NVIDIA DOCA Installation Guide for Linux

This guide details the necessary steps to set up NVIDIA DOCA in your Linux environment.

Introduction

Installation of the NVIDIA® BlueField® networking platform software requires following the following step-by-step procedure.

Supported Platforms

Supported BlueField DPUs and SuperNICs

The following NVIDIA® BlueField® DPUs and SuperNICs are supported with DOCA:

Supported ConnectX NICs

The following NVIDIA® ConnectX® NICs are supported with DOCA on the host:

• ConnectX-7 devices

• ConnectX-6 Dx devices

• ConnectX-6 Lx devices

• ConnectX-6 devices

Hardware Prerequisites

This quick start guide assumes that an NVIDIA® BlueField® networking platform has been installed in a server according to the instructions detailed in your DPU's hardware user guide.

DOCA Packages

Supported Operating System Distributions

The default operating system of the BlueField DPU (Arm) is Ubuntu 22.04.

The supported operating systems on the host machine per DOCA profile are the following:

BlueField Networking Platform Image Installation

This guide provides the minimal instructions for setting up DOCA on a standard system.

Installation Files

Uninstalling Software from Host

If an older DOCA software version is installed on your host, make sure to uninstall it before proceeding with the installation of the new version:

$ for f in $( dpkg --list | grep doca | awk '{print $2}' ); do echo $f ; apt remove --purge $f -y ; done
$ ofed_uninstall.sh --force
$ sudo apt-get autoremove

host# for f in $(rpm -qa | grep -i doca ) ; do yum -y remove $f; done
host# ofed_uninstall.sh --force
host# yum autoremove
host# yum makecache

Then perform the following steps:

1. Download NVIDIA's RPM-GPG-KEY-Mellanox-SHA256 key:

   # wget http://www.mellanox.com/downloads/ofed/RPM-GPG-KEY-Mellanox-SHA256
   --2018-01-25 13:52:30--  http://www.mellanox.com/downloads/ofed/RPM-GPG-KEY-Mellanox-SHA256
   Resolving www.mellanox.com... 72.3.194.0
   Connecting to www.mellanox.com|72.3.194.0|:80... connected.
   HTTP request sent, awaiting response... 200 OK
   Length: 1354 (1.3K) [text/plain]
   Saving to: ?RPM-GPG-KEY-Mellanox-SHA256?
    
   100%[=================================================>] 1,354       --.-K/s   in 0s
    
   2018-01-25 13:52:30 (247 MB/s) - ?RPM-GPG-KEY-Mellanox-SHA256? saved [1354/1354]
   

   
   

2. Install the key:

   # sudo rpm --import RPM-GPG-KEY-Mellanox-SHA256
   warning: rpmts_HdrFromFdno: Header V3 DSA/SHA1 Signature, key ID 6224c050: NOKEY
   Retrieving key from file:///repos/MLNX_OFED//RPM-GPG-KEY-Mellanox
   Importing GPG key 0x6224C050:
    Userid: "Mellanox Technologies (Mellanox Technologies - Signing Key v2) "
    From  : /repos/MLNX_OFED//RPM-GPG-KEY-Mellanox-SHA256
   Is this ok [y/N]:
   

   
   

3. Verify that the key was successfully imported:

   # rpm -q gpg-pubkey --qf '%{NAME}-%{VERSION}-%{RELEASE}\t%{SUMMARY}\n' | grep Mellanox
   gpg-pubkey-a9e4b643-520791ba    gpg(Mellanox Technologies )
   

   
   

Installing Prerequisites on Host for Target BlueField

Install doca-tools to manage and flash the BlueField DPU or SuperNIC.

Determining BlueField Device ID

It is important to learn your BlueField's device-id for performing some of the software installations or upgrades in this guide.

To determine the device ID of the DPUs and SuperNICs on your setup, run:

host# mst start
host# mst status -v

Example output:

MST modules:
------------
    MST PCI module is not loaded
    MST PCI configuration module loaded
PCI devices:
------------
DEVICE_TYPE             MST                           PCI       RDMA            NET                       NUMA
BlueField2(rev:1)       /dev/mst/mt41686_pciconf0.1   3b:00.1   mlx5_1          net-ens1f1                0
 
BlueField2(rev:1)       /dev/mst/mt41686_pciconf0     3b:00.0   mlx5_0          net-ens1f0                0

BlueField3(rev:1)       /dev/mst/mt41692_pciconf0.1   e2:00.1   mlx5_1          net-ens7f1np1             4

BlueField3(rev:1)       /dev/mst/mt41692_pciconf0     e2:00.0   mlx5_0          net-ens7f0np0             4

The device IDs for the BlueField-2 and BlueField-3 networking platforms in this example are /dev/mst/mt41686_pciconf0 and /dev/mst/mt41692_pciconf0 respectively.

