Re: [PATCH 2/3] grant-table: refactor grant copy to reduce duplicate code

["Jan Beulich" <JBeulich@suse.com>]

>>> On 22.01.15 at 15:42, <david.vrabel@citrix.com> wrote:
> On 22/01/15 14:24, Jan Beulich wrote:
>>>>> On 20.01.15 at 19:19, <david.vrabel@citrix.com> wrote:
>>> +static int gnttab_copy_buf(const struct gnttab_copy *op,
>>> +                           struct gnttab_copy_buf *dest,
>>> +                           const struct gnttab_copy_buf *src)
>>> +{
>>> +    s16 rc;
>> 
>> An s16 local variable used as return value in a function returning
>> int is kind of odd. Elsewhere there are also cases where the
>> function return types are also s16. I don't think that's particularly
>> efficient, and hence I think it should be changed in the places
>> where you add new helpers even if the original code used s16
>> for that purpose.
> 
> I was (trying to) use s16 where we return a GNTST_* value instead of the
> usual -ERRNO.  I can change them all to int if this is preferred
> (although I not sure I get your efficiency argument).

Operations on 16-bit registers are less efficient on x86 (due to
the extra operand size prefix byte). On ARM I don't even know
how they get carried out by the compiler, but there surely is no
way to access the low 16 bits of a register by other than
memory loads or stores.

Jan