Re: [PATCH] net: Add support for handling queueing in hardware

[Eric Dumazet <eric.dumazet@gmail.com>]

On Wed, 2011-07-13 at 08:52 -0500, Andy Fleming wrote:
> The QDisc code does a bunch of locking which is unnecessary if
> you have hardware which handles all of the queueing. Add
> support for this, and skip over all of the queueing code if
> the feature is enabled on a given device, which breaks QDisc
> support on dpaa_eth, and also coopts the FCOE feature bit.
> 
> Signed-off-by: Andy Fleming <afleming@freescale.com>
> Signed-off-by: Ben Collins <ben.c@servergy.com>
> Cc: netdev@vger.kernel.org
> ---
>  include/linux/netdev_features.h | 2 ++
>  net/core/dev.c                  | 6 ++++++
>  2 files changed, 8 insertions(+)
> 
> diff --git a/include/linux/netdev_features.h b/include/linux/netdev_features.h
> index 3dd3934..ffb4587 100644
> --- a/include/linux/netdev_features.h
> +++ b/include/linux/netdev_features.h
> @@ -56,6 +56,7 @@ enum {
>  	NETIF_F_LOOPBACK_BIT,		/* Enable loopback */
>  	NETIF_F_RXFCS_BIT,		/* Append FCS to skb pkt data */
>  	NETIF_F_RXALL_BIT,		/* Receive errored frames too */
> +	NETIF_F_HW_QDISC_BIT,		/* Supports hardware Qdisc */
>  
>  	/*
>  	 * Add your fresh new feature above and remember to update
> @@ -80,6 +81,7 @@ enum {
>  #define NETIF_F_GSO_ROBUST	__NETIF_F(GSO_ROBUST)
>  #define NETIF_F_HIGHDMA		__NETIF_F(HIGHDMA)
>  #define NETIF_F_HW_CSUM		__NETIF_F(HW_CSUM)
> +#define NETIF_F_HW_QDISC	__NETIF_F(HW_QDISC)
>  #define NETIF_F_HW_VLAN_FILTER	__NETIF_F(HW_VLAN_FILTER)
>  #define NETIF_F_HW_VLAN_RX	__NETIF_F(HW_VLAN_RX)
>  #define NETIF_F_HW_VLAN_TX	__NETIF_F(HW_VLAN_TX)
> diff --git a/net/core/dev.c b/net/core/dev.c
> index dffbef7..6818b18 100644
> --- a/net/core/dev.c
> +++ b/net/core/dev.c
> @@ -2743,6 +2743,12 @@ int dev_queue_xmit(struct sk_buff *skb)
>  
>  	skb_update_prio(skb);
>  
> +	if (dev->features & NETIF_F_HW_QDISC) {
> +		txq = netdev_pick_tx(dev, skb);
> +		rc = dev_hard_start_xmit(skb, dev, txq);
> +		goto out;
> +	}
> +
>  	txq = netdev_pick_tx(dev, skb);
>  	q = rcu_dereference_bh(txq->qdisc);
>  

Nobody forces you to use a qdisc.

And if your device really is lockless, it can use NETIF_F_LLTX feature.