Re: [Qemu-devel] XSAVES in GET_SUPPORTED_CPUID (was Re: [PATCH] target-i386: add Skylake-Client cpu mode)

[Paolo Bonzini]

On 12/05/2016 14:03, Eduardo Habkost wrote:
> > Item number 2 on the other hand means that it's okay to add Skylake CPU
> > models without XSAVES.  Because of the large number of kernels in the
> > wild that block XSAVES, I'm inclined to do that.
>
> Agreed. Now, should we name the CPU model without XSAVES
> "Skylake" or "Skylake-noXSAVES"? I'm inclined towards the latter,
> to follow the same pattern we used for "Haswell-noTSX".

Do we have a plan to add Skylake with XSAVES?  I think no, so it should
be fine to do

        .name = "Skylake",

        .features[FEATURE_XSAVE] = CPUID_XSAVE_XSAVEOPT |
                CPUID_XSAVE_XGETBV1 | CPUID_XSAVE_XSAVEC
                /* omitting CPUID_XSAVE_XSAVES because... */

Haswell-noTSX was added only because we already had a model with TSX.
If we hadn't we probably would have had:

- Haswell without TSX

- Broadwell without TSX

- Broadwell-EX with TSX (or something like that).

Thanks,

Paolo