From mboxrd@z Thu Jan 1 00:00:00 1970 From: Ben Greear Subject: Re: [PATCH] veth: remove hardware checksum feature Date: Wed, 07 Aug 2013 17:23:44 -0700 Message-ID: <5202E510.9060309@candelatech.com> References: <51F15E50.8080208@guap.ru> <5202E153.4060202@candelatech.com> <1375920752.4004.71.camel@edumazet-glaptop> Mime-Version: 1.0 Content-Type: text/plain; charset=UTF-8; format=flowed Content-Transfer-Encoding: QUOTED-PRINTABLE Cc: "Vitaly E. Lavrov" , netdev@vger.kernel.org, davem@davemloft.net To: Eric Dumazet Return-path: Received: from mail.candelatech.com ([208.74.158.172]:41488 "EHLO ns3.lanforge.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S933288Ab3HHAXw (ORCPT ); Wed, 7 Aug 2013 20:23:52 -0400 In-Reply-To: <1375920752.4004.71.camel@edumazet-glaptop> Sender: netdev-owner@vger.kernel.org List-ID: On 08/07/2013 05:12 PM, Eric Dumazet wrote: > On Wed, 2013-08-07 at 17:07 -0700, Ben Greear wrote: >> On 07/25/2013 10:20 AM, Vitaly E. Lavrov wrote: >>> The network device VETH can't support the feature NETIF_F_HW_CSUM. >>> All locally generated packets have invalid checksum. >>> Wrong commit http://git.kernel.org/cgit/linux/kernel/git/torvalds/l= inux.git/commit/?id=3D8093315a91340bca52549044975d8c7f673b28a1 ( kernel= 3.9.0 ) >>> >>> Workaround "ethtool -K vethX tx off" >>> >>> Possible patch: >>> >>> diff --git a/drivers/net/veth.c b/drivers/net/veth.c >>> index 177f911..3db97da 100644 >>> --- a/drivers/net/veth.c >>> +++ b/drivers/net/veth.c >>> @@ -254,7 +254,7 @@ static const struct net_device_ops veth_netdev_= ops =3D { >>> }; >>> >>> #define VETH_FEATURES (NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_A= LL_TSO | \ >>> - NETIF_F_HW_CSUM | NETIF_F_RXCSUM | NETIF_F_H= IGHDMA | \ >>> + NETIF_F_RXCSUM | NETIF_F_HIGHDMA | \ >>> NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_= CTAG_RX | \ >>> NETIF_F_HW_VLAN_STAG_TX | NETIF_F_HW_VLAN_= STAG_RX ) >> >> I see the same problem. >> >> My test case is a bit complicated, but the gist is that I have a VET= H >> pair, one with IP (veth1), one connected to a bridge-like-thing (vet= h2). >> >> The UDP frames sent on veth1 appear on veth2, and when I sniff veth2= , >> the packets show broken checksum. The work-around mentioned in Vita= ly's >> email above fixes the problem for me (I did not try the patch yet). >> >> Eric: You responded originally that you needed more info. If >> my explanation above is not sufficient, please let me know what >> you need... >> > > tcpdump is known to display wrong checksums, its not a reason to disa= ble > tx checksums on our interfaces and kill performance. I am receiving the packet into user space by reading veth2 using a packet socket, and then writing that packet out to eth6 (e100e). As far as I can tell, it reads from veth2 with bad checksum and then goes onto the wire with bad checksum. > -K Don't attempt to verify IP, TCP, or UDP checksums. T= his > is useful for interfaces that perform some or all of those checksum > calculation in hardware; other=E2=80=90 > wise, all outgoing TCP checksums will be flagged as ba= d. Is it ever valid to *read* a packet with bad checksum though? I though= t the bogus bad hw-checksum issue was only on the tx-side as far as sniff= ing goes? Thanks, Ben --=20 Ben Greear Candela Technologies Inc http://www.candelatech.com