Re: [PATCH for-4.6] xen/public: arm: Use __typeof__ rather than typeof

[Julien Grall <julien.grall@citrix.com>]

Hi Jan,

On 27/10/2015 08:07, Jan Beulich wrote:
>>>> On 26.10.15 at 19:08, <julien.grall@citrix.com> wrote:
>> On 23/10/15 15:55, Ian Campbell wrote:
>>> On Fri, 2015-10-23 at 15:44 +0100, Julien Grall wrote:
>>>> Thank you for the explanation. I think we can do the same as x86 does
>>>> i.e:
>>>>
>>>> #define set_xen_guest_handle_raw(hnd, val)                  \
>>>>      do { if ( sizeof(hnd) == 8 ) *(uint64_t *)&(hnd) = 0;   \
>>>>           (hnd).p = val;                                     \
>>>>      } while ( 0 )
>>>
>>> This evaluates hnd twice, which I assumed we wanted to avoid.
>>>
>>> But if that is OK for x86 in this situation then there is no harm doing it
>>> on ARM too[0].
>>>
>>> But in that case I think we would just do
>>> 	(hnd).q = val ; (hnd).p = val
>>> rather than messing with &, casts and *.
>>
>> Which is, based on the ISO C spec [1], unspecified. See 6.2.6.1#7:
>>
>> "When a value is stored in a member of an object of union type, the
>> bytes of the object representation that do not correspond to that member
>> but do correspond to other members take unspecified values, but the
>> value of the union object shall not thereby become a trap
>> representation."
>
> Question is - why would you need both assignments? Afaics the
> .q one would suffice.

Agree .q would suffice only and only if we replace all the usage of .p 
by .q with a cast.

Otherwise, accessing .p just after .q may be implementation-defined (see 
[1]).

Anyway, as you sent a patch to drop get_xen_guest_handle, I will see to 
implement Ian Jackson suggestion. I.e:

typedef union {
  uint64_t actual;
  type *for_check }

#define set_xen_guest_handle_raw(hnd, val) 	\
   do { 						\
	hnd.actual = (uint64_t)(val);		\
	sizeof((val) == ((hnd).for_check)
   }

Regards,

[1]  http://www.open-std.org/jtc1/sc22/wg14/www/docs/dr_283.htm

-- 
Julien Grall