Re: [Xen-devel] [PATCH v3 2/5] x86emul: don't assume a memory operand

["Jan Beulich" <JBeulich@xxxxxxxx>]

>>> On 06.12.16 at 17:49, <andrew.cooper3@xxxxxxxxxx> wrote:
> On 06/12/16 11:13, Jan Beulich wrote:
>> @@ -2359,7 +2360,7 @@ x86_decode(
>>          }
>>      }
>>  
>> -    if ( override_seg != -1 && ea.type == OP_MEM )
>> +    if ( override_seg != x86_seg_none )
>>          ea.mem.seg = override_seg;
> 
> Could we get away with asserting ea.type == OP_MEM if override_seg is
> set, to help validate our assumptions about state?  (Possibly even
> passing #UD back in the non-debug case)

That would be wrong - we'd be asserting guest controlled state.
There's nothing preventing a segment override to be present on
instructions without memory operands. And for example string
insns don't have OP_MEM set despite having (implicit) memory
operands (after all that's the hole reason for the change here
[but not the patch as a hole], as the following PV priv-op patch
requires the segment override to take effect on OUTS). Nor
would such be correct for conditional branches, where some of
the segment overrides have a different meaning (necessarily
ignored by the emulator).

> Otherwise, Reviewed-by: Andrew Cooper <andrew.cooper3@xxxxxxxxxx>

Let me know of the applicability of this in the light of the above.

Jan


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