[PATCH] improve mvneta (Marvell 370) driver TCPv6 performance

[PATCH] improve mvneta (Marvell 370) driver TCPv6 performance

[Phil Hofer]

TCP performance over IPv6 was only about 60% of IPv4 performance. This
patch makes up most (but not all) of the difference.

Signed off by: Thomas Petazzoni (thomas.petazzoni@free-electrons.com)

diff --git a/drivers/net/ethernet/marvell/mvneta.c
b/drivers/net/ethernet/marvell/mvneta.c
index 5bdf782..926c948 100644
--- a/drivers/net/ethernet/marvell/mvneta.c
+++ b/drivers/net/ethernet/marvell/mvneta.c
@@ -1235,13 +1235,17 @@ static u32 mvneta_txq_desc_csum(int l3_offs,
int l3_proto,
        command =  l3_offs    << MVNETA_TX_L3_OFF_SHIFT;
        command |= ip_hdr_len << MVNETA_TX_IP_HLEN_SHIFT;

-       if (l3_proto == htons(ETH_P_IP))
-               command |= MVNETA_TXD_IP_CSUM;
-       else
+       /* IPv6 packet header checksumming
+        * is not supported, but TCPv6 header
+        * checksumming *is* supported.
+        */
+       if (l3_proto == htons(ETH_P_IPV6))
                command |= MVNETA_TX_L3_IP6;
+       else
+               command |= MVNETA_TXD_IP_CSUM;

        if (l4_proto == IPPROTO_TCP)
-               command |=  MVNETA_TX_L4_CSUM_FULL;
+               command |= MVNETA_TX_L4_CSUM_FULL;
        else if (l4_proto == IPPROTO_UDP)
                command |= MVNETA_TX_L4_UDP | MVNETA_TX_L4_CSUM_FULL;
        else
@@ -1288,13 +1292,11 @@ static void mvneta_rx_error(struct mvneta_port *pp,
 static void mvneta_rx_csum(struct mvneta_port *pp, u32 status,
                           struct sk_buff *skb)
 {
-       if ((status & MVNETA_RXD_L3_IP4) &&
-           (status & MVNETA_RXD_L4_CSUM_OK)) {
+       if (status & MVNETA_RXD_L4_CSUM_OK) {
                skb->csum = 0;
                skb->ip_summed = CHECKSUM_UNNECESSARY;
                return;
        }
-
        skb->ip_summed = CHECKSUM_NONE;
 }

@@ -3129,7 +3131,7 @@ static int mvneta_probe(struct platform_device *pdev)

        netif_napi_add(dev, &pp->napi, mvneta_poll, NAPI_POLL_WEIGHT);

-       dev->features = NETIF_F_SG | NETIF_F_IP_CSUM | NETIF_F_TSO;
+       dev->features = NETIF_F_SG| NETIF_F_IP_CSUM |
NETIF_F_IPV6_CSUM | NETIF_F_TSO;
        dev->hw_features |= dev->features;
        dev->vlan_features |= dev->features;
        dev->priv_flags |= IFF_UNICAST_FLT;

Re: [PATCH] improve mvneta (Marvell 370) driver TCPv6 performance

[Thomas Petazzoni]

Hello Phil,

On Fri, 26 Jun 2015 11:12:03 -0700, Phil Hofer wrote:
> TCP performance over IPv6 was only about 60% of IPv4 performance. This
> patch makes up most (but not all) of the difference.
> 
> Signed off by: Thomas Petazzoni (thomas.petazzoni@free-electrons.com)

What makes you think you can use my Signed-off-by for a patch that I
haven't written? It should be your Signed-off-by here, certainly not
mine.

Also, the e-mail address should be between <...>, not between (...).

Thanks!

Thomas
-- 
Thomas Petazzoni, CTO, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com

Re: [PATCH] improve mvneta (Marvell 370) driver TCPv6 performance

[Phil Hofer]

My apologies.

I have a follow-up on the way with some of Willy's suggested changes as well.

On Fri, Jun 26, 2015 at 11:52 AM, Thomas Petazzoni
<thomas.petazzoni@free-electrons.com> wrote:
> Hello Phil,
>
> On Fri, 26 Jun 2015 11:12:03 -0700, Phil Hofer wrote:
>> TCP performance over IPv6 was only about 60% of IPv4 performance. This
>> patch makes up most (but not all) of the difference.
>>
>> Signed off by: Thomas Petazzoni (thomas.petazzoni@free-electrons.com)
>
> What makes you think you can use my Signed-off-by for a patch that I
> haven't written? It should be your Signed-off-by here, certainly not
> mine.
>
> Also, the e-mail address should be between <...>, not between (...).
>
> Thanks!
>
> Thomas
> --
> Thomas Petazzoni, CTO, Free Electrons
> Embedded Linux, Kernel and Android engineering
> http://free-electrons.com