[Qemu-devel] MSRC001_102C on EPYC (was Re: [PATCH v3] target-i386/cpu: Add new EPYC CPU model)

[Eduardo Habkost <ehabkost@redhat.com>]

Hi,

On Tue, Aug 15, 2017 at 12:00:51PM -0500, Brijesh Singh wrote:
> Add a new base CPU model called 'EPYC' to model processors from AMD EPYC
> family (which includes EPYC 76xx,75xx,74xx, 73xx and 72xx).
> 
> The following features bits have been added/removed compare to Opteron_G5
> 
> Added: monitor, movbe, rdrand, mmxext, ffxsr, rdtscp, cr8legacy, osvw,
>        fsgsbase, bmi1, avx2, smep, bmi2, rdseed, adx, smap, clfshopt, sha
>        xsaveopt, xsavec, xgetbv1, arat
> 
> Removed: xop, fma4, tbm
> 
[...]
> +    {
> +        .name = "EPYC",
> +        .level = 0xd,
> +        .vendor = CPUID_VENDOR_AMD,
> +        .family = 23,
> +        .model = 1,
> +        .stepping = 2,

These f/m/s values trigger model-specific code in Windows 10
guests[1], and I couldn't find any public information that allow
us to fix the problem.

Windows 10 tries to set bit 15 of MSRC001_102C, in code that
looks like workarounds for CPU Erratas.

I found a Revision Guide for family 17h[2], but it has no mention
of MSRC001_102C at all.

Can AMD help us fix this?

If we are unable to fix it, I plan to work around it by changing
EPYC's family/model/stepping to the values in Opteron_G5 on QEMU
3.0.

[1] Details can be seen at:
    https://bugzilla.redhat.com/show_bug.cgi?id=1592276
    https://bugzilla.redhat.com/show_bug.cgi?id=1593190#c12
[2] https://developer.amd.com/wp-content/resources/55449_1.12.pdf

-- 
Eduardo