On Thu Sep 18, 2025 at 8:09 AM KST, Jakub Kicinski wrote:

Thank you for reviewing! and sorry for the delayed reply.
There have been quite a lot of things on my end recently.

On Wed, 17 Sep 2025 03:33:05 +0900 Yeounsu Moon wrote:

quoted

@@ -965,14 +965,11 @@ receive_packet (struct net_device *dev)
 			struct sk_buff *skb;
 
 			/* Small skbuffs for short packets */
-			if (pkt_len > copy_thresh) {
-				dma_unmap_single(&np->pdev->dev,
-						 desc_to_dma(desc),
-						 np->rx_buf_sz,
-						 DMA_FROM_DEVICE);
-				skb_put(skb = np->rx_skbuff[entry], pkt_len);
-				np->rx_skbuff[entry] = NULL;
-			} else if ((skb = netdev_alloc_skb_ip_align(dev, pkt_len))) {
+			if (pkt_len <= copy_thresh) {
+				skb = netdev_alloc_skb_ip_align(dev, pkt_len);
+				if (!skb)
+					goto fallback_to_normal_path;

The goto looks pretty awkward.

	skb = NULL;
	if (pkt_len <= copy_thresh)
		skb = netdev_alloc_skb_ip_align(dev, pkt_len);
	if (!skb) {
		// existing non-copy path
	} else {
		// existing copybreak path
	}

I totally agree with your point. However, the two cases handle `skb` and
`rx_skbuff` differently depending on the `copy_thresh` condition,
regardless of whether `skb` is NULL or not.

This patch is only intended to gracefully handle the failure case when `skb`
allocation fails.

	Yeounsu Moon