Linux Virtualization - Jan 2016 - [PATCH RFC] virtio_balloon: refill by config handler

Peter Mladek reported that balloon might never refill completely after
restore.  This is because fill_balloon is only called once there.
Calling fill_balloon repeatedly seems too aggressive,
especially in light of the fact that it sleeps on failure: let's
wake the config change handler and fill it asynchronously.

Reported-by: Petr Mladek <pmladek at suse.com>
Signed-off-by: Michael S. Tsirkin <mst at redhat.com>
---

I was unable to test this - for some reason my test VM
doesn't resume (with or without the patch).
Petr, does this work for you?

 drivers/virtio/virtio_balloon.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/virtio/virtio_balloon.c b/drivers/virtio/virtio_balloon.c
index 7efc329..ee29473 100644
--- a/drivers/virtio/virtio_balloon.c
+++ b/drivers/virtio/virtio_balloon.c
@@ -589,7 +589,7 @@ static int virtballoon_restore(struct virtio_device *vdev)
 
 	virtio_device_ready(vdev);
 
-	fill_balloon(vb, towards_target(vb));
+	wake_up(&vb->config_change);
 	update_balloon_size(vb);
 	return 0;
 }
-- 
MST

On Fri 2016-01-01 12:13:48, Michael S. Tsirkin wrote:
> Peter Mladek reported that balloon might never refill completely after
> restore.  This is because fill_balloon is only called once there.
> Calling fill_balloon repeatedly seems too aggressive,
> especially in light of the fact that it sleeps on failure: let's
> wake the config change handler and fill it asynchronously.
> 
> Reported-by: Petr Mladek <pmladek at suse.com>
> Signed-off-by: Michael S. Tsirkin <mst at redhat.com>
> ---
> 
> I was unable to test this - for some reason my test VM
> doesn't resume (with or without the patch).
> Petr, does this work for you?
Ah, I have just realized that the problem happens only after the
conversion to the workqueue. It works without this patch with the
current code. The balloon kthread is waken when released from
__refrigerator().

I am sorry for the noise. I wanted to split the conversion to
workqueues into two patches to make it better readable. We need to
queue the balloon resizing work here. I wrongly assumed that the
original kthread must have the same problem.

Best Regards,
Petr
>  drivers/virtio/virtio_balloon.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/drivers/virtio/virtio_balloon.c
b/drivers/virtio/virtio_balloon.c
> index 7efc329..ee29473 100644
> --- a/drivers/virtio/virtio_balloon.c
> +++ b/drivers/virtio/virtio_balloon.c
> @@ -589,7 +589,7 @@ static int virtballoon_restore(struct virtio_device
*vdev)
>  
>  	virtio_device_ready(vdev);
>  
> -	fill_balloon(vb, towards_target(vb));
> +	wake_up(&vb->config_change);
>  	update_balloon_size(vb);
>  	return 0;
>  }
> -- 
> MST

On Tue, Jan 05, 2016 at 03:37:33PM +0100, Petr Mladek
wrote:
> On Fri 2016-01-01 12:13:48, Michael S. Tsirkin wrote:
> > Peter Mladek reported that balloon might never refill completely after
> > restore.  This is because fill_balloon is only called once there.
> > Calling fill_balloon repeatedly seems too aggressive,
> > especially in light of the fact that it sleeps on failure: let's
> > wake the config change handler and fill it asynchronously.
> > 
> > Reported-by: Petr Mladek <pmladek at suse.com>
> > Signed-off-by: Michael S. Tsirkin <mst at redhat.com>
> > ---
> > 
> > I was unable to test this - for some reason my test VM
> > doesn't resume (with or without the patch).
> > Petr, does this work for you?
> 
> Ah, I have just realized that the problem happens only after the
> conversion to the workqueue. It works without this patch with the
> current code. The balloon kthread is waken when released from
> __refrigerator().
In fact, maybe we should drop fill_balloon and
update_balloon_size here?
Resume will finish faster ...
> I am sorry for the noise. I wanted to split the conversion to
> workqueues into two patches to make it better readable. We need to
> queue the balloon resizing work here. I wrongly assumed that the
> original kthread must have the same problem.
> 
> Best Regards,
> Petr
> 
> >  drivers/virtio/virtio_balloon.c | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> > 
> > diff --git a/drivers/virtio/virtio_balloon.c
b/drivers/virtio/virtio_balloon.c
> > index 7efc329..ee29473 100644
> > --- a/drivers/virtio/virtio_balloon.c
> > +++ b/drivers/virtio/virtio_balloon.c
> > @@ -589,7 +589,7 @@ static int virtballoon_restore(struct
virtio_device *vdev)
> >  
> >  	virtio_device_ready(vdev);
> >  
> > -	fill_balloon(vb, towards_target(vb));
> > +	wake_up(&vb->config_change);
> >  	update_balloon_size(vb);
> >  	return 0;
> >  }
> > -- 
> > MST