Skip to content

Xilinx-CNS/onload

Repository files navigation

OpenOnload®️

Onload®️ is a high performance user-level network stack, which accelerates TCP and UDP network I/O for applications using the BSD sockets on Linux.

Features

  • Low Application-to-application latency.
  • Binary compatible with existing applications.
  • Open Source (GPLv2.0 and BSD-2-Clause).

OpenOnload comprises a user-level shared library that intercepts network- related system calls and implements the protocol stack, and supporting kernel modules. It is compatible with the full system call API, including those aspects that are usually problematic for user-level networking, such as fork(), exec(), passing sockets through Unix domain sockets, and advancing the protocol when the application is not scheduled.

Installation and Quick Start Guide

OpenOnload is distributed as source code. Instructions for building, packaging and installing may be found in DEVELOPING.md

If there are any interfaces to be accelerated by Onload which are not AMD Solarflare interfaces, execute the following to register those interfaces to be accelerated using AF_XDP:

echo ens2f0 > /sys/module/sfc_resource/afxdp/register

The application to be Onloaded should be launched by prefixing the command line with onload.

Support

The publicly-hosted repository is a community-supported project. When raising issues on this repository it is expected that users will be running from the head of the git tree to pick up recent changes, not using official versions of Onload that were typically released before recent breaking kernel changes appeared as that is likely to lead to many duplicate issues being raised. Incompatibilities introduced by recent kernel versions are likely to be fixed rapidly here in this repository.

Supported releases of OpenOnload are available from https://www.xilinx.com/support/download/nic-software-and-drivers.html#open. Please raise issues on supported releases of Onload with support-nic@amd.com.

Compatible Linux kernels and distributions

This source tree is compatible with the following Linux kernels and distributions:

  • Debian 12
  • Ubuntu LTS 22.04, 24.04
  • Red Hat Enterprise Linux 8.6 - 8.10
  • Red Hat Enterprise Linux 9.0 - 9.4
  • kernel.org Linux kernels 5.11 - 6.9

Onload with AMD Solarflare NICs

Onload provides optimum networking acceleration and additional features using the native ef_vi hardware interface provided by AMD Solarflare network adapters compared to using Linux's AF_XDP mechanism. In this mode kernel and driver support for AF_XDP is not required.

A version of the 'sfc' net driver for AMD Solarflare network adapters is included.

Compatible AMD Solarflare network adapters

The following adapters are able to support OpenOnload without AF_XDP:

  • SFN8522, SFN8542, SFN8042
  • X2522, X2522-25G, X2541
  • X3522

Onload with AF_XDP

OpenOnload can accelerate applications on non-Solarflare network adapters with support for AF_XDP.

Compatible network adapters, drivers and operating systems

To support zero-copy, Onload needs AF_XDP network adapter drivers to implement the necessary AF_XDP primitives. Typically the latest drivers from the network adapter vendors will support these primitives.

The AF_XDP support is a community-supported work in progress that is not currently at release quality.

If a netdriver does not support AF_XDP in native mode, Onload will try to use generic XDP mode when registering an interface. To make it work, one has to set up userland helper before registering the interface:

$ make -C ./src/tools/bpf_link_helper/
$ echo $(realpath ./src/tools/bpf_link_helper/bpf-link-helper) | sudo tee /sys/module/sfc_resource/parameters/bpf_link_helper

Building without AMD Solarflare NICs, for AF_XDP only

OpenOnload can be built without SFC driver:

  • make: use HAVE_SFC=0 variable;
  • onload_build & onload_install: use --no-sfc parameter;
  • onload_tool reload: use --onload-only parameter.

Also, it can be built without EFCT and AUX support:

  • make: use HAVE_EFCT=0 variable;
  • onload_build: use --no-efct parameter.

Contributions

Please see CONTRIBUTING.md

Copyright

This file: (c) Copyright 2020-2024 Advanced Micro Devices, Inc.