I didn't want to remove the default parsing of tunnel as VxLan because I thought it might be used, Instead I moved it to the end, which makes it detect all supported tunnel through udp_dst_port, And only if no tunnel was matched it would default to VxLan. That was the reason geneve weren't detected and parsed as vxlan instead, which is the bug I was trying to solve.
We can take help/input from i40 maintainers for it.
Hi Beilei Xing, For setting packet_type as Tunnel, what criteria is used by i40 driver. Is it only udp_dst port or any other parameters also.
-----Original Message----- From: Singh, Aman Deep <aman.deep.singh@intel.com> Sent: Thursday, January 20, 2022 12:47 PM To: Matan Azrad <matan@nvidia.com>; Ferruh Yigit <ferruh.yigit@intel.com>; Raja Zidane <rzidane@nvidia.com>; dev@dpdk.org Cc: stable@dpdk.org Subject: Re: [PATCH] app/testpmd: fix GENEVE parsing in csum forward mode External email: Use caution opening links or attachments On 1/18/2022 6:49 PM, Matan Azrad wrote: app/test-pmd/csumonly.c | 16 ++++++++++------ 1 file changed, 10 insertions(+), 6 deletions(-) diff --git a/app/test-pmd/csumonly.c b/app/test-pmd/csumonly.c index 2aeea243b6..fe810fecdd 100644 --- a/app/test-pmd/csumonly.c +++ b/app/test-pmd/csumonly.c @@ -254,7 +254,10 @@ parse_gtp(struct rte_udp_hdr *udp_hdr, info->l2_len += RTE_ETHER_GTP_HLEN; } -/* Parse a vxlan header */ +/* + * Parse a vxlan header. + * If a tunnel is detected in 'pkt_type' it will be parsed by default as vxlan. + */ static void parse_vxlan(struct rte_udp_hdr *udp_hdr, struct testpmd_offload_info *info, @@ -912,17 +915,18 @@ pkt_burst_checksum_forward(struct fwd_stream *fs) RTE_MBUF_F_TX_TUNNEL_VXLAN_GPE; goto tunnel_update; } - parse_vxlan(udp_hdr, &info, - m->packet_type); + parse_geneve(udp_hdr, &info); if (info.is_tunnel) { tx_ol_flags |= - RTE_MBUF_F_TX_TUNNEL_VXLAN; + + RTE_MBUF_F_TX_TUNNEL_GENEVE; goto tunnel_update; } - parse_geneve(udp_hdr, &info); + /* Always keep last. */ + parse_vxlan(udp_hdr, &info, + m->packet_type); if (info.is_tunnel) { tx_ol_flags |= - RTE_MBUF_F_TX_TUNNEL_GENEVE; + + RTE_MBUF_F_TX_TUNNEL_VXLAN; goto tunnel_update; } } else if (info.l4_proto == IPPROTO_GRE) {-----Original Message----- From: Ferruh Yigit <ferruh.yigit@intel.com> Sent: Tuesday, January 18, 2022 3:03 PM To: Matan Azrad <matan@nvidia.com>; Raja Zidane <rzidane@nvidia.com>; dev@dpdk.org Cc: stable@dpdk.org Subject: Re: [PATCH] app/testpmd: fix GENEVE parsing in csum forward mode External email: Use caution opening links or attachments On 1/18/2022 12:55 PM, Matan Azrad wrote:-----Original Message----- From: Ferruh Yigit <ferruh.yigit@intel.com> Sent: Tuesday, January 18, 2022 2:28 PM To: Matan Azrad <matan@nvidia.com>; Raja Zidane <rzidane@nvidia.com>; dev@dpdk.org Cc: stable@dpdk.org Subject: Re: [PATCH] app/testpmd: fix GENEVE parsing in csum forward mode External email: Use caution opening links or attachments On 1/18/2022 11:27 AM, Matan Azrad wrote:-----Original Message----- From: Ferruh Yigit <ferruh.yigit@intel.com> Sent: Tuesday, January 18, 2022 11:52 AM To: Raja Zidane <rzidane@nvidia.com>; dev@dpdk.org Cc: Matan Azrad <matan@nvidia.com>; stable@dpdk.org Subject: Re: [PATCH] app/testpmd: fix GENEVE parsing in csum forward mode External email: Use caution opening links or attachments On 12/5/2021 3:44 AM, Raja Zidane wrote:The csum FWD mode parses any received packet to set mbuf offloads for the transmitting burst, mainly in the checksum/TSO areas. In the case of a tunnel header, the csum FWD tries to detect known tunnels by the standard definition using the header'sdata and fallback to check the packet type in the mbuf to see if the Rx port driver already sign the packet as a tunnel. In the fallback case, the csum assumes the tunnel is VXLAN and parses the tunnel as VXLAN.As far as I can see there is a VXLAN port check in 'parse_vxlan()', why it is not helping?The problem is not the vxlan check but the tunnel type in mbuf that caused thepacket to be detected as vxlan(default) before checking GENEVE tunnel case. Check is as following: if (udp_hdr->dst_port != _htons(RTE_VXLAN_DEFAULT_PORT) && RTE_ETH_IS_TUNNEL_PKT(pkt_type) == 0) return; Do you what is the intention for the "RTE_ETH_IS_TUNNEL_PKT(pkt_type) ==0"check? Why vxlan parsing doesn't stop when it is not default port?Maybe some drivers set the tunnel type for vxlan packets coming after non-standard vxlan port. But checking the tunnel flag to say that it is vxlan is too broad, isn't it? And this is the problem you are having. Can there be any way to detect and check non-standard vxlan port?Maybe yes, but it is probably more complex solusion. See this patch: https://patches.dpdk.org/project/dpdk/patch/1423819371-24222-13-git-se nd-email-olivier.matz@6wind.com/ comments there: 1. check udp destination port, 4789 is the default vxlan port (rfc7348). 2. currently, this flag is set by i40e only if the packet is vxlan And maybe another driver assumes more ports here. Collecting all the non-standard ports can start from i40 maintainers😊I think, here we should check for supported tunnel types only. The i40 driver setting a Tunnel flag, does not means that it VxLan type only. As in your case those were GENEVE packets getting treated as VxLan. Making parse_vxlan() udp_dst_port specific can avoid this issue.When the GENEVE tunnel was added to the known tunnels in csum, its parsing trial was wrongly located after the pkt type detection, causing the csum to parse the GENEVE header as VXLAN when the Rx port set the tunnel packet type. Locate the GENEVE parsing trial before the packet type detection. Fixes: ea0e711b8ae0 ("app/testpmd: add GENEVE parsing") Cc: stable@dpdk.org Signed-off-by: Raja Zidane <rzidane@nvidia.com> --- Acked-by: Matan Azrad <matan@nvidia.com>Ack should be before '---' to be part of the commit log, otherwise it is dropped when applied as comment.