KVM: VMX: Execute WBINVD to keep data consistency with assigned devices

Some guest device driver may leverage the "Non-Snoop" I/O, and explicitly
WBINVD or CLFLUSH to a RAM space. Since migration may occur before WBINVD or
CLFLUSH, we need to maintain data consistency either by:
1: flushing cache (wbinvd) when the guest is scheduled out if there is no
wbinvd exit, or
2: execute wbinvd on all dirty physical CPUs when guest wbinvd exits.

Signed-off-by: Yaozu (Eddie) Dong <eddie.dong@intel.com>
Signed-off-by: Sheng Yang <sheng@linux.intel.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>

5 files changed, 67 insertions, 2 deletions

diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
index a57cdeacc4d..2bda62485c4 100644
--- a/arch/x86/include/asm/kvm_host.h
+++ b/arch/x86/include/asm/kvm_host.h
@@ -15,6 +15,7 @@
 #include <linux/mm.h>
 #include <linux/mmu_notifier.h>
 #include <linux/tracepoint.h>
+#include <linux/cpumask.h>
 
 #include <linux/kvm.h>
 #include <linux/kvm_para.h>
@@ -358,6 +359,8 @@ struct kvm_vcpu_arch {
 
 	/* fields used by HYPER-V emulation */
 	u64 hv_vapic;
+
+	cpumask_var_t wbinvd_dirty_mask;
 };
 
 struct kvm_arch {
@@ -514,6 +517,8 @@ struct kvm_x86_ops {
 
 	void (*set_supported_cpuid)(u32 func, struct kvm_cpuid_entry2 *entry);
 
+	bool (*has_wbinvd_exit)(void);
+
 	const struct trace_print_flags *exit_reasons_str;
 };
 
@@ -571,6 +576,7 @@ void kvm_emulate_cpuid(struct kvm_vcpu *vcpu);
 int kvm_emulate_halt(struct kvm_vcpu *vcpu);
 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address);
 int emulate_clts(struct kvm_vcpu *vcpu);
+int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu);
 
 void kvm_get_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg);
 int kvm_load_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector, int seg);
diff --git a/arch/x86/kvm/emulate.c b/arch/x86/kvm/emulate.c
index abb8cec420a..e8bdddc4509 100644
--- a/arch/x86/kvm/emulate.c
+++ b/arch/x86/kvm/emulate.c
@@ -3138,8 +3138,11 @@ twobyte_insn:
 		emulate_clts(ctxt->vcpu);
 		c->dst.type = OP_NONE;
 		break;
-	case 0x08:		/* invd */
 	case 0x09:		/* wbinvd */
+		kvm_emulate_wbinvd(ctxt->vcpu);
+		c->dst.type = OP_NONE;
+		break;
+	case 0x08:		/* invd */
 	case 0x0d:		/* GrpP (prefetch) */
 	case 0x18:		/* Grp16 (prefetch/nop) */
 		c->dst.type = OP_NONE;
diff --git a/arch/x86/kvm/svm.c b/arch/x86/kvm/svm.c
index 587b99d37d4..56c9b6bd765 100644
--- a/arch/x86/kvm/svm.c
+++ b/arch/x86/kvm/svm.c
@@ -3424,6 +3424,11 @@ static bool svm_rdtscp_supported(void)
 	return false;
 }
 
+static bool svm_has_wbinvd_exit(void)
+{
+	return true;
+}
+
 static void svm_fpu_deactivate(struct kvm_vcpu *vcpu)
 {
 	struct vcpu_svm *svm = to_svm(vcpu);
@@ -3508,6 +3513,8 @@ static struct kvm_x86_ops svm_x86_ops = {
 	.rdtscp_supported = svm_rdtscp_supported,
 
 	.set_supported_cpuid = svm_set_supported_cpuid,
+
+	.has_wbinvd_exit = svm_has_wbinvd_exit,
 };
 
 static int __init svm_init(void)
diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
index 661c6e199b4..4dfb1dc09c8 100644
--- a/arch/x86/kvm/vmx.c
+++ b/arch/x86/kvm/vmx.c
@@ -412,6 +412,12 @@ static inline bool cpu_has_virtual_nmis(void)
 	return vmcs_config.pin_based_exec_ctrl & PIN_BASED_VIRTUAL_NMIS;
 }
 
+static inline bool cpu_has_vmx_wbinvd_exit(void)
+{
+	return vmcs_config.cpu_based_2nd_exec_ctrl &
+		SECONDARY_EXEC_WBINVD_EXITING;
+}
+
 static inline bool report_flexpriority(void)
 {
 	return flexpriority_enabled;
@@ -3397,7 +3403,7 @@ static int handle_invlpg(struct kvm_vcpu *vcpu)
 static int handle_wbinvd(struct kvm_vcpu *vcpu)
 {
 	skip_emulated_instruction(vcpu);
-	/* TODO: Add support for VT-d/pass-through device */
+	kvm_emulate_wbinvd(vcpu);
 	return 1;
 }
 
@@ -4347,6 +4353,8 @@ static struct kvm_x86_ops vmx_x86_ops = {
 	.rdtscp_supported = vmx_rdtscp_supported,
 
 	.set_supported_cpuid = vmx_set_supported_cpuid,
+
+	.has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
 };
 
 static int __init vmx_init(void)
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index 27322d34123..3d72fc06705 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -1783,8 +1783,28 @@ out:
 	return r;
 }
 
+static void wbinvd_ipi(void *garbage)
+{
+	wbinvd();
+}
+
+static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
+{
+	return vcpu->kvm->arch.iommu_domain &&
+		!(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY);
+}
+
 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 {
+	/* Address WBINVD may be executed by guest */
+	if (need_emulate_wbinvd(vcpu)) {
+		if (kvm_x86_ops->has_wbinvd_exit())
+			cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
+		else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
+			smp_call_function_single(vcpu->cpu,
+					wbinvd_ipi, NULL, 1);
+	}
+
 	kvm_x86_ops->vcpu_load(vcpu, cpu);
 	if (unlikely(per_cpu(cpu_tsc_khz, cpu) == 0)) {
 		unsigned long khz = cpufreq_quick_get(cpu);
@@ -3660,6 +3680,21 @@ int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
 	return X86EMUL_CONTINUE;
 }
 
+int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
+{
+	if (!need_emulate_wbinvd(vcpu))
+		return X86EMUL_CONTINUE;
+
+	if (kvm_x86_ops->has_wbinvd_exit()) {
+		smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
+				wbinvd_ipi, NULL, 1);
+		cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
+	}
+	wbinvd();
+	return X86EMUL_CONTINUE;
+}
+EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
+
 int emulate_clts(struct kvm_vcpu *vcpu)
 {
 	kvm_x86_ops->set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
@@ -5263,6 +5298,7 @@ void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
 		vcpu->arch.time_page = NULL;
 	}
 
+	free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
 	fx_free(vcpu);
 	kvm_x86_ops->vcpu_free(vcpu);
 }
@@ -5392,7 +5428,12 @@ int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
 	}
 	vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
 
+	if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL))
+		goto fail_free_mce_banks;
+
 	return 0;
+fail_free_mce_banks:
+	kfree(vcpu->arch.mce_banks);
 fail_free_lapic:
 	kvm_free_lapic(vcpu);
 fail_mmu_destroy: