virtio_net: xsk: prevent disable tx napi
authorXuan Zhuo <xuanzhuo@linux.alibaba.com>
Tue, 12 Nov 2024 01:29:25 +0000 (09:29 +0800)
committerJakub Kicinski <kuba@kernel.org>
Sat, 16 Nov 2024 02:46:55 +0000 (18:46 -0800)
Since xsk's TX queue is consumed by TX NAPI, if sq is bound to xsk, then
we must stop tx napi from being disabled.

Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Link: https://patch.msgid.link/20241112012928.102478-11-xuanzhuo@linux.alibaba.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
drivers/net/virtio_net.c

index 6cd9fdb..d5b8567 100644 (file)
@@ -5090,7 +5090,7 @@ static int virtnet_set_coalesce(struct net_device *dev,
                                struct netlink_ext_ack *extack)
 {
        struct virtnet_info *vi = netdev_priv(dev);
-       int ret, queue_number, napi_weight;
+       int ret, queue_number, napi_weight, i;
        bool update_napi = false;
 
        /* Can't change NAPI weight if the link is up */
@@ -5119,6 +5119,14 @@ static int virtnet_set_coalesce(struct net_device *dev,
                return ret;
 
        if (update_napi) {
+               /* xsk xmit depends on the tx napi. So if xsk is active,
+                * prevent modifications to tx napi.
+                */
+               for (i = queue_number; i < vi->max_queue_pairs; i++) {
+                       if (vi->sq[i].xsk_pool)
+                               return -EBUSY;
+               }
+
                for (; queue_number < vi->max_queue_pairs; queue_number++)
                        vi->sq[queue_number].napi.weight = napi_weight;
        }