Installing Software on Host

1. Install DOCA local repo package for host: 

   The following table provides instructions for installing the DOCA host repo on your device depending on your OS and desired profile.

   
   

   OS	Profile	Instructions
   Deb-based	doca-all	Download the DOCA host repo from section "NVIDIA DOCA Installation Guide for Linux | id (2.6.0)NVIDIADOCAInstallationGuideforLinux InstallationFiles" for the host. Unpack the deb repo. Run: host# dpkg -i <repo_file> Perform apt update. Run:  host# apt-get update If the kernel version on your host is not supported (not shown under "NVIDIA DOCA Installation Guide for Linux | id (2.6.0)NVIDIADOCAInstallationGuideforLinux SupportedOperatingSystemDistributions"), refer to section "NVIDIA DOCA Installation Guide for Linux | id (2.6.0)NVIDIADOCAInstallationGuideforLinux DOCAExtraPackage". Ensure that the kernel headers installed match the version of the currently running kernel.  If the build directory exists in under /lib/modules/$(uname -r)/build, then the kernel headers are installed. Run apt install for DOCA SDK, DOCA runtime, DOCA tools:  host# sudo apt install -y doca-all
   	doca-networking	Download the DOCA host repo from section "NVIDIA DOCA Installation Guide for Linux | id (2.6.0)NVIDIADOCAInstallationGuideforLinux InstallationFiles" for the host. Unpack the deb repo. Run:  host# dpkg -i <repo_file> Perform apt update. Run:  host# apt-get update If the kernel version on your host is not supported (not shown under "NVIDIA DOCA Installation Guide for Linux | id (2.6.0)NVIDIADOCAInstallationGuideforLinux SupportedOperatingSystemDistributions"), refer to section "NVIDIA DOCA Installation Guide for Linux | id (2.6.0)NVIDIADOCAInstallationGuideforLinux DOCAExtraPackage". Ensure that the kernel headers installed match the version of the currently running kernel. If the build directory exists in under /lib/modules/$(uname -r)/build, then the kernel headers are installed. Run apt install for DOCA SDK, DOCA runtime, DOCA tools:  host# sudo apt install -y doca-networking
   	doca-ofed	Download the DOCA host repo from section "NVIDIA DOCA Installation Guide for Linux | id (2.6.0)NVIDIADOCAInstallationGuideforLinux InstallationFiles" for the host. Unpack the deb repo. Run:  host# sudo dpkg -i doca-host-repo-ubuntu<version>_amd64.deb Perform apt update. Run:   host# sudo apt-get update If the kernel version on your host is not supported (not shown under "NVIDIA DOCA Installation Guide for Linux | id (2.6.0)NVIDIADOCAInstallationGuideforLinux SupportedOperatingSystemDistributions"), refer to section "NVIDIA DOCA Installation Guide for Linux | id (2.6.0)NVIDIADOCAInstallationGuideforLinux DOCAExtraPackage". Ensure that the kernel headers installed match the version of the currently running kernel. If the build directory exists in under /lib/modules/$(uname -r)/build, then the kernel headers are installed. Install doca-ofed. Run:  host# sudo apt install -y doca-ofed mlnx-fw-updater
