When EPT is used for nested guest we need to re-init MMU as shadow
EPT MMU (nested_ept_init_mmu_context() does that). When we return back
from L2 to L1 kvm_mmu_reset_context() in nested_vmx_load_cr3() resets
MMU back to normal TDP mode. Add a special 'guest_mmu' so we can use
separate root caches; the improved hit rate is not very important for
single vCPU performance, but it avoids contention on the mmu_lock for
many vCPUs.
On the nested CPUID benchmark, with 16 vCPUs, an L2->L1->L2 vmexit
goes from 42k to 26k cycles.
Signed-off-by: Vitaly Kuznetsov <vkuznets@xxxxxxxxxx>
Signed-off-by: Paolo Bonzini <pbonzini@xxxxxxxxxx>
---
Changes since v1:
- drop now unneded local vmx variable in vmx_free_vcpu_nested
[Sean Christopherson]
---
arch/x86/include/asm/kvm_host.h | 3 +++
arch/x86/kvm/mmu.c | 15 +++++++++++----
arch/x86/kvm/vmx.c | 27 ++++++++++++++++++---------
3 files changed, 32 insertions(+), 13 deletions(-)
diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
index 404c3438827b..a3829869353b 100644
--- a/arch/x86/include/asm/kvm_host.h
+++ b/arch/x86/include/asm/kvm_host.h
@@ -539,6 +539,9 @@ struct kvm_vcpu_arch {
/* Non-nested MMU for L1 */
struct kvm_mmu root_mmu;
+ /* L1 MMU when running nested */
+ struct kvm_mmu guest_mmu;
+
/*
* Paging state of an L2 guest (used for nested npt)
*
diff --git a/arch/x86/kvm/mmu.c b/arch/x86/kvm/mmu.c
index 4491b8894337..96c2a0b3eb53 100644
--- a/arch/x86/kvm/mmu.c
+++ b/arch/x86/kvm/mmu.c
@@ -4967,8 +4967,10 @@ EXPORT_SYMBOL_GPL(kvm_mmu_load);
void kvm_mmu_unload(struct kvm_vcpu *vcpu)
{
- kvm_mmu_free_roots(vcpu, vcpu->arch.mmu, KVM_MMU_ROOTS_ALL);
- WARN_ON(VALID_PAGE(vcpu->arch.mmu->root_hpa));
+ kvm_mmu_free_roots(vcpu, &vcpu->arch.root_mmu, KVM_MMU_ROOTS_ALL);
+ WARN_ON(VALID_PAGE(vcpu->arch.root_mmu.root_hpa));
+ kvm_mmu_free_roots(vcpu, &vcpu->arch.guest_mmu, KVM_MMU_ROOTS_ALL);
+ WARN_ON(VALID_PAGE(vcpu->arch.guest_mmu.root_hpa));
}
EXPORT_SYMBOL_GPL(kvm_mmu_unload);
@@ -5407,13 +5409,18 @@ int kvm_mmu_create(struct kvm_vcpu *vcpu)
vcpu->arch.mmu = &vcpu->arch.root_mmu;
vcpu->arch.walk_mmu = &vcpu->arch.root_mmu;
+
vcpu->arch.root_mmu.root_hpa = INVALID_PAGE;
vcpu->arch.root_mmu.translate_gpa = translate_gpa;
- vcpu->arch.nested_mmu.translate_gpa = translate_nested_gpa;
-
for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
vcpu->arch.root_mmu.prev_roots[i] = KVM_MMU_ROOT_INFO_INVALID;
+ vcpu->arch.guest_mmu.root_hpa = INVALID_PAGE;
+ vcpu->arch.guest_mmu.translate_gpa = translate_gpa;
+ for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
+ vcpu->arch.guest_mmu.prev_roots[i] = KVM_MMU_ROOT_INFO_INVALID;
+
+ vcpu->arch.nested_mmu.translate_gpa = translate_nested_gpa;
return alloc_mmu_pages(vcpu);
}
diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
index 2d55adab52de..93ff08136fc1 100644
--- a/arch/x86/kvm/vmx.c
+++ b/arch/x86/kvm/vmx.c
@@ -8468,8 +8468,10 @@ static inline void nested_release_vmcs12(struct vcpu_vmx *vmx)
* Free whatever needs to be freed from vmx->nested when L1 goes down, or
* just stops using VMX.
