Re: [BUG] SLOB breaks Crypto

[Adrian-Ken Rueegsegger <ken@codelabs.ch>]

Herbert Xu wrote:
> On Tue, May 18, 2010 at 10:17:35AM +0200, Adrian-Ken Rueegsegger wrote:
>> As noted in my other mail [1] it seems like the HMAC tests trigger these
>> errors.
> 
> Thanks for all the detective work!
> 
> I think the problem is this changeset:
> 
> commit 6eb7228421c01ba48a6a88a7a5b3e71cfb70d4a9
> Author: Herbert Xu <herbert@gondor.apana.org.au>
> Date:   Tue Jan 8 17:16:44 2008 +1100
> 
>     [CRYPTO] api: Set default CRYPTO_MINALIGN to unsigned long long
>     
>     Thanks to David Miller for pointing out that the SLAB (or SLOB/SLUB)
>     cache uses the alignment of unsigned long long if the architecture
>     kmalloc/slab alignment macros are not defined.
>     
>     This patch changes the CRYPTO_MINALIGN so that it uses the same default
>     value.
>     
>     Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
> 
> diff --git a/include/linux/crypto.h b/include/linux/crypto.h
> index 0aba104..5e02d1b 100644
> --- a/include/linux/crypto.h
> +++ b/include/linux/crypto.h
> @@ -90,13 +90,11 @@
>  #define CRYPTO_MINALIGN ARCH_KMALLOC_MINALIGN
>  #elif defined(ARCH_SLAB_MINALIGN)
>  #define CRYPTO_MINALIGN ARCH_SLAB_MINALIGN
> +#else
> +#define CRYPTO_MINALIGN __alignof__(unsigned long long)
>  #endif
>  
> -#ifdef CRYPTO_MINALIGN
>  #define CRYPTO_MINALIGN_ATTR __attribute__ ((__aligned__(CRYPTO_MINALIGN)))
> -#else
> -#define CRYPTO_MINALIGN_ATTR
> -#endif
>  
>  struct scatterlist;
>  struct crypto_ablkcipher;
> 
> So evidently the assumption made in this change does not work on
> SLOB since it only guarantees __alignof__(unsigned long).
> 
> I think the simplest fix is to revert this changeset.

When doing the revert it is necessary to either have
ARCH_KMALLOC_MINALIGN defined or explicitly define CRYPTO_MINALIGN in
the case where it is not. Otherwise shash compilation fails because it
needs CRYPTO_MINALIGN.

Regards,
Adrian