Skip to contentSkip to conversation

Add support for XSAVES to AMD and unify it with Intel

PATCH Applied Aaron Lewis 2 participants 12 messages Oct 21, 2019 – Oct 22, 2019

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

v3 cover Add support for XSAVES to AMD and unify it with Intel

Unify AMD's and Intel's approach for supporting XSAVES. To do this change Intel's approach from using the MSR-load areas to writing the guest/host values to IA32_XSS on a VM-enter/VM-exit. Switching to this strategy allows for a common approach between both AMD and Intel. Additionally, define svm_xsaves_supported() based on AMD's feedback, and add vcpu->arch.xsaves_enabled to track whether XSAVES is enabled in the guest.

This change sets up IA32_XSS to be a non-zero value in the future, which may happen sooner than later with support for guest CET feature being added.

v2 -> v3:
- Remove guest_xcr0_loaded from kvm_vcpu.
- Add vcpu->arch.xsaves_enabled.
- Add staged rollout to load the hardware IA32_XSS MSR with guest/host values on VM-entry and VM-exit:
1) Introduce vcpu->arch->xsaves_enabled.
2) Add svm implementation for switching between guest and host IA32_XSS.
3) Add vmx implementation for switching between guest and host IA32_XSS.
4) Remove svm and vmx implementation and add it to common code.

v1 -> v2:
- Add the flag xsaves_enabled to kvm_vcpu_arch to track when XSAVES is enabled in the guest, whether or not XSAVES is enumerated in the guest CPUID.
- Remove code that sets the X86_FEATURE_XSAVES bit in the guest CPUID which was added in patch "Enumerate XSAVES in guest CPUID when it is available to the guest". As a result we no longer need that patch.
- Added a comment to kvm_set_msr_common to describe how to save/restore PT MSRS without using XSAVES/XRSTORS.
- Added more comments to the "Add support for XSAVES on AMD" patch.
- Replaced vcpu_set_msr_expect_result() with _vcpu_set_msr() in the test library.

Aaron Lewis (9):
KVM: x86: Introduce vcpu->arch.xsaves_enabled
KVM: VMX: Fix conditions for guest IA32_XSS support
KVM: x86: Remove unneeded kvm_vcpu variable, guest_xcr0_loaded
KVM: SVM: Use wrmsr for switching between guest and host IA32_XSS on AMD
KVM: VMX: Use wrmsr for switching between guest and host IA32_XSS on Intel
KVM: x86: Move IA32_XSS-swapping on VM-entry/VM-exit to common x86 code
kvm: x86: Move IA32_XSS to kvm_{get,set}_msr_common
kvm: svm: Update svm_xsaves_supported
kvm: tests: Add test to verify MSR_IA32_XSS

 arch/x86/include/asm/kvm_host.h               |  1 +
 arch/x86/kvm/svm.c                            |  9 ++-
 arch/x86/kvm/vmx/vmx.c                        | 41 ++--------
 arch/x86/kvm/x86.c                            | 52 ++++++++++---
 arch/x86/kvm/x86.h                            |  4 +-
 include/linux/kvm_host.h                      |  1 -
 tools/testing/selftests/kvm/.gitignore        |  1 +
 tools/testing/selftests/kvm/Makefile          |  1 +
 .../selftests/kvm/include/x86_64/processor.h  |  7 +-
 .../selftests/kvm/lib/x86_64/processor.c      | 72 +++++++++++++++---
 .../selftests/kvm/x86_64/xss_msr_test.c       | 76 +++++++++++++++++++
 11 files changed, 205 insertions(+), 60 deletions(-)
 create mode 100644 tools/testing/selftests/kvm/x86_64/xss_msr_test.c
Signature · 1 line
2.23.0.866.gb869b98d4c-goog
#1 lore raw

PATCH v3 1/9 KVM: x86: Introduce vcpu->arch.xsaves_enabled

Cache whether XSAVES is enabled in the guest by adding xsaves_enabled to vcpu->arch.

Change-Id If4638e0901c28a4494dad2e103e2c075e8ab5d68
---
 arch/x86/include/asm/kvm_host.h | 1 +
 arch/x86/kvm/svm.c              | 3 +++
 arch/x86/kvm/vmx/vmx.c          | 5 +++++
 3 files changed, 9 insertions(+)
arch/x86/include/asm/kvm_host.h +1 -0
index 50eb430b0ad8..634c2598e389 100644
@@ -562,6 +562,7 @@ struct kvm_vcpu_arch {
 	u64 smbase;
 	u64 smi_count;
 	bool tpr_access_reporting;
+	bool xsaves_enabled;
 	u64 ia32_xss;
 	u64 microcode_version;
 	u64 arch_capabilities;
arch/x86/kvm/svm.c +3 -0
index f8ecb6df5106..f64041368594 100644
@@ -5887,6 +5887,9 @@ static void svm_cpuid_update(struct kvm_vcpu *vcpu)
 {
 	struct vcpu_svm *svm = to_svm(vcpu);
 
+	vcpu->arch.xsaves_enabled = guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
+				    boot_cpu_has(X86_FEATURE_XSAVES);
+
 	/* Update nrips enabled cache */
 	svm->nrips_enabled = !!guest_cpuid_has(&svm->vcpu, X86_FEATURE_NRIPS);
 
arch/x86/kvm/vmx/vmx.c +5 -0
index e7970a2e8eae..34525af44353 100644
@@ -4040,6 +4040,8 @@ static void vmx_compute_secondary_exec_control(struct vcpu_vmx *vmx)
 			guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
 			guest_cpuid_has(vcpu, X86_FEATURE_XSAVES);
 
