[PATCH 3/3] KVM: SVM: Lazy fpu for npt

[Avi Kivity <avi@redhat.com>]

If two conditions apply:
 - no bits outside TS and EM differ between the host and guest cr0
 - the fpu is active

then we can activate the selective cr0 write intercept and drop the
unconditional cr0 read and write intercept, and allow the guest to run
with the host fpu state.  This reduces the heavyweight context switch
when npt is enabled.

Signed-off-by: Avi Kivity <avi@redhat.com>
---
 arch/x86/kvm/svm.c |   70 +++++++++++++++++++++++++--------------------------
 1 files changed, 34 insertions(+), 36 deletions(-)

diff --git a/arch/x86/kvm/svm.c b/arch/x86/kvm/svm.c
index f4418e2..7f3d890 100644
--- a/arch/x86/kvm/svm.c
+++ b/arch/x86/kvm/svm.c
@@ -571,6 +571,7 @@ static void init_vmcb(struct vcpu_svm *svm)
 	control->intercept = 	(1ULL << INTERCEPT_INTR) |
 				(1ULL << INTERCEPT_NMI) |
 				(1ULL << INTERCEPT_SMI) |
+				(1ULL << INTERCEPT_SELECTIVE_CR0) |
 				(1ULL << INTERCEPT_CPUID) |
 				(1ULL << INTERCEPT_INVD) |
 				(1ULL << INTERCEPT_HLT) |
@@ -643,10 +644,8 @@ static void init_vmcb(struct vcpu_svm *svm)
 		control->intercept &= ~((1ULL << INTERCEPT_TASK_SWITCH) |
 					(1ULL << INTERCEPT_INVLPG));
 		control->intercept_exceptions &= ~(1 << PF_VECTOR);
-		control->intercept_cr_read &= ~(INTERCEPT_CR0_MASK|
-						INTERCEPT_CR3_MASK);
-		control->intercept_cr_write &= ~(INTERCEPT_CR0_MASK|
-						 INTERCEPT_CR3_MASK);
+		control->intercept_cr_read &= ~INTERCEPT_CR3_MASK;
+		control->intercept_cr_write &= ~INTERCEPT_CR3_MASK;
 		save->g_pat = 0x0007040600070406ULL;
 		save->cr3 = 0;
 		save->cr4 = 0;
@@ -965,6 +964,27 @@ static void svm_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
 {
 }
 
+static void update_cr0_intercept(struct vcpu_svm *svm)
+{
+	ulong gcr0 = svm->vcpu.arch.cr0;
+	u64 *hcr0 = &svm->vmcb->save.cr0;
+
+	if (!svm->vcpu.fpu_active)
+		*hcr0 |= SVM_CR0_SELECTIVE_MASK;
+	else
+		*hcr0 = (*hcr0 & ~SVM_CR0_SELECTIVE_MASK)
+			| (gcr0 & SVM_CR0_SELECTIVE_MASK);
+
+
+	if (gcr0 == *hcr0 && svm->vcpu.fpu_active) {
+		svm->vmcb->control.intercept_cr_read &= ~INTERCEPT_CR0_MASK;
+		svm->vmcb->control.intercept_cr_write &= ~INTERCEPT_CR0_MASK;
+	} else {
+		svm->vmcb->control.intercept_cr_read |= INTERCEPT_CR0_MASK;
+		svm->vmcb->control.intercept_cr_write |= INTERCEPT_CR0_MASK;
+	}
+}
+
 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 {
 	struct vcpu_svm *svm = to_svm(vcpu);
@@ -982,12 +1002,11 @@ static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 		}
 	}
 #endif
-	if (npt_enabled)
-		goto set;
-
 	vcpu->arch.cr0 = cr0;
-	cr0 |= X86_CR0_PG | X86_CR0_WP;
-set:
+
+	if (!npt_enabled)
+		cr0 |= X86_CR0_PG | X86_CR0_WP;
+
 	/*
 	 * re-enable caching here because the QEMU bios
 	 * does not do it - this results in some delay at
@@ -995,6 +1014,7 @@ set:
 	 */
 	cr0 &= ~(X86_CR0_CD | X86_CR0_NW);
 	svm->vmcb->save.cr0 = cr0;
+	update_cr0_intercept(svm);
 }
 
 static void svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
@@ -1240,11 +1260,8 @@ static int ud_interception(struct vcpu_svm *svm)
 static int nm_interception(struct vcpu_svm *svm)
 {
 	svm->vmcb->control.intercept_exceptions &= ~(1 << NM_VECTOR);
-	if (!kvm_read_cr0_bits(&svm->vcpu, X86_CR0_TS))
-		svm->vmcb->save.cr0 &= ~X86_CR0_TS;
-	else
-		svm->vmcb->save.cr0 |= X86_CR0_TS;
 	svm->vcpu.fpu_active = 1;
+	update_cr0_intercept(svm);
 
 	return 1;
 }
@@ -2297,7 +2314,7 @@ static int (*svm_exit_handlers[])(struct vcpu_svm *svm) = {
 	[SVM_EXIT_READ_CR3]           		= emulate_on_interception,
 	[SVM_EXIT_READ_CR4]           		= emulate_on_interception,
 	[SVM_EXIT_READ_CR8]           		= emulate_on_interception,
-	/* for now: */
+	[SVM_EXIT_CR0_SEL_WRITE]		= emulate_on_interception,
 	[SVM_EXIT_WRITE_CR0]          		= emulate_on_interception,
 	[SVM_EXIT_WRITE_CR3]          		= emulate_on_interception,
 	[SVM_EXIT_WRITE_CR4]          		= emulate_on_interception,
@@ -2383,21 +2400,10 @@ static int handle_exit(struct kvm_vcpu *vcpu)
 
 	svm_complete_interrupts(svm);
 
-	if (npt_enabled) {
-		int mmu_reload = 0;
-		if ((kvm_read_cr0_bits(vcpu, X86_CR0_PG) ^ svm->vmcb->save.cr0)
-		    & X86_CR0_PG) {
-			svm_set_cr0(vcpu, svm->vmcb->save.cr0);
-			mmu_reload = 1;
-		}
+	if (!(svm->vmcb->control.intercept_cr_write & INTERCEPT_CR0_MASK))
 		vcpu->arch.cr0 = svm->vmcb->save.cr0;
+	if (npt_enabled)
 		vcpu->arch.cr3 = svm->vmcb->save.cr3;
-		if (mmu_reload) {
-			kvm_mmu_reset_context(vcpu);
-			kvm_mmu_load(vcpu);
-		}
-	}
-
 
 	if (svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
 		kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
@@ -2580,8 +2586,6 @@ static void svm_flush_tlb(struct kvm_vcpu *vcpu)
 
 static void svm_prepare_guest_switch(struct kvm_vcpu *vcpu)
 {
-	if (npt_enabled)
-		vcpu->fpu_active = 1;
 }
 
 static inline void sync_cr8_to_lapic(struct kvm_vcpu *vcpu)
@@ -2920,14 +2924,8 @@ static void svm_fpu_deactivate(struct kvm_vcpu *vcpu)
 {
 	struct vcpu_svm *svm = to_svm(vcpu);
 
-	if (npt_enabled) {
-		/* hack: npt requires active fpu at this time */
-		vcpu->fpu_active = 1;
-		return;
-	}
-
+	update_cr0_intercept(svm);
 	svm->vmcb->control.intercept_exceptions |= 1 << NM_VECTOR;
-	svm->vmcb->save.cr0 |= X86_CR0_TS;
 }
 
 static struct kvm_x86_ops svm_x86_ops = {
-- 
1.6.5.3