Re: [PATCH v2 4/4] KVM: arm64: Use config_lock to protect vgic state

From: Oliver Upton <oliver.upton@linux.dev>
To: Marc Zyngier <maz@kernel.org>
Cc: James Morse <james.morse@arm.com>,
	Suzuki K Poulose <suzuki.poulose@arm.com>,
	kvmarm@lists.linux.dev, Zenghui Yu <yuzenghui@huawei.com>,
	linux-arm-kernel@lists.infradead.org,
	Sean Christopherson <seanjc@google.com>
Subject: Re: [PATCH v2 4/4] KVM: arm64: Use config_lock to protect vgic state
Date: Thu, 23 Mar 2023 19:18:05 +0000	[thread overview]
Message-ID: <ZByl7ZB835jySq/x@linux.dev> (raw)
In-Reply-To: <87o7olgfjs.wl-maz@kernel.org>

On Wed, Mar 22, 2023 at 12:02:15PM +0000, Marc Zyngier wrote:
> On Thu, 16 Mar 2023 21:14:12 +0000,
> Oliver Upton <oliver.upton@linux.dev> wrote:

[...]

> > --- a/arch/arm64/kvm/vgic/vgic-its.c
> > +++ b/arch/arm64/kvm/vgic/vgic-its.c
> > @@ -2043,7 +2043,10 @@ static int vgic_its_attr_regs_access(struct kvm_device *dev,
> >  	if (offset & align)
> >  		return -EINVAL;
> >  
> > -	mutex_lock(&dev->kvm->lock);
> > +	if (!lock_all_vcpus(dev->kvm))
> > +		return -EBUSY;
> > +
> > +	mutex_lock(&dev->kvm->arch.config_lock);
> 
> Huh, that's fishy. The whole "lock the VM and the lock the individual
> vcpus" is there to prevent a concurrent creation of a vcpu while we're
> doing stuff that affects them all. Allowing a new vcpu to come online
> while this sequence is happening is ... unexpected.
> 
> Why do we need to drop this initial lock? I'd expect them to be
> completely cumulative.

Urgh.. Yes, you're right. I'll go with kvm->lock -> lock_all_vcpus() ->
kvm->config_lock in the next spin to guard against the vCPU creation
race.

-- 
Thanks,
Oliver