+		vcpu->arch.xsaves_enabled = xsaves_enabled;
+
 		if (!xsaves_enabled)
 			exec_control &= ~SECONDARY_EXEC_XSAVES;
 
@@ -7093,6 +7095,9 @@ static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
 {
 	struct vcpu_vmx *vmx = to_vmx(vcpu);
 
+	/* xsaves_enabled is recomputed in vmx_compute_secondary_exec_control(). */
+	vcpu->arch.xsaves_enabled = false;
+
 	if (cpu_has_secondary_exec_ctrls()) {
 		vmx_compute_secondary_exec_control(vmx);
 		vmcs_set_secondary_exec_control(vmx);
Signature · 1 line
2.23.0.866.gb869b98d4c-goog
#2 lore raw

PATCH v3 2/9 KVM: VMX: Fix conditions for guest IA32_XSS support

In reply to Aaron Lewis

Volume 4 of the SDM says that IA32_XSS is supported
if CPUID(EAX=0DH,ECX=1):EAX.XSS[bit 3] is set, so only the X86_FEATURE_XSAVES check is necessary (X86_FEATURE_XSAVES is the Linux name for CPUID(EAX=0DH,ECX=1):EAX.XSS[bit 3]).

Fixes 4d763b168e9c5 ("KVM: VMX: check CPUID before allowing read/write of IA32_XSS")
Change-Id I9059b9f2e3595e4b09a4cdcf14b933b22ebad419
---
 arch/x86/kvm/vmx/vmx.c | 24 +++++++++++-------------
 1 file changed, 11 insertions(+), 13 deletions(-)
arch/x86/kvm/vmx/vmx.c +11 -13
index 34525af44353..a9b070001c3e 100644
@@ -1821,10 +1821,8 @@ static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
 		return vmx_get_vmx_msr(&vmx->nested.msrs, msr_info->index,
 				       &msr_info->data);
 	case MSR_IA32_XSS:
-		if (!vmx_xsaves_supported() ||
-		    (!msr_info->host_initiated &&
-		     !(guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
-		       guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))))
+		if (!msr_info->host_initiated &&
+		    !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
 			return 1;
 		msr_info->data = vcpu->arch.ia32_xss;
 		break;
@@ -2064,10 +2062,8 @@ static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
 			return 1;
 		return vmx_set_vmx_msr(vcpu, msr_index, data);
 	case MSR_IA32_XSS:
-		if (!vmx_xsaves_supported() ||
-		    (!msr_info->host_initiated &&
-		     !(guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
-		       guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))))
+		if (!msr_info->host_initiated &&
+		    !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
 			return 1;
 		/*
 		 * The only supported bit as of Skylake is bit 8, but
@@ -2076,11 +2072,13 @@ static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
 		if (data != 0)
 			return 1;
 		vcpu->arch.ia32_xss = data;
-		if (vcpu->arch.ia32_xss != host_xss)
-			add_atomic_switch_msr(vmx, MSR_IA32_XSS,
-				vcpu->arch.ia32_xss, host_xss, false);
-		else
-			clear_atomic_switch_msr(vmx, MSR_IA32_XSS);
+		if (vcpu->arch.xsaves_enabled) {
+			if (vcpu->arch.ia32_xss != host_xss)
+				add_atomic_switch_msr(vmx, MSR_IA32_XSS,
+					vcpu->arch.ia32_xss, host_xss, false);
+			else
+				clear_atomic_switch_msr(vmx, MSR_IA32_XSS);
+		}
 		break;
 	case MSR_IA32_RTIT_CTL:
 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
Signature · 1 line
2.23.0.866.gb869b98d4c-goog
maintainer
#3 lore raw

Re: [PATCH v3 2/9] KVM: VMX: Fix conditions for guest IA32_XSS support

Quoted 53 lines · Aaron Lewis

Volume 4 of the SDM says that IA32_XSS is supported
if CPUID(EAX=0DH,ECX=1):EAX.XSS[bit 3] is set, so only the X86_FEATURE_XSAVES check is necessary (X86_FEATURE_XSAVES is the Linux name for CPUID(EAX=0DH,ECX=1):EAX.XSS[bit 3]).

