On Wed, Dec 07, 2011 at 04:05:29PM +0530, Amit Shah wrote:

On (Wed) 07 Dec 2011 [12:28:24], Michael S. Tsirkin wrote:

quoted

On Wed, Dec 07, 2011 at 01:18:47AM +0530, Amit Shah wrote:

quoted

diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
index 697a0fc..1378f3c 100644
--- a/drivers/net/virtio_net.c
+++ b/drivers/net/virtio_net.c

@@ -1151,6 +1151,38 @@ static void __devexit virtnet_remove(struct virtio_device *vdev)
 	free_netdev(vi->dev);
 }
 
+#ifdef CONFIG_PM
+static int virtnet_freeze(struct virtio_device *vdev)
+{
+	struct virtnet_info *vi = vdev->priv;
+
+	netif_device_detach(vi->dev);
+	if (netif_running(vi->dev))
+		napi_disable(&vi->napi);
+

Could refill_work still be running at this point?

Yes, it could.  So moving the cancel_delayed_work_sync() before
disabling napi would work fine?

No, because napi poll can schedule that.
Further, refill can reschedule itself.

It also looks like we have a bug in virtio net cleanup now:
cancel_delayed_work_sync is called after unregister, so
it will be calling napi API on an invalid device.
And, if it schedules itself it will run after device is gone.

I think we need some locking to fix this.

 Anything else that might similar treatment?

		Amit

-- 
MST