[PATCH] KVM/X86: Check input sreg values before loading vcpu

[PATCH] KVM/X86: Check input sreg values before loading vcpu

[Tianyu Lan]

From: Lan Tianyu <Tianyu.Lan@microsoft.com>

This patch is to check sreg value first and then load vcpu in order
to avoid redundant loading/putting vcpu.

Signed-off-by: Lan Tianyu <Tianyu.Lan@microsoft.com>
---
 arch/x86/kvm/x86.c | 14 +++++++-------
 1 file changed, 7 insertions(+), 7 deletions(-)

diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index c8a0b54..46da9ec 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -7671,6 +7671,10 @@ EXPORT_SYMBOL_GPL(kvm_task_switch);
 
 int kvm_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
 {
+	if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
+			(sregs->cr4 & X86_CR4_OSXSAVE))
+		return -EINVAL;
+
 	if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
 		/*
 		 * When EFER.LME and CR0.PG are set, the processor is in
@@ -7701,14 +7705,10 @@ int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
 	struct desc_ptr dt;
 	int ret = -EINVAL;
 
-	vcpu_load(vcpu);
-
-	if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
-			(sregs->cr4 & X86_CR4_OSXSAVE))
-		goto out;
-
 	if (kvm_valid_sregs(vcpu, sregs))
-		goto out;
+		return ret;
+
+	vcpu_load(vcpu);
 
 	apic_base_msr.data = sregs->apic_base;
 	apic_base_msr.host_initiated = true;
-- 
2.7.4

Re: [PATCH] KVM/X86: Check input sreg values before loading vcpu

[rkrcmar]

2018-02-27 06:57+0000, Tianyu Lan:
> From: Lan Tianyu <Tianyu.Lan@microsoft.com>
> 
> This patch is to check sreg value first and then load vcpu in order
> to avoid redundant loading/putting vcpu.
> 
> Signed-off-by: Lan Tianyu <Tianyu.Lan@microsoft.com>
> ---

Patch "KVM: x86: KVM_CAP_SYNC_REGS" made significant changes to the
sregs setter, so the patch cannot be applied in current form.

I think that moving the X86_CR4_OSXSAVE check to guest_cpuid_has still
makes sense, but avoiding the vcpu_load/put would produce worse code
elsewhere and avoiding the load/put is not critical as any error is
probably going to be the end for this VM.

Thanks.

Re: [PATCH] KVM/X86: Check input sreg values before loading vcpu

[Tianyu Lan]

Hi Radim:
	Thanks for your review.

On 3/9/2018 12:15 AM, rkrcmar@redhat.com wrote:
> 2018-02-27 06:57+0000, Tianyu Lan:
>> From: Lan Tianyu <Tianyu.Lan@microsoft.com>
>>
>> This patch is to check sreg value first and then load vcpu in order
>> to avoid redundant loading/putting vcpu.
>>
>> Signed-off-by: Lan Tianyu <Tianyu.Lan@microsoft.com>
>> ---
> 
> Patch "KVM: x86: KVM_CAP_SYNC_REGS" made significant changes to the
> sregs setter, so the patch cannot be applied in current form.
> 
> I think that moving the X86_CR4_OSXSAVE check to guest_cpuid_has still
> makes sense, but avoiding the vcpu_load/put would produce worse code
> elsewhere and avoiding the load/put is not critical as any error is
> probably going to be the end for this VM.
> 

OK. I will update patch.