Fixes 4d763b168e9c5 ("KVM: VMX: check CPUID before allowing read/write of IA32_XSS")
Change-Id I9059b9f2e3595e4b09a4cdcf14b933b22ebad419
---
 arch/x86/kvm/vmx/vmx.c | 24 +++++++++++-------------
 1 file changed, 11 insertions(+), 13 deletions(-)
arch/x86/kvm/vmx/vmx.c +4 -13
index 34525af44353..a9b070001c3e 100644
@@ -1821,10 +1821,8 @@ static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
 		return vmx_get_vmx_msr(&vmx->nested.msrs, msr_info->index,
 				       &msr_info->data);
 	case MSR_IA32_XSS:
-		if (!vmx_xsaves_supported() ||
-		    (!msr_info->host_initiated &&
-		     !(guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
-		       guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))))
+		if (!msr_info->host_initiated &&
+		    !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
 			return 1;
 		msr_info->data = vcpu->arch.ia32_xss;
 		break;
@@ -2064,10 +2062,8 @@ static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
 			return 1;
 		return vmx_set_vmx_msr(vcpu, msr_index, data);
 	case MSR_IA32_XSS:
-		if (!vmx_xsaves_supported() ||
-		    (!msr_info->host_initiated &&
-		     !(guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
-		       guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))))
+		if (!msr_info->host_initiated &&
+		    !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
 			return 1;
 		/*
 		 * The only supported bit as of Skylake is bit 8, but
@@ -2076,11 +2072,13 @@ static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
 		if (data != 0)
 			return 1;
 		vcpu->arch.ia32_xss = data;
-		if (vcpu->arch.ia32_xss != host_xss)
-			add_atomic_switch_msr(vmx, MSR_IA32_XSS,
-				vcpu->arch.ia32_xss, host_xss, false);
-		else
-			clear_atomic_switch_msr(vmx, MSR_IA32_XSS);

arch/x86/kvm/vmx/vmx.c static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info) · line 2084

Quoted 10 lines
arch/x86/kvm/vmx/vmx.c +7 -0
+		if (vcpu->arch.xsaves_enabled) {
+			if (vcpu->arch.ia32_xss != host_xss)
+				add_atomic_switch_msr(vmx, MSR_IA32_XSS,
+					vcpu->arch.ia32_xss, host_xss, false);
+			else
+				clear_atomic_switch_msr(vmx, MSR_IA32_XSS);
+		}
 		break;
 	case MSR_IA32_RTIT_CTL:
 		if ((pt_mode != PT_MODE_HOST_GUEST) ||

The last hunk technically doesn't belong in this patch, but okay.

Paolo

#4 lore raw

PATCH v3 3/9 KVM: x86: Remove unneeded kvm_vcpu variable, guest_xcr0_loaded

In reply to Aaron Lewis

The kvm_vcpu variable, guest_xcr0_loaded, is a waste of an 'int' and a conditional branch. VMX and SVM are the only users, and both unconditionally pair kvm_load_guest_xcr0() with kvm_put_guest_xcr0() making this check unnecessary. Without this variable, the predicates in kvm_load_guest_xcr0 and kvm_put_guest_xcr0 should match.

Change-Id I7b1eb9b62969d7bbb2850f27e42f863421641b23
---
 arch/x86/kvm/x86.c       | 16 +++++-----------
 include/linux/kvm_host.h |  1 -
 2 files changed, 5 insertions(+), 12 deletions(-)
arch/x86/kvm/x86.c +5 -11
index 661e2bf38526..39eac7b2aa01 100644
@@ -815,22 +815,16 @@ EXPORT_SYMBOL_GPL(kvm_lmsw);
 void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
 {
 	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
-			!vcpu->guest_xcr0_loaded) {
-		/* kvm_set_xcr() also depends on this */
-		if (vcpu->arch.xcr0 != host_xcr0)
-			xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
-		vcpu->guest_xcr0_loaded = 1;
-	}
+	    vcpu->arch.xcr0 != host_xcr0)
+		xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
 }
 EXPORT_SYMBOL_GPL(kvm_load_guest_xcr0);
 
 void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
 {
-	if (vcpu->guest_xcr0_loaded) {
-		if (vcpu->arch.xcr0 != host_xcr0)
-			xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
-		vcpu->guest_xcr0_loaded = 0;
-	}
+	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
+	    vcpu->arch.xcr0 != host_xcr0)
+		xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
 }
 EXPORT_SYMBOL_GPL(kvm_put_guest_xcr0);
 
include/linux/kvm_host.h +0 -1
index 719fc3e15ea4..d2017302996c 100644
@@ -278,7 +278,6 @@ struct kvm_vcpu {
 	struct mutex mutex;
 	struct kvm_run *run;
 
-	int guest_xcr0_loaded;
 	struct swait_queue_head wq;
 	struct pid __rcu *pid;
 	int sigset_active;
Signature · 1 line
2.23.0.866.gb869b98d4c-goog
#5 lore raw

PATCH v3 4/9 KVM: SVM: Use wrmsr for switching between guest and host IA32_XSS on AMD

In reply to Aaron Lewis

When the guest can execute the XSAVES/XRSTORS instructions, set the hardware IA32_XSS MSR to guest/host values on VM-entry/VM-exit.

Note that vcpu->arch.ia32_xss is currently guaranteed to be 0 on AMD, since there is no way to change it.

