Re: [Qemu-devel] [PATCH 3/6] crypto: introduce a xts_uint128 data type

[Alberto Garcia <berto@igalia.com>]

On Tue 09 Oct 2018 04:58:39 PM CEST, Daniel P. Berrangé wrote:
>> > @@ -85,7 +90,7 @@ void xts_decrypt(const void *datactx,
>> >                   uint8_t *dst,
>> >                   const uint8_t *src)
>> >  {
>> > -    uint8_t PP[XTS_BLOCK_SIZE], CC[XTS_BLOCK_SIZE], T[XTS_BLOCK_SIZE];
>> > +    xts_uint128 PP, CC, T;
>> >      unsigned long i, m, mo, lim;
>> 
>>    [...]
>> 
>> >          /* Pm = first length % XTS_BLOCK_SIZE bytes of PP */
>> >          for (i = 0; i < mo; i++) {
>> > -            CC[i] = src[XTS_BLOCK_SIZE + i];
>> > -            dst[XTS_BLOCK_SIZE + i] = PP[i];
>> > +            ((uint8_t *)&CC)[i] = src[XTS_BLOCK_SIZE + i];
>> > +            dst[XTS_BLOCK_SIZE + i] = ((uint8_t *)&PP)[i];
>> >          }
>> 
>> On second thoughts, these casts are a bit cumbersome. I wonder if it
>> isn't better to keep the array a uint8_t[] and only treat it as
>> xts_uint128 in the places where you actually do 64-bit operations
>> (xts_uint128_xor, xts_mult_x).
>
> I had done that originally, but it just shifts ugly casts from one
> place to another place in the code.

Does it really? There's a dozen casts to uint8_t * in different
places. If you use uint_8[] you would only need something like this:

static void xts_mult_x(uint8_t *I8)
{
    xts_uint128 *I = (xts_uint128 *) I8;
    /* ... the rest of the function remains the same ... */
}

And something similar in xts_uint128_xor(), which could be an inline
function instead of a macro.

Berto