Re: [Xen-devel] [PATCH v2 16/16] x86emul: don't assume a memory operand

["Jan Beulich" <JBeulich@xxxxxxxx>]

>>> On 29.09.16 at 23:12, <andrew.cooper3@xxxxxxxxxx> wrote:
> On 28/09/16 09:19, Jan Beulich wrote:
>> @@ -2216,7 +2217,7 @@ x86_decode(
>>          }
>>      }
>>  
>> -    if ( override_seg != -1 && ea.type == OP_MEM )
>> +    if ( override_seg != x86_seg_none )
> 
> I don't see why the "ea.type == OP_MEM" should be dropped at this
> point.  We have already set ea.type appropriately for memory
> instructions by this point, and it does open up the case where
> instructions which would have triggered XSA-123 get incorrect
> information reported if queried with x86_insn_operand_ea()

The need to remove this actually became apparent with the
testing I did for the priv-op handling, namely for OUTS with a
segment override: When we had (before the patch here)
ea.type start out as OP_MEM, the conditional above was true
_unless_ ea.type got changed later on. With it now (properly
imo) starting out as OP_NONE, instructions not changing it to
OP_MEM (like all the string ones) would not get the segment
override applied anymore.

And no, x86_insn_operand_ea() returns x86_seg_none when
ea.type is anything other than OP_MEM.

Jan


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