Re: [PATCH 1/1] balloon: stop inflate balloon after oom notification

["Michael S. Tsirkin" <mst@redhat.com>]

On Fri, Sep 09, 2016 at 04:54:44PM +0300, Denis V. Lunev wrote:
> From: Konstantin Neumoin <kneumoin@virtuozzo.com>
> 
> At this moment oom notification in balloon does not work as expected.
> After virtballoon_oom_notify there is an infinitive loop:
>  - virtballoon_oom_notify was called and balloon was deflated
>  - balloon get notification that config was changed, compare target and
>    actual and try to reach target again.
> 
> This patch adds global variable fail_counter which indicates that oom has
> been happened. We check that fail_counter was changed between calls
> update_balloon_size_func. In this case we should not try to inflate balloon
> even if actual != target.

Doesn't look right to me. What if it triggered an hour ago?
We really need mm core to expose an oom_in_progress flag that we can test.

More implementation comments below.

> 
> Signed-off-by: Konstantin Neumoin <kneumoin@virtuozzo.com>
> Signed-off-by: Denis V. Lunev <den@openvz.org>
> CC: Michael S. Tsirkin <mst@redhat.com>
> ---
>  drivers/virtio/virtio_balloon.c | 15 +++++++++++++++
>  1 file changed, 15 insertions(+)
> 
> diff --git a/drivers/virtio/virtio_balloon.c b/drivers/virtio/virtio_balloon.c
> index 4e7003d..253bf05 100644
> --- a/drivers/virtio/virtio_balloon.c
> +++ b/drivers/virtio/virtio_balloon.c
> @@ -50,6 +50,8 @@ MODULE_PARM_DESC(oom_pages, "pages to free on OOM");
>  static struct vfsmount *balloon_mnt;
>  #endif
>  
> +static unsigned long fail_count;
> +
>  struct virtio_balloon {
>  	struct virtio_device *vdev;
>  	struct virtqueue *inflate_vq, *deflate_vq, *stats_vq;
> @@ -361,6 +363,8 @@ static int virtballoon_oom_notify(struct notifier_block *self,
>  	unsigned long *freed;
>  	unsigned num_freed_pages;
>  
> +	fail_count++;
> +
>  	vb = container_of(self, struct virtio_balloon, nb);
>  	if (!virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_DEFLATE_ON_OOM))
>  		return NOTIFY_OK;

OOM should not interfere with ballooning unless
VIRTIO_BALLOON_F_DEFLATE_ON_OOM has been negotiated.


> @@ -386,11 +390,22 @@ static void update_balloon_size_func(struct work_struct *work)
>  {
>  	struct virtio_balloon *vb;
>  	s64 diff;
> +	static unsigned long fc;
> +
> +	if (fc == 0)
> +		fc = fail_count;

Why?

>  
>  	vb = container_of(work, struct virtio_balloon,
>  			  update_balloon_size_work);
>  	diff = towards_target(vb);
>  
> +	if (fc != fail_count) {
> +		fc = fail_count;

Unlikely to work correctly if there are multiple balloon devices.


> +		/* Don't inflate balloon after oom notification */
> +		if (diff > 0)
> +			return;
> +	}
> +
>  	if (diff > 0)
>  		diff -= fill_balloon(vb, diff);
>  	else if (diff < 0)

I'd rather make it per-device.
Absence of memory ordering primitives of any kind also
looks suspicious.

> -- 
> 2.7.4