KVM: SVM: Flush Hyper-V TLB when required

The Hyper-V "EnlightenedNptTlb" enlightenment is always enabled when KVM
is running on top of Hyper-V and Hyper-V exposes support for it (which
is always). On AMD CPUs this enlightenment results in ASID invalidations
not flushing TLB entries derived from the NPT. To force the underlying
(L0) hypervisor to rebuild its shadow page tables, an explicit hypercall
is needed.

The original KVM implementation of Hyper-V's "EnlightenedNptTlb" on SVM
only added remote TLB flush hooks. This worked out fine for a while, as
sufficient remote TLB flushes where being issued in KVM to mask the
problem. Since v5.17, changes in the TDP code reduced the number of
flushes and the out-of-sync TLB prevents guests from booting
successfully.

Split svm_flush_tlb_current() into separate callbacks for the 3 cases
(guest/all/current), and issue the required Hyper-V hypercall when a
Hyper-V TLB flush is needed. The most important case where the TLB flush
was missing is when loading a new PGD, which is followed by what is now
svm_flush_tlb_current().

Cc: [email protected] # v5.17+
Fixes: 1e0c7d4 ("KVM: SVM: hyper-v: Remote TLB flush for SVM")
Link: https://lore.kernel.org/lkml/[email protected]/
Suggested-by: Sean Christopherson <[email protected]>
Signed-off-by: Jeremi Piotrowski <[email protected]>
Reviewed-by: Vitaly Kuznetsov <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

arch/x86/kvm/kvm_onhyperv.h

@@ -12,6 +12,11 @@ int hv_remote_flush_tlb_with_range(struct kvm *kvm,
 int hv_remote_flush_tlb(struct kvm *kvm);
 void hv_track_root_tdp(struct kvm_vcpu *vcpu, hpa_t root_tdp);
 #else /* !CONFIG_HYPERV */
+static inline int hv_remote_flush_tlb(struct kvm *kvm)
+{
+	return -EOPNOTSUPP;
+}
+
 static inline void hv_track_root_tdp(struct kvm_vcpu *vcpu, hpa_t root_tdp)
 {
 }

arch/x86/kvm/svm/svm.c

@@ -3729,7 +3729,7 @@ static void svm_enable_nmi_window(struct kvm_vcpu *vcpu)
 	svm->vmcb->save.rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF);
 }
 
-static void svm_flush_tlb_current(struct kvm_vcpu *vcpu)
+static void svm_flush_tlb_asid(struct kvm_vcpu *vcpu)
 {
 	struct vcpu_svm *svm = to_svm(vcpu);
 
@@ -3753,6 +3753,37 @@ static void svm_flush_tlb_current(struct kvm_vcpu *vcpu)
 		svm->current_vmcb->asid_generation--;
 }
 
+static void svm_flush_tlb_current(struct kvm_vcpu *vcpu)
+{
+	hpa_t root_tdp = vcpu->arch.mmu->root.hpa;
+
+	/*
+	 * When running on Hyper-V with EnlightenedNptTlb enabled, explicitly
+	 * flush the NPT mappings via hypercall as flushing the ASID only
+	 * affects virtual to physical mappings, it does not invalidate guest
+	 * physical to host physical mappings.
+	 */
+	if (svm_hv_is_enlightened_tlb_enabled(vcpu) && VALID_PAGE(root_tdp))
+		hyperv_flush_guest_mapping(root_tdp);
+
+	svm_flush_tlb_asid(vcpu);
+}
+
+static void svm_flush_tlb_all(struct kvm_vcpu *vcpu)
+{
+	/*
+	 * When running on Hyper-V with EnlightenedNptTlb enabled, remote TLB
+	 * flushes should be routed to hv_remote_flush_tlb() without requesting
+	 * a "regular" remote flush.  Reaching this point means either there's
+	 * a KVM bug or a prior hv_remote_flush_tlb() call failed, both of
+	 * which might be fatal to the guest.  Yell, but try to recover.
+	 */
+	if (WARN_ON_ONCE(svm_hv_is_enlightened_tlb_enabled(vcpu)))
+		hv_remote_flush_tlb(vcpu->kvm);
+
+	svm_flush_tlb_asid(vcpu);
+}
+
 static void svm_flush_tlb_gva(struct kvm_vcpu *vcpu, gva_t gva)
 {
 	struct vcpu_svm *svm = to_svm(vcpu);
@@ -4745,10 +4776,10 @@ static struct kvm_x86_ops svm_x86_ops __initdata = {
 	.set_rflags = svm_set_rflags,
 	.get_if_flag = svm_get_if_flag,
 
-	.flush_tlb_all = svm_flush_tlb_current,
+	.flush_tlb_all = svm_flush_tlb_all,
 	.flush_tlb_current = svm_flush_tlb_current,
 	.flush_tlb_gva = svm_flush_tlb_gva,
-	.flush_tlb_guest = svm_flush_tlb_current,
+	.flush_tlb_guest = svm_flush_tlb_asid,
 
 	.vcpu_pre_run = svm_vcpu_pre_run,
 	.vcpu_run = svm_vcpu_run,

arch/x86/kvm/svm/svm_onhyperv.h

@@ -6,6 +6,8 @@
 #ifndef __ARCH_X86_KVM_SVM_ONHYPERV_H__
 #define __ARCH_X86_KVM_SVM_ONHYPERV_H__
 
+#include <asm/mshyperv.h>
+
 #if IS_ENABLED(CONFIG_HYPERV)
 
 #include "kvm_onhyperv.h"
@@ -15,6 +17,14 @@ static struct kvm_x86_ops svm_x86_ops;
 
 int svm_hv_enable_l2_tlb_flush(struct kvm_vcpu *vcpu);
 
+static inline bool svm_hv_is_enlightened_tlb_enabled(struct kvm_vcpu *vcpu)
+{
+	struct hv_vmcb_enlightenments *hve = &to_svm(vcpu)->vmcb->control.hv_enlightenments;
+
+	return ms_hyperv.nested_features & HV_X64_NESTED_ENLIGHTENED_TLB &&
+	       !!hve->hv_enlightenments_control.enlightened_npt_tlb;
+}
+
 static inline void svm_hv_init_vmcb(struct vmcb *vmcb)
 {
 	struct hv_vmcb_enlightenments *hve = &vmcb->control.hv_enlightenments;
@@ -80,6 +90,11 @@ static inline void svm_hv_update_vp_id(struct vmcb *vmcb, struct kvm_vcpu *vcpu)
 }
 #else
 
+static inline bool svm_hv_is_enlightened_tlb_enabled(struct kvm_vcpu *vcpu)
+{
+	return false;
+}
+
 static inline void svm_hv_init_vmcb(struct vmcb *vmcb)
 {
 }