   RPM-based	doca-all	RHEL users need a valid subscription to install packages. See "NVIDIA DOCA Installation Guide for Linux | id (2.6.0)NVIDIADOCAInstallationGuideforLinux HowToOpenRedHatAccount". Install the following software dependencies. Run: For RH 8.x, run:  host# yum -y install https://dl.fedoraproject.org/pub/epel/epel-release-latest-8.noarch.rpm For Alinux 3.x (Alibaba Cloud Linux), run:  host# yum -y install https://dl.fedoraproject.org/pub/epel/epel-release-latest-8.noarch.rpm For Rocky 8.x, run:  host# sudo dnf makecache host# sudo dnf install epel-release host# sudo dnf install -y yum-utils host# sudo dnf config-manager --enable PowerTools host# sudo dnf clean dbcache For CentOS 8.x, run:  host# sudo dnf install epel-release host# sudo dnf config-manager --set-enabled PowerTools Download the DOCA host repo from section "NVIDIA DOCA Installation Guide for Linux | id (2.6.0)NVIDIADOCAInstallationGuideforLinux InstallationFiles" for the host. Unpack the rpm repo. Run:  host# rpm -Uvh <repo_file>.rpm Perform yum update. Run:  host# sudo yum makecache If the kernel version on your host is not supported (not shown under "NVIDIA DOCA Installation Guide for Linux | id (2.6.0)NVIDIADOCAInstallationGuideforLinux SupportedOperatingSystemDistributions"), refer to section "NVIDIA DOCA Installation Guide for Linux | id (2.6.0)NVIDIADOCAInstallationGuideforLinux DOCAExtraPackage". Run yum install for DOCA SDK, DOCA runtime, DOCA tools:  host# sudo yum install -y doca-all
   	doca-networking	Install the following software dependencies. Run: For RH 8.x, run:  host# yum -y install https://dl.fedoraproject.org/pub/epel/epel-release-latest-8.noarch.rpm For Alinux 3.x (Alibaba Cloud Linux), run:  host# yum -y install https://dl.fedoraproject.org/pub/epel/epel-release-latest-8.noarch.rpm For Rocky 8.x, run: host# sudo dnf makecache host# sudo dnf install epel-release host# sudo dnf install -y yum-utils host# sudo dnf config-manager --enable PowerTools host# sudo dnf clean dbcache For CentOS 8.x, run:  host# sudo dnf install epel-release host# sudo dnf config-manager --set-enabled PowerTools For Oracle Linux 8.7, run:  host# yum config-manager --set-enabled ol8_codeready_builder Download the DOCA host repo from section "NVIDIA DOCA Installation Guide for Linux | id (2.6.0)NVIDIADOCAInstallationGuideforLinux InstallationFiles" for the host. Unpack the rpm repo. Run:  host# rpm -Uvh <repo_file>.rpm Perform yum update. Run:  host# sudo yum makecache If the kernel version on your host is not supported (not shown under "NVIDIA DOCA Installation Guide for Linux | id (2.6.0)NVIDIADOCAInstallationGuideforLinux SupportedOperatingSystemDistributions"), refer to section "NVIDIA DOCA Installation Guide for Linux | id (2.6.0)NVIDIADOCAInstallationGuideforLinux DOCAExtraPackage". Run yum install for DOCA SDK, DOCA runtime, DOCA tools:  host# sudo yum install -y doca-networking
   	doca-ofed	Download the DOCA host repo from section "NVIDIA DOCA Installation Guide for Linux | id (2.6.0)NVIDIADOCAInstallationGuideforLinux InstallationFiles" for the host. Unpack the RPM repo. Run:  host# sudo rpm -Uvh doca-host-repo-rhel<version>.x86_64.rpm Install doca-extra. Run:   host# sudo yum/dnf install -y doca-extra If the kernel version on your host is not supported (not shown under "NVIDIA DOCA Installation Guide for Linux | id (2.6.0)NVIDIADOCAInstallationGuideforLinux SupportedOperatingSystemDistributions"), refer to section "NVIDIA DOCA Installation Guide for Linux | id (2.6.0)NVIDIADOCAInstallationGuideforLinux DOCAExtraPackage". Install doca-ofed. Run:   host# sudo yum install -y doca-ofed mlnx-fw-updater

