[PATCH] x86: simplify sync_test_bit()

[PATCH] x86: simplify sync_test_bit()

[Jan Beulich]

There really is no need for a redundant implementation here, just keep
the alternative name for allowing consumers to use consistent naming.

Signed-off-by: Jan Beulich <jbeulich@novell.com>

--- a/include/asm-x86/sync_bitops.h
+++ b/include/asm-x86/sync_bitops.h
@@ -130,26 +130,7 @@ static inline int sync_test_and_change_b
 	return oldbit;
 }
 
-static __always_inline int sync_constant_test_bit(int nr, const volatile unsigned long *addr)
-{
-	return ((1UL << (nr & 31)) &
-		(((const volatile unsigned int *)addr)[nr >> 5])) != 0;
-}
-
-static inline int sync_var_test_bit(int nr, const volatile unsigned long * addr)
-{
-	int oldbit;
-
-	__asm__ __volatile__("btl %2,%1\n\tsbbl %0,%0"
-			     :"=r" (oldbit)
-			     :"m" (ADDR),"Ir" (nr));
-	return oldbit;
-}
-
-#define sync_test_bit(nr,addr)			\
-	(__builtin_constant_p(nr) ?		\
-	 sync_constant_test_bit((nr),(addr)) :	\
-	 sync_var_test_bit((nr),(addr)))
+#define sync_test_bit test_bit
 
 #undef ADDR
 

Re: [PATCH] x86: simplify sync_test_bit()

[Jeremy Fitzhardinge]

Jan Beulich wrote:
> There really is no need for a redundant implementation here, just keep
> the alternative name for allowing consumers to use consistent naming.
>
> Signed-off-by: Jan Beulich <jbeulich@novell.com>
>
> --- a/include/asm-x86/sync_bitops.h
> +++ b/include/asm-x86/sync_bitops.h
> @@ -130,26 +130,7 @@ static inline int sync_test_and_change_b
>  	return oldbit;
>  }
>  
> -static __always_inline int sync_constant_test_bit(int nr, const volatile unsigned long *addr)
> -{
> -	return ((1UL << (nr & 31)) &
> -		(((const volatile unsigned int *)addr)[nr >> 5])) != 0;
> -}
> -
> -static inline int sync_var_test_bit(int nr, const volatile unsigned long * addr)
> -{
> -	int oldbit;
> -
> -	__asm__ __volatile__("btl %2,%1\n\tsbbl %0,%0"
> -			     :"=r" (oldbit)
> -			     :"m" (ADDR),"Ir" (nr));
> -	return oldbit;
> -}
> -
> -#define sync_test_bit(nr,addr)			\
> -	(__builtin_constant_p(nr) ?		\
> -	 sync_constant_test_bit((nr),(addr)) :	\
> -	 sync_var_test_bit((nr),(addr)))
> +#define sync_test_bit test_bit
Hm,

#define sync_test_bit(nr, addr) test_bit(nr, addr)

would be better, but seems reasonable to me. Or even an inline for 
consistency.

J

Re: [PATCH] x86: simplify sync_test_bit()

[Jan Beulich]

>#define sync_test_bit(nr, addr) test_bit(nr, addr)
>
>would be better, but seems reasonable to me. Or even an inline for 
>consistency.

I'm usually intentionally using just the names, without parameters and
not as inline, in such alias definitions so that in case the name gets used
as a function pointer (arguably unlikely here) there's not going to be
any missing definition or duplicate function instantiation. But from a
functionality point of view, either of the alternatives you suggest is
of course as good.

Jan

Re: [PATCH] x86: simplify sync_test_bit()

[Jeremy Fitzhardinge]

Jan Beulich wrote:
> I'm usually intentionally using just the names, without parameters and
> not as inline, in such alias definitions so that in case the name gets used
> as a function pointer (arguably unlikely here) there's not going to be
> any missing definition or duplicate function instantiation. But from a
> functionality point of view, either of the alternatives you suggest is
> of course as good.

I'm especially wary of the naked #define, since it will affect any 
instance of sync_test_bit:

struct bit_thingy {
	unsigned sync_test_bit:1;
	unsigned other_test_bit:1;
	...
};

J

Re: [PATCH] x86: simplify sync_test_bit()

[Ingo Molnar]

* Jan Beulich <jbeulich@novell.com> wrote:

> There really is no need for a redundant implementation here, just keep 
> the alternative name for allowing consumers to use consistent naming.

> -	__asm__ __volatile__("btl %2,%1\n\tsbbl %0,%0"
> -			     :"=r" (oldbit)
> -			     :"m" (ADDR),"Ir" (nr));

> +#define sync_test_bit test_bit

thanks, applied.

	Ingo

Re: [PATCH] x86: simplify sync_test_bit()

[Jeremy Fitzhardinge]

Ingo Molnar wrote:
> * Jan Beulich <jbeulich@novell.com> wrote:
>
>   
>> There really is no need for a redundant implementation here, just keep 
>> the alternative name for allowing consumers to use consistent naming.
>>     
>
>   
>> -	__asm__ __volatile__("btl %2,%1\n\tsbbl %0,%0"
>> -			     :"=r" (oldbit)
>> -			     :"m" (ADDR),"Ir" (nr));
>>     
>
>   
>> +#define sync_test_bit test_bit
>>     
>
> thanks, applied.

Please use this instead.

    J

Subject: x86: use macro parameters on sync_test_bit

Using a naked parameterless macro could lead to other tokens being
unexpectedly replaced.

Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
---
 include/asm-x86/sync_bitops.h |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

===================================================================
--- a/include/asm-x86/sync_bitops.h
+++ b/include/asm-x86/sync_bitops.h
@@ -123,7 +123,7 @@
 	return oldbit;
 }
 
-#define sync_test_bit test_bit
+#define sync_test_bit(nr, addr) test_bit(nr, addr)
 
 #undef ADDR
 

Re: [PATCH] x86: simplify sync_test_bit()

[Ingo Molnar]

* Jeremy Fitzhardinge <jeremy@goop.org> wrote:

> Subject: x86: use macro parameters on sync_test_bit
>
> Using a naked parameterless macro could lead to other tokens being 
> unexpectedly replaced.

> -#define sync_test_bit test_bit
> +#define sync_test_bit(nr, addr) test_bit(nr, addr)

thanks Jeremy, applied.

	Ingo