Re: [PATCH] kvm tools: Wait for all VCPU threads to exit while rebooting

[Sasha Levin <levinsasha928@gmail.com>]

On Thu, 2011-08-11 at 13:02 +0300, Pekka Enberg wrote:
> On Thu, Aug 11, 2011 at 12:47 PM, Sasha Levin <levinsasha928@gmail.com> wrote:
> > This patch changes kvm_cpu__reboot() behaviour to block until all VCPU
> > threads have ended, this allows us to assume that the guest is stopped
> > when the function has returned.
> >
> > This fixes errors on close caused by releasing KVM_RUN structure while
> > VCPUs were still running.
> >
> > Signed-off-by: Sasha Levin <levinsasha928@gmail.com>
> > ---
> >  tools/kvm/kvm-cpu.c |   10 ++++++++--
> >  1 files changed, 8 insertions(+), 2 deletions(-)
> >
> > diff --git a/tools/kvm/kvm-cpu.c b/tools/kvm/kvm-cpu.c
> > index 2f5d23c..64c0997 100644
> > --- a/tools/kvm/kvm-cpu.c
> > +++ b/tools/kvm/kvm-cpu.c
> > @@ -421,7 +421,13 @@ static void kvm_cpu__handle_coalesced_mmio(struct kvm_cpu *cpu)
> >
> >  void kvm_cpu__reboot(void)
> >  {
> > -       pthread_kill(kvm_cpus[0]->thread, SIGKVMEXIT);
> > +       int i;
> > +
> > +       for (i = 0; i < KVM_NR_CPUS; i++)
> > +               if (kvm_cpus[i]) {
> > +                       pthread_kill(kvm_cpus[i]->thread, SIGKVMEXIT);
> > +                       pthread_join(kvm_cpus[i]->thread, NULL);
> > +               }
> 
> Wouldn't it be better to first pthread_kill() all CPUs first and do
> pthread_join() in separate loop? That should be faster especially if
> there's lots of CPUs.

Actually, we don't even need a pthread_kill() there. I'll send a new
version :)

-- 

Sasha.