2024-01-24 09:10:50

by Sebastian Ene

[permalink] [raw]
Subject: [PATCH v2] KVM: arm64: Fix circular locking dependency

The rule inside kvm enforces that the vcpu->mutex is taken *inside*
kvm->lock. The rule is violated by the pkvm_create_hyp_vm() which acquires
the kvm->lock while already holding the vcpu->mutex lock from
kvm_vcpu_ioctl(). Avoid the circular locking dependency altogether by
protecting the hyp vm handle with the config_lock, much like we already
do for other forms of VM-scoped data.

Signed-off-by: Sebastian Ene <[email protected]>
Cc: [email protected]
Reviewed-by: Oliver Upton <[email protected]>
---
arch/arm64/kvm/pkvm.c | 27 +++++++++++++++++----------
1 file changed, 17 insertions(+), 10 deletions(-)

diff --git a/arch/arm64/kvm/pkvm.c b/arch/arm64/kvm/pkvm.c
index 8350fb8fee0b..b7be96a53597 100644
--- a/arch/arm64/kvm/pkvm.c
+++ b/arch/arm64/kvm/pkvm.c
@@ -101,6 +101,17 @@ void __init kvm_hyp_reserve(void)
hyp_mem_base);
}

+static void __pkvm_destroy_hyp_vm(struct kvm *host_kvm)
+{
+ if (host_kvm->arch.pkvm.handle) {
+ WARN_ON(kvm_call_hyp_nvhe(__pkvm_teardown_vm,
+ host_kvm->arch.pkvm.handle));
+ }
+
+ host_kvm->arch.pkvm.handle = 0;
+ free_hyp_memcache(&host_kvm->arch.pkvm.teardown_mc);
+}
+
/*
* Allocates and donates memory for hypervisor VM structs at EL2.
*
@@ -181,7 +192,7 @@ static int __pkvm_create_hyp_vm(struct kvm *host_kvm)
return 0;

destroy_vm:
- pkvm_destroy_hyp_vm(host_kvm);
+ __pkvm_destroy_hyp_vm(host_kvm);
return ret;
free_vm:
free_pages_exact(hyp_vm, hyp_vm_sz);
@@ -194,23 +205,19 @@ int pkvm_create_hyp_vm(struct kvm *host_kvm)
{
int ret = 0;

- mutex_lock(&host_kvm->lock);
+ mutex_lock(&host_kvm->arch.config_lock);
if (!host_kvm->arch.pkvm.handle)
ret = __pkvm_create_hyp_vm(host_kvm);
- mutex_unlock(&host_kvm->lock);
+ mutex_unlock(&host_kvm->arch.config_lock);

return ret;
}

void pkvm_destroy_hyp_vm(struct kvm *host_kvm)
{
- if (host_kvm->arch.pkvm.handle) {
- WARN_ON(kvm_call_hyp_nvhe(__pkvm_teardown_vm,
- host_kvm->arch.pkvm.handle));
- }
-
- host_kvm->arch.pkvm.handle = 0;
- free_hyp_memcache(&host_kvm->arch.pkvm.teardown_mc);
+ mutex_lock(&host_kvm->arch.config_lock);
+ __pkvm_destroy_hyp_vm(host_kvm);
+ mutex_unlock(&host_kvm->arch.config_lock);
}

int pkvm_init_host_vm(struct kvm *host_kvm)
--
2.43.0.429.g432eaa2c6b-goog



2024-01-30 21:40:59

by Marc Zyngier

[permalink] [raw]
Subject: Re: [PATCH v2] KVM: arm64: Fix circular locking dependency

On Wed, 24 Jan 2024 09:10:28 +0000, Sebastian Ene wrote:
> The rule inside kvm enforces that the vcpu->mutex is taken *inside*
> kvm->lock. The rule is violated by the pkvm_create_hyp_vm() which acquires
> the kvm->lock while already holding the vcpu->mutex lock from
> kvm_vcpu_ioctl(). Avoid the circular locking dependency altogether by
> protecting the hyp vm handle with the config_lock, much like we already
> do for other forms of VM-scoped data.

Applied to fixes, thanks!

[1/1] KVM: arm64: Fix circular locking dependency
commit: 10c02aad111df02088d1a81792a709f6a7eca6cc

Cheers,

M.
--
Without deviation from the norm, progress is not possible.