doca-comch

작성자: nvidia

사용자가 호스트 + BlueField 쌍에서 DOCA Comch 작업을 직접 수행할 때 이 스킬을 사용하세요 — 호스트 ↔ DPU PCIe 제어 평면 메시징을 설정하고, 서버 선택…

npx skills add https://github.com/nvidia/skills --skill doca-comch

DOCA Comch

Where to start: This skill assumes DOCA is already installed and the user is doing hands-on Comch work on a BlueField + host pair with DOCA. Open TASKS.md if the user wants to do something (configure / build / modify / run / test / debug); open CAPABILITIES.md when the question is what can Comch express on this version. If the user has not installed DOCA yet, route to doca-setup first. If the user is asking "is Comch even on this DOCA" because the docs they read mention doca-comm-channel, route to CAPABILITIES.md ## Version compatibility for the 2.5 rename.

Example questions this skill answers well

The CLASSES of Comch questions this skill is built to answer, each with one worked example. The agent should treat the class as the load-bearing piece — the worked example is a single instance.

  • "How do I bring up a Comch channel between host and DPU?" — worked example: "server side on the DPU, client side on the host, exchange a first control message". Answered by the role-selection + lifecycle workflow in TASKS.md ## configure + CAPABILITIES.md ## Capabilities and modes server-vs-client table.
  • "How do I move bulk data over Comch with low CPU?" — worked example: "stream a 64 KiB chunk every 100 µs from the host driver to a DPU agent". Answered by the producer / consumer fast-path described in CAPABILITIES.md ## Capabilities and modes
    • the channel-setup workflow in TASKS.md ## configure step 4, with the "slow-path vs fast-path" selection rule.
  • "What is the maximum message size I can send?" — worked example: "can I send a 4 MiB control message in one shot?". Answered by the capability-query rule (doca_comch_cap_get_max_msg_size) in CAPABILITIES.md ## Capabilities and modes
  • "Why doesn't the DPU side see the representor?" — worked example: "doca_comch_server_create returns DOCA_ERROR_NOT_PERMITTED on a freshly imaged BlueField". Answered by the representor + permission overlay in CAPABILITIES.md ## Safety policy
  • "Is this Comch capability on my installed DOCA version?" — worked example: "is doca_comch_producer in DOCA 2.6.0, or do I still need the old slow-path API?". Answered by the version-compatibility overlay in CAPABILITIES.md ## Version compatibility, which cross-links the canonical detection chain in doca-version and adds the Comch-specific 2.5 rename rule.
  • "What does this DOCA_ERROR_* from a Comch call mean and which layer caused it?" — worked example: "DOCA_ERROR_AGAIN on submitting a doca_comch_task_send via doca_task_submit". Answered by the Comch overlay on the cross-library taxonomy in CAPABILITIES.md ## Error taxonomy

Audience

This skill serves external developers building applications that consume the DOCA Comch library — i.e., users whose code calls doca_comch_* (directly in C/C++, or through FFI/bindings from another language) to exchange control or data messages between a host process and a BlueField agent over PCIe. It is not for NVIDIA developers contributing to DOCA Comch itself.

Language scope. DOCA Comch ships as a C library with pkg-config module name doca-comch. The shipped samples are written in C. C and C++ consumers are the canonical case; the worked examples in TASKS.md assume that path. Other-language consumers (Rust, Go, Python, …) consume the same *.so through FFI or language-specific bindings; the skill's contribution in that case is to keep the role-split, lifecycle, capability-discovery, permission, and error-taxonomy guidance language-neutral, and to route the agent to the public C ABI as the authoritative surface that any wrapper will eventually call.

When to load this skill

Load this skill when the user is doing hands-on DOCA Comch work, in any language. Concretely:

  • Initializing a doca_comch_server on the DPU side or a doca_comch_client on the host side, on a representor or PCIe address the host can reach.
  • Configuring at least one of: a recv callback for slow-path messages, a producer for fast-path outbound data, a consumer for fast-path inbound data, before doca_ctx_start().
  • Reading or setting comch properties via doca_comch_set_* and doca_comch_cap_get_* — max message size, max number of clients (server side), producer / consumer queue sizing.
  • Establishing a connection between the two sides and reacting to connection-state transitions via the connection callbacks registered before start.
  • Choosing between the slow-path (send-task / recv-callback, message-oriented, lower throughput, single-call simplicity) and the fast-path (producer / consumer, asynchronous, much higher throughput, two-context setup).
  • Debugging a DOCA_ERROR_* returned from a Comch call (lifecycle vs. permission vs. capability vs. would-block) and the connection state machine across the host / DPU pair.
  • Designing or extending non-C bindings (Rust, Go, Python, …) that wrap the Comch C ABI — for the lifecycle, role-split, permission, and capability rules the wrapper must honor.

Do not load this skill for general DOCA orientation, install of DOCA itself, or non-Comch library questions. For those, use doca-public-knowledge-map.

What this skill provides

This is a thin loader. The body keeps only the orientation needed to pick the right next file. The substantive Comch-specific material lives in two companion files:

  • CAPABILITIES.md — what Comch can express on this version: the server vs client role split, the slow-path send-task / recv-callback surface, the producer / consumer fast-path surface, the capability-query surface (doca_comch_cap_get_*), the Comch error taxonomy (mapped onto the cross-library DOCA_ERROR_* set), the observability surface (connection state callbacks, task completion callbacks), and the safety policy that gates representor and permission decisions.
  • TASKS.md — step-by-step workflows for the six in-scope Comch verbs: configure, build, modify, run, test, debug. Plus a Deferred task verbs block that points out-of-scope questions at the right next skill.

