Re: [PATCH 8/8] i386: bitops: smp_mb__{before, after}_clear_bit() definitions

[Nick Piggin <nickpiggin@yahoo.com.au>]

Satyam Sharma wrote:
> From: Satyam Sharma <ssatyam@cse.iitk.ac.in>
> 
> [8/8] i386: bitops: smp_mb__{before, after}_clear_bit() definitions
> 
>>From Documentation/atomic_ops.txt, those archs that require explicit
> memory barriers around clear_bit() must also implement these two interfaces.
> However, for i386, clear_bit() is a strict, locked, atomic and
> un-reorderable operation and includes an implicit memory barrier already.
> 
> But these two functions have been wrongly defined as "barrier()" which is
> a pointless _compiler optimization_ barrier, and only serves to make gcc
> not do legitimate optimizations that it could have otherwise done.
> 
> So let's make these proper no-ops, because that's exactly what we require
> these to be on the i386 platform.

No. clear_bit is not a compiler barrier on i386, thus smp_mb__before/after
must be.


> 
> Signed-off-by: Satyam Sharma <ssatyam@cse.iitk.ac.in>
> Cc: David Howells <dhowells@redhat.com>
> Cc: Nick Piggin <nickpiggin@yahoo.com.au>
> 
> ---
> 
> [ A similar optimization needs to be done in the atomic.h also.
>   Will submit that patch shortly. ]
> 
>  include/asm-i386/bitops.h |    4 ++--
>  1 files changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/include/asm-i386/bitops.h b/include/asm-i386/bitops.h
> index 4f1fda5..42999eb 100644
> --- a/include/asm-i386/bitops.h
> +++ b/include/asm-i386/bitops.h
> @@ -106,8 +106,8 @@ static inline void __clear_bit(int nr, unsigned long *addr)
>   * Bit operations are already serializing on x86.
>   * These must still be defined here for API completeness.
>   */
> -#define smp_mb__before_clear_bit()	barrier()
> -#define smp_mb__after_clear_bit()	barrier()
> +#define smp_mb__before_clear_bit()	do {} while (0)
> +#define smp_mb__after_clear_bit()	do {} while (0)
>  
>  /**
>   * __change_bit - Toggle a bit in memory
> 


-- 
SUSE Labs, Novell Inc.