[PATCH] KVM: VMX: Pass cr0.mp through to the guest when the fpu is active

[PATCH] KVM: VMX: Pass cr0.mp through to the guest when the fpu is active

[Avi Kivity]

When cr0.mp is clear, the guest doesn't expect a #NM in response to
a WAIT instruction.  Because we always keep cr0.mp set, it will get
a #NM, and potentially be confused.

Fix by keeping cr0.mp set only when the fpu is inactive, and passing
it through when inactive.

Reported-by: Lorenzo Martignoni <martignlo@gmail.com>
Analyzed-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
---
 arch/x86/kvm/vmx.c |   15 +++++++++------
 1 files changed, 9 insertions(+), 6 deletions(-)

diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
index d27ff6a..9befb53 100644
--- a/arch/x86/kvm/vmx.c
+++ b/arch/x86/kvm/vmx.c
@@ -66,7 +66,7 @@ module_param(emulate_invalid_guest_state, bool, S_IRUGO);
 #define KVM_GUEST_CR0_MASK						\
 	(KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST				\
-	(X86_CR0_WP | X86_CR0_NE | X86_CR0_MP)
+	(X86_CR0_WP | X86_CR0_NE)
 #define KVM_VM_CR0_ALWAYS_ON						\
 	(KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
 #define KVM_CR4_GUEST_OWNED_BITS				      \
@@ -796,12 +796,15 @@ static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
 
 static void vmx_fpu_activate(struct kvm_vcpu *vcpu)
 {
+	ulong cr0;
+
 	if (vcpu->fpu_active)
 		return;
 	vcpu->fpu_active = 1;
-	vmcs_clear_bits(GUEST_CR0, X86_CR0_TS);
-	if (kvm_read_cr0_bits(vcpu, X86_CR0_TS))
-		vmcs_set_bits(GUEST_CR0, X86_CR0_TS);
+	cr0 = vmcs_readl(GUEST_CR0);
+	cr0 &= ~(X86_CR0_TS | X86_CR0_MP);
+	cr0 |= kvm_read_cr0_bits(vcpu, X86_CR0_TS | X86_CR0_MP);
+	vmcs_writel(GUEST_CR0, cr0);
 	update_exception_bitmap(vcpu);
 	vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
 	vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
@@ -812,7 +815,7 @@ static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu)
 	ulong old_ts, old_cr0;
 
 	old_ts = kvm_read_cr0_bits(vcpu, X86_CR0_TS);
-	vmcs_set_bits(GUEST_CR0, X86_CR0_TS);
+	vmcs_set_bits(GUEST_CR0, X86_CR0_TS | X86_CR0_MP);
 	update_exception_bitmap(vcpu);
 	vcpu->arch.cr0_guest_owned_bits = 0;
 	vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
@@ -1765,7 +1768,7 @@ static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 		ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
 
 	if (!vcpu->fpu_active)
-		hw_cr0 |= X86_CR0_TS;
+		hw_cr0 |= X86_CR0_TS | X86_CR0_MP;
 
 	vmcs_writel(CR0_READ_SHADOW, cr0);
 	vmcs_writel(GUEST_CR0, hw_cr0);
-- 
1.6.5.3

Re: [PATCH] KVM: VMX: Pass cr0.mp through to the guest when the fpu is active

[Avi Kivity]

On 01/24/2010 04:26 PM, Avi Kivity wrote:
> When cr0.mp is clear, the guest doesn't expect a #NM in response to
> a WAIT instruction.  Because we always keep cr0.mp set, it will get
> a #NM, and potentially be confused.
>
> Fix by keeping cr0.mp set only when the fpu is inactive, and passing
> it through when inactive.
>    

Note: since all reasonable guests set cr0.mp, and those that don't 
shouldn't issue fwait, there's no need to queue this for .33 or stable.

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

Re: [PATCH] KVM: VMX: Pass cr0.mp through to the guest when the fpu is active

[Marcelo Tosatti]

On Sun, Jan 24, 2010 at 04:26:40PM +0200, Avi Kivity wrote:
> When cr0.mp is clear, the guest doesn't expect a #NM in response to
> a WAIT instruction.  Because we always keep cr0.mp set, it will get
> a #NM, and potentially be confused.
> 
> Fix by keeping cr0.mp set only when the fpu is inactive, and passing
> it through when inactive.
> 
> Reported-by: Lorenzo Martignoni <martignlo@gmail.com>
> Analyzed-by: Gleb Natapov <gleb@redhat.com>
> Signed-off-by: Avi Kivity <avi@redhat.com>

Applied, thanks.