[PATCH 0/2] KVM: x86: nSVM: fixes for SYSENTER emulation

[PATCH 0/2] KVM: x86: nSVM: fixes for SYSENTER emulation

[Maxim Levitsky]

This is a result of a deep rabbit hole dive in regard to why
currently the nested migration of 32 bit guests
is totally broken on AMD.

It turns out that due to slight differences between the original AMD64
implementation and the Intel's remake, SYSENTER instruction behaves a
bit differently on Intel, and to support migration from Intel to AMD we
try to emulate those differences away.

Sadly that collides with virtual vmload/vmsave feature that is used in nesting.
The problem was that when it is enabled,
on migration (and otherwise when userspace reads MSR_IA32_SYSENTER_{EIP|ESP},
wrong value were returned, which leads to #DF in the
nested guest when the wrong value is loaded back.

The patch I prepared carefully fixes this, by mostly disabling that
SYSCALL emulation when we don't spoof the Intel's vendor ID, and if we do,
and yet somehow SVM is enabled (this is a very rare edge case), then
virtual vmload/save is force disabled.

V2: incorporated review feedback from Paulo.

Best regards,
        Maxim Levitsky

Maxim Levitsky (2):
  KVM: x86: add guest_cpuid_is_intel
  KVM: nSVM: improve SYSENTER emulation on AMD

 arch/x86/kvm/cpuid.h   |  8 ++++
 arch/x86/kvm/svm/svm.c | 99 +++++++++++++++++++++++++++---------------
 arch/x86/kvm/svm/svm.h |  6 +--
 3 files changed, 76 insertions(+), 37 deletions(-)

-- 
2.26.2

[PATCH 1/2] KVM: x86: add guest_cpuid_is_intel

[Maxim Levitsky]

This is similar to existing 'guest_cpuid_is_amd_or_hygon'

Signed-off-by: Maxim Levitsky <mlevitsk@redhat.com>
---
 arch/x86/kvm/cpuid.h | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/arch/x86/kvm/cpuid.h b/arch/x86/kvm/cpuid.h
index 2a0c5064497f..ded84d244f19 100644
--- a/arch/x86/kvm/cpuid.h
+++ b/arch/x86/kvm/cpuid.h
@@ -248,6 +248,14 @@ static inline bool guest_cpuid_is_amd_or_hygon(struct kvm_vcpu *vcpu)
 		is_guest_vendor_hygon(best->ebx, best->ecx, best->edx));
 }
 
+static inline bool guest_cpuid_is_intel(struct kvm_vcpu *vcpu)
+{
+	struct kvm_cpuid_entry2 *best;
+
+	best = kvm_find_cpuid_entry(vcpu, 0, 0);
+	return best && is_guest_vendor_intel(best->ebx, best->ecx, best->edx);
+}
+
 static inline int guest_cpuid_family(struct kvm_vcpu *vcpu)
 {
 	struct kvm_cpuid_entry2 *best;
-- 
2.26.2

[PATCH 2/2] KVM: nSVM: improve SYSENTER emulation on AMD

[Maxim Levitsky]

Currently to support Intel->AMD migration, if CPU vendor is GenuineIntel,
we emulate the full 64 value for MSR_IA32_SYSENTER_{EIP|ESP}
msrs, and we also emulate the sysenter/sysexit instruction in long mode.

(Emulator does still refuse to emulate sysenter in 64 bit mode, on the
ground that the code for that wasn't tested and likely has no users)

However when virtual vmload/vmsave is enabled, the vmload instruction will
update these 32 bit msrs without triggering their msr intercept,
which will lead to having stale values in kvm's shadow copy of these msrs,
which relies on the intercept to be up to date.

Fix/optimize this by doing the following:

1. Enable the MSR intercepts for SYSENTER MSRs iff vendor=GenuineIntel
   (This is both a tiny optimization and also ensures that in case
   the guest cpu vendor is AMD, the msrs will be 32 bit wide as
   AMD defined).

2. Store only high 32 bit part of these msrs on interception and combine
   it with hardware msr value on intercepted read/writes
   iff vendor=GenuineIntel.

3. Disable vmload/vmsave virtualization if vendor=GenuineIntel.
   (It is somewhat insane to set vendor=GenuineIntel and still enable
   SVM for the guest but well whatever).
   Then zero the high 32 bit parts when kvm intercepts and emulates vmload.

Thanks a lot to Paulo Bonzini for helping me with fixing this in the most
correct way.

