Re: [PATCH] x86/cpuid: expose AVX512_4VNNIW and AVX512_4FMAPS features to kvm guest

[Paolo Bonzini]

On 28/10/2016 12:13, Luc, Piotr wrote:
> On Fri, 2016-10-28 at 17:12 +0800, He Chen wrote:
>> The spec can be found in Intel Software Developer Manual or in
>> Instruction Set Extensions Programming Reference.
>>
>> Signed-off-by: Luwei Kang <luwei.kang@xxxxxxxxx>
>> Signed-off-by: He Chen <he.chen@xxxxxxxxxxxxxxx>
>> ---
>>  arch/x86/kvm/cpuid.c | 7 ++++++-
>>  1 file changed, 6 insertions(+), 1 deletion(-)
>>
>> diff --git a/arch/x86/kvm/cpuid.c b/arch/x86/kvm/cpuid.c
>> index afa7bbb..328b169 100644
>> --- a/arch/x86/kvm/cpuid.c
>> +++ b/arch/x86/kvm/cpuid.c
>> @@ -376,6 +376,10 @@ static inline int __do_cpuid_ent(struct
>> kvm_cpuid_entry2 *entry, u32 function,
>>  	/* cpuid 7.0.ecx*/
>>  	const u32 kvm_cpuid_7_0_ecx_x86_features = F(PKU) | 0
>> /*OSPKE*/;
>>  
>> +	/* cpuid 7.0.edx*/
>> +	const u32 kvm_cpuid_7_0_edx_x86_features =
>> +        0x4 /* AVX512-4VNNIW */ | 0x8 /* AVX512-4FMAPS */;
>> +
>>  	/* all calls to cpuid_count() should be made on the same cpu
>> */
>>  	get_cpu();
>>  
>> @@ -458,12 +462,13 @@ static inline int __do_cpuid_ent(struct
>> kvm_cpuid_entry2 *entry, u32 function,
>>  			/* PKU is not yet implemented for shadow
>> paging. */
>>  			if (!tdp_enabled)
>>  				entry->ecx &= ~F(PKU);
>> +            entry->edx &= kvm_cpuid_7_0_edx_x86_features;
> 
> The cpu_mask() is missed here.
> I understand that it doesn't work for this scattered features but the
> bits in edx must be zeroed if corresponding flags were cleared in
> fpu__xstate_clear_all_cpu_caps.
> So this implies more work unfortunately.

So if the x86 folks would retract their objection and accept a new
cpufeature array element it would be nice, because KVM could just do

	cpuid_mask(&entry->edx, CPUID_7_0_EDX);

Otherwise, if you add a cpuid_count_edx function to processor.h then one
can do:

	entry_>edx &= cpuid_count_edx(7, 0);

which is decent too.

Thanks,

Paolo

>>  		} else {
>>  			entry->ebx = 0;
>>  			entry->ecx = 0;
>> +            entry->edx = 0;
>>  		}
>>  		entry->eax = 0;
>> -		entry->edx = 0;
>>  		break;
>>  	}
>>  	case 9: