Re: [Xen-devel] [PATCH RFC 10/31] xen/x86: Calculate HVM featureset

[Andrew Cooper <andrew.cooper3@xxxxxxxxxx>]

On 22/12/15 17:11, Jan Beulich wrote:
>>>> On 16.12.15 at 22:24, <andrew.cooper3@xxxxxxxxxx> wrote:
>> @@ -22,6 +24,27 @@ void __init calculate_featuresets(void)
>>  
>>      /* Unconditionally claim to be able to set the hypervisor bit. */
>>      __set_bit(X86_FEATURE_HYPERVISOR, pv_featureset);
>> +
>> +    /* HVM featureset. */
>> +    if ( hvm_enabled )
>> +    {
>> +        const uint32_t *hvm_featuremask = hvm_funcs.hap_supported
>> +            ? hvm_hap_featuremask : hvm_shadow_featuremask;
>> +
>> +        for ( i = 0; i < ARRAY_SIZE(hvm_featureset); ++i )
>> +            hvm_featureset[i] = host_featureset[i] & hvm_featuremask[i];
>> +
>> +        /* Unconditionally claim to be able to set the hypervisor bit. */
>> +        __set_bit(X86_FEATURE_HYPERVISOR, hvm_featureset);
>> +
>> +        /*
>> +         * On AMD, PV guests are entirely unable to use 'sysenter' as Xen 
>> runs
>> +         * in long mode, but HVM guests are able if running in protected 
>> mode.
>> +         */
>> +        if ( (boot_cpu_data.x86_vendor == X86_VENDOR_AMD) &&
>> +             !test_bit(X86_FEATURE_SEP, boot_cpu_data.x86_capability) )
> Is the ! correct here?

Yes - init_amd() deliberately clobbers the feature.

> And please use cpu_has_sep.

Actually, thinking about it, the check isn't quite correct.  We need to
only advertise the feature if it was available before init_amd()
clobbered it.

~Andrew

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