[Devel] [PATCH vz7.46 1/7] KVM: hyperv: fix locking of struct kvm_hv fields

[Roman Kagan]

From: Paolo Bonzini <pbonzini at redhat.com>

Introduce a new mutex to avoid an AB-BA deadlock between kvm->lock and
vcpu->mutex.  Protect accesses in kvm_hv_setup_tsc_page too, as suggested
by Roman.

Reported-by: Dmitry Vyukov <dvyukov at google.com>
Reviewed-by: Roman Kagan <rkagan at virtuozzo.com>
Signed-off-by: Paolo Bonzini <pbonzini at redhat.com>
(cherry picked from commit 3f5ad8be3713572f3946b69eb376206153d0ea2d)
Signed-off-by: Roman Kagan <rkagan at virtuozzo.com>
---
 Documentation/virtual/kvm/locking.txt | 16 +++++++++++++++-
 arch/x86/include/asm/kvm_host.h       |  1 +
 arch/x86/kvm/hyperv.c                 | 24 +++++++++++++++---------
 arch/x86/kvm/x86.c                    |  1 +
 4 files changed, 32 insertions(+), 10 deletions(-)

diff --git a/Documentation/virtual/kvm/locking.txt b/Documentation/virtual/kvm/locking.txt
index 19f94a6b9bb0..fd38c21b7c5c 100644
--- a/Documentation/virtual/kvm/locking.txt
+++ b/Documentation/virtual/kvm/locking.txt
@@ -4,7 +4,21 @@ KVM Lock Overview
 1. Acquisition Orders
 ---------------------
 
-(to be written)
+The acquisition orders for mutexes are as follows:
+
+- kvm->lock is taken outside vcpu->mutex
+
+- kvm->lock is taken outside kvm->slots_lock and kvm->irq_lock
+
+- kvm->slots_lock is taken outside kvm->irq_lock, though acquiring
+  them together is quite rare.
+
+On x86, vcpu->mutex is taken outside kvm->arch.hyperv.hv_lock.
+
+For spinlocks, kvm_lock is taken outside kvm->mmu_lock.
+
+Everything else is a leaf: no other lock is taken inside the critical
+sections.
 
 2: Exception
 ------------
diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
index feaec5007610..bcf17448dcab 100644
--- a/arch/x86/include/asm/kvm_host.h
+++ b/arch/x86/include/asm/kvm_host.h
@@ -668,6 +668,7 @@ struct kvm_apic_map {
 
 /* Hyper-V emulation context */
 struct kvm_hv {
+	struct mutex hv_lock;
 	u64 hv_guest_os_id;
 	u64 hv_hypercall;
 	u64 hv_tsc_page;
diff --git a/arch/x86/kvm/hyperv.c b/arch/x86/kvm/hyperv.c
index d32c41855edd..380e238cac7e 100644
--- a/arch/x86/kvm/hyperv.c
+++ b/arch/x86/kvm/hyperv.c
@@ -878,6 +878,10 @@ void kvm_hv_setup_tsc_page(struct kvm *kvm,
 	if (!(hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE))
 		return;
 
+	mutex_lock(&kvm->arch.hyperv.hv_lock);
+	if (!(hv->hv_tsc_page & HV_X64_MSR_TSC_REFERENCE_ENABLE))
+		goto out_unlock;
+
 	gfn = hv->hv_tsc_page >> HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT;
 	/*
 	 * Because the TSC parameters only vary when there is a
@@ -885,7 +889,7 @@ void kvm_hv_setup_tsc_page(struct kvm *kvm,
 	 */
 	if (unlikely(kvm_read_guest(kvm, gfn_to_gpa(gfn),
 				    &tsc_seq, sizeof(tsc_seq))))
-		return;
+		goto out_unlock;
 
 	/*
 	 * While we're computing and writing the parameters, force the
@@ -894,15 +898,15 @@ void kvm_hv_setup_tsc_page(struct kvm *kvm,
 	hv->tsc_ref.tsc_sequence = 0;
 	if (kvm_write_guest(kvm, gfn_to_gpa(gfn),
 			    &hv->tsc_ref, sizeof(hv->tsc_ref.tsc_sequence)))
-		return;
+		goto out_unlock;
 
 	if (!compute_tsc_page_parameters(hv_clock, &hv->tsc_ref))
-		return;
+		goto out_unlock;
 
 	/* Ensure sequence is zero before writing the rest of the struct.  */
 	smp_wmb();
 	if (kvm_write_guest(kvm, gfn_to_gpa(gfn), &hv->tsc_ref, sizeof(hv->tsc_ref)))
-		return;
+		goto out_unlock;
 
 	/*
 	 * Now switch to the TSC page mechanism by writing the sequence.
@@ -917,6 +921,8 @@ void kvm_hv_setup_tsc_page(struct kvm *kvm,
 	hv->tsc_ref.tsc_sequence = tsc_seq;
 	kvm_write_guest(kvm, gfn_to_gpa(gfn),
 			&hv->tsc_ref, sizeof(hv->tsc_ref.tsc_sequence));
+out_unlock:
+	mutex_unlock(&kvm->arch.hyperv.hv_lock);
 }
 
 static int kvm_hv_set_msr_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data,
@@ -1166,9 +1172,9 @@ int kvm_hv_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
 	if (kvm_hv_msr_partition_wide(msr)) {
 		int r;
 
-		mutex_lock(&vcpu->kvm->lock);
+		mutex_lock(&vcpu->kvm->arch.hyperv.hv_lock);
 		r = kvm_hv_set_msr_pw(vcpu, msr, data, host);
-		mutex_unlock(&vcpu->kvm->lock);
+		mutex_unlock(&vcpu->kvm->arch.hyperv.hv_lock);
 		return r;
 	} else
 		return kvm_hv_set_msr(vcpu, msr, data, host);
@@ -1179,9 +1185,9 @@ int kvm_hv_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
 	if (kvm_hv_msr_partition_wide(msr)) {
 		int r;
 
-		mutex_lock(&vcpu->kvm->lock);
+		mutex_lock(&vcpu->kvm->arch.hyperv.hv_lock);
 		r = kvm_hv_get_msr_pw(vcpu, msr, pdata);
-		mutex_unlock(&vcpu->kvm->lock);
+		mutex_unlock(&vcpu->kvm->arch.hyperv.hv_lock);
 		return r;
 	} else
 		return kvm_hv_get_msr(vcpu, msr, pdata);
@@ -1189,7 +1195,7 @@ int kvm_hv_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
 
 bool kvm_hv_hypercall_enabled(struct kvm *kvm)
 {
-	return kvm->arch.hyperv.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
+	return READ_ONCE(kvm->arch.hyperv.hv_hypercall) & HV_X64_MSR_HYPERCALL_ENABLE;
 }
 
 static void kvm_hv_hypercall_set_result(struct kvm_vcpu *vcpu, u64 result)
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index 04c94ff0b06c..5db5b179b47b 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -7875,6 +7875,7 @@ int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
 
 	raw_spin_lock_init(&kvm->arch.tsc_write_lock);
 	mutex_init(&kvm->arch.apic_map_lock);
+	mutex_init(&kvm->arch.hyperv.hv_lock);
 	spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
 
 	kvm->arch.kvmclock_offset = -ktime_get_boot_ns();
-- 
2.14.3