Re: [Xen-devel] [PATCH 08/27] x86/hvm: Dispatch cpuid_viridian_leaves() from guest_cpuid()

["Jan Beulich" <JBeulich@xxxxxxxx>]

>>> On 04.01.17 at 16:36, <andrew.cooper3@xxxxxxxxxx> wrote:
> On 04/01/17 15:24, Jan Beulich wrote:
>>>>> On 04.01.17 at 13:39, <andrew.cooper3@xxxxxxxxxx> wrote:
>>> One check against EFER_SVME is replaced with the more appropriate 
>>> cpu_has_svm,
>>> when determining whether MSR bitmaps are available.
>> I don't think this is correct - start_svm() may fail, in which case
>> the CPUID flag doesn't get cleared, yet EFER.SVME also doesn't
>> get set. How about comparing hvm_funcs (if not NULL) ->name
>> against "SVM"?
> 
> Hmm.  This shows that the same logical bug is present in the vmx side. 

Oh, indeed. I had assumed vmx_secondary_exec_control & Co
would remain zero until after the last failure point, or get zeroed
in case of failure. I think we need to make that happen, or else
I think we have the same problem elsewhere. For the actual
CPUID feature flags, otoh, I don't think we should clear them in
these failure cases though.

Jan


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