Change-Id Id51a782462086e6d7a3ab621838e200f1c005afd
---
 arch/x86/kvm/svm.c | 23 +++++++++++++++++++++++
 1 file changed, 23 insertions(+)
arch/x86/kvm/svm.c +23 -0
index f64041368594..2702ebba24ba 100644
@@ -115,6 +115,8 @@ MODULE_DEVICE_TABLE(x86cpu, svm_cpu_id);
 
 static bool erratum_383_found __read_mostly;
 
+static u64 __read_mostly host_xss;
+
 static const u32 host_save_user_msrs[] = {
 #ifdef CONFIG_X86_64
 	MSR_STAR, MSR_LSTAR, MSR_CSTAR, MSR_SYSCALL_MASK, MSR_KERNEL_GS_BASE,
@@ -1400,6 +1402,9 @@ static __init int svm_hardware_setup(void)
 			pr_info("Virtual GIF supported\n");
 	}
 
+	if (boot_cpu_has(X86_FEATURE_XSAVES))
+		rdmsrl(MSR_IA32_XSS, host_xss);
+
 	return 0;
 
 err:
@@ -5590,6 +5595,22 @@ static void svm_cancel_injection(struct kvm_vcpu *vcpu)
 	svm_complete_interrupts(svm);
 }
 
+static void svm_load_guest_xss(struct kvm_vcpu *vcpu)
+{
+	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
+	    vcpu->arch.xsaves_enabled &&
+	    vcpu->arch.ia32_xss != host_xss)
+		wrmsrl(MSR_IA32_XSS, vcpu->arch.ia32_xss);
+}
+
+static void svm_load_host_xss(struct kvm_vcpu *vcpu)
+{
+	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
+	    vcpu->arch.xsaves_enabled &&
+	    vcpu->arch.ia32_xss != host_xss)
+		wrmsrl(MSR_IA32_XSS, host_xss);
+}
+
 static void svm_vcpu_run(struct kvm_vcpu *vcpu)
 {
 	struct vcpu_svm *svm = to_svm(vcpu);
@@ -5629,6 +5650,7 @@ static void svm_vcpu_run(struct kvm_vcpu *vcpu)
 
 	clgi();
 	kvm_load_guest_xcr0(vcpu);
+	svm_load_guest_xss(vcpu);
 
 	if (lapic_in_kernel(vcpu) &&
 		vcpu->arch.apic->lapic_timer.timer_advance_ns)
@@ -5778,6 +5800,7 @@ static void svm_vcpu_run(struct kvm_vcpu *vcpu)
 	if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI))
 		kvm_before_interrupt(&svm->vcpu);
 
+	svm_load_host_xss(vcpu);
 	kvm_put_guest_xcr0(vcpu);
 	stgi();
 
Signature · 1 line
2.23.0.866.gb869b98d4c-goog
#6 lore raw

PATCH v3 5/9 KVM: VMX: Use wrmsr for switching between guest and host IA32_XSS on Intel

In reply to Aaron Lewis

When the guest can execute the XSAVES/XRSTORS instructions, use wrmsr to set the hardware IA32_XSS MSR to guest/host values on VM-entry/VM-exit, rather than the MSR-load areas. By using the same approach as AMD, we will be able to use a common implementation for both (in the next patch).

Change-Id I9447d104b2615c04e39e4af0c911e1e7309bf464
---
 arch/x86/kvm/vmx/vmx.c | 25 ++++++++++++++++++-------
 1 file changed, 18 insertions(+), 7 deletions(-)
arch/x86/kvm/vmx/vmx.c +18 -7
index a9b070001c3e..f3cd2e372c4a 100644
@@ -2072,13 +2072,6 @@ static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
 		if (data != 0)
 			return 1;
 		vcpu->arch.ia32_xss = data;
-		if (vcpu->arch.xsaves_enabled) {
-			if (vcpu->arch.ia32_xss != host_xss)
-				add_atomic_switch_msr(vmx, MSR_IA32_XSS,
-					vcpu->arch.ia32_xss, host_xss, false);
-			else
-				clear_atomic_switch_msr(vmx, MSR_IA32_XSS);
-		}
 		break;
 	case MSR_IA32_RTIT_CTL:
 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
@@ -6492,6 +6485,22 @@ void vmx_update_host_rsp(struct vcpu_vmx *vmx, unsigned long host_rsp)
 	}
 }
 
+static void vmx_load_guest_xss(struct kvm_vcpu *vcpu)
+{
+	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
+	    vcpu->arch.xsaves_enabled &&
+	    vcpu->arch.ia32_xss != host_xss)
+		wrmsrl(MSR_IA32_XSS, vcpu->arch.ia32_xss);
+}
+
+static void vmx_load_host_xss(struct kvm_vcpu *vcpu)
+{
+	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
+	    vcpu->arch.xsaves_enabled &&
+	    vcpu->arch.ia32_xss != host_xss)
+		wrmsrl(MSR_IA32_XSS, host_xss);
+}
+
 bool __vmx_vcpu_run(struct vcpu_vmx *vmx, unsigned long *regs, bool launched);
 
 static void vmx_vcpu_run(struct kvm_vcpu *vcpu)
@@ -6543,6 +6552,7 @@ static void vmx_vcpu_run(struct kvm_vcpu *vcpu)
 		vmx_set_interrupt_shadow(vcpu, 0);
 
 	kvm_load_guest_xcr0(vcpu);
+	vmx_load_guest_xss(vcpu);
 
 	if (static_cpu_has(X86_FEATURE_PKU) &&
 	    kvm_read_cr4_bits(vcpu, X86_CR4_PKE) &&
@@ -6649,6 +6659,7 @@ static void vmx_vcpu_run(struct kvm_vcpu *vcpu)
 			__write_pkru(vmx->host_pkru);
 	}
 
+	vmx_load_host_xss(vcpu);
 	kvm_put_guest_xcr0(vcpu);
 
 	vmx->nested.nested_run_pending = 0;
Signature · 1 line
2.23.0.866.gb869b98d4c-goog
#7 lore raw

PATCH v3 6/9 KVM: x86: Move IA32_XSS-swapping on VM-entry/VM-exit to common x86 code

In reply to Aaron Lewis

Hoist the vendor-specific code related to loading the hardware IA32_XSS MSR with guest/host values on VM-entry/VM-exit to common x86 code.

