diff mbox

[1/2] KVM: x86: disable MPX if host did not enable MPX XSAVE features

Message ID 1457437481-65784-2-git-send-email-pbonzini@redhat.com (mailing list archive)
State New, archived
Headers show

Commit Message

Paolo Bonzini March 8, 2016, 11:44 a.m. UTC 
When eager FPU is disabled, KVM will still see the MPX bit in CPUID and
presumably the MPX vmentry and vmexit controls.  However, it will not
be able to expose the MPX XSAVE features to the guest, because the guest's
accessible XSAVE features are always a subset of host_xcr0.

In this case, we should disable the MPX CPUID bit, the BNDCFGS MSR,
and the MPX vmentry and vmexit controls for nested virtualization.
It is then unnecessary to enable guest eager FPU if the guest has the
MPX CPUID bit set.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
 arch/x86/kvm/cpuid.c | 13 ++++++++++---
 arch/x86/kvm/cpuid.h |  9 +--------
 arch/x86/kvm/vmx.c   | 13 ++++++-------
 3 files changed, 17 insertions(+), 18 deletions(-)

Comments

Paolo Bonzini March 11, 2016, 12:09 p.m. UTC | #1 
On 11/03/2016 03:37, Yang Zhang wrote:
>> @@ -97,7 +104,7 @@ int kvm_update_cpuid(struct kvm_vcpu *vcpu)
>>   	if (best && (best->eax & (F(XSAVES) | F(XSAVEC))))
>>   		best->ebx = xstate_required_size(vcpu->arch.xcr0, true);
>>   
>> -	vcpu->arch.eager_fpu = use_eager_fpu() || guest_cpuid_has_mpx(vcpu);
>> +	vcpu->arch.eager_fpu = use_eager_fpu();
> 
> Hi Paolo,
> 
> As i mentioned on another thread, force KVM to use eager fpu
> unconditionally may introduce the performance regression. Though the
> cost for eager fpu is very small especially in modern CPU, it still
> cannot be ignored on old platform.

This patch doesn't change anything in that respect.  It doesn't enable
eager FPU in any case where it wasn't already enabled before the patch.

All this patch does is hide MPX completely to the guests (just like it's
hidden on the host) if the host is using lazy FPU.

> And we have observed some performance
> decrease on those platforms according the result from some experiments
> which did several years ago.

Indeed after the merge window I plan to benchmark KVM on old systems
(pre-XSAVE) to see if there is a negative benefit from eager FPU.

Thanks,

Paolo
--
To unsubscribe from this list: send the line "unsubscribe kvm" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Yang Zhang March 24, 2016, 1:06 p.m. UTC | #2 
On 2016/3/11 20:09, Paolo Bonzini wrote:

Sorry for the late reply.

>
>
> On 11/03/2016 03:37, Yang Zhang wrote:
>>> @@ -97,7 +104,7 @@ int kvm_update_cpuid(struct kvm_vcpu *vcpu)
>>>    	if (best && (best->eax & (F(XSAVES) | F(XSAVEC))))
>>>    		best->ebx = xstate_required_size(vcpu->arch.xcr0, true);
>>>
>>> -	vcpu->arch.eager_fpu = use_eager_fpu() || guest_cpuid_has_mpx(vcpu);
>>> +	vcpu->arch.eager_fpu = use_eager_fpu();
>>
>> Hi Paolo,
>>
>> As i mentioned on another thread, force KVM to use eager fpu
>> unconditionally may introduce the performance regression. Though the
>> cost for eager fpu is very small especially in modern CPU, it still
>> cannot be ignored on old platform.
>
> This patch doesn't change anything in that respect.  It doesn't enable
> eager FPU in any case where it wasn't already enabled before the patch.

