Re: [PATCH] dm vdo: use kernel byteswapping routines instead of GCC ones

[Matthew Sakai <msakai@redhat.com>]

On 3/20/24 23:52, Eric Biggers wrote:
> On Wed, Mar 20, 2024 at 05:44:05PM -0400, Matthew Sakai wrote:
>>   static __always_inline u64 getblock64(const u64 *p, int i)
>>   {
>> -#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
>> -	return p[i];
>> -#elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
>> -	return __builtin_bswap64(p[i]);
>> -#else
>> -#error "can't figure out byte order"
>> -#endif
>> +	return le64_to_cpup(&p[i]);
>>   }
>>   
>>   static __always_inline void putblock64(u64 *p, int i, u64 value)
>>   {
>> -#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
>> -	p[i] = value;
>> -#elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
>> -	p[i] = __builtin_bswap64(value);
>> -#else
>> -#error "can't figure out byte order"
>> -#endif
>> +	p[i] = cpu_to_le64(value);
>>   }
> 
> This is very broken.  What you're actually looking for is get_unaligned_le64()
> and put_unaligned_le64().  And they should be folded directly into the caller.
> 
> - Eric
> 

Thanks for the suggestion. We'll work on a v2 to include this.

Matt