Change-Id Ic6e3430833955b98eb9b79ae6715cf2a3fdd6d82
---
 arch/x86/kvm/svm.c     | 27 ++-------------------------
 arch/x86/kvm/vmx/vmx.c | 27 ++-------------------------
 arch/x86/kvm/x86.c     | 38 ++++++++++++++++++++++++++++----------
 arch/x86/kvm/x86.h     |  4 ++--
 4 files changed, 34 insertions(+), 62 deletions(-)
arch/x86/kvm/svm.c +2 -25
index 2702ebba24ba..36d1cfd45c60 100644
@@ -115,8 +115,6 @@ MODULE_DEVICE_TABLE(x86cpu, svm_cpu_id);
 
 static bool erratum_383_found __read_mostly;
 
-static u64 __read_mostly host_xss;
-
 static const u32 host_save_user_msrs[] = {
 #ifdef CONFIG_X86_64
 	MSR_STAR, MSR_LSTAR, MSR_CSTAR, MSR_SYSCALL_MASK, MSR_KERNEL_GS_BASE,
@@ -1402,9 +1400,6 @@ static __init int svm_hardware_setup(void)
 			pr_info("Virtual GIF supported\n");
 	}
 
-	if (boot_cpu_has(X86_FEATURE_XSAVES))
-		rdmsrl(MSR_IA32_XSS, host_xss);
-
 	return 0;
 
 err:
@@ -5595,22 +5590,6 @@ static void svm_cancel_injection(struct kvm_vcpu *vcpu)
 	svm_complete_interrupts(svm);
 }
 
-static void svm_load_guest_xss(struct kvm_vcpu *vcpu)
-{
-	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
-	    vcpu->arch.xsaves_enabled &&
-	    vcpu->arch.ia32_xss != host_xss)
-		wrmsrl(MSR_IA32_XSS, vcpu->arch.ia32_xss);
-}
-
-static void svm_load_host_xss(struct kvm_vcpu *vcpu)
-{
-	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
-	    vcpu->arch.xsaves_enabled &&
-	    vcpu->arch.ia32_xss != host_xss)
-		wrmsrl(MSR_IA32_XSS, host_xss);
-}
-
 static void svm_vcpu_run(struct kvm_vcpu *vcpu)
 {
 	struct vcpu_svm *svm = to_svm(vcpu);
@@ -5649,8 +5628,7 @@ static void svm_vcpu_run(struct kvm_vcpu *vcpu)
 	svm->vmcb->save.cr2 = vcpu->arch.cr2;
 
 	clgi();
-	kvm_load_guest_xcr0(vcpu);
-	svm_load_guest_xss(vcpu);
+	kvm_load_guest_xsave_state(vcpu);
 
 	if (lapic_in_kernel(vcpu) &&
 		vcpu->arch.apic->lapic_timer.timer_advance_ns)
@@ -5800,8 +5778,7 @@ static void svm_vcpu_run(struct kvm_vcpu *vcpu)
 	if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI))
 		kvm_before_interrupt(&svm->vcpu);
 
-	svm_load_host_xss(vcpu);
-	kvm_put_guest_xcr0(vcpu);
+	kvm_load_host_xsave_state(vcpu);
 	stgi();
 
 	/* Any pending NMI will happen here */
arch/x86/kvm/vmx/vmx.c +2 -25
index f3cd2e372c4a..f7d292ac9921 100644
@@ -106,8 +106,6 @@ module_param(enable_apicv, bool, S_IRUGO);
 static bool __read_mostly nested = 1;
 module_param(nested, bool, S_IRUGO);
 
-static u64 __read_mostly host_xss;
-
 bool __read_mostly enable_pml = 1;
 module_param_named(pml, enable_pml, bool, S_IRUGO);
 
@@ -6485,22 +6483,6 @@ void vmx_update_host_rsp(struct vcpu_vmx *vmx, unsigned long host_rsp)
 	}
 }
 
-static void vmx_load_guest_xss(struct kvm_vcpu *vcpu)
-{
-	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
-	    vcpu->arch.xsaves_enabled &&
-	    vcpu->arch.ia32_xss != host_xss)
-		wrmsrl(MSR_IA32_XSS, vcpu->arch.ia32_xss);
-}
-
-static void vmx_load_host_xss(struct kvm_vcpu *vcpu)
-{
-	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
-	    vcpu->arch.xsaves_enabled &&
-	    vcpu->arch.ia32_xss != host_xss)
-		wrmsrl(MSR_IA32_XSS, host_xss);
-}
-
 bool __vmx_vcpu_run(struct vcpu_vmx *vmx, unsigned long *regs, bool launched);
 
 static void vmx_vcpu_run(struct kvm_vcpu *vcpu)
@@ -6551,8 +6533,7 @@ static void vmx_vcpu_run(struct kvm_vcpu *vcpu)
 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
 		vmx_set_interrupt_shadow(vcpu, 0);
 
