Configure and debug DOCA Ethernet RX/TX packet queues on BlueField DPUs and ConnectX NICs.
npx skills add https://github.com/NVIDIA/skills --skill doca-eth
Please help me install this Agent Skill. Skill: doca-eth Source: https://github.com/NVIDIA/skills/tree/main/skills/doca-eth Command: npx skills add https://github.com/NVIDIA/skills --skill doca-eth
Building line-rate packet I/O with the DOCA Ethernet library requires choosing the right RX queue type, sizing queues against per-device capability limits, and diagnosing silent failures like an RX queue that receives no packets. This Skill guides agents through those decisions using the real doca_eth_rxq_* / doca_eth_txq_* API surface instead of hallucinated symbols.
doca_eth_rxq and doca_eth_txq contexts on ports, representors, and SFs.enum doca_eth_rxq_type values (_REGULAR, _CYCLIC, _MANAGED_MEMPOOL, _SHARED_MEMPOOL) with mandatory doca_eth_*_cap_* queries before assuming device support.DOCA_ERROR_* taxonomy plus a precondition matrix (device access, port link state, traffic steering) for the silent-RX-queue failure mode.doca_eth_rxq starts cleanly but no packets arrive; the Skill walks the steering and promiscuous-mode preconditions and routes flow-rule programming to the doca-flow skill rather than misdiagnosing an Ethernet bug.Ask your agent to help bring up a doca_eth_rxq and doca_eth_txq on a BlueField representor and choose the right RX queue type for your traffic shape.
High-intent search queries and answers about installing and using this skill.
Verify device access, port link state, and a steering plan first, then create doca_eth_rxq and doca_eth_txq as independent DOCA Core contexts. Set the RX type and burst size within capability limits before calling doca_ctx_start on each queue.
Choose based on data shape: REGULAR for per-packet task handling, CYCLIC for fixed-MTU line-rate ingress into a preallocated ring, MANAGED_MEMPOOL for library-managed buffers, and SHARED_MEMPOOL for multi-queue shared buffers. Always confirm support with doca_eth_rxq_cap_is_type_supported.
A started RX queue without steering is silent, not broken. Packets arrive only if a DOCA Flow rule steers traffic to the queue or the interface is in kernel promiscuous mode; also verify the port reports linkup via devlink or ip link.
DOCA_ERROR_AGAIN on TX submission means the send queue is full, not a hardware failure. Drain completions by calling doca_pe_progress before resubmitting, or raise queue sizing within the device capability limit.
Checksum offload is device-conditional and must be verified with doca_eth_txq_cap_is_l3_chksum_offload_supported against the active doca_devinfo. Requesting it on a device that does not advertise it fails with DOCA_ERROR_NOT_SUPPORTED at start.
Use doca-flow for programming steering rules that decide which packets reach which RX queue, including Flow pipes, RSS, and hairpin. DOCA Ethernet covers only the queue surface; it consumes steering output but does not program it.
davidfowlInitialize and configure the Aspire framework for an existing application.
stolyarchukGuide C++ engineers in designing and optimizing DOCA Flow pipelines on NVIDIA SmartNICs.
air-gappedAutomate NVIDIA datacenter GPU bring-up on Ubuntu 24.04 with health validation.
vikingokftCompute WooCommerce Memberships access, drip timing, grants, and discounts correctly via official hooks and APIs.