On Wed, Jun 24, 2015 at 10:31:09PM -0300, Julio Faracco wrote:

quoted

2015-06-24 3:10 GMT-03:00 Michael S. Tsirkin [off-list ref]:

quoted

On Tue, Jun 23, 2015 at 10:44:29PM -0300, Julio Faracco wrote:

quoted

virtio_net paravirtualized driver does not have a tx_timeout() function
to
guarantee that the driver will recover properly after receiving a
timeout
during a transmission of a packet. This patch add this feature and
throw a
timeout exception after 5 HZ. Considering some tests, this is the best
time to use here.

Signed-off-by: Julio Faracco <redacted>
Cc: Jason Wang <jasowang@redhat.com>

Looks like a bunch of locks and flushes are missing in this patch.  IMHO
that's just too painful with current hardware.  IMO the right thing to
do here is to add ability to reset specific queues to hardware.

I agree, Michael. This model is the default one resetting the device
due to transmission timeout.
To have a better performance, only some queues must be reset.

It's not a question of performance. You would need to write
a bunch of code anyway. Why not do it in the hypervisor
so guest can simply write into a register and reset
a ring?


BTW now that I think about it, this requires Jason's
patches that introduce the tx interrupt, otherwise
packet will timeout simply because no packets are sent.

I am trying to understand how TX interrupt patches will help
here? This function will be called when driver fails to send 
packets. Even before TX interrupt patches, packets are flowing.

Is my understanding wrong some where?

quoted

quoted

quoted

---
 drivers/net/virtio_net.c |   69
 +++++++++++++++++++++++++++++++++++++++++++++-
 1 file changed, 68 insertions(+), 1 deletion(-)

diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
index 63c7810..75ac45c 100644
--- a/drivers/net/virtio_net.c
+++ b/drivers/net/virtio_net.c

@@ -135,6 +135,9 @@ struct virtnet_info {
      /* Work struct for config space updates */
      struct work_struct config_work;

+     /* Work struct for resetting the virtio-net driver. */
+     struct work_struct reset_task;
+
      /* Does the affinity hint is set for virtqueues? */
      bool affinity_hint_set;

@@ -1394,6 +1397,18 @@ static int virtnet_change_mtu(struct net_device

*dev, int new_mtu)
      return 0;
 }

+static void virtnet_tx_timeout(struct net_device *dev)
+{
+     struct virtnet_info *vi = netdev_priv(dev);
+
+     dev_warn(&dev->dev, "TX Timeout exception with latency: %ld\n",
+              jiffies - dev_trans_start(dev));
+
+     schedule_work(&vi->reset_task);

What if after this triggers user does something
to the device (e.g. attempts to remove it)?
Or if a packet is transmitted or used?

At some point, this work must be canceled.
Yes, you are right. Specially, when the driver is being removed.

quoted

quoted

+}
+
+static void virtnet_reset_task(struct work_struct *work);
+
 static const struct net_device_ops virtnet_netdev = {
      .ndo_open            = virtnet_open,
      .ndo_stop            = virtnet_close,

@@ -1405,6 +1420,7 @@ static const struct net_device_ops virtnet_netdev

= {
      .ndo_get_stats64     = virtnet_stats,
      .ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid,
      .ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid,
+     .ndo_tx_timeout      = virtnet_tx_timeout,
 #ifdef CONFIG_NET_POLL_CONTROLLER
      .ndo_poll_controller = virtnet_netpoll,
 #endif

@@ -1750,6 +1766,7 @@ static int virtnet_probe(struct virtio_device

*vdev)
      dev->netdev_ops = &virtnet_netdev;
      dev->features = NETIF_F_HIGHDMA;

+     dev->watchdog_timeo = 5 * HZ;
      dev->ethtool_ops = &virtnet_ethtool_ops;
      SET_NETDEV_DEV(dev, &vdev->dev);

@@ -1811,6 +1828,7 @@ static int virtnet_probe(struct virtio_device

*vdev)
      }

      INIT_WORK(&vi->config_work, virtnet_config_changed_work);
+     INIT_WORK(&vi->reset_task, virtnet_reset_task);

      /* If we can receive ANY GSO packets, we must allocate large
      ones. */
      if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) ||

@@ -1891,7 +1909,7 @@ static int virtnet_probe(struct virtio_device

*vdev)
              netif_carrier_on(dev);
      }

-     pr_debug("virtnet: registered device %s with %d RX and TX
vq's\n",
+     pr_debug("virtio_net: registered device %s with %d RX and TX
vq's\n",
               dev->name, max_queue_pairs);

      return 0;

@@ -2001,6 +2019,55 @@ static int virtnet_restore(struct virtio_device

*vdev)
 }
 #endif

+static void virtnet_reset_task(struct work_struct *work)
+{
+     struct virtnet_info *vi =
+             container_of(work, struct virtnet_info, reset_task);
+     struct net_device *dev = vi->dev;
+     struct virtio_device *vdev = vi->vdev;
+     int err, i;
+
+     flush_work(&vi->config_work);
+
+     netif_device_detach(vi->dev);
+     cancel_delayed_work_sync(&vi->refill);
+
+     if (netif_running(vi->dev)) {
+             for (i = 0; i < vi->max_queue_pairs; i++) {
+                     napi_disable(&vi->rq[i].napi);
+                     napi_hash_del(&vi->rq[i].napi);
+                     netif_napi_del(&vi->rq[i].napi);
+             }
+     }
+
+     remove_vq_common(vi);
+
+     dev->stats.tx_errors++;
+
+     err = init_vqs(vi);
+     if (err) {
+             dev_warn(&dev->dev, "virtio_net: virtqueue initialization
failed.\n");
+             return;
+     }
+
+     virtio_device_ready(vdev);
+
+     if (netif_running(vi->dev)) {
+             for (i = 0; i < vi->curr_queue_pairs; i++)
+                     if (!try_fill_recv(vi, &vi->rq[i], GFP_KERNEL))
+                             schedule_delayed_work(&vi->refill, 0);
+
+             for (i = 0; i < vi->max_queue_pairs; i++)
+                     virtnet_napi_enable(&vi->rq[i]);
+     }
+
+     netif_device_attach(vi->dev);
+
+     rtnl_lock();
+     virtnet_set_queues(vi, vi->curr_queue_pairs);
+     rtnl_unlock();

Won't this lose a bunch of state, like mac addresses,
multicast, rx mode, etc etc?

I will rebase this patch with the properly changes and locks.
After, I will resend it.

Thanks for your opinion, Michael.

quoted

quoted

+}
+
 static struct virtio_device_id id_table[] = {
      { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
      { 0 },
--
1.7.10.4

--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html