   
   

2. Load the drivers:

   host# sudo /etc/init.d/openibd restart
   

   
   

3. Initialize MST. Run: 

   host# sudo mst start
   

   
   

4. Reset the nvconfig parameters to their default values: 

   host# sudo mlxconfig -d <device-id> -y reset
   
   Reset configuration for device <device-id>? (y/n) [n] : y
   Applying... Done!
   -I- Please reboot machine to load new configurations.
   

   
   

   To learn the BlueField's device ID, refer to section "Determining Bluefield Device ID".

   
   

5. Skip this step if your BlueField DPU or SuperNIC is Ethernet only. Please refer to NVIDIA DOCA Installation Guide for Linux | id (2.6.0)NVIDIADOCAInstallationGuideforLinux SupportedPlatforms to learn your Bluefield type.

   If you have a VPI networking platform, the default link type of the ports will be configured to IB. To verify your link type, run: 

   host# sudo mst start
   host# sudo mlxconfig -d <device-id> -e q | grep -i link_type
   Configurations:                              Default         Current         Next Boot
   *        LINK_TYPE_P1                        IB(1)           ETH(2)          IB(1)
   *        LINK_TYPE_P2                        IB(1)           ETH(2)          IB(1)
   

   
   

    If your BlueField is Ethernet capable only, then the sudo mlxconfig -d <device> command will not provide an output.

   If the current link type is set to IB, run the following command to change it to Ethernet: 

   host# sudo mlxconfig -d <device-id> s LINK_TYPE_P1=2 LINK_TYPE_P2=2
   

   
   

6. Verify that RShim is active. 

   host# sudo systemctl status rshim
   

   This command is expected to display active (running). If RShim service does not launch automatically, run: 

   host# sudo systemctl enable rshim
   host# sudo systemctl start rshim
   

   
   

7. Assign a dynamic IP to tmfifo_net0 interface (RShim host interface). 

   Skip this step if you are installing the DOCA image on multiple DPUs.

   
   

   host# ifconfig tmfifo_net0 192.168.100.1 netmask 255.255.255.252 up
   

   
   

DOCA Extra Package

If the kernel version on on your host is not supported (not shown under "NVIDIA DOCA Installation Guide for Linux | id (2.6.0)NVIDIADOCAInstallationGuideforLinux SupportedOperatingSystemDistributions"), two options are available:

• Switch to a compatible kernel.

• Install doca-extra package:

  1. Run:

     host# sudo apt install -y doca-tools doca-extra
     

     
     

  2. Execute the doca-kernel-support script which downloads the appropriate missing packages from your kernel to support DOCA: 

     host# sudo /opt/mellanox/doca/tools/doca-kernel-support
     

     
     

     doca-kernel-support does not support customized or unofficial kernels.

     
     

Installing Software on DPU

Users have two options for installing DOCA on the DPU:

• Upgrading the full DOCA image on the DPU (recommended) – this option overwrites the entire boot partition

• Upgrading DOCA local repo package on the DPU – this option upgrades DOCA components without overwriting the boot partition. Use this option to preserve configurations or files on the DPU itself.

Installing Full DOCA Image on DPU via Host

Option 1 – No Pre-defined Password

BFB installation is executed as follows: 

host# sudo bfb-install --rshim <rshimN> --bfb <image_path.bfb>

Where rshimN is rshim0 if you only have one Bluefield. You may run the following command to verify:

host# ls -la /dev/ | grep rshim

Option 2 – Set Pre-defined Password

Ubuntu users can provide a unique password that will be applied at the end of the BlueField software image installation. This password needs to be defined in a bf.cfg configuration file.

To set the password for the "ubuntu" user:

1. Create password hash. Run: 

   host# openssl passwd -1
   Password:
   Verifying - Password:
   $1$3B0RIrfX$TlHry93NFUJzg3Nya00rE1
   

   
   

2. Add the password hash in quotes to the bf.cfg file: 

   host# sudo vim bf.cfg
   ubuntu_PASSWORD='$1$3B0RIrfX$TlHry93NFUJzg3Nya00rE1'
   

   When running the installation command, use the --config flag to provide the file containing the password: 

   host# sudo bfb-install --rshim <rshimN> --bfb <image_path.bfb> --config bf.cfg
   

   
   

   If --config is not used, then upon first login to the BlueField device, users will be asked to update their password.

   The following is an example of Ubuntu installation assuming the pv Linux tool has been installed (to view the installation progress). 

   host# sudo bfb-install --rshim rshim0 --bfb DOCA_<version>-aarch64.bfb --config bf.cfg
   Pushing bfb
   1.08GiB 0:00:57 [19.5MiB/s] [      <=>    ]
   Collecting BlueField booting status. Press Ctrl+C to stop…
   INFO[BL2]: start
   INFO[BL2]: DDR POST passed
   INFO[BL2]: UEFI loaded
   INFO[BL31]: start
   INFO[BL31]: runtime
   INFO[UEFI]: eMMC init
   INFO[UEFI]: eMMC probed
   INFO[UEFI]: PCIe enum start
   INFO[UEFI]: PCIe enum end
   INFO[MISC]: Ubuntu installation started
   INFO[MISC]: Installation finished
   INFO[MISC]: Rebooting...
   