-	kvm_load_guest_xcr0(vcpu);
-	vmx_load_guest_xss(vcpu);
+	kvm_load_guest_xsave_state(vcpu);
 
 	if (static_cpu_has(X86_FEATURE_PKU) &&
 	    kvm_read_cr4_bits(vcpu, X86_CR4_PKE) &&
@@ -6659,8 +6640,7 @@ static void vmx_vcpu_run(struct kvm_vcpu *vcpu)
 			__write_pkru(vmx->host_pkru);
 	}
 
-	vmx_load_host_xss(vcpu);
-	kvm_put_guest_xcr0(vcpu);
+	kvm_load_host_xsave_state(vcpu);
 
 	vmx->nested.nested_run_pending = 0;
 	vmx->idt_vectoring_info = 0;
@@ -7615,9 +7595,6 @@ static __init int hardware_setup(void)
 		WARN_ONCE(host_bndcfgs, "KVM: BNDCFGS in host will be lost");
 	}
 
-	if (boot_cpu_has(X86_FEATURE_XSAVES))
-		rdmsrl(MSR_IA32_XSS, host_xss);
-
 	if (!cpu_has_vmx_vpid() || !cpu_has_vmx_invvpid() ||
 	    !(cpu_has_vmx_invvpid_single() || cpu_has_vmx_invvpid_global()))
 		enable_vpid = 0;
arch/x86/kvm/x86.c +28 -10
index 39eac7b2aa01..259a30e4d3a9 100644
@@ -176,6 +176,8 @@ struct kvm_shared_msrs {
 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
 static struct kvm_shared_msrs __percpu *shared_msrs;
 
+static u64 __read_mostly host_xss;
+
 struct kvm_stats_debugfs_item debugfs_entries[] = {
 	{ "pf_fixed", VCPU_STAT(pf_fixed) },
 	{ "pf_guest", VCPU_STAT(pf_guest) },
@@ -812,21 +814,34 @@ void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 }
 EXPORT_SYMBOL_GPL(kvm_lmsw);
 
-void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
+void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu)
 {
-	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
-	    vcpu->arch.xcr0 != host_xcr0)
-		xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
+	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
+
+		if (vcpu->arch.xcr0 != host_xcr0)
+			xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
+
+		if (vcpu->arch.xsaves_enabled &&
+		    vcpu->arch.ia32_xss != host_xss)
+			wrmsrl(MSR_IA32_XSS, vcpu->arch.ia32_xss);
+	}
 }
-EXPORT_SYMBOL_GPL(kvm_load_guest_xcr0);
+EXPORT_SYMBOL_GPL(kvm_load_guest_xsave_state);
 
-void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
+void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu)
 {
-	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
-	    vcpu->arch.xcr0 != host_xcr0)
-		xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
+	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
+
+		if (vcpu->arch.xcr0 != host_xcr0)
+			xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
+
+		if (vcpu->arch.xsaves_enabled &&
+		    vcpu->arch.ia32_xss != host_xss)
+			wrmsrl(MSR_IA32_XSS, host_xss);
+	}
+
 }
-EXPORT_SYMBOL_GPL(kvm_put_guest_xcr0);
+EXPORT_SYMBOL_GPL(kvm_load_host_xsave_state);
 
 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
 {
@@ -9287,6 +9302,9 @@ int kvm_arch_hardware_setup(void)
 		kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
 	}
 
+	if (boot_cpu_has(X86_FEATURE_XSAVES))
+		rdmsrl(MSR_IA32_XSS, host_xss);
+
 	kvm_init_msr_list();
 	return 0;
 }
arch/x86/kvm/x86.h +2 -2
index dbf7442a822b..250c2c932e46 100644
@@ -366,7 +366,7 @@ static inline bool kvm_pat_valid(u64 data)
 	return (data | ((data & 0x0202020202020202ull) << 1)) == data;
 }
 
-void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu);
-void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu);
+void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu);
+void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu);
 
 #endif
Signature · 1 line
2.23.0.866.gb869b98d4c-goog
#8 lore raw

PATCH v3 7/9 kvm: x86: Move IA32_XSS to kvm_{get,set}_msr_common

In reply to Aaron Lewis

Hoist support for RDMSR/WRMSR of IA32_XSS from vmx into common code so that it can be used for svm as well.

Right now, kvm only allows the guest IA32_XSS to be zero, so the guest's usage of XSAVES will be exactly the same as XSAVEC.

