[PATCH net-next v2] net: stmmac: improve TSO/GSO queue selection

[Lorenzo Bianconi]

> GSO (TSO/USO) frames are currently always sent through queue 0 since
> the driver assumed there was no way to determine which queues are TSO
> capable. This unconditionally pins all GSO traffic to queue 0, bypassing
> the per-flow queue distribution and any XPS setup.
> 
> Instead, pick the queue through netdev_pick_tx() and fall back to queue 0
> only when the selected queue cannot provide hardware checksumming (COE
> disabled) or, for the GSO types offloaded by hardware, cannot run TSO
> (TBS enabled), since TSO and TBS cannot coexist on the same channel (see
> stmmac_tso_channel_permitted()).
> 
> While at it, base the TSO capability check on priv->gso_enabled_types
> rather than a hardcoded GSO type mask, so it matches the TSO/USO
> capability actually used by the transmit path (e.g. UDP segmentation is
> only offloaded on GMAC4).
> 
> Signed-off-by: Lorenzo Bianconi <lorenzo.bianconi at oss.qualcomm.com>

please drop this version, I will post v3 to fix some pending issues.

Regards,
Lorenzo

> ---
> Changes in v2:
> - Take into account when checksum offload is not supported.
> - Link to v1: https://lore.kernel.org/r/20260807-stmmac_select_queue-tso-fix-v1-1-4c4787bad02b@oss.qualcomm.com
> ---
>  drivers/net/ethernet/stmicro/stmmac/stmmac_main.c | 18 ++++++++++--------
>  1 file changed, 10 insertions(+), 8 deletions(-)
> 
> diff --git a/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c b/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
> index b2b7d0242dd3..7a4bd0292966 100644
> --- a/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
> +++ b/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
> @@ -6445,19 +6445,21 @@ static int stmmac_setup_tc(struct net_device *ndev, enum tc_setup_type type,
>  static u16 stmmac_select_queue(struct net_device *dev, struct sk_buff *skb,
>  			       struct net_device *sb_dev)
>  {
> -	int gso = skb_shinfo(skb)->gso_type;
> +	u32 queue = netdev_pick_tx(dev, skb, NULL) % dev->real_num_tx_queues;
> +	struct stmmac_priv *priv = netdev_priv(dev);
>  
> -	if (gso & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6 | SKB_GSO_UDP_L4)) {
> -		/*
> -		 * There is no way to determine the number of TSO/USO
> -		 * capable Queues. Let's use always the Queue 0
> -		 * because if TSO/USO is supported then at least this
> -		 * one will be capable.
> +	if (skb_is_gso(skb) &&
> +	    (priv->plat->tx_queues_cfg[queue].coe_unsupported ||
> +	     ((skb_shinfo(skb)->gso_type & priv->gso_enabled_types) &&
> +	      !stmmac_tso_channel_permitted(priv, queue)))) {
> +		/* GSO frames need hardware checksumming. Moreover, hardware
> +		 * TSO/USO can't run on COE-less or TBS queues. In this case
> +		 * fall back to the queue reserved for TSO (queue 0).
>  		 */
>  		return 0;
>  	}
>  
> -	return netdev_pick_tx(dev, skb, NULL) % dev->real_num_tx_queues;
> +	return queue;
>  }
>  
>  static int stmmac_set_mac_address(struct net_device *ndev, void *addr)
> 
> ---
> base-commit: 8ac4255c1e0c83d2e1559a18b8918673116fc8d6
> change-id: 20260807-stmmac_select_queue-tso-fix-06cc586d4022
> 
> Best regards,
> -- 
> Lorenzo Bianconi <lorenzo.bianconi at oss.qualcomm.com>
> 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 228 bytes
Desc: not available
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20260812/916b2232/attachment.sig>