Re: [Xen-devel] [PATCH] x86/vvmx: Improvements to INVEPT instruction handling

[Andrew Cooper <andrew.cooper3@xxxxxxxxxx>]

On 08/02/17 07:46, Tian, Kevin wrote:
>> From: Andrew Cooper [mailto:andrew.cooper3@xxxxxxxxxx]
>> Sent: Tuesday, February 07, 2017 12:55 AM
>>
>>  * Latch current once at the start.
>>  * Avoid the memory operand read for INVEPT_ALL_CONTEXT.  Experimentally, 
>> this
>>    is how hardware behaves, and avoids an unnecessary pagewalk.
>>  * Reject Reg/Reg encodings of the instruction.
>>  * Audit eptp against maxphysaddr.
>>  * Introduce and use VMX_INSN_INVALID_INV_OPERAND to correct the vmfail
>>    semantics.
>>  * Add extra newlines for clarity
>>
>> Also, introduce some TODOs for further checks which should be performed.
>> These checks are hard to perform at the moment, as there is no easy way to 
>> see
>> which MSR values where given to the guest.
>>
>> Signed-off-by: Andrew Cooper <andrew.cooper3@xxxxxxxxxx>
> Acked-by: Kevin Tian <kevin.tian@xxxxxxxxx>

Actually, it turns out that a combination of 2b2793d3 and f438b1c5 is
entirely broken for 32bit hypervisors, and this patch was an accidental
bugfix.

decode_vmx_inst() reads using the default memory operand size, meaning
that a 32bit code segment executing INVEPT only fills in the bottom half
of &eptp.

~Andrew

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