Re: [PATCH kvm-unit-tests] access: check SMEP on prefetch pte path

[Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>]

On 06/25/2011 10:44 AM, Yang, Wei Y wrote:
> 
>>
>>> +void set_cr4_smep(int smep)
>>> +{
>>> +    unsigned long cr4 = read_cr4();
>>> +
>>> +    cr4 &= ~CR4_SMEP_MASK;
>>> +    if (smep)
>>> +	cr4 |= CR4_SMEP_MASK;
>>> +    write_cr4(cr4);
>>> +}
>>> +
>>
>> It can work if the box does not support SMEP?
> 
> It will report unhandled exception 13 in access.out which
> we count as errors either. Do we need verify it before
> running the smep test case?
> 

It can generate the wrong report that is not caused by the fault of KVM,
we'd better avoid it.

>>
>>>  void set_efer_nx(int nx)
>>>  {
>>>      unsigned long long efer;
>>> @@ -176,7 +188,7 @@ void ac_test_init(ac_test_t *at, void *virt)
>>>
>>>  int ac_test_bump_one(ac_test_t *at)
>>>  {
>>> -    for (int i = 0; i < NR_AC_FLAGS; ++i)
>>> +    for (int i = 0; i < NR_AC_FLAGS-1; ++i)
>>
>> Why not test "SMEP" for all test case?
> 
> It's actually the last question. :)
> The page where the current code stays is user page.
> If enabled SMEP, we must add some code to convert the current page
> into kernel page, or the page fault will occur again and again. And it seems
> no need to test SMEP for all test case. We just need fetch access which we
> do in a separate case.
> 

Oh, i see, thanks for your explanation, i think we'd better test SMEP for
all kinds of access, but i don't have strong opinion :-)