Re: [PATCH] net: mv643xx_eth: Add GRO support

[Willy Tarreau <w@1wt.eu>]

Hi,

On Thu, Apr 11, 2013 at 02:40:23PM +0200, Sebastian Hesselbarth wrote:
> This patch adds GRO support to mv643xx_eth by making it invoke
> napi_gro_receive instead of netif_receive_skb.
> 
> Signed-off-by: Soeren Moch <smoch@web.de>
> Signed-off-by: Sebastian Hesselbarth <sebastian.hesselbarth@gmail.com>
> ---
> Cc: "David S. Miller" <davem@davemloft.net>
> Cc: Lennert Buytenhek <buytenh@wantstofly.org>
> Cc: Andrew Lunn <andrew@lunn.ch>
> Cc: Jason Cooper <jason@lakedaemon.net>
> Cc: Florian Fainelli <florian@openwrt.org>
> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
> Cc: Paul Mackerras <paulus@samba.org>
> Cc: Dale Farnsworth <dale@farnsworth.org>
> Cc: netdev@vger.kernel.org
> Cc: linux-arm-kernel@lists.infradead.org
> Cc: linuxppc-dev@lists.ozlabs.org
> Cc: linux-kernel@vger.kernel.org
> ---
>  drivers/net/ethernet/marvell/mv643xx_eth.c |    2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/drivers/net/ethernet/marvell/mv643xx_eth.c b/drivers/net/ethernet/marvell/mv643xx_eth.c
> index 305038f..c850d04 100644
> --- a/drivers/net/ethernet/marvell/mv643xx_eth.c
> +++ b/drivers/net/ethernet/marvell/mv643xx_eth.c
> @@ -604,7 +604,7 @@ static int rxq_process(struct rx_queue *rxq, int budget)
>  			lro_receive_skb(&rxq->lro_mgr, skb, (void *)cmd_sts);
>  			lro_flush_needed = 1;
>  		} else
> -			netif_receive_skb(skb);
> +			napi_gro_receive(&mp->napi, skb);
>  
>  		continue;

I remember having experimented with this on 3.6 a few months ago with this
driver and finally switching back to something like this instead which
showed better performance on my tests :

               if (skb->ip_summed == CHECKSUM_UNNECESSARY)
                       napi_gro_receive(napi, skb);
               else
                       netif_receive_skb(skb);

Unfortunately I don't have more details as my commit message was rather
short due to this resulting from experimentation. Did you verify that
you did not lose any performance in various workloads ? I was playing
with bridges at this time, it's possible that I got better performance
on bridging with netif_receive_skb() than with napi_gro_receive().

Regards,
Willy