[Qemu-devel] Re: [PATCH 4/4] target-xxx: Use fprintf_function (format checking)

[Stefan Weil <weil@mail.berlios.de>]

Am 01.11.2010 12:03, schrieb Paolo Bonzini:
> On 11/01/2010 11:27 AM, TeLeMan wrote:
>> On Mon, Nov 1, 2010 at 18:14, Paolo Bonzini<pbonzini@redhat.com>  wrote:
>>> On 11/01/2010 10:50 AM, TeLeMan wrote:
>>>>>>
>>>>>> I think this patch is not right. Outputting 64bits data is not
>>>>>> necessary on 32bits mode.
>>>>>
>>>>> Do you speak of 32 bit hosts or 32 bit targets?
>>>>
>>>> 32bit mode of x64
>>>
>>> There is no such thing as a 32 bit host on x64, only 64-bit hosts that
>>> haven't turned on long mode.  So printing 64 bits is correct for those.
>>
>> If so, why the above crX is printed by 32 bits?
>
> There are two issues.  One is what type specifier to use (and it is a 
> correctness issue), the other is what width to use (and it is an 
> aesthetics issue).  The patch fixes the correctness issue and makes 
> the aesthetic part worse.
>
> I agree that a better fix would be to cast to uint32_t as it is done 
> for crX, but this patch is anyway better than nothing because right 
> now DR7 is printed incorrectly _exactly on 64-bit guests running on 
> 32-bit mode_.
>
> An even better fix than uint32_t would be to introduce TARGET_FMT_8lx 
> (which maps to "%08"PRI_x64) so that, if for some reason the high 
> 32-bit are not zero, they will be shown.
>
> Paolo

Yes. We already had a similar discussion about TARGET_FMT_PLX, see
http://www.mail-archive.com/qemu-devel@nongnu.org/msg42977.html.

There I suggested to define PRIxTPA (format specifier for a target
physical address). We could also add a PRIxTUL (for target_ulong),
so it would be possible to output "%04" PRIxTUL or "%08" PRIxTUL.
Thus we could avoid the need for TARGET_FMT_8lx, TARGET_FMT_4lx
and maybe even TARGET_FMT_2lx.

Kind regards,
Stefan