Re: [PATCH v3] KVM: remove buggy vcpu id check on vcpu creation

["Radim Krčmář" <rkrcmar@redhat.com>]

2016-04-20 19:53+0200, Greg Kurz:
> On Wed, 20 Apr 2016 19:27:06 +0200
> Radim Krčmář <rkrcmar@redhat.com> wrote:
>> 2016-04-20 18:09+0100, James Hogan:
>> > On Wed, Apr 20, 2016 at 07:02:10PM +0200, Radim Krčmář wrote:  
>> >> 2016-04-20 17:44+0200, Greg Kurz:  
>> >> > diff --git a/arch/mips/kvm/mips.c b/arch/mips/kvm/mips.c
>> >> > index 70ef1a43c114..0278ea146db5 100644
>> >> > --- a/arch/mips/kvm/mips.c
>> >> > +++ b/arch/mips/kvm/mips.c
>> >> > @@ -248,9 +248,14 @@ struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
>> >> >  	int err, size, offset;
>> >> >  	void *gebase;
>> >> >  	int i;
>> >> > +	struct kvm_vcpu *vcpu;
>> >> >  
>> >> > -	struct kvm_vcpu *vcpu = kzalloc(sizeof(struct kvm_vcpu), GFP_KERNEL);
>> >> > +	if (id >= KVM_MAX_VCPUS) {
>> >> > +		err = -EINVAL;
>> >> > +		goto out;  
>> >> 
>> >> 'vcpu' looks undefined at this point, so kfree in 'out:' may bug.  
>> > 
>> > Thats out_free_cpu I think?  
>> 
>> My bad, it is.  Thank you!
>> 
> 
> I kept the goto based construct because it was done this way for kzalloc().
> but I agree that 'return ERR_PTR(-EINVAL)' may look more explicit.
> 
> Worth a v4 ?

No, it is consistent with kzalloc fault handling this way.

I was going to queue it, but found an issue with kvm_get_vcpu_by_id()
that would allow the guest to create multiple VCPUs with the same id,
which led to an unfortunate discourse on KVM API.
(Please see a new thread.)