Re: [PATCH 2/2] xen/x86: Introduce a new VMASSIST for architectural behaviour of iopl

[Andrew Cooper <andrew.cooper3@citrix.com>]

On 17/03/16 11:00, Jan Beulich wrote:
>>>> On 17.03.16 at 11:45, <andrew.cooper3@citrix.com> wrote:
>> On 17/03/16 10:25, Jan Beulich wrote:
>>>>>> On 16.03.16 at 21:05, <andrew.cooper3@citrix.com> wrote:
>>>> @@ -1742,8 +1742,10 @@ static void load_segments(struct vcpu *n)
>>>>              cs_and_mask = (unsigned short)regs->cs |
>>>>                  ((unsigned int)vcpu_info(n, evtchn_upcall_mask) << 16);
>>>>              /* Fold upcall mask into RFLAGS.IF. */
>>>> -            eflags  = regs->_eflags & ~X86_EFLAGS_IF;
>>>> +            eflags  = regs->_eflags & ~(X86_EFLAGS_IF|X86_EFLAGS_IOPL);
>>> This and ...
>>>
>>>> @@ -1788,8 +1790,10 @@ static void load_segments(struct vcpu *n)
>>>>              ((unsigned long)vcpu_info(n, evtchn_upcall_mask) << 32);
>>>>  
>>>>          /* Fold upcall mask into RFLAGS.IF. */
>>>> -        rflags  = regs->rflags & ~X86_EFLAGS_IF;
>>>> +        rflags  = regs->rflags & ~(X86_EFLAGS_IF|X86_EFLAGS_IOPL);
>>> ... this is not really necessary (but also not wrong) - the actual
>>> EFLAGS.IOPL is always zero (and assumed to be so by code
>>> further down from the respective adjustments you make). For
>>> consistency's sake it might be better to either drop the changes
>>> here, or also adjust the two places masking regs->eflags.
>> I will adjust the others.  I would prefer not to rely on the assumption
>> that it is actually 0.
> But you realize that if it wasn't zero, we'd have a security issue?

Indeed.  But as this adjustment is literally free for us to use, making
Xen a little more robust in the (hopefully never) case were IOPL ends up
not being 0.

~Andrew

> (This notwithstanding I'm fine with both directions, as indicated
> before.)
>
> Jan
>


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