Re: [net-next v7 07/10] net: bnxt: Implement software USO

[Joe Damato]

On Wed, Apr 01, 2026 at 05:35:14PM -0700, Eric Dumazet wrote:
> On Wed, Apr 1, 2026 at 4:38 PM Joe Damato <joe@xxxxxxx> wrote:
> >

[...]

> > +       /* Zero the csum fields so tso_build_hdr will propagate zeroes into
> > +        * every segment header. HW csum offload will recompute from scratch.
> > +        */
> 
> We might need a call to skb_cow_head(skb, 0) before changing ->check
> (or anything in skb->head)
> 
> Alternative would be to perform the clears after each tso_build_hdr()
> and leave skb->head untouched.

Thanks for the careful review; I appreciate your time and energy.

I'll remove the existing clears you pointed and perform the clear after each
tso_build_hdr() as you suggested with something like:

@@ -103,6 +96,7 @@ netdev_tx_t bnxt_sw_udp_gso_xmit(struct bnxt *bp,
                unsigned int offset;
                dma_addr_t dma_addr;
                struct tx_bd *txbd;
+               struct udphdr *uh;
                void *this_hdr;
                int bd_count;
                __le32 csum;
@@ -116,6 +110,17 @@ netdev_tx_t bnxt_sw_udp_gso_xmit(struct bnxt *bp,

                tso_build_hdr(skb, this_hdr, &tso, seg_payload, last);

+               /* Zero stale csum fields copied from the original skb;
+                * HW offload recomputes from scratch.
+                */
+               uh = this_hdr + skb_transport_offset(skb);
+               uh->check = 0;
+               if (!tso.ipv6) {
+                       struct iphdr *iph = this_hdr + skb_network_offset(skb);
+
+                       iph->check = 0;
+               }