net/virtio: fix crash when configured twice

[ upstream commit 52bd03e969e9b38a7357287aece2488c7f92158f ]

When first attempt to configure a device with RX interrupt enabled fails
for some reason (e.g. because "Multiple intr vector not supported"),
second attempt to configure the device with RX interrupt disabled and
feature set unchanged will succeed but will leave virtio queues not
allocated. Accessing the queues will cause a segfault.

First attempt:
  - virtio_dev_configure()
    - virtio_init_device() is called to reinit the device because
      "dev->data->dev_conf.intr_conf.rxq" is "1"
      - virtio_configure_intr() fails and returns an error
      - virtio_free_queues() frees previously allocated virtio queues
    - virtio_init_device() fails and returns an error
  - virtio_dev_configure() fails and returns an error

Second attempt:
  - virtio_dev_configure()
    - This time virtio_init_device() is not called, virtio queues
      are not allocated

With this fix, reinit the device during configuration if virtio queues
are not allocated.

Fixes: 2b38151 ("net/virtio: fix queue memory leak on error")

Signed-off-by: Alexander Chernavin <achernavin@netgate.com>
Reviewed-by: Chenbo Xia <chenbo.xia@intel.com>

drivers/net/virtio/virtio_ethdev.c

@@ -2129,6 +2129,13 @@ virtio_dev_configure(struct rte_eth_dev *dev)
 			return ret;
 	}
 
+	/* if queues are not allocated, reinit the device */
+	if (hw->vqs == NULL) {
+		ret = virtio_init_device(dev, hw->req_guest_features);
+		if (ret < 0)
+			return ret;
+	}
+
 	if ((rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
 			    DEV_RX_OFFLOAD_TCP_CKSUM)) &&
 		!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {