[PATCH v2] KVM: SEV: Mark nested locking of vcpu->mutex

[PATCH v2] KVM: SEV: Mark nested locking of vcpu->mutex

[Peter Gonda]

svm_vm_migrate_from() uses sev_lock_vcpus_for_migration() to lock all
source and target vcpu->mutex. Mark the nested subclasses to avoid false
positives from lockdep.

Warning example:
============================================
WARNING: possible recursive locking detected
5.17.0-dbg-DEV #15 Tainted: G           O
--------------------------------------------
sev_migrate_tes/18859 is trying to acquire lock:
ffff8d672d484238 (&vcpu->mutex){+.+.}-{3:3}, at: sev_lock_vcpus_for_migration+0x7e/0x150
but task is already holding lock:
ffff8d67703f81f8 (&vcpu->mutex){+.+.}-{3:3}, at: sev_lock_vcpus_for_migration+0x7e/0x150
other info that might help us debug this:
 Possible unsafe locking scenario:
       CPU0
       ----
  lock(&vcpu->mutex);
  lock(&vcpu->mutex);
 *** DEADLOCK ***
 May be due to missing lock nesting notation
3 locks held by sev_migrate_tes/18859:
 #0: ffff9302f91323b8 (&kvm->lock){+.+.}-{3:3}, at: sev_vm_move_enc_context_from+0x96/0x740
 #1: ffff9302f906a3b8 (&kvm->lock/1){+.+.}-{3:3}, at: sev_vm_move_enc_context_from+0xae/0x740
 #2: ffff8d67703f81f8 (&vcpu->mutex){+.+.}-{3:3}, at: sev_lock_vcpus_for_migration+0x7e/0x150

Fixes: b56639318bb2b ("KVM: SEV: Add support for SEV intra host migration")
Reported-by: John Sperbeck<jsperbeck@google.com>
Suggested-by: David Rientjes <rientjes@google.com>
Suggested-by: Sean Christopherson <seanjc@google.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: kvm@vger.kernel.org
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Peter Gonda <pgonda@google.com>

---

Tested by running sev_migrate_tests with lockdep enabled. Before we see
a warning from sev_lock_vcpus_for_migration(). After we get no warnings.

---
 arch/x86/kvm/svm/sev.c | 15 +++++++++++----
 1 file changed, 11 insertions(+), 4 deletions(-)

diff --git a/arch/x86/kvm/svm/sev.c b/arch/x86/kvm/svm/sev.c
index 75fa6dd268f0..673e1ee2cfc9 100644
--- a/arch/x86/kvm/svm/sev.c
+++ b/arch/x86/kvm/svm/sev.c
@@ -1591,14 +1591,21 @@ static void sev_unlock_two_vms(struct kvm *dst_kvm, struct kvm *src_kvm)
 	atomic_set_release(&src_sev->migration_in_progress, 0);
 }
 
+#define SEV_MIGRATION_SOURCE 0
+#define SEV_MIGRATION_TARGET 1
 
-static int sev_lock_vcpus_for_migration(struct kvm *kvm)
+/*
+ * To avoid lockdep warnings callers should pass @vm argument with either
+ * SEV_MIGRATION_SOURCE or SEV_MIGRATE_TARGET. This allows subclassing of all
+ * vCPU mutex locks.
+ */
+static int sev_lock_vcpus_for_migration(struct kvm *kvm, int vm)
 {
 	struct kvm_vcpu *vcpu;
 	unsigned long i, j;
 
 	kvm_for_each_vcpu(i, vcpu, kvm) {
-		if (mutex_lock_killable(&vcpu->mutex))
+		if (mutex_lock_killable_nested(&vcpu->mutex, i * 2 + vm))
 			goto out_unlock;
 	}
 
@@ -1745,10 +1752,10 @@ int sev_vm_move_enc_context_from(struct kvm *kvm, unsigned int source_fd)
 		charged = true;
 	}
 
-	ret = sev_lock_vcpus_for_migration(kvm);
+	ret = sev_lock_vcpus_for_migration(kvm, SEV_MIGRATION_SOURCE);
 	if (ret)
 		goto out_dst_cgroup;
-	ret = sev_lock_vcpus_for_migration(source_kvm);
+	ret = sev_lock_vcpus_for_migration(source_kvm, SEV_MIGRATION_TARGET);
 	if (ret)
 		goto out_dst_vcpu;
 
-- 
2.35.1.1094.g7c7d902a7c-goog

Re: [PATCH v2] KVM: SEV: Mark nested locking of vcpu->mutex

[Sean Christopherson]