Change-Id Ie4b0f777d71e428fbee6e82071ac2d7618e9bb40
---
 arch/x86/kvm/vmx/vmx.c | 18 ------------------
 arch/x86/kvm/x86.c     | 20 ++++++++++++++++++++
 2 files changed, 20 insertions(+), 18 deletions(-)
arch/x86/kvm/vmx/vmx.c +0 -18
index f7d292ac9921..b29511d63971 100644
@@ -1818,12 +1818,6 @@ static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
 			return 1;
 		return vmx_get_vmx_msr(&vmx->nested.msrs, msr_info->index,
 				       &msr_info->data);
-	case MSR_IA32_XSS:
-		if (!msr_info->host_initiated &&
-		    !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
-			return 1;
-		msr_info->data = vcpu->arch.ia32_xss;
-		break;
 	case MSR_IA32_RTIT_CTL:
 		if (pt_mode != PT_MODE_HOST_GUEST)
 			return 1;
@@ -2059,18 +2053,6 @@ static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
 		if (!nested_vmx_allowed(vcpu))
 			return 1;
 		return vmx_set_vmx_msr(vcpu, msr_index, data);
-	case MSR_IA32_XSS:
-		if (!msr_info->host_initiated &&
-		    !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
-			return 1;
-		/*
-		 * The only supported bit as of Skylake is bit 8, but
-		 * it is not supported on KVM.
-		 */
-		if (data != 0)
-			return 1;
-		vcpu->arch.ia32_xss = data;
-		break;
 	case MSR_IA32_RTIT_CTL:
 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
 			vmx_rtit_ctl_check(vcpu, data) ||
arch/x86/kvm/x86.c +20 -0
index 259a30e4d3a9..cbbf792a04c1 100644
@@ -2697,6 +2697,20 @@ int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
 	case MSR_IA32_TSC:
 		kvm_write_tsc(vcpu, msr_info);
 		break;
+	case MSR_IA32_XSS:
+		if (!msr_info->host_initiated &&
+		    !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
+			return 1;
+		/*
+		 * We do support PT if kvm_x86_ops->pt_supported(), but we do
+		 * not support IA32_XSS[bit 8]. Guests will have to use
+		 * RDMSR/WRMSR rather than XSAVES/XRSTORS to save/restore PT
+		 * MSRs.
+		 */
+		if (data != 0)
+			return 1;
+		vcpu->arch.ia32_xss = data;
+		break;
 	case MSR_SMI_COUNT:
 		if (!msr_info->host_initiated)
 			return 1;
@@ -3027,6 +3041,12 @@ int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
 		return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
 				   msr_info->host_initiated);
+	case MSR_IA32_XSS:
+		if (!msr_info->host_initiated &&
+		    !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
+			return 1;
+		msr_info->data = vcpu->arch.ia32_xss;
+		break;
 	case MSR_K7_CLK_CTL:
 		/*
 		 * Provide expected ramp-up count for K7. All other
Signature · 1 line
2.23.0.866.gb869b98d4c-goog
#9 lore raw

PATCH v3 8/9 kvm: svm: Update svm_xsaves_supported

In reply to Aaron Lewis

AMD CPUs now support XSAVES in a limited fashion (they require IA32_XSS to be zero).

AMD has no equivalent of Intel's "Enable XSAVES/XRSTORS" VM-execution control. Instead, XSAVES is always available to the guest when supported on the host.

Change-Id I40dc2c682eb0d38c2208d95d5eb7bbb6c47f6317
---
 arch/x86/kvm/svm.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
arch/x86/kvm/svm.c +1 -1
index 36d1cfd45c60..dd8a6418d56c 100644
@@ -5965,7 +5965,7 @@ static bool svm_mpx_supported(void)
 
 static bool svm_xsaves_supported(void)
 {
-	return false;
+	return boot_cpu_has(X86_FEATURE_XSAVES);
 }
 
 static bool svm_umip_emulated(void)
Signature · 1 line
2.23.0.866.gb869b98d4c-goog
#10 lore raw

PATCH v3 9/9 kvm: tests: Add test to verify MSR_IA32_XSS

In reply to Aaron Lewis

Ensure that IA32_XSS appears in KVM_GET_MSR_INDEX_LIST if it can be set to a non-zero value.

Change-Id Ia2d644f69e2d6d8c27d7e0a7a45c2bf9c42bf5ff
---
 tools/testing/selftests/kvm/.gitignore        |  1 +
 tools/testing/selftests/kvm/Makefile          |  1 +
 .../selftests/kvm/include/x86_64/processor.h  |  7 +-
 .../selftests/kvm/lib/x86_64/processor.c      | 72 +++++++++++++++---
 .../selftests/kvm/x86_64/xss_msr_test.c       | 76 +++++++++++++++++++
 5 files changed, 147 insertions(+), 10 deletions(-)
 create mode 100644 tools/testing/selftests/kvm/x86_64/xss_msr_test.c
tools/testing/selftests/kvm/.gitignore +1 -0
index b35da375530a..6e9ec34f8124 100644
@@ -11,6 +11,7 @@
 /x86_64/vmx_close_while_nested_test
 /x86_64/vmx_set_nested_state_test
 /x86_64/vmx_tsc_adjust_test
+/x86_64/xss_msr_test
 /clear_dirty_log_test
 /dirty_log_test
 /kvm_create_max_vcpus
tools/testing/selftests/kvm/Makefile +1 -0
index c5ec868fa1e5..3138a916574a 100644
@@ -25,6 +25,7 @@ TEST_GEN_PROGS_x86_64 += x86_64/vmx_close_while_nested_test
 TEST_GEN_PROGS_x86_64 += x86_64/vmx_dirty_log_test
 TEST_GEN_PROGS_x86_64 += x86_64/vmx_set_nested_state_test
 TEST_GEN_PROGS_x86_64 += x86_64/vmx_tsc_adjust_test
+TEST_GEN_PROGS_x86_64 += x86_64/xss_msr_test
 TEST_GEN_PROGS_x86_64 += clear_dirty_log_test
 TEST_GEN_PROGS_x86_64 += dirty_log_test
 TEST_GEN_PROGS_x86_64 += kvm_create_max_vcpus
tools/testing/selftests/kvm/include/x86_64/processor.h +6 -1
index ff234018219c..635ee6c33ad2 100644
@@ -308,6 +308,8 @@ struct kvm_x86_state *vcpu_save_state(struct kvm_vm *vm, uint32_t vcpuid);
 void vcpu_load_state(struct kvm_vm *vm, uint32_t vcpuid,
 		     struct kvm_x86_state *state);
 