   
   

Installing Full DOCA Image on Multiple Networking Platforms

On a host with multiple networking platforms, the BFB image can be installed on all of them using the multi-bfb-install script.

host# ./multi-bfb-install --bfb <bfb-file> --password <password>

This script detects the number of RShim devices and configures them statically.

• For Ubuntu – the script creates a configuration file /etc/netplan/20-tmfifo.yaml

• For CentOS/RH 8.0 and 8.2 – the script installs the bridge-utils package to use the brctl command, creates the tm-br bridge, and connects all RShim interfaces to it

After the installation is complete, the configuration of the bridge and each RShim interface can be observed using ifconfig. The expected result is to see the IP on the tm-br bridge configured to 192.168.100.1 with subnet 255.255.255.0.

ssh ubuntu@192.168.100.2

The script burns a new MAC address to each BlueField and configures a new IP, 192.168.100.x, as described earlier.

Installing DOCA Local Repo Package on BlueField

1. Download the DOCA SDK, DOCA Runtime, and DOCA Tools package from section NVIDIA DOCA Installation Guide for Linux | id (2.6.0)NVIDIADOCAInstallationGuideforLinux InstallationFiles.

2. Copy deb repo package into BlueField. Run: 

   host# sudo scp -r doca-repo-aarch64-ubuntu2204-local_<version>_arm64.deb ubuntu@192.168.100.2:/tmp/
   

   
   

3. Unpack the deb repo. Run: 

   dpu# sudo dpkg -i doca-dpu-repo-ubuntu2204-local_<version>_arm64.deb
   

   
   

4. Run apt update. 

   dpu# sudo apt-get update
   

   
   

5. Run apt install for DOCA Runtime, DOCA Tools, and DOCA SDK: 

   dpu# sudo apt install doca-runtime
   dpu# sudo apt install doca-tools
   dpu# sudo apt install doca-sdk
   

   
   

Upgrading Firmware

To upgrade firmware:

1. SSH to your BlueField device via 192.168.100.2 (preconfigured).

   If multiple BlueField's are installed, the tmfifo IP interface does not have to be 192.168.100.2. The last octate changes and depends on the RShim number.

   The default credentials for Ubuntu are as follows: 

   Username	Password
   ubuntu	ubuntu or a unique password that you set in bf.cfg

   For example: 

   host# ssh ubuntu@192.168.100.2
   Password: <configured-password>
   

   
   

2. Upgrade firmware in BlueField networking platform. Run: 

   dpu# sudo /opt/mellanox/mlnx-fw-updater/mlnx_fw_updater.pl --force-fw-update
   

   Example output: 

   Device #1:
   ----------
   
     Device Type:      BlueField-2
     [...]
     Versions:         Current        Available
        FW             <Old_FW>       <New_FW>
   

   
   

3. For the firmware upgrade to take effect:

   1. Run the following command on BlueField and host: 

      dpu# sudo mst start
      

      
      

   2. Otherwise, trigger reset by running the following: 

      dpu# sudo mlxfwreset -d /dev/mst/mt41686_pciconf0 --sync 1 -y reset
      

      If the mlxfwreset command fails, perform host power cycle.  

      The entire BlueField networking platform will experience reset.

      
      

Post-installation Procedure

1. Restart the driver. Run: 

   host# sudo /etc/init.d/openibd restart
   Unloading HCA driver:                                      [  OK  ]
   Loading HCA driver and Access Layer:                       [  OK  ]
   

   
   

2. Configure the physical function (PF) interfaces. 

   host# sudo ifconfig <interface-1> <network-1/mask> up
   host# sudo ifconfig <interface-2> <network-2/mask> up
   

   For example: 

   host# sudo ifconfig p2p1 192.168.200.32/24 up
   host# sudo ifconfig p2p2 192.168.201.32/24 up
   

   Pings between the source and destination should now be operational.