This patch fixes nested migration of 32 bit nested guests, that was
broken because incorrect cached values of SYSENTER msrs were stored in
the migration stream if L1 changed these msrs with
vmload prior to L2 entry.

Signed-off-by: Maxim Levitsky <mlevitsk@redhat.com>
---
 arch/x86/kvm/svm/svm.c | 99 +++++++++++++++++++++++++++---------------
 arch/x86/kvm/svm/svm.h |  6 +--
 2 files changed, 68 insertions(+), 37 deletions(-)

diff --git a/arch/x86/kvm/svm/svm.c b/arch/x86/kvm/svm/svm.c
index 271196400495..6c39b0cd6ec6 100644
--- a/arch/x86/kvm/svm/svm.c
+++ b/arch/x86/kvm/svm/svm.c
@@ -95,6 +95,8 @@ static const struct svm_direct_access_msrs {
 } direct_access_msrs[MAX_DIRECT_ACCESS_MSRS] = {
 	{ .index = MSR_STAR,				.always = true  },
 	{ .index = MSR_IA32_SYSENTER_CS,		.always = true  },
+	{ .index = MSR_IA32_SYSENTER_EIP,		.always = false },
+	{ .index = MSR_IA32_SYSENTER_ESP,		.always = false },
 #ifdef CONFIG_X86_64
 	{ .index = MSR_GS_BASE,				.always = true  },
 	{ .index = MSR_FS_BASE,				.always = true  },
@@ -1258,16 +1260,6 @@ static void init_vmcb(struct kvm_vcpu *vcpu)
 	if (kvm_vcpu_apicv_active(vcpu))
 		avic_init_vmcb(svm);
 
-	/*
-	 * If hardware supports Virtual VMLOAD VMSAVE then enable it
-	 * in VMCB and clear intercepts to avoid #VMEXIT.
-	 */
-	if (vls) {
-		svm_clr_intercept(svm, INTERCEPT_VMLOAD);
-		svm_clr_intercept(svm, INTERCEPT_VMSAVE);
-		svm->vmcb->control.virt_ext |= VIRTUAL_VMLOAD_VMSAVE_ENABLE_MASK;
-	}
-
 	if (vgif) {
 		svm_clr_intercept(svm, INTERCEPT_STGI);
 		svm_clr_intercept(svm, INTERCEPT_CLGI);
@@ -2133,9 +2125,11 @@ static int vmload_vmsave_interception(struct kvm_vcpu *vcpu, bool vmload)
 
 	ret = kvm_skip_emulated_instruction(vcpu);
 
-	if (vmload)
+	if (vmload) {
 		nested_svm_vmloadsave(vmcb12, svm->vmcb);
-	else
+		svm->sysenter_eip_hi = 0;
+		svm->sysenter_esp_hi = 0;
+	} else
 		nested_svm_vmloadsave(svm->vmcb, vmcb12);
 
 	kvm_vcpu_unmap(vcpu, &map, true);
@@ -2677,10 +2671,14 @@ static int svm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
 		msr_info->data = svm->vmcb01.ptr->save.sysenter_cs;
 		break;
 	case MSR_IA32_SYSENTER_EIP:
-		msr_info->data = svm->sysenter_eip;
+		msr_info->data = (u32)svm->vmcb01.ptr->save.sysenter_eip;
+		if (guest_cpuid_is_intel(vcpu))
+			msr_info->data |= (u64)svm->sysenter_eip_hi << 32;
 		break;
 	case MSR_IA32_SYSENTER_ESP:
-		msr_info->data = svm->sysenter_esp;
+		msr_info->data = svm->vmcb01.ptr->save.sysenter_esp;
+		if (guest_cpuid_is_intel(vcpu))
+			msr_info->data |= (u64)svm->sysenter_esp_hi << 32;
 		break;
 	case MSR_TSC_AUX:
 		if (!boot_cpu_has(X86_FEATURE_RDTSCP))
@@ -2885,12 +2883,19 @@ static int svm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
 		svm->vmcb01.ptr->save.sysenter_cs = data;
 		break;
 	case MSR_IA32_SYSENTER_EIP:
-		svm->sysenter_eip = data;
-		svm->vmcb01.ptr->save.sysenter_eip = data;
+		svm->vmcb01.ptr->save.sysenter_eip = (u32)data;
+		/*
+		 * We only intercept the MSR_IA32_SYSENTER_{EIP|ESP} msrs
+		 * when we spoof an Intel vendor ID (for cross vendor migration).
+		 * In this case we use this intercept to track the high
+		 * 32 bit part of these msrs to support Intel's
+		 * implementation of SYSENTER/SYSEXIT.
+		 */
+		svm->sysenter_eip_hi = guest_cpuid_is_intel(vcpu) ? (data >> 32) : 0;
 		break;
 	case MSR_IA32_SYSENTER_ESP:
-		svm->sysenter_esp = data;
-		svm->vmcb01.ptr->save.sysenter_esp = data;
+		svm->vmcb01.ptr->save.sysenter_esp = (u32)data;
+		svm->sysenter_esp_hi = guest_cpuid_is_intel(vcpu) ? (data >> 32) : 0;
 		break;
 	case MSR_TSC_AUX:
 		if (!boot_cpu_has(X86_FEATURE_RDTSCP))
@@ -4009,24 +4014,50 @@ static void svm_vcpu_after_set_cpuid(struct kvm_vcpu *vcpu)
 			vcpu->arch.reserved_gpa_bits &= ~(1UL << (best->ebx & 0x3f));
 	}
 
