Re: [PATCH] KVM: Add CPUID support for VIA CPU

[Avi Kivity <avi@redhat.com>]

On 04/14/2011 06:14 AM, BrillyWu@viatech.com.cn wrote:
>   On 04/13/2011 02:05 PM, BrillyWu@viatech.com.cn wrote:
> >  >  >>
> >  >  >>   +	/* cpuid 0xC0000001.edx */
> >  >  >>  +	const u32 kvm_supported_word5_x86_features =
> >  >  >>   +		F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
> >  >  >>   +		F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
> >  >  >>   +		F(PMM) | F(PMM_EN);
> >  >  >>   +
> >
> >  >  >   Are all of these features save wrt save/restore? (do they all act
> >  >  >on  state in standard registers?)  Do they need any control register
> >  >  >bits  to be active or MSRs to configure?
> >
> >  >  These features depend on instructions for the PadLock hardware engine of VIA CPU.
> >  >  The PadLock instructions just act on standard registers like general X86 instructions, and the features have been enabled when the CPU leave the factory, so there is no need to activate any control register bits or configure MSRs.
>
> >  I see there is a dependency on EFLAGS[30].  Does a VM entry clear this bit?  If not, we have to do it ourselves.
>
> Yes, PadLock hardware engine has some association with EFLAGS[30], but it just required that the EFLAGS[30] should be set to "0"
> before using PadLock ACE instructions. It is recommanded that execute instruction sequence "pushf;popf" to clear this bit before using
> ACE instructions.
> AFAIK, the VM entry never sets the EFLAGS[30] bit, so it seems that we do not have to do it ourselves.

I don't think we need to.  kvm kernel code doesn't use padlock; other 
sources which might set EFLAGS[30] are:

- the guest; but VMEXIT clears EFLAGS
- userspace; but syscall/sysenter/int instructions clear EFLAGS[30]
- another kernel thread; there is likely a popf in the context switch 
path somewhere (we should verify this)

-- 
error compiling committee.c: too many arguments to function