Hi Jay Wang, Thank you so much, I will give a try with the patch and let you know the result.
Thanks, Nousi On Thu, 20 Nov 2025 at 03:15, Jay Wang <[email protected]> wrote: > Hi Nousi, > > > > I can confirm that I was able to reproduce your issue in my local test > environment(*Ampere Altra Max 96 Cores* with *the Intel E810-C 2x ports > NIC*). > > The issue also persists in the latest VPP codebase, so I spent some time > investigating it. > > From what I observed, commit[1] appears to be the root cause. This commit > introduces NEON intrinsics to add support for the IAVF Rx vector path on > Arm platforms in DPDK. > > I don’t want to bother you with the implementation details of the Rx > vector function[2]. But in short, the issue occurs because the function > writes the *pkt_len* field to the wrong location in the *rte_mbuf* when > copying from the hardware descriptor. > > I’ve created a quick workaround for VPP 25.02(see attached patch) and > tested it with 64B/1024B packets. It seems to work fine. Please give it a > try and let me know if it resolves the issue on your side. > > > > VPP 25.02 works fine on X86 systems because it uses a different Rx vector > code path. > > Interestingly, I observed a similar packet length issue when testing the > X86 counterpart of Rx vector code path[3]. > > > > Thanks, > > Jay Wang > > > > [1] https://gerrit.fd.io/r/c/vpp/+/37017 > > [2] > https://github.com/DPDK/dpdk/blob/main/drivers/net/intel/iavf/iavf_rxtx_vec_neon.c#L144 > > [3] > https://github.com/DPDK/dpdk/blob/main/drivers/net/intel/iavf/iavf_rxtx_vec_sse.c#L391 > > *From:* [email protected] <[email protected]> *On Behalf Of *nousi > via lists.fd.io > *Sent:* Tuesday, November 11, 2025 2:22 PM > *To:* [email protected] > *Subject:* [vpp-dev] [VPP 25.02 | ARM64 | E810-C | IAVF] Packets showing > length 65535 in DPDK input node (works fine on VPP 22.02) > > > > Hi everyone, > > I’m seeing a packet length issue after upgrading to *VPP 25.02* on *ARM64 > (Ampere® Altra® Q64-30 @ 3.0GHz)* with an *Intel E810-C (4x ports)* NIC > using the *IAVF driver*. > > All packets received in the *dpdk-input node* show a *length of 65535 > bytes*, which is clearly incorrect. This is visible in both *trace* and * > pcap* captures. > > Here’s what I’ve observed so far: > > - The issue only appears with *VPP 25.02* on *ARM64*. > - The *same platform (hardware + NIC)* works perfectly with *VPP 22.02* > — packet lengths are reported correctly. > - On *X86 *systems, *VPP 25.02* also works fine (no packet length > issue). > - So it seems specific to *VPP 25.02 + ARM64 + IAVF driver* > combination. > > *Setup details:* > > - *CPU:* Ampere® Altra® Q64-30 (ARM64) > - *NIC:* Intel E810-C (4 ports) > - *Driver:* IAVF (DPDK) > - *VPP version:* 25.02 > - *Symptom:* All packets show length = 65535 in dpdk-input node > > Has anyone else seen this behavior or is aware of any related changes in * > VPP 25.02* or the * IAVF/DPDK layer* for ARM64? > Would appreciate any pointers, patches, or debug hints — especially if > someone has tested E810/IAVF on ARM with newer VPP versions. > > Thanks a lot in advance! > > *Nousi.* > IMPORTANT NOTICE: The contents of this email and any attachments are > confidential and may also be privileged. If you are not the intended > recipient, please notify the sender immediately and do not disclose the > contents to any other person, use it for any purpose, or store or copy the > information in any medium. Thank you. > -- *Thanks & Regards,B.Nousilal,*
-=-=-=-=-=-=-=-=-=-=-=- Links: You receive all messages sent to this group. View/Reply Online (#26532): https://lists.fd.io/g/vpp-dev/message/26532 Mute This Topic: https://lists.fd.io/mt/116238372/21656 Group Owner: [email protected] Unsubscribe: https://lists.fd.io/g/vpp-dev/leave/14379924/21656/631435203/xyzzy [[email protected]] -=-=-=-=-=-=-=-=-=-=-=-
