On Tue, May 07, 2013 at 05:29:09PM +0300, Gleb Natapov wrote:
> On Tue, May 07, 2013 at 01:07:06PM +0200, Cornelia Huck wrote:
> > On Tue, 7 May 2013 11:21:09 +0200
> > Cornelia Huck <cornelia.huck@xxxxxxxxxx> wrote:
> >
> > > On Tue, 7 May 2013 12:12:09 +0300
> > > Gleb Natapov <gleb@xxxxxxxxxx> wrote:
> > >
> > > > On Tue, May 07, 2013 at 05:10:41PM +0800, Asias He wrote:
> > > > > On Tue, May 07, 2013 at 11:42:07AM +0300, Gleb Natapov wrote:
> > > > > > On Tue, May 07, 2013 at 04:36:50PM +0800, Asias He wrote:
> > > > > > > On Tue, May 07, 2013 at 11:18:38AM +0300, Gleb Natapov wrote:
> > > > > > > > On Tue, May 07, 2013 at 04:14:50PM +0800, Asias He wrote:
> > > > > > > > > On Tue, May 07, 2013 at 10:55:36AM +0300, Gleb Natapov wrote:
> > > > > > > > > > On Tue, May 07, 2013 at 09:37:30AM +0800, Asias He wrote:
> > > > > > > > > > > Hi,
> > > > > > > > > > >
> > > > > > > > > > > I am seeing this with linus/master. Any ideas?
> > > > > > > > > > >
> > > > > > > > > > How reproducible it this? What HEAD are you seeing this with?
> > > > > > > > >
> > > > > > > > > Almost always. Start a guest with vhost-net on.
> > > > > > > > >
> > > > > > > > It happens during start or on VM exit? The trace shows do_exit()
> > > > > > > >
> > > > > > > > > QEMU : e3351000cd682200835763caca87adf708ed1c65
> > > > > > > > > KERNEL: 51a26ae7a14b85c99c9be470c2d28eeeba0f26a3
> > > > > > > > >
> > > > > > > > Can you try with kvm.git next branch?
> > > > > > >
> > > > > > > With next branch db6ae6158186a17165ef990bda2895ae7594b039,
> > > > > > >
> > > > > > Can you revert a0f155e9646d5f1c263f6f9aae880151100243bb and try again?
> > > > >
> > > > > The issue is gone with a0f155e9646d5f1c263f6f9aae880151100243bb
> > > > > reverted.
> > > > >
> > > > Cornelia, any ideas?
> > >
> > > irqfd_deactivate before kvm_init or after kvm_exit? Let me look...
> > >
> >
> > Hm, no idea.
> >
> For am quick look is seems that after the patch irqfd_cleanup_wq is
> initialized as part of kvm-intel/kvm-amd module, but before the patch is
> was initialized as part of kvm module. The later is how it should be,
> but I do not see why would it cause the problem.
I have a fix for this now. Will send out patch shortly.
> > Asias, could you try whether the patch below yields any interesting
> > backtraces?
> >
> > diff --git a/virt/kvm/eventfd.c b/virt/kvm/eventfd.c
> > index 64ee720..59c3810 100644
> > --- a/virt/kvm/eventfd.c
> > +++ b/virt/kvm/eventfd.c
> > @@ -92,6 +92,8 @@ struct _irqfd {
> > };
> >
> > static struct workqueue_struct *irqfd_cleanup_wq;
> > +static int irqfd_ready;
> > +static unsigned long irqfd_cnt;
> >
> > static void
> > irqfd_inject(struct work_struct *work)
> > @@ -410,6 +412,8 @@ kvm_irqfd_assign(struct kvm *kvm, struct kvm_irqfd *args)
> > */
> > fput(file);
> >
> > + irqfd_cnt++;
> > +
> > return 0;
> >
> > fail:
> > @@ -483,12 +487,16 @@ kvm_irqfd_deassign(struct kvm *kvm, struct kvm_irqfd *args)
> > */
> > flush_workqueue(irqfd_cleanup_wq);
> >
> > + irqfd_cnt--;
> > +
> > return 0;
> > }
> >
> > int
> > kvm_irqfd(struct kvm *kvm, struct kvm_irqfd *args)
> > {
> > + WARN_ON(!irqfd_ready);
> > +
> > if (args->flags & ~(KVM_IRQFD_FLAG_DEASSIGN | KVM_IRQFD_FLAG_RESAMPLE))
> > return -EINVAL;
> >
> > @@ -548,15 +556,22 @@ void kvm_irq_routing_update(struct kvm *kvm,
> > */
> > int kvm_irqfd_init(void)
> > {
> > + WARN_ON(irqfd_cnt);
> > +
> > irqfd_cleanup_wq = create_singlethread_workqueue("kvm-irqfd-cleanup");
> > if (!irqfd_cleanup_wq)
> > return -ENOMEM;
> >
> > + irqfd_ready = 1;
> > +
> > return 0;
> > }
> >
> > void kvm_irqfd_exit(void)
> > {
> > + WARN_ON(irqfd_cnt);
> > + irqfd_ready = 0;
> > +
> > destroy_workqueue(irqfd_cleanup_wq);
> > }
> > #endif
>
> --
> Gleb.
--
Asias
--
To unsubscribe from this list: send the line "unsubscribe kvm" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
