skills.rest
NVIDIA
NVIDIAOFFICIAL
Explore all skills by this author →
WebsiteGitHub

doca-bf3-deployment

Guides BlueField-3 platform bring-up via RShim BFB push, TMFIFO recovery, and install verification.

3.2k|370|Updated Feb 25, 2026
Search Tags:#firmware#doca#dpu#bluefield-3#bfb-install#platform bring-up#rshim
One-click install
npx skills add https://github.com/NVIDIA/skills --skill doca-bf3-deployment
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: doca-bf3-deployment
Source: https://github.com/NVIDIA/skills/tree/main/skills/doca-bf3-deployment
Command: npx skills add https://github.com/NVIDIA/skills --skill doca-bf3-deployment

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

Bringing up a BlueField-3 DPU from a powered card to a healthy, workload-ready platform is error-prone: bfb-install can exit 0 on a partial install, TMFIFO pings can lie via local loopback, host PFs can vanish after a push, and host-side versus Arm-side DOCA installs are easily confused. This Skill gives an AI agent the documented RShim/BFB bring-up procedure, a six-state failure classifier, and strict safety routing for every mutating burn.

Core Features & Use Cases

  • BFB push over RShim: Walks the operator through disambiguating /dev/rshim<N> on multi-DPU hosts, authoring bf.cfg from the live BSP-manual schema, and pushing the BlueField bundle with bfb-install without inventing flags or filenames.
  • Six-state BlueField classifier: Diagnoses a BF3 that did not come back after a push across installer-still-running, uefi-only, linux-up-tmfifo-down, tmfifo-up-ssh-down, arm-ok-host-pfs-unbound, and host-bf-version-mismatch states, with ordered recovery.
  • Install verification and version matching: Confirms which BFB landed via cat /etc/mlnx-release and bfver, reads per-device firmware with flint, and re-closes the four-way version match before handing off to deployment skills.
  • Use Case: An operator runs bfb-install, sees exit code 0, but the DPU never reaches the ready marker; the agent parses the console log for the NIC firmware update failure line, classifies the state, and walks the documented recovery instead of declaring success.

Quick Start

Ask your agent to push a BFB image to your BlueField-3 over RShim and verify the install, and it will load this skill to walk the documented bring-up sequence.

Frequently Asked Questions about doca-bf3-deployment

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I push a BFB image to a BlueField-3 over RShim?▼

Use bfb-install from the host, targeting the correct /dev/rshim<N> device identified by cross-matching the DEV_NAME field in each rshim misc file against lspci output. Author bf.cfg from the live BSP-manual schema, and never trust exit code 0 alone; parse the console for the documented Linux up and DPU is ready markers.

Why does bfb-install exit 0 but my BlueField-3 never comes back?▼

bfb-install can return 0 on a partial install; the canonical signature is an Ubuntu installation completed line followed by an INFO[MISC] NIC firmware update failed message. Classify the DPU with the six-state classifier, starting by checking whether the installer is still running or the boot stalled at UEFI.

Why does ping 192.168.100.2 work but SSH to the BlueField fails?▼

The ping may be hitting the host's local loopback instead of the DPU. Run ip route get 192.168.100.2 before pinging; dev lo or a local address binding is the failure signature, while dev tmfifo_net0 or dev tm-br confirms healthy egress to the BF3.

Is DOCA installed on the host or on the BlueField Arm side?▼

Both sides carry DOCA: the host runs DOCA-Host packages installed via apt or dnf, while the Arm side runs the DOCA userland baked into the BFB image. Read the host version with pkg-config --modversion doca-common and the Arm side with cat /etc/mlnx-release and bfver.

Does this skill cover BlueField-4 bring-up?▼

No. BlueField-4 uses the BMC-Redfish provisioning path, which is handled by the sibling doca-bf4-deployment skill. This skill covers only the classic RShim/BFB path used by BlueField-3.

Who owns safety for mlxconfig mode changes and firmware burns?▼

All mutating operations, including BFB reflashes, mlxconfig set mode flips, and firmware burns, are governed by the doca-hardware-safety meta-policy loaded alongside this skill. This skill provides only the BF3-specific sequencing, never the burn discipline itself.

Related Skills

Explore More Software Engineering Skills→
air-gappedair-gapped
5

nvidia-datacenter-bringup

Automate NVIDIA datacenter GPU bring-up on Ubuntu 24.04 with health validation.

Official
Advanced
fred-drakefred-drake
27

infrastructure

Deploy NixOS configurations with Colmena across Proxmox LXC containers and bare-metal hosts.

Community
Advanced
NVIDIANVIDIA
14

k8s-launch-kit-config

Explain, create, and edit k8s-launch-kit configuration files for Kubernetes network deployment.

Official
Advanced
soulmachinesoulmachine
3

asus-esc8000-firmware-upgrade

Automate BMC and BIOS firmware updates on ASUS ESC8000 servers via Redfish and HPM APIs.

Community
Advanced