On 16/08/16 15:57, Jan Beulich wrote:
>>>> On 16.08.16 at 16:08, <andrew.coop...@citrix.com> wrote:
>> On 16/08/16 10:32, Jan Beulich wrote:
>>> Just like said in commit c0bc0adf24 ("x86emul: use DstEax where
>>> possible"): While it avoids just a few instructions, we should
>>> nevertheless make use of generic code as much as possible.
>>>
>>> Signed-off-by: Jan Beulich <jbeul...@suse.com>
>> This does reduce the amount of code, but it isn't strictly true. The
>> mul and div instructions are DstEaxEdx, as are a number of other
>> instructions.
>>
>> We shouldn't end up with special casing the eax part because we have an
>> easy literal for it, but leaving the edx hard coded because that is
>> easier to express in the current code.
> I think the code reduction is nevertheless worth it, and reduction
> here can only help readability imo. Would you be okay if I added
> a comment to the place where the DstEax gets set here? (Note
> that DstEdxEax wouldn't be true for 8-bit operations, so I'd rather
> not use this as another alias or even a completely new operand
> kind description. And please also remember that the tables don't
> express all operands in all cases anyway - just consider
> SHLD/SHRD.)
The other option would be to use DstNone and explicitly fill in
_regs.eax, which avoids all the code to play with dst, and matches how
rdtsc/rdmsr/wrmsr currently work.
~Andrew
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
https://lists.xen.org/xen-devel
