Re: [PATCH net] net: lan966x: Free skb when PTP TX timestamp request fails in xmit

From: Daniel Machon

Date: Fri Nov 20 2026 - 04:58:12 EST


> When all LAN966X_MAX_PTP_ID timestamp slots are in flight,
> lan966x_ptp_txtstamp_request() returns -EBUSY without queuing or freeing
> skb, and lan966x_port_xmit() returns that negative errno directly from
> .ndo_start_xmit. Because dev_xmit_complete() treats negative return codes
> (< NET_XMIT_MASK) as packets already consumed by the driver, skb is never
> freed or requeued, permanently leaking the sk_buff and its socket
> sk_wmem_alloc accounting.
>
> Free skb with dev_kfree_skb_any(), increment tx_dropped, and return
> NETDEV_TX_OK when lan966x_ptp_txtstamp_request() fails.
>
> Found via an x86_64 LLVM=1 (Clang 21.1.8) allyesconfig audit with
> -Wimplicit-int-enum-cast enabled; fix and changelog drafted with an AI
> coding assistant. Build-tested under LLVM=1 W=1 (-Werror) and
> -Wimplicit-int-enum-cast; not tested on Microchip LAN966x hardware.
>
> Fixes: 77eecf25bd9d ("net: lan966x: Update extraction/injection for timestamping")
> Signed-off-by: Ian Rogers <irogers@xxxxxxxxxx>
> ---
> .../net/ethernet/microchip/lan966x/lan966x_main.c | 14 +++++++++-----
> 1 file changed, 9 insertions(+), 5 deletions(-)
>
> diff --git a/drivers/net/ethernet/microchip/lan966x/lan966x_main.c b/drivers/net/ethernet/microchip/lan966x/lan966x_main.c
> index b8696fc7e9b9..9c722d23e0bb 100644
> --- a/drivers/net/ethernet/microchip/lan966x/lan966x_main.c
> +++ b/drivers/net/ethernet/microchip/lan966x/lan966x_main.c
> @@ -392,6 +392,7 @@ static netdev_tx_t lan966x_port_xmit(struct sk_buff *skb,
> struct lan966x_port *port = netdev_priv(dev);
> struct lan966x *lan966x = port->lan966x;
> __be32 ifh[IFH_LEN];
> + netdev_tx_t ret;
> int err;
>
> memset(ifh, 0x0, sizeof(__be32) * IFH_LEN);
> @@ -404,8 +405,11 @@ static netdev_tx_t lan966x_port_xmit(struct sk_buff *skb,
>
> if (port->lan966x->ptp && skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) {
> err = lan966x_ptp_txtstamp_request(port, skb);
> - if (err)
> - return err;
> + if (err) {
> + dev_kfree_skb_any(skb);
> + dev->stats.tx_dropped++;
> + return NETDEV_TX_OK;
> + }
>
> lan966x_ifh_set_rew_op(ifh, LAN966X_SKB_CB(skb)->rew_op);
> lan966x_ifh_set_oam_type(ifh, LAN966X_SKB_CB(skb)->pdu_type);
> @@ -414,12 +418,12 @@ static netdev_tx_t lan966x_port_xmit(struct sk_buff *skb,
>
> spin_lock(&lan966x->tx_lock);
> if (port->lan966x->fdma)
> - err = lan966x->ops->fdma_xmit(skb, ifh, dev);
> + ret = lan966x->ops->fdma_xmit(skb, ifh, dev);
> else
> - err = lan966x_port_ifh_xmit(skb, ifh, dev);
> + ret = lan966x_port_ifh_xmit(skb, ifh, dev);
> spin_unlock(&lan966x->tx_lock);
>
> - return err;
> + return ret;
> }
>
> static int lan966x_port_change_mtu(struct net_device *dev, int new_mtu)
> --
> 2.56.1.708.g77570a6a1d-goog
>

Hi Ian,

Thanks for the fix. The issue is real, and also on my backlog of work to
upstream.

Just a heads up, I expect the AI review to flag issues in this area. The lan966x
PTP TX path has a number of known problems that I am are going to address in a
PTP series soon. I'll go through the findings once the review is in and follow
up on this one.

/Daniel