Re: [PATCH v3 15/20] ARM: LPAE: use phys_addr_t instead of unsigned long for physical addresses

[Catalin Marinas <catalin.marinas@arm.com>]

On 17 November 2010 11:18, Arnd Bergmann <arnd@arndb.de> wrote:
> On Wednesday 17 November 2010, Catalin Marinas wrote:
>> >
>> > It may make sense to mark it const volatile, as that's the most
>> > permissive set of attributes possible (iow, it'll accept any pointer
>> > without warning.)  However, might be an idea to check what other
>> > arches do.
>>
>> The majority of architectures use volatile and a few add both const
>> and volatile.
>>
>> I usually use the generic headers as reference for function prototypes
>> and that's just using volatile without a const (cc'ing Arnd for any
>> view on this).
>
> const volatile sounds right here, I can also change the generic header
> if we agree to do the same in arm.

That's what I plan to add, though it may be better to push as a
separate patch to fix compiler warnings now. I don't know when the
LPAE patches would go in.

--- a/arch/arm/include/asm/memory.h
+++ b/arch/arm/include/asm/memory.h
@@ -189,7 +189,7 @@
  * translation for translating DMA addresses.  Use the driver
  * DMA support - see dma-mapping.h.
  */
-static inline phys_addr_t virt_to_phys(void *x)
+static inline phys_addr_t virt_to_phys(const volatile void *x)
 {
 	return __virt_to_phys((unsigned long)(x));
 }

-- 
Catalin