I mean why not keep the old way that only activate the eager_fpu while 
guest sees the MPX bit in CPUID, like:

vcpu->arch.eager_fpu = use_eager_fpu() && guest_cpuid_has_mpx(vcpu);

Besides, vmx_fpu_activate is called when do vcpu_reset. So it seems no 
need to call fpu_activate() here again.
Paolo Bonzini March 24, 2016, 1:07 p.m. UTC | #3 
On 24/03/2016 14:06, Yang Zhang wrote:
> 
> I mean why not keep the old way that only activate the eager_fpu while
> guest sees the MPX bit in CPUID, like:
> 
> vcpu->arch.eager_fpu = use_eager_fpu() && guest_cpuid_has_mpx(vcpu);

If the host uses eager FPU you can assume that it's faster than lazy FPU.

Paolo

> Besides, vmx_fpu_activate is called when do vcpu_reset. So it seems no
> need to call fpu_activate() here again.
--
To unsubscribe from this list: send the line "unsubscribe kvm" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
diff mbox

Patch

diff --git a/arch/x86/kvm/cpuid.c b/arch/x86/kvm/cpuid.c
index 6525e926f566..fa241d4fda98 100644
--- a/arch/x86/kvm/cpuid.c
+++ b/arch/x86/kvm/cpuid.c
@@ -46,11 +46,18 @@  static u32 xstate_required_size(u64 xstate_bv, bool compacted)
 	return ret;
 }
 
+bool kvm_mpx_supported(void)
+{
+	return ((host_xcr0 & (XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR))
+		 && kvm_x86_ops->mpx_supported());
+}
+EXPORT_SYMBOL_GPL(kvm_mpx_supported);
+
 u64 kvm_supported_xcr0(void)
 {
 	u64 xcr0 = KVM_SUPPORTED_XCR0 & host_xcr0;
 
-	if (!kvm_x86_ops->mpx_supported())
+	if (!kvm_mpx_supported())
 		xcr0 &= ~(XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR);
 
 	return xcr0;
@@ -97,7 +104,7 @@  int kvm_update_cpuid(struct kvm_vcpu *vcpu)
 	if (best && (best->eax & (F(XSAVES) | F(XSAVEC))))
 		best->ebx = xstate_required_size(vcpu->arch.xcr0, true);
 
-	vcpu->arch.eager_fpu = use_eager_fpu() || guest_cpuid_has_mpx(vcpu);
+	vcpu->arch.eager_fpu = use_eager_fpu();
 	if (vcpu->arch.eager_fpu)
 		kvm_x86_ops->fpu_activate(vcpu);
 
@@ -295,7 +302,7 @@  static inline int __do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
 #endif
 	unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
 	unsigned f_invpcid = kvm_x86_ops->invpcid_supported() ? F(INVPCID) : 0;
-	unsigned f_mpx = kvm_x86_ops->mpx_supported() ? F(MPX) : 0;
+	unsigned f_mpx = kvm_mpx_supported() ? F(MPX) : 0;
 	unsigned f_xsaves = kvm_x86_ops->xsaves_supported() ? F(XSAVES) : 0;
 
 	/* cpuid 1.edx */
diff --git a/arch/x86/kvm/cpuid.h b/arch/x86/kvm/cpuid.h
index c8eda1498121..66a6581724ad 100644
--- a/arch/x86/kvm/cpuid.h
+++ b/arch/x86/kvm/cpuid.h
@@ -5,6 +5,7 @@ 
 #include <asm/cpu.h>
 
 int kvm_update_cpuid(struct kvm_vcpu *vcpu);
+bool kvm_mpx_supported(void);
 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
 					      u32 function, u32 index);
 int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid,
@@ -135,14 +136,6 @@  static inline bool guest_cpuid_has_rtm(struct kvm_vcpu *vcpu)
 	return best && (best->ebx & bit(X86_FEATURE_RTM));
 }
 
