KVM: x86: nSVM: implement nested TSC scaling

This was tested by booting a nested guest with TSC=1Ghz,
observing the clocks, and doing about 100 cycles of migration.

Note that qemu patch is needed to support migration because
of a new MSR that needs to be placed in the migration state.

The patch will be sent to the qemu mailing list soon.

Signed-off-by: Maxim Levitsky <mlevitsk@redhat.com>
Message-Id: <20210914154825.104886-14-mlevitsk@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

5 files changed, 63 insertions, 6 deletions

diff --git a/arch/x86/kvm/svm/nested.c b/arch/x86/kvm/svm/nested.c
index ec3e926b20dd..31fd4bd334c2 100644
--- a/arch/x86/kvm/svm/nested.c
+++ b/arch/x86/kvm/svm/nested.c
@@ -538,8 +538,17 @@ static void nested_vmcb02_prepare_control(struct vcpu_svm *svm)
 	if (nested_npt_enabled(svm))
 		nested_svm_init_mmu_context(vcpu);
 
-	svm->vmcb->control.tsc_offset = vcpu->arch.tsc_offset =
-		vcpu->arch.l1_tsc_offset + svm->nested.ctl.tsc_offset;
+	vcpu->arch.tsc_offset = kvm_calc_nested_tsc_offset(
+			vcpu->arch.l1_tsc_offset,
+			svm->nested.ctl.tsc_offset,
+			svm->tsc_ratio_msr);
+
+	svm->vmcb->control.tsc_offset = vcpu->arch.tsc_offset;
+
+	if (svm->tsc_ratio_msr != kvm_default_tsc_scaling_ratio) {
+		WARN_ON(!svm->tsc_scaling_enabled);
+		nested_svm_update_tsc_ratio_msr(vcpu);
+	}
 
 	svm->vmcb->control.int_ctl             =
 		(svm->nested.ctl.int_ctl & int_ctl_vmcb12_bits) |
@@ -824,6 +833,12 @@ int nested_svm_vmexit(struct vcpu_svm *svm)
 		vmcb_mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
 	}
 
+	if (svm->tsc_ratio_msr != kvm_default_tsc_scaling_ratio) {
+		WARN_ON(!svm->tsc_scaling_enabled);
+		vcpu->arch.tsc_scaling_ratio = vcpu->arch.l1_tsc_scaling_ratio;
+		svm_write_tsc_multiplier(vcpu, vcpu->arch.tsc_scaling_ratio);
+	}
+
 	svm->nested.ctl.nested_cr3 = 0;
 
 	/*
@@ -1211,6 +1226,16 @@ int nested_svm_exit_special(struct vcpu_svm *svm)
 	return NESTED_EXIT_CONTINUE;
 }
 
+void nested_svm_update_tsc_ratio_msr(struct kvm_vcpu *vcpu)
+{
+	struct vcpu_svm *svm = to_svm(vcpu);
+
+	vcpu->arch.tsc_scaling_ratio =
+		kvm_calc_nested_tsc_multiplier(vcpu->arch.l1_tsc_scaling_ratio,
+					       svm->tsc_ratio_msr);
+	svm_write_tsc_multiplier(vcpu, vcpu->arch.tsc_scaling_ratio);
+}
+
 static int svm_get_nested_state(struct kvm_vcpu *vcpu,
 				struct kvm_nested_state __user *user_kvm_nested_state,
 				u32 user_data_size)
diff --git a/arch/x86/kvm/svm/svm.c b/arch/x86/kvm/svm/svm.c
index 1ff904653a3e..89077160d463 100644
--- a/arch/x86/kvm/svm/svm.c
+++ b/arch/x86/kvm/svm/svm.c
@@ -940,6 +940,9 @@ static __init void svm_set_cpu_caps(void)
 		if (npt_enabled)
 			kvm_cpu_cap_set(X86_FEATURE_NPT);
 
+		if (tsc_scaling)
+			kvm_cpu_cap_set(X86_FEATURE_TSCRATEMSR);
+
 		/* Nested VM can receive #VMEXIT instead of triggering #GP */
 		kvm_cpu_cap_set(X86_FEATURE_SVME_ADDR_CHK);
 	}
@@ -1132,7 +1135,9 @@ static u64 svm_get_l2_tsc_offset(struct kvm_vcpu *vcpu)
 
 static u64 svm_get_l2_tsc_multiplier(struct kvm_vcpu *vcpu)
 {
-	return kvm_default_tsc_scaling_ratio;
+	struct vcpu_svm *svm = to_svm(vcpu);
+
+	return svm->tsc_ratio_msr;
 }
 
 static void svm_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
@@ -1144,7 +1149,7 @@ static void svm_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
 	vmcb_mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
 }
 