-	if (!kvm_vcpu_apicv_active(vcpu))
-		return;
+	if (kvm_vcpu_apicv_active(vcpu)) {
+		/*
+		 * AVIC does not work with an x2APIC mode guest. If the X2APIC feature
+		 * is exposed to the guest, disable AVIC.
+		 */
+		if (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC))
+			kvm_request_apicv_update(vcpu->kvm, false,
+						 APICV_INHIBIT_REASON_X2APIC);
 
-	/*
-	 * AVIC does not work with an x2APIC mode guest. If the X2APIC feature
-	 * is exposed to the guest, disable AVIC.
-	 */
-	if (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC))
-		kvm_request_apicv_update(vcpu->kvm, false,
-					 APICV_INHIBIT_REASON_X2APIC);
+		/*
+		 * Currently, AVIC does not work with nested virtualization.
+		 * So, we disable AVIC when cpuid for SVM is set in the L1 guest.
+		 */
+		if (nested && guest_cpuid_has(vcpu, X86_FEATURE_SVM))
+			kvm_request_apicv_update(vcpu->kvm, false,
+						 APICV_INHIBIT_REASON_NESTED);
+	}
 
-	/*
-	 * Currently, AVIC does not work with nested virtualization.
-	 * So, we disable AVIC when cpuid for SVM is set in the L1 guest.
-	 */
-	if (nested && guest_cpuid_has(vcpu, X86_FEATURE_SVM))
-		kvm_request_apicv_update(vcpu->kvm, false,
-					 APICV_INHIBIT_REASON_NESTED);
+	if (guest_cpuid_is_intel(vcpu)) {
+		/*
+		 * We must intercept SYSENTER_EIP and SYSENTER_ESP
+		 * accesses because the processor only stores 32 bits.
+		 * For the same reason we cannot use virtual VMLOAD/VMSAVE.
+		 */
+		svm_set_intercept(svm, INTERCEPT_VMLOAD);
+		svm_set_intercept(svm, INTERCEPT_VMSAVE);
+		svm->vmcb->control.virt_ext &= ~VIRTUAL_VMLOAD_VMSAVE_ENABLE_MASK;
+
+		set_msr_interception(vcpu, svm->msrpm, MSR_IA32_SYSENTER_EIP, 0, 0);
+		set_msr_interception(vcpu, svm->msrpm, MSR_IA32_SYSENTER_ESP, 0, 0);
+	} else {
+		/*
+		 * If hardware supports Virtual VMLOAD VMSAVE then enable it
+		 * in VMCB and clear intercepts to avoid #VMEXIT.
+		 */
+		if (vls) {
+			svm_clr_intercept(svm, INTERCEPT_VMLOAD);
+			svm_clr_intercept(svm, INTERCEPT_VMSAVE);
+			svm->vmcb->control.virt_ext |= VIRTUAL_VMLOAD_VMSAVE_ENABLE_MASK;
+		}
+		/* No need to intercept these MSRs */
+		set_msr_interception(vcpu, svm->msrpm, MSR_IA32_SYSENTER_EIP, 1, 1);
+		set_msr_interception(vcpu, svm->msrpm, MSR_IA32_SYSENTER_ESP, 1, 1);
+	}
 }
 
 static bool svm_has_wbinvd_exit(void)
diff --git a/arch/x86/kvm/svm/svm.h b/arch/x86/kvm/svm/svm.h
index 8e276c4fb33d..fffdd5fb446d 100644
--- a/arch/x86/kvm/svm/svm.h
+++ b/arch/x86/kvm/svm/svm.h
@@ -28,7 +28,7 @@ static const u32 host_save_user_msrs[] = {
 };
 #define NR_HOST_SAVE_USER_MSRS ARRAY_SIZE(host_save_user_msrs)
 