On Tue, Apr 05, 2022, Peter Gonda wrote:
> diff --git a/arch/x86/kvm/svm/sev.c b/arch/x86/kvm/svm/sev.c
> index 75fa6dd268f0..673e1ee2cfc9 100644
> --- a/arch/x86/kvm/svm/sev.c
> +++ b/arch/x86/kvm/svm/sev.c
> @@ -1591,14 +1591,21 @@ static void sev_unlock_two_vms(struct kvm *dst_kvm, struct kvm *src_kvm)
>  	atomic_set_release(&src_sev->migration_in_progress, 0);
>  }
>  
> +#define SEV_MIGRATION_SOURCE 0
> +#define SEV_MIGRATION_TARGET 1
>  
> -static int sev_lock_vcpus_for_migration(struct kvm *kvm)
> +/*
> + * To avoid lockdep warnings callers should pass @vm argument with either

I think it's important to call that these are false positives, saying "avoid
lockdep warnings" suggests we're intentionally not fixing bugs :-)

> + * SEV_MIGRATION_SOURCE or SEV_MIGRATE_TARGET. This allows subclassing of all
> + * vCPU mutex locks.
> + */

If we use an enum, that'll make the param self-documenting.  And we can also use
that to eliminate the remaining magic number '2'.  E.g. this as fixup.

---
 arch/x86/kvm/svm/sev.c | 20 +++++++++++++-------
 1 file changed, 13 insertions(+), 7 deletions(-)

diff --git a/arch/x86/kvm/svm/sev.c b/arch/x86/kvm/svm/sev.c
index 673e1ee2cfc9..1e07d5d3f85a 100644
--- a/arch/x86/kvm/svm/sev.c
+++ b/arch/x86/kvm/svm/sev.c
@@ -1591,21 +1591,27 @@ static void sev_unlock_two_vms(struct kvm *dst_kvm, struct kvm *src_kvm)
 	atomic_set_release(&src_sev->migration_in_progress, 0);
 }

-#define SEV_MIGRATION_SOURCE 0
-#define SEV_MIGRATION_TARGET 1

 /*
- * To avoid lockdep warnings callers should pass @vm argument with either
- * SEV_MIGRATION_SOURCE or SEV_MIGRATE_TARGET. This allows subclassing of all
- * vCPU mutex locks.
+ * To suppress lockdep false positives, subclass all vCPU mutex locks by
+ * assigning even numbers to the source vCPUs and odd numbers to destination
+ * vCPUs based on the vCPU's index.
  */
-static int sev_lock_vcpus_for_migration(struct kvm *kvm, int vm)
+enum sev_migration_role {
+	SEV_MIGRATION_SOURCE = 0,
+	SEV_MIGRATION_TARGET,
+	SEV_NR_MIGRATION_ROLES,
+};
+
+static int sev_lock_vcpus_for_migration(struct kvm *kvm,
+					enum sev_migration_role role)
 {
 	struct kvm_vcpu *vcpu;
 	unsigned long i, j;

 	kvm_for_each_vcpu(i, vcpu, kvm) {
-		if (mutex_lock_killable_nested(&vcpu->mutex, i * 2 + vm))
+		if (mutex_lock_killable_nested(&vcpu->mutex,
+					       i * SEV_NR_MIGRATION_ROLES + role))
 			goto out_unlock;
 	}


base-commit: 6600ddafa53b35fd5c869aff4a5efb981ed06955
--

Re: [PATCH v2] KVM: SEV: Mark nested locking of vcpu->mutex

[Peter Gonda]

On Wed, Apr 6, 2022 at 12:06 PM Sean Christopherson <seanjc@google.com> wrote:
>
> On Tue, Apr 05, 2022, Peter Gonda wrote:
> > diff --git a/arch/x86/kvm/svm/sev.c b/arch/x86/kvm/svm/sev.c
> > index 75fa6dd268f0..673e1ee2cfc9 100644
> > --- a/arch/x86/kvm/svm/sev.c
> > +++ b/arch/x86/kvm/svm/sev.c
> > @@ -1591,14 +1591,21 @@ static void sev_unlock_two_vms(struct kvm *dst_kvm, struct kvm *src_kvm)
> >       atomic_set_release(&src_sev->migration_in_progress, 0);
> >  }
> >
> > +#define SEV_MIGRATION_SOURCE 0
> > +#define SEV_MIGRATION_TARGET 1
> >
> > -static int sev_lock_vcpus_for_migration(struct kvm *kvm)
> > +/*
> > + * To avoid lockdep warnings callers should pass @vm argument with either
>
> I think it's important to call that these are false positives, saying "avoid
> lockdep warnings" suggests we're intentionally not fixing bugs :-)
>
> > + * SEV_MIGRATION_SOURCE or SEV_MIGRATE_TARGET. This allows subclassing of all
> > + * vCPU mutex locks.
> > + */
>
> If we use an enum, that'll make the param self-documenting.  And we can also use
> that to eliminate the remaining magic number '2'.  E.g. this as fixup.

Sounds good. I took this fixup into the V3. Thanks.