[PATCH 2/2] [RFCv3] arm: add half-word __xchg

[mathieu.desnoyers@polymtl.ca (Mathieu Desnoyers)]

* Alexander Shishkin (virtuoso at slind.org) wrote:
> On systems where ldrexh/strexh are not available,
>   * for pre-v6 systems, use a generic local version,
>   * for v6 without v6K, emulate xchg2 using 32-bit cmpxchg()
>     (it is not yet clear if xchg1 has to be emulated on such
>     systems as well, thus the "size" parameter).
> 
> The __xchg_generic() function is based on the code that Jamie
> posted earlier.
> 
> Signed-off-by: Alexander Shishkin <virtuoso@slind.org>
> CC: linux-arm-kernel-bounces at lists.infradead.org
> CC: Imre Deak <imre.deak@nokia.com>
> CC: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
> CC: Jamie Lokier <jamie@shareable.org>
> ---
>  arch/arm/include/asm/system.h |   56 +++++++++++++++++++++++++++++++++++++++++
>  1 files changed, 56 insertions(+), 0 deletions(-)
> 
> diff --git a/arch/arm/include/asm/system.h b/arch/arm/include/asm/system.h
> index d65b2f5..7a5983f 100644
> --- a/arch/arm/include/asm/system.h
> +++ b/arch/arm/include/asm/system.h
> @@ -218,6 +218,39 @@ do {									\
>  	last = __switch_to(prev,task_thread_info(prev), task_thread_info(next));	\
>  } while (0)
>  
> +#if __LINUX_ARM_ARCH__ >= 6
> +
> +#include <linux/align.h>
> +
> +static inline unsigned long __cmpxchg(volatile void *ptr, unsigned long old,
> +				      unsigned long new, int size);
> +
> +/*
> + * emulate __xchg() using 32-bit __cmpxchg()
> + */
> +static inline unsigned long __xchg_generic(unsigned long x,
> +						 volatile void *ptr, int size)
> +{
> +	unsigned long *ptrbig = object_align_floor((unsigned long *)ptr);

ptrbig could be renamed to something more relevant. Maybe ptralign ?

> +	int shift = ((unsigned)ptr - (unsigned)ptrbig) * 8;

"unsigned int" everywhere above would be cleaner. Also, it's worth
checking the generated assembly: does gcc perform the transformation:

  * 8   ->   << 3    automatically ?

> +	unsigned long mask, add, ret;
> +
> +	mask = ~(((1 << (size * 8)) - 1) << shift);

Maybe better to do:

+	mask = ~((((1UL << 3) << size) - 1) << shift);

But, other question: what assumptions are you doing about endianness
here ? I recall that ARM supports switchable endianness. Dunno about the
Linux-specific case though.

Thanks,

Mathieu

> +	add = x << shift;
> +
> +	ret = *ptrbig;
> +	while (1) {
> +		unsigned long tmp = __cmpxchg(ptrbig, ret, (ret & mask) | add,
> +					      4);
> +		if (tmp == ret)
> +			break;
> +		ret = tmp;
> +	}
> +
> +	return ret;
> +}
> +#endif
> +
>  #if defined(CONFIG_CPU_SA1100) || defined(CONFIG_CPU_SA110)
>  /*
>   * On the StrongARM, "swp" is terminally broken since it bypasses the
> @@ -262,6 +295,22 @@ static inline unsigned long __xchg(unsigned long x, volatile void *ptr, int size
>  			: "r" (x), "r" (ptr)
>  			: "memory", "cc");
>  		break;
> +#ifdef CONFIG_CPU_32v6K
> +	case 2:
> +		asm volatile("@	__xchg2\n"
> +		"1:	ldrexh	%0, [%3]\n"
> +		"	strexh	%1, %2, [%3]\n"
> +		"	teq	%1, #0\n"
> +		"	bne	1b"
> +			: "=&r" (ret), "=&r" (tmp)
> +			: "r" (x), "r" (ptr)
> +			: "memory", "cc");
> +		break;
> +#else
> +	case 2:
> +		ret = __xchg_generic(x, ptr, 2);
> +		break;
> +#endif
>  	case 4:
>  		asm volatile("@	__xchg4\n"
>  		"1:	ldrex	%0, [%3]\n"
> @@ -283,6 +332,13 @@ static inline unsigned long __xchg(unsigned long x, volatile void *ptr, int size
>  		raw_local_irq_restore(flags);
>  		break;
>  
> +	case 2:
> +		raw_local_irq_save(flags);
> +		ret = *(volatile unsigned short *)ptr;
> +		*(volatile unsigned short *)ptr = x;
> +		raw_local_irq_restore(flags);
> +		break;
> +
>  	case 4:
>  		raw_local_irq_save(flags);
>  		ret = *(volatile unsigned long *)ptr;
> -- 
> 1.7.1.1.g15764
> 

-- 
Mathieu Desnoyers
Operating System Efficiency R&D Consultant
EfficiOS Inc.
http://www.efficios.com