Upgrading BlueField Using Standard Linux Tools

This dpu-upgrade procedure enables upgrading DOCA components using standard Linux tools (e.g., apt update and yum update). This process utilizes native package manager repositories to upgrade DPUs without the need for a full installation, and has the following benefits:

• Only updates components that include modifications

  • Configurable – user can select specific components (e.g., UEFI-ATF, NIC-FW)

• Includes upgrade of:

  • DOCA drivers and libraries 

  • DOCA reference applications

  • BSP (UEFI/ATF) upgrade while maintaining the configuration

  • NIC firmware upgrade while maintaining the configuration

• Does not:

  • Impact user binaries

  • Upgrade non-Ubuntu OS kernels

  • Upgrade DPU BMC firmware

• After completion of DPU upgrade:

  • If NIC firmware was not updated, perform DPU Arm reset (software reset / reboot DPU)

  • If NIC firmware was updated, perform firmware reset (mlxfwreset) or perform a graceful shutdown and power cycle

<dpu> $ apt remove --purge mlxbf-bootimages* -y

<dpu> $ apt update
<dpu> $ apt install gnupg2

<dpu> $ export DOCA_REPO="https://linux.mellanox.com/public/repo/doca/2.6.0/ubuntu22.04/dpu-arm64"

<dpu> $ curl $DOCA_REPO/GPG-KEY-Mellanox.pub | gpg --dearmor > /etc/apt/trusted.gpg.d/GPG-KEY-Mellanox.pub

<dpu> $ echo "deb [signed-by=/etc/apt/trusted.gpg.d/GPG-KEY-Mellanox.pub] $DOCA_REPO ./" > /etc/apt/sources.list.d/doca.list

<dpu> $ apt update

<dpu> $ apt install mlxbf-bootimages-signed

<dpu> $ bfrec

<dpu> $ apt install mlnx-fw-updater-signed.aarch64

<dpu> $ export RUN_FW_UPDATER=no

<dpu> $ apt upgrade

<dpu> $ mlxfwreset -d /dev/mst/mt*_pciconf0 -y -l 3 --sync 1 r

<dpu> $ yum -y remove mlxbf-bootimages*
<dpu> $ yum makecache

<dpu> $ export DOCA_REPO="https://linux.mellanox.com/public/repo/doca/2.6.0/rhel8.6/dpu-arm64/"

echo "[doca] 
name=DOCA Online Repo 
baseurl=$DOCA_REPO 
enabled=1 
gpgcheck=0 
priority=10 
cost=10" > /etc/yum.repos.d/doca.repo

<dpu> $ yum makecache

<dpu> $ yum install mlxbf-bootimages-signed.aarch64 mlxbf-bfscripts

<dpu> $ bfrec

<dpu> $ yum install mlnx-fw-updater-signed.aarch64

<dpu> $ export RUN_FW_UPDATER=no

sudo /opt/mellanox/mlnx-fw-updater/mlnx_fw_updater.pl --force-fw-update

<dpu> $ yum upgrade --nobest

<dpu> $ mlxfwreset -d /dev/mst/mt*_pciconf0 -y -l 3 --sync 1 r

Building Your Own BFB Installation Image

Users wishing to build their own customized BlueField OS image can use the BFB build environment. Please refer to the bfb-build project in this GitHub webpage for more information. 

Setting Up Build Environment for Developers

For full instructions about setting up a development environment, refer to the NVIDIA DOCA Developer Guide.

Additional SDKs for DOCA

Installing CUDA on NVIDIA Converged Accelerator

NVIDIA® CUDA® is a parallel computing platform and programming model developed by NVIDIA for general computing GPUs.

This section details the necessary steps to set up CUDA on your environment. This section assumes that a BFB image has already been installed on your environment.
To install CUDA on your converged accelerator:

1. Download and install the latest NVIDIA Data Center GPU driver.

2. Download and install CUDA 

   The CUDA version tested to work with DOCA SDK is 11.8.0.

   
   

Configuring Operation Mode

There are two modes that the NVIDIA Converged Accelerator may operate in:

• Standard mode (default) – the BlueField networking platform and the GPU operate separately

• BlueField-X mode – the GPU is exposed to BlueField and is no longer visible on the host

To verify which mode the system is operating in, run: 

host# sudo mst start
host# sudo mlxconfig -d <device-id> q PCI_DOWNSTREAM_PORT_OWNER[4]

• Standard mode output: 

  Device #1:
  […]
  Configurations:                              Next Boot
           PCI_DOWNSTREAM_PORT_OWNER[4]        DEVICE_DEFAULT(0)
  
  
  

  
  

• BlueField-X mode output:

  Device #1:
  […]
  Configurations:                              Next Boot
           PCI_DOWNSTREAM_PORT_OWNER[4]        EMBEDDED_CPU(15)
  

  
  

To configure BlueField-X mode, run: 

host# mlxconfig -d <device-id> s PCI_DOWNSTREAM_PORT_OWNER[4]=0xF

To configure standard mode, run: 

host# mlxconfig -d <device-id> s PCI_DOWNSTREAM_PORT_OWNER[4]=0x0

Power cycle is required for configuration to take effect. For power cycle the host run:

host# ipmitool power cycle

Downloading and Installing CUDA Toolkit and Driver

This section details the necessary steps to set up CUDA on your environment. It assumes that a BFB image has already been installed on your environment.

