On 3/1/26 12:39 AM, Michael S. Tsirkin wrote:

On Sat, Feb 28, 2026 at 05:41:22AM -0800, Vishwanath Seshagiri wrote:

quoted

@@ -2446,7 +2349,7 @@ static struct sk_buff *virtnet_skb_append_frag(struct sk_buff *head_skb,
  
  	offset = buf - page_address(page);
  	if (skb_can_coalesce(curr_skb, num_skb_frags, page, offset)) {
-		put_page(page);
+		page_pool_put_page(rq->page_pool, page, -1, true);
  		skb_coalesce_rx_frag(curr_skb, num_skb_frags - 1,
  				     len, truesize);
  	} else {

This part seems wrong to me: can this not be called from
xsk_append_merge_buffer?  Because that one seems to still allocate with
napi_alloc_frag.

Ack. will fix this in v9.