Re: RE: [PATCH 01/13] Nested Virtualization: tools

[Andre Przywara <andre.przywara@amd.com>]

Dong, Eddie wrote:
> Andre Przywara wrote:
>> Dong, Eddie wrote:
>>> Dong, Eddie wrote:
>>>> # HG changeset patch
>>>> # User cegger
>>>> # Date 1283345869 -7200
>>>> tools: Add nestedhvm guest config option
>>>>
>>>> diff -r 80ef08613ec2 -r ecec3d163efa tools/libxc/xc_cpuid_x86.c
>>>> --- a/tools/libxc/xc_cpuid_x86.c
>>>> +++ b/tools/libxc/xc_cpuid_x86.c
>>>> @@ -30,7 +30,7 @@
>>>>  #define set_bit(idx, dst)   ((dst) |= (1u << ((idx) & 31)))
>>>>
>>>>  #define DEF_MAX_BASE 0x0000000du
>>>> -#define DEF_MAX_EXT  0x80000008u
>>>> +#define DEF_MAX_EXT  0x8000000au
>>> How can this make Intel CPU happy?
>>> You may refer to my previous comments in V2.
>> Correct me if I am wrong, but this is only a max boundary:
>> tools/libxc/xc_cpuid_x86.c:234
>>      case 0x80000000:
>>          if ( regs[0] > DEF_MAX_EXT )
>>              regs[0] = DEF_MAX_EXT;
>>          break;
>> So if an Intel CPU returns 0x80000008 here, this will be in the
>> regs[0] field and thus any higher value in DEF_MAX_EXT does not
>> affect the guest's CPUID response.
>> So as long as Intel CPUs don't return higher values which don't match
>> the AMD assignment (which is extremely unlikely), extending
>> DEF_MAX_EXT is fine.
>>
> But it is called as MAX_EXT and will cause some unreasonable setup of leaves.
Where? If DEF_MAX_EXT would be 0x8FFFFFFF, CPUID would still return 
0x80000008 on Intel CPUs. I don't see any leaves setup because of a 
changed DEF_MAX_EXT value. CPUID will just return the smaller value of 
(DEF_MAX_EXT,native CPUID), any leafs beyond the value will not be 
populated in xc_cpuid_apply_policy() and thus will not appear in the 
HV's struct domain->arch.cpuids array used for delivering CPUID results 
to guests.


> May you split the MACRO to _AMD & _INTEL, or a dynamic variable depending on CPU brand like Keir suggested? 
I guess that is not needed. The leaf is properly handled in the 
{amd,intel}_xc_cpuid_policy() filters, which will only be called on the 
respective CPUs.

Regards,
Andre.

-- 
Andre Przywara
AMD-Operating System Research Center (OSRC), Dresden, Germany
Tel: +49 351 448-3567-12