vhost: fix enabling vring per queue
authorVictor Kaplansky <victork@redhat.com>
Tue, 24 Nov 2015 07:25:35 +0000 (15:25 +0800)
committerThomas Monjalon <thomas.monjalon@6wind.com>
Tue, 24 Nov 2015 20:34:11 +0000 (21:34 +0100)
The VHOST_USER_SET_VRING_ENABLE request was sent for each queue-pair.
However, it's changed to be sent per queue in the queue-pair by QEMU
commit dc3db6ad ("vhost-user: start/stop all rings"). The change
is reasonable, as we send all other request per queue, instead of
queue-pair.

Hence we should do proper changes to adapt to the QEMU change here.
Otherwise, a segfault will be triggered when last TX queue was enabled.

Signed-off-by: Victor Kaplansky <victork@redhat.com>
Signed-off-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
lib/librte_vhost/vhost_user/virtio-net-user.c

index 99da029..2934d1c 100644 (file)
@@ -323,7 +323,6 @@ user_set_vring_enable(struct vhost_device_ctx ctx,
                      struct vhost_vring_state *state)
 {
        struct virtio_net *dev = get_device(ctx);
-       uint16_t base_idx = state->index;
        int enable = (int)state->num;
 
        RTE_LOG(INFO, VHOST_CONFIG,
@@ -331,12 +330,10 @@ user_set_vring_enable(struct vhost_device_ctx ctx,
                enable, state->index);
 
        if (notify_ops->vring_state_changed) {
-               notify_ops->vring_state_changed(dev, base_idx / VIRTIO_QNUM,
-                                               enable);
+               notify_ops->vring_state_changed(dev, state->index, enable);
        }
 
-       dev->virtqueue[base_idx + VIRTIO_RXQ]->enabled = enable;
-       dev->virtqueue[base_idx + VIRTIO_TXQ]->enabled = enable;
+       dev->virtqueue[state->index]->enabled = enable;
 
        return 0;
 }