Re: [Xen-devel] [PATCH 09/10] tools/x86emul: Advertise more CPUID features for testing purposes

[Andrew Cooper <andrew.cooper3@xxxxxxxxxx>]

On 27/03/17 14:42, Jan Beulich wrote:
>>>> On 27.03.17 at 15:03, <andrew.cooper3@xxxxxxxxxx> wrote:
>> On 27/03/17 13:56, George Dunlap wrote:
>>> On 27/03/17 13:13, Jan Beulich wrote:
>>>>>>> On 27.03.17 at 13:20, <george.dunlap@xxxxxxxxxx> wrote:
>>>>> On 27/03/17 10:56, Andrew Cooper wrote:
>>>>>> Signed-off-by: Andrew Cooper <andrew.cooper3@xxxxxxxxxx>
>>>>>> ---
>>>>>> CC: Jan Beulich <JBeulich@xxxxxxxx>
>>>>>> CC: George Dunlap <george.dunlap@xxxxxxxxxxxxx>
>>>>>> CC: Ian Jackson <Ian.Jackson@xxxxxxxxxxxxx>
>>>>>> CC: Wei Liu <wei.liu2@xxxxxxxxxx>
>>>>>> ---
>>>>>>  tools/tests/x86_emulator/x86_emulate.c | 41 
>>>>>> ++++++++++++++++++++++++----------
>>>>>>  1 file changed, 29 insertions(+), 12 deletions(-)
>>>>>>
>>>>>> diff --git a/tools/tests/x86_emulator/x86_emulate.c 
>>>>> b/tools/tests/x86_emulator/x86_emulate.c
>>>>>> index cea0595..2c49954 100644
>>>>>> --- a/tools/tests/x86_emulator/x86_emulate.c
>>>>>> +++ b/tools/tests/x86_emulator/x86_emulate.c
>>>>>> @@ -73,20 +73,37 @@ int emul_test_cpuid(
>>>>>>           : "a" (leaf), "c" (subleaf));
>>>>>>  Oh, s
>>>>>>      /*
>>>>>> -     * The emulator doesn't itself use MOVBE, so we can always run the
>>>>>> -     * respective tests.
>>>>>> +     * Some instructions and features can be emulated without specific
>>>>>> +     * hardware support.  These features are unconditionally reported 
>>>>>> here,
>>>>>> +     * for testing and fuzzing-coverage purposes.
>>>>> But similarly to my question in patch 10 -- is there any chance that the
>>>>> emulator will ever be called with a cpuid callback that returns 'false"
>>>>> for these?  If so, isn't there therefore a chance that there will be
>>>>> some sort of bug which only triggers if these bits are set to 'false'?
>>>> I think I've suggested before that the cpuid hook should actually
>>>> return void, as it can't possibly fail (now that CPUID faulting is
>>>> being handled in generic code).
>>> This isn't about failing so much as it is about reporting the presence /
>>> absence of hardware features.  With this patch, cpuid unconditionally
>>> advertises the presence of a number of features (MOVBE, rtm, ADCX/ADOX,
>>> &c) because the emulation will work even if the features aren't actually
>>> present in hardware.  I'm suggesting that we may want to make sure that
>>> we test *both* the "feature is present" path, *and* the "feature is
>>> missing" path.
>> I have some plans to make this happen, but it isn't easy with the
>> existing infrastructure.  In the meantime, It is more important to get
>> better coverage.
> Can't we simply grab enough bits of input data to cover the ones
> of interest here, store them away, and use that instead of the
> hard coded 1s here?

Yes, but I also want to cover hiding features which are present on the
CPU as well, to test those early #UD paths.

~Andrew

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