*/
-static void free_nested(struct vcpu_vmx *vmx)
+static void free_nested(struct kvm_vcpu *vcpu)
{
+ struct vcpu_vmx *vmx = to_vmx(vcpu);
+
if (!vmx->nested.vmxon && !vmx->nested.smm.vmxon)
return;
@@ -8502,6 +8504,8 @@ static void free_nested(struct vcpu_vmx *vmx)
vmx->nested.pi_desc = NULL;
}
+ kvm_mmu_free_roots(vcpu, &vcpu->arch.guest_mmu, KVM_MMU_ROOTS_ALL);
+
free_loaded_vmcs(&vmx->nested.vmcs02);
}
@@ -8510,7 +8514,7 @@ static int handle_vmoff(struct kvm_vcpu *vcpu)
{
if (!nested_vmx_check_permission(vcpu))
return 1;
- free_nested(to_vmx(vcpu));
+ free_nested(vcpu);
nested_vmx_succeed(vcpu);
return kvm_skip_emulated_instruction(vcpu);
}
@@ -8541,6 +8545,8 @@ static int handle_vmclear(struct kvm_vcpu *vcpu)
if (vmptr == vmx->nested.current_vmptr)
nested_release_vmcs12(vmx);
+ kvm_mmu_free_roots(vcpu, &vcpu->arch.guest_mmu, KVM_MMU_ROOTS_ALL);
+
kvm_vcpu_write_guest(vcpu,
vmptr + offsetof(struct vmcs12, launch_state),
&zero, sizeof(zero));
@@ -8924,6 +8930,9 @@ static int handle_vmptrld(struct kvm_vcpu *vcpu)
}
nested_release_vmcs12(vmx);
+
+ kvm_mmu_free_roots(vcpu, &vcpu->arch.guest_mmu,
+ KVM_MMU_ROOTS_ALL);
/*
* Load VMCS12 from guest memory since it is not already
* cached.
@@ -10976,12 +10985,10 @@ static void vmx_switch_vmcs(struct kvm_vcpu *vcpu, struct loaded_vmcs *vmcs)
*/
static void vmx_free_vcpu_nested(struct kvm_vcpu *vcpu)
{
- struct vcpu_vmx *vmx = to_vmx(vcpu);
-
- vcpu_load(vcpu);
- vmx_switch_vmcs(vcpu, &vmx->vmcs01);
- free_nested(vmx);
- vcpu_put(vcpu);
+ vcpu_load(vcpu);
+ vmx_switch_vmcs(vcpu, &to_vmx(vcpu)->vmcs01);
+ free_nested(vcpu);
+ vcpu_put(vcpu);
}
static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
@@ -11331,6 +11338,7 @@ static int nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
if (!valid_ept_address(vcpu, nested_ept_get_cr3(vcpu)))
return 1;
+ vcpu->arch.mmu = &vcpu->arch.guest_mmu;
kvm_init_shadow_ept_mmu(vcpu,
to_vmx(vcpu)->nested.msrs.ept_caps &
VMX_EPT_EXECUTE_ONLY_BIT,
@@ -11346,6 +11354,7 @@ static int nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
static void nested_ept_uninit_mmu_context(struct kvm_vcpu *vcpu)
{
+ vcpu->arch.mmu = &vcpu->arch.root_mmu;
vcpu->arch.walk_mmu = &vcpu->arch.root_mmu;
}
@@ -13421,7 +13430,7 @@ static void vmx_leave_nested(struct kvm_vcpu *vcpu)
to_vmx(vcpu)->nested.nested_run_pending = 0;
nested_vmx_vmexit(vcpu, -1, 0, 0);
}
- free_nested(to_vmx(vcpu));
+ free_nested(vcpu);
}
/*
--
2.17.1