The skill assumes a host + BlueField pair where DOCA is already installed at the standard location and the user has the privileges their public install profile expects (in particular, sudo on the DPU side to see the host representor). It does not cover installing DOCA — that path goes through doca-setup.

What this skill deliberately does not ship

This skill is agent guidance, not a samples or templates bundle. To keep the boundary clean, it deliberately does not contain — and pull requests should not add:

  • Pre-written DOCA Comch application source code, in any language. The verified Comch source code is the shipped C samples at /opt/mellanox/doca/samples/doca_comch/<name>/. The agent's job is to route the user to those files and prescribe a minimum-diff modification on them via the universal modify-a-sample workflow in doca-programming-guide, layered with the Comch-specific overrides in TASKS.md ## modify.
  • Standalone build manifests (meson.build, CMakeLists.txt, Cargo.toml, …) parked inside the skill. The agent constructs the build manifest in the user's project directory against the user's installed DOCA, where pkg-config --modversion doca-comch is the source of truth.
  • A samples/, bindings/, or reference/ subtree of any kind. A mock or incomplete artifact in this skill's tree, even one labeled "reference", is misleading: users will read it as buildable.

Loading order

  1. Read this SKILL.md first to confirm the user's question is in scope.
  2. For the Comch capability matrix, role split, slow-path / fast-path surfaces, permission policy, error taxonomy, observability, and safety policy, see CAPABILITIES.md.
  3. For step-by-step workflows — configure, build, modify, run, test, debug — see TASKS.md.

Both companion files cross-link to each other, doca-version for the canonical version-handling rules, and doca-public-knowledge-map whenever the right answer is "look it up in the public docs or the installed package layout" rather than "Comch-specific guidance".

Related skills

  • doca-public-knowledge-map — the routing table for every public DOCA documentation source and the on-disk layout of an installed DOCA package. The Comch URL slug is DOCA-Comch (DOCA 2.5+), not doca-comm-channel.
  • doca-setup — env preparation, install verification, representor visibility checks, and the I have no install yet path with the public NGC DOCA container. This skill assumes its preconditions are satisfied.
  • doca-version — canonical DOCA version-handling rules. This skill's ## Version compatibility cross-links the four-way match rule and adds the Comch-specific 2.5 rename overlay.
  • doca-structured-tools-contract — the bundle's structured-tools precedence rule (detect / prefer / fall back / report). The Command appendix in TASKS.md honors this contract.
  • doca-programming-guide — general DOCA programming patterns shared by every library: the canonical pkg-config + meson build pattern, the universal modify-a-shipped-sample first-app workflow, the universal lifecycle, the cross-library DOCA_ERROR_* taxonomy, and the program-side debug order. This skill layers Comch specifics on top.
  • doca-debug — the cross-cutting debug ladder (install / version / build / link / runtime / program / driver). Comch-specific debug (lifecycle violations, representor visibility, slow-path vs fast-path queue-full symptoms) overlays on top of that ladder.

nvidia의 다른 스킬

compileiq-debug
nvidia
무언가 잘못되었을 때 사용: Search()가 멈추거나, 모든 평가가 INVALID_SCORE를 반환하거나, 점수가 개선되지 않거나, 모든 설정이 동일한 숫자를 반환하거나, ptxas 오류 등이 발생할 때
create-github-pr
nvidia
gh CLI를 사용하여 GitHub 풀 리퀘스트를 생성합니다. 사용자가 새 PR을 만들거나, 코드 리뷰를 제출하거나, 풀 리퀘스트를 열고자 할 때 사용합니다. 트리거 키워드 -…
nemoclaw-maintainer-cross-issue-sweep
nvidia
다른 열린 이슈들을 스캔하여 주어진 PR이 함께 수정하거나 실수로 망가뜨릴 수 있는 이슈를 찾습니다. 인접 수정 기회와 모순 위험을 file:line…과 함께 출력합니다.
fhir-basics
nvidia
에이전트에게 FHIR R4 API의 작동 방식, 사용 가능한 리소스, 검색 매개변수를 사용한 쿼리 방법, 모든 응답 형식을 올바르게 파싱하는 방법을 가르칩니다…
compileiq-validate-result
nvidia
검색이 완료된 후, 속도 향상을 청구하거나 ACF를 발송하기 전에 사용합니다. dump_results CSV를 로드하고, 상위 K개 후보(단일 목표)를 추출합니다…
changelog-audit
nvidia
릴리스 전에 Warp CHANGELOG.md를 감사합니다: 누락된 항목 복구, 사용자 영향별 정렬, 항목 언어 다듬기, 줄 바꿈, (릴리스 브랜치 모드) 비교 업데이트…
maintain-dynamic-plugins
nvidia
NeMo Relay 동적 플러그인 로더, 매니페스트, Rust 네이티브 SDK, gRPC 워커 프로토콜, Python 워커 SDK, 문서, 테스트 및 릴리스 워크플로 커버리지를 유지 관리합니다.
dgx-diagnose
nvidia
일반적인 DGX Station GB300 문제 진단 — CUDA 충돌, 잘못된 GPU 타겟팅, vLLM/SGLang 컨테이너 버그, MIG 상태 문제, NVLink/Fabric Manager 오류,…