[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <MN2PR18MB23674C19A0AF33951791F0F0B0C20@MN2PR18MB2367.namprd18.prod.outlook.com>
Date: Sun, 28 Jul 2019 15:22:16 +0000
From: Stefan Chulski <stefanc@...vell.com>
To: Matteo Croce <mcroce@...hat.com>,
Antoine Tenart <antoine.tenart@...tlin.com>,
Marcin Wojtas <mw@...ihalf.com>,
"Maxime Chevallier" <maxime.chevallier@...tlin.com>
CC: netdev <netdev@...r.kernel.org>,
LKML <linux-kernel@...r.kernel.org>,
"David S . Miller" <davem@...emloft.net>
Subject: RE: [EXT] Re: [PATCH net-next] mvpp2: document HW checksum behaviour
> Hi all,
>
> probably dev->vlan_features is safe to keep the CSUM features to avoid
> unnecessary calculation in some cases, but I have another question.
> Does the PP2 hardware support checksumming within any offset? I replaced
> 'NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM' with NETIF_F_HW_CSUM and
> then stacked 5 VxLANS on top of a mvpp2 device, to have the last IP header
> at offset 264:
>
> ip link set $dev up
> ip addr add 192.168.0.$last/24 dev $dev
>
> for i in {1..5}; do
> ip link add vx$i type vxlan id $i dstport 4789 remote 192.168.$((i-
> 1)).$other
> ip link set vx$i up
> ip addr add 192.168.$i.$last/24 dev vx$i done
>
> 00:51:82:11:22:00 > 3c:fd:fe:9c:60:6c, ethertype IPv4 (0x0800), length 348:
> 192.168.0.1.33625 > 192.168.0.2.4789: VXLAN, flags [I] (0x08), vni 1
> 02:25:60:da:87:03 > 92:20:05:45:3d:d3, ethertype IPv4 (0x0800), length 298:
> 192.168.1.1.33625 > 192.168.1.2.4789: VXLAN, flags [I] (0x08), vni 2
> 12:20:97:15:8f:aa > 66:08:23:c7:72:ea, ethertype IPv4 (0x0800), length 248:
> 192.168.2.1.33625 > 192.168.2.2.4789: VXLAN, flags [I] (0x08), vni 3
> c6:1c:b9:fd:9d:28 > 22:ca:cb:6a:ea:68, ethertype IPv4 (0x0800), length 198:
> 192.168.3.1.33625 > 192.168.3.2.4789: VXLAN, flags [I] (0x08), vni 4
> 02:34:5f:45:a5:9d > d2:4e:d4:d7:42:31, ethertype IPv4 (0x0800), length 148:
> 192.168.4.1.34504 > 192.168.4.2.4789: VXLAN, flags [I] (0x08), vni 5
> a2:99:fd:9c:1b:05 > 5a:81:3b:fc:6a:07, ethertype IPv4 (0x0800), length 98:
> 192.168.5.1 > 192.168.5.2: ICMP echo request, id 1654, seq 156, length 64
>
> It seems that the HW is capable of doing it, can someone with a datasheet
> confirm this?
L3_offset in TX descriptor has 7 bits, so beginning of Layer3 should be less than 128 Bytes.
Stefan,
Regards.
Powered by blists - more mailing lists