Re: [PATCH 2/2] kvmtool: assume dead vcpus are paused too

[Will Deacon <will.deacon@arm.com>]

On Wed, Nov 04, 2015 at 06:51:12PM -0500, Sasha Levin wrote:
> On 11/04/2015 06:51 AM, Will Deacon wrote:
> > +	mutex_lock(&pause_lock);
> > +
> > +	/* The kvm->cpus array contains a null pointer in the last location */
> > +	for (i = 0; ; i++) {
> > +		if (kvm->cpus[i])
> > +			pthread_kill(kvm->cpus[i]->thread, SIGKVMEXIT);
> > +		else
> > +			break;
> > +	}
> > +
> > +	kvm__continue(kvm);
> 
> In this scenario: if we grabbed pause_lock, signaled vcpu0 to exit, and it did
> before we called kvm__continue(), we won't end up releasing pause_lock, which
> might cause a lockup later, no?

Hmm, yeah, maybe that should be an explicit mutex_unlock rather than a
call to kvm__continue.

Will