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 <[email protected]> Signed-off-by: Paolo Bonzini <[email protected]> --- arch/x86/include/asm/kvm_host.h | 3 +++ arch/x86/kvm/mmu.c | 15 +++++++++++---- arch/x86/kvm/vmx.c | 27 +++++++++++++++++++-------- 3 files changed, 33 insertions(+), 12 deletions(-) diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h index fc6038b4ed6b..3f53235ee2a8 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 113feceec686..96593cd285db 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 0e56b2237a42..29af40b9239f 100644 --- a/arch/x86/kvm/vmx.c +++ b/arch/x86/kvm/vmx.c @@ -8438,8 +8438,10 @@ 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; @@ -8472,6 +8474,8 @@ 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); } @@ -8480,7 +8484,7 @@ 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); } @@ -8511,6 +8515,8 @@ 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)); @@ -8894,6 +8900,9 @@ 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. @@ -10926,12 +10935,12 @@ 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); + 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, &vmx->vmcs01); + free_nested(vcpu); + vcpu_put(vcpu); } static void vmx_free_vcpu(struct kvm_vcpu *vcpu) @@ -11281,6 +11290,7 @@ 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, @@ -11296,6 +11306,7 @@ 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; } @@ -13363,7 +13374,7 @@ 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); } /* -- 2.14.4
