On Wed, May 08, 2024 at 07:39:20AM +0100, Marc Zyngier wrote:
> On Tue, 07 May 2024 20:29:12 +0100,
> Oliver Upton <oliver.upton@xxxxxxxxx> wrote:
> >
> > A particularly annoying userspace could create a vCPU after KVM has
> > computed mpidr_data for the VM, either by racing against VGIC
> > initialization or having a userspace irqchip.
> >
> > In any case, this means mpidr_data no longer fully describes the VM, and
> > attempts to find the new vCPU with kvm_mpidr_to_vcpu() will fail. The
> > fix is to discard mpidr_data altogether, as it is only a performance
> > optimization and not required for correctness. In all likelihood KVM
> > will recompute the mappings when KVM_RUN is called on the new vCPU.
> >
> > Note that reads of mpidr_data are not guarded by a lock; promote to RCU
> > to cope with the possibility of mpidr_data being invalidated at runtime.
> >
> > Fixes: 54a8006d0b49 ("KVM: arm64: Fast-track kvm_mpidr_to_vcpu() when mpidr_data is available")
> > Signed-off-by: Oliver Upton <oliver.upton@xxxxxxxxx>
> > ---
> > arch/arm64/kvm/arm.c | 49 ++++++++++++++++++++++++++++++++++++--------
> > 1 file changed, 40 insertions(+), 9 deletions(-)
> >
> > diff --git a/arch/arm64/kvm/arm.c b/arch/arm64/kvm/arm.c
> > index c4a0a35e02c7..0d845131a0e0 100644
> > --- a/arch/arm64/kvm/arm.c
> > +++ b/arch/arm64/kvm/arm.c
> > @@ -195,6 +195,22 @@ void kvm_arch_create_vm_debugfs(struct kvm *kvm)
> > kvm_sys_regs_create_debugfs(kvm);
> > }
> >
> > +static void kvm_destroy_mpidr_data(struct kvm *kvm)
> > +{
> > + struct kvm_mpidr_data *data;
> > +
> > + mutex_lock(&kvm->arch.config_lock);
> > +
> > + data = rcu_dereference_raw(kvm->arch.mpidr_data);
>
> I'm slightly worried by this. Why can't we use the "cooked" version?
> If anything I'd like to see a comment about this, as it is usually
> frowned upon.
No reason other than my own laziness... This really should be:
rcu_dereference_protected(kvm->arch.mpidr_data,
lockdep_is_held(&kvm->arch.config_lock));
since we're behind the update-side lock.
--
Thanks,
Oliver