-static void svm_write_tsc_multiplier(struct kvm_vcpu *vcpu, u64 multiplier)
+void svm_write_tsc_multiplier(struct kvm_vcpu *vcpu, u64 multiplier)
 {
 	wrmsrl(MSR_AMD64_TSC_RATIO, multiplier);
 }
@@ -1356,7 +1361,6 @@ static void init_vmcb(struct kvm_vcpu *vcpu)
 	vmcb_mark_all_dirty(svm->vmcb);
 
 	enable_gif(svm);
-
 }
 
 static void __svm_vcpu_reset(struct kvm_vcpu *vcpu)
@@ -1367,6 +1371,7 @@ static void __svm_vcpu_reset(struct kvm_vcpu *vcpu)
 
 	svm_init_osvw(vcpu);
 	vcpu->arch.microcode_version = 0x01000065;
+	svm->tsc_ratio_msr = kvm_default_tsc_scaling_ratio;
 
 	if (sev_es_guest(vcpu->kvm))
 		sev_es_vcpu_reset(svm);
@@ -2718,6 +2723,11 @@ static int svm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
 	struct vcpu_svm *svm = to_svm(vcpu);
 
 	switch (msr_info->index) {
+	case MSR_AMD64_TSC_RATIO:
+		if (!msr_info->host_initiated && !svm->tsc_scaling_enabled)
+			return 1;
+		msr_info->data = svm->tsc_ratio_msr;
+		break;
 	case MSR_STAR:
 		msr_info->data = svm->vmcb01.ptr->save.star;
 		break;
@@ -2867,6 +2877,19 @@ static int svm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
 	u32 ecx = msr->index;
 	u64 data = msr->data;
 	switch (ecx) {
+	case MSR_AMD64_TSC_RATIO:
+		if (!msr->host_initiated && !svm->tsc_scaling_enabled)
+			return 1;
+
+		if (data & TSC_RATIO_RSVD)
+			return 1;
+
+		svm->tsc_ratio_msr = data;
+
+		if (svm->tsc_scaling_enabled && is_guest_mode(vcpu))
+			nested_svm_update_tsc_ratio_msr(vcpu);
+
+		break;
 	case MSR_IA32_CR_PAT:
 		if (!kvm_mtrr_valid(vcpu, MSR_IA32_CR_PAT, data))
 			return 1;
@@ -4062,6 +4085,8 @@ static void svm_vcpu_after_set_cpuid(struct kvm_vcpu *vcpu)
 	svm->nrips_enabled = kvm_cpu_cap_has(X86_FEATURE_NRIPS) &&
 			     guest_cpuid_has(vcpu, X86_FEATURE_NRIPS);
 
+	svm->tsc_scaling_enabled = tsc_scaling && guest_cpuid_has(vcpu, X86_FEATURE_TSCRATEMSR);
+
 	svm_recalc_instruction_intercepts(vcpu, svm);
 
 	/* For sev guests, the memory encryption bit is not reserved in CR3.  */
diff --git a/arch/x86/kvm/svm/svm.h b/arch/x86/kvm/svm/svm.h
index c19859e62dec..0d7bbe548ac3 100644
--- a/arch/x86/kvm/svm/svm.h
+++ b/arch/x86/kvm/svm/svm.h
@@ -140,6 +140,8 @@ struct vcpu_svm {
 	u64 next_rip;
 
 	u64 spec_ctrl;
+
+	u64 tsc_ratio_msr;
 	/*
 	 * Contains guest-controlled bits of VIRT_SPEC_CTRL, which will be
 	 * translated into the appropriate L2_CFG bits on the host to
@@ -160,7 +162,8 @@ struct vcpu_svm {
 	unsigned long int3_rip;
 
 	/* cached guest cpuid flags for faster access */
-	bool nrips_enabled	: 1;
+	bool nrips_enabled                : 1;
+	bool tsc_scaling_enabled          : 1;
 
 	u32 ldr_reg;
 	u32 dfr_reg;
@@ -483,6 +486,8 @@ int nested_svm_check_permissions(struct kvm_vcpu *vcpu);
 int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
 			       bool has_error_code, u32 error_code);
 int nested_svm_exit_special(struct vcpu_svm *svm);
+void nested_svm_update_tsc_ratio_msr(struct kvm_vcpu *vcpu);
+void svm_write_tsc_multiplier(struct kvm_vcpu *vcpu, u64 multiplier);
 void nested_load_control_from_vmcb12(struct vcpu_svm *svm,
 				     struct vmcb_control_area *control);
 void nested_sync_control_from_vmcb02(struct vcpu_svm *svm);
diff --git a/arch/x86/kvm/vmx/vmx.c b/arch/x86/kvm/vmx/vmx.c
index b8bfb46d58f4..1c8b2b6e7ed9 100644
--- a/arch/x86/kvm/vmx/vmx.c
+++ b/arch/x86/kvm/vmx/vmx.c
@@ -6435,6 +6435,7 @@ static bool vmx_has_emulated_msr(struct kvm *kvm, u32 index)
 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
 		return nested;
 	case MSR_AMD64_VIRT_SPEC_CTRL:
+	case MSR_AMD64_TSC_RATIO:
 		/* This is AMD only.  */
 		return false;
 	default:
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index ee1b9e168d46..c028ff7d6fb9 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -1361,6 +1361,7 @@ static const u32 emulated_msrs_all[] = {
 	MSR_PLATFORM_INFO,
 	MSR_MISC_FEATURES_ENABLES,
 	MSR_AMD64_VIRT_SPEC_CTRL,
+	MSR_AMD64_TSC_RATIO,
 	MSR_IA32_POWER_CTL,
 	MSR_IA32_UCODE_REV,