Re: [Qemu-devel] [PATCH v3 0/3] target-i386: add memory protection-key support

[Paolo Bonzini <pbonzini@redhat.com>]

On 18/11/2015 18:51, Eduardo Habkost wrote:
> Is every CPU supporting PKU guaranteed to have
> CPUID.(EAX=0DH,ECX=9):EBX = 0xa80?

We asked Intel a while ago when reorganizing XSAVE support in KVM and
QEMU.  Unfortunately, Intel is not listing the offsets anymore in the
documentation, but they confirmed at the time that offsets won't change.

http://lists.xen.org/archives/html/xen-devel/2013-09/msg00484.html is
pretty clear in this respect:

>> [adding H. Peter Anvin... the context is whether the layout of the
>> XSAVE/XRSTOR area is fixed, including the offset of each separate
>> Ext_SAVE_Area].
> 
> It is.
> 
>> So please Intel, pretty please do not modify the XSAVE offsets, and
>> clarify this as soon as possible.
> 
> They will not change.
> 
>         -hpa


This of course doesn't mean that the 0xa80 is correct; it only means
that if it is correct, it will always stay correct. :)

It makes sense that 0xa80 is correct, since ECX=8 is a supervisor state
(thus never saved by XSAVE/XSAVEOPT).

The fact that standard format does not account for supervisor states
means that supporting supervisor states in KVM might encounter a few
small complications.  In particular we might have to modify KVM and QEMU
to support compacted format in KVM_GET/SET_XSAVE.  For now the only
supervisor state is just a bunch of MSRs, so it would not be necessary
to include it in KVM_GET/SET_XSAVE.  Problem averted for now, then.

> Where is the PKRU state offset/layout documented?

Volume 1 of the SDM, section 13.5.7 ("PKRU State") documents the layout.
 (Careful: chapter 13 of volume 1 is "Managing State Using the XSAVE
Feature Set"; chapter 13 of volume 3 is "System Programming for
Instruction Set Extensions and Processor Extended States", and also
deals with XSAVE).

Paolo