jetson-customize-pcie

par nvidia

Activation/désactivation PCIe par contrôleur, voies et vitesse de liaison pour une carte personnalisée Jetson Thor ou Orin via ODMDATA + superposition kernel-DT. Ne PAS utiliser pour les voies UPHY…

npx skills add https://github.com/nvidia/skills --skill jetson-customize-pcie

Customize PCIe (per-controller status / lanes / speed)

Overview

PCIe on Tegra264 (Thor, pcie@C0..C5) and Tegra234 (Orin, pcie@C0..C10) is split across multiple controllers that share the UPHY lane pool with USB3 / MGBE / UFS. Each controller's runtime behavior is determined by two surfaces, both required:

SurfaceTargetAuthoritative for
ODMDATA pcie@N_status=… (+ pcie@N_max-link-speed, pcie@N_pcie-mode, pcie@N_clk-scheme, pcie-cN-endpoint-enable)/pcie/pcie@N in BPMP DTBUPHY lane power, refclk gating, controller-side power rails
Kernel-DT overlay on &pcieN/bus@0/pcie@<addr> in kernel DTBKernel probe, lane width, link speed, RC/EP mode

Skipping the kernel overlay on a disable lets the kernel probe a powered-down PHY (link timeouts in dmesg). Skipping the ODMDATA token on a disable leaves BPMP holding the PHY hot.

Agentic, not table-driven — no controller table, no questions.json. Every controller, lane width, schematic-routed receptacle, and authoritative DT node address is discovered at runtime from the docs + DTB + carrier pinmap.

The output is a kernel-DT overlay commit only. Per-controller fragment@N blocks are appended to the composite custom overlay .dts per ../../references/bsp-customization-kernel-dtb.md and committed to the bsp_sources/ hardware repo. /jetson-build-source compiles the composite to .dtbo and owns its Makefile + flash-conf registration.

This skill MUST NOT edit ODMDATA="...". All ODMDATA tokens (pcie@N_status=…, pcie@N_max-link-speed, pcie@N_pcie-mode, pcie@N_clk-scheme, pcie-cN-endpoint-enable, plus the uphyX-config-N surface tokens and UPHY_CONFIG="" clear) are emitted by /jetson-customize-uphy in a single atomic commit on the carrier flash-conf fork. The allocation table this skill consumes from the UPHY sidecar already tells the operator which controllers are okay / disabled / per-lane sized; this skill only translates that table into kernel-DT overlay fragments and verifies that the overlay agrees with the ODMDATA already committed by customize-uphy (consistency check in Step 8 — disagreement is reported, not silently fixed).

When to invoke

  • The user says "configure PCIe", "enable PCIe controller", "set PCIe num-lanes", "change PCIe link speed", or asks to flip a pcie@N_status token.
  • A specific PCIe slot or M.2 receptacle doesn't enumerate after flash, OR the link trains at the wrong width / speed.
  • jetson-customize-uphy ran and re-allocated lanes across PCIe controllers (e.g. switched from uphy0-config-7 to uphy0-config-6 enabling PCIe C3); the per-controller side now needs to be brought up.
  • jetson-customize-mgbe reports the QSFP path is wired but the kernel doesn't probe its PCIe-side companion (rare; XFI configurations).

Prerequisites:

  • Active profile with reference_devkit: + custom_carrier: blocks.
  • <source.root_path>/Linux_for_Tegra/.git exists (/jetson-init-source).
  • /jetson-derive-carrier has run — carrier flash-conf fork is in the overlay tracker.
  • /jetson-customize-uphy has run — its JSON sidecar at <workspace>/target-platform/<profile-stem>.jetson-customize-uphy.json drives the per-controller enable decision.
  • Source-of-truth docs registered or supplied at prompt: Adaptation Guide, Module Design Guide, SoC TRM.
  • When custom_carrier: is present, both documents.custom_carrier_schematic AND documents.custom_carrier_pinmux_xls are REQUIRED. Refuse the run if either is missing — routing on a custom carrier cannot be guessed. Reference-devkit-only profiles skip this check.
  • dtc on PATH.

