On 03/12/2014 00:05, Radim Krčmář wrote:
> 2014-12-02 14:09+0100, Paolo Bonzini:
>> - EAX=0Dh, ECX=1: output registers EBX/ECX/EDX are reserved.
>
> (As good as reserved without XSAVES/IA32_XSS.)
>
>> - EAX=0Dh, ECX>1: output register ECX is zero for all the CPUID leaves
>> we support, because variable "supported" comes from XCR0 and not XSS.
>> However, only bits above 0 are reserved. Output register EDX is reserved.
>
> (Yes. Well, EDX is 0 when the sub-leaf is invalid.)
>
>> Source: Intel Architecture Instruction Set Extensions Programming
>> Reference, ref. number 319433-022
>>
>> Signed-off-by: Paolo Bonzini <pbonzini@xxxxxxxxxx>
>> --
>> arch/x86/kvm/cpuid.c | 13 ++++++++++---
>> 1 file changed, 10 insertions(+), 3 deletions(-)
>>
>> diff --git a/arch/x86/kvm/cpuid.c b/arch/x86/kvm/cpuid.c
>> index 0d919bc33b02..b1366743a728 100644
>> --- a/arch/x86/kvm/cpuid.c
>> +++ b/arch/x86/kvm/cpuid.c
>> @@ -470,10 +470,17 @@ static inline int __do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
>> goto out;
>>
>> do_cpuid_1_ent(&entry[i], function, idx);
>> - if (idx == 1)
>> + if (idx == 1) {
>> entry[i].eax &= kvm_supported_word10_x86_features;
>> - else if (entry[i].eax == 0 || !(supported & mask))
>> - continue;
>> + entry[i].ebx = 0;
>> + entry[i].ecx = 0;
>> + } else {
>> + if (entry[i].eax == 0 || !(supported & mask))
>> + continue;
>> + WARN_ON_ONCE(entry[i].ecx & 1);
>> + entry[i].ecx &= 1;
>
> ECX Bit 0 is set if the sub-leaf index, n, maps to a valid bit in the
> IA32_XSS MSR and bit 0 is clear if n maps to a valid bit in XCR0.
>
> ECX should be set to 0 instead, we definitely don't map to a valid bit
> in IA32_XSS now.
Well, there is a WARN just above. :) But I can change it to zero instead.
> (Having only one part of cpuid ready for it is weird ...)
>
>> + }
>> + entry[i].edx = 0;
>> entry[i].flags |=
>> KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
>
> (Unrelated, I have yet to understand how this flag translates
> * If ECX contains an invalid sub-leaf index, EAX/EBX/ECX/EDX return 0.)
If the index is invalid, entry[i].eax is zero and we do not return
anything at all.
Paolo
>> ++*nent;
>
> Forcing a change of the XSAVES implementation is a likely purpose of
> this patch and it is correct after changing the ecx handling, so then,
>
> Reviewed-by: Radim Krčmář <rkrcmar@xxxxxxxxxx>
>
--
To unsubscribe from this list: send the line "unsubscribe kvm" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