+struct kvm_msr_list *kvm_get_msr_index_list(void);
+
 struct kvm_cpuid2 *kvm_get_supported_cpuid(void);
 void vcpu_set_cpuid(struct kvm_vm *vm, uint32_t vcpuid,
 		    struct kvm_cpuid2 *cpuid);
@@ -322,10 +324,13 @@ kvm_get_supported_cpuid_entry(uint32_t function)
 }
 
 uint64_t vcpu_get_msr(struct kvm_vm *vm, uint32_t vcpuid, uint64_t msr_index);
+int _vcpu_set_msr(struct kvm_vm *vm, uint32_t vcpuid, uint64_t msr_index,
+		  uint64_t msr_value);
 void vcpu_set_msr(struct kvm_vm *vm, uint32_t vcpuid, uint64_t msr_index,
 	  	  uint64_t msr_value);
 
-uint32_t kvm_get_cpuid_max(void);
+uint32_t kvm_get_cpuid_max_basic(void);
+uint32_t kvm_get_cpuid_max_extended(void);
 void kvm_get_cpu_address_width(unsigned int *pa_bits, unsigned int *va_bits);
 
 /*
tools/testing/selftests/kvm/lib/x86_64/processor.c +63 -9

121 lines not on this page · raw message

tools/testing/selftests/kvm/x86_64/xss_msr_test.c new file +76 -0

76 lines not on this page · raw message

Signature · 1 line
2.23.0.866.gb869b98d4c-goog
maintainer Applied
#11 lore raw
In reply to Aaron Lewis
Quoted 45 lines · Aaron Lewis

Unify AMD's and Intel's approach for supporting XSAVES. To do this change Intel's approach from using the MSR-load areas to writing the guest/host values to IA32_XSS on a VM-enter/VM-exit. Switching to this strategy allows for a common approach between both AMD and Intel. Additionally, define svm_xsaves_supported() based on AMD's feedback, and add vcpu->arch.xsaves_enabled to track whether XSAVES is enabled in the guest.

This change sets up IA32_XSS to be a non-zero value in the future, which may happen sooner than later with support for guest CET feature being added.

v2 -> v3:
- Remove guest_xcr0_loaded from kvm_vcpu.
- Add vcpu->arch.xsaves_enabled.
- Add staged rollout to load the hardware IA32_XSS MSR with guest/host values on VM-entry and VM-exit:
1) Introduce vcpu->arch->xsaves_enabled.
2) Add svm implementation for switching between guest and host IA32_XSS.
3) Add vmx implementation for switching between guest and host IA32_XSS.
4) Remove svm and vmx implementation and add it to common code.

v1 -> v2:
- Add the flag xsaves_enabled to kvm_vcpu_arch to track when XSAVES is enabled in the guest, whether or not XSAVES is enumerated in the guest CPUID.
- Remove code that sets the X86_FEATURE_XSAVES bit in the guest CPUID which was added in patch "Enumerate XSAVES in guest CPUID when it is available to the guest". As a result we no longer need that patch.
- Added a comment to kvm_set_msr_common to describe how to save/restore PT MSRS without using XSAVES/XRSTORS.
- Added more comments to the "Add support for XSAVES on AMD" patch.
- Replaced vcpu_set_msr_expect_result() with _vcpu_set_msr() in the test library.

Aaron Lewis (9):
KVM: x86: Introduce vcpu->arch.xsaves_enabled
KVM: VMX: Fix conditions for guest IA32_XSS support
KVM: x86: Remove unneeded kvm_vcpu variable, guest_xcr0_loaded
KVM: SVM: Use wrmsr for switching between guest and host IA32_XSS on AMD
KVM: VMX: Use wrmsr for switching between guest and host IA32_XSS on Intel
KVM: x86: Move IA32_XSS-swapping on VM-entry/VM-exit to common x86 code
kvm: x86: Move IA32_XSS to kvm_{get,set}_msr_common
kvm: svm: Update svm_xsaves_supported
kvm: tests: Add test to verify MSR_IA32_XSS

Quoted 14 lines
 arch/x86/include/asm/kvm_host.h               |  1 +
 arch/x86/kvm/svm.c                            |  9 ++-
 arch/x86/kvm/vmx/vmx.c                        | 41 ++--------
 arch/x86/kvm/x86.c                            | 52 ++++++++++---
 arch/x86/kvm/x86.h                            |  4 +-
 include/linux/kvm_host.h                      |  1 -
 tools/testing/selftests/kvm/.gitignore        |  1 +
 tools/testing/selftests/kvm/Makefile          |  1 +
 .../selftests/kvm/include/x86_64/processor.h  |  7 +-
 .../selftests/kvm/lib/x86_64/processor.c      | 72 +++++++++++++++---
 .../selftests/kvm/x86_64/xss_msr_test.c       | 76 +++++++++++++++++++
 11 files changed, 205 insertions(+), 60 deletions(-)
 create mode 100644 tools/testing/selftests/kvm/x86_64/xss_msr_test.c

Queued, thanks.

Paolo