Procedure (summary)

Full step-by-step walkthrough lives in references/procedure.md. High-level flow:

  1. Resolve active target + open source-of-truth documents (incl. <carrier-pinmap>, <ref-dtb>, <uphy-state>). Refuse if <uphy-state> is missing.
  2. Diff PCIe topology — devkit vs custom carrier — by decompiling <ref-dtb> and grepping the schematic for PEX<N>_* net labels.
  3. AskUserQuestion (multiSelect) — which controllers to customize.
  4. Per-controller verification: pinmap + schematic + pin_verifier.py for PE<N>_CLKREQ_L, PE<N>_RST_L, optional PE<N>_WAKE_L.
  5. Auto-derive per-controller plan (enable from <uphy-state>, lanes / speed from Adaptation Guide, mode hard-pinned to "rc") → mandatory confirm-or-customize gate.
  6. Append per-controller fragment@N blocks (marker /* custom-bsp: pcie:pcie@<addr> */) to the composite custom overlay .dts in bsp_sources/. Pre-flight dtc + fdtoverlay. Commit via the workflow's preview gate. Do not edit ODMDATA/jetson-customize-uphy already emitted pcie@N_status=…, pcie@N_max-link-speed, pcie@N_pcie-mode, pcie@N_clk-scheme, and pcie-cN-endpoint-enable in its single atomic ODMDATA commit. This skill only translates the per-controller plan into kernel-DT overlay fragments.
  7. (Step folded into Step 6 — overlay-only emission.)
  8. Cross-check ODMDATA vs overlay consistency. On a contradictory row, stop and ask the user how to recover the two commits. Never run git reset --hard autonomously.
  9. Write run-state JSON sidecar at <workspace>/target-platform/<profile-stem>.jetson-customize-pcie.json
    • summary, then drive the downstream next-step chain via sequential AskUserQuestion prompts per references/procedure.md Step 9. Never substitute a printed "Next step: …" line for the prompts.

Limitations

  • Mode hard-pinned to RC. Endpoint mode is only emitted when the operator passes mode_override="ep" in Step 5c.
  • enable is derived, not asked. UPHY-allocated controllers are mandatorily okay; non-allocated are mandatorily disabled.
  • No upstream BSP edits. Output lands in Linux_for_Tegra/ + bsp_sources/ only.
  • Pre-flight overlay merge is a sanity check, not the production build. /jetson-build-source is authoritative.
  • Flash-conf overlay registration is out of scope. Owned by /jetson-build-source Step 5.0a.

Troubleshooting

  • <uphy-state> missing → run /jetson-customize-uphy first.
  • Slot doesn't enumerate after flash → check dmesg | grep pcie; re-verify ODMDATA pcie@<N>_status=okay and the overlay fragment agree (Step 8 table in references/procedure.md).
  • Link trains at wrong width → confirm UPHY config in <uphy-state> allocates the expected lane count; the kernel fragment's num-lanes must match.
  • compatible mismatch → fix the composite root, not the fragment. UEFI plugin-manager silently skips on mismatch.
  • Contradictory ODMDATA-vs-overlay row → ask the user; do not auto-git reset --hard. See gotchas.
  • Common pitfalls — see references/gotchas.md (RC pinning, node- address sourcing, stock-disabled controllers, intra-file handoff with jetson-customize-uphy).

References

Plus de skills de nvidia

compileiq-debug
nvidia
Utilisez quand quelque chose ne va pas : Search() bloque, toutes les évaluations retournent INVALID_SCORE, les scores ne s'améliorent pas, chaque configuration retourne le même nombre, erreurs ptxas…
create-github-pr
nvidia
Créer des pull requests GitHub en utilisant l'interface en ligne de commande gh. Utiliser lorsque l'utilisateur souhaite créer une nouvelle PR, soumettre du code pour révision, ou ouvrir une pull request. Mots-clés de déclenchement -…
nemoclaw-maintainer-cross-issue-sweep
nvidia
Analyse les autres problèmes ouverts pour trouver ceux qu’une PR donnée pourrait également corriger ou casser accidentellement. Génère des opportunités de correctifs adjacents et des risques de contradiction avec fichier:ligne…
fhir-basics
nvidia
Apprend aux agents comment fonctionnent les API FHIR R4, quelles ressources sont disponibles, comment les interroger avec des paramètres de recherche, et comment analyser correctement tous les formats de réponse…
compileiq-validate-result
nvidia
Utiliser APRÈS qu'une recherche soit terminée et AVANT de réclamer un accélérateur ou d'expédier un ACF. Charge le CSV dump_results, extrait les K meilleurs candidats (mono-objectif)…
changelog-audit
nvidia
Auditer le CHANGELOG.md de Warp avant une publication : récupérer les entrées perdues, trier par impact utilisateur, affiner le langage des entrées, ajuster les retours à la ligne et (en mode branche de publication) mettre à jour la comparaison…
maintain-dynamic-plugins
nvidia
Maintenir les chargeurs de plugins dynamiques NeMo Relay, les manifestes, les SDK natifs Rust, le protocole worker gRPC, le SDK worker Python, la documentation, les tests et la couverture du workflow de publication
dgx-diagnose
nvidia
Diagnostiquer les problèmes courants du DGX Station GB300 — plantages CUDA, ciblage incorrect du GPU, bugs de conteneur vLLM/SGLang, problèmes d'état MIG, erreurs NVLink/Fabric Manager,…