[RFC PATCH] kvm/x86: Keep root hpa in prev_roots as much as possible

[Lai Jiangshan <jiangshanlai@gmail.com>]

From: Lai Jiangshan <laijs@linux.alibaba.com>

Pagetable roots in prev_roots[] are likely to be reused soon and
there is no much overhead to keep it with a new need_sync field
introduced.

With the help of the new need_sync field, pagetable roots are
kept as much as possible, and they will be re-synced before reused
instead of being dropped.

Signed-off-by: Lai Jiangshan <laijs@linux.alibaba.com>
---

This patch is just for RFC.
  Is the idea Ok?
  If the idea is Ok, we need to reused one bit from pgd or hpa
    as need_sync to save memory.  Which one is better?

 arch/x86/include/asm/kvm_host.h |  3 ++-
 arch/x86/kvm/mmu/mmu.c          |  6 ++++++
 arch/x86/kvm/vmx/nested.c       | 12 ++++--------
 arch/x86/kvm/x86.c              |  9 +++++----
 4 files changed, 17 insertions(+), 13 deletions(-)

diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
index 55efbacfc244..19a337cf7aa6 100644
--- a/arch/x86/include/asm/kvm_host.h
+++ b/arch/x86/include/asm/kvm_host.h
@@ -354,10 +354,11 @@ struct rsvd_bits_validate {
 struct kvm_mmu_root_info {
 	gpa_t pgd;
 	hpa_t hpa;
+	bool need_sync;
 };
 
 #define KVM_MMU_ROOT_INFO_INVALID \
-	((struct kvm_mmu_root_info) { .pgd = INVALID_PAGE, .hpa = INVALID_PAGE })
+	((struct kvm_mmu_root_info) { .pgd = INVALID_PAGE, .hpa = INVALID_PAGE, .need_sync = true})
 
 #define KVM_MMU_NUM_PREV_ROOTS 3
 
diff --git a/arch/x86/kvm/mmu/mmu.c b/arch/x86/kvm/mmu/mmu.c
index 5e60b00e8e50..147827135549 100644
--- a/arch/x86/kvm/mmu/mmu.c
+++ b/arch/x86/kvm/mmu/mmu.c
@@ -3878,6 +3878,7 @@ static bool cached_root_available(struct kvm_vcpu *vcpu, gpa_t new_pgd,
 
 	root.pgd = mmu->root_pgd;
 	root.hpa = mmu->root_hpa;
+	root.need_sync = false;
 
 	if (is_root_usable(&root, new_pgd, new_role))
 		return true;
@@ -3892,6 +3893,11 @@ static bool cached_root_available(struct kvm_vcpu *vcpu, gpa_t new_pgd,
 	mmu->root_hpa = root.hpa;
 	mmu->root_pgd = root.pgd;
 
+	if (i < KVM_MMU_NUM_PREV_ROOTS && root.need_sync) {
+		kvm_make_request(KVM_REQ_MMU_SYNC, vcpu);
+		kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
+	}
+
 	return i < KVM_MMU_NUM_PREV_ROOTS;
 }
 
diff --git a/arch/x86/kvm/vmx/nested.c b/arch/x86/kvm/vmx/nested.c
index 6058a65a6ede..ab7069ac6dc5 100644
--- a/arch/x86/kvm/vmx/nested.c
+++ b/arch/x86/kvm/vmx/nested.c
@@ -5312,7 +5312,7 @@ static int handle_invept(struct kvm_vcpu *vcpu)
 {
 	struct vcpu_vmx *vmx = to_vmx(vcpu);
 	u32 vmx_instruction_info, types;
-	unsigned long type, roots_to_free;
+	unsigned long type;
 	struct kvm_mmu *mmu;
 	gva_t gva;
 	struct x86_exception e;
@@ -5361,29 +5361,25 @@ static int handle_invept(struct kvm_vcpu *vcpu)
 			return nested_vmx_fail(vcpu,
 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
 
-		roots_to_free = 0;
 		if (nested_ept_root_matches(mmu->root_hpa, mmu->root_pgd,
 					    operand.eptp))
-			roots_to_free |= KVM_MMU_ROOT_CURRENT;
+			kvm_mmu_free_roots(vcpu, mmu, KVM_MMU_ROOT_CURRENT);
 
 		for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++) {
 			if (nested_ept_root_matches(mmu->prev_roots[i].hpa,
 						    mmu->prev_roots[i].pgd,
 						    operand.eptp))
-				roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
+				mmu->prev_roots[i].need_sync = true;
 		}
 		break;
 	case VMX_EPT_EXTENT_GLOBAL:
-		roots_to_free = KVM_MMU_ROOTS_ALL;
+		kvm_mmu_free_roots(vcpu, mmu, KVM_MMU_ROOTS_ALL);
 		break;
 	default:
 		BUG();
 		break;
 	}
 
-	if (roots_to_free)
-		kvm_mmu_free_roots(vcpu, mmu, roots_to_free);
-
 	return nested_vmx_succeed(vcpu);
 }
 
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index bbc4e04e67ad..1f5617ec6b34 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -11680,7 +11680,6 @@ int kvm_handle_invpcid(struct kvm_vcpu *vcpu, unsigned long type, gva_t gva)
 	bool pcid_enabled;
 	struct x86_exception e;
 	unsigned i;
-	unsigned long roots_to_free = 0;
 	struct {
 		u64 pcid;
 		u64 gla;
@@ -11722,9 +11721,8 @@ int kvm_handle_invpcid(struct kvm_vcpu *vcpu, unsigned long type, gva_t gva)
 		for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
 			if (kvm_get_pcid(vcpu, vcpu->arch.mmu->prev_roots[i].pgd)
 			    == operand.pcid)
-				roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
+				vcpu->arch.mmu->prev_roots[i].need_sync = true;
 
-		kvm_mmu_free_roots(vcpu, vcpu->arch.mmu, roots_to_free);
 		/*
 		 * If neither the current cr3 nor any of the prev_roots use the
 		 * given PCID, then nothing needs to be done here because a
@@ -11743,7 +11741,10 @@ int kvm_handle_invpcid(struct kvm_vcpu *vcpu, unsigned long type, gva_t gva)
 
 		fallthrough;
 	case INVPCID_TYPE_ALL_INCL_GLOBAL:
-		kvm_make_request(KVM_REQ_MMU_RELOAD, vcpu);
+		kvm_mmu_sync_roots(vcpu);
+		kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
+		for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
+			vcpu->arch.mmu->prev_roots[i].need_sync = true;
 		return kvm_skip_emulated_instruction(vcpu);
 
 	default:
-- 
2.19.1.6.gb485710b