Re: [PATCH v2 2/4] KVM: arm64: vgic: Explicitly implement vgic_dist::ready ordering

[Keir Fraser <keirf@google.com>]

On Sat, Jul 19, 2025 at 10:15:56AM +0800, Yao Yuan wrote:
> On Fri, Jul 18, 2025 at 08:00:17AM -0700, Sean Christopherson wrote:
> > On Thu, Jul 17, 2025, Yao Yuan wrote:
> > > On Wed, Jul 16, 2025 at 11:07:35AM +0800, Keir Fraser wrote:
> > > > In preparation to remove synchronize_srcu() from MMIO registration,
> > > > remove the distributor's dependency on this implicit barrier by
> > > > direct acquire-release synchronization on the flag write and its
> > > > lock-free check.
> > > >
> > > > Signed-off-by: Keir Fraser <keirf@google.com>
> > > > ---
> > > >  arch/arm64/kvm/vgic/vgic-init.c | 11 ++---------
> > > >  1 file changed, 2 insertions(+), 9 deletions(-)
> > > >
> > > > diff --git a/arch/arm64/kvm/vgic/vgic-init.c b/arch/arm64/kvm/vgic/vgic-init.c
> > > > index 502b65049703..bc83672e461b 100644
> > > > --- a/arch/arm64/kvm/vgic/vgic-init.c
> > > > +++ b/arch/arm64/kvm/vgic/vgic-init.c
> > > > @@ -567,7 +567,7 @@ int kvm_vgic_map_resources(struct kvm *kvm)
> > > >  	gpa_t dist_base;
> > > >  	int ret = 0;
> > > >
> > > > -	if (likely(dist->ready))
> > > > +	if (likely(smp_load_acquire(&dist->ready)))
> > > >  		return 0;
> > > >
> > > >  	mutex_lock(&kvm->slots_lock);
> > > > @@ -598,14 +598,7 @@ int kvm_vgic_map_resources(struct kvm *kvm)
> > > >  		goto out_slots;
> > > >  	}
> > > >
> > > > -	/*
> > > > -	 * kvm_io_bus_register_dev() guarantees all readers see the new MMIO
> > > > -	 * registration before returning through synchronize_srcu(), which also
> > > > -	 * implies a full memory barrier. As such, marking the distributor as
> > > > -	 * 'ready' here is guaranteed to be ordered after all vCPUs having seen
> > > > -	 * a completely configured distributor.
> > > > -	 */
> > > > -	dist->ready = true;
> > > > +	smp_store_release(&dist->ready, true);
> > >
> > > No need the store-release and load-acquire for replacing
> > > synchronize_srcu_expedited() w/ call_srcu() IIUC:
> >
> > This isn't about using call_srcu(), because it's not actually about kvm->buses.
> > This code is concerned with ensuring that all stores to kvm->arch.vgic are ordered
> > before the store to set kvm->arch.vgic.ready, so that vCPUs never see "ready==true"
> > with a half-baked distributor.
> >
> > In the current code, kvm_vgic_map_resources() relies on the synchronize_srcu() in
> > kvm_io_bus_register_dev() to provide the ordering guarantees.  Switching to
> > smp_store_release() + smp_load_acquire() removes the dependency on the
> > synchronize_srcu() so that the synchronize_srcu() call can be safely removed.
> 
> Yes, I understand this and agree with your point.
> 
> Just for discusstion: I thought it should also work even w/o
> introduce the load acqure + store release after switch to
> call_srcu(): The smp_mb() in call_srcu() order the all store
> to kvm->arch.vgic before store kvm->arch.vgic.ready in
> current implementation.

The load-acquire would still be required, to ensure that accesses to
kvm->arch.vgic do not get reordered earlier than the lock-free check
of kvm->arch.vgic.ready. Otherwise that CPU could see that the vgic is
initialised, but then use speculated reads of uninitialised vgic state.

> >