.
> > 
> > Reported-by: Phil Sutter <phil@xxxxxx>
> > Suggested-by: Eric Dumazet <eric.dumazet@xxxxxxxxx>
> > Signed-off-by: Herbert Xu <herbert@xxxxxxxxxxxxxxxxxxx>
> 
> Applied, thanks Herbert.

Sorry Dave but you'll have to revert this because I've been able
to trigger the following crash with the patch:

Testing concurrent rhashtable access from 50 threads
------------[ cut here ]------------
kernel BUG at ../mm/vmalloc.c:1337!
invalid opcode: 0000 [#1] PREEMPT SMP 

The reason is that because I was testing insertions with BH disabled,
and __vmalloc doesn't like that, even with GFP_ATOMIC.  As we
obviously want to continue to support rhashtable users inserting
entries with BH disabled, we'll have to look for an alternate
solution.

Thanks,
-- 
Email: Herbert Xu <herbert@xxxxxxxxxxxxxxxxxxx>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt
--
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/