-#define MAX_DIRECT_ACCESS_MSRS	18
+#define MAX_DIRECT_ACCESS_MSRS	20
 #define MSRPM_OFFSETS	16
 extern u32 msrpm_offsets[MSRPM_OFFSETS] __read_mostly;
 extern bool npt_enabled;
@@ -116,8 +116,8 @@ struct vcpu_svm {
 	struct kvm_vmcb_info *current_vmcb;
 	struct svm_cpu_data *svm_data;
 	u32 asid;
-	uint64_t sysenter_esp;
-	uint64_t sysenter_eip;
+	u32 sysenter_esp_hi;
+	u32 sysenter_eip_hi;
 	uint64_t tsc_aux;
 
 	u64 msr_decfg;
-- 
2.26.2

Re: [PATCH 0/2] KVM: x86: nSVM: fixes for SYSENTER emulation

[Maxim Levitsky]

On Thu, 2021-04-01 at 14:16 +0300, Maxim Levitsky wrote:
> This is a result of a deep rabbit hole dive in regard to why
> currently the nested migration of 32 bit guests
> is totally broken on AMD.

Please ignore this patch series, I didn't update the patch version.

Best regards,
	Maxim Levitsky

> 
> It turns out that due to slight differences between the original AMD64
> implementation and the Intel's remake, SYSENTER instruction behaves a
> bit differently on Intel, and to support migration from Intel to AMD we
> try to emulate those differences away.
> 
> Sadly that collides with virtual vmload/vmsave feature that is used in nesting.
> The problem was that when it is enabled,
> on migration (and otherwise when userspace reads MSR_IA32_SYSENTER_{EIP|ESP},
> wrong value were returned, which leads to #DF in the
> nested guest when the wrong value is loaded back.
> 
> The patch I prepared carefully fixes this, by mostly disabling that
> SYSCALL emulation when we don't spoof the Intel's vendor ID, and if we do,
> and yet somehow SVM is enabled (this is a very rare edge case), then
> virtual vmload/save is force disabled.
> 
> V2: incorporated review feedback from Paulo.
> 
> Best regards,
>         Maxim Levitsky
> 
> Maxim Levitsky (2):
>   KVM: x86: add guest_cpuid_is_intel
>   KVM: nSVM: improve SYSENTER emulation on AMD
> 
>  arch/x86/kvm/cpuid.h   |  8 ++++
>  arch/x86/kvm/svm/svm.c | 99 +++++++++++++++++++++++++++---------------
>  arch/x86/kvm/svm/svm.h |  6 +--
>  3 files changed, 76 insertions(+), 37 deletions(-)
> 
> -- 
> 2.26.2
> 

[PATCH 0/2] KVM: x86: nSVM: fixes for SYSENTER emulation

[Maxim Levitsky]

This is a result of deep rabbit hole dive in regard to why currently the
nested migration of 32 bit guests is totally broken on AMD.

It turns out that due to slight differences between the original AMD64
implementation and the Intel's remake, SYSENTER instruction behaves a
bit differently on Intel, and to support migration from Intel to AMD we
try to emulate those differences away.

Sadly that collides with virtual vmload/vmsave feature that is used in nesting
such as on migration (and otherwise when userspace
reads MSR_IA32_SYSENTER_EIP/MSR_IA32_SYSENTER_ESP), wrong value is returned,
which leads to #DF in the nested guest when the wrong value is loaded back.

The patch I prepared carefully fixes this, by mostly disabling that SYSCALL
emulation when we don't spoof Intel's vendor ID, and if we do, and yet somehow
SVM is enabled (this is very rare corner case), then virtual vmload/save is
force disabled.

Best regards,
	Maxim Levitsky

Maxim Levitsky (2):
  KVM: x86: add guest_cpuid_is_intel
  KVM: nSVM: improve SYSENTER emulation on AMD

 arch/x86/kvm/cpuid.h   |  8 ++++
 arch/x86/kvm/svm/svm.c | 97 ++++++++++++++++++++++++++++--------------
 arch/x86/kvm/svm/svm.h |  7 +--
 3 files changed, 77 insertions(+), 35 deletions(-)

-- 
2.26.2