-static inline bool guest_cpuid_has_mpx(struct kvm_vcpu *vcpu)
-{
-	struct kvm_cpuid_entry2 *best;
-
-	best = kvm_find_cpuid_entry(vcpu, 7, 0);
-	return best && (best->ebx & bit(X86_FEATURE_MPX));
-}
-
 static inline bool guest_cpuid_has_pcommit(struct kvm_vcpu *vcpu)
 {
 	struct kvm_cpuid_entry2 *best;
diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
index 46154dac71e6..e512aa7ed874 100644
--- a/arch/x86/kvm/vmx.c
+++ b/arch/x86/kvm/vmx.c
@@ -861,7 +861,6 @@  static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu);
 static u64 construct_eptp(unsigned long root_hpa);
 static void kvm_cpu_vmxon(u64 addr);
 static void kvm_cpu_vmxoff(void);
-static bool vmx_mpx_supported(void);
 static bool vmx_xsaves_supported(void);
 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr);
 static void vmx_set_segment(struct kvm_vcpu *vcpu,
@@ -2595,7 +2594,7 @@  static void nested_vmx_setup_ctls_msrs(struct vcpu_vmx *vmx)
 		VM_EXIT_LOAD_IA32_EFER | VM_EXIT_SAVE_IA32_EFER |
 		VM_EXIT_SAVE_VMX_PREEMPTION_TIMER | VM_EXIT_ACK_INTR_ON_EXIT;
 
-	if (vmx_mpx_supported())
+	if (kvm_mpx_supported())
 		vmx->nested.nested_vmx_exit_ctls_high |= VM_EXIT_CLEAR_BNDCFGS;
 
 	/* We support free control of debug control saving. */
@@ -2616,7 +2615,7 @@  static void nested_vmx_setup_ctls_msrs(struct vcpu_vmx *vmx)
 		VM_ENTRY_LOAD_IA32_PAT;
 	vmx->nested.nested_vmx_entry_ctls_high |=
 		(VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR | VM_ENTRY_LOAD_IA32_EFER);
-	if (vmx_mpx_supported())
+	if (kvm_mpx_supported())
 		vmx->nested.nested_vmx_entry_ctls_high |= VM_ENTRY_LOAD_BNDCFGS;
 
 	/* We support free control of debug control loading. */
@@ -2860,7 +2859,7 @@  static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
 		msr_info->data = vmcs_readl(GUEST_SYSENTER_ESP);
 		break;
 	case MSR_IA32_BNDCFGS:
-		if (!vmx_mpx_supported())
+		if (!kvm_mpx_supported())
 			return 1;
 		msr_info->data = vmcs_read64(GUEST_BNDCFGS);
 		break;
@@ -2937,7 +2936,7 @@  static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
 		vmcs_writel(GUEST_SYSENTER_ESP, data);
 		break;
 	case MSR_IA32_BNDCFGS:
-		if (!vmx_mpx_supported())
+		if (!kvm_mpx_supported())
 			return 1;
 		vmcs_write64(GUEST_BNDCFGS, data);
 		break;
@@ -3410,7 +3409,7 @@  static void init_vmcs_shadow_fields(void)
 	for (i = j = 0; i < max_shadow_read_write_fields; i++) {
 		switch (shadow_read_write_fields[i]) {
 		case GUEST_BNDCFGS:
-			if (!vmx_mpx_supported())
+			if (!kvm_mpx_supported())
 				continue;
 			break;
 		default:
@@ -10265,7 +10264,7 @@  static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
 	vmcs12->guest_sysenter_cs = vmcs_read32(GUEST_SYSENTER_CS);
 	vmcs12->guest_sysenter_esp = vmcs_readl(GUEST_SYSENTER_ESP);
 	vmcs12->guest_sysenter_eip = vmcs_readl(GUEST_SYSENTER_EIP);
-	if (vmx_mpx_supported())
+	if (kvm_mpx_supported())
 		vmcs12->guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
 	if (nested_cpu_has_xsaves(vmcs12))
 		vmcs12->xss_exit_bitmap = vmcs_read64(XSS_EXIT_BITMAP);