Re: [Xen-devel] [Embedded-pv-devel] [PATCH v9] xen: add para-virtual sound interface header files

[Artem Mygaiev <artem_mygaiev@xxxxxxxx>]

On 24.11.16 18:08, Julien Grall wrote:
> On 24/11/16 15:50, Artem Mygaiev wrote:
>> On 24.11.16 17:09, Julien Grall wrote:
>>> Hi Artem,
>>>
>>> On 24/11/16 14:58, Artem Mygaiev wrote:
>>>> On 24.11.16 15:31, Jan Beulich wrote:
>>>>>>>> On 24.11.16 at 14:18, <artem_mygaiev@xxxxxxxx> wrote:
>>>>>> On 24.11.16 15:04, Jan Beulich wrote:
>>>>>>>> In my case I need it to at least define Linux specific __packed
>>>>>>>> attribute which is not supported by Win AFAIK.
>>>>>>> ... needs to avoid using platform specific constructs (or making
>>>>>>> other assumptions on the platform). I can only re-iterate: Please
>>>>>>> follow the model other PV protocols already present (without
>>>>>>> repeating their mistakes, if at all possible).
>>>>>> But if we don't pack (or enforce some specific alignment) data
>>>>>> structures used between domains they may have different alignment on
>>>>>> different domains, no?
>>>>> Let me repeat what I've said before: Please take a look at existing
>>>>> interface headers. By properly placing fields and adding explicit
>>>>> padding where needed, you can avoid such layout differences.
>>>>
>>>> Don't get me wrong - I have checked the i/f headers before. My concern
>>>> is very generic: C standard defines alignment as an
>>>> *implementation-defined* integer, so current implementation is based
>>>> under two assumptions on the guest platforms: a) guest compiler
>>>> does not
>>>> have "packing" enabled by default and b) alignment follows typical
>>>> values (1 byte for char, 2 bytes for short, ...). While this is
>>>> true in
>>>> most cases, with some embedded RTOS this may become an issue.
>>>
>>> I am not sure why you mentioned embedded RTOS, this is specific to the
>>> compiler and not an OS.
>>>
>> Indeed, my statement is not quite correct. It is just that there are
>> proprietary RTOSes bundled with IDEs with own or customized tools,
>> compilers, etc.
>
> And their own ABI? It sounds crazy to me that someone may decide to
> use padding even when field are naturally aligned.
No, no, that would be overkill... I am just talking about system-wide
alignment configuration for performance/size balance (reduce padding,
not increase).

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
https://lists.xen.org/xen-devel