[PATCH] Define smp_call_function_mask() on uniprocessor builds

[PATCH] Define smp_call_function_mask() on uniprocessor builds

[Avi Kivity]

Rather than #ifdef users of smp_call_function_mask(), define it as an
empty macro to avoid build errors.  The function explicitly prohibits
callers from specifying the current cpu, so nothing needs to be done.

This unbreaks uniprocessor KVM builds.

Signed-off-by: Avi Kivity <avi@qumranet.com>

diff --git a/include/linux/smp.h b/include/linux/smp.h
index 259a13c..24e2e31 100644
--- a/include/linux/smp.h
+++ b/include/linux/smp.h
@@ -108,6 +108,9 @@ static inline void smp_send_reschedule(int cpu) { }
 	0;			\
 })
 
+#define smp_call_function_mask(mask, func, info, wait) \
+	do { (void)(mask); (void)(func); (void)(info); (void)(wait); } while (0)
+
 #endif /* !SMP */
 
 /*

Re: [PATCH] Define smp_call_function_mask() on uniprocessor builds

[Andrew Morton]

On Wed, 24 Oct 2007 10:51:25 +0200 Avi Kivity <avi@qumranet.com> wrote:

> Rather than #ifdef users of smp_call_function_mask(), define it as an
> empty macro to avoid build errors.  The function explicitly prohibits
> callers from specifying the current cpu, so nothing needs to be done.
> 
> This unbreaks uniprocessor KVM builds.
> 
> Signed-off-by: Avi Kivity <avi@qumranet.com>
> 
> diff --git a/include/linux/smp.h b/include/linux/smp.h
> index 259a13c..24e2e31 100644
> --- a/include/linux/smp.h
> +++ b/include/linux/smp.h
> @@ -108,6 +108,9 @@ static inline void smp_send_reschedule(int cpu) { }
>  	0;			\
>  })
>  
> +#define smp_call_function_mask(mask, func, info, wait) \
> +	do { (void)(mask); (void)(func); (void)(info); (void)(wait); } while (0)
> +
>  #endif /* !SMP */
>  
>  /*

The real smp_call_function_mask() returns int, so this version should do so
also.

An inlined C function will provide the needed references to the arguments -
no need for those funny (void) thingies.  And it will provide typechecking.

Re: [PATCH] Define smp_call_function_mask() on uniprocessor builds

[Avi Kivity]

Andrew Morton wrote:
> On Wed, 24 Oct 2007 10:51:25 +0200 Avi Kivity <avi@qumranet.com> wrote:
>
>   
>> Rather than #ifdef users of smp_call_function_mask(), define it as an
>> empty macro to avoid build errors.  The function explicitly prohibits
>> callers from specifying the current cpu, so nothing needs to be done.
>>
>> This unbreaks uniprocessor KVM builds.
>>
>> Signed-off-by: Avi Kivity <avi@qumranet.com>
>>
>> diff --git a/include/linux/smp.h b/include/linux/smp.h
>> index 259a13c..24e2e31 100644
>> --- a/include/linux/smp.h
>> +++ b/include/linux/smp.h
>> @@ -108,6 +108,9 @@ static inline void smp_send_reschedule(int cpu) { }
>>  	0;			\
>>  })
>>  
>> +#define smp_call_function_mask(mask, func, info, wait) \
>> +	do { (void)(mask); (void)(func); (void)(info); (void)(wait); } while (0)
>> +
>>  #endif /* !SMP */
>>  
>>  /*
>>     
>
> The real smp_call_function_mask() returns int, so this version should do so
> also.
>
>   

Right.  Ingo sent a patch which does this.

[I can't think of a way to use that int though]

> An inlined C function will provide the needed references to the arguments -
> no need for those funny (void) thingies.  And it will provide typechecking.
>   

Last time I did this (on smp_call_function_single()), the build broke on 
a zillion-and-a-half archs due to include file hell.  That got patched 
into a macro. I saw the cpumask_t in there and got cold feet.

-- 
error compiling committee.c: too many arguments to function