Re: [PATCH v2 05/12] VMX/altp2m: add code to support EPTP switching and #VE.

[Ed White <edmund.h.white@intel.com>]

On 06/24/2015 04:59 AM, Andrew Cooper wrote:
>> +
>> +    if ( !veinfo )
>> +        return 0;
>> +
>> +    if ( veinfo->semaphore != 0 )
>> +        goto out;
> 
> The semantics of this semaphore are not clearly spelled out in the
> manual.  The only information I can locate concerning this field is in
> note in 25.5.6.1 which says:
> 
> "Delivery of virtualization exceptions writes the value FFFFFFFFH to
> offset 4 in the virtualization-exception informa-
> tion area (see Section 25.5.6.2). Thus, once a virtualization exception
> occurs, another can occur only if software
> clears this field."
> 
> I presume this should be taken to mean "software writes 0 to this
> field", but some clarification would be nice.
> 

Immediately above that, where the conditions required to deliver #VE
are discussed, it says "the 32 bits at offset 4 in the
virtualization-exception information area are all 0". Hardware never
writes anything other than FFFFFFFFH there, so only software can make
that be so.

>> +
>> +            if ( !p2m_find_altp2m_by_eptp(v->domain, eptp, &idx) )
>> +            {
>> +                gdprintk(XENLOG_ERR, "EPTP not found in alternate p2m list\n");
>> +                domain_crash(v->domain);
>> +            }
>> +        }
>> +
> 
> Is it worth checking that idx is plausible at this point, before blindly
> writing it back into the vcpu structure?

I'm not sure I follow your logic. In the case where the hardware supports
EPTP_INDEX, the hardware itself is asserting that the index is valid. In the
case quoted above, if the index isn't valid p2m_find_altp2m_by_eptp() will
fail.

Ed