Re: [CFT PATCH v2 0/2] KVM: support XSAVES usage in the host

[Paolo Bonzini <pbonzini@redhat.com>]

On 25/11/2014 11:13, Wanpeng Li wrote:
> Hi Paolo,
> On Mon, Nov 24, 2014 at 05:43:32PM +0100, Paolo Bonzini wrote:
>> The first patch ensures that XSAVES is not exposed in the guest until
>> we emulate MSR_IA32_XSS.  The second exports XSAVE data in the correct
>> format.
>>
>> I tested these on a non-XSAVES system so they should not be completely
>> broken, but I need some help.  I am not even sure which XSAVE states
>> are _not_ enabled, and thus compacted, in Linux.
>>
>> Note that these patches do not add support for XSAVES in the guest yet,
>> since MSR_IA32_XSS is not emulated.
>>
>> If they fix the bug Nadav reported, I'll add Reported-by and commit.
> 
> I test this patchset w/ your "KVM: x86: export get_xsave_addr" patch on 
> Skylake and guest hang during boot. The guest screen show "Probing EDD 
> (edd=off to disable)... ok", and no more dump.

Anything in dmesg?  Can you try this patch on top?

diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index 373b0ab9a32e..ca26681455c2 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -6955,6 +6955,9 @@ int fx_init(struct kvm_vcpu *vcpu)
 		return err;

 	fpu_finit(&vcpu->arch.guest_fpu);
+	if (cpu_has_xsaves)
+		vcpu->arch.guest_fpu.state->xsave.xsave_hdr.xcomp_bv =
+			host_xcr0 | XSTATE_COMPACTION_ENABLED;

 	/*
 	 * Ensure guest xcr0 is valid for loading

Paolo