1. Install CUDA by visiting the CUDA Toolkit Downloads webpage.

   Select the Linux distribution and version relevant for your environment.

   
   

   This section shows the native compilation option either on x86 or aarch64 hosts.

   
   

2. Test that the driver installation completed successfully. Run: 

   dpu# nvidia-smi
   
   Tue Apr  5 13:37:59 2022       
   +-----------------------------------------------------------------------------+
   | NVIDIA-SMI 510.47.03    Driver Version: 510.47.03    CUDA Version: 11.8     |
   |-------------------------------+----------------------+----------------------+
   | GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
   | Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
   |                               |                      |               MIG M. |
   |===============================+======================+======================|
   |   0  NVIDIA BF A10       Off  | 00000000:06:00.0 Off |                    0 |
   |  0%   43C    P0    N/A / 225W |      0MiB / 23028MiB |      0%      Default |
   |                               |                      |                  N/A |
   +-------------------------------+----------------------+----------------------+
                                                                                  
   +-----------------------------------------------------------------------------+
   | Processes:                                                                  |
   |  GPU   GI   CI        PID   Type   Process name                  GPU Memory |
   |        ID   ID                                                   Usage      |
   |=============================================================================|
   |  No running processes found                                                 |
   +-----------------------------------------------------------------------------+
   

   
   

3. Verify that the installation completed successfully.

   1. Download CUDA samples repo. Run: 

      dpu# git clone https://github.com/NVIDIA/cuda-samples.git
      

      
      

   2. Build and run vectorAdd CUDA sample. Run: 

      dpu# cd cuda-samples/Samples/0_Introduction/vectorAdd
      dpu# make
      dpu# ./vectorAdd
      

      
      

   If the vectorAdd sample works as expected, it should output "Test Passed".

   
   

   If it seems that the GPU is slow or stuck, stop execution and run: 

   dpu# sudo setpci -v -d ::0302 800.L=201 # CPL_VC0 = 32
   

   
   

   
   

GPUDirect RDMA

For information on GPUDirect RMDA and more, refer to DOCA GPUNetIO documentation.

Installing Rivermax on BlueField Networking Platform 

NVIDIA Rivermax offers a unique IP-based solution for any media and data streaming use case.

This section provides the steps to install Rivermax assuming that a BFB image has already been installed on your environment.

Downloading Rivermax Driver

1. Navigate to the NVIDIA Rivermax SDK product page.

2. Register to be able to download the driver package using the JOIN button at the top of the page.

3. Download the appropriate driver package according to your BFB under the "Linux" subsection. For example, for Ubuntu 22.04 BFB, download rivermax_ubuntu2204_<version>.tar.gz.

Installing Rivermax Driver

1. Copy the .tgz file to BlueField:
   

   host# sudo scp -r rivermax_ubuntu2204_<version>.tar.gz ubuntu@192.168.100.2:/tmp/
   

   
   

2. Extract the Rivermax file:

   dpu# sudo tar xzf rivermax_ubuntu2204_<version>.tar.gz
   

   
   

3. Install the Rivermax driver package:

   dpu# cd <rivermax-version>/Ubuntu.22.04/deb-dist/aarch64/
   dpu# sudo dpkg -i rivermax_<version>.deb
   

   
   

Installing Rivermax Libraries from DOCA

Rivermax libraries are compatibles with DOCA components and can be found inside the doca-dpu-repo.

1. Unpack the doca-dpu-repo:
   

   dpu# sudo dpkg -i doca-dpu-repo-ubuntu2204-local_<version>_arm64.deb
   

   
   

2. Run apt update:

   dpu# sudo apt-get update
   

   
   

3. Install the Rivermax libraries:

   dpu# sudo apt install doca-rmax-libs
   dpu# sudo apt install libdoca-rmax-libs-dev
   

   
   

For additional details and guidelines, please visit the NVIDIA Rivermax SDK product page